auto import from //depot/cupcake/@135843
diff --git a/vm/mterp/out/InterpAsm-allstubs.S b/vm/mterp/out/InterpAsm-allstubs.S
new file mode 100644
index 0000000..a6973ae
--- /dev/null
+++ b/vm/mterp/out/InterpAsm-allstubs.S
@@ -0,0 +1,35 @@
+/*
+ * This file was generated automatically by gen-mterp.py for 'allstubs'.
+ *
+ * --> DO NOT EDIT <--
+ */
+
+
+    .global dvmAsmInstructionStart
+    .type   dvmAsmInstructionStart, %function
+dvmAsmInstructionStart = .L_OP_NOP
+    .text
+
+    .balign 64
+.L_OP_NOP:   /* dummy */
+
+    .balign 64
+    .size   dvmAsmInstructionStart, .-dvmAsmInstructionStart
+    .global dvmAsmInstructionEnd
+dvmAsmInstructionEnd:
+
+/*
+ * ===========================================================================
+ *  Sister implementations
+ * ===========================================================================
+ */
+    .global dvmAsmSisterStart
+    .type   dvmAsmSisterStart, %function
+    .text
+    .balign 4
+dvmAsmSisterStart:
+
+    .size   dvmAsmSisterStart, .-dvmAsmSisterStart
+    .global dvmAsmSisterEnd
+dvmAsmSisterEnd:
+
diff --git a/vm/mterp/out/InterpAsm-armv4.S b/vm/mterp/out/InterpAsm-armv4.S
new file mode 100644
index 0000000..3de7aea
--- /dev/null
+++ b/vm/mterp/out/InterpAsm-armv4.S
@@ -0,0 +1,10012 @@
+/*
+ * This file was generated automatically by gen-mterp.py for 'armv4'.
+ *
+ * --> DO NOT EDIT <--
+ */
+
+/* File: armv5te/header.S */
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ * ARMv5 definitions and declarations.
+ */
+
+/*
+ARM EABI general notes:
+
+r0-r3 hold first 4 args to a method; they are not preserved across method calls
+r4-r8 are available for general use
+r9 is given special treatment in some situations, but not for us
+r10 (sl) seems to be generally available
+r11 (fp) is used by gcc (unless -fomit-frame-pointer is set)
+r12 (ip) is scratch -- not preserved across method calls
+r13 (sp) should be managed carefully in case a signal arrives
+r14 (lr) must be preserved
+r15 (pc) can be tinkered with directly
+
+r0 holds returns of <= 4 bytes
+r0-r1 hold returns of 8 bytes, low word in r0
+
+Callee must save/restore r4+ (except r12) if it modifies them.
+
+Stack is "full descending".  Only the arguments that don't fit in the first 4
+registers are placed on the stack.  "sp" points at the first stacked argument
+(i.e. the 5th arg).
+
+VFP: single-precision results in s0, double-precision results in d0.
+
+In the EABI, "sp" must be 64-bit aligned on entry to a function, and any
+64-bit quantities (long long, double) must be 64-bit aligned.
+*/
+
+/*
+Mterp and ARM notes:
+
+The following registers have fixed assignments:
+
+  reg nick      purpose
+  r4  rPC       interpreted program counter, used for fetching instructions
+  r5  rFP       interpreted frame pointer, used for accessing locals and args
+  r6  rGLUE     MterpGlue pointer
+  r7  rIBASE    interpreted instruction base pointer, used for computed goto
+  r8  rINST     first 16-bit code unit of current instruction
+
+Macros are provided for common operations.  Each macro MUST emit only
+one instruction to make instruction-counting easier.  They MUST NOT alter
+unspecified registers or condition codes.
+*/
+
+/* single-purpose registers, given names for clarity */
+#define rPC     r4
+#define rFP     r5
+#define rGLUE   r6
+#define rIBASE  r7
+#define rINST   r8
+
+/* save/restore the PC and/or FP from the glue struct */
+#define LOAD_PC_FROM_GLUE()     ldr     rPC, [rGLUE, #offGlue_pc]
+#define SAVE_PC_TO_GLUE()       str     rPC, [rGLUE, #offGlue_pc]
+#define LOAD_FP_FROM_GLUE()     ldr     rFP, [rGLUE, #offGlue_fp]
+#define SAVE_FP_TO_GLUE()       str     rFP, [rGLUE, #offGlue_fp]
+#define LOAD_PC_FP_FROM_GLUE()  ldmia   rGLUE, {rPC, rFP}
+#define SAVE_PC_FP_TO_GLUE()    stmia   rGLUE, {rPC, rFP}
+
+/*
+ * "export" the PC to the stack frame, f/b/o future exception objects.  Must
+ * be done *before* something calls dvmThrowException.
+ *
+ * In C this is "SAVEAREA_FROM_FP(fp)->xtra.currentPc = pc", i.e.
+ * fp - sizeof(StackSaveArea) + offsetof(SaveArea, xtra.currentPc)
+ *
+ * It's okay to do this more than once.
+ */
+#define EXPORT_PC() \
+    str     rPC, [rFP, #(-sizeofStackSaveArea + offStackSaveArea_currentPc)]
+
+/*
+ * Given a frame pointer, find the stack save area.
+ *
+ * In C this is "((StackSaveArea*)(_fp) -1)".
+ */
+#define SAVEAREA_FROM_FP(_reg, _fpreg) \
+    sub     _reg, _fpreg, #sizeofStackSaveArea
+
+/*
+ * Fetch the next instruction from rPC into rINST.  Does not advance rPC.
+ */
+#define FETCH_INST()            ldrh    rINST, [rPC]
+
+/*
+ * Fetch the next instruction from the specified offset.  Advances rPC
+ * to point to the next instruction.  "_count" is in 16-bit code units.
+ *
+ * Because of the limited size of immediate constants on ARM, this is only
+ * suitable for small forward movements (i.e. don't try to implement "goto"
+ * with this).
+ *
+ * This must come AFTER anything that can throw an exception, or the
+ * exception catch may miss.  (This also implies that it must come after
+ * EXPORT_PC().)
+ */
+#define FETCH_ADVANCE_INST(_count) ldrh    rINST, [rPC, #(_count*2)]!
+
+/*
+ * Fetch the next instruction from an offset specified by _reg.  Updates
+ * rPC to point to the next instruction.  "_reg" must specify the distance
+ * in bytes, *not* 16-bit code units, and may be a signed value.
+ *
+ * We want to write "ldrh rINST, [rPC, _reg, lsl #2]!", but some of the
+ * bits that hold the shift distance are used for the half/byte/sign flags.
+ * In some cases we can pre-double _reg for free, so we require a byte offset
+ * here.
+ */
+#define FETCH_ADVANCE_INST_RB(_reg) ldrh    rINST, [rPC, _reg]!
+
+/*
+ * Fetch a half-word code unit from an offset past the current PC.  The
+ * "_count" value is in 16-bit code units.  Does not advance rPC.
+ *
+ * The "_S" variant works the same but treats the value as signed.
+ */
+#define FETCH(_reg, _count)     ldrh    _reg, [rPC, #(_count*2)]
+#define FETCH_S(_reg, _count)   ldrsh   _reg, [rPC, #(_count*2)]
+
+/*
+ * Fetch one byte from an offset past the current PC.  Pass in the same
+ * "_count" as you would for FETCH, and an additional 0/1 indicating which
+ * byte of the halfword you want (lo/hi).
+ */
+#define FETCH_B(_reg, _count, _byte) ldrb     _reg, [rPC, #(_count*2+_byte)]
+
+/*
+ * Put the instruction's opcode field into the specified register.
+ */
+#define GET_INST_OPCODE(_reg)   and     _reg, rINST, #255
+
+/*
+ * Begin executing the opcode in _reg.  Because this only jumps within the
+ * interpreter, we don't have to worry about pre-ARMv5 THUMB interwork.
+ */
+#define GOTO_OPCODE(_reg)       add     pc, rIBASE, _reg, lsl #6
+
+/*
+ * Get/set the 32-bit value from a Dalvik register.
+ */
+#define GET_VREG(_reg, _vreg)   ldr     _reg, [rFP, _vreg, lsl #2]
+#define SET_VREG(_reg, _vreg)   str     _reg, [rFP, _vreg, lsl #2]
+
+/*
+ * This is a #include, not a %include, because we want the C pre-processor
+ * to expand the macros into assembler assignment statements.
+ */
+#include "../common/asm-constants.h"
+
+
+/* File: armv5te/platform.S */
+/*
+ * ===========================================================================
+ *  CPU-version-specific defines
+ * ===========================================================================
+ */
+
+/*
+ * Macro for "LDR PC,xxx", which is not allowed pre-ARMv5.  Essentially a
+ * one-way branch.
+ *
+ * May modify IP.  Does not modify LR.
+ */
+.macro  LDR_PC source
+    ldr     pc, \source
+.endm
+
+/*
+ * Macro for "MOV LR,PC / LDR PC,xxx", which is not allowed pre-ARMv5.
+ * Jump to subroutine.
+ *
+ * May modify IP and LR.
+ */
+.macro  LDR_PC_LR source
+    mov     lr, pc
+    ldr     pc, \source
+.endm
+
+/*
+ * Macro for "LDMFD SP!, {...regs...,PC}".
+ *
+ * May modify IP and LR.
+ */
+.macro  LDMFD_PC regs
+    ldmfd   sp!, {\regs,pc}
+.endm
+
+
+/* File: armv5te/entry.S */
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ * Interpreter entry point.
+ */
+
+/*
+ * We don't have formal stack frames, so gdb scans upward in the code
+ * to find the start of the function (a label with the %function type),
+ * and then looks at the next few instructions to figure out what
+ * got pushed onto the stack.  From this it figures out how to restore
+ * the registers, including PC, for the previous stack frame.  If gdb
+ * sees a non-function label, it stops scanning, so either we need to
+ * have nothing but assembler-local labels between the entry point and
+ * the break, or we need to fake it out.
+ *
+ * When this is defined, we add some stuff to make gdb less confused.
+ */
+#define ASSIST_DEBUGGER 1
+
+    .text
+    .align  2
+    .global dvmMterpStdRun
+    .type   dvmMterpStdRun, %function
+
+/*
+ * On entry:
+ *  r0  MterpGlue* glue
+ *
+ * This function returns a boolean "changeInterp" value.  The return comes
+ * via a call to dvmMterpStdBail().
+ */
+dvmMterpStdRun:
+#define MTERP_ENTRY1 \
+    .save {r4-r10,fp,lr}; \
+    stmfd   sp!, {r4-r10,fp,lr}         @ save 9 regs
+#define MTERP_ENTRY2 \
+    .pad    #4; \
+    sub     sp, sp, #4                  @ align 64
+
+    .fnstart
+    MTERP_ENTRY1
+    MTERP_ENTRY2
+
+    /* save stack pointer, add magic word for debuggerd */
+    str     sp, [r0, #offGlue_bailPtr]  @ save SP for eventual return
+
+    /* set up "named" registers, figure out entry point */
+    mov     rGLUE, r0                   @ set rGLUE
+    ldrb    r1, [r0, #offGlue_entryPoint]   @ InterpEntry enum is char
+    LOAD_PC_FP_FROM_GLUE()              @ load rPC and rFP from "glue"
+    adr     rIBASE, dvmAsmInstructionStart  @ set rIBASE
+    cmp     r1, #kInterpEntryInstr      @ usual case?
+    bne     .Lnot_instr                 @ no, handle it
+
+    /* start executing the instruction at rPC */
+    FETCH_INST()                        @ load rINST from rPC
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+.Lnot_instr:
+    cmp     r1, #kInterpEntryReturn     @ were we returning from a method?
+    beq     common_returnFromMethod
+
+.Lnot_return:
+    cmp     r1, #kInterpEntryThrow      @ were we throwing an exception?
+    beq     common_exceptionThrown
+
+.Lbad_arg:
+    ldr     r0, strBadEntryPoint
+    @ r1 holds value of entryPoint
+    bl      printf
+    bl      dvmAbort
+    .fnend
+
+
+    .global dvmMterpStdBail
+    .type   dvmMterpStdBail, %function
+
+/*
+ * Restore the stack pointer and PC from the save point established on entry.
+ * This is essentially the same as a longjmp, but should be cheaper.  The
+ * last instruction causes us to return to whoever called dvmMterpStdRun.
+ *
+ * We pushed some registers on the stack in dvmMterpStdRun, then saved
+ * SP and LR.  Here we restore SP, restore the registers, and then restore
+ * LR to PC.
+ *
+ * On entry:
+ *  r0  MterpGlue* glue
+ *  r1  bool changeInterp
+ */
+dvmMterpStdBail:
+    ldr     sp, [r0, #offGlue_bailPtr]      @ sp<- saved SP
+    mov     r0, r1                          @ return the changeInterp value
+    add     sp, sp, #4                      @ un-align 64
+    LDMFD_PC "r4-r10,fp"                    @ restore 9 regs and return
+
+
+/*
+ * String references.
+ */
+strBadEntryPoint:
+    .word   .LstrBadEntryPoint
+
+
+
+    .global dvmAsmInstructionStart
+    .type   dvmAsmInstructionStart, %function
+dvmAsmInstructionStart = .L_OP_NOP
+    .text
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NOP: /* 0x00 */
+/* File: armv5te/OP_NOP.S */
+    FETCH_ADVANCE_INST(1)               @ advance to next instr, load rINST
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    GOTO_OPCODE(ip)                     @ execute it
+
+#ifdef ASSIST_DEBUGGER
+    /* insert fake function header to help gdb find the stack frame */
+    .type   dalvik_inst, %function
+dalvik_inst:
+    .fnstart
+    MTERP_ENTRY1
+    MTERP_ENTRY2
+    .fnend
+#endif
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE: /* 0x01 */
+/* File: armv5te/OP_MOVE.S */
+    /* for move, move-object, long-to-int */
+    /* op vA, vB */
+    mov     r1, rINST, lsr #12          @ r1<- B from 15:12
+    mov     r0, rINST, lsr #8           @ r0<- A from 11:8
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[B]
+    and     r0, r0, #15
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[A]<- r2
+    GOTO_OPCODE(ip)                     @ execute next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_FROM16: /* 0x02 */
+/* File: armv5te/OP_MOVE_FROM16.S */
+    /* for: move/from16, move-object/from16 */
+    /* op vAA, vBBBB */
+    FETCH(r1, 1)                        @ r1<- BBBB
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[BBBB]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[AA]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_16: /* 0x03 */
+/* File: armv5te/OP_MOVE_16.S */
+    /* for: move/16, move-object/16 */
+    /* op vAAAA, vBBBB */
+    FETCH(r1, 2)                        @ r1<- BBBB
+    FETCH(r0, 1)                        @ r0<- AAAA
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[BBBB]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[AAAA]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_WIDE: /* 0x04 */
+/* File: armv5te/OP_MOVE_WIDE.S */
+    /* move-wide vA, vB */
+    /* NOTE: regs can overlap, e.g. "move v6,v7" or "move v7,v6" */
+    mov     r2, rINST, lsr #8           @ r2<- A(+)
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r2, r2, #15
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[A]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- fp[B]
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r2, {r0-r1}                 @ fp[A]<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_WIDE_FROM16: /* 0x05 */
+/* File: armv5te/OP_MOVE_WIDE_FROM16.S */
+    /* move-wide/from16 vAA, vBBBB */
+    /* NOTE: regs can overlap, e.g. "move v6,v7" or "move v7,v6" */
+    FETCH(r3, 1)                        @ r3<- BBBB
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[BBBB]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[AA]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- fp[BBBB]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r2, {r0-r1}                 @ fp[AA]<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_WIDE_16: /* 0x06 */
+/* File: armv5te/OP_MOVE_WIDE_16.S */
+    /* move-wide/16 vAAAA, vBBBB */
+    /* NOTE: regs can overlap, e.g. "move v6,v7" or "move v7,v6" */
+    FETCH(r3, 2)                        @ r3<- BBBB
+    FETCH(r2, 1)                        @ r2<- AAAA
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[BBBB]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[AAAA]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- fp[BBBB]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r2, {r0-r1}                 @ fp[AAAA]<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_OBJECT: /* 0x07 */
+/* File: armv5te/OP_MOVE_OBJECT.S */
+/* File: armv5te/OP_MOVE.S */
+    /* for move, move-object, long-to-int */
+    /* op vA, vB */
+    mov     r1, rINST, lsr #12          @ r1<- B from 15:12
+    mov     r0, rINST, lsr #8           @ r0<- A from 11:8
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[B]
+    and     r0, r0, #15
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[A]<- r2
+    GOTO_OPCODE(ip)                     @ execute next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_OBJECT_FROM16: /* 0x08 */
+/* File: armv5te/OP_MOVE_OBJECT_FROM16.S */
+/* File: armv5te/OP_MOVE_FROM16.S */
+    /* for: move/from16, move-object/from16 */
+    /* op vAA, vBBBB */
+    FETCH(r1, 1)                        @ r1<- BBBB
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[BBBB]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[AA]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_OBJECT_16: /* 0x09 */
+/* File: armv5te/OP_MOVE_OBJECT_16.S */
+/* File: armv5te/OP_MOVE_16.S */
+    /* for: move/16, move-object/16 */
+    /* op vAAAA, vBBBB */
+    FETCH(r1, 2)                        @ r1<- BBBB
+    FETCH(r0, 1)                        @ r0<- AAAA
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[BBBB]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[AAAA]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_RESULT: /* 0x0a */
+/* File: armv5te/OP_MOVE_RESULT.S */
+    /* for: move-result, move-result-object */
+    /* op vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    ldr     r0, [rGLUE, #offGlue_retval]    @ r0<- glue->retval.i
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[AA]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_RESULT_WIDE: /* 0x0b */
+/* File: armv5te/OP_MOVE_RESULT_WIDE.S */
+    /* move-result-wide vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    add     r3, rGLUE, #offGlue_retval  @ r3<- &glue->retval
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[AA]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- retval.j
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r2, {r0-r1}                 @ fp[AA]<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_RESULT_OBJECT: /* 0x0c */
+/* File: armv5te/OP_MOVE_RESULT_OBJECT.S */
+/* File: armv5te/OP_MOVE_RESULT.S */
+    /* for: move-result, move-result-object */
+    /* op vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    ldr     r0, [rGLUE, #offGlue_retval]    @ r0<- glue->retval.i
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[AA]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_EXCEPTION: /* 0x0d */
+/* File: armv5te/OP_MOVE_EXCEPTION.S */
+    /* move-exception vAA */
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- glue->self
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    ldr     r3, [r0, #offThread_exception]  @ r3<- dvmGetException bypass
+    mov     r1, #0                      @ r1<- 0
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    SET_VREG(r3, r2)                    @ fp[AA]<- exception obj
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offThread_exception]  @ dvmClearException bypass
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RETURN_VOID: /* 0x0e */
+/* File: armv5te/OP_RETURN_VOID.S */
+    b       common_returnFromMethod
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RETURN: /* 0x0f */
+/* File: armv5te/OP_RETURN.S */
+    /*
+     * Return a 32-bit value.  Copies the return value into the "glue"
+     * structure, then jumps to the return handler.
+     *
+     * for: return, return-object
+     */
+    /* op vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    GET_VREG(r0, r2)                    @ r0<- vAA
+    str     r0, [rGLUE, #offGlue_retval] @ retval.i <- vAA
+    b       common_returnFromMethod
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RETURN_WIDE: /* 0x10 */
+/* File: armv5te/OP_RETURN_WIDE.S */
+    /*
+     * Return a 64-bit value.  Copies the return value into the "glue"
+     * structure, then jumps to the return handler.
+     */
+    /* return-wide vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[AA]
+    add     r3, rGLUE, #offGlue_retval  @ r3<- &glue->retval
+    ldmia   r2, {r0-r1}                 @ r0/r1 <- vAA/vAA+1
+    stmia   r3, {r0-r1}                 @ retval<- r0/r1
+    b       common_returnFromMethod
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RETURN_OBJECT: /* 0x11 */
+/* File: armv5te/OP_RETURN_OBJECT.S */
+/* File: armv5te/OP_RETURN.S */
+    /*
+     * Return a 32-bit value.  Copies the return value into the "glue"
+     * structure, then jumps to the return handler.
+     *
+     * for: return, return-object
+     */
+    /* op vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    GET_VREG(r0, r2)                    @ r0<- vAA
+    str     r0, [rGLUE, #offGlue_retval] @ retval.i <- vAA
+    b       common_returnFromMethod
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_4: /* 0x12 */
+/* File: armv5te/OP_CONST_4.S */
+    /* const/4 vA, #+B */
+    mov     r1, rINST, lsl #16          @ r1<- Bxxx0000
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mov     r1, r1, asr #28             @ r1<- sssssssB (sign-extended)
+    and     r0, r0, #15
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    SET_VREG(r1, r0)                    @ fp[A]<- r1
+    GOTO_OPCODE(ip)                     @ execute next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_16: /* 0x13 */
+/* File: armv5te/OP_CONST_16.S */
+    /* const/16 vAA, #+BBBB */
+    FETCH_S(r0, 1)                      @ r0<- ssssBBBB (sign-extended)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r0, r3)                    @ vAA<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST: /* 0x14 */
+/* File: armv5te/OP_CONST.S */
+    /* const vAA, #+BBBBbbbb */
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH(r0, 1)                        @ r0<- bbbb (low)
+    FETCH(r1, 2)                        @ r1<- BBBB (high)
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    orr     r0, r0, r1, lsl #16         @ r0<- BBBBbbbb
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r3)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_HIGH16: /* 0x15 */
+/* File: armv5te/OP_CONST_HIGH16.S */
+    /* const/high16 vAA, #+BBBB0000 */
+    FETCH(r0, 1)                        @ r0<- 0000BBBB (zero-extended)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    mov     r0, r0, lsl #16             @ r0<- BBBB0000
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r0, r3)                    @ vAA<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_WIDE_16: /* 0x16 */
+/* File: armv5te/OP_CONST_WIDE_16.S */
+    /* const-wide/16 vAA, #+BBBB */
+    FETCH_S(r0, 1)                      @ r0<- ssssBBBB (sign-extended)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    mov     r1, r0, asr #31             @ r1<- ssssssss
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r3, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_WIDE_32: /* 0x17 */
+/* File: armv5te/OP_CONST_WIDE_32.S */
+    /* const-wide/32 vAA, #+BBBBbbbb */
+    FETCH(r0, 1)                        @ r0<- 0000bbbb (low)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH_S(r2, 2)                      @ r2<- ssssBBBB (high)
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    orr     r0, r0, r2, lsl #16         @ r0<- BBBBbbbb
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[AA]
+    mov     r1, r0, asr #31             @ r1<- ssssssss
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r3, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_WIDE: /* 0x18 */
+/* File: armv5te/OP_CONST_WIDE.S */
+    /* const-wide vAA, #+HHHHhhhhBBBBbbbb */
+    FETCH(r0, 1)                        @ r0<- bbbb (low)
+    FETCH(r1, 2)                        @ r1<- BBBB (low middle)
+    FETCH(r2, 3)                        @ r2<- hhhh (high middle)
+    orr     r0, r0, r1, lsl #16         @ r0<- BBBBbbbb (low word)
+    FETCH(r3, 4)                        @ r3<- HHHH (high)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    orr     r1, r2, r3, lsl #16         @ r1<- HHHHhhhh (high word)
+    FETCH_ADVANCE_INST(5)               @ advance rPC, load rINST
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_WIDE_HIGH16: /* 0x19 */
+/* File: armv5te/OP_CONST_WIDE_HIGH16.S */
+    /* const-wide/high16 vAA, #+BBBB000000000000 */
+    FETCH(r1, 1)                        @ r1<- 0000BBBB (zero-extended)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    mov     r0, #0                      @ r0<- 00000000
+    mov     r1, r1, lsl #16             @ r1<- BBBB0000
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r3, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_STRING: /* 0x1a */
+/* File: armv5te/OP_CONST_STRING.S */
+    /* const/string vAA, String@BBBB */
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- glue->methodClassDex
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    ldr     r2, [r2, #offDvmDex_pResStrings]   @ r2<- dvmDex->pResStrings
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- pResStrings[BBBB]
+    cmp     r0, #0                      @ not yet resolved?
+    beq     .LOP_CONST_STRING_resolve
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_STRING_JUMBO: /* 0x1b */
+/* File: armv5te/OP_CONST_STRING_JUMBO.S */
+    /* const/string vAA, String@BBBBBBBB */
+    FETCH(r0, 1)                        @ r0<- bbbb (low)
+    FETCH(r1, 2)                        @ r1<- BBBB (high)
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- glue->methodClassDex
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    ldr     r2, [r2, #offDvmDex_pResStrings]   @ r2<- dvmDex->pResStrings
+    orr     r1, r0, r1, lsl #16         @ r1<- BBBBbbbb
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- pResStrings[BBBB]
+    cmp     r0, #0
+    beq     .LOP_CONST_STRING_JUMBO_resolve
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_CLASS: /* 0x1c */
+/* File: armv5te/OP_CONST_CLASS.S */
+    /* const/class vAA, Class@BBBB */
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- glue->methodClassDex
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    ldr     r2, [r2, #offDvmDex_pResClasses]   @ r2<- dvmDex->pResClasses
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- pResClasses[BBBB]
+    cmp     r0, #0                      @ not yet resolved?
+    beq     .LOP_CONST_CLASS_resolve
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MONITOR_ENTER: /* 0x1d */
+/* File: armv5te/OP_MONITOR_ENTER.S */
+    /*
+     * Synchronize on an object.
+     */
+    /* monitor-enter vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    GET_VREG(r1, r2)                    @ r1<- vAA (object)
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- glue->self
+    cmp     r1, #0                      @ null object?
+#ifdef WITH_MONITOR_TRACKING
+    EXPORT_PC()                         @ export PC so we can grab stack trace
+#endif
+    beq     common_errNullObject        @ null object, throw an exception
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    bl      dvmLockObject               @ call(self, obj)
+#ifdef WITH_DEADLOCK_PREDICTION /* implies WITH_MONITOR_TRACKING */
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- glue->self
+    ldr     r1, [r0, #offThread_exception] @ check for exception
+    cmp     r1, #0
+    bne     common_exceptionThrown      @ exception raised, bail out
+#endif
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MONITOR_EXIT: /* 0x1e */
+/* File: armv5te/OP_MONITOR_EXIT.S */
+    /*
+     * Unlock an object.
+     *
+     * Exceptions that occur when unlocking a monitor need to appear as
+     * if they happened at the following instruction.  See the Dalvik
+     * instruction spec.
+     */
+    /* monitor-exit vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    EXPORT_PC()                         @ before fetch: export the PC
+    GET_VREG(r1, r2)                    @ r1<- vAA (object)
+    cmp     r1, #0                      @ null object?
+    beq     common_errNullObject        @ yes
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- glue->self
+    bl      dvmUnlockObject             @ r0<- success for unlock(self, obj)
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yes, exception is pending
+    FETCH_ADVANCE_INST(1)               @ before throw: advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CHECK_CAST: /* 0x1f */
+/* File: armv5te/OP_CHECK_CAST.S */
+    /*
+     * Check to see if a cast from one class to another is allowed.
+     */
+    /* check-cast vAA, class@BBBB */
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH(r2, 1)                        @ r2<- BBBB
+    GET_VREG(r9, r3)                    @ r9<- object
+    ldr     r0, [rGLUE, #offGlue_methodClassDex]    @ r0<- pDvmDex
+    cmp     r9, #0                      @ is object null?
+    ldr     r0, [r0, #offDvmDex_pResClasses]    @ r0<- pDvmDex->pResClasses
+    beq     .LOP_CHECK_CAST_okay            @ null obj, cast always succeeds
+    ldr     r1, [r0, r2, lsl #2]        @ r1<- resolved class
+    ldr     r0, [r9, #offObject_clazz]  @ r0<- obj->clazz
+    cmp     r1, #0                      @ have we resolved this before?
+    beq     .LOP_CHECK_CAST_resolve         @ not resolved, do it now
+.LOP_CHECK_CAST_resolved:
+    cmp     r0, r1                      @ same class (trivial success)?
+    bne     .LOP_CHECK_CAST_fullcheck       @ no, do full check
+.LOP_CHECK_CAST_okay:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INSTANCE_OF: /* 0x20 */
+/* File: armv5te/OP_INSTANCE_OF.S */
+    /*
+     * Check to see if an object reference is an instance of a class.
+     *
+     * Most common situation is a non-null object, being compared against
+     * an already-resolved class.
+     */
+    /* instance-of vA, vB, class@CCCC */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB (object)
+    and     r9, r9, #15                 @ r9<- A
+    cmp     r0, #0                      @ is object null?
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- pDvmDex
+    beq     .LOP_INSTANCE_OF_store           @ null obj, not an instance, store r0
+    FETCH(r3, 1)                        @ r3<- CCCC
+    ldr     r2, [r2, #offDvmDex_pResClasses]    @ r2<- pDvmDex->pResClasses
+    ldr     r1, [r2, r3, lsl #2]        @ r1<- resolved class
+    ldr     r0, [r0, #offObject_clazz]  @ r0<- obj->clazz
+    cmp     r1, #0                      @ have we resolved this before?
+    beq     .LOP_INSTANCE_OF_resolve         @ not resolved, do it now
+.LOP_INSTANCE_OF_resolved: @ r0=obj->clazz, r1=resolved class
+    cmp     r0, r1                      @ same class (trivial success)?
+    beq     .LOP_INSTANCE_OF_trivial         @ yes, trivial finish
+    b       .LOP_INSTANCE_OF_fullcheck       @ no, do full check
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ARRAY_LENGTH: /* 0x21 */
+/* File: armv5te/OP_ARRAY_LENGTH.S */
+    /*
+     * Return the length of an array.
+     */
+    mov     r1, rINST, lsr #12          @ r1<- B
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    GET_VREG(r0, r1)                    @ r0<- vB (object ref)
+    and     r2, r2, #15                 @ r2<- A
+    cmp     r0, #0                      @ is object null?
+    beq     common_errNullObject        @ yup, fail
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- array length
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r3, r2)                    @ vB<- length
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEW_INSTANCE: /* 0x22 */
+/* File: armv5te/OP_NEW_INSTANCE.S */
+    /*
+     * Create a new instance of a class.
+     */
+    /* new-instance vAA, class@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResClasses]    @ r3<- pDvmDex->pResClasses
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved class
+    EXPORT_PC()                         @ req'd for init, resolve, alloc
+    cmp     r0, #0                      @ already resolved?
+    beq     .LOP_NEW_INSTANCE_resolve         @ no, resolve it now
+.LOP_NEW_INSTANCE_resolved:   @ r0=class
+    ldrb    r1, [r0, #offClassObject_status]    @ r1<- ClassStatus enum
+    cmp     r1, #CLASS_INITIALIZED      @ has class been initialized?
+    bne     .LOP_NEW_INSTANCE_needinit        @ no, init class now
+.LOP_NEW_INSTANCE_initialized: @ r0=class
+    ldr     r3, [r0, #offClassObject_accessFlags]   @ r3<- clazz->accessFlags
+    tst     r3, #(ACC_INTERFACE|ACC_ABSTRACT)   @ abstract or interface?
+    mov     r1, #ALLOC_DONT_TRACK       @ flags for alloc call
+    beq     .LOP_NEW_INSTANCE_finish          @ concrete class, continue
+    b       .LOP_NEW_INSTANCE_abstract        @ fail
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEW_ARRAY: /* 0x23 */
+/* File: armv5te/OP_NEW_ARRAY.S */
+    /*
+     * Allocate an array of objects, specified with the array class
+     * and a count.
+     *
+     * The verifier guarantees that this is an array class, so we don't
+     * check for it here.
+     */
+    /* new-array vA, vB, class@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    FETCH(r2, 1)                        @ r2<- CCCC
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    GET_VREG(r1, r0)                    @ r1<- vB (array length)
+    ldr     r3, [r3, #offDvmDex_pResClasses]    @ r3<- pDvmDex->pResClasses
+    cmp     r1, #0                      @ check length
+    ldr     r0, [r3, r2, lsl #2]        @ r0<- resolved class
+    bmi     common_errNegativeArraySize @ negative length, bail
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ req'd for resolve, alloc
+    bne     .LOP_NEW_ARRAY_finish          @ resolved, continue
+    b       .LOP_NEW_ARRAY_resolve         @ do resolve now
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FILLED_NEW_ARRAY: /* 0x24 */
+/* File: armv5te/OP_FILLED_NEW_ARRAY.S */
+    /*
+     * Create a new array with elements filled from registers.
+     *
+     * for: filled-new-array, filled-new-array/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, type@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResClasses]    @ r3<- pDvmDex->pResClasses
+    EXPORT_PC()                         @ need for resolve and alloc
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved class
+    mov     r10, rINST, lsr #8          @ r10<- AA or BA
+    cmp     r0, #0                      @ already resolved?
+    bne     .LOP_FILLED_NEW_ARRAY_continue        @ yes, continue on
+8:  ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    mov     r2, #0                      @ r2<- false
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- call(clazz, ref)
+    cmp     r0, #0                      @ got null?
+    beq     common_exceptionThrown      @ yes, handle exception
+    b       .LOP_FILLED_NEW_ARRAY_continue
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FILLED_NEW_ARRAY_RANGE: /* 0x25 */
+/* File: armv5te/OP_FILLED_NEW_ARRAY_RANGE.S */
+/* File: armv5te/OP_FILLED_NEW_ARRAY.S */
+    /*
+     * Create a new array with elements filled from registers.
+     *
+     * for: filled-new-array, filled-new-array/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, type@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResClasses]    @ r3<- pDvmDex->pResClasses
+    EXPORT_PC()                         @ need for resolve and alloc
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved class
+    mov     r10, rINST, lsr #8          @ r10<- AA or BA
+    cmp     r0, #0                      @ already resolved?
+    bne     .LOP_FILLED_NEW_ARRAY_RANGE_continue        @ yes, continue on
+8:  ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    mov     r2, #0                      @ r2<- false
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- call(clazz, ref)
+    cmp     r0, #0                      @ got null?
+    beq     common_exceptionThrown      @ yes, handle exception
+    b       .LOP_FILLED_NEW_ARRAY_RANGE_continue
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FILL_ARRAY_DATA: /* 0x26 */
+/* File: armv5te/OP_FILL_ARRAY_DATA.S */
+    /* fill-array-data vAA, +BBBBBBBB */
+    FETCH(r0, 1)                        @ r0<- bbbb (lo)
+    FETCH(r1, 2)                        @ r1<- BBBB (hi)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    orr     r1, r0, r1, lsl #16         @ r1<- BBBBbbbb
+    GET_VREG(r0, r3)                    @ r0<- vAA (array object)
+    add     r1, rPC, r1, lsl #1         @ r1<- PC + BBBBbbbb*2 (array data off.)
+    EXPORT_PC();
+    bl      dvmInterpHandleFillArrayData@ fill the array with predefined data
+    cmp     r0, #0                      @ 0 means an exception is thrown
+    beq     common_exceptionThrown      @ has exception
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_THROW: /* 0x27 */
+/* File: armv5te/OP_THROW.S */
+    /*
+     * Throw an exception object in the current thread.
+     */
+    /* throw vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    GET_VREG(r1, r2)                    @ r1<- vAA (exception object)
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- glue->self
+    cmp     r1, #0                      @ null object?
+    beq     common_errNullObject        @ yes, throw an NPE instead
+    @ bypass dvmSetException, just store it
+    str     r1, [r0, #offThread_exception]  @ thread->exception<- obj
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_GOTO: /* 0x28 */
+/* File: armv5te/OP_GOTO.S */
+    /*
+     * Unconditional branch, 8-bit offset.
+     *
+     * The branch distance is a signed code-unit offset, which we need to
+     * double to get a byte offset.
+     */
+    /* goto +AA */
+    mov     r0, rINST, lsl #16          @ r0<- AAxx0000
+    movs    r9, r0, asr #24             @ r9<- ssssssAA (sign-extended)
+    mov     r9, r9, lsl #1              @ r9<- byte offset
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+    FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_GOTO_16: /* 0x29 */
+/* File: armv5te/OP_GOTO_16.S */
+    /*
+     * Unconditional branch, 16-bit offset.
+     *
+     * The branch distance is a signed code-unit offset, which we need to
+     * double to get a byte offset.
+     */
+    /* goto/16 +AAAA */
+    FETCH_S(r0, 1)                      @ r0<- ssssAAAA (sign-extended)
+    movs    r9, r0, asl #1              @ r9<- byte offset, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+    FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_GOTO_32: /* 0x2a */
+/* File: armv5te/OP_GOTO_32.S */
+    /*
+     * Unconditional branch, 32-bit offset.
+     *
+     * The branch distance is a signed code-unit offset, which we need to
+     * double to get a byte offset.
+     *
+     * Unlike most opcodes, this one is allowed to branch to itself, so
+     * our "backward branch" test must be "<=0" instead of "<0".  The ORRS
+     * instruction doesn't affect the V flag, so we need to clear it
+     * explicitly.
+     */
+    /* goto/32 +AAAAAAAA */
+    FETCH(r0, 1)                        @ r0<- aaaa (lo)
+    FETCH(r1, 2)                        @ r1<- AAAA (hi)
+    cmp     ip, ip                      @ (clear V flag during stall)
+    orrs    r0, r0, r1, lsl #16         @ r0<- AAAAaaaa, check sign
+    mov     r9, r0, asl #1              @ r9<- byte offset
+    ble     common_backwardBranch       @ backward branch, do periodic checks
+    FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_PACKED_SWITCH: /* 0x2b */
+/* File: armv5te/OP_PACKED_SWITCH.S */
+    /*
+     * Handle a packed-switch or sparse-switch instruction.  In both cases
+     * we decode it and hand it off to a helper function.
+     *
+     * We don't really expect backward branches in a switch statement, but
+     * they're perfectly legal, so we check for them here.
+     *
+     * for: packed-switch, sparse-switch
+     */
+    /* op vAA, +BBBB */
+    FETCH(r0, 1)                        @ r0<- bbbb (lo)
+    FETCH(r1, 2)                        @ r1<- BBBB (hi)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    orr     r0, r0, r1, lsl #16         @ r0<- BBBBbbbb
+    GET_VREG(r1, r3)                    @ r1<- vAA
+    add     r0, rPC, r0, lsl #1         @ r0<- PC + BBBBbbbb*2
+    bl      dvmInterpHandlePackedSwitch                       @ r0<- code-unit branch offset
+    movs    r9, r0, asl #1              @ r9<- branch byte offset, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+    beq     common_backwardBranch       @ (want to use BLE but V is unknown)
+    FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPARSE_SWITCH: /* 0x2c */
+/* File: armv5te/OP_SPARSE_SWITCH.S */
+/* File: armv5te/OP_PACKED_SWITCH.S */
+    /*
+     * Handle a packed-switch or sparse-switch instruction.  In both cases
+     * we decode it and hand it off to a helper function.
+     *
+     * We don't really expect backward branches in a switch statement, but
+     * they're perfectly legal, so we check for them here.
+     *
+     * for: packed-switch, sparse-switch
+     */
+    /* op vAA, +BBBB */
+    FETCH(r0, 1)                        @ r0<- bbbb (lo)
+    FETCH(r1, 2)                        @ r1<- BBBB (hi)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    orr     r0, r0, r1, lsl #16         @ r0<- BBBBbbbb
+    GET_VREG(r1, r3)                    @ r1<- vAA
+    add     r0, rPC, r0, lsl #1         @ r0<- PC + BBBBbbbb*2
+    bl      dvmInterpHandleSparseSwitch                       @ r0<- code-unit branch offset
+    movs    r9, r0, asl #1              @ r9<- branch byte offset, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+    beq     common_backwardBranch       @ (want to use BLE but V is unknown)
+    FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMPL_FLOAT: /* 0x2d */
+/* File: armv5te/OP_CMPL_FLOAT.S */
+    /*
+     * Compare two floating-point values.  Puts 0, 1, or -1 into the
+     * destination register based on the results of the comparison.
+     *
+     * Provide a "naninst" instruction that puts 1 or -1 into r1 depending
+     * on what value we'd like to return when one of the operands is NaN.
+     *
+     * The operation we're implementing is:
+     *   if (x == y)
+     *     return 0;
+     *   else if (x < y)
+     *     return -1;
+     *   else if (x > y)
+     *     return 1;
+     *   else
+     *     return {-1,1};  // one or both operands was NaN
+     *
+     * The straightforward implementation requires 3 calls to functions
+     * that return a result in r0.  We can do it with two calls if our
+     * EABI library supports __aeabi_cfcmple (only one if we want to check
+     * for NaN directly):
+     *   check x <= y
+     *     if <, return -1
+     *     if ==, return 0
+     *   check y <= x
+     *     if <, return 1
+     *   return {-1,1}
+     *
+     * for: cmpl-float, cmpg-float
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    GET_VREG(r9, r2)                    @ r9<- vBB
+    GET_VREG(r10, r3)                   @ r10<- vCC
+    mov     r0, r9                      @ copy to arg registers
+    mov     r1, r10
+    bl      __aeabi_cfcmple             @ cmp <=: C clear if <, Z set if eq
+    bhi     .LOP_CMPL_FLOAT_gt_or_nan       @ C set and Z clear, disambiguate
+    mvncc   r1, #0                      @ (less than) r1<- -1
+    moveq   r1, #0                      @ (equal) r1<- 0, trumps less than
+.LOP_CMPL_FLOAT_finish:
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r3)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMPG_FLOAT: /* 0x2e */
+/* File: armv5te/OP_CMPG_FLOAT.S */
+/* File: armv5te/OP_CMPL_FLOAT.S */
+    /*
+     * Compare two floating-point values.  Puts 0, 1, or -1 into the
+     * destination register based on the results of the comparison.
+     *
+     * Provide a "naninst" instruction that puts 1 or -1 into r1 depending
+     * on what value we'd like to return when one of the operands is NaN.
+     *
+     * The operation we're implementing is:
+     *   if (x == y)
+     *     return 0;
+     *   else if (x < y)
+     *     return -1;
+     *   else if (x > y)
+     *     return 1;
+     *   else
+     *     return {-1,1};  // one or both operands was NaN
+     *
+     * The straightforward implementation requires 3 calls to functions
+     * that return a result in r0.  We can do it with two calls if our
+     * EABI library supports __aeabi_cfcmple (only one if we want to check
+     * for NaN directly):
+     *   check x <= y
+     *     if <, return -1
+     *     if ==, return 0
+     *   check y <= x
+     *     if <, return 1
+     *   return {-1,1}
+     *
+     * for: cmpl-float, cmpg-float
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    GET_VREG(r9, r2)                    @ r9<- vBB
+    GET_VREG(r10, r3)                   @ r10<- vCC
+    mov     r0, r9                      @ copy to arg registers
+    mov     r1, r10
+    bl      __aeabi_cfcmple             @ cmp <=: C clear if <, Z set if eq
+    bhi     .LOP_CMPG_FLOAT_gt_or_nan       @ C set and Z clear, disambiguate
+    mvncc   r1, #0                      @ (less than) r1<- -1
+    moveq   r1, #0                      @ (equal) r1<- 0, trumps less than
+.LOP_CMPG_FLOAT_finish:
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r3)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMPL_DOUBLE: /* 0x2f */
+/* File: armv5te/OP_CMPL_DOUBLE.S */
+    /*
+     * Compare two floating-point values.  Puts 0, 1, or -1 into the
+     * destination register based on the results of the comparison.
+     *
+     * Provide a "naninst" instruction that puts 1 or -1 into r1 depending
+     * on what value we'd like to return when one of the operands is NaN.
+     *
+     * See OP_CMPL_FLOAT for an explanation.
+     *
+     * For: cmpl-double, cmpg-double
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r9, r0, #255                @ r9<- BB
+    mov     r10, r0, lsr #8             @ r10<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[BB]
+    add     r10, rFP, r10, lsl #2       @ r10<- &fp[CC]
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r10, {r2-r3}                @ r2/r3<- vCC/vCC+1
+    bl      __aeabi_cdcmple             @ cmp <=: C clear if <, Z set if eq
+    bhi     .LOP_CMPL_DOUBLE_gt_or_nan       @ C set and Z clear, disambiguate
+    mvncc   r1, #0                      @ (less than) r1<- -1
+    moveq   r1, #0                      @ (equal) r1<- 0, trumps less than
+.LOP_CMPL_DOUBLE_finish:
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r3)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMPG_DOUBLE: /* 0x30 */
+/* File: armv5te/OP_CMPG_DOUBLE.S */
+/* File: armv5te/OP_CMPL_DOUBLE.S */
+    /*
+     * Compare two floating-point values.  Puts 0, 1, or -1 into the
+     * destination register based on the results of the comparison.
+     *
+     * Provide a "naninst" instruction that puts 1 or -1 into r1 depending
+     * on what value we'd like to return when one of the operands is NaN.
+     *
+     * See OP_CMPL_FLOAT for an explanation.
+     *
+     * For: cmpl-double, cmpg-double
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r9, r0, #255                @ r9<- BB
+    mov     r10, r0, lsr #8             @ r10<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[BB]
+    add     r10, rFP, r10, lsl #2       @ r10<- &fp[CC]
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r10, {r2-r3}                @ r2/r3<- vCC/vCC+1
+    bl      __aeabi_cdcmple             @ cmp <=: C clear if <, Z set if eq
+    bhi     .LOP_CMPG_DOUBLE_gt_or_nan       @ C set and Z clear, disambiguate
+    mvncc   r1, #0                      @ (less than) r1<- -1
+    moveq   r1, #0                      @ (equal) r1<- 0, trumps less than
+.LOP_CMPG_DOUBLE_finish:
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r3)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMP_LONG: /* 0x31 */
+/* File: armv5te/OP_CMP_LONG.S */
+    /*
+     * Compare two 64-bit values.  Puts 0, 1, or -1 into the destination
+     * register based on the results of the comparison.
+     *
+     * We load the full values with LDM, but in practice many values could
+     * be resolved by only looking at the high word.  This could be made
+     * faster or slower by splitting the LDM into a pair of LDRs.
+     *
+     * If we just wanted to set condition flags, we could do this:
+     *  subs    ip, r0, r2
+     *  sbcs    ip, r1, r3
+     *  subeqs  ip, r0, r2
+     * Leaving { <0, 0, >0 } in ip.  However, we have to set it to a specific
+     * integer value, which we can do with 2 conditional mov/mvn instructions
+     * (set 1, set -1; if they're equal we already have 0 in ip), giving
+     * us a constant 5-cycle path plus a branch at the end to the
+     * instruction epilogue code.  The multi-compare approach below needs
+     * 2 or 3 cycles + branch if the high word doesn't match, 6 + branch
+     * in the worst case (the 64-bit values are equal).
+     */
+    /* cmp-long vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    cmp     r1, r3                      @ compare (vBB+1, vCC+1)
+    blt     .LOP_CMP_LONG_less            @ signed compare on high part
+    bgt     .LOP_CMP_LONG_greater
+    subs    r1, r0, r2                  @ r1<- r0 - r2
+    bhi     .LOP_CMP_LONG_greater         @ unsigned compare on low part
+    bne     .LOP_CMP_LONG_less
+    b       .LOP_CMP_LONG_finish          @ equal; r1 already holds 0
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_EQ: /* 0x32 */
+/* File: armv5te/OP_IF_EQ.S */
+/* File: armv5te/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r3, r1)                    @ r3<- vB
+    GET_VREG(r2, r0)                    @ r2<- vA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, r3                      @ compare (vA, vB)
+    bne  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ yes, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_NE: /* 0x33 */
+/* File: armv5te/OP_IF_NE.S */
+/* File: armv5te/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r3, r1)                    @ r3<- vB
+    GET_VREG(r2, r0)                    @ r2<- vA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, r3                      @ compare (vA, vB)
+    beq  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ yes, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_LT: /* 0x34 */
+/* File: armv5te/OP_IF_LT.S */
+/* File: armv5te/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r3, r1)                    @ r3<- vB
+    GET_VREG(r2, r0)                    @ r2<- vA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, r3                      @ compare (vA, vB)
+    bge  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ yes, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_GE: /* 0x35 */
+/* File: armv5te/OP_IF_GE.S */
+/* File: armv5te/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r3, r1)                    @ r3<- vB
+    GET_VREG(r2, r0)                    @ r2<- vA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, r3                      @ compare (vA, vB)
+    blt  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ yes, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_GT: /* 0x36 */
+/* File: armv5te/OP_IF_GT.S */
+/* File: armv5te/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r3, r1)                    @ r3<- vB
+    GET_VREG(r2, r0)                    @ r2<- vA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, r3                      @ compare (vA, vB)
+    ble  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ yes, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_LE: /* 0x37 */
+/* File: armv5te/OP_IF_LE.S */
+/* File: armv5te/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r3, r1)                    @ r3<- vB
+    GET_VREG(r2, r0)                    @ r2<- vA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, r3                      @ compare (vA, vB)
+    bgt  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ yes, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_EQZ: /* 0x38 */
+/* File: armv5te/OP_IF_EQZ.S */
+/* File: armv5te/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    bne  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_NEZ: /* 0x39 */
+/* File: armv5te/OP_IF_NEZ.S */
+/* File: armv5te/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    beq  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_LTZ: /* 0x3a */
+/* File: armv5te/OP_IF_LTZ.S */
+/* File: armv5te/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    bge  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_GEZ: /* 0x3b */
+/* File: armv5te/OP_IF_GEZ.S */
+/* File: armv5te/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    blt  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_GTZ: /* 0x3c */
+/* File: armv5te/OP_IF_GTZ.S */
+/* File: armv5te/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    ble  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_LEZ: /* 0x3d */
+/* File: armv5te/OP_IF_LEZ.S */
+/* File: armv5te/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    bgt  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_3E: /* 0x3e */
+/* File: armv5te/OP_UNUSED_3E.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_3F: /* 0x3f */
+/* File: armv5te/OP_UNUSED_3F.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_40: /* 0x40 */
+/* File: armv5te/OP_UNUSED_40.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_41: /* 0x41 */
+/* File: armv5te/OP_UNUSED_41.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_42: /* 0x42 */
+/* File: armv5te/OP_UNUSED_42.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_43: /* 0x43 */
+/* File: armv5te/OP_UNUSED_43.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET: /* 0x44 */
+/* File: armv5te/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #2     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldr   r2, [r0, #offArrayObject_contents]  @ r2<- vBB[vCC]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r9)                    @ vAA<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_WIDE: /* 0x45 */
+/* File: armv4/OP_AGET_WIDE.S */
+    /*
+     * Array get, 64 bits.  vAA <- vBB[vCC].
+     *
+     * Arrays of long/double are 64-bit aligned, so it's okay to use LDRD.
+     */
+    /* aget-wide vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #3          @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcc     .LOP_AGET_WIDE_finish          @ okay, continue below
+    b       common_errArrayIndex        @ index >= length, bail
+    @ May want to swap the order of these two branches depending on how the
+    @ branch prediction (if any) handles conditional forward branches vs.
+    @ unconditional forward branches.
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_OBJECT: /* 0x46 */
+/* File: armv5te/OP_AGET_OBJECT.S */
+/* File: armv5te/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #2     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldr   r2, [r0, #offArrayObject_contents]  @ r2<- vBB[vCC]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r9)                    @ vAA<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_BOOLEAN: /* 0x47 */
+/* File: armv5te/OP_AGET_BOOLEAN.S */
+/* File: armv5te/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #0     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldrb   r2, [r0, #offArrayObject_contents]  @ r2<- vBB[vCC]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r9)                    @ vAA<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_BYTE: /* 0x48 */
+/* File: armv5te/OP_AGET_BYTE.S */
+/* File: armv5te/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #0     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldrsb   r2, [r0, #offArrayObject_contents]  @ r2<- vBB[vCC]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r9)                    @ vAA<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_CHAR: /* 0x49 */
+/* File: armv5te/OP_AGET_CHAR.S */
+/* File: armv5te/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #1     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldrh   r2, [r0, #offArrayObject_contents]  @ r2<- vBB[vCC]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r9)                    @ vAA<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_SHORT: /* 0x4a */
+/* File: armv5te/OP_AGET_SHORT.S */
+/* File: armv5te/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #1     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldrsh   r2, [r0, #offArrayObject_contents]  @ r2<- vBB[vCC]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r9)                    @ vAA<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT: /* 0x4b */
+/* File: armv5te/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA.
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aput, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #2     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r9)                    @ r2<- vAA
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r2, [r0, #offArrayObject_contents]  @ vBB[vCC]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_WIDE: /* 0x4c */
+/* File: armv4/OP_APUT_WIDE.S */
+    /*
+     * Array put, 64 bits.  vBB[vCC] <- vAA.
+     */
+    /* aput-wide vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #3          @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    bcc     .LOP_APUT_WIDE_finish          @ okay, continue below
+    b       common_errArrayIndex        @ index >= length, bail
+    @ May want to swap the order of these two branches depending on how the
+    @ branch prediction (if any) handles conditional forward branches vs.
+    @ unconditional forward branches.
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_OBJECT: /* 0x4d */
+/* File: armv5te/OP_APUT_OBJECT.S */
+    /*
+     * Store an object into an array.  vBB[vCC] <- vAA.
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    GET_VREG(r1, r2)                    @ r1<- vBB (array object)
+    GET_VREG(r0, r3)                    @ r0<- vCC (requested index)
+    cmp     r1, #0                      @ null array object?
+    GET_VREG(r9, r9)                    @ r9<- vAA
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r1, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r10, r1, r0, lsl #2         @ r10<- arrayObj + index*width
+    cmp     r0, r3                      @ compare unsigned index, length
+    bcc     .LOP_APUT_OBJECT_finish          @ we're okay, continue on
+    b       common_errArrayIndex        @ index >= length, bail
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_BOOLEAN: /* 0x4e */
+/* File: armv5te/OP_APUT_BOOLEAN.S */
+/* File: armv5te/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA.
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aput, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #0     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r9)                    @ r2<- vAA
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strb  r2, [r0, #offArrayObject_contents]  @ vBB[vCC]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_BYTE: /* 0x4f */
+/* File: armv5te/OP_APUT_BYTE.S */
+/* File: armv5te/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA.
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aput, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #0     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r9)                    @ r2<- vAA
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strb  r2, [r0, #offArrayObject_contents]  @ vBB[vCC]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_CHAR: /* 0x50 */
+/* File: armv5te/OP_APUT_CHAR.S */
+/* File: armv5te/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA.
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aput, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #1     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r9)                    @ r2<- vAA
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strh  r2, [r0, #offArrayObject_contents]  @ vBB[vCC]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_SHORT: /* 0x51 */
+/* File: armv5te/OP_APUT_SHORT.S */
+/* File: armv5te/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA.
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aput, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #1     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r9)                    @ r2<- vAA
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strh  r2, [r0, #offArrayObject_contents]  @ vBB[vCC]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET: /* 0x52 */
+/* File: armv5te/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_finish
+    b       common_exceptionThrown
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_WIDE: /* 0x53 */
+/* File: armv4/OP_IGET_WIDE.S */
+    /*
+     * Wide 32-bit instance field get.
+     */
+    /* iget-wide vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_WIDE_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method] @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_WIDE_finish
+    b       common_exceptionThrown
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_OBJECT: /* 0x54 */
+/* File: armv5te/OP_IGET_OBJECT.S */
+/* File: armv5te/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_OBJECT_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_OBJECT_finish
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_BOOLEAN: /* 0x55 */
+/* File: armv5te/OP_IGET_BOOLEAN.S */
+@include "armv5te/OP_IGET.S" { "load":"ldrb", "sqnum":"1" }
+/* File: armv5te/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_BOOLEAN_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_BOOLEAN_finish
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_BYTE: /* 0x56 */
+/* File: armv5te/OP_IGET_BYTE.S */
+@include "armv5te/OP_IGET.S" { "load":"ldrsb", "sqnum":"2" }
+/* File: armv5te/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_BYTE_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_BYTE_finish
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_CHAR: /* 0x57 */
+/* File: armv5te/OP_IGET_CHAR.S */
+@include "armv5te/OP_IGET.S" { "load":"ldrh", "sqnum":"3" }
+/* File: armv5te/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_CHAR_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_CHAR_finish
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_SHORT: /* 0x58 */
+/* File: armv5te/OP_IGET_SHORT.S */
+@include "armv5te/OP_IGET.S" { "load":"ldrsh", "sqnum":"4" }
+/* File: armv5te/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_SHORT_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_SHORT_finish
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT: /* 0x59 */
+/* File: armv5te/OP_IPUT.S */
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_WIDE: /* 0x5a */
+/* File: armv4/OP_IPUT_WIDE.S */
+    /* iput-wide vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_WIDE_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method] @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_WIDE_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_OBJECT: /* 0x5b */
+/* File: armv5te/OP_IPUT_OBJECT.S */
+/* File: armv5te/OP_IPUT.S */
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_OBJECT_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_OBJECT_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_BOOLEAN: /* 0x5c */
+/* File: armv5te/OP_IPUT_BOOLEAN.S */
+@include "armv5te/OP_IPUT.S" { "store":"strb", "sqnum":"1" }
+/* File: armv5te/OP_IPUT.S */
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_BOOLEAN_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_BOOLEAN_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_BYTE: /* 0x5d */
+/* File: armv5te/OP_IPUT_BYTE.S */
+@include "armv5te/OP_IPUT.S" { "store":"strb", "sqnum":"2" }
+/* File: armv5te/OP_IPUT.S */
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_BYTE_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_BYTE_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_CHAR: /* 0x5e */
+/* File: armv5te/OP_IPUT_CHAR.S */
+@include "armv5te/OP_IPUT.S" { "store":"strh", "sqnum":"3" }
+/* File: armv5te/OP_IPUT.S */
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_CHAR_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_CHAR_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_SHORT: /* 0x5f */
+/* File: armv5te/OP_IPUT_SHORT.S */
+@include "armv5te/OP_IPUT.S" { "store":"strh", "sqnum":"4" }
+/* File: armv5te/OP_IPUT.S */
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_SHORT_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_SHORT_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET: /* 0x60 */
+/* File: armv5te/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_resolve         @ yes, do resolve
+.LOP_SGET_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_WIDE: /* 0x61 */
+/* File: armv4/OP_SGET_WIDE.S */
+    /*
+     * 64-bit SGET handler.
+     */
+    /* sget-wide vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_WIDE_resolve         @ yes, do resolve
+.LOP_SGET_WIDE_finish:
+    mov     r1, rINST, lsr #8           @ r1<- AA
+    add     r0, r0, #offStaticField_value
+    ldmia   r0, {r2-r3}                 @ r2/r3<- field value (aligned)
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[AA]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    stmia   r1, {r2-r3}                 @ vAA/vAA+1<- r2/r3
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_OBJECT: /* 0x62 */
+/* File: armv5te/OP_SGET_OBJECT.S */
+/* File: armv5te/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_OBJECT_resolve         @ yes, do resolve
+.LOP_SGET_OBJECT_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_BOOLEAN: /* 0x63 */
+/* File: armv5te/OP_SGET_BOOLEAN.S */
+/* File: armv5te/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_BOOLEAN_resolve         @ yes, do resolve
+.LOP_SGET_BOOLEAN_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_BYTE: /* 0x64 */
+/* File: armv5te/OP_SGET_BYTE.S */
+/* File: armv5te/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_BYTE_resolve         @ yes, do resolve
+.LOP_SGET_BYTE_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_CHAR: /* 0x65 */
+/* File: armv5te/OP_SGET_CHAR.S */
+/* File: armv5te/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_CHAR_resolve         @ yes, do resolve
+.LOP_SGET_CHAR_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_SHORT: /* 0x66 */
+/* File: armv5te/OP_SGET_SHORT.S */
+/* File: armv5te/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_SHORT_resolve         @ yes, do resolve
+.LOP_SGET_SHORT_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT: /* 0x67 */
+/* File: armv5te/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_resolve         @ yes, do resolve
+.LOP_SPUT_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_WIDE: /* 0x68 */
+/* File: armv4/OP_SPUT_WIDE.S */
+    /*
+     * 64-bit SPUT handler.
+     */
+    /* sput-wide vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_WIDE_resolve         @ yes, do resolve
+.LOP_SPUT_WIDE_finish: @ field ptr in r0, AA in r9
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldmia   r9, {r2-r3}                 @ r2/r3<- vAA/vAA+1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    add     r0, r0, #offStaticField_value
+    stmia   r0, {r2-r3}                 @ field<- vAA/vAA+1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_OBJECT: /* 0x69 */
+/* File: armv5te/OP_SPUT_OBJECT.S */
+/* File: armv5te/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_OBJECT_resolve         @ yes, do resolve
+.LOP_SPUT_OBJECT_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_BOOLEAN: /* 0x6a */
+/* File: armv5te/OP_SPUT_BOOLEAN.S */
+/* File: armv5te/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_BOOLEAN_resolve         @ yes, do resolve
+.LOP_SPUT_BOOLEAN_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_BYTE: /* 0x6b */
+/* File: armv5te/OP_SPUT_BYTE.S */
+/* File: armv5te/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_BYTE_resolve         @ yes, do resolve
+.LOP_SPUT_BYTE_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_CHAR: /* 0x6c */
+/* File: armv5te/OP_SPUT_CHAR.S */
+/* File: armv5te/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_CHAR_resolve         @ yes, do resolve
+.LOP_SPUT_CHAR_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_SHORT: /* 0x6d */
+/* File: armv5te/OP_SPUT_SHORT.S */
+/* File: armv5te/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_SHORT_resolve         @ yes, do resolve
+.LOP_SPUT_SHORT_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_VIRTUAL: /* 0x6e */
+/* File: armv5te/OP_INVOKE_VIRTUAL.S */
+    /*
+     * Handle a virtual method call.
+     *
+     * for: invoke-virtual, invoke-virtual/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved baseMethod
+    .if     (!0)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    bne     .LOP_INVOKE_VIRTUAL_continue        @ yes, continue on
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    mov     r2, #METHOD_VIRTUAL         @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    bne     .LOP_INVOKE_VIRTUAL_continue        @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_SUPER: /* 0x6f */
+/* File: armv5te/OP_INVOKE_SUPER.S */
+    /*
+     * Handle a "super" method call.
+     *
+     * for: invoke-super, invoke-super/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    .if     (!0)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    GET_VREG(r2, r10)                   @ r2<- "this" ptr
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved baseMethod
+    cmp     r2, #0                      @ null "this"?
+    ldr     r9, [rGLUE, #offGlue_method] @ r9<- current method
+    beq     common_errNullObject        @ null "this", throw exception
+    cmp     r0, #0                      @ already resolved?
+    ldr     r9, [r9, #offMethod_clazz]  @ r9<- method->clazz
+    EXPORT_PC()                         @ must export for invoke
+    bne     .LOP_INVOKE_SUPER_continue        @ resolved, continue on
+    b       .LOP_INVOKE_SUPER_resolve         @ do resolve now
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_DIRECT: /* 0x70 */
+/* File: armv5te/OP_INVOKE_DIRECT.S */
+    /*
+     * Handle a direct method call.
+     *
+     * (We could defer the "is 'this' pointer null" test to the common
+     * method invocation code, and use a flag to indicate that static
+     * calls don't count.  If we do this as part of copying the arguments
+     * out we could avoiding loading the first arg twice.)
+     *
+     * for: invoke-direct, invoke-direct/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved methodToCall
+    .if     (!0)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    GET_VREG(r2, r10)                   @ r2<- "this" ptr
+    beq     .LOP_INVOKE_DIRECT_resolve         @ not resolved, do it now
+.LOP_INVOKE_DIRECT_finish:
+    cmp     r2, #0                      @ null "this" ref?
+    bne     common_invokeMethodNoRange   @ no, continue on
+    b       common_errNullObject        @ yes, throw exception
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_STATIC: /* 0x71 */
+/* File: armv5te/OP_INVOKE_STATIC.S */
+    /*
+     * Handle a static method call.
+     *
+     * for: invoke-static, invoke-static/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved methodToCall
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    bne     common_invokeMethodNoRange @ yes, continue on
+0:  ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    mov     r2, #METHOD_STATIC          @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    bne     common_invokeMethodNoRange @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_INTERFACE: /* 0x72 */
+/* File: armv5te/OP_INVOKE_INTERFACE.S */
+    /*
+     * Handle an interface method call.
+     *
+     * for: invoke-interface, invoke-interface/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r2, 2)                        @ r2<- FEDC or CCCC
+    FETCH(r1, 1)                        @ r1<- BBBB
+    .if     (!0)
+    and     r2, r2, #15                 @ r2<- C (or stays CCCC)
+    .endif
+    EXPORT_PC()                         @ must export for invoke
+    GET_VREG(r0, r2)                    @ r0<- first arg ("this")
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- methodClassDex
+    cmp     r0, #0                      @ null obj?
+    ldr     r2, [rGLUE, #offGlue_method]  @ r2<- method
+    beq     common_errNullObject        @ yes, fail
+    ldr     r0, [r0, #offObject_clazz]  @ r0<- thisPtr->clazz
+    bl      dvmFindInterfaceMethodInCache @ r0<- call(class, ref, method, dex)
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yes, handle exception
+    b       common_invokeMethodNoRange @ jump to common handler 
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_73: /* 0x73 */
+/* File: armv5te/OP_UNUSED_73.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_VIRTUAL_RANGE: /* 0x74 */
+/* File: armv5te/OP_INVOKE_VIRTUAL_RANGE.S */
+/* File: armv5te/OP_INVOKE_VIRTUAL.S */
+    /*
+     * Handle a virtual method call.
+     *
+     * for: invoke-virtual, invoke-virtual/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved baseMethod
+    .if     (!1)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    bne     .LOP_INVOKE_VIRTUAL_RANGE_continue        @ yes, continue on
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    mov     r2, #METHOD_VIRTUAL         @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    bne     .LOP_INVOKE_VIRTUAL_RANGE_continue        @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_SUPER_RANGE: /* 0x75 */
+/* File: armv5te/OP_INVOKE_SUPER_RANGE.S */
+/* File: armv5te/OP_INVOKE_SUPER.S */
+    /*
+     * Handle a "super" method call.
+     *
+     * for: invoke-super, invoke-super/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    .if     (!1)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    GET_VREG(r2, r10)                   @ r2<- "this" ptr
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved baseMethod
+    cmp     r2, #0                      @ null "this"?
+    ldr     r9, [rGLUE, #offGlue_method] @ r9<- current method
+    beq     common_errNullObject        @ null "this", throw exception
+    cmp     r0, #0                      @ already resolved?
+    ldr     r9, [r9, #offMethod_clazz]  @ r9<- method->clazz
+    EXPORT_PC()                         @ must export for invoke
+    bne     .LOP_INVOKE_SUPER_RANGE_continue        @ resolved, continue on
+    b       .LOP_INVOKE_SUPER_RANGE_resolve         @ do resolve now
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_DIRECT_RANGE: /* 0x76 */
+/* File: armv5te/OP_INVOKE_DIRECT_RANGE.S */
+/* File: armv5te/OP_INVOKE_DIRECT.S */
+    /*
+     * Handle a direct method call.
+     *
+     * (We could defer the "is 'this' pointer null" test to the common
+     * method invocation code, and use a flag to indicate that static
+     * calls don't count.  If we do this as part of copying the arguments
+     * out we could avoiding loading the first arg twice.)
+     *
+     * for: invoke-direct, invoke-direct/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved methodToCall
+    .if     (!1)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    GET_VREG(r2, r10)                   @ r2<- "this" ptr
+    beq     .LOP_INVOKE_DIRECT_RANGE_resolve         @ not resolved, do it now
+.LOP_INVOKE_DIRECT_RANGE_finish:
+    cmp     r2, #0                      @ null "this" ref?
+    bne     common_invokeMethodRange   @ no, continue on
+    b       common_errNullObject        @ yes, throw exception
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_STATIC_RANGE: /* 0x77 */
+/* File: armv5te/OP_INVOKE_STATIC_RANGE.S */
+/* File: armv5te/OP_INVOKE_STATIC.S */
+    /*
+     * Handle a static method call.
+     *
+     * for: invoke-static, invoke-static/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved methodToCall
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    bne     common_invokeMethodRange @ yes, continue on
+0:  ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    mov     r2, #METHOD_STATIC          @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    bne     common_invokeMethodRange @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_INTERFACE_RANGE: /* 0x78 */
+/* File: armv5te/OP_INVOKE_INTERFACE_RANGE.S */
+/* File: armv5te/OP_INVOKE_INTERFACE.S */
+    /*
+     * Handle an interface method call.
+     *
+     * for: invoke-interface, invoke-interface/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r2, 2)                        @ r2<- FEDC or CCCC
+    FETCH(r1, 1)                        @ r1<- BBBB
+    .if     (!1)
+    and     r2, r2, #15                 @ r2<- C (or stays CCCC)
+    .endif
+    EXPORT_PC()                         @ must export for invoke
+    GET_VREG(r0, r2)                    @ r0<- first arg ("this")
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- methodClassDex
+    cmp     r0, #0                      @ null obj?
+    ldr     r2, [rGLUE, #offGlue_method]  @ r2<- method
+    beq     common_errNullObject        @ yes, fail
+    ldr     r0, [r0, #offObject_clazz]  @ r0<- thisPtr->clazz
+    bl      dvmFindInterfaceMethodInCache @ r0<- call(class, ref, method, dex)
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yes, handle exception
+    b       common_invokeMethodRange @ jump to common handler 
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_79: /* 0x79 */
+/* File: armv5te/OP_UNUSED_79.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_7A: /* 0x7a */
+/* File: armv5te/OP_UNUSED_7A.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEG_INT: /* 0x7b */
+/* File: armv5te/OP_NEG_INT.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    rsb     r0, r0, #0                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NOT_INT: /* 0x7c */
+/* File: armv5te/OP_NOT_INT.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mvn     r0, r0                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEG_LONG: /* 0x7d */
+/* File: armv5te/OP_NEG_LONG.S */
+/* File: armv5te/unopWide.S */
+    /*
+     * Generic 64-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0/r1".
+     * This could be an ARM instruction or a function call.
+     *
+     * For: neg-long, not-long, neg-double, long-to-double, double-to-long
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vAA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    rsbs    r0, r0, #0                           @ optional op; may set condition codes
+    rsc     r1, r1, #0                              @ r0/r1<- op, r2-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NOT_LONG: /* 0x7e */
+/* File: armv5te/OP_NOT_LONG.S */
+/* File: armv5te/unopWide.S */
+    /*
+     * Generic 64-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0/r1".
+     * This could be an ARM instruction or a function call.
+     *
+     * For: neg-long, not-long, neg-double, long-to-double, double-to-long
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vAA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mvn     r0, r0                           @ optional op; may set condition codes
+    mvn     r1, r1                              @ r0/r1<- op, r2-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEG_FLOAT: /* 0x7f */
+/* File: armv5te/OP_NEG_FLOAT.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    add     r0, r0, #0x80000000                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEG_DOUBLE: /* 0x80 */
+/* File: armv5te/OP_NEG_DOUBLE.S */
+/* File: armv5te/unopWide.S */
+    /*
+     * Generic 64-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0/r1".
+     * This could be an ARM instruction or a function call.
+     *
+     * For: neg-long, not-long, neg-double, long-to-double, double-to-long
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vAA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    add     r1, r1, #0x80000000                              @ r0/r1<- op, r2-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_LONG: /* 0x81 */
+/* File: armv5te/OP_INT_TO_LONG.S */
+/* File: armv5te/unopWider.S */
+    /*
+     * Generic 32bit-to-64bit unary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = op r0", where
+     * "result" is a 64-bit quantity in r0/r1.
+     *
+     * For: int-to-long, int-to-double, float-to-long, float-to-double
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r3)                    @ r0<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mov     r1, r0, asr #31                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vA/vA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-11 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_FLOAT: /* 0x82 */
+/* File: armv5te/OP_INT_TO_FLOAT.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    bl      __aeabi_i2f                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_DOUBLE: /* 0x83 */
+/* File: armv5te/OP_INT_TO_DOUBLE.S */
+/* File: armv5te/unopWider.S */
+    /*
+     * Generic 32bit-to-64bit unary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = op r0", where
+     * "result" is a 64-bit quantity in r0/r1.
+     *
+     * For: int-to-long, int-to-double, float-to-long, float-to-double
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r3)                    @ r0<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    bl      __aeabi_i2d                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vA/vA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-11 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_LONG_TO_INT: /* 0x84 */
+/* File: armv5te/OP_LONG_TO_INT.S */
+/* we ignore the high word, making this equivalent to a 32-bit reg move */
+/* File: armv5te/OP_MOVE.S */
+    /* for move, move-object, long-to-int */
+    /* op vA, vB */
+    mov     r1, rINST, lsr #12          @ r1<- B from 15:12
+    mov     r0, rINST, lsr #8           @ r0<- A from 11:8
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[B]
+    and     r0, r0, #15
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[A]<- r2
+    GOTO_OPCODE(ip)                     @ execute next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_LONG_TO_FLOAT: /* 0x85 */
+/* File: armv5te/OP_LONG_TO_FLOAT.S */
+/* File: armv5te/unopNarrower.S */
+    /*
+     * Generic 64bit-to-32bit unary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = op r0/r1", where
+     * "result" is a 32-bit quantity in r0.
+     *
+     * For: long-to-float, double-to-int, double-to-float
+     *
+     * (This would work for long-to-int, but that instruction is actually
+     * an exact match for OP_MOVE.)
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    and     r9, r9, #15
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vB/vB+1
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_l2f                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-11 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_LONG_TO_DOUBLE: /* 0x86 */
+/* File: armv5te/OP_LONG_TO_DOUBLE.S */
+/* File: armv5te/unopWide.S */
+    /*
+     * Generic 64-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0/r1".
+     * This could be an ARM instruction or a function call.
+     *
+     * For: neg-long, not-long, neg-double, long-to-double, double-to-long
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vAA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_l2d                              @ r0/r1<- op, r2-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FLOAT_TO_INT: /* 0x87 */
+/* File: armv5te/OP_FLOAT_TO_INT.S */
+/* EABI appears to have Java-style conversions of +inf/-inf/NaN */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    bl      __aeabi_f2iz                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+#if 0
+@include "armv5te/unop.S" {"instr":"bl      f2i_doconv"}
+@break
+/*
+ * Convert the float in r0 to an int in r0.
+ *
+ * We have to clip values to int min/max per the specification.  The
+ * expected common case is a "reasonable" value that converts directly
+ * to modest integer.  The EABI convert function isn't doing this for us.
+ */
+f2i_doconv:
+    stmfd   sp!, {r4, lr}
+    mov     r1, #0x4f000000             @ (float)maxint
+    mov     r4, r0
+    bl      __aeabi_fcmpge              @ is arg >= maxint?
+    cmp     r0, #0                      @ nonzero == yes
+    mvnne   r0, #0x80000000             @ return maxint (7fffffff)
+    ldmnefd sp!, {r4, pc}
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, #0xcf000000             @ (float)minint
+    bl      __aeabi_fcmple              @ is arg <= minint?
+    cmp     r0, #0                      @ nonzero == yes
+    movne   r0, #0x80000000             @ return minint (80000000)
+    ldmnefd sp!, {r4, pc}
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r4
+    bl      __aeabi_fcmpeq              @ is arg == self?
+    cmp     r0, #0                      @ zero == no
+    ldmeqfd sp!, {r4, pc}               @ return zero for NaN
+
+    mov     r0, r4                      @ recover arg
+    bl      __aeabi_f2iz                @ convert float to int
+    ldmfd   sp!, {r4, pc}
+#endif
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FLOAT_TO_LONG: /* 0x88 */
+/* File: armv5te/OP_FLOAT_TO_LONG.S */
+@include "armv5te/unopWider.S" {"instr":"bl      __aeabi_f2lz"}
+/* File: armv5te/unopWider.S */
+    /*
+     * Generic 32bit-to-64bit unary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = op r0", where
+     * "result" is a 64-bit quantity in r0/r1.
+     *
+     * For: int-to-long, int-to-double, float-to-long, float-to-double
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r3)                    @ r0<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    bl      f2l_doconv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vA/vA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-11 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FLOAT_TO_DOUBLE: /* 0x89 */
+/* File: armv5te/OP_FLOAT_TO_DOUBLE.S */
+/* File: armv5te/unopWider.S */
+    /*
+     * Generic 32bit-to-64bit unary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = op r0", where
+     * "result" is a 64-bit quantity in r0/r1.
+     *
+     * For: int-to-long, int-to-double, float-to-long, float-to-double
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r3)                    @ r0<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    bl      __aeabi_f2d                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vA/vA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-11 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DOUBLE_TO_INT: /* 0x8a */
+/* File: armv5te/OP_DOUBLE_TO_INT.S */
+/* EABI appears to have Java-style conversions of +inf/-inf/NaN */
+/* File: armv5te/unopNarrower.S */
+    /*
+     * Generic 64bit-to-32bit unary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = op r0/r1", where
+     * "result" is a 32-bit quantity in r0.
+     *
+     * For: long-to-float, double-to-int, double-to-float
+     *
+     * (This would work for long-to-int, but that instruction is actually
+     * an exact match for OP_MOVE.)
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    and     r9, r9, #15
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vB/vB+1
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_d2iz                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-11 instructions */
+
+
+#if 0
+@include "armv5te/unopNarrower.S" {"instr":"bl      d2i_doconv"}
+@break
+/*
+ * Convert the double in r0/r1 to an int in r0.
+ *
+ * We have to clip values to int min/max per the specification.  The
+ * expected common case is a "reasonable" value that converts directly
+ * to modest integer.  The EABI convert function isn't doing this for us.
+ */
+d2i_doconv:
+    stmfd   sp!, {r4, r5, lr}           @ save regs
+    ldr     r2, .LOP_DOUBLE_TO_INT_maxlo       @ (double)maxint, lo
+    ldr     r3, .LOP_DOUBLE_TO_INT_maxhi       @ (double)maxint, hi
+    sub     sp, sp, #4                  @ align for EABI
+    mov     r4, r0                      @ save r0
+    mov     r5, r1                      @  and r1
+    bl      __aeabi_dcmpge              @ is arg >= maxint?
+    cmp     r0, #0                      @ nonzero == yes
+    mvnne   r0, #0x80000000             @ return maxint (7fffffff)
+    bne     1f
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r5
+    ldr     r3, .LOP_DOUBLE_TO_INT_min         @ (double)minint, hi
+    mov     r2, #0                      @ (double)minint, lo
+    bl      __aeabi_dcmple              @ is arg <= minint?
+    cmp     r0, #0                      @ nonzero == yes
+    movne   r0, #0x80000000             @ return minint (80000000)
+    bne     1f
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r5
+    mov     r2, r4                      @ compare against self
+    mov     r3, r5
+    bl      __aeabi_dcmpeq              @ is arg == self?
+    cmp     r0, #0                      @ zero == no
+    beq     1f                          @ return zero for NaN
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r5
+    bl      __aeabi_d2iz                @ convert double to int
+
+1:
+    add     sp, sp, #4
+    ldmfd   sp!, {r4, r5, pc}
+
+.LOP_DOUBLE_TO_INT_maxlo:
+    .word   0xffc00000                  @ maxint, as a double (low word)
+.LOP_DOUBLE_TO_INT_maxhi:
+    .word   0x41dfffff                  @ maxint, as a double (high word)
+.LOP_DOUBLE_TO_INT_min:
+    .word   0xc1e00000                  @ minint, as a double (high word)
+#endif
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DOUBLE_TO_LONG: /* 0x8b */
+/* File: armv5te/OP_DOUBLE_TO_LONG.S */
+@include "armv5te/unopWide.S" {"instr":"bl      __aeabi_d2lz"}
+/* File: armv5te/unopWide.S */
+    /*
+     * Generic 64-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0/r1".
+     * This could be an ARM instruction or a function call.
+     *
+     * For: neg-long, not-long, neg-double, long-to-double, double-to-long
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vAA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      d2l_doconv                              @ r0/r1<- op, r2-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-13 instructions */
+
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DOUBLE_TO_FLOAT: /* 0x8c */
+/* File: armv5te/OP_DOUBLE_TO_FLOAT.S */
+/* File: armv5te/unopNarrower.S */
+    /*
+     * Generic 64bit-to-32bit unary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = op r0/r1", where
+     * "result" is a 32-bit quantity in r0.
+     *
+     * For: long-to-float, double-to-int, double-to-float
+     *
+     * (This would work for long-to-int, but that instruction is actually
+     * an exact match for OP_MOVE.)
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    and     r9, r9, #15
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vB/vB+1
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_d2f                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-11 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_BYTE: /* 0x8d */
+/* File: armv5te/OP_INT_TO_BYTE.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+    mov     r0, r0, asl #24                           @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mov     r0, r0, asr #24                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_CHAR: /* 0x8e */
+/* File: armv5te/OP_INT_TO_CHAR.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+    mov     r0, r0, asl #16                           @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mov     r0, r0, lsr #16                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_SHORT: /* 0x8f */
+/* File: armv5te/OP_INT_TO_SHORT.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+    mov     r0, r0, asl #16                           @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mov     r0, r0, asr #16                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_INT: /* 0x90 */
+/* File: armv5te/OP_ADD_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    add     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_INT: /* 0x91 */
+/* File: armv5te/OP_SUB_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    sub     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_INT: /* 0x92 */
+/* File: armv5te/OP_MUL_INT.S */
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    mul     r0, r1, r0                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_INT: /* 0x93 */
+/* File: armv5te/OP_DIV_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 1
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl     __aeabi_idiv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_INT: /* 0x94 */
+/* File: armv5te/OP_REM_INT.S */
+/* idivmod returns quotient in r0 and remainder in r1 */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 1
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_idivmod                              @ r1<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r1, r9)               @ vAA<- r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_INT: /* 0x95 */
+/* File: armv5te/OP_AND_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    and     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_INT: /* 0x96 */
+/* File: armv5te/OP_OR_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    orr     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_INT: /* 0x97 */
+/* File: armv5te/OP_XOR_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    eor     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_INT: /* 0x98 */
+/* File: armv5te/OP_SHL_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, asl r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_INT: /* 0x99 */
+/* File: armv5te/OP_SHR_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, asr r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_INT: /* 0x9a */
+/* File: armv5te/OP_USHR_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, lsr r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_LONG: /* 0x9b */
+/* File: armv5te/OP_ADD_LONG.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    adds    r0, r0, r2                           @ optional op; may set condition codes
+    adc     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_LONG: /* 0x9c */
+/* File: armv5te/OP_SUB_LONG.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    subs    r0, r0, r2                           @ optional op; may set condition codes
+    sbc     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_LONG: /* 0x9d */
+/* File: armv5te/OP_MUL_LONG.S */
+    /*
+     * Signed 64-bit integer multiply.
+     *
+     * Consider WXxYZ (r1r0 x r3r2) with a long multiply:
+     *        WX
+     *      x YZ
+     *  --------
+     *     ZW ZX
+     *  YW YX
+     *
+     * The low word of the result holds ZX, the high word holds
+     * (ZW+YX) + (the high overflow from ZX).  YW doesn't matter because
+     * it doesn't fit in the low 64 bits.
+     *
+     * Unlike most ARM math operations, multiply instructions have
+     * restrictions on using the same register more than once (Rd and Rm
+     * cannot be the same).
+     */
+    /* mul-long vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    mul     ip, r2, r1                  @  ip<- ZxW
+    umull   r9, r10, r2, r0             @  r9/r10 <- ZxX
+    mla     r2, r0, r3, ip              @  r2<- YxX + (ZxW)
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    add     r10, r2, r10                @  r10<- r10 + low(ZxW + (YxX))
+    add     r0, rFP, r0, lsl #2         @ r0<- &fp[AA]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    b       .LOP_MUL_LONG_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_LONG: /* 0x9e */
+/* File: armv5te/OP_DIV_LONG.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 1
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_ldivmod                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_LONG: /* 0x9f */
+/* File: armv5te/OP_REM_LONG.S */
+/* ldivmod returns quotient in r0/r1 and remainder in r2/r3 */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 1
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_ldivmod                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r2,r3}     @ vAA/vAA+1<- r2/r3
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_LONG: /* 0xa0 */
+/* File: armv5te/OP_AND_LONG.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    and     r0, r0, r2                           @ optional op; may set condition codes
+    and     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_LONG: /* 0xa1 */
+/* File: armv5te/OP_OR_LONG.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    orr     r0, r0, r2                           @ optional op; may set condition codes
+    orr     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_LONG: /* 0xa2 */
+/* File: armv5te/OP_XOR_LONG.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    eor     r0, r0, r2                           @ optional op; may set condition codes
+    eor     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_LONG: /* 0xa3 */
+/* File: armv5te/OP_SHL_LONG.S */
+    /*
+     * Long integer shift.  This is different from the generic 32/64-bit
+     * binary operations because vAA/vBB are 64-bit but vCC (the shift
+     * distance) is 32-bit.  Also, Dalvik requires us to mask off the low
+     * 6 bits of the shift distance.
+     */
+    /* shl-long vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r3, r0, #255                @ r3<- BB
+    mov     r0, r0, lsr #8              @ r0<- CC
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[BB]
+    GET_VREG(r2, r0)                    @ r2<- vCC
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    and     r2, r2, #63                 @ r2<- r2 & 0x3f
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+
+    mov     r1, r1, asl r2              @  r1<- r1 << r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r1, r1, r0, lsr r3          @  r1<- r1 | (r0 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    movpl   r1, r0, asl ip              @  if r2 >= 32, r1<- r0 << (r2-32)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    b       .LOP_SHL_LONG_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_LONG: /* 0xa4 */
+/* File: armv5te/OP_SHR_LONG.S */
+    /*
+     * Long integer shift.  This is different from the generic 32/64-bit
+     * binary operations because vAA/vBB are 64-bit but vCC (the shift
+     * distance) is 32-bit.  Also, Dalvik requires us to mask off the low
+     * 6 bits of the shift distance.
+     */
+    /* shr-long vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r3, r0, #255                @ r3<- BB
+    mov     r0, r0, lsr #8              @ r0<- CC
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[BB]
+    GET_VREG(r2, r0)                    @ r2<- vCC
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    and     r2, r2, #63                 @ r0<- r0 & 0x3f
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+
+    mov     r0, r0, lsr r2              @  r0<- r2 >> r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r0, r0, r1, asl r3          @  r0<- r0 | (r1 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    movpl   r0, r1, asr ip              @  if r2 >= 32, r0<-r1 >> (r2-32)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    b       .LOP_SHR_LONG_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_LONG: /* 0xa5 */
+/* File: armv5te/OP_USHR_LONG.S */
+    /*
+     * Long integer shift.  This is different from the generic 32/64-bit
+     * binary operations because vAA/vBB are 64-bit but vCC (the shift
+     * distance) is 32-bit.  Also, Dalvik requires us to mask off the low
+     * 6 bits of the shift distance.
+     */
+    /* ushr-long vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r3, r0, #255                @ r3<- BB
+    mov     r0, r0, lsr #8              @ r0<- CC
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[BB]
+    GET_VREG(r2, r0)                    @ r2<- vCC
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    and     r2, r2, #63                 @ r0<- r0 & 0x3f
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+
+    mov     r0, r0, lsr r2              @  r0<- r2 >> r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r0, r0, r1, asl r3          @  r0<- r0 | (r1 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    movpl   r0, r1, lsr ip              @  if r2 >= 32, r0<-r1 >>> (r2-32)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    b       .LOP_USHR_LONG_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_FLOAT: /* 0xa6 */
+/* File: armv5te/OP_ADD_FLOAT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_fadd                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_FLOAT: /* 0xa7 */
+/* File: armv5te/OP_SUB_FLOAT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_fsub                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_FLOAT: /* 0xa8 */
+/* File: armv5te/OP_MUL_FLOAT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_fmul                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_FLOAT: /* 0xa9 */
+/* File: armv5te/OP_DIV_FLOAT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_fdiv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_FLOAT: /* 0xaa */
+/* File: armv5te/OP_REM_FLOAT.S */
+/* EABI doesn't define a float remainder function, but libm does */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      fmodf                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_DOUBLE: /* 0xab */
+/* File: armv5te/OP_ADD_DOUBLE.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_dadd                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_DOUBLE: /* 0xac */
+/* File: armv5te/OP_SUB_DOUBLE.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_dsub                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_DOUBLE: /* 0xad */
+/* File: armv5te/OP_MUL_DOUBLE.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_dmul                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_DOUBLE: /* 0xae */
+/* File: armv5te/OP_DIV_DOUBLE.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_ddiv                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_DOUBLE: /* 0xaf */
+/* File: armv5te/OP_REM_DOUBLE.S */
+/* EABI doesn't define a double remainder function, but libm does */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      fmod                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_INT_2ADDR: /* 0xb0 */
+/* File: armv5te/OP_ADD_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    add     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_INT_2ADDR: /* 0xb1 */
+/* File: armv5te/OP_SUB_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    sub     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_INT_2ADDR: /* 0xb2 */
+/* File: armv5te/OP_MUL_INT_2ADDR.S */
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    mul     r0, r1, r0                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_INT_2ADDR: /* 0xb3 */
+/* File: armv5te/OP_DIV_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 1
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl     __aeabi_idiv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_INT_2ADDR: /* 0xb4 */
+/* File: armv5te/OP_REM_INT_2ADDR.S */
+/* idivmod returns quotient in r0 and remainder in r1 */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 1
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_idivmod                              @ r1<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r1, r9)               @ vAA<- r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_INT_2ADDR: /* 0xb5 */
+/* File: armv5te/OP_AND_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    and     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_INT_2ADDR: /* 0xb6 */
+/* File: armv5te/OP_OR_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    orr     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_INT_2ADDR: /* 0xb7 */
+/* File: armv5te/OP_XOR_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    eor     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_INT_2ADDR: /* 0xb8 */
+/* File: armv5te/OP_SHL_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, asl r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_INT_2ADDR: /* 0xb9 */
+/* File: armv5te/OP_SHR_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, asr r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_INT_2ADDR: /* 0xba */
+/* File: armv5te/OP_USHR_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, lsr r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_LONG_2ADDR: /* 0xbb */
+/* File: armv5te/OP_ADD_LONG_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    adds    r0, r0, r2                           @ optional op; may set condition codes
+    adc     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_LONG_2ADDR: /* 0xbc */
+/* File: armv5te/OP_SUB_LONG_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    subs    r0, r0, r2                           @ optional op; may set condition codes
+    sbc     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_LONG_2ADDR: /* 0xbd */
+/* File: armv5te/OP_MUL_LONG_2ADDR.S */
+    /*
+     * Signed 64-bit integer multiply, "/2addr" version.
+     *
+     * See OP_MUL_LONG for an explanation.
+     *
+     * We get a little tight on registers, so to avoid looking up &fp[A]
+     * again we stuff it into rINST.
+     */
+    /* mul-long/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     rINST, rFP, r9, lsl #2      @ rINST<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   rINST, {r0-r1}              @ r0/r1<- vAA/vAA+1
+    mul     ip, r2, r1                  @  ip<- ZxW
+    umull   r9, r10, r2, r0             @  r9/r10 <- ZxX
+    mla     r2, r0, r3, ip              @  r2<- YxX + (ZxW)
+    mov     r0, rINST                   @ r0<- &fp[A] (free up rINST)
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    add     r10, r2, r10                @  r10<- r10 + low(ZxW + (YxX))
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r0, {r9-r10}                @ vAA/vAA+1<- r9/r10
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_LONG_2ADDR: /* 0xbe */
+/* File: armv5te/OP_DIV_LONG_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 1
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_ldivmod                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_LONG_2ADDR: /* 0xbf */
+/* File: armv5te/OP_REM_LONG_2ADDR.S */
+/* ldivmod returns quotient in r0/r1 and remainder in r2/r3 */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 1
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_ldivmod                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r2,r3}     @ vAA/vAA+1<- r2/r3
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_LONG_2ADDR: /* 0xc0 */
+/* File: armv5te/OP_AND_LONG_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    and     r0, r0, r2                           @ optional op; may set condition codes
+    and     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_LONG_2ADDR: /* 0xc1 */
+/* File: armv5te/OP_OR_LONG_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    orr     r0, r0, r2                           @ optional op; may set condition codes
+    orr     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_LONG_2ADDR: /* 0xc2 */
+/* File: armv5te/OP_XOR_LONG_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    eor     r0, r0, r2                           @ optional op; may set condition codes
+    eor     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_LONG_2ADDR: /* 0xc3 */
+/* File: armv5te/OP_SHL_LONG_2ADDR.S */
+    /*
+     * Long integer shift, 2addr version.  vA is 64-bit value/result, vB is
+     * 32-bit shift distance.
+     */
+    /* shl-long/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r2, r3)                    @ r2<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    and     r2, r2, #63                 @ r2<- r2 & 0x3f
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+
+    mov     r1, r1, asl r2              @  r1<- r1 << r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r1, r1, r0, lsr r3          @  r1<- r1 | (r0 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    movpl   r1, r0, asl ip              @  if r2 >= 32, r1<- r0 << (r2-32)
+    mov     r0, r0, asl r2              @  r0<- r0 << r2
+    b       .LOP_SHL_LONG_2ADDR_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_LONG_2ADDR: /* 0xc4 */
+/* File: armv5te/OP_SHR_LONG_2ADDR.S */
+    /*
+     * Long integer shift, 2addr version.  vA is 64-bit value/result, vB is
+     * 32-bit shift distance.
+     */
+    /* shr-long/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r2, r3)                    @ r2<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    and     r2, r2, #63                 @ r2<- r2 & 0x3f
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+
+    mov     r0, r0, lsr r2              @  r0<- r2 >> r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r0, r0, r1, asl r3          @  r0<- r0 | (r1 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    movpl   r0, r1, asr ip              @  if r2 >= 32, r0<-r1 >> (r2-32)
+    mov     r1, r1, asr r2              @  r1<- r1 >> r2
+    b       .LOP_SHR_LONG_2ADDR_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_LONG_2ADDR: /* 0xc5 */
+/* File: armv5te/OP_USHR_LONG_2ADDR.S */
+    /*
+     * Long integer shift, 2addr version.  vA is 64-bit value/result, vB is
+     * 32-bit shift distance.
+     */
+    /* ushr-long/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r2, r3)                    @ r2<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    and     r2, r2, #63                 @ r2<- r2 & 0x3f
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+
+    mov     r0, r0, lsr r2              @  r0<- r2 >> r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r0, r0, r1, asl r3          @  r0<- r0 | (r1 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    movpl   r0, r1, lsr ip              @  if r2 >= 32, r0<-r1 >>> (r2-32)
+    mov     r1, r1, lsr r2              @  r1<- r1 >>> r2
+    b       .LOP_USHR_LONG_2ADDR_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_FLOAT_2ADDR: /* 0xc6 */
+/* File: armv5te/OP_ADD_FLOAT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_fadd                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_FLOAT_2ADDR: /* 0xc7 */
+/* File: armv5te/OP_SUB_FLOAT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_fsub                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_FLOAT_2ADDR: /* 0xc8 */
+/* File: armv5te/OP_MUL_FLOAT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_fmul                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_FLOAT_2ADDR: /* 0xc9 */
+/* File: armv5te/OP_DIV_FLOAT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_fdiv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_FLOAT_2ADDR: /* 0xca */
+/* File: armv5te/OP_REM_FLOAT_2ADDR.S */
+/* EABI doesn't define a float remainder function, but libm does */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      fmodf                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_DOUBLE_2ADDR: /* 0xcb */
+/* File: armv5te/OP_ADD_DOUBLE_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_dadd                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_DOUBLE_2ADDR: /* 0xcc */
+/* File: armv5te/OP_SUB_DOUBLE_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_dsub                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_DOUBLE_2ADDR: /* 0xcd */
+/* File: armv5te/OP_MUL_DOUBLE_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_dmul                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_DOUBLE_2ADDR: /* 0xce */
+/* File: armv5te/OP_DIV_DOUBLE_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_ddiv                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_DOUBLE_2ADDR: /* 0xcf */
+/* File: armv5te/OP_REM_DOUBLE_2ADDR.S */
+/* EABI doesn't define a double remainder function, but libm does */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      fmod                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_INT_LIT16: /* 0xd0 */
+/* File: armv5te/OP_ADD_INT_LIT16.S */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    add     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RSUB_INT: /* 0xd1 */
+/* File: armv5te/OP_RSUB_INT.S */
+/* this op is "rsub-int", but can be thought of as "rsub-int/lit16" */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    rsb     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_INT_LIT16: /* 0xd2 */
+/* File: armv5te/OP_MUL_INT_LIT16.S */
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    mul     r0, r1, r0                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_INT_LIT16: /* 0xd3 */
+/* File: armv5te/OP_DIV_INT_LIT16.S */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 1
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    bl     __aeabi_idiv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_INT_LIT16: /* 0xd4 */
+/* File: armv5te/OP_REM_INT_LIT16.S */
+/* idivmod returns quotient in r0 and remainder in r1 */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 1
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    bl      __aeabi_idivmod                              @ r1<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r1, r9)               @ vAA<- r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_INT_LIT16: /* 0xd5 */
+/* File: armv5te/OP_AND_INT_LIT16.S */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    and     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_INT_LIT16: /* 0xd6 */
+/* File: armv5te/OP_OR_INT_LIT16.S */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    orr     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_INT_LIT16: /* 0xd7 */
+/* File: armv5te/OP_XOR_INT_LIT16.S */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    eor     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_INT_LIT8: /* 0xd8 */
+/* File: armv5te/OP_ADD_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    add     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RSUB_INT_LIT8: /* 0xd9 */
+/* File: armv5te/OP_RSUB_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    rsb     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_INT_LIT8: /* 0xda */
+/* File: armv5te/OP_MUL_INT_LIT8.S */
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    mul     r0, r1, r0                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_INT_LIT8: /* 0xdb */
+/* File: armv5te/OP_DIV_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 1
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl     __aeabi_idiv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_INT_LIT8: /* 0xdc */
+/* File: armv5te/OP_REM_INT_LIT8.S */
+/* idivmod returns quotient in r0 and remainder in r1 */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 1
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_idivmod                              @ r1<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r1, r9)               @ vAA<- r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_INT_LIT8: /* 0xdd */
+/* File: armv5te/OP_AND_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    and     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_INT_LIT8: /* 0xde */
+/* File: armv5te/OP_OR_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    orr     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_INT_LIT8: /* 0xdf */
+/* File: armv5te/OP_XOR_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    eor     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_INT_LIT8: /* 0xe0 */
+/* File: armv5te/OP_SHL_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, asl r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_INT_LIT8: /* 0xe1 */
+/* File: armv5te/OP_SHR_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, asr r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_INT_LIT8: /* 0xe2 */
+/* File: armv5te/OP_USHR_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, lsr r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E3: /* 0xe3 */
+/* File: armv5te/OP_UNUSED_E3.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E4: /* 0xe4 */
+/* File: armv5te/OP_UNUSED_E4.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E5: /* 0xe5 */
+/* File: armv5te/OP_UNUSED_E5.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E6: /* 0xe6 */
+/* File: armv5te/OP_UNUSED_E6.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E7: /* 0xe7 */
+/* File: armv5te/OP_UNUSED_E7.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E8: /* 0xe8 */
+/* File: armv5te/OP_UNUSED_E8.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E9: /* 0xe9 */
+/* File: armv5te/OP_UNUSED_E9.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EA: /* 0xea */
+/* File: armv5te/OP_UNUSED_EA.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EB: /* 0xeb */
+/* File: armv5te/OP_UNUSED_EB.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EC: /* 0xec */
+/* File: armv5te/OP_UNUSED_EC.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_ED: /* 0xed */
+/* File: armv5te/OP_UNUSED_ED.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_EXECUTE_INLINE: /* 0xee */
+/* File: armv5te/OP_EXECUTE_INLINE.S */
+    /*
+     * Execute a "native inline" instruction.
+     *
+     * We need to call:
+     *  dvmPerformInlineOp4Std(arg0, arg1, arg2, arg3, &retval, ref)
+     *
+     * The first four args are in r0-r3, but the last two must be pushed
+     * onto the stack.
+     */
+    /* [opt] execute-inline vAA, {vC, vD, vE, vF}, inline@BBBB */
+    FETCH(r10, 1)                       @ r10<- BBBB
+    add     r1, rGLUE, #offGlue_retval  @ r1<- &glue->retval
+    EXPORT_PC()                         @ can throw
+    sub     sp, sp, #8                  @ make room for arg(s)
+    mov     r0, rINST, lsr #12          @ r0<- B
+    str     r1, [sp]                    @ push &glue->retval
+    bl      .LOP_EXECUTE_INLINE_continue        @ make call; will return after
+    add     sp, sp, #8                  @ pop stack
+    cmp     r0, #0                      @ test boolean result of inline
+    beq     common_exceptionThrown      @ returned false, handle exception
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EF: /* 0xef */
+/* File: armv5te/OP_UNUSED_EF.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_DIRECT_EMPTY: /* 0xf0 */
+/* File: armv5te/OP_INVOKE_DIRECT_EMPTY.S */
+    /*
+     * invoke-direct-empty is a no-op in a "standard" interpreter.
+     */
+    FETCH_ADVANCE_INST(3)               @ advance to next instr, load rINST
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    GOTO_OPCODE(ip)                     @ execute it
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_F1: /* 0xf1 */
+/* File: armv5te/OP_UNUSED_F1.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_QUICK: /* 0xf2 */
+/* File: armv5te/OP_IGET_QUICK.S */
+    /* For: iget-quick, iget-object-quick */
+    /* op vA, vB, offset@CCCC */
+    mov     r2, rINST, lsr #12          @ r2<- B
+    GET_VREG(r3, r2)                    @ r3<- object we're operating on
+    FETCH(r1, 1)                        @ r1<- field byte offset
+    cmp     r3, #0                      @ check object for null
+    mov     r2, rINST, lsr #8           @ r2<- A(+)
+    beq     common_errNullObject        @ object was null
+    ldr     r0, [r3, r1]                @ r0<- obj.field (always 32 bits)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_WIDE_QUICK: /* 0xf3 */
+/* File: armv4/OP_IGET_WIDE_QUICK.S */
+    /* iget-wide-quick vA, vB, offset@CCCC */
+    mov     r2, rINST, lsr #12          @ r2<- B
+    GET_VREG(r3, r2)                    @ r3<- object we're operating on
+    FETCH(r1, 1)                        @ r1<- field byte offset
+    cmp     r3, #0                      @ check object for null
+    mov     r2, rINST, lsr #8           @ r2<- A(+)
+    beq     common_errNullObject        @ object was null
+    add     r9, r3, r1                  @ r9<- object + offset
+    ldmia   r9, {r0-r1}                 @ r0/r1<- obj.field (64 bits, aligned)
+    and     r2, r2, #15                 @ r2<- A
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    add     r3, rFP, r2, lsl #2         @ r3<- &fp[A]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r3, {r0-r1}                 @ fp[A]<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_OBJECT_QUICK: /* 0xf4 */
+/* File: armv5te/OP_IGET_OBJECT_QUICK.S */
+/* File: armv5te/OP_IGET_QUICK.S */
+    /* For: iget-quick, iget-object-quick */
+    /* op vA, vB, offset@CCCC */
+    mov     r2, rINST, lsr #12          @ r2<- B
+    GET_VREG(r3, r2)                    @ r3<- object we're operating on
+    FETCH(r1, 1)                        @ r1<- field byte offset
+    cmp     r3, #0                      @ check object for null
+    mov     r2, rINST, lsr #8           @ r2<- A(+)
+    beq     common_errNullObject        @ object was null
+    ldr     r0, [r3, r1]                @ r0<- obj.field (always 32 bits)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_QUICK: /* 0xf5 */
+/* File: armv5te/OP_IPUT_QUICK.S */
+    /* For: iput-quick, iput-object-quick */
+    /* op vA, vB, offset@CCCC */
+    mov     r2, rINST, lsr #12          @ r2<- B
+    GET_VREG(r3, r2)                    @ r3<- fp[B], the object pointer
+    FETCH(r1, 1)                        @ r1<- field byte offset
+    cmp     r3, #0                      @ check object for null
+    mov     r2, rINST, lsr #8           @ r2<- A(+)
+    beq     common_errNullObject        @ object was null
+    and     r2, r2, #15
+    GET_VREG(r0, r2)                    @ r0<- fp[A]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    str     r0, [r3, r1]                @ obj.field (always 32 bits)<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_WIDE_QUICK: /* 0xf6 */
+/* File: armv4/OP_IPUT_WIDE_QUICK.S */
+    /* iput-wide-quick vA, vB, offset@CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A(+)
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r2, r1)                    @ r2<- fp[B], the object pointer
+    add     r3, rFP, r0, lsl #2         @ r3<- &fp[A]
+    cmp     r2, #0                      @ check object for null
+    ldmia   r3, {r0-r1}                 @ r0/r1<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH(r3, 1)                        @ r3<- field byte offset
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    add     r2, r2, r3                  @ r2<- object + byte offset
+    stmia   r2, {r0-r1}                 @ obj.field (64 bits, aligned)<- r0/r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_OBJECT_QUICK: /* 0xf7 */
+/* File: armv5te/OP_IPUT_OBJECT_QUICK.S */
+/* File: armv5te/OP_IPUT_QUICK.S */
+    /* For: iput-quick, iput-object-quick */
+    /* op vA, vB, offset@CCCC */
+    mov     r2, rINST, lsr #12          @ r2<- B
+    GET_VREG(r3, r2)                    @ r3<- fp[B], the object pointer
+    FETCH(r1, 1)                        @ r1<- field byte offset
+    cmp     r3, #0                      @ check object for null
+    mov     r2, rINST, lsr #8           @ r2<- A(+)
+    beq     common_errNullObject        @ object was null
+    and     r2, r2, #15
+    GET_VREG(r0, r2)                    @ r0<- fp[A]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    str     r0, [r3, r1]                @ obj.field (always 32 bits)<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_VIRTUAL_QUICK: /* 0xf8 */
+/* File: armv5te/OP_INVOKE_VIRTUAL_QUICK.S */
+    /*
+     * Handle an optimized virtual method call.
+     *
+     * for: [opt] invoke-virtual-quick, invoke-virtual-quick/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r3, 2)                        @ r3<- FEDC or CCCC
+    FETCH(r1, 1)                        @ r1<- BBBB
+    .if     (!0)
+    and     r3, r3, #15                 @ r3<- C (or stays CCCC)
+    .endif
+    GET_VREG(r2, r3)                    @ r2<- vC ("this" ptr)
+    cmp     r2, #0                      @ is "this" null?
+    beq     common_errNullObject        @ null "this", throw exception
+    ldr     r2, [r2, #offObject_clazz]  @ r2<- thisPtr->clazz
+    ldr     r2, [r2, #offClassObject_vtable]    @ r2<- thisPtr->clazz->vtable
+    EXPORT_PC()                         @ invoke must export
+    ldr     r0, [r2, r1, lsl #2]        @ r3<- vtable[BBBB]
+    bl      common_invokeMethodNoRange @ continue on
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_VIRTUAL_QUICK_RANGE: /* 0xf9 */
+/* File: armv5te/OP_INVOKE_VIRTUAL_QUICK_RANGE.S */
+/* File: armv5te/OP_INVOKE_VIRTUAL_QUICK.S */
+    /*
+     * Handle an optimized virtual method call.
+     *
+     * for: [opt] invoke-virtual-quick, invoke-virtual-quick/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r3, 2)                        @ r3<- FEDC or CCCC
+    FETCH(r1, 1)                        @ r1<- BBBB
+    .if     (!1)
+    and     r3, r3, #15                 @ r3<- C (or stays CCCC)
+    .endif
+    GET_VREG(r2, r3)                    @ r2<- vC ("this" ptr)
+    cmp     r2, #0                      @ is "this" null?
+    beq     common_errNullObject        @ null "this", throw exception
+    ldr     r2, [r2, #offObject_clazz]  @ r2<- thisPtr->clazz
+    ldr     r2, [r2, #offClassObject_vtable]    @ r2<- thisPtr->clazz->vtable
+    EXPORT_PC()                         @ invoke must export
+    ldr     r0, [r2, r1, lsl #2]        @ r3<- vtable[BBBB]
+    bl      common_invokeMethodRange @ continue on
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_SUPER_QUICK: /* 0xfa */
+/* File: armv5te/OP_INVOKE_SUPER_QUICK.S */
+    /*
+     * Handle an optimized "super" method call.
+     *
+     * for: [opt] invoke-super-quick, invoke-super-quick/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    .if     (!0)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r2, [r2, #offMethod_clazz]  @ r2<- method->clazz
+    EXPORT_PC()                         @ must export for invoke
+    ldr     r2, [r2, #offClassObject_super]     @ r2<- method->clazz->super
+    GET_VREG(r3, r10)                   @ r3<- "this"
+    ldr     r2, [r2, #offClassObject_vtable]    @ r2<- ...clazz->super->vtable
+    cmp     r3, #0                      @ null "this" ref?
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- super->vtable[BBBB]
+    beq     common_errNullObject        @ "this" is null, throw exception
+    bl      common_invokeMethodNoRange @ continue on
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_SUPER_QUICK_RANGE: /* 0xfb */
+/* File: armv5te/OP_INVOKE_SUPER_QUICK_RANGE.S */
+/* File: armv5te/OP_INVOKE_SUPER_QUICK.S */
+    /*
+     * Handle an optimized "super" method call.
+     *
+     * for: [opt] invoke-super-quick, invoke-super-quick/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    .if     (!1)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r2, [r2, #offMethod_clazz]  @ r2<- method->clazz
+    EXPORT_PC()                         @ must export for invoke
+    ldr     r2, [r2, #offClassObject_super]     @ r2<- method->clazz->super
+    GET_VREG(r3, r10)                   @ r3<- "this"
+    ldr     r2, [r2, #offClassObject_vtable]    @ r2<- ...clazz->super->vtable
+    cmp     r3, #0                      @ null "this" ref?
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- super->vtable[BBBB]
+    beq     common_errNullObject        @ "this" is null, throw exception
+    bl      common_invokeMethodRange @ continue on
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FC: /* 0xfc */
+/* File: armv5te/OP_UNUSED_FC.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FD: /* 0xfd */
+/* File: armv5te/OP_UNUSED_FD.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FE: /* 0xfe */
+/* File: armv5te/OP_UNUSED_FE.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FF: /* 0xff */
+/* File: armv5te/OP_UNUSED_FF.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+
+    .balign 64
+    .size   dvmAsmInstructionStart, .-dvmAsmInstructionStart
+    .global dvmAsmInstructionEnd
+dvmAsmInstructionEnd:
+
+/*
+ * ===========================================================================
+ *  Sister implementations
+ * ===========================================================================
+ */
+    .global dvmAsmSisterStart
+    .type   dvmAsmSisterStart, %function
+    .text
+    .balign 4
+dvmAsmSisterStart:
+
+/* continuation for OP_CONST_STRING */
+
+    /*
+     * Continuation if the String has not yet been resolved.
+     *  r1: BBBB (String ref)
+     *  r9: target register
+     */
+.LOP_CONST_STRING_resolve:
+    EXPORT_PC()
+    ldr     r0, [rGLUE, #offGlue_method] @ r0<- glue->method
+    ldr     r0, [r0, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveString            @ r0<- String reference
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yup, handle the exception
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_CONST_STRING_JUMBO */
+
+    /*
+     * Continuation if the String has not yet been resolved.
+     *  r1: BBBBBBBB (String ref)
+     *  r9: target register
+     */
+.LOP_CONST_STRING_JUMBO_resolve:
+    EXPORT_PC()
+    ldr     r0, [rGLUE, #offGlue_method] @ r0<- glue->method
+    ldr     r0, [r0, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveString            @ r0<- String reference
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yup, handle the exception
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_CONST_CLASS */
+
+    /*
+     * Continuation if the Class has not yet been resolved.
+     *  r1: BBBB (Class ref)
+     *  r9: target register
+     */
+.LOP_CONST_CLASS_resolve:
+    EXPORT_PC()
+    ldr     r0, [rGLUE, #offGlue_method] @ r0<- glue->method
+    mov     r2, #1                      @ r2<- true
+    ldr     r0, [r0, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- Class reference
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yup, handle the exception
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_CHECK_CAST */
+
+    /*
+     * Trivial test failed, need to perform full check.  This is common.
+     *  r0 holds obj->clazz
+     *  r1 holds class resolved from BBBB
+     *  r9 holds object
+     */
+.LOP_CHECK_CAST_fullcheck:
+    bl      dvmInstanceofNonTrivial     @ r0<- boolean result
+    cmp     r0, #0                      @ failed?
+    bne     .LOP_CHECK_CAST_okay            @ no, success
+
+    @ A cast has failed.  We need to throw a ClassCastException with the
+    @ class of the object that failed to be cast.
+    EXPORT_PC()                         @ about to throw
+    ldr     r3, [r9, #offObject_clazz]  @ r3<- obj->clazz
+    ldr     r0, .LstrClassCastExceptionPtr
+    ldr     r1, [r3, #offClassObject_descriptor] @ r1<- obj->clazz->descriptor
+    bl      dvmThrowExceptionWithClassMessage
+    b       common_exceptionThrown
+
+    /*
+     * Resolution required.  This is the least-likely path.
+     *
+     *  r2 holds BBBB
+     *  r9 holds object
+     */
+.LOP_CHECK_CAST_resolve:
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    mov     r1, r2                      @ r1<- BBBB
+    mov     r2, #0                      @ r2<- false
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- resolved ClassObject ptr
+    cmp     r0, #0                      @ got null?
+    beq     common_exceptionThrown      @ yes, handle exception
+    mov     r1, r0                      @ r1<- class resolved from BBB
+    ldr     r0, [r9, #offObject_clazz]  @ r0<- obj->clazz
+    b       .LOP_CHECK_CAST_resolved        @ pick up where we left off
+
+.LstrClassCastExceptionPtr:
+    .word   .LstrClassCastException
+
+
+/* continuation for OP_INSTANCE_OF */
+
+    /*
+     * Trivial test failed, need to perform full check.  This is common.
+     *  r0 holds obj->clazz
+     *  r1 holds class resolved from BBBB
+     *  r9 holds A
+     */
+.LOP_INSTANCE_OF_fullcheck:
+    bl      dvmInstanceofNonTrivial     @ r0<- boolean result
+    @ fall through to OP_INSTANCE_OF_store
+
+    /*
+     * r0 holds boolean result
+     * r9 holds A
+     */
+.LOP_INSTANCE_OF_store:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r0, r9)                    @ vA<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+    /*
+     * Trivial test succeeded, save and bail.
+     *  r9 holds A
+     */
+.LOP_INSTANCE_OF_trivial:
+    mov     r0, #1                      @ indicate success
+    @ could b OP_INSTANCE_OF_store, but copying is faster and cheaper
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r0, r9)                    @ vA<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+    /*
+     * Resolution required.  This is the least-likely path.
+     *
+     *  r3 holds BBBB
+     *  r9 holds A
+     */
+.LOP_INSTANCE_OF_resolve:
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [rGLUE, #offGlue_method]    @ r0<- glue->method
+    mov     r1, r3                      @ r1<- BBBB
+    mov     r2, #1                      @ r2<- true
+    ldr     r0, [r0, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- resolved ClassObject ptr
+    cmp     r0, #0                      @ got null?
+    beq     common_exceptionThrown      @ yes, handle exception
+    mov     r1, r0                      @ r1<- class resolved from BBB
+    mov     r3, rINST, lsr #12          @ r3<- B
+    GET_VREG(r0, r3)                    @ r0<- vB (object)
+    ldr     r0, [r0, #offObject_clazz]  @ r0<- obj->clazz
+    b       .LOP_INSTANCE_OF_resolved        @ pick up where we left off
+
+
+/* continuation for OP_NEW_INSTANCE */
+
+    .balign 32                          @ minimize cache lines
+.LOP_NEW_INSTANCE_finish: @ r0=class
+    bl      dvmAllocObject              @ r0<- new object
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yes, handle the exception
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r3)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+    /*
+     * Class initialization required.
+     *
+     *  r0 holds class object
+     */
+.LOP_NEW_INSTANCE_needinit:
+    mov     r9, r0                      @ save r0
+    bl      dvmInitClass                @ initialize class
+    cmp     r0, #0                      @ check boolean result
+    mov     r0, r9                      @ restore r0
+    bne     .LOP_NEW_INSTANCE_initialized     @ success, continue
+    b       common_exceptionThrown      @ failed, deal with init exception
+
+    /*
+     * Resolution required.  This is the least-likely path.
+     *
+     *  r1 holds BBBB
+     */
+.LOP_NEW_INSTANCE_resolve:
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    mov     r2, #0                      @ r2<- false
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- resolved ClassObject ptr
+    cmp     r0, #0                      @ got null?
+    bne     .LOP_NEW_INSTANCE_resolved        @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+    /*
+     * We can't instantiate an abstract class or interface, so throw an
+     * InstantiationError with the class descriptor as the message.
+     *
+     *  r0 holds class object
+     */
+.LOP_NEW_INSTANCE_abstract:
+    ldr     r1, [r0, #offClassObject_descriptor]
+    ldr     r0, .LstrInstantiationErrorPtr
+    bl      dvmThrowExceptionWithClassMessage
+    b       common_exceptionThrown
+
+.LstrInstantiationErrorPtr:
+    .word   .LstrInstantiationError
+
+
+/* continuation for OP_NEW_ARRAY */
+
+
+    /*
+     * Resolve class.  (This is an uncommon case.)
+     *
+     *  r1 holds array length
+     *  r2 holds class ref CCCC
+     */
+.LOP_NEW_ARRAY_resolve:
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    mov     r9, r1                      @ r9<- length (save)
+    mov     r1, r2                      @ r1<- CCCC
+    mov     r2, #0                      @ r2<- false
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- call(clazz, ref)
+    cmp     r0, #0                      @ got null?
+    mov     r1, r9                      @ r1<- length (restore)
+    beq     common_exceptionThrown      @ yes, handle exception
+    @ fall through to OP_NEW_ARRAY_finish
+
+    /*
+     * Finish allocation.
+     *
+     *  r0 holds class
+     *  r1 holds array length
+     */
+.LOP_NEW_ARRAY_finish:
+    mov     r2, #ALLOC_DONT_TRACK       @ don't track in local refs table
+    bl      dvmAllocArrayByClass        @ r0<- call(clazz, length, flags)
+    cmp     r0, #0                      @ failed?
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    beq     common_exceptionThrown      @ yes, handle the exception
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ vA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_FILLED_NEW_ARRAY */
+
+    /*
+     * On entry:
+     *  r0 holds array class
+     *  r10 holds AA or BA
+     */
+.LOP_FILLED_NEW_ARRAY_continue:
+    ldr     r3, [r0, #offClassObject_descriptor] @ r3<- arrayClass->descriptor
+    mov     r2, #ALLOC_DONT_TRACK       @ r2<- alloc flags
+    ldrb    r3, [r3, #1]                @ r3<- descriptor[1]
+    .if     0
+    mov     r1, r10                     @ r1<- AA (length)
+    .else
+    mov     r1, r10, lsr #4             @ r1<- B (length)
+    .endif
+    cmp     r3, #'I'                    @ array of ints?
+    cmpne   r3, #'L'                    @ array of objects?
+    cmpne   r3, #'['                    @ array of arrays?
+    mov     r9, r1                      @ save length in r9
+    bne     .LOP_FILLED_NEW_ARRAY_notimpl         @ no, not handled yet
+    bl      dvmAllocArrayByClass        @ r0<- call(arClass, length, flags)
+    cmp     r0, #0                      @ null return?
+    beq     common_exceptionThrown      @ alloc failed, handle exception
+
+    FETCH(r1, 2)                        @ r1<- FEDC or CCCC
+    str     r0, [rGLUE, #offGlue_retval]    @ retval.l <- new array
+    add     r0, r0, #offArrayObject_contents @ r0<- newArray->contents
+    subs    r9, r9, #1                  @ length--, check for neg
+    FETCH_ADVANCE_INST(3)               @ advance to next instr, load rINST
+    bmi     2f                          @ was zero, bail
+
+    @ copy values from registers into the array
+    @ r0=array, r1=CCCC/FEDC, r9=length (from AA or B), r10=AA/BA
+    .if     0
+    add     r2, rFP, r1, lsl #2         @ r2<- &fp[CCCC]
+1:  ldr     r3, [r2], #4                @ r3<- *r2++
+    subs    r9, r9, #1                  @ count--
+    str     r3, [r0], #4                @ *contents++ = vX
+    bpl     1b
+    @ continue at 2
+    .else
+    cmp     r9, #4                      @ length was initially 5?
+    and     r2, r10, #15                @ r2<- A
+    bne     1f                          @ <= 4 args, branch
+    GET_VREG(r3, r2)                    @ r3<- vA
+    sub     r9, r9, #1                  @ count--
+    str     r3, [r0, #16]               @ contents[4] = vA
+1:  and     r2, r1, #15                 @ r2<- F/E/D/C
+    GET_VREG(r3, r2)                    @ r3<- vF/vE/vD/vC
+    mov     r1, r1, lsr #4              @ r1<- next reg in low 4
+    subs    r9, r9, #1                  @ count--
+    str     r3, [r0], #4                @ *contents++ = vX
+    bpl     1b
+    @ continue at 2
+    .endif
+
+2:
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    GOTO_OPCODE(ip)                     @ execute it
+
+    /*
+     * Throw an exception indicating that we have not implemented this
+     * mode of filled-new-array.
+     */
+.LOP_FILLED_NEW_ARRAY_notimpl:
+    ldr     r0, .L_strInternalError
+    ldr     r1, .L_strFilledNewArrayNotImpl
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+    .if     (!0)                 @ define in one or the other, not both
+.L_strFilledNewArrayNotImpl:
+    .word   .LstrFilledNewArrayNotImpl
+.L_strInternalError:
+    .word   .LstrInternalError
+    .endif
+
+
+/* continuation for OP_FILLED_NEW_ARRAY_RANGE */
+
+    /*
+     * On entry:
+     *  r0 holds array class
+     *  r10 holds AA or BA
+     */
+.LOP_FILLED_NEW_ARRAY_RANGE_continue:
+    ldr     r3, [r0, #offClassObject_descriptor] @ r3<- arrayClass->descriptor
+    mov     r2, #ALLOC_DONT_TRACK       @ r2<- alloc flags
+    ldrb    r3, [r3, #1]                @ r3<- descriptor[1]
+    .if     1
+    mov     r1, r10                     @ r1<- AA (length)
+    .else
+    mov     r1, r10, lsr #4             @ r1<- B (length)
+    .endif
+    cmp     r3, #'I'                    @ array of ints?
+    cmpne   r3, #'L'                    @ array of objects?
+    cmpne   r3, #'['                    @ array of arrays?
+    mov     r9, r1                      @ save length in r9
+    bne     .LOP_FILLED_NEW_ARRAY_RANGE_notimpl         @ no, not handled yet
+    bl      dvmAllocArrayByClass        @ r0<- call(arClass, length, flags)
+    cmp     r0, #0                      @ null return?
+    beq     common_exceptionThrown      @ alloc failed, handle exception
+
+    FETCH(r1, 2)                        @ r1<- FEDC or CCCC
+    str     r0, [rGLUE, #offGlue_retval]    @ retval.l <- new array
+    add     r0, r0, #offArrayObject_contents @ r0<- newArray->contents
+    subs    r9, r9, #1                  @ length--, check for neg
+    FETCH_ADVANCE_INST(3)               @ advance to next instr, load rINST
+    bmi     2f                          @ was zero, bail
+
+    @ copy values from registers into the array
+    @ r0=array, r1=CCCC/FEDC, r9=length (from AA or B), r10=AA/BA
+    .if     1
+    add     r2, rFP, r1, lsl #2         @ r2<- &fp[CCCC]
+1:  ldr     r3, [r2], #4                @ r3<- *r2++
+    subs    r9, r9, #1                  @ count--
+    str     r3, [r0], #4                @ *contents++ = vX
+    bpl     1b
+    @ continue at 2
+    .else
+    cmp     r9, #4                      @ length was initially 5?
+    and     r2, r10, #15                @ r2<- A
+    bne     1f                          @ <= 4 args, branch
+    GET_VREG(r3, r2)                    @ r3<- vA
+    sub     r9, r9, #1                  @ count--
+    str     r3, [r0, #16]               @ contents[4] = vA
+1:  and     r2, r1, #15                 @ r2<- F/E/D/C
+    GET_VREG(r3, r2)                    @ r3<- vF/vE/vD/vC
+    mov     r1, r1, lsr #4              @ r1<- next reg in low 4
+    subs    r9, r9, #1                  @ count--
+    str     r3, [r0], #4                @ *contents++ = vX
+    bpl     1b
+    @ continue at 2
+    .endif
+
+2:
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    GOTO_OPCODE(ip)                     @ execute it
+
+    /*
+     * Throw an exception indicating that we have not implemented this
+     * mode of filled-new-array.
+     */
+.LOP_FILLED_NEW_ARRAY_RANGE_notimpl:
+    ldr     r0, .L_strInternalError
+    ldr     r1, .L_strFilledNewArrayNotImpl
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+    .if     (!1)                 @ define in one or the other, not both
+.L_strFilledNewArrayNotImpl:
+    .word   .LstrFilledNewArrayNotImpl
+.L_strInternalError:
+    .word   .LstrInternalError
+    .endif
+
+
+/* continuation for OP_CMPL_FLOAT */
+
+    @ Test for NaN with a second comparison.  EABI forbids testing bit
+    @ patterns, and we can't represent 0x7fc00000 in immediate form, so
+    @ make the library call.
+.LOP_CMPL_FLOAT_gt_or_nan:
+    mov     r1, r9                      @ reverse order
+    mov     r0, r10
+    bl      __aeabi_cfcmple             @ r0<- Z set if eq, C clear if <
+    @bleq    common_abort
+    movcc   r1, #1                      @ (greater than) r1<- 1
+    bcc     .LOP_CMPL_FLOAT_finish
+    mvn     r1, #0                            @ r1<- 1 or -1 for NaN
+    b       .LOP_CMPL_FLOAT_finish
+
+
+#if 0       /* "clasic" form */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    GET_VREG(r9, r2)                    @ r9<- vBB
+    GET_VREG(r10, r3)                   @ r10<- vCC
+    mov     r0, r9                      @ r0<- vBB
+    mov     r1, r10                     @ r1<- vCC
+    bl      __aeabi_fcmpeq              @ r0<- (vBB == vCC)
+    cmp     r0, #0                      @ equal?
+    movne   r1, #0                      @ yes, result is 0
+    bne     OP_CMPL_FLOAT_finish
+    mov     r0, r9                      @ r0<- vBB
+    mov     r1, r10                     @ r1<- vCC
+    bl      __aeabi_fcmplt              @ r0<- (vBB < vCC)
+    cmp     r0, #0                      @ less than?
+    b       OP_CMPL_FLOAT_continue
+@%break
+
+OP_CMPL_FLOAT_continue:
+    mvnne   r1, #0                      @ yes, result is -1
+    bne     OP_CMPL_FLOAT_finish
+    mov     r0, r9                      @ r0<- vBB
+    mov     r1, r10                     @ r1<- vCC
+    bl      __aeabi_fcmpgt              @ r0<- (vBB > vCC)
+    cmp     r0, #0                      @ greater than?
+    beq     OP_CMPL_FLOAT_nan               @ no, must be NaN
+    mov     r1, #1                      @ yes, result is 1
+    @ fall through to _finish
+
+OP_CMPL_FLOAT_finish:
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r3)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+    /*
+     * This is expected to be uncommon, so we double-branch (once to here,
+     * again back to _finish).
+     */
+OP_CMPL_FLOAT_nan:
+    mvn     r1, #0                            @ r1<- 1 or -1 for NaN
+    b       OP_CMPL_FLOAT_finish
+
+#endif
+
+
+/* continuation for OP_CMPG_FLOAT */
+
+    @ Test for NaN with a second comparison.  EABI forbids testing bit
+    @ patterns, and we can't represent 0x7fc00000 in immediate form, so
+    @ make the library call.
+.LOP_CMPG_FLOAT_gt_or_nan:
+    mov     r1, r9                      @ reverse order
+    mov     r0, r10
+    bl      __aeabi_cfcmple             @ r0<- Z set if eq, C clear if <
+    @bleq    common_abort
+    movcc   r1, #1                      @ (greater than) r1<- 1
+    bcc     .LOP_CMPG_FLOAT_finish
+    mov     r1, #1                            @ r1<- 1 or -1 for NaN
+    b       .LOP_CMPG_FLOAT_finish
+
+
+#if 0       /* "clasic" form */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    GET_VREG(r9, r2)                    @ r9<- vBB
+    GET_VREG(r10, r3)                   @ r10<- vCC
+    mov     r0, r9                      @ r0<- vBB
+    mov     r1, r10                     @ r1<- vCC
+    bl      __aeabi_fcmpeq              @ r0<- (vBB == vCC)
+    cmp     r0, #0                      @ equal?
+    movne   r1, #0                      @ yes, result is 0
+    bne     OP_CMPG_FLOAT_finish
+    mov     r0, r9                      @ r0<- vBB
+    mov     r1, r10                     @ r1<- vCC
+    bl      __aeabi_fcmplt              @ r0<- (vBB < vCC)
+    cmp     r0, #0                      @ less than?
+    b       OP_CMPG_FLOAT_continue
+@%break
+
+OP_CMPG_FLOAT_continue:
+    mvnne   r1, #0                      @ yes, result is -1
+    bne     OP_CMPG_FLOAT_finish
+    mov     r0, r9                      @ r0<- vBB
+    mov     r1, r10                     @ r1<- vCC
+    bl      __aeabi_fcmpgt              @ r0<- (vBB > vCC)
+    cmp     r0, #0                      @ greater than?
+    beq     OP_CMPG_FLOAT_nan               @ no, must be NaN
+    mov     r1, #1                      @ yes, result is 1
+    @ fall through to _finish
+
+OP_CMPG_FLOAT_finish:
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r3)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+    /*
+     * This is expected to be uncommon, so we double-branch (once to here,
+     * again back to _finish).
+     */
+OP_CMPG_FLOAT_nan:
+    mov     r1, #1                            @ r1<- 1 or -1 for NaN
+    b       OP_CMPG_FLOAT_finish
+
+#endif
+
+
+/* continuation for OP_CMPL_DOUBLE */
+
+    @ Test for NaN with a second comparison.  EABI forbids testing bit
+    @ patterns, and we can't represent 0x7fc00000 in immediate form, so
+    @ make the library call.
+.LOP_CMPL_DOUBLE_gt_or_nan:
+    ldmia   r10, {r0-r1}                @ reverse order
+    ldmia   r9, {r2-r3}
+    bl      __aeabi_cdcmple             @ r0<- Z set if eq, C clear if <
+    @bleq    common_abort
+    movcc   r1, #1                      @ (greater than) r1<- 1
+    bcc     .LOP_CMPL_DOUBLE_finish
+    mvn     r1, #0                            @ r1<- 1 or -1 for NaN
+    b       .LOP_CMPL_DOUBLE_finish
+
+
+/* continuation for OP_CMPG_DOUBLE */
+
+    @ Test for NaN with a second comparison.  EABI forbids testing bit
+    @ patterns, and we can't represent 0x7fc00000 in immediate form, so
+    @ make the library call.
+.LOP_CMPG_DOUBLE_gt_or_nan:
+    ldmia   r10, {r0-r1}                @ reverse order
+    ldmia   r9, {r2-r3}
+    bl      __aeabi_cdcmple             @ r0<- Z set if eq, C clear if <
+    @bleq    common_abort
+    movcc   r1, #1                      @ (greater than) r1<- 1
+    bcc     .LOP_CMPG_DOUBLE_finish
+    mov     r1, #1                            @ r1<- 1 or -1 for NaN
+    b       .LOP_CMPG_DOUBLE_finish
+
+
+/* continuation for OP_CMP_LONG */
+
+.LOP_CMP_LONG_less:
+    mvn     r1, #0                      @ r1<- -1
+    @ Want to cond code the next mov so we can avoid branch, but don't see it;
+    @ instead, we just replicate the tail end.
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r9)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+.LOP_CMP_LONG_greater:
+    mov     r1, #1                      @ r1<- 1
+    @ fall through to _finish
+
+.LOP_CMP_LONG_finish:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r9)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_AGET_WIDE */
+
+.LOP_AGET_WIDE_finish:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    add     r0, r0, #offArrayObject_contents
+    ldmia   r0, {r2-r3}                 @ r2/r3 <- vBB[vCC]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r2-r3}                 @ vAA/vAA+1<- r2/r3
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_APUT_WIDE */
+
+.LOP_APUT_WIDE_finish:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldmia   r9, {r2-r3}                 @ r2/r3<- vAA/vAA+1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    add     r0, #offArrayObject_contents
+    stmia   r0, {r2-r3}                 @ vBB[vCC] <- r2/r3
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_APUT_OBJECT */
+    /*
+     * On entry:
+     *  r1 = vBB (arrayObj)
+     *  r9 = vAA (obj)
+     *  r10 = offset into array (vBB + vCC * width)
+     */
+.LOP_APUT_OBJECT_finish:
+    cmp     r9, #0                      @ storing null reference?
+    beq     .LOP_APUT_OBJECT_skip_check      @ yes, skip type checks
+    ldr     r0, [r9, #offObject_clazz]  @ r0<- obj->clazz
+    ldr     r1, [r1, #offObject_clazz]  @ r1<- arrayObj->clazz
+    bl      dvmCanPutArrayElement       @ test object type vs. array type
+    cmp     r0, #0                      @ okay?
+    beq     common_errArrayStore        @ no
+.LOP_APUT_OBJECT_skip_check:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r9, [r10, #offArrayObject_contents] @ vBB[vCC]<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_finish:
+    @bl      common_squeak0
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    ldr   r0, [r9, r3]                @ r0<- obj.field (8/16/32 bits)
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET_WIDE */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_WIDE_finish:
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    add     r9, r9, r3                  @ r9<- obj + field offset
+    ldmia   r9, {r0-r1}                 @ r0/r1<- obj.field (64-bit align ok)
+    and     r2, r2, #15                 @ r2<- A
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    add     r3, rFP, r2, lsl #2         @ r3<- &fp[A]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r3, {r0-r1}                 @ fp[A]<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET_OBJECT */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_OBJECT_finish:
+    @bl      common_squeak0
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    ldr   r0, [r9, r3]                @ r0<- obj.field (8/16/32 bits)
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET_BOOLEAN */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_BOOLEAN_finish:
+    @bl      common_squeak1
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    ldr   r0, [r9, r3]                @ r0<- obj.field (8/16/32 bits)
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET_BYTE */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_BYTE_finish:
+    @bl      common_squeak2
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    ldr   r0, [r9, r3]                @ r0<- obj.field (8/16/32 bits)
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET_CHAR */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_CHAR_finish:
+    @bl      common_squeak3
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    ldr   r0, [r9, r3]                @ r0<- obj.field (8/16/32 bits)
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET_SHORT */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_SHORT_finish:
+    @bl      common_squeak4
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    ldr   r0, [r9, r3]                @ r0<- obj.field (8/16/32 bits)
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_finish:
+    @bl      common_squeak0
+    mov     r1, rINST, lsr #8           @ r1<- A+
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    and     r1, r1, #15                 @ r1<- A
+    cmp     r9, #0                      @ check object for null
+    GET_VREG(r0, r1)                    @ r0<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT_WIDE */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_WIDE_finish:
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    cmp     r9, #0                      @ check object for null
+    and     r2, r2, #15                 @ r2<- A
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    add     r2, rFP, r2, lsl #2         @ r3<- &fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldmia   r2, {r0-r1}                 @ r0/r1<- fp[A]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    add     r9, r9, r3                  @ r9<- object + byte offset
+    stmia   r9, {r0-r1}                 @ obj.field (64 bits, aligned)<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT_OBJECT */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_OBJECT_finish:
+    @bl      common_squeak0
+    mov     r1, rINST, lsr #8           @ r1<- A+
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    and     r1, r1, #15                 @ r1<- A
+    cmp     r9, #0                      @ check object for null
+    GET_VREG(r0, r1)                    @ r0<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT_BOOLEAN */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_BOOLEAN_finish:
+    @bl      common_squeak1
+    mov     r1, rINST, lsr #8           @ r1<- A+
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    and     r1, r1, #15                 @ r1<- A
+    cmp     r9, #0                      @ check object for null
+    GET_VREG(r0, r1)                    @ r0<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT_BYTE */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_BYTE_finish:
+    @bl      common_squeak2
+    mov     r1, rINST, lsr #8           @ r1<- A+
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    and     r1, r1, #15                 @ r1<- A
+    cmp     r9, #0                      @ check object for null
+    GET_VREG(r0, r1)                    @ r0<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT_CHAR */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_CHAR_finish:
+    @bl      common_squeak3
+    mov     r1, rINST, lsr #8           @ r1<- A+
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    and     r1, r1, #15                 @ r1<- A
+    cmp     r9, #0                      @ check object for null
+    GET_VREG(r0, r1)                    @ r0<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT_SHORT */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_SHORT_finish:
+    @bl      common_squeak4
+    mov     r1, rINST, lsr #8           @ r1<- A+
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    and     r1, r1, #15                 @ r1<- A
+    cmp     r9, #0                      @ check object for null
+    GET_VREG(r0, r1)                    @ r0<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_SGET */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SGET_WIDE */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_WIDE_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_WIDE_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SGET_OBJECT */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_OBJECT_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_OBJECT_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SGET_BOOLEAN */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_BOOLEAN_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_BOOLEAN_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SGET_BYTE */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_BYTE_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_BYTE_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SGET_CHAR */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_CHAR_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_CHAR_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SGET_SHORT */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_SHORT_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_SHORT_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SPUT_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT_WIDE */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     *  r9: &fp[AA]
+     */
+.LOP_SPUT_WIDE_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_WIDE_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT_OBJECT */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SPUT_OBJECT_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_OBJECT_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT_BOOLEAN */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SPUT_BOOLEAN_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_BOOLEAN_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT_BYTE */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SPUT_BYTE_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_BYTE_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT_CHAR */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SPUT_CHAR_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_CHAR_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT_SHORT */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SPUT_SHORT_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_SHORT_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_INVOKE_VIRTUAL */
+
+    /*
+     * At this point:
+     *  r0 = resolved base method
+     *  r10 = C or CCCC (index of first arg, which is the "this" ptr)
+     */
+.LOP_INVOKE_VIRTUAL_continue:
+    GET_VREG(r1, r10)                   @ r1<- "this" ptr
+    ldrh    r2, [r0, #offMethod_methodIndex]    @ r2<- baseMethod->methodIndex
+    cmp     r1, #0                      @ is "this" null?
+    beq     common_errNullObject        @ null "this", throw exception
+    ldr     r3, [r1, #offObject_clazz]  @ r1<- thisPtr->clazz
+    ldr     r3, [r3, #offClassObject_vtable]    @ r3<- thisPtr->clazz->vtable
+    ldr     r0, [r3, r2, lsl #2]        @ r3<- vtable[methodIndex]
+    bl      common_invokeMethodNoRange @ continue on
+
+
+/* continuation for OP_INVOKE_SUPER */
+
+    /*
+     * At this point:
+     *  r0 = resolved base method
+     *  r9 = method->clazz
+     */
+.LOP_INVOKE_SUPER_continue:
+    ldr     r1, [r9, #offClassObject_super]     @ r1<- method->clazz->super
+    ldrh    r2, [r0, #offMethod_methodIndex]    @ r2<- baseMethod->methodIndex
+    ldr     r3, [r1, #offClassObject_vtableCount]   @ r3<- super->vtableCount
+    EXPORT_PC()                         @ must export for invoke
+    cmp     r2, r3                      @ compare (methodIndex, vtableCount)
+    bcs     .LOP_INVOKE_SUPER_nsm             @ method not present in superclass
+    ldr     r1, [r1, #offClassObject_vtable]    @ r1<- ...clazz->super->vtable
+    ldr     r0, [r1, r2, lsl #2]        @ r3<- vtable[methodIndex]
+    bl      common_invokeMethodNoRange @ continue on
+
+.LOP_INVOKE_SUPER_resolve:
+    mov     r0, r9                      @ r0<- method->clazz
+    mov     r2, #METHOD_VIRTUAL         @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    bne     .LOP_INVOKE_SUPER_continue        @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+    /*
+     * Throw a NoSuchMethodError with the method name as the message.
+     *  r0 = resolved base method
+     */
+.LOP_INVOKE_SUPER_nsm:
+    ldr     r1, [r0, #offMethod_name]   @ r1<- method name
+    b       common_errNoSuchMethod
+
+
+/* continuation for OP_INVOKE_DIRECT */
+
+    /*
+     * On entry:
+     *  r1 = reference (BBBB or CCCC)
+     *  r10 = "this" register
+     */
+.LOP_INVOKE_DIRECT_resolve:
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    mov     r2, #METHOD_DIRECT          @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    GET_VREG(r2, r10)                   @ r2<- "this" ptr (reload)
+    bne     .LOP_INVOKE_DIRECT_finish          @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+
+/* continuation for OP_INVOKE_VIRTUAL_RANGE */
+
+    /*
+     * At this point:
+     *  r0 = resolved base method
+     *  r10 = C or CCCC (index of first arg, which is the "this" ptr)
+     */
+.LOP_INVOKE_VIRTUAL_RANGE_continue:
+    GET_VREG(r1, r10)                   @ r1<- "this" ptr
+    ldrh    r2, [r0, #offMethod_methodIndex]    @ r2<- baseMethod->methodIndex
+    cmp     r1, #0                      @ is "this" null?
+    beq     common_errNullObject        @ null "this", throw exception
+    ldr     r3, [r1, #offObject_clazz]  @ r1<- thisPtr->clazz
+    ldr     r3, [r3, #offClassObject_vtable]    @ r3<- thisPtr->clazz->vtable
+    ldr     r0, [r3, r2, lsl #2]        @ r3<- vtable[methodIndex]
+    bl      common_invokeMethodRange @ continue on
+
+
+/* continuation for OP_INVOKE_SUPER_RANGE */
+
+    /*
+     * At this point:
+     *  r0 = resolved base method
+     *  r9 = method->clazz
+     */
+.LOP_INVOKE_SUPER_RANGE_continue:
+    ldr     r1, [r9, #offClassObject_super]     @ r1<- method->clazz->super
+    ldrh    r2, [r0, #offMethod_methodIndex]    @ r2<- baseMethod->methodIndex
+    ldr     r3, [r1, #offClassObject_vtableCount]   @ r3<- super->vtableCount
+    EXPORT_PC()                         @ must export for invoke
+    cmp     r2, r3                      @ compare (methodIndex, vtableCount)
+    bcs     .LOP_INVOKE_SUPER_RANGE_nsm             @ method not present in superclass
+    ldr     r1, [r1, #offClassObject_vtable]    @ r1<- ...clazz->super->vtable
+    ldr     r0, [r1, r2, lsl #2]        @ r3<- vtable[methodIndex]
+    bl      common_invokeMethodRange @ continue on
+
+.LOP_INVOKE_SUPER_RANGE_resolve:
+    mov     r0, r9                      @ r0<- method->clazz
+    mov     r2, #METHOD_VIRTUAL         @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    bne     .LOP_INVOKE_SUPER_RANGE_continue        @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+    /*
+     * Throw a NoSuchMethodError with the method name as the message.
+     *  r0 = resolved base method
+     */
+.LOP_INVOKE_SUPER_RANGE_nsm:
+    ldr     r1, [r0, #offMethod_name]   @ r1<- method name
+    b       common_errNoSuchMethod
+
+
+/* continuation for OP_INVOKE_DIRECT_RANGE */
+
+    /*
+     * On entry:
+     *  r1 = reference (BBBB or CCCC)
+     *  r10 = "this" register
+     */
+.LOP_INVOKE_DIRECT_RANGE_resolve:
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    mov     r2, #METHOD_DIRECT          @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    GET_VREG(r2, r10)                   @ r2<- "this" ptr (reload)
+    bne     .LOP_INVOKE_DIRECT_RANGE_finish          @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+
+/* continuation for OP_FLOAT_TO_LONG */
+/*
+ * Convert the float in r0 to a long in r0/r1.
+ *
+ * We have to clip values to long min/max per the specification.  The
+ * expected common case is a "reasonable" value that converts directly
+ * to modest integer.  The EABI convert function isn't doing this for us.
+ */
+f2l_doconv:
+    stmfd   sp!, {r4, lr}
+    mov     r1, #0x5f000000             @ (float)maxlong
+    mov     r4, r0
+    bl      __aeabi_fcmpge              @ is arg >= maxlong?
+    cmp     r0, #0                      @ nonzero == yes
+    mvnne   r0, #0                      @ return maxlong (7fffffff)
+    mvnne   r1, #0x80000000
+    ldmnefd sp!, {r4, pc}
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, #0xdf000000             @ (float)minlong
+    bl      __aeabi_fcmple              @ is arg <= minlong?
+    cmp     r0, #0                      @ nonzero == yes
+    movne   r0, #0                      @ return minlong (80000000)
+    movne   r1, #0x80000000
+    ldmnefd sp!, {r4, pc}
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r4
+    bl      __aeabi_fcmpeq              @ is arg == self?
+    cmp     r0, #0                      @ zero == no
+    moveq   r1, #0                      @ return zero for NaN
+    ldmeqfd sp!, {r4, pc}
+
+    mov     r0, r4                      @ recover arg
+    bl      __aeabi_f2lz                @ convert float to long
+    ldmfd   sp!, {r4, pc}
+
+
+/* continuation for OP_DOUBLE_TO_LONG */
+/*
+ * Convert the double in r0/r1 to a long in r0/r1.
+ *
+ * We have to clip values to long min/max per the specification.  The
+ * expected common case is a "reasonable" value that converts directly
+ * to modest integer.  The EABI convert function isn't doing this for us.
+ */
+d2l_doconv:
+    stmfd   sp!, {r4, r5, lr}           @ save regs
+    ldr     r3, .LOP_DOUBLE_TO_LONG_max         @ (double)maxlong, hi
+    sub     sp, sp, #4                  @ align for EABI
+    mov     r2, #0                      @ (double)maxlong, lo
+    mov     r4, r0                      @ save r0
+    mov     r5, r1                      @  and r1
+    bl      __aeabi_dcmpge              @ is arg >= maxlong?
+    cmp     r0, #0                      @ nonzero == yes
+    mvnne   r0, #0                      @ return maxlong (7fffffffffffffff)
+    mvnne   r1, #0x80000000
+    bne     1f
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r5
+    ldr     r3, .LOP_DOUBLE_TO_LONG_min         @ (double)minlong, hi
+    mov     r2, #0                      @ (double)minlong, lo
+    bl      __aeabi_dcmple              @ is arg <= minlong?
+    cmp     r0, #0                      @ nonzero == yes
+    movne   r0, #0                      @ return minlong (8000000000000000)
+    movne   r1, #0x80000000
+    bne     1f
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r5
+    mov     r2, r4                      @ compare against self
+    mov     r3, r5
+    bl      __aeabi_dcmpeq              @ is arg == self?
+    cmp     r0, #0                      @ zero == no
+    moveq   r1, #0                      @ return zero for NaN
+    beq     1f
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r5
+    bl      __aeabi_d2lz                @ convert double to long
+
+1:
+    add     sp, sp, #4
+    ldmfd   sp!, {r4, r5, pc}
+
+.LOP_DOUBLE_TO_LONG_max:
+    .word   0x43e00000                  @ maxlong, as a double (high word)
+.LOP_DOUBLE_TO_LONG_min:
+    .word   0xc3e00000                  @ minlong, as a double (high word)
+
+
+/* continuation for OP_MUL_LONG */
+
+.LOP_MUL_LONG_finish:
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r0, {r9-r10}                @ vAA/vAA+1<- r9/r10
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_SHL_LONG */
+
+.LOP_SHL_LONG_finish:
+    mov     r0, r0, asl r2              @  r0<- r0 << r2
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_SHR_LONG */
+
+.LOP_SHR_LONG_finish:
+    mov     r1, r1, asr r2              @  r1<- r1 >> r2
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_USHR_LONG */
+
+.LOP_USHR_LONG_finish:
+    mov     r1, r1, lsr r2              @  r1<- r1 >>> r2
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_SHL_LONG_2ADDR */
+
+.LOP_SHL_LONG_2ADDR_finish:
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_SHR_LONG_2ADDR */
+
+.LOP_SHR_LONG_2ADDR_finish:
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_USHR_LONG_2ADDR */
+
+.LOP_USHR_LONG_2ADDR_finish:
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_EXECUTE_INLINE */
+
+    /*
+     * Extract args, call function.
+     *  r0 = #of args (0-4)
+     *  r10 = call index
+     *  lr = return addr, above  [DO NOT bl out of here w/o preserving LR]
+     *
+     * Other ideas:
+     * - Use a jump table from the main piece to jump directly into the
+     *   AND/LDR pairs.  Costs a data load, saves a branch.
+     * - Have five separate pieces that do the loading, so we can work the
+     *   interleave a little better.  Increases code size.
+     */
+.LOP_EXECUTE_INLINE_continue:
+    rsb     r0, r0, #4                  @ r0<- 4-r0
+    FETCH(r9, 2)                        @ r9<- FEDC
+    add     pc, pc, r0, lsl #3          @ computed goto, 2 instrs each
+    bl      common_abort                @ (skipped due to ARM prefetch)
+4:  and     ip, r9, #0xf000             @ isolate F
+    ldr     r3, [rFP, ip, lsr #10]      @ r3<- vF (shift right 12, left 2)
+3:  and     ip, r9, #0x0f00             @ isolate E
+    ldr     r2, [rFP, ip, lsr #6]       @ r2<- vE
+2:  and     ip, r9, #0x00f0             @ isolate D
+    ldr     r1, [rFP, ip, lsr #2]       @ r1<- vD
+1:  and     ip, r9, #0x000f             @ isolate C
+    ldr     r0, [rFP, ip, lsl #2]       @ r0<- vC
+0:
+    ldr     r9, .LOP_EXECUTE_INLINE_table       @ table of InlineOperation
+    LDR_PC  "[r9, r10, lsl #4]"         @ sizeof=16, "func" is first entry
+    @ (not reached)
+
+.LOP_EXECUTE_INLINE_table:
+    .word   gDvmInlineOpsTable
+
+
+    .size   dvmAsmSisterStart, .-dvmAsmSisterStart
+    .global dvmAsmSisterEnd
+dvmAsmSisterEnd:
+
+/* File: armv5te/footer.S */
+/*
+ * ===========================================================================
+ *  Common subroutines and data
+ * ===========================================================================
+ */
+
+    .text
+    .align  2
+
+/*
+ * Common code when a backward branch is taken.
+ *
+ * On entry:
+ *  r9 is PC adjustment *in bytes*
+ */
+common_backwardBranch:
+    mov     r0, #kInterpEntryInstr
+    bl      common_periodicChecks
+    FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/*
+ * Need to see if the thread needs to be suspended or debugger/profiler
+ * activity has begun.
+ *
+ * TODO: if JDWP isn't running, zero out pDebuggerActive pointer so we don't
+ * have to do the second ldr.
+ *
+ * TODO: reduce this so we're just checking a single location.
+ *
+ * On entry:
+ *  r0 is reentry type, e.g. kInterpEntryInstr
+ *  r9 is trampoline PC adjustment *in bytes*
+ */
+common_periodicChecks:
+    ldr     r3, [rGLUE, #offGlue_pSelfSuspendCount] @ r3<- &suspendCount
+
+#if defined(WITH_DEBUGGER)
+    ldr     r1, [rGLUE, #offGlue_pDebuggerActive]   @ r1<- &debuggerActive
+#endif
+#if defined(WITH_PROFILER)
+    ldr     r2, [rGLUE, #offGlue_pActiveProfilers]  @ r2<- &activeProfilers
+#endif
+
+    ldr     r3, [r3]                    @ r3<- suspendCount (int)
+
+#if defined(WITH_DEBUGGER)
+    ldrb    r1, [r1]                    @ r1<- debuggerActive (boolean)
+#endif
+#if defined (WITH_PROFILER)
+    ldr     r2, [r2]                    @ r2<- activeProfilers (int)
+#endif
+
+    cmp     r3, #0                      @ suspend pending?
+    bne     2f                          @ yes, check suspend
+
+#if defined(WITH_DEBUGGER) || defined(WITH_PROFILER)
+# if defined(WITH_DEBUGGER) && defined(WITH_PROFILER)
+    orrs    r1, r1, r2                  @ r1<- r1 | r2
+    cmp     r1, #0                      @ debugger attached or profiler started?
+# elif defined(WITH_DEBUGGER)
+    cmp     r1, #0                      @ debugger attached?
+# elif defined(WITH_PROFILER)
+    cmp     r2, #0                      @ profiler started?
+# endif
+    bne     3f                          @ debugger/profiler, switch interp
+#endif
+
+    bx      lr                          @ nothing to do, return
+
+2:  @ check suspend
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- glue->self
+    b       dvmCheckSuspendPending      @ suspend if necessary, then return
+
+3:  @ debugger/profiler enabled, bail out
+    add     rPC, rPC, r9                @ update rPC
+    str     r0, [rGLUE, #offGlue_entryPoint]
+    mov     r1, #1                      @ "want switch" = true
+    b       common_gotoBail
+
+
+/*
+ * The equivalent of "goto bail", this calls through the "bail handler".
+ *
+ * State registers will be saved to the "glue" area before bailing.
+ *
+ * On entry:
+ *  r1 is "bool changeInterp", indicating if we want to switch to the
+ *     other interpreter or just bail all the way out
+ */
+common_gotoBail:
+    SAVE_PC_FP_TO_GLUE()                @ export state to "glue"
+    mov     r0, rGLUE                   @ r0<- glue ptr
+    b       dvmMterpStdBail             @ call(glue, changeInterp)
+
+    @add     r1, r1, #1                  @ using (boolean+1)
+    @add     r0, rGLUE, #offGlue_jmpBuf  @ r0<- &glue->jmpBuf
+    @bl      _longjmp                    @ does not return
+    @bl      common_abort
+
+
+/*
+ * Common code for method invocation with range.
+ *
+ * On entry:
+ *  r0 is "Method* methodToCall", the method we're trying to call
+ */
+common_invokeMethodRange:
+.LinvokeNewRange:
+    @ prepare to copy args to "outs" area of current frame
+    movs    r2, rINST, lsr #8           @ r2<- AA (arg count) -- test for zero
+    SAVEAREA_FROM_FP(r10, rFP)          @ r10<- stack save area
+    beq     .LinvokeArgsDone            @ if no args, skip the rest
+    FETCH(r1, 2)                        @ r1<- CCCC
+
+    @ r0=methodToCall, r1=CCCC, r2=count, r10=outs
+    @ (very few methods have > 10 args; could unroll for common cases)
+    add     r3, rFP, r1, lsl #2         @ r3<- &fp[CCCC]
+    sub     r10, r10, r2, lsl #2        @ r10<- "outs" area, for call args
+1:  ldr     r1, [r3], #4                @ val = *fp++
+    subs    r2, r2, #1                  @ count--
+    str     r1, [r10], #4               @ *outs++ = val
+    bne     1b                          @ ...while count != 0
+    b       .LinvokeArgsDone
+
+/*
+ * Common code for method invocation without range.
+ *
+ * On entry:
+ *  r0 is "Method* methodToCall", the method we're trying to call
+ */
+common_invokeMethodNoRange:
+.LinvokeNewNoRange:
+    @ prepare to copy args to "outs" area of current frame
+    movs    r2, rINST, lsr #12          @ r2<- B (arg count) -- test for zero
+    SAVEAREA_FROM_FP(r10, rFP)          @ r10<- stack save area
+    beq     .LinvokeArgsDone            @ if no args, skip the rest
+    FETCH(r1, 2)                        @ r1<- GFED
+
+    @ r0=methodToCall, r1=GFED, r2=count, r10=outs
+.LinvokeNonRange:
+    rsb     r2, r2, #5                  @ r2<- 5-r2
+    add     pc, pc, r2, lsl #4          @ computed goto, 4 instrs each
+    bl      common_abort                @ (skipped due to ARM prefetch)
+5:  and     ip, rINST, #0x0f00          @ isolate A
+    ldr     r3, [rFP, ip, lsr #6]       @ r3<- vA (shift right 8, left 2)
+    mov     r0, r0                      @ nop
+    str     r3, [r10, #-4]!             @ *--outs = vA
+4:  and     ip, r1, #0xf000             @ isolate G
+    ldr     r3, [rFP, ip, lsr #10]      @ r3<- vG (shift right 12, left 2)
+    mov     r0, r0                      @ nop
+    str     r3, [r10, #-4]!             @ *--outs = vG
+3:  and     ip, r1, #0x0f00             @ isolate F
+    ldr     r3, [rFP, ip, lsr #6]       @ r3<- vF
+    mov     r0, r0                      @ nop
+    str     r3, [r10, #-4]!             @ *--outs = vF
+2:  and     ip, r1, #0x00f0             @ isolate E
+    ldr     r3, [rFP, ip, lsr #2]       @ r3<- vE
+    mov     r0, r0                      @ nop
+    str     r3, [r10, #-4]!             @ *--outs = vE
+1:  and     ip, r1, #0x000f             @ isolate D
+    ldr     r3, [rFP, ip, lsl #2]       @ r3<- vD
+    mov     r0, r0                      @ nop
+    str     r3, [r10, #-4]!             @ *--outs = vD
+0:  @ fall through to .LinvokeArgsDone
+
+.LinvokeArgsDone: @ r0=methodToCall
+    @ find space for the new stack frame, check for overflow
+    SAVEAREA_FROM_FP(r1, rFP)           @ r1<- stack save area
+    ldrh    r2, [r0, #offMethod_registersSize]  @ r2<- methodToCall->regsSize
+    ldrh    r3, [r0, #offMethod_outsSize]   @ r3<- methodToCall->outsSize
+    sub     r1, r1, r2, lsl #2          @ r1<- newFp (old savearea - regsSize)
+    SAVEAREA_FROM_FP(r10, r1)           @ r10<- newSaveArea
+@    bl      common_dumpRegs
+    ldr     r9, [rGLUE, #offGlue_interpStackEnd]    @ r9<- interpStackEnd
+    sub     r3, r10, r3, lsl #2         @ r3<- bottom (newsave - outsSize)
+    cmp     r3, r9                      @ bottom < interpStackEnd?
+    blt     .LstackOverflow             @ yes, this frame will overflow stack
+
+    @ set up newSaveArea
+#ifdef EASY_GDB
+    SAVEAREA_FROM_FP(ip, rFP)           @ ip<- stack save area
+    str     ip, [r10, #offStackSaveArea_prevSave]
+#endif
+    str     rFP, [r10, #offStackSaveArea_prevFrame]
+    str     rPC, [r10, #offStackSaveArea_savedPc]
+    str     r0, [r10, #offStackSaveArea_method]
+
+    ldr     r3, [r0, #offMethod_accessFlags] @ r3<- methodToCall->accessFlags
+    tst     r3, #ACC_NATIVE
+    bne     .LinvokeNative
+
+    /*
+    stmfd   sp!, {r0-r3}
+    bl      common_printNewline
+    mov     r0, rFP
+    mov     r1, #0
+    bl      dvmDumpFp
+    ldmfd   sp!, {r0-r3}
+    stmfd   sp!, {r0-r3}
+    mov     r0, r1
+    mov     r1, r10
+    bl      dvmDumpFp
+    bl      common_printNewline
+    ldmfd   sp!, {r0-r3}
+    */
+
+    @ Update "glue" values for the new method
+    @ r0=methodToCall, r1=newFp
+    ldr     r3, [r0, #offMethod_clazz]      @ r3<- method->clazz
+    str     r0, [rGLUE, #offGlue_method]    @ glue->method = methodToCall
+    ldr     r3, [r3, #offClassObject_pDvmDex] @ r3<- method->clazz->pDvmDex
+    ldr     rPC, [r0, #offMethod_insns]     @ rPC<- method->insns
+    str     r3, [rGLUE, #offGlue_methodClassDex] @ glue->methodClassDex = ...
+    ldr     r2, [rGLUE, #offGlue_self]      @ r2<- glue->self
+    FETCH_INST()                            @ load rINST from rPC
+    mov     rFP, r1                         @ fp = newFp
+    GET_INST_OPCODE(ip)                     @ extract opcode from rINST
+    str     r1, [r2, #offThread_curFrame]   @ self->curFrame = newFp
+    GOTO_OPCODE(ip)                         @ jump to next instruction
+
+.LinvokeNative:
+    @ Prep for the native call
+    @ r0=methodToCall, r1=newFp, r10=newSaveArea
+    ldr     r3, [rGLUE, #offGlue_self]      @ r3<- glue->self
+    ldr     r9, [r3, #offThread_jniLocal_nextEntry] @ r9<- thread->refNext
+    str     r1, [r3, #offThread_curFrame]   @ self->curFrame = newFp
+    str     r9, [r10, #offStackSaveArea_localRefTop] @newFp->localRefTop=refNext
+    mov     r9, r3                      @ r9<- glue->self (preserve)
+
+    mov     r2, r0                      @ r2<- methodToCall
+    mov     r0, r1                      @ r0<- newFp (points to args)
+    add     r1, rGLUE, #offGlue_retval  @ r1<- &retval
+
+#ifdef ASSIST_DEBUGGER
+    /* insert fake function header to help gdb find the stack frame */
+    b       .Lskip
+    .type   dalvik_mterp, %function
+dalvik_mterp:
+    .fnstart
+    MTERP_ENTRY1
+    MTERP_ENTRY2
+.Lskip:
+#endif
+
+    @mov     lr, pc                      @ set return addr
+    @ldr     pc, [r2, #offMethod_nativeFunc] @ pc<- methodToCall->nativeFunc
+    LDR_PC_LR "[r2, #offMethod_nativeFunc]"
+
+    @ native return; r9=self, r10=newSaveArea
+    @ equivalent to dvmPopJniLocals
+    ldr     r0, [r10, #offStackSaveArea_localRefTop] @ r0<- newSave->localRefTop
+    ldr     r1, [r9, #offThread_exception] @ check for exception
+    str     rFP, [r9, #offThread_curFrame]  @ self->curFrame = fp
+    cmp     r1, #0                      @ null?
+    str     r0, [r9, #offThread_jniLocal_nextEntry] @ self->refNext<- r0
+    bne     common_exceptionThrown      @ no, handle exception
+
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+.LstackOverflow:
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- self
+    bl      dvmHandleStackOverflow
+    b       common_exceptionThrown
+#ifdef ASSIST_DEBUGGER
+    .fnend
+#endif
+
+
+    /*
+     * Common code for method invocation, calling through "glue code".
+     *
+     * TODO: now that we have range and non-range invoke handlers, this
+     *       needs to be split into two.  Maybe just create entry points
+     *       that set r9 and jump here?
+     *
+     * On entry:
+     *  r0 is "Method* methodToCall", the method we're trying to call
+     *  r9 is "bool methodCallRange", indicating if this is a /range variant
+     */
+     .if    0
+.LinvokeOld:
+    sub     sp, sp, #8                  @ space for args + pad
+    FETCH(ip, 2)                        @ ip<- FEDC or CCCC
+    mov     r2, r0                      @ A2<- methodToCall
+    mov     r0, rGLUE                   @ A0<- glue
+    SAVE_PC_FP_TO_GLUE()                @ export state to "glue"
+    mov     r1, r9                      @ A1<- methodCallRange
+    mov     r3, rINST, lsr #8           @ A3<- AA
+    str     ip, [sp, #0]                @ A4<- ip
+    bl      dvmMterp_invokeMethod       @ call the C invokeMethod
+    add     sp, sp, #8                  @ remove arg area
+    b       common_resumeAfterGlueCall  @ continue to next instruction
+    .endif
+
+
+
+/*
+ * Common code for handling a return instruction.
+ *
+ * This does not return.
+ */
+common_returnFromMethod:
+.LreturnNew:
+    mov     r0, #kInterpEntryReturn
+    mov     r9, #0
+    bl      common_periodicChecks
+
+    SAVEAREA_FROM_FP(r0, rFP)           @ r0<- saveArea (old)
+    ldr     rFP, [r0, #offStackSaveArea_prevFrame] @ fp = saveArea->prevFrame
+    ldr     r2, [rFP, #(offStackSaveArea_method - sizeofStackSaveArea)]
+                                        @ r2<- method we're returning to
+    cmp     r2, #0                      @ is this a break frame?
+    mov     r1, #0                      @ "want switch" = false
+    beq     common_gotoBail             @ break frame, bail out completely
+
+    ldr     rPC, [r0, #offStackSaveArea_savedPc] @ pc = saveArea->savedPc
+    ldr     r3, [rGLUE, #offGlue_self]      @ r3<- glue->self
+    str     r2, [rGLUE, #offGlue_method]    @ glue->method = newSave->method
+    str     rFP, [r3, #offThread_curFrame]  @ self->curFrame = fp
+    ldr     r1, [r2, #offMethod_clazz]      @ r1<- method->clazz
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    ldr     r1, [r1, #offClassObject_pDvmDex]   @ r1<- method->clazz->pDvmDex
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [rGLUE, #offGlue_methodClassDex]
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+    /*
+     * Return handling, calls through "glue code".
+     */
+     .if    0
+.LreturnOld:
+    SAVE_PC_FP_TO_GLUE()                @ export state
+    mov     r0, rGLUE                   @ arg to function
+    bl      dvmMterp_returnFromMethod
+    b       common_resumeAfterGlueCall
+    .endif
+
+
+/*
+ * Somebody has thrown an exception.  Handle it.
+ *
+ * If the exception processing code returns to us (instead of falling
+ * out of the interpreter), continue with whatever the next instruction
+ * now happens to be.
+ *
+ * This does not return.
+ */
+common_exceptionThrown:
+.LexceptionNew:
+    mov     r0, #kInterpEntryThrow
+    mov     r9, #0
+    bl      common_periodicChecks
+
+    ldr     r10, [rGLUE, #offGlue_self] @ r10<- glue->self
+    ldr     r9, [r10, #offThread_exception] @ r9<- self->exception
+    mov     r1, r10                     @ r1<- self
+    mov     r0, r9                      @ r0<- exception
+    bl      dvmAddTrackedAlloc          @ don't let the exception be GCed
+    mov     r3, #0                      @ r3<- NULL
+    str     r3, [r10, #offThread_exception] @ self->exception = NULL
+
+    /* set up args and a local for "&fp" */
+    /* (str sp, [sp, #-4]!  would be perfect here, but is discouraged) */
+    str     rFP, [sp, #-4]!             @ *--sp = fp
+    mov     ip, sp                      @ ip<- &fp
+    mov     r3, #0                      @ r3<- false
+    str     ip, [sp, #-4]!              @ *--sp = &fp
+    ldr     r1, [rGLUE, #offGlue_method] @ r1<- glue->method
+    mov     r0, r10                     @ r0<- self
+    ldr     r1, [r1, #offMethod_insns]  @ r1<- method->insns
+    mov     r2, r9                      @ r2<- exception
+    sub     r1, rPC, r1                 @ r1<- pc - method->insns
+    mov     r1, r1, asr #1              @ r1<- offset in code units
+
+    /* call, r0 gets catchRelPc (a code-unit offset) */
+    bl      dvmFindCatchBlock           @ call(self, relPc, exc, scan?, &fp)
+
+    /* fix earlier stack overflow if necessary; may trash rFP */
+    ldrb    r1, [r10, #offThread_stackOverflowed]
+    cmp     r1, #0                      @ did we overflow earlier?
+    beq     1f                          @ no, skip ahead
+    mov     rFP, r0                     @ save relPc result in rFP
+    mov     r0, r10                     @ r0<- self
+    bl      dvmCleanupStackOverflow     @ call(self)
+    mov     r0, rFP                     @ restore result
+1:
+
+    /* update frame pointer and check result from dvmFindCatchBlock */
+    ldr     rFP, [sp, #4]               @ retrieve the updated rFP
+    cmp     r0, #0                      @ is catchRelPc < 0?
+    add     sp, sp, #8                  @ restore stack
+    bmi     .LnotCaughtLocally
+
+    /* adjust locals to match self->curFrame and updated PC */
+    SAVEAREA_FROM_FP(r1, rFP)           @ r1<- new save area
+    ldr     r1, [r1, #offStackSaveArea_method] @ r1<- new method
+    str     r1, [rGLUE, #offGlue_method]    @ glue->method = new method
+    ldr     r2, [r1, #offMethod_clazz]      @ r2<- method->clazz
+    ldr     r3, [r1, #offMethod_insns]      @ r3<- method->insns
+    ldr     r2, [r2, #offClassObject_pDvmDex] @ r2<- method->clazz->pDvmDex
+    add     rPC, r3, r0, asl #1             @ rPC<- method->insns + catchRelPc
+    str     r2, [rGLUE, #offGlue_methodClassDex] @ glue->pDvmDex = meth...
+
+    /* release the tracked alloc on the exception */
+    mov     r0, r9                      @ r0<- exception
+    mov     r1, r10                     @ r1<- self
+    bl      dvmReleaseTrackedAlloc      @ release the exception
+
+    /* restore the exception if the handler wants it */
+    FETCH_INST()                        @ load rINST from rPC
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    cmp     ip, #OP_MOVE_EXCEPTION      @ is it "move-exception"?
+    streq   r9, [r10, #offThread_exception] @ yes, restore the exception
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+.LnotCaughtLocally: @ r9=exception, r10=self
+    /* fix stack overflow if necessary */
+    ldrb    r1, [r10, #offThread_stackOverflowed]
+    cmp     r1, #0                      @ did we overflow earlier?
+    movne   r0, r10                     @ if yes: r0<- self
+    blne    dvmCleanupStackOverflow     @ if yes: call(self)
+
+    @ may want to show "not caught locally" debug messages here
+#if DVM_SHOW_EXCEPTION >= 2
+    /* call __android_log_print(prio, tag, format, ...) */
+    /* "Exception %s from %s:%d not caught locally" */
+    @ dvmLineNumFromPC(method, pc - method->insns)
+    ldr     r0, [rGLUE, #offGlue_method]
+    ldr     r1, [r0, #offMethod_insns]
+    sub     r1, rPC, r1
+    asr     r1, r1, #1
+    bl      dvmLineNumFromPC
+    str     r0, [sp, #-4]!
+    @ dvmGetMethodSourceFile(method)
+    ldr     r0, [rGLUE, #offGlue_method]
+    bl      dvmGetMethodSourceFile
+    str     r0, [sp, #-4]!
+    @ exception->clazz->descriptor
+    ldr     r3, [r9, #offObject_clazz]
+    ldr     r3, [r3, #offClassObject_descriptor]
+    @
+    ldr     r2, strExceptionNotCaughtLocally
+    ldr     r1, strLogTag
+    mov     r0, #3                      @ LOG_DEBUG
+    bl      __android_log_print
+#endif
+    str     r9, [r10, #offThread_exception] @ restore exception
+    mov     r0, r9                      @ r0<- exception
+    mov     r1, r10                     @ r1<- self
+    bl      dvmReleaseTrackedAlloc      @ release the exception
+    mov     r1, #0                      @ "want switch" = false
+    b       common_gotoBail             @ bail out
+
+
+    /*
+     * Exception handling, calls through "glue code".
+     */
+    .if     0
+.LexceptionOld:
+    SAVE_PC_FP_TO_GLUE()                @ export state
+    mov     r0, rGLUE                   @ arg to function
+    bl      dvmMterp_exceptionThrown
+    b       common_resumeAfterGlueCall
+    .endif
+
+
+/*
+ * After returning from a "glued" function, pull out the updated
+ * values and start executing at the next instruction.
+ */
+common_resumeAfterGlueCall:
+    LOAD_PC_FP_FROM_GLUE()              @ pull rPC and rFP out of glue
+    FETCH_INST()                        @ load rINST from rPC
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/*
+ * Invalid array index.
+ */
+common_errArrayIndex:
+    EXPORT_PC()
+    ldr     r0, strArrayIndexException
+    mov     r1, #0
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+/*
+ * Invalid array value.
+ */
+common_errArrayStore:
+    EXPORT_PC()
+    ldr     r0, strArrayStoreException
+    mov     r1, #0
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+/*
+ * Integer divide or mod by zero.
+ */
+common_errDivideByZero:
+    EXPORT_PC()
+    ldr     r0, strArithmeticException
+    ldr     r1, strDivideByZero
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+/*
+ * Attempt to allocate an array with a negative size.
+ */
+common_errNegativeArraySize:
+    EXPORT_PC()
+    ldr     r0, strNegativeArraySizeException
+    mov     r1, #0
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+/*
+ * Invocation of a non-existent method.
+ */
+common_errNoSuchMethod:
+    EXPORT_PC()
+    ldr     r0, strNoSuchMethodError
+    mov     r1, #0
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+/*
+ * We encountered a null object when we weren't expecting one.  We
+ * export the PC, throw a NullPointerException, and goto the exception
+ * processing code.
+ */
+common_errNullObject:
+    EXPORT_PC()
+    ldr     r0, strNullPointerException
+    mov     r1, #0
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+/*
+ * For debugging, cause an immediate fault.  The source address will
+ * be in lr (use a bl instruction to jump here).
+ */
+common_abort:
+    ldr     pc, .LdeadFood
+.LdeadFood:
+    .word   0xdeadf00d
+
+/*
+ * Spit out a "we were here", preserving all registers.  (The attempt
+ * to save ip won't work, but we need to save an even number of
+ * registers for EABI 64-bit stack alignment.)
+ */
+    .macro  SQUEAK num
+common_squeak\num:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    ldr     r0, strSqueak
+    mov     r1, #\num
+    bl      printf
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+    .endm
+
+    SQUEAK  0
+    SQUEAK  1
+    SQUEAK  2
+    SQUEAK  3
+    SQUEAK  4
+    SQUEAK  5
+
+/*
+ * Spit out the number in r0, preserving registers.
+ */
+common_printNum:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    mov     r1, r0
+    ldr     r0, strSqueak
+    bl      printf
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+
+/*
+ * Print a newline, preserving registers.
+ */
+common_printNewline:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    ldr     r0, strNewline
+    bl      printf
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+
+    /*
+     * Print the 32-bit quantity in r0 as a hex value, preserving registers.
+     */
+common_printHex:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    mov     r1, r0
+    ldr     r0, strPrintHex
+    bl      printf
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+
+/*
+ * Print the 64-bit quantity in r0-r1, preserving registers.
+ */
+common_printLong:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    mov     r3, r1
+    mov     r2, r0
+    ldr     r0, strPrintLong
+    bl      printf
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+
+/*
+ * Print full method info.  Pass the Method* in r0.  Preserves regs.
+ */
+common_printMethod:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bl      dvmMterpPrintMethod
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+
+/*
+ * Call a C helper function that dumps regs and possibly some
+ * additional info.  Requires the C function to be compiled in.
+ */
+    .if     0
+common_dumpRegs:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bl      dvmMterpDumpArmRegs
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+    .endif
+
+
+/*
+ * String references, must be close to the code that uses them.
+ */
+    .align  2
+strArithmeticException:
+    .word   .LstrArithmeticException
+strArrayIndexException:
+    .word   .LstrArrayIndexException
+strArrayStoreException:
+    .word   .LstrArrayStoreException
+strDivideByZero:
+    .word   .LstrDivideByZero
+strNegativeArraySizeException:
+    .word   .LstrNegativeArraySizeException
+strNoSuchMethodError:
+    .word   .LstrNoSuchMethodError
+strNullPointerException:
+    .word   .LstrNullPointerException
+
+strLogTag:
+    .word   .LstrLogTag
+strExceptionNotCaughtLocally:
+    .word   .LstrExceptionNotCaughtLocally
+
+strNewline:
+    .word   .LstrNewline
+strSqueak:
+    .word   .LstrSqueak
+strPrintHex:
+    .word   .LstrPrintHex
+strPrintLong:
+    .word   .LstrPrintLong
+
+/*
+ * Zero-terminated ASCII string data.
+ *
+ * On ARM we have two choices: do like gcc does, and LDR from a .word
+ * with the address, or use an ADR pseudo-op to get the address
+ * directly.  ADR saves 4 bytes and an indirection, but it's using a
+ * PC-relative addressing mode and hence has a limited range, which
+ * makes it not work well with mergeable string sections.
+ */
+    .section .rodata.str1.4,"aMS",%progbits,1
+
+.LstrBadEntryPoint:
+    .asciz  "Bad entry point %d\n"
+.LstrArithmeticException:
+    .asciz  "Ljava/lang/ArithmeticException;"
+.LstrArrayIndexException:
+    .asciz  "Ljava/lang/ArrayIndexOutOfBoundsException;"
+.LstrArrayStoreException:
+    .asciz  "Ljava/lang/ArrayStoreException;"
+.LstrClassCastException:
+    .asciz  "Ljava/lang/ClassCastException;"
+.LstrDivideByZero:
+    .asciz  "divide by zero"
+.LstrFilledNewArrayNotImpl:
+    .asciz  "filled-new-array only implemented for objects and 'int'"
+.LstrInternalError:
+    .asciz  "Ljava/lang/InternalError;"
+.LstrInstantiationError:
+    .asciz  "Ljava/lang/InstantiationError;"
+.LstrNegativeArraySizeException:
+    .asciz  "Ljava/lang/NegativeArraySizeException;"
+.LstrNoSuchMethodError:
+    .asciz  "Ljava/lang/NoSuchMethodError;"
+.LstrNullPointerException:
+    .asciz  "Ljava/lang/NullPointerException;"
+
+.LstrLogTag:
+    .asciz  "mterp"
+.LstrExceptionNotCaughtLocally:
+    .asciz  "Exception %s from %s:%d not caught locally\n"
+
+.LstrNewline:
+    .asciz  "\n"
+.LstrSqueak:
+    .asciz  "<%d>"
+.LstrPrintHex:
+    .asciz  "<0x%x>"
+.LstrPrintLong:
+    .asciz  "<%lld>"
+
+
diff --git a/vm/mterp/out/InterpAsm-armv5te.S b/vm/mterp/out/InterpAsm-armv5te.S
new file mode 100644
index 0000000..9987ff5
--- /dev/null
+++ b/vm/mterp/out/InterpAsm-armv5te.S
@@ -0,0 +1,10006 @@
+/*
+ * This file was generated automatically by gen-mterp.py for 'armv5te'.
+ *
+ * --> DO NOT EDIT <--
+ */
+
+/* File: armv5te/header.S */
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ * ARMv5 definitions and declarations.
+ */
+
+/*
+ARM EABI general notes:
+
+r0-r3 hold first 4 args to a method; they are not preserved across method calls
+r4-r8 are available for general use
+r9 is given special treatment in some situations, but not for us
+r10 (sl) seems to be generally available
+r11 (fp) is used by gcc (unless -fomit-frame-pointer is set)
+r12 (ip) is scratch -- not preserved across method calls
+r13 (sp) should be managed carefully in case a signal arrives
+r14 (lr) must be preserved
+r15 (pc) can be tinkered with directly
+
+r0 holds returns of <= 4 bytes
+r0-r1 hold returns of 8 bytes, low word in r0
+
+Callee must save/restore r4+ (except r12) if it modifies them.
+
+Stack is "full descending".  Only the arguments that don't fit in the first 4
+registers are placed on the stack.  "sp" points at the first stacked argument
+(i.e. the 5th arg).
+
+VFP: single-precision results in s0, double-precision results in d0.
+
+In the EABI, "sp" must be 64-bit aligned on entry to a function, and any
+64-bit quantities (long long, double) must be 64-bit aligned.
+*/
+
+/*
+Mterp and ARM notes:
+
+The following registers have fixed assignments:
+
+  reg nick      purpose
+  r4  rPC       interpreted program counter, used for fetching instructions
+  r5  rFP       interpreted frame pointer, used for accessing locals and args
+  r6  rGLUE     MterpGlue pointer
+  r7  rIBASE    interpreted instruction base pointer, used for computed goto
+  r8  rINST     first 16-bit code unit of current instruction
+
+Macros are provided for common operations.  Each macro MUST emit only
+one instruction to make instruction-counting easier.  They MUST NOT alter
+unspecified registers or condition codes.
+*/
+
+/* single-purpose registers, given names for clarity */
+#define rPC     r4
+#define rFP     r5
+#define rGLUE   r6
+#define rIBASE  r7
+#define rINST   r8
+
+/* save/restore the PC and/or FP from the glue struct */
+#define LOAD_PC_FROM_GLUE()     ldr     rPC, [rGLUE, #offGlue_pc]
+#define SAVE_PC_TO_GLUE()       str     rPC, [rGLUE, #offGlue_pc]
+#define LOAD_FP_FROM_GLUE()     ldr     rFP, [rGLUE, #offGlue_fp]
+#define SAVE_FP_TO_GLUE()       str     rFP, [rGLUE, #offGlue_fp]
+#define LOAD_PC_FP_FROM_GLUE()  ldmia   rGLUE, {rPC, rFP}
+#define SAVE_PC_FP_TO_GLUE()    stmia   rGLUE, {rPC, rFP}
+
+/*
+ * "export" the PC to the stack frame, f/b/o future exception objects.  Must
+ * be done *before* something calls dvmThrowException.
+ *
+ * In C this is "SAVEAREA_FROM_FP(fp)->xtra.currentPc = pc", i.e.
+ * fp - sizeof(StackSaveArea) + offsetof(SaveArea, xtra.currentPc)
+ *
+ * It's okay to do this more than once.
+ */
+#define EXPORT_PC() \
+    str     rPC, [rFP, #(-sizeofStackSaveArea + offStackSaveArea_currentPc)]
+
+/*
+ * Given a frame pointer, find the stack save area.
+ *
+ * In C this is "((StackSaveArea*)(_fp) -1)".
+ */
+#define SAVEAREA_FROM_FP(_reg, _fpreg) \
+    sub     _reg, _fpreg, #sizeofStackSaveArea
+
+/*
+ * Fetch the next instruction from rPC into rINST.  Does not advance rPC.
+ */
+#define FETCH_INST()            ldrh    rINST, [rPC]
+
+/*
+ * Fetch the next instruction from the specified offset.  Advances rPC
+ * to point to the next instruction.  "_count" is in 16-bit code units.
+ *
+ * Because of the limited size of immediate constants on ARM, this is only
+ * suitable for small forward movements (i.e. don't try to implement "goto"
+ * with this).
+ *
+ * This must come AFTER anything that can throw an exception, or the
+ * exception catch may miss.  (This also implies that it must come after
+ * EXPORT_PC().)
+ */
+#define FETCH_ADVANCE_INST(_count) ldrh    rINST, [rPC, #(_count*2)]!
+
+/*
+ * Fetch the next instruction from an offset specified by _reg.  Updates
+ * rPC to point to the next instruction.  "_reg" must specify the distance
+ * in bytes, *not* 16-bit code units, and may be a signed value.
+ *
+ * We want to write "ldrh rINST, [rPC, _reg, lsl #2]!", but some of the
+ * bits that hold the shift distance are used for the half/byte/sign flags.
+ * In some cases we can pre-double _reg for free, so we require a byte offset
+ * here.
+ */
+#define FETCH_ADVANCE_INST_RB(_reg) ldrh    rINST, [rPC, _reg]!
+
+/*
+ * Fetch a half-word code unit from an offset past the current PC.  The
+ * "_count" value is in 16-bit code units.  Does not advance rPC.
+ *
+ * The "_S" variant works the same but treats the value as signed.
+ */
+#define FETCH(_reg, _count)     ldrh    _reg, [rPC, #(_count*2)]
+#define FETCH_S(_reg, _count)   ldrsh   _reg, [rPC, #(_count*2)]
+
+/*
+ * Fetch one byte from an offset past the current PC.  Pass in the same
+ * "_count" as you would for FETCH, and an additional 0/1 indicating which
+ * byte of the halfword you want (lo/hi).
+ */
+#define FETCH_B(_reg, _count, _byte) ldrb     _reg, [rPC, #(_count*2+_byte)]
+
+/*
+ * Put the instruction's opcode field into the specified register.
+ */
+#define GET_INST_OPCODE(_reg)   and     _reg, rINST, #255
+
+/*
+ * Begin executing the opcode in _reg.  Because this only jumps within the
+ * interpreter, we don't have to worry about pre-ARMv5 THUMB interwork.
+ */
+#define GOTO_OPCODE(_reg)       add     pc, rIBASE, _reg, lsl #6
+
+/*
+ * Get/set the 32-bit value from a Dalvik register.
+ */
+#define GET_VREG(_reg, _vreg)   ldr     _reg, [rFP, _vreg, lsl #2]
+#define SET_VREG(_reg, _vreg)   str     _reg, [rFP, _vreg, lsl #2]
+
+/*
+ * This is a #include, not a %include, because we want the C pre-processor
+ * to expand the macros into assembler assignment statements.
+ */
+#include "../common/asm-constants.h"
+
+
+/* File: armv5te/platform.S */
+/*
+ * ===========================================================================
+ *  CPU-version-specific defines
+ * ===========================================================================
+ */
+
+/*
+ * Macro for "LDR PC,xxx", which is not allowed pre-ARMv5.  Essentially a
+ * one-way branch.
+ *
+ * May modify IP.  Does not modify LR.
+ */
+.macro  LDR_PC source
+    ldr     pc, \source
+.endm
+
+/*
+ * Macro for "MOV LR,PC / LDR PC,xxx", which is not allowed pre-ARMv5.
+ * Jump to subroutine.
+ *
+ * May modify IP and LR.
+ */
+.macro  LDR_PC_LR source
+    mov     lr, pc
+    ldr     pc, \source
+.endm
+
+/*
+ * Macro for "LDMFD SP!, {...regs...,PC}".
+ *
+ * May modify IP and LR.
+ */
+.macro  LDMFD_PC regs
+    ldmfd   sp!, {\regs,pc}
+.endm
+
+
+/* File: armv5te/entry.S */
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ * Interpreter entry point.
+ */
+
+/*
+ * We don't have formal stack frames, so gdb scans upward in the code
+ * to find the start of the function (a label with the %function type),
+ * and then looks at the next few instructions to figure out what
+ * got pushed onto the stack.  From this it figures out how to restore
+ * the registers, including PC, for the previous stack frame.  If gdb
+ * sees a non-function label, it stops scanning, so either we need to
+ * have nothing but assembler-local labels between the entry point and
+ * the break, or we need to fake it out.
+ *
+ * When this is defined, we add some stuff to make gdb less confused.
+ */
+#define ASSIST_DEBUGGER 1
+
+    .text
+    .align  2
+    .global dvmMterpStdRun
+    .type   dvmMterpStdRun, %function
+
+/*
+ * On entry:
+ *  r0  MterpGlue* glue
+ *
+ * This function returns a boolean "changeInterp" value.  The return comes
+ * via a call to dvmMterpStdBail().
+ */
+dvmMterpStdRun:
+#define MTERP_ENTRY1 \
+    .save {r4-r10,fp,lr}; \
+    stmfd   sp!, {r4-r10,fp,lr}         @ save 9 regs
+#define MTERP_ENTRY2 \
+    .pad    #4; \
+    sub     sp, sp, #4                  @ align 64
+
+    .fnstart
+    MTERP_ENTRY1
+    MTERP_ENTRY2
+
+    /* save stack pointer, add magic word for debuggerd */
+    str     sp, [r0, #offGlue_bailPtr]  @ save SP for eventual return
+
+    /* set up "named" registers, figure out entry point */
+    mov     rGLUE, r0                   @ set rGLUE
+    ldrb    r1, [r0, #offGlue_entryPoint]   @ InterpEntry enum is char
+    LOAD_PC_FP_FROM_GLUE()              @ load rPC and rFP from "glue"
+    adr     rIBASE, dvmAsmInstructionStart  @ set rIBASE
+    cmp     r1, #kInterpEntryInstr      @ usual case?
+    bne     .Lnot_instr                 @ no, handle it
+
+    /* start executing the instruction at rPC */
+    FETCH_INST()                        @ load rINST from rPC
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+.Lnot_instr:
+    cmp     r1, #kInterpEntryReturn     @ were we returning from a method?
+    beq     common_returnFromMethod
+
+.Lnot_return:
+    cmp     r1, #kInterpEntryThrow      @ were we throwing an exception?
+    beq     common_exceptionThrown
+
+.Lbad_arg:
+    ldr     r0, strBadEntryPoint
+    @ r1 holds value of entryPoint
+    bl      printf
+    bl      dvmAbort
+    .fnend
+
+
+    .global dvmMterpStdBail
+    .type   dvmMterpStdBail, %function
+
+/*
+ * Restore the stack pointer and PC from the save point established on entry.
+ * This is essentially the same as a longjmp, but should be cheaper.  The
+ * last instruction causes us to return to whoever called dvmMterpStdRun.
+ *
+ * We pushed some registers on the stack in dvmMterpStdRun, then saved
+ * SP and LR.  Here we restore SP, restore the registers, and then restore
+ * LR to PC.
+ *
+ * On entry:
+ *  r0  MterpGlue* glue
+ *  r1  bool changeInterp
+ */
+dvmMterpStdBail:
+    ldr     sp, [r0, #offGlue_bailPtr]      @ sp<- saved SP
+    mov     r0, r1                          @ return the changeInterp value
+    add     sp, sp, #4                      @ un-align 64
+    LDMFD_PC "r4-r10,fp"                    @ restore 9 regs and return
+
+
+/*
+ * String references.
+ */
+strBadEntryPoint:
+    .word   .LstrBadEntryPoint
+
+
+
+    .global dvmAsmInstructionStart
+    .type   dvmAsmInstructionStart, %function
+dvmAsmInstructionStart = .L_OP_NOP
+    .text
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NOP: /* 0x00 */
+/* File: armv5te/OP_NOP.S */
+    FETCH_ADVANCE_INST(1)               @ advance to next instr, load rINST
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    GOTO_OPCODE(ip)                     @ execute it
+
+#ifdef ASSIST_DEBUGGER
+    /* insert fake function header to help gdb find the stack frame */
+    .type   dalvik_inst, %function
+dalvik_inst:
+    .fnstart
+    MTERP_ENTRY1
+    MTERP_ENTRY2
+    .fnend
+#endif
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE: /* 0x01 */
+/* File: armv5te/OP_MOVE.S */
+    /* for move, move-object, long-to-int */
+    /* op vA, vB */
+    mov     r1, rINST, lsr #12          @ r1<- B from 15:12
+    mov     r0, rINST, lsr #8           @ r0<- A from 11:8
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[B]
+    and     r0, r0, #15
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[A]<- r2
+    GOTO_OPCODE(ip)                     @ execute next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_FROM16: /* 0x02 */
+/* File: armv5te/OP_MOVE_FROM16.S */
+    /* for: move/from16, move-object/from16 */
+    /* op vAA, vBBBB */
+    FETCH(r1, 1)                        @ r1<- BBBB
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[BBBB]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[AA]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_16: /* 0x03 */
+/* File: armv5te/OP_MOVE_16.S */
+    /* for: move/16, move-object/16 */
+    /* op vAAAA, vBBBB */
+    FETCH(r1, 2)                        @ r1<- BBBB
+    FETCH(r0, 1)                        @ r0<- AAAA
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[BBBB]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[AAAA]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_WIDE: /* 0x04 */
+/* File: armv5te/OP_MOVE_WIDE.S */
+    /* move-wide vA, vB */
+    /* NOTE: regs can overlap, e.g. "move v6,v7" or "move v7,v6" */
+    mov     r2, rINST, lsr #8           @ r2<- A(+)
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r2, r2, #15
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[A]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- fp[B]
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r2, {r0-r1}                 @ fp[A]<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_WIDE_FROM16: /* 0x05 */
+/* File: armv5te/OP_MOVE_WIDE_FROM16.S */
+    /* move-wide/from16 vAA, vBBBB */
+    /* NOTE: regs can overlap, e.g. "move v6,v7" or "move v7,v6" */
+    FETCH(r3, 1)                        @ r3<- BBBB
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[BBBB]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[AA]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- fp[BBBB]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r2, {r0-r1}                 @ fp[AA]<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_WIDE_16: /* 0x06 */
+/* File: armv5te/OP_MOVE_WIDE_16.S */
+    /* move-wide/16 vAAAA, vBBBB */
+    /* NOTE: regs can overlap, e.g. "move v6,v7" or "move v7,v6" */
+    FETCH(r3, 2)                        @ r3<- BBBB
+    FETCH(r2, 1)                        @ r2<- AAAA
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[BBBB]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[AAAA]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- fp[BBBB]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r2, {r0-r1}                 @ fp[AAAA]<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_OBJECT: /* 0x07 */
+/* File: armv5te/OP_MOVE_OBJECT.S */
+/* File: armv5te/OP_MOVE.S */
+    /* for move, move-object, long-to-int */
+    /* op vA, vB */
+    mov     r1, rINST, lsr #12          @ r1<- B from 15:12
+    mov     r0, rINST, lsr #8           @ r0<- A from 11:8
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[B]
+    and     r0, r0, #15
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[A]<- r2
+    GOTO_OPCODE(ip)                     @ execute next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_OBJECT_FROM16: /* 0x08 */
+/* File: armv5te/OP_MOVE_OBJECT_FROM16.S */
+/* File: armv5te/OP_MOVE_FROM16.S */
+    /* for: move/from16, move-object/from16 */
+    /* op vAA, vBBBB */
+    FETCH(r1, 1)                        @ r1<- BBBB
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[BBBB]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[AA]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_OBJECT_16: /* 0x09 */
+/* File: armv5te/OP_MOVE_OBJECT_16.S */
+/* File: armv5te/OP_MOVE_16.S */
+    /* for: move/16, move-object/16 */
+    /* op vAAAA, vBBBB */
+    FETCH(r1, 2)                        @ r1<- BBBB
+    FETCH(r0, 1)                        @ r0<- AAAA
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[BBBB]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[AAAA]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_RESULT: /* 0x0a */
+/* File: armv5te/OP_MOVE_RESULT.S */
+    /* for: move-result, move-result-object */
+    /* op vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    ldr     r0, [rGLUE, #offGlue_retval]    @ r0<- glue->retval.i
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[AA]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_RESULT_WIDE: /* 0x0b */
+/* File: armv5te/OP_MOVE_RESULT_WIDE.S */
+    /* move-result-wide vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    add     r3, rGLUE, #offGlue_retval  @ r3<- &glue->retval
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[AA]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- retval.j
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r2, {r0-r1}                 @ fp[AA]<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_RESULT_OBJECT: /* 0x0c */
+/* File: armv5te/OP_MOVE_RESULT_OBJECT.S */
+/* File: armv5te/OP_MOVE_RESULT.S */
+    /* for: move-result, move-result-object */
+    /* op vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    ldr     r0, [rGLUE, #offGlue_retval]    @ r0<- glue->retval.i
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[AA]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_EXCEPTION: /* 0x0d */
+/* File: armv5te/OP_MOVE_EXCEPTION.S */
+    /* move-exception vAA */
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- glue->self
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    ldr     r3, [r0, #offThread_exception]  @ r3<- dvmGetException bypass
+    mov     r1, #0                      @ r1<- 0
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    SET_VREG(r3, r2)                    @ fp[AA]<- exception obj
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offThread_exception]  @ dvmClearException bypass
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RETURN_VOID: /* 0x0e */
+/* File: armv5te/OP_RETURN_VOID.S */
+    b       common_returnFromMethod
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RETURN: /* 0x0f */
+/* File: armv5te/OP_RETURN.S */
+    /*
+     * Return a 32-bit value.  Copies the return value into the "glue"
+     * structure, then jumps to the return handler.
+     *
+     * for: return, return-object
+     */
+    /* op vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    GET_VREG(r0, r2)                    @ r0<- vAA
+    str     r0, [rGLUE, #offGlue_retval] @ retval.i <- vAA
+    b       common_returnFromMethod
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RETURN_WIDE: /* 0x10 */
+/* File: armv5te/OP_RETURN_WIDE.S */
+    /*
+     * Return a 64-bit value.  Copies the return value into the "glue"
+     * structure, then jumps to the return handler.
+     */
+    /* return-wide vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[AA]
+    add     r3, rGLUE, #offGlue_retval  @ r3<- &glue->retval
+    ldmia   r2, {r0-r1}                 @ r0/r1 <- vAA/vAA+1
+    stmia   r3, {r0-r1}                 @ retval<- r0/r1
+    b       common_returnFromMethod
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RETURN_OBJECT: /* 0x11 */
+/* File: armv5te/OP_RETURN_OBJECT.S */
+/* File: armv5te/OP_RETURN.S */
+    /*
+     * Return a 32-bit value.  Copies the return value into the "glue"
+     * structure, then jumps to the return handler.
+     *
+     * for: return, return-object
+     */
+    /* op vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    GET_VREG(r0, r2)                    @ r0<- vAA
+    str     r0, [rGLUE, #offGlue_retval] @ retval.i <- vAA
+    b       common_returnFromMethod
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_4: /* 0x12 */
+/* File: armv5te/OP_CONST_4.S */
+    /* const/4 vA, #+B */
+    mov     r1, rINST, lsl #16          @ r1<- Bxxx0000
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mov     r1, r1, asr #28             @ r1<- sssssssB (sign-extended)
+    and     r0, r0, #15
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    SET_VREG(r1, r0)                    @ fp[A]<- r1
+    GOTO_OPCODE(ip)                     @ execute next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_16: /* 0x13 */
+/* File: armv5te/OP_CONST_16.S */
+    /* const/16 vAA, #+BBBB */
+    FETCH_S(r0, 1)                      @ r0<- ssssBBBB (sign-extended)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r0, r3)                    @ vAA<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST: /* 0x14 */
+/* File: armv5te/OP_CONST.S */
+    /* const vAA, #+BBBBbbbb */
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH(r0, 1)                        @ r0<- bbbb (low)
+    FETCH(r1, 2)                        @ r1<- BBBB (high)
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    orr     r0, r0, r1, lsl #16         @ r0<- BBBBbbbb
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r3)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_HIGH16: /* 0x15 */
+/* File: armv5te/OP_CONST_HIGH16.S */
+    /* const/high16 vAA, #+BBBB0000 */
+    FETCH(r0, 1)                        @ r0<- 0000BBBB (zero-extended)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    mov     r0, r0, lsl #16             @ r0<- BBBB0000
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r0, r3)                    @ vAA<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_WIDE_16: /* 0x16 */
+/* File: armv5te/OP_CONST_WIDE_16.S */
+    /* const-wide/16 vAA, #+BBBB */
+    FETCH_S(r0, 1)                      @ r0<- ssssBBBB (sign-extended)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    mov     r1, r0, asr #31             @ r1<- ssssssss
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r3, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_WIDE_32: /* 0x17 */
+/* File: armv5te/OP_CONST_WIDE_32.S */
+    /* const-wide/32 vAA, #+BBBBbbbb */
+    FETCH(r0, 1)                        @ r0<- 0000bbbb (low)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH_S(r2, 2)                      @ r2<- ssssBBBB (high)
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    orr     r0, r0, r2, lsl #16         @ r0<- BBBBbbbb
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[AA]
+    mov     r1, r0, asr #31             @ r1<- ssssssss
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r3, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_WIDE: /* 0x18 */
+/* File: armv5te/OP_CONST_WIDE.S */
+    /* const-wide vAA, #+HHHHhhhhBBBBbbbb */
+    FETCH(r0, 1)                        @ r0<- bbbb (low)
+    FETCH(r1, 2)                        @ r1<- BBBB (low middle)
+    FETCH(r2, 3)                        @ r2<- hhhh (high middle)
+    orr     r0, r0, r1, lsl #16         @ r0<- BBBBbbbb (low word)
+    FETCH(r3, 4)                        @ r3<- HHHH (high)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    orr     r1, r2, r3, lsl #16         @ r1<- HHHHhhhh (high word)
+    FETCH_ADVANCE_INST(5)               @ advance rPC, load rINST
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_WIDE_HIGH16: /* 0x19 */
+/* File: armv5te/OP_CONST_WIDE_HIGH16.S */
+    /* const-wide/high16 vAA, #+BBBB000000000000 */
+    FETCH(r1, 1)                        @ r1<- 0000BBBB (zero-extended)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    mov     r0, #0                      @ r0<- 00000000
+    mov     r1, r1, lsl #16             @ r1<- BBBB0000
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r3, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_STRING: /* 0x1a */
+/* File: armv5te/OP_CONST_STRING.S */
+    /* const/string vAA, String@BBBB */
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- glue->methodClassDex
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    ldr     r2, [r2, #offDvmDex_pResStrings]   @ r2<- dvmDex->pResStrings
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- pResStrings[BBBB]
+    cmp     r0, #0                      @ not yet resolved?
+    beq     .LOP_CONST_STRING_resolve
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_STRING_JUMBO: /* 0x1b */
+/* File: armv5te/OP_CONST_STRING_JUMBO.S */
+    /* const/string vAA, String@BBBBBBBB */
+    FETCH(r0, 1)                        @ r0<- bbbb (low)
+    FETCH(r1, 2)                        @ r1<- BBBB (high)
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- glue->methodClassDex
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    ldr     r2, [r2, #offDvmDex_pResStrings]   @ r2<- dvmDex->pResStrings
+    orr     r1, r0, r1, lsl #16         @ r1<- BBBBbbbb
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- pResStrings[BBBB]
+    cmp     r0, #0
+    beq     .LOP_CONST_STRING_JUMBO_resolve
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_CLASS: /* 0x1c */
+/* File: armv5te/OP_CONST_CLASS.S */
+    /* const/class vAA, Class@BBBB */
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- glue->methodClassDex
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    ldr     r2, [r2, #offDvmDex_pResClasses]   @ r2<- dvmDex->pResClasses
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- pResClasses[BBBB]
+    cmp     r0, #0                      @ not yet resolved?
+    beq     .LOP_CONST_CLASS_resolve
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MONITOR_ENTER: /* 0x1d */
+/* File: armv5te/OP_MONITOR_ENTER.S */
+    /*
+     * Synchronize on an object.
+     */
+    /* monitor-enter vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    GET_VREG(r1, r2)                    @ r1<- vAA (object)
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- glue->self
+    cmp     r1, #0                      @ null object?
+#ifdef WITH_MONITOR_TRACKING
+    EXPORT_PC()                         @ export PC so we can grab stack trace
+#endif
+    beq     common_errNullObject        @ null object, throw an exception
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    bl      dvmLockObject               @ call(self, obj)
+#ifdef WITH_DEADLOCK_PREDICTION /* implies WITH_MONITOR_TRACKING */
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- glue->self
+    ldr     r1, [r0, #offThread_exception] @ check for exception
+    cmp     r1, #0
+    bne     common_exceptionThrown      @ exception raised, bail out
+#endif
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MONITOR_EXIT: /* 0x1e */
+/* File: armv5te/OP_MONITOR_EXIT.S */
+    /*
+     * Unlock an object.
+     *
+     * Exceptions that occur when unlocking a monitor need to appear as
+     * if they happened at the following instruction.  See the Dalvik
+     * instruction spec.
+     */
+    /* monitor-exit vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    EXPORT_PC()                         @ before fetch: export the PC
+    GET_VREG(r1, r2)                    @ r1<- vAA (object)
+    cmp     r1, #0                      @ null object?
+    beq     common_errNullObject        @ yes
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- glue->self
+    bl      dvmUnlockObject             @ r0<- success for unlock(self, obj)
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yes, exception is pending
+    FETCH_ADVANCE_INST(1)               @ before throw: advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CHECK_CAST: /* 0x1f */
+/* File: armv5te/OP_CHECK_CAST.S */
+    /*
+     * Check to see if a cast from one class to another is allowed.
+     */
+    /* check-cast vAA, class@BBBB */
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH(r2, 1)                        @ r2<- BBBB
+    GET_VREG(r9, r3)                    @ r9<- object
+    ldr     r0, [rGLUE, #offGlue_methodClassDex]    @ r0<- pDvmDex
+    cmp     r9, #0                      @ is object null?
+    ldr     r0, [r0, #offDvmDex_pResClasses]    @ r0<- pDvmDex->pResClasses
+    beq     .LOP_CHECK_CAST_okay            @ null obj, cast always succeeds
+    ldr     r1, [r0, r2, lsl #2]        @ r1<- resolved class
+    ldr     r0, [r9, #offObject_clazz]  @ r0<- obj->clazz
+    cmp     r1, #0                      @ have we resolved this before?
+    beq     .LOP_CHECK_CAST_resolve         @ not resolved, do it now
+.LOP_CHECK_CAST_resolved:
+    cmp     r0, r1                      @ same class (trivial success)?
+    bne     .LOP_CHECK_CAST_fullcheck       @ no, do full check
+.LOP_CHECK_CAST_okay:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INSTANCE_OF: /* 0x20 */
+/* File: armv5te/OP_INSTANCE_OF.S */
+    /*
+     * Check to see if an object reference is an instance of a class.
+     *
+     * Most common situation is a non-null object, being compared against
+     * an already-resolved class.
+     */
+    /* instance-of vA, vB, class@CCCC */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB (object)
+    and     r9, r9, #15                 @ r9<- A
+    cmp     r0, #0                      @ is object null?
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- pDvmDex
+    beq     .LOP_INSTANCE_OF_store           @ null obj, not an instance, store r0
+    FETCH(r3, 1)                        @ r3<- CCCC
+    ldr     r2, [r2, #offDvmDex_pResClasses]    @ r2<- pDvmDex->pResClasses
+    ldr     r1, [r2, r3, lsl #2]        @ r1<- resolved class
+    ldr     r0, [r0, #offObject_clazz]  @ r0<- obj->clazz
+    cmp     r1, #0                      @ have we resolved this before?
+    beq     .LOP_INSTANCE_OF_resolve         @ not resolved, do it now
+.LOP_INSTANCE_OF_resolved: @ r0=obj->clazz, r1=resolved class
+    cmp     r0, r1                      @ same class (trivial success)?
+    beq     .LOP_INSTANCE_OF_trivial         @ yes, trivial finish
+    b       .LOP_INSTANCE_OF_fullcheck       @ no, do full check
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ARRAY_LENGTH: /* 0x21 */
+/* File: armv5te/OP_ARRAY_LENGTH.S */
+    /*
+     * Return the length of an array.
+     */
+    mov     r1, rINST, lsr #12          @ r1<- B
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    GET_VREG(r0, r1)                    @ r0<- vB (object ref)
+    and     r2, r2, #15                 @ r2<- A
+    cmp     r0, #0                      @ is object null?
+    beq     common_errNullObject        @ yup, fail
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- array length
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r3, r2)                    @ vB<- length
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEW_INSTANCE: /* 0x22 */
+/* File: armv5te/OP_NEW_INSTANCE.S */
+    /*
+     * Create a new instance of a class.
+     */
+    /* new-instance vAA, class@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResClasses]    @ r3<- pDvmDex->pResClasses
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved class
+    EXPORT_PC()                         @ req'd for init, resolve, alloc
+    cmp     r0, #0                      @ already resolved?
+    beq     .LOP_NEW_INSTANCE_resolve         @ no, resolve it now
+.LOP_NEW_INSTANCE_resolved:   @ r0=class
+    ldrb    r1, [r0, #offClassObject_status]    @ r1<- ClassStatus enum
+    cmp     r1, #CLASS_INITIALIZED      @ has class been initialized?
+    bne     .LOP_NEW_INSTANCE_needinit        @ no, init class now
+.LOP_NEW_INSTANCE_initialized: @ r0=class
+    ldr     r3, [r0, #offClassObject_accessFlags]   @ r3<- clazz->accessFlags
+    tst     r3, #(ACC_INTERFACE|ACC_ABSTRACT)   @ abstract or interface?
+    mov     r1, #ALLOC_DONT_TRACK       @ flags for alloc call
+    beq     .LOP_NEW_INSTANCE_finish          @ concrete class, continue
+    b       .LOP_NEW_INSTANCE_abstract        @ fail
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEW_ARRAY: /* 0x23 */
+/* File: armv5te/OP_NEW_ARRAY.S */
+    /*
+     * Allocate an array of objects, specified with the array class
+     * and a count.
+     *
+     * The verifier guarantees that this is an array class, so we don't
+     * check for it here.
+     */
+    /* new-array vA, vB, class@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    FETCH(r2, 1)                        @ r2<- CCCC
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    GET_VREG(r1, r0)                    @ r1<- vB (array length)
+    ldr     r3, [r3, #offDvmDex_pResClasses]    @ r3<- pDvmDex->pResClasses
+    cmp     r1, #0                      @ check length
+    ldr     r0, [r3, r2, lsl #2]        @ r0<- resolved class
+    bmi     common_errNegativeArraySize @ negative length, bail
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ req'd for resolve, alloc
+    bne     .LOP_NEW_ARRAY_finish          @ resolved, continue
+    b       .LOP_NEW_ARRAY_resolve         @ do resolve now
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FILLED_NEW_ARRAY: /* 0x24 */
+/* File: armv5te/OP_FILLED_NEW_ARRAY.S */
+    /*
+     * Create a new array with elements filled from registers.
+     *
+     * for: filled-new-array, filled-new-array/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, type@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResClasses]    @ r3<- pDvmDex->pResClasses
+    EXPORT_PC()                         @ need for resolve and alloc
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved class
+    mov     r10, rINST, lsr #8          @ r10<- AA or BA
+    cmp     r0, #0                      @ already resolved?
+    bne     .LOP_FILLED_NEW_ARRAY_continue        @ yes, continue on
+8:  ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    mov     r2, #0                      @ r2<- false
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- call(clazz, ref)
+    cmp     r0, #0                      @ got null?
+    beq     common_exceptionThrown      @ yes, handle exception
+    b       .LOP_FILLED_NEW_ARRAY_continue
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FILLED_NEW_ARRAY_RANGE: /* 0x25 */
+/* File: armv5te/OP_FILLED_NEW_ARRAY_RANGE.S */
+/* File: armv5te/OP_FILLED_NEW_ARRAY.S */
+    /*
+     * Create a new array with elements filled from registers.
+     *
+     * for: filled-new-array, filled-new-array/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, type@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResClasses]    @ r3<- pDvmDex->pResClasses
+    EXPORT_PC()                         @ need for resolve and alloc
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved class
+    mov     r10, rINST, lsr #8          @ r10<- AA or BA
+    cmp     r0, #0                      @ already resolved?
+    bne     .LOP_FILLED_NEW_ARRAY_RANGE_continue        @ yes, continue on
+8:  ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    mov     r2, #0                      @ r2<- false
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- call(clazz, ref)
+    cmp     r0, #0                      @ got null?
+    beq     common_exceptionThrown      @ yes, handle exception
+    b       .LOP_FILLED_NEW_ARRAY_RANGE_continue
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FILL_ARRAY_DATA: /* 0x26 */
+/* File: armv5te/OP_FILL_ARRAY_DATA.S */
+    /* fill-array-data vAA, +BBBBBBBB */
+    FETCH(r0, 1)                        @ r0<- bbbb (lo)
+    FETCH(r1, 2)                        @ r1<- BBBB (hi)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    orr     r1, r0, r1, lsl #16         @ r1<- BBBBbbbb
+    GET_VREG(r0, r3)                    @ r0<- vAA (array object)
+    add     r1, rPC, r1, lsl #1         @ r1<- PC + BBBBbbbb*2 (array data off.)
+    EXPORT_PC();
+    bl      dvmInterpHandleFillArrayData@ fill the array with predefined data
+    cmp     r0, #0                      @ 0 means an exception is thrown
+    beq     common_exceptionThrown      @ has exception
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_THROW: /* 0x27 */
+/* File: armv5te/OP_THROW.S */
+    /*
+     * Throw an exception object in the current thread.
+     */
+    /* throw vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    GET_VREG(r1, r2)                    @ r1<- vAA (exception object)
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- glue->self
+    cmp     r1, #0                      @ null object?
+    beq     common_errNullObject        @ yes, throw an NPE instead
+    @ bypass dvmSetException, just store it
+    str     r1, [r0, #offThread_exception]  @ thread->exception<- obj
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_GOTO: /* 0x28 */
+/* File: armv5te/OP_GOTO.S */
+    /*
+     * Unconditional branch, 8-bit offset.
+     *
+     * The branch distance is a signed code-unit offset, which we need to
+     * double to get a byte offset.
+     */
+    /* goto +AA */
+    mov     r0, rINST, lsl #16          @ r0<- AAxx0000
+    movs    r9, r0, asr #24             @ r9<- ssssssAA (sign-extended)
+    mov     r9, r9, lsl #1              @ r9<- byte offset
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+    FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_GOTO_16: /* 0x29 */
+/* File: armv5te/OP_GOTO_16.S */
+    /*
+     * Unconditional branch, 16-bit offset.
+     *
+     * The branch distance is a signed code-unit offset, which we need to
+     * double to get a byte offset.
+     */
+    /* goto/16 +AAAA */
+    FETCH_S(r0, 1)                      @ r0<- ssssAAAA (sign-extended)
+    movs    r9, r0, asl #1              @ r9<- byte offset, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+    FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_GOTO_32: /* 0x2a */
+/* File: armv5te/OP_GOTO_32.S */
+    /*
+     * Unconditional branch, 32-bit offset.
+     *
+     * The branch distance is a signed code-unit offset, which we need to
+     * double to get a byte offset.
+     *
+     * Unlike most opcodes, this one is allowed to branch to itself, so
+     * our "backward branch" test must be "<=0" instead of "<0".  The ORRS
+     * instruction doesn't affect the V flag, so we need to clear it
+     * explicitly.
+     */
+    /* goto/32 +AAAAAAAA */
+    FETCH(r0, 1)                        @ r0<- aaaa (lo)
+    FETCH(r1, 2)                        @ r1<- AAAA (hi)
+    cmp     ip, ip                      @ (clear V flag during stall)
+    orrs    r0, r0, r1, lsl #16         @ r0<- AAAAaaaa, check sign
+    mov     r9, r0, asl #1              @ r9<- byte offset
+    ble     common_backwardBranch       @ backward branch, do periodic checks
+    FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_PACKED_SWITCH: /* 0x2b */
+/* File: armv5te/OP_PACKED_SWITCH.S */
+    /*
+     * Handle a packed-switch or sparse-switch instruction.  In both cases
+     * we decode it and hand it off to a helper function.
+     *
+     * We don't really expect backward branches in a switch statement, but
+     * they're perfectly legal, so we check for them here.
+     *
+     * for: packed-switch, sparse-switch
+     */
+    /* op vAA, +BBBB */
+    FETCH(r0, 1)                        @ r0<- bbbb (lo)
+    FETCH(r1, 2)                        @ r1<- BBBB (hi)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    orr     r0, r0, r1, lsl #16         @ r0<- BBBBbbbb
+    GET_VREG(r1, r3)                    @ r1<- vAA
+    add     r0, rPC, r0, lsl #1         @ r0<- PC + BBBBbbbb*2
+    bl      dvmInterpHandlePackedSwitch                       @ r0<- code-unit branch offset
+    movs    r9, r0, asl #1              @ r9<- branch byte offset, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+    beq     common_backwardBranch       @ (want to use BLE but V is unknown)
+    FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPARSE_SWITCH: /* 0x2c */
+/* File: armv5te/OP_SPARSE_SWITCH.S */
+/* File: armv5te/OP_PACKED_SWITCH.S */
+    /*
+     * Handle a packed-switch or sparse-switch instruction.  In both cases
+     * we decode it and hand it off to a helper function.
+     *
+     * We don't really expect backward branches in a switch statement, but
+     * they're perfectly legal, so we check for them here.
+     *
+     * for: packed-switch, sparse-switch
+     */
+    /* op vAA, +BBBB */
+    FETCH(r0, 1)                        @ r0<- bbbb (lo)
+    FETCH(r1, 2)                        @ r1<- BBBB (hi)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    orr     r0, r0, r1, lsl #16         @ r0<- BBBBbbbb
+    GET_VREG(r1, r3)                    @ r1<- vAA
+    add     r0, rPC, r0, lsl #1         @ r0<- PC + BBBBbbbb*2
+    bl      dvmInterpHandleSparseSwitch                       @ r0<- code-unit branch offset
+    movs    r9, r0, asl #1              @ r9<- branch byte offset, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+    beq     common_backwardBranch       @ (want to use BLE but V is unknown)
+    FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMPL_FLOAT: /* 0x2d */
+/* File: armv5te/OP_CMPL_FLOAT.S */
+    /*
+     * Compare two floating-point values.  Puts 0, 1, or -1 into the
+     * destination register based on the results of the comparison.
+     *
+     * Provide a "naninst" instruction that puts 1 or -1 into r1 depending
+     * on what value we'd like to return when one of the operands is NaN.
+     *
+     * The operation we're implementing is:
+     *   if (x == y)
+     *     return 0;
+     *   else if (x < y)
+     *     return -1;
+     *   else if (x > y)
+     *     return 1;
+     *   else
+     *     return {-1,1};  // one or both operands was NaN
+     *
+     * The straightforward implementation requires 3 calls to functions
+     * that return a result in r0.  We can do it with two calls if our
+     * EABI library supports __aeabi_cfcmple (only one if we want to check
+     * for NaN directly):
+     *   check x <= y
+     *     if <, return -1
+     *     if ==, return 0
+     *   check y <= x
+     *     if <, return 1
+     *   return {-1,1}
+     *
+     * for: cmpl-float, cmpg-float
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    GET_VREG(r9, r2)                    @ r9<- vBB
+    GET_VREG(r10, r3)                   @ r10<- vCC
+    mov     r0, r9                      @ copy to arg registers
+    mov     r1, r10
+    bl      __aeabi_cfcmple             @ cmp <=: C clear if <, Z set if eq
+    bhi     .LOP_CMPL_FLOAT_gt_or_nan       @ C set and Z clear, disambiguate
+    mvncc   r1, #0                      @ (less than) r1<- -1
+    moveq   r1, #0                      @ (equal) r1<- 0, trumps less than
+.LOP_CMPL_FLOAT_finish:
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r3)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMPG_FLOAT: /* 0x2e */
+/* File: armv5te/OP_CMPG_FLOAT.S */
+/* File: armv5te/OP_CMPL_FLOAT.S */
+    /*
+     * Compare two floating-point values.  Puts 0, 1, or -1 into the
+     * destination register based on the results of the comparison.
+     *
+     * Provide a "naninst" instruction that puts 1 or -1 into r1 depending
+     * on what value we'd like to return when one of the operands is NaN.
+     *
+     * The operation we're implementing is:
+     *   if (x == y)
+     *     return 0;
+     *   else if (x < y)
+     *     return -1;
+     *   else if (x > y)
+     *     return 1;
+     *   else
+     *     return {-1,1};  // one or both operands was NaN
+     *
+     * The straightforward implementation requires 3 calls to functions
+     * that return a result in r0.  We can do it with two calls if our
+     * EABI library supports __aeabi_cfcmple (only one if we want to check
+     * for NaN directly):
+     *   check x <= y
+     *     if <, return -1
+     *     if ==, return 0
+     *   check y <= x
+     *     if <, return 1
+     *   return {-1,1}
+     *
+     * for: cmpl-float, cmpg-float
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    GET_VREG(r9, r2)                    @ r9<- vBB
+    GET_VREG(r10, r3)                   @ r10<- vCC
+    mov     r0, r9                      @ copy to arg registers
+    mov     r1, r10
+    bl      __aeabi_cfcmple             @ cmp <=: C clear if <, Z set if eq
+    bhi     .LOP_CMPG_FLOAT_gt_or_nan       @ C set and Z clear, disambiguate
+    mvncc   r1, #0                      @ (less than) r1<- -1
+    moveq   r1, #0                      @ (equal) r1<- 0, trumps less than
+.LOP_CMPG_FLOAT_finish:
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r3)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMPL_DOUBLE: /* 0x2f */
+/* File: armv5te/OP_CMPL_DOUBLE.S */
+    /*
+     * Compare two floating-point values.  Puts 0, 1, or -1 into the
+     * destination register based on the results of the comparison.
+     *
+     * Provide a "naninst" instruction that puts 1 or -1 into r1 depending
+     * on what value we'd like to return when one of the operands is NaN.
+     *
+     * See OP_CMPL_FLOAT for an explanation.
+     *
+     * For: cmpl-double, cmpg-double
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r9, r0, #255                @ r9<- BB
+    mov     r10, r0, lsr #8             @ r10<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[BB]
+    add     r10, rFP, r10, lsl #2       @ r10<- &fp[CC]
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r10, {r2-r3}                @ r2/r3<- vCC/vCC+1
+    bl      __aeabi_cdcmple             @ cmp <=: C clear if <, Z set if eq
+    bhi     .LOP_CMPL_DOUBLE_gt_or_nan       @ C set and Z clear, disambiguate
+    mvncc   r1, #0                      @ (less than) r1<- -1
+    moveq   r1, #0                      @ (equal) r1<- 0, trumps less than
+.LOP_CMPL_DOUBLE_finish:
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r3)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMPG_DOUBLE: /* 0x30 */
+/* File: armv5te/OP_CMPG_DOUBLE.S */
+/* File: armv5te/OP_CMPL_DOUBLE.S */
+    /*
+     * Compare two floating-point values.  Puts 0, 1, or -1 into the
+     * destination register based on the results of the comparison.
+     *
+     * Provide a "naninst" instruction that puts 1 or -1 into r1 depending
+     * on what value we'd like to return when one of the operands is NaN.
+     *
+     * See OP_CMPL_FLOAT for an explanation.
+     *
+     * For: cmpl-double, cmpg-double
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r9, r0, #255                @ r9<- BB
+    mov     r10, r0, lsr #8             @ r10<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[BB]
+    add     r10, rFP, r10, lsl #2       @ r10<- &fp[CC]
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r10, {r2-r3}                @ r2/r3<- vCC/vCC+1
+    bl      __aeabi_cdcmple             @ cmp <=: C clear if <, Z set if eq
+    bhi     .LOP_CMPG_DOUBLE_gt_or_nan       @ C set and Z clear, disambiguate
+    mvncc   r1, #0                      @ (less than) r1<- -1
+    moveq   r1, #0                      @ (equal) r1<- 0, trumps less than
+.LOP_CMPG_DOUBLE_finish:
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r3)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMP_LONG: /* 0x31 */
+/* File: armv5te/OP_CMP_LONG.S */
+    /*
+     * Compare two 64-bit values.  Puts 0, 1, or -1 into the destination
+     * register based on the results of the comparison.
+     *
+     * We load the full values with LDM, but in practice many values could
+     * be resolved by only looking at the high word.  This could be made
+     * faster or slower by splitting the LDM into a pair of LDRs.
+     *
+     * If we just wanted to set condition flags, we could do this:
+     *  subs    ip, r0, r2
+     *  sbcs    ip, r1, r3
+     *  subeqs  ip, r0, r2
+     * Leaving { <0, 0, >0 } in ip.  However, we have to set it to a specific
+     * integer value, which we can do with 2 conditional mov/mvn instructions
+     * (set 1, set -1; if they're equal we already have 0 in ip), giving
+     * us a constant 5-cycle path plus a branch at the end to the
+     * instruction epilogue code.  The multi-compare approach below needs
+     * 2 or 3 cycles + branch if the high word doesn't match, 6 + branch
+     * in the worst case (the 64-bit values are equal).
+     */
+    /* cmp-long vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    cmp     r1, r3                      @ compare (vBB+1, vCC+1)
+    blt     .LOP_CMP_LONG_less            @ signed compare on high part
+    bgt     .LOP_CMP_LONG_greater
+    subs    r1, r0, r2                  @ r1<- r0 - r2
+    bhi     .LOP_CMP_LONG_greater         @ unsigned compare on low part
+    bne     .LOP_CMP_LONG_less
+    b       .LOP_CMP_LONG_finish          @ equal; r1 already holds 0
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_EQ: /* 0x32 */
+/* File: armv5te/OP_IF_EQ.S */
+/* File: armv5te/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r3, r1)                    @ r3<- vB
+    GET_VREG(r2, r0)                    @ r2<- vA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, r3                      @ compare (vA, vB)
+    bne  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ yes, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_NE: /* 0x33 */
+/* File: armv5te/OP_IF_NE.S */
+/* File: armv5te/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r3, r1)                    @ r3<- vB
+    GET_VREG(r2, r0)                    @ r2<- vA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, r3                      @ compare (vA, vB)
+    beq  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ yes, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_LT: /* 0x34 */
+/* File: armv5te/OP_IF_LT.S */
+/* File: armv5te/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r3, r1)                    @ r3<- vB
+    GET_VREG(r2, r0)                    @ r2<- vA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, r3                      @ compare (vA, vB)
+    bge  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ yes, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_GE: /* 0x35 */
+/* File: armv5te/OP_IF_GE.S */
+/* File: armv5te/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r3, r1)                    @ r3<- vB
+    GET_VREG(r2, r0)                    @ r2<- vA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, r3                      @ compare (vA, vB)
+    blt  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ yes, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_GT: /* 0x36 */
+/* File: armv5te/OP_IF_GT.S */
+/* File: armv5te/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r3, r1)                    @ r3<- vB
+    GET_VREG(r2, r0)                    @ r2<- vA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, r3                      @ compare (vA, vB)
+    ble  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ yes, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_LE: /* 0x37 */
+/* File: armv5te/OP_IF_LE.S */
+/* File: armv5te/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r3, r1)                    @ r3<- vB
+    GET_VREG(r2, r0)                    @ r2<- vA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, r3                      @ compare (vA, vB)
+    bgt  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ yes, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_EQZ: /* 0x38 */
+/* File: armv5te/OP_IF_EQZ.S */
+/* File: armv5te/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    bne  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_NEZ: /* 0x39 */
+/* File: armv5te/OP_IF_NEZ.S */
+/* File: armv5te/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    beq  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_LTZ: /* 0x3a */
+/* File: armv5te/OP_IF_LTZ.S */
+/* File: armv5te/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    bge  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_GEZ: /* 0x3b */
+/* File: armv5te/OP_IF_GEZ.S */
+/* File: armv5te/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    blt  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_GTZ: /* 0x3c */
+/* File: armv5te/OP_IF_GTZ.S */
+/* File: armv5te/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    ble  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_LEZ: /* 0x3d */
+/* File: armv5te/OP_IF_LEZ.S */
+/* File: armv5te/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    bgt  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_3E: /* 0x3e */
+/* File: armv5te/OP_UNUSED_3E.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_3F: /* 0x3f */
+/* File: armv5te/OP_UNUSED_3F.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_40: /* 0x40 */
+/* File: armv5te/OP_UNUSED_40.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_41: /* 0x41 */
+/* File: armv5te/OP_UNUSED_41.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_42: /* 0x42 */
+/* File: armv5te/OP_UNUSED_42.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_43: /* 0x43 */
+/* File: armv5te/OP_UNUSED_43.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET: /* 0x44 */
+/* File: armv5te/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #2     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldr   r2, [r0, #offArrayObject_contents]  @ r2<- vBB[vCC]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r9)                    @ vAA<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_WIDE: /* 0x45 */
+/* File: armv5te/OP_AGET_WIDE.S */
+    /*
+     * Array get, 64 bits.  vAA <- vBB[vCC].
+     *
+     * Arrays of long/double are 64-bit aligned, so it's okay to use LDRD.
+     */
+    /* aget-wide vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #3          @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcc     .LOP_AGET_WIDE_finish          @ okay, continue below
+    b       common_errArrayIndex        @ index >= length, bail
+    @ May want to swap the order of these two branches depending on how the
+    @ branch prediction (if any) handles conditional forward branches vs.
+    @ unconditional forward branches.
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_OBJECT: /* 0x46 */
+/* File: armv5te/OP_AGET_OBJECT.S */
+/* File: armv5te/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #2     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldr   r2, [r0, #offArrayObject_contents]  @ r2<- vBB[vCC]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r9)                    @ vAA<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_BOOLEAN: /* 0x47 */
+/* File: armv5te/OP_AGET_BOOLEAN.S */
+/* File: armv5te/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #0     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldrb   r2, [r0, #offArrayObject_contents]  @ r2<- vBB[vCC]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r9)                    @ vAA<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_BYTE: /* 0x48 */
+/* File: armv5te/OP_AGET_BYTE.S */
+/* File: armv5te/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #0     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldrsb   r2, [r0, #offArrayObject_contents]  @ r2<- vBB[vCC]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r9)                    @ vAA<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_CHAR: /* 0x49 */
+/* File: armv5te/OP_AGET_CHAR.S */
+/* File: armv5te/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #1     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldrh   r2, [r0, #offArrayObject_contents]  @ r2<- vBB[vCC]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r9)                    @ vAA<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_SHORT: /* 0x4a */
+/* File: armv5te/OP_AGET_SHORT.S */
+/* File: armv5te/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #1     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldrsh   r2, [r0, #offArrayObject_contents]  @ r2<- vBB[vCC]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r9)                    @ vAA<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT: /* 0x4b */
+/* File: armv5te/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA.
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aput, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #2     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r9)                    @ r2<- vAA
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r2, [r0, #offArrayObject_contents]  @ vBB[vCC]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_WIDE: /* 0x4c */
+/* File: armv5te/OP_APUT_WIDE.S */
+    /*
+     * Array put, 64 bits.  vBB[vCC] <- vAA.
+     *
+     * Arrays of long/double are 64-bit aligned, so it's okay to use STRD.
+     */
+    /* aput-wide vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #3          @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    bcc     .LOP_APUT_WIDE_finish          @ okay, continue below
+    b       common_errArrayIndex        @ index >= length, bail
+    @ May want to swap the order of these two branches depending on how the
+    @ branch prediction (if any) handles conditional forward branches vs.
+    @ unconditional forward branches.
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_OBJECT: /* 0x4d */
+/* File: armv5te/OP_APUT_OBJECT.S */
+    /*
+     * Store an object into an array.  vBB[vCC] <- vAA.
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    GET_VREG(r1, r2)                    @ r1<- vBB (array object)
+    GET_VREG(r0, r3)                    @ r0<- vCC (requested index)
+    cmp     r1, #0                      @ null array object?
+    GET_VREG(r9, r9)                    @ r9<- vAA
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r1, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r10, r1, r0, lsl #2         @ r10<- arrayObj + index*width
+    cmp     r0, r3                      @ compare unsigned index, length
+    bcc     .LOP_APUT_OBJECT_finish          @ we're okay, continue on
+    b       common_errArrayIndex        @ index >= length, bail
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_BOOLEAN: /* 0x4e */
+/* File: armv5te/OP_APUT_BOOLEAN.S */
+/* File: armv5te/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA.
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aput, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #0     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r9)                    @ r2<- vAA
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strb  r2, [r0, #offArrayObject_contents]  @ vBB[vCC]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_BYTE: /* 0x4f */
+/* File: armv5te/OP_APUT_BYTE.S */
+/* File: armv5te/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA.
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aput, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #0     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r9)                    @ r2<- vAA
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strb  r2, [r0, #offArrayObject_contents]  @ vBB[vCC]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_CHAR: /* 0x50 */
+/* File: armv5te/OP_APUT_CHAR.S */
+/* File: armv5te/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA.
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aput, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #1     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r9)                    @ r2<- vAA
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strh  r2, [r0, #offArrayObject_contents]  @ vBB[vCC]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_SHORT: /* 0x51 */
+/* File: armv5te/OP_APUT_SHORT.S */
+/* File: armv5te/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA.
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aput, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #1     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r9)                    @ r2<- vAA
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strh  r2, [r0, #offArrayObject_contents]  @ vBB[vCC]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET: /* 0x52 */
+/* File: armv5te/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_finish
+    b       common_exceptionThrown
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_WIDE: /* 0x53 */
+/* File: armv5te/OP_IGET_WIDE.S */
+    /*
+     * Wide 32-bit instance field get.
+     */
+    /* iget-wide vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_WIDE_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method] @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_WIDE_finish
+    b       common_exceptionThrown
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_OBJECT: /* 0x54 */
+/* File: armv5te/OP_IGET_OBJECT.S */
+/* File: armv5te/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_OBJECT_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_OBJECT_finish
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_BOOLEAN: /* 0x55 */
+/* File: armv5te/OP_IGET_BOOLEAN.S */
+@include "armv5te/OP_IGET.S" { "load":"ldrb", "sqnum":"1" }
+/* File: armv5te/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_BOOLEAN_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_BOOLEAN_finish
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_BYTE: /* 0x56 */
+/* File: armv5te/OP_IGET_BYTE.S */
+@include "armv5te/OP_IGET.S" { "load":"ldrsb", "sqnum":"2" }
+/* File: armv5te/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_BYTE_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_BYTE_finish
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_CHAR: /* 0x57 */
+/* File: armv5te/OP_IGET_CHAR.S */
+@include "armv5te/OP_IGET.S" { "load":"ldrh", "sqnum":"3" }
+/* File: armv5te/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_CHAR_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_CHAR_finish
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_SHORT: /* 0x58 */
+/* File: armv5te/OP_IGET_SHORT.S */
+@include "armv5te/OP_IGET.S" { "load":"ldrsh", "sqnum":"4" }
+/* File: armv5te/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_SHORT_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_SHORT_finish
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT: /* 0x59 */
+/* File: armv5te/OP_IPUT.S */
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_WIDE: /* 0x5a */
+/* File: armv5te/OP_IPUT_WIDE.S */
+    /* iput-wide vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_WIDE_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method] @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_WIDE_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_OBJECT: /* 0x5b */
+/* File: armv5te/OP_IPUT_OBJECT.S */
+/* File: armv5te/OP_IPUT.S */
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_OBJECT_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_OBJECT_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_BOOLEAN: /* 0x5c */
+/* File: armv5te/OP_IPUT_BOOLEAN.S */
+@include "armv5te/OP_IPUT.S" { "store":"strb", "sqnum":"1" }
+/* File: armv5te/OP_IPUT.S */
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_BOOLEAN_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_BOOLEAN_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_BYTE: /* 0x5d */
+/* File: armv5te/OP_IPUT_BYTE.S */
+@include "armv5te/OP_IPUT.S" { "store":"strb", "sqnum":"2" }
+/* File: armv5te/OP_IPUT.S */
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_BYTE_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_BYTE_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_CHAR: /* 0x5e */
+/* File: armv5te/OP_IPUT_CHAR.S */
+@include "armv5te/OP_IPUT.S" { "store":"strh", "sqnum":"3" }
+/* File: armv5te/OP_IPUT.S */
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_CHAR_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_CHAR_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_SHORT: /* 0x5f */
+/* File: armv5te/OP_IPUT_SHORT.S */
+@include "armv5te/OP_IPUT.S" { "store":"strh", "sqnum":"4" }
+/* File: armv5te/OP_IPUT.S */
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_SHORT_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_SHORT_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET: /* 0x60 */
+/* File: armv5te/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_resolve         @ yes, do resolve
+.LOP_SGET_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_WIDE: /* 0x61 */
+/* File: armv5te/OP_SGET_WIDE.S */
+    /*
+     * 64-bit SGET handler.
+     */
+    /* sget-wide vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_WIDE_resolve         @ yes, do resolve
+.LOP_SGET_WIDE_finish:
+    mov     r1, rINST, lsr #8           @ r1<- AA
+    ldrd    r2, [r0, #offStaticField_value] @ r2/r3<- field value (aligned)
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[AA]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    stmia   r1, {r2-r3}                 @ vAA/vAA+1<- r2/r3
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_OBJECT: /* 0x62 */
+/* File: armv5te/OP_SGET_OBJECT.S */
+/* File: armv5te/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_OBJECT_resolve         @ yes, do resolve
+.LOP_SGET_OBJECT_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_BOOLEAN: /* 0x63 */
+/* File: armv5te/OP_SGET_BOOLEAN.S */
+/* File: armv5te/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_BOOLEAN_resolve         @ yes, do resolve
+.LOP_SGET_BOOLEAN_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_BYTE: /* 0x64 */
+/* File: armv5te/OP_SGET_BYTE.S */
+/* File: armv5te/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_BYTE_resolve         @ yes, do resolve
+.LOP_SGET_BYTE_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_CHAR: /* 0x65 */
+/* File: armv5te/OP_SGET_CHAR.S */
+/* File: armv5te/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_CHAR_resolve         @ yes, do resolve
+.LOP_SGET_CHAR_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_SHORT: /* 0x66 */
+/* File: armv5te/OP_SGET_SHORT.S */
+/* File: armv5te/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_SHORT_resolve         @ yes, do resolve
+.LOP_SGET_SHORT_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT: /* 0x67 */
+/* File: armv5te/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_resolve         @ yes, do resolve
+.LOP_SPUT_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_WIDE: /* 0x68 */
+/* File: armv5te/OP_SPUT_WIDE.S */
+    /*
+     * 64-bit SPUT handler.
+     */
+    /* sput-wide vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_WIDE_resolve         @ yes, do resolve
+.LOP_SPUT_WIDE_finish: @ field ptr in r0, AA in r9
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldmia   r9, {r2-r3}                 @ r2/r3<- vAA/vAA+1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strd    r2, [r0, #offStaticField_value] @ field<- vAA/vAA+1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_OBJECT: /* 0x69 */
+/* File: armv5te/OP_SPUT_OBJECT.S */
+/* File: armv5te/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_OBJECT_resolve         @ yes, do resolve
+.LOP_SPUT_OBJECT_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_BOOLEAN: /* 0x6a */
+/* File: armv5te/OP_SPUT_BOOLEAN.S */
+/* File: armv5te/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_BOOLEAN_resolve         @ yes, do resolve
+.LOP_SPUT_BOOLEAN_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_BYTE: /* 0x6b */
+/* File: armv5te/OP_SPUT_BYTE.S */
+/* File: armv5te/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_BYTE_resolve         @ yes, do resolve
+.LOP_SPUT_BYTE_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_CHAR: /* 0x6c */
+/* File: armv5te/OP_SPUT_CHAR.S */
+/* File: armv5te/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_CHAR_resolve         @ yes, do resolve
+.LOP_SPUT_CHAR_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_SHORT: /* 0x6d */
+/* File: armv5te/OP_SPUT_SHORT.S */
+/* File: armv5te/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_SHORT_resolve         @ yes, do resolve
+.LOP_SPUT_SHORT_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_VIRTUAL: /* 0x6e */
+/* File: armv5te/OP_INVOKE_VIRTUAL.S */
+    /*
+     * Handle a virtual method call.
+     *
+     * for: invoke-virtual, invoke-virtual/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved baseMethod
+    .if     (!0)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    bne     .LOP_INVOKE_VIRTUAL_continue        @ yes, continue on
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    mov     r2, #METHOD_VIRTUAL         @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    bne     .LOP_INVOKE_VIRTUAL_continue        @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_SUPER: /* 0x6f */
+/* File: armv5te/OP_INVOKE_SUPER.S */
+    /*
+     * Handle a "super" method call.
+     *
+     * for: invoke-super, invoke-super/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    .if     (!0)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    GET_VREG(r2, r10)                   @ r2<- "this" ptr
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved baseMethod
+    cmp     r2, #0                      @ null "this"?
+    ldr     r9, [rGLUE, #offGlue_method] @ r9<- current method
+    beq     common_errNullObject        @ null "this", throw exception
+    cmp     r0, #0                      @ already resolved?
+    ldr     r9, [r9, #offMethod_clazz]  @ r9<- method->clazz
+    EXPORT_PC()                         @ must export for invoke
+    bne     .LOP_INVOKE_SUPER_continue        @ resolved, continue on
+    b       .LOP_INVOKE_SUPER_resolve         @ do resolve now
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_DIRECT: /* 0x70 */
+/* File: armv5te/OP_INVOKE_DIRECT.S */
+    /*
+     * Handle a direct method call.
+     *
+     * (We could defer the "is 'this' pointer null" test to the common
+     * method invocation code, and use a flag to indicate that static
+     * calls don't count.  If we do this as part of copying the arguments
+     * out we could avoiding loading the first arg twice.)
+     *
+     * for: invoke-direct, invoke-direct/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved methodToCall
+    .if     (!0)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    GET_VREG(r2, r10)                   @ r2<- "this" ptr
+    beq     .LOP_INVOKE_DIRECT_resolve         @ not resolved, do it now
+.LOP_INVOKE_DIRECT_finish:
+    cmp     r2, #0                      @ null "this" ref?
+    bne     common_invokeMethodNoRange   @ no, continue on
+    b       common_errNullObject        @ yes, throw exception
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_STATIC: /* 0x71 */
+/* File: armv5te/OP_INVOKE_STATIC.S */
+    /*
+     * Handle a static method call.
+     *
+     * for: invoke-static, invoke-static/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved methodToCall
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    bne     common_invokeMethodNoRange @ yes, continue on
+0:  ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    mov     r2, #METHOD_STATIC          @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    bne     common_invokeMethodNoRange @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_INTERFACE: /* 0x72 */
+/* File: armv5te/OP_INVOKE_INTERFACE.S */
+    /*
+     * Handle an interface method call.
+     *
+     * for: invoke-interface, invoke-interface/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r2, 2)                        @ r2<- FEDC or CCCC
+    FETCH(r1, 1)                        @ r1<- BBBB
+    .if     (!0)
+    and     r2, r2, #15                 @ r2<- C (or stays CCCC)
+    .endif
+    EXPORT_PC()                         @ must export for invoke
+    GET_VREG(r0, r2)                    @ r0<- first arg ("this")
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- methodClassDex
+    cmp     r0, #0                      @ null obj?
+    ldr     r2, [rGLUE, #offGlue_method]  @ r2<- method
+    beq     common_errNullObject        @ yes, fail
+    ldr     r0, [r0, #offObject_clazz]  @ r0<- thisPtr->clazz
+    bl      dvmFindInterfaceMethodInCache @ r0<- call(class, ref, method, dex)
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yes, handle exception
+    b       common_invokeMethodNoRange @ jump to common handler 
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_73: /* 0x73 */
+/* File: armv5te/OP_UNUSED_73.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_VIRTUAL_RANGE: /* 0x74 */
+/* File: armv5te/OP_INVOKE_VIRTUAL_RANGE.S */
+/* File: armv5te/OP_INVOKE_VIRTUAL.S */
+    /*
+     * Handle a virtual method call.
+     *
+     * for: invoke-virtual, invoke-virtual/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved baseMethod
+    .if     (!1)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    bne     .LOP_INVOKE_VIRTUAL_RANGE_continue        @ yes, continue on
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    mov     r2, #METHOD_VIRTUAL         @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    bne     .LOP_INVOKE_VIRTUAL_RANGE_continue        @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_SUPER_RANGE: /* 0x75 */
+/* File: armv5te/OP_INVOKE_SUPER_RANGE.S */
+/* File: armv5te/OP_INVOKE_SUPER.S */
+    /*
+     * Handle a "super" method call.
+     *
+     * for: invoke-super, invoke-super/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    .if     (!1)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    GET_VREG(r2, r10)                   @ r2<- "this" ptr
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved baseMethod
+    cmp     r2, #0                      @ null "this"?
+    ldr     r9, [rGLUE, #offGlue_method] @ r9<- current method
+    beq     common_errNullObject        @ null "this", throw exception
+    cmp     r0, #0                      @ already resolved?
+    ldr     r9, [r9, #offMethod_clazz]  @ r9<- method->clazz
+    EXPORT_PC()                         @ must export for invoke
+    bne     .LOP_INVOKE_SUPER_RANGE_continue        @ resolved, continue on
+    b       .LOP_INVOKE_SUPER_RANGE_resolve         @ do resolve now
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_DIRECT_RANGE: /* 0x76 */
+/* File: armv5te/OP_INVOKE_DIRECT_RANGE.S */
+/* File: armv5te/OP_INVOKE_DIRECT.S */
+    /*
+     * Handle a direct method call.
+     *
+     * (We could defer the "is 'this' pointer null" test to the common
+     * method invocation code, and use a flag to indicate that static
+     * calls don't count.  If we do this as part of copying the arguments
+     * out we could avoiding loading the first arg twice.)
+     *
+     * for: invoke-direct, invoke-direct/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved methodToCall
+    .if     (!1)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    GET_VREG(r2, r10)                   @ r2<- "this" ptr
+    beq     .LOP_INVOKE_DIRECT_RANGE_resolve         @ not resolved, do it now
+.LOP_INVOKE_DIRECT_RANGE_finish:
+    cmp     r2, #0                      @ null "this" ref?
+    bne     common_invokeMethodRange   @ no, continue on
+    b       common_errNullObject        @ yes, throw exception
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_STATIC_RANGE: /* 0x77 */
+/* File: armv5te/OP_INVOKE_STATIC_RANGE.S */
+/* File: armv5te/OP_INVOKE_STATIC.S */
+    /*
+     * Handle a static method call.
+     *
+     * for: invoke-static, invoke-static/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved methodToCall
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    bne     common_invokeMethodRange @ yes, continue on
+0:  ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    mov     r2, #METHOD_STATIC          @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    bne     common_invokeMethodRange @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_INTERFACE_RANGE: /* 0x78 */
+/* File: armv5te/OP_INVOKE_INTERFACE_RANGE.S */
+/* File: armv5te/OP_INVOKE_INTERFACE.S */
+    /*
+     * Handle an interface method call.
+     *
+     * for: invoke-interface, invoke-interface/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r2, 2)                        @ r2<- FEDC or CCCC
+    FETCH(r1, 1)                        @ r1<- BBBB
+    .if     (!1)
+    and     r2, r2, #15                 @ r2<- C (or stays CCCC)
+    .endif
+    EXPORT_PC()                         @ must export for invoke
+    GET_VREG(r0, r2)                    @ r0<- first arg ("this")
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- methodClassDex
+    cmp     r0, #0                      @ null obj?
+    ldr     r2, [rGLUE, #offGlue_method]  @ r2<- method
+    beq     common_errNullObject        @ yes, fail
+    ldr     r0, [r0, #offObject_clazz]  @ r0<- thisPtr->clazz
+    bl      dvmFindInterfaceMethodInCache @ r0<- call(class, ref, method, dex)
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yes, handle exception
+    b       common_invokeMethodRange @ jump to common handler 
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_79: /* 0x79 */
+/* File: armv5te/OP_UNUSED_79.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_7A: /* 0x7a */
+/* File: armv5te/OP_UNUSED_7A.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEG_INT: /* 0x7b */
+/* File: armv5te/OP_NEG_INT.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    rsb     r0, r0, #0                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NOT_INT: /* 0x7c */
+/* File: armv5te/OP_NOT_INT.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mvn     r0, r0                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEG_LONG: /* 0x7d */
+/* File: armv5te/OP_NEG_LONG.S */
+/* File: armv5te/unopWide.S */
+    /*
+     * Generic 64-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0/r1".
+     * This could be an ARM instruction or a function call.
+     *
+     * For: neg-long, not-long, neg-double, long-to-double, double-to-long
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vAA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    rsbs    r0, r0, #0                           @ optional op; may set condition codes
+    rsc     r1, r1, #0                              @ r0/r1<- op, r2-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NOT_LONG: /* 0x7e */
+/* File: armv5te/OP_NOT_LONG.S */
+/* File: armv5te/unopWide.S */
+    /*
+     * Generic 64-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0/r1".
+     * This could be an ARM instruction or a function call.
+     *
+     * For: neg-long, not-long, neg-double, long-to-double, double-to-long
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vAA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mvn     r0, r0                           @ optional op; may set condition codes
+    mvn     r1, r1                              @ r0/r1<- op, r2-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEG_FLOAT: /* 0x7f */
+/* File: armv5te/OP_NEG_FLOAT.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    add     r0, r0, #0x80000000                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEG_DOUBLE: /* 0x80 */
+/* File: armv5te/OP_NEG_DOUBLE.S */
+/* File: armv5te/unopWide.S */
+    /*
+     * Generic 64-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0/r1".
+     * This could be an ARM instruction or a function call.
+     *
+     * For: neg-long, not-long, neg-double, long-to-double, double-to-long
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vAA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    add     r1, r1, #0x80000000                              @ r0/r1<- op, r2-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_LONG: /* 0x81 */
+/* File: armv5te/OP_INT_TO_LONG.S */
+/* File: armv5te/unopWider.S */
+    /*
+     * Generic 32bit-to-64bit unary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = op r0", where
+     * "result" is a 64-bit quantity in r0/r1.
+     *
+     * For: int-to-long, int-to-double, float-to-long, float-to-double
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r3)                    @ r0<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mov     r1, r0, asr #31                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vA/vA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-11 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_FLOAT: /* 0x82 */
+/* File: armv5te/OP_INT_TO_FLOAT.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    bl      __aeabi_i2f                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_DOUBLE: /* 0x83 */
+/* File: armv5te/OP_INT_TO_DOUBLE.S */
+/* File: armv5te/unopWider.S */
+    /*
+     * Generic 32bit-to-64bit unary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = op r0", where
+     * "result" is a 64-bit quantity in r0/r1.
+     *
+     * For: int-to-long, int-to-double, float-to-long, float-to-double
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r3)                    @ r0<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    bl      __aeabi_i2d                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vA/vA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-11 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_LONG_TO_INT: /* 0x84 */
+/* File: armv5te/OP_LONG_TO_INT.S */
+/* we ignore the high word, making this equivalent to a 32-bit reg move */
+/* File: armv5te/OP_MOVE.S */
+    /* for move, move-object, long-to-int */
+    /* op vA, vB */
+    mov     r1, rINST, lsr #12          @ r1<- B from 15:12
+    mov     r0, rINST, lsr #8           @ r0<- A from 11:8
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[B]
+    and     r0, r0, #15
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[A]<- r2
+    GOTO_OPCODE(ip)                     @ execute next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_LONG_TO_FLOAT: /* 0x85 */
+/* File: armv5te/OP_LONG_TO_FLOAT.S */
+/* File: armv5te/unopNarrower.S */
+    /*
+     * Generic 64bit-to-32bit unary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = op r0/r1", where
+     * "result" is a 32-bit quantity in r0.
+     *
+     * For: long-to-float, double-to-int, double-to-float
+     *
+     * (This would work for long-to-int, but that instruction is actually
+     * an exact match for OP_MOVE.)
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    and     r9, r9, #15
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vB/vB+1
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_l2f                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-11 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_LONG_TO_DOUBLE: /* 0x86 */
+/* File: armv5te/OP_LONG_TO_DOUBLE.S */
+/* File: armv5te/unopWide.S */
+    /*
+     * Generic 64-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0/r1".
+     * This could be an ARM instruction or a function call.
+     *
+     * For: neg-long, not-long, neg-double, long-to-double, double-to-long
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vAA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_l2d                              @ r0/r1<- op, r2-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FLOAT_TO_INT: /* 0x87 */
+/* File: armv5te/OP_FLOAT_TO_INT.S */
+/* EABI appears to have Java-style conversions of +inf/-inf/NaN */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    bl      __aeabi_f2iz                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+#if 0
+@include "armv5te/unop.S" {"instr":"bl      f2i_doconv"}
+@break
+/*
+ * Convert the float in r0 to an int in r0.
+ *
+ * We have to clip values to int min/max per the specification.  The
+ * expected common case is a "reasonable" value that converts directly
+ * to modest integer.  The EABI convert function isn't doing this for us.
+ */
+f2i_doconv:
+    stmfd   sp!, {r4, lr}
+    mov     r1, #0x4f000000             @ (float)maxint
+    mov     r4, r0
+    bl      __aeabi_fcmpge              @ is arg >= maxint?
+    cmp     r0, #0                      @ nonzero == yes
+    mvnne   r0, #0x80000000             @ return maxint (7fffffff)
+    ldmnefd sp!, {r4, pc}
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, #0xcf000000             @ (float)minint
+    bl      __aeabi_fcmple              @ is arg <= minint?
+    cmp     r0, #0                      @ nonzero == yes
+    movne   r0, #0x80000000             @ return minint (80000000)
+    ldmnefd sp!, {r4, pc}
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r4
+    bl      __aeabi_fcmpeq              @ is arg == self?
+    cmp     r0, #0                      @ zero == no
+    ldmeqfd sp!, {r4, pc}               @ return zero for NaN
+
+    mov     r0, r4                      @ recover arg
+    bl      __aeabi_f2iz                @ convert float to int
+    ldmfd   sp!, {r4, pc}
+#endif
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FLOAT_TO_LONG: /* 0x88 */
+/* File: armv5te/OP_FLOAT_TO_LONG.S */
+@include "armv5te/unopWider.S" {"instr":"bl      __aeabi_f2lz"}
+/* File: armv5te/unopWider.S */
+    /*
+     * Generic 32bit-to-64bit unary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = op r0", where
+     * "result" is a 64-bit quantity in r0/r1.
+     *
+     * For: int-to-long, int-to-double, float-to-long, float-to-double
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r3)                    @ r0<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    bl      f2l_doconv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vA/vA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-11 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FLOAT_TO_DOUBLE: /* 0x89 */
+/* File: armv5te/OP_FLOAT_TO_DOUBLE.S */
+/* File: armv5te/unopWider.S */
+    /*
+     * Generic 32bit-to-64bit unary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = op r0", where
+     * "result" is a 64-bit quantity in r0/r1.
+     *
+     * For: int-to-long, int-to-double, float-to-long, float-to-double
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r3)                    @ r0<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    bl      __aeabi_f2d                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vA/vA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-11 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DOUBLE_TO_INT: /* 0x8a */
+/* File: armv5te/OP_DOUBLE_TO_INT.S */
+/* EABI appears to have Java-style conversions of +inf/-inf/NaN */
+/* File: armv5te/unopNarrower.S */
+    /*
+     * Generic 64bit-to-32bit unary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = op r0/r1", where
+     * "result" is a 32-bit quantity in r0.
+     *
+     * For: long-to-float, double-to-int, double-to-float
+     *
+     * (This would work for long-to-int, but that instruction is actually
+     * an exact match for OP_MOVE.)
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    and     r9, r9, #15
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vB/vB+1
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_d2iz                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-11 instructions */
+
+
+#if 0
+@include "armv5te/unopNarrower.S" {"instr":"bl      d2i_doconv"}
+@break
+/*
+ * Convert the double in r0/r1 to an int in r0.
+ *
+ * We have to clip values to int min/max per the specification.  The
+ * expected common case is a "reasonable" value that converts directly
+ * to modest integer.  The EABI convert function isn't doing this for us.
+ */
+d2i_doconv:
+    stmfd   sp!, {r4, r5, lr}           @ save regs
+    ldr     r2, .LOP_DOUBLE_TO_INT_maxlo       @ (double)maxint, lo
+    ldr     r3, .LOP_DOUBLE_TO_INT_maxhi       @ (double)maxint, hi
+    sub     sp, sp, #4                  @ align for EABI
+    mov     r4, r0                      @ save r0
+    mov     r5, r1                      @  and r1
+    bl      __aeabi_dcmpge              @ is arg >= maxint?
+    cmp     r0, #0                      @ nonzero == yes
+    mvnne   r0, #0x80000000             @ return maxint (7fffffff)
+    bne     1f
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r5
+    ldr     r3, .LOP_DOUBLE_TO_INT_min         @ (double)minint, hi
+    mov     r2, #0                      @ (double)minint, lo
+    bl      __aeabi_dcmple              @ is arg <= minint?
+    cmp     r0, #0                      @ nonzero == yes
+    movne   r0, #0x80000000             @ return minint (80000000)
+    bne     1f
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r5
+    mov     r2, r4                      @ compare against self
+    mov     r3, r5
+    bl      __aeabi_dcmpeq              @ is arg == self?
+    cmp     r0, #0                      @ zero == no
+    beq     1f                          @ return zero for NaN
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r5
+    bl      __aeabi_d2iz                @ convert double to int
+
+1:
+    add     sp, sp, #4
+    ldmfd   sp!, {r4, r5, pc}
+
+.LOP_DOUBLE_TO_INT_maxlo:
+    .word   0xffc00000                  @ maxint, as a double (low word)
+.LOP_DOUBLE_TO_INT_maxhi:
+    .word   0x41dfffff                  @ maxint, as a double (high word)
+.LOP_DOUBLE_TO_INT_min:
+    .word   0xc1e00000                  @ minint, as a double (high word)
+#endif
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DOUBLE_TO_LONG: /* 0x8b */
+/* File: armv5te/OP_DOUBLE_TO_LONG.S */
+@include "armv5te/unopWide.S" {"instr":"bl      __aeabi_d2lz"}
+/* File: armv5te/unopWide.S */
+    /*
+     * Generic 64-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0/r1".
+     * This could be an ARM instruction or a function call.
+     *
+     * For: neg-long, not-long, neg-double, long-to-double, double-to-long
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vAA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      d2l_doconv                              @ r0/r1<- op, r2-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-13 instructions */
+
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DOUBLE_TO_FLOAT: /* 0x8c */
+/* File: armv5te/OP_DOUBLE_TO_FLOAT.S */
+/* File: armv5te/unopNarrower.S */
+    /*
+     * Generic 64bit-to-32bit unary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = op r0/r1", where
+     * "result" is a 32-bit quantity in r0.
+     *
+     * For: long-to-float, double-to-int, double-to-float
+     *
+     * (This would work for long-to-int, but that instruction is actually
+     * an exact match for OP_MOVE.)
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    and     r9, r9, #15
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vB/vB+1
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_d2f                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-11 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_BYTE: /* 0x8d */
+/* File: armv5te/OP_INT_TO_BYTE.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+    mov     r0, r0, asl #24                           @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mov     r0, r0, asr #24                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_CHAR: /* 0x8e */
+/* File: armv5te/OP_INT_TO_CHAR.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+    mov     r0, r0, asl #16                           @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mov     r0, r0, lsr #16                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_SHORT: /* 0x8f */
+/* File: armv5te/OP_INT_TO_SHORT.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+    mov     r0, r0, asl #16                           @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mov     r0, r0, asr #16                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_INT: /* 0x90 */
+/* File: armv5te/OP_ADD_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    add     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_INT: /* 0x91 */
+/* File: armv5te/OP_SUB_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    sub     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_INT: /* 0x92 */
+/* File: armv5te/OP_MUL_INT.S */
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    mul     r0, r1, r0                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_INT: /* 0x93 */
+/* File: armv5te/OP_DIV_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 1
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl     __aeabi_idiv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_INT: /* 0x94 */
+/* File: armv5te/OP_REM_INT.S */
+/* idivmod returns quotient in r0 and remainder in r1 */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 1
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_idivmod                              @ r1<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r1, r9)               @ vAA<- r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_INT: /* 0x95 */
+/* File: armv5te/OP_AND_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    and     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_INT: /* 0x96 */
+/* File: armv5te/OP_OR_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    orr     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_INT: /* 0x97 */
+/* File: armv5te/OP_XOR_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    eor     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_INT: /* 0x98 */
+/* File: armv5te/OP_SHL_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, asl r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_INT: /* 0x99 */
+/* File: armv5te/OP_SHR_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, asr r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_INT: /* 0x9a */
+/* File: armv5te/OP_USHR_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, lsr r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_LONG: /* 0x9b */
+/* File: armv5te/OP_ADD_LONG.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    adds    r0, r0, r2                           @ optional op; may set condition codes
+    adc     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_LONG: /* 0x9c */
+/* File: armv5te/OP_SUB_LONG.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    subs    r0, r0, r2                           @ optional op; may set condition codes
+    sbc     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_LONG: /* 0x9d */
+/* File: armv5te/OP_MUL_LONG.S */
+    /*
+     * Signed 64-bit integer multiply.
+     *
+     * Consider WXxYZ (r1r0 x r3r2) with a long multiply:
+     *        WX
+     *      x YZ
+     *  --------
+     *     ZW ZX
+     *  YW YX
+     *
+     * The low word of the result holds ZX, the high word holds
+     * (ZW+YX) + (the high overflow from ZX).  YW doesn't matter because
+     * it doesn't fit in the low 64 bits.
+     *
+     * Unlike most ARM math operations, multiply instructions have
+     * restrictions on using the same register more than once (Rd and Rm
+     * cannot be the same).
+     */
+    /* mul-long vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    mul     ip, r2, r1                  @  ip<- ZxW
+    umull   r9, r10, r2, r0             @  r9/r10 <- ZxX
+    mla     r2, r0, r3, ip              @  r2<- YxX + (ZxW)
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    add     r10, r2, r10                @  r10<- r10 + low(ZxW + (YxX))
+    add     r0, rFP, r0, lsl #2         @ r0<- &fp[AA]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    b       .LOP_MUL_LONG_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_LONG: /* 0x9e */
+/* File: armv5te/OP_DIV_LONG.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 1
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_ldivmod                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_LONG: /* 0x9f */
+/* File: armv5te/OP_REM_LONG.S */
+/* ldivmod returns quotient in r0/r1 and remainder in r2/r3 */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 1
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_ldivmod                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r2,r3}     @ vAA/vAA+1<- r2/r3
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_LONG: /* 0xa0 */
+/* File: armv5te/OP_AND_LONG.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    and     r0, r0, r2                           @ optional op; may set condition codes
+    and     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_LONG: /* 0xa1 */
+/* File: armv5te/OP_OR_LONG.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    orr     r0, r0, r2                           @ optional op; may set condition codes
+    orr     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_LONG: /* 0xa2 */
+/* File: armv5te/OP_XOR_LONG.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    eor     r0, r0, r2                           @ optional op; may set condition codes
+    eor     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_LONG: /* 0xa3 */
+/* File: armv5te/OP_SHL_LONG.S */
+    /*
+     * Long integer shift.  This is different from the generic 32/64-bit
+     * binary operations because vAA/vBB are 64-bit but vCC (the shift
+     * distance) is 32-bit.  Also, Dalvik requires us to mask off the low
+     * 6 bits of the shift distance.
+     */
+    /* shl-long vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r3, r0, #255                @ r3<- BB
+    mov     r0, r0, lsr #8              @ r0<- CC
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[BB]
+    GET_VREG(r2, r0)                    @ r2<- vCC
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    and     r2, r2, #63                 @ r2<- r2 & 0x3f
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+
+    mov     r1, r1, asl r2              @  r1<- r1 << r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r1, r1, r0, lsr r3          @  r1<- r1 | (r0 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    movpl   r1, r0, asl ip              @  if r2 >= 32, r1<- r0 << (r2-32)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    b       .LOP_SHL_LONG_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_LONG: /* 0xa4 */
+/* File: armv5te/OP_SHR_LONG.S */
+    /*
+     * Long integer shift.  This is different from the generic 32/64-bit
+     * binary operations because vAA/vBB are 64-bit but vCC (the shift
+     * distance) is 32-bit.  Also, Dalvik requires us to mask off the low
+     * 6 bits of the shift distance.
+     */
+    /* shr-long vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r3, r0, #255                @ r3<- BB
+    mov     r0, r0, lsr #8              @ r0<- CC
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[BB]
+    GET_VREG(r2, r0)                    @ r2<- vCC
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    and     r2, r2, #63                 @ r0<- r0 & 0x3f
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+
+    mov     r0, r0, lsr r2              @  r0<- r2 >> r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r0, r0, r1, asl r3          @  r0<- r0 | (r1 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    movpl   r0, r1, asr ip              @  if r2 >= 32, r0<-r1 >> (r2-32)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    b       .LOP_SHR_LONG_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_LONG: /* 0xa5 */
+/* File: armv5te/OP_USHR_LONG.S */
+    /*
+     * Long integer shift.  This is different from the generic 32/64-bit
+     * binary operations because vAA/vBB are 64-bit but vCC (the shift
+     * distance) is 32-bit.  Also, Dalvik requires us to mask off the low
+     * 6 bits of the shift distance.
+     */
+    /* ushr-long vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r3, r0, #255                @ r3<- BB
+    mov     r0, r0, lsr #8              @ r0<- CC
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[BB]
+    GET_VREG(r2, r0)                    @ r2<- vCC
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    and     r2, r2, #63                 @ r0<- r0 & 0x3f
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+
+    mov     r0, r0, lsr r2              @  r0<- r2 >> r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r0, r0, r1, asl r3          @  r0<- r0 | (r1 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    movpl   r0, r1, lsr ip              @  if r2 >= 32, r0<-r1 >>> (r2-32)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    b       .LOP_USHR_LONG_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_FLOAT: /* 0xa6 */
+/* File: armv5te/OP_ADD_FLOAT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_fadd                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_FLOAT: /* 0xa7 */
+/* File: armv5te/OP_SUB_FLOAT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_fsub                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_FLOAT: /* 0xa8 */
+/* File: armv5te/OP_MUL_FLOAT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_fmul                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_FLOAT: /* 0xa9 */
+/* File: armv5te/OP_DIV_FLOAT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_fdiv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_FLOAT: /* 0xaa */
+/* File: armv5te/OP_REM_FLOAT.S */
+/* EABI doesn't define a float remainder function, but libm does */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      fmodf                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_DOUBLE: /* 0xab */
+/* File: armv5te/OP_ADD_DOUBLE.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_dadd                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_DOUBLE: /* 0xac */
+/* File: armv5te/OP_SUB_DOUBLE.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_dsub                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_DOUBLE: /* 0xad */
+/* File: armv5te/OP_MUL_DOUBLE.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_dmul                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_DOUBLE: /* 0xae */
+/* File: armv5te/OP_DIV_DOUBLE.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_ddiv                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_DOUBLE: /* 0xaf */
+/* File: armv5te/OP_REM_DOUBLE.S */
+/* EABI doesn't define a double remainder function, but libm does */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      fmod                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_INT_2ADDR: /* 0xb0 */
+/* File: armv5te/OP_ADD_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    add     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_INT_2ADDR: /* 0xb1 */
+/* File: armv5te/OP_SUB_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    sub     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_INT_2ADDR: /* 0xb2 */
+/* File: armv5te/OP_MUL_INT_2ADDR.S */
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    mul     r0, r1, r0                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_INT_2ADDR: /* 0xb3 */
+/* File: armv5te/OP_DIV_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 1
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl     __aeabi_idiv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_INT_2ADDR: /* 0xb4 */
+/* File: armv5te/OP_REM_INT_2ADDR.S */
+/* idivmod returns quotient in r0 and remainder in r1 */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 1
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_idivmod                              @ r1<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r1, r9)               @ vAA<- r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_INT_2ADDR: /* 0xb5 */
+/* File: armv5te/OP_AND_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    and     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_INT_2ADDR: /* 0xb6 */
+/* File: armv5te/OP_OR_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    orr     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_INT_2ADDR: /* 0xb7 */
+/* File: armv5te/OP_XOR_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    eor     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_INT_2ADDR: /* 0xb8 */
+/* File: armv5te/OP_SHL_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, asl r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_INT_2ADDR: /* 0xb9 */
+/* File: armv5te/OP_SHR_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, asr r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_INT_2ADDR: /* 0xba */
+/* File: armv5te/OP_USHR_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, lsr r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_LONG_2ADDR: /* 0xbb */
+/* File: armv5te/OP_ADD_LONG_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    adds    r0, r0, r2                           @ optional op; may set condition codes
+    adc     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_LONG_2ADDR: /* 0xbc */
+/* File: armv5te/OP_SUB_LONG_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    subs    r0, r0, r2                           @ optional op; may set condition codes
+    sbc     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_LONG_2ADDR: /* 0xbd */
+/* File: armv5te/OP_MUL_LONG_2ADDR.S */
+    /*
+     * Signed 64-bit integer multiply, "/2addr" version.
+     *
+     * See OP_MUL_LONG for an explanation.
+     *
+     * We get a little tight on registers, so to avoid looking up &fp[A]
+     * again we stuff it into rINST.
+     */
+    /* mul-long/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     rINST, rFP, r9, lsl #2      @ rINST<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   rINST, {r0-r1}              @ r0/r1<- vAA/vAA+1
+    mul     ip, r2, r1                  @  ip<- ZxW
+    umull   r9, r10, r2, r0             @  r9/r10 <- ZxX
+    mla     r2, r0, r3, ip              @  r2<- YxX + (ZxW)
+    mov     r0, rINST                   @ r0<- &fp[A] (free up rINST)
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    add     r10, r2, r10                @  r10<- r10 + low(ZxW + (YxX))
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r0, {r9-r10}                @ vAA/vAA+1<- r9/r10
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_LONG_2ADDR: /* 0xbe */
+/* File: armv5te/OP_DIV_LONG_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 1
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_ldivmod                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_LONG_2ADDR: /* 0xbf */
+/* File: armv5te/OP_REM_LONG_2ADDR.S */
+/* ldivmod returns quotient in r0/r1 and remainder in r2/r3 */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 1
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_ldivmod                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r2,r3}     @ vAA/vAA+1<- r2/r3
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_LONG_2ADDR: /* 0xc0 */
+/* File: armv5te/OP_AND_LONG_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    and     r0, r0, r2                           @ optional op; may set condition codes
+    and     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_LONG_2ADDR: /* 0xc1 */
+/* File: armv5te/OP_OR_LONG_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    orr     r0, r0, r2                           @ optional op; may set condition codes
+    orr     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_LONG_2ADDR: /* 0xc2 */
+/* File: armv5te/OP_XOR_LONG_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    eor     r0, r0, r2                           @ optional op; may set condition codes
+    eor     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_LONG_2ADDR: /* 0xc3 */
+/* File: armv5te/OP_SHL_LONG_2ADDR.S */
+    /*
+     * Long integer shift, 2addr version.  vA is 64-bit value/result, vB is
+     * 32-bit shift distance.
+     */
+    /* shl-long/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r2, r3)                    @ r2<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    and     r2, r2, #63                 @ r2<- r2 & 0x3f
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+
+    mov     r1, r1, asl r2              @  r1<- r1 << r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r1, r1, r0, lsr r3          @  r1<- r1 | (r0 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    movpl   r1, r0, asl ip              @  if r2 >= 32, r1<- r0 << (r2-32)
+    mov     r0, r0, asl r2              @  r0<- r0 << r2
+    b       .LOP_SHL_LONG_2ADDR_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_LONG_2ADDR: /* 0xc4 */
+/* File: armv5te/OP_SHR_LONG_2ADDR.S */
+    /*
+     * Long integer shift, 2addr version.  vA is 64-bit value/result, vB is
+     * 32-bit shift distance.
+     */
+    /* shr-long/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r2, r3)                    @ r2<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    and     r2, r2, #63                 @ r2<- r2 & 0x3f
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+
+    mov     r0, r0, lsr r2              @  r0<- r2 >> r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r0, r0, r1, asl r3          @  r0<- r0 | (r1 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    movpl   r0, r1, asr ip              @  if r2 >= 32, r0<-r1 >> (r2-32)
+    mov     r1, r1, asr r2              @  r1<- r1 >> r2
+    b       .LOP_SHR_LONG_2ADDR_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_LONG_2ADDR: /* 0xc5 */
+/* File: armv5te/OP_USHR_LONG_2ADDR.S */
+    /*
+     * Long integer shift, 2addr version.  vA is 64-bit value/result, vB is
+     * 32-bit shift distance.
+     */
+    /* ushr-long/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r2, r3)                    @ r2<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    and     r2, r2, #63                 @ r2<- r2 & 0x3f
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+
+    mov     r0, r0, lsr r2              @  r0<- r2 >> r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r0, r0, r1, asl r3          @  r0<- r0 | (r1 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    movpl   r0, r1, lsr ip              @  if r2 >= 32, r0<-r1 >>> (r2-32)
+    mov     r1, r1, lsr r2              @  r1<- r1 >>> r2
+    b       .LOP_USHR_LONG_2ADDR_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_FLOAT_2ADDR: /* 0xc6 */
+/* File: armv5te/OP_ADD_FLOAT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_fadd                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_FLOAT_2ADDR: /* 0xc7 */
+/* File: armv5te/OP_SUB_FLOAT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_fsub                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_FLOAT_2ADDR: /* 0xc8 */
+/* File: armv5te/OP_MUL_FLOAT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_fmul                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_FLOAT_2ADDR: /* 0xc9 */
+/* File: armv5te/OP_DIV_FLOAT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_fdiv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_FLOAT_2ADDR: /* 0xca */
+/* File: armv5te/OP_REM_FLOAT_2ADDR.S */
+/* EABI doesn't define a float remainder function, but libm does */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      fmodf                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_DOUBLE_2ADDR: /* 0xcb */
+/* File: armv5te/OP_ADD_DOUBLE_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_dadd                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_DOUBLE_2ADDR: /* 0xcc */
+/* File: armv5te/OP_SUB_DOUBLE_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_dsub                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_DOUBLE_2ADDR: /* 0xcd */
+/* File: armv5te/OP_MUL_DOUBLE_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_dmul                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_DOUBLE_2ADDR: /* 0xce */
+/* File: armv5te/OP_DIV_DOUBLE_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_ddiv                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_DOUBLE_2ADDR: /* 0xcf */
+/* File: armv5te/OP_REM_DOUBLE_2ADDR.S */
+/* EABI doesn't define a double remainder function, but libm does */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      fmod                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_INT_LIT16: /* 0xd0 */
+/* File: armv5te/OP_ADD_INT_LIT16.S */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    add     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RSUB_INT: /* 0xd1 */
+/* File: armv5te/OP_RSUB_INT.S */
+/* this op is "rsub-int", but can be thought of as "rsub-int/lit16" */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    rsb     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_INT_LIT16: /* 0xd2 */
+/* File: armv5te/OP_MUL_INT_LIT16.S */
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    mul     r0, r1, r0                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_INT_LIT16: /* 0xd3 */
+/* File: armv5te/OP_DIV_INT_LIT16.S */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 1
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    bl     __aeabi_idiv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_INT_LIT16: /* 0xd4 */
+/* File: armv5te/OP_REM_INT_LIT16.S */
+/* idivmod returns quotient in r0 and remainder in r1 */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 1
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    bl      __aeabi_idivmod                              @ r1<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r1, r9)               @ vAA<- r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_INT_LIT16: /* 0xd5 */
+/* File: armv5te/OP_AND_INT_LIT16.S */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    and     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_INT_LIT16: /* 0xd6 */
+/* File: armv5te/OP_OR_INT_LIT16.S */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    orr     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_INT_LIT16: /* 0xd7 */
+/* File: armv5te/OP_XOR_INT_LIT16.S */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    eor     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_INT_LIT8: /* 0xd8 */
+/* File: armv5te/OP_ADD_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    add     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RSUB_INT_LIT8: /* 0xd9 */
+/* File: armv5te/OP_RSUB_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    rsb     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_INT_LIT8: /* 0xda */
+/* File: armv5te/OP_MUL_INT_LIT8.S */
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    mul     r0, r1, r0                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_INT_LIT8: /* 0xdb */
+/* File: armv5te/OP_DIV_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 1
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl     __aeabi_idiv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_INT_LIT8: /* 0xdc */
+/* File: armv5te/OP_REM_INT_LIT8.S */
+/* idivmod returns quotient in r0 and remainder in r1 */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 1
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_idivmod                              @ r1<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r1, r9)               @ vAA<- r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_INT_LIT8: /* 0xdd */
+/* File: armv5te/OP_AND_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    and     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_INT_LIT8: /* 0xde */
+/* File: armv5te/OP_OR_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    orr     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_INT_LIT8: /* 0xdf */
+/* File: armv5te/OP_XOR_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    eor     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_INT_LIT8: /* 0xe0 */
+/* File: armv5te/OP_SHL_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, asl r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_INT_LIT8: /* 0xe1 */
+/* File: armv5te/OP_SHR_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, asr r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_INT_LIT8: /* 0xe2 */
+/* File: armv5te/OP_USHR_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, lsr r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E3: /* 0xe3 */
+/* File: armv5te/OP_UNUSED_E3.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E4: /* 0xe4 */
+/* File: armv5te/OP_UNUSED_E4.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E5: /* 0xe5 */
+/* File: armv5te/OP_UNUSED_E5.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E6: /* 0xe6 */
+/* File: armv5te/OP_UNUSED_E6.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E7: /* 0xe7 */
+/* File: armv5te/OP_UNUSED_E7.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E8: /* 0xe8 */
+/* File: armv5te/OP_UNUSED_E8.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E9: /* 0xe9 */
+/* File: armv5te/OP_UNUSED_E9.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EA: /* 0xea */
+/* File: armv5te/OP_UNUSED_EA.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EB: /* 0xeb */
+/* File: armv5te/OP_UNUSED_EB.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EC: /* 0xec */
+/* File: armv5te/OP_UNUSED_EC.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_ED: /* 0xed */
+/* File: armv5te/OP_UNUSED_ED.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_EXECUTE_INLINE: /* 0xee */
+/* File: armv5te/OP_EXECUTE_INLINE.S */
+    /*
+     * Execute a "native inline" instruction.
+     *
+     * We need to call:
+     *  dvmPerformInlineOp4Std(arg0, arg1, arg2, arg3, &retval, ref)
+     *
+     * The first four args are in r0-r3, but the last two must be pushed
+     * onto the stack.
+     */
+    /* [opt] execute-inline vAA, {vC, vD, vE, vF}, inline@BBBB */
+    FETCH(r10, 1)                       @ r10<- BBBB
+    add     r1, rGLUE, #offGlue_retval  @ r1<- &glue->retval
+    EXPORT_PC()                         @ can throw
+    sub     sp, sp, #8                  @ make room for arg(s)
+    mov     r0, rINST, lsr #12          @ r0<- B
+    str     r1, [sp]                    @ push &glue->retval
+    bl      .LOP_EXECUTE_INLINE_continue        @ make call; will return after
+    add     sp, sp, #8                  @ pop stack
+    cmp     r0, #0                      @ test boolean result of inline
+    beq     common_exceptionThrown      @ returned false, handle exception
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EF: /* 0xef */
+/* File: armv5te/OP_UNUSED_EF.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_DIRECT_EMPTY: /* 0xf0 */
+/* File: armv5te/OP_INVOKE_DIRECT_EMPTY.S */
+    /*
+     * invoke-direct-empty is a no-op in a "standard" interpreter.
+     */
+    FETCH_ADVANCE_INST(3)               @ advance to next instr, load rINST
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    GOTO_OPCODE(ip)                     @ execute it
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_F1: /* 0xf1 */
+/* File: armv5te/OP_UNUSED_F1.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_QUICK: /* 0xf2 */
+/* File: armv5te/OP_IGET_QUICK.S */
+    /* For: iget-quick, iget-object-quick */
+    /* op vA, vB, offset@CCCC */
+    mov     r2, rINST, lsr #12          @ r2<- B
+    GET_VREG(r3, r2)                    @ r3<- object we're operating on
+    FETCH(r1, 1)                        @ r1<- field byte offset
+    cmp     r3, #0                      @ check object for null
+    mov     r2, rINST, lsr #8           @ r2<- A(+)
+    beq     common_errNullObject        @ object was null
+    ldr     r0, [r3, r1]                @ r0<- obj.field (always 32 bits)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_WIDE_QUICK: /* 0xf3 */
+/* File: armv5te/OP_IGET_WIDE_QUICK.S */
+    /* iget-wide-quick vA, vB, offset@CCCC */
+    mov     r2, rINST, lsr #12          @ r2<- B
+    GET_VREG(r3, r2)                    @ r3<- object we're operating on
+    FETCH(r1, 1)                        @ r1<- field byte offset
+    cmp     r3, #0                      @ check object for null
+    mov     r2, rINST, lsr #8           @ r2<- A(+)
+    beq     common_errNullObject        @ object was null
+    ldrd    r0, [r3, r1]                @ r0<- obj.field (64 bits, aligned)
+    and     r2, r2, #15
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    add     r3, rFP, r2, lsl #2         @ r3<- &fp[A]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r3, {r0-r1}                 @ fp[A]<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_OBJECT_QUICK: /* 0xf4 */
+/* File: armv5te/OP_IGET_OBJECT_QUICK.S */
+/* File: armv5te/OP_IGET_QUICK.S */
+    /* For: iget-quick, iget-object-quick */
+    /* op vA, vB, offset@CCCC */
+    mov     r2, rINST, lsr #12          @ r2<- B
+    GET_VREG(r3, r2)                    @ r3<- object we're operating on
+    FETCH(r1, 1)                        @ r1<- field byte offset
+    cmp     r3, #0                      @ check object for null
+    mov     r2, rINST, lsr #8           @ r2<- A(+)
+    beq     common_errNullObject        @ object was null
+    ldr     r0, [r3, r1]                @ r0<- obj.field (always 32 bits)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_QUICK: /* 0xf5 */
+/* File: armv5te/OP_IPUT_QUICK.S */
+    /* For: iput-quick, iput-object-quick */
+    /* op vA, vB, offset@CCCC */
+    mov     r2, rINST, lsr #12          @ r2<- B
+    GET_VREG(r3, r2)                    @ r3<- fp[B], the object pointer
+    FETCH(r1, 1)                        @ r1<- field byte offset
+    cmp     r3, #0                      @ check object for null
+    mov     r2, rINST, lsr #8           @ r2<- A(+)
+    beq     common_errNullObject        @ object was null
+    and     r2, r2, #15
+    GET_VREG(r0, r2)                    @ r0<- fp[A]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    str     r0, [r3, r1]                @ obj.field (always 32 bits)<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_WIDE_QUICK: /* 0xf6 */
+/* File: armv5te/OP_IPUT_WIDE_QUICK.S */
+    /* iput-wide-quick vA, vB, offset@CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A(+)
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r2, r1)                    @ r2<- fp[B], the object pointer
+    add     r3, rFP, r0, lsl #2         @ r3<- &fp[A]
+    cmp     r2, #0                      @ check object for null
+    ldmia   r3, {r0-r1}                 @ r0/r1<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH(r3, 1)                        @ r3<- field byte offset
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    strd    r0, [r2, r3]                @ obj.field (64 bits, aligned)<- r0/r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_OBJECT_QUICK: /* 0xf7 */
+/* File: armv5te/OP_IPUT_OBJECT_QUICK.S */
+/* File: armv5te/OP_IPUT_QUICK.S */
+    /* For: iput-quick, iput-object-quick */
+    /* op vA, vB, offset@CCCC */
+    mov     r2, rINST, lsr #12          @ r2<- B
+    GET_VREG(r3, r2)                    @ r3<- fp[B], the object pointer
+    FETCH(r1, 1)                        @ r1<- field byte offset
+    cmp     r3, #0                      @ check object for null
+    mov     r2, rINST, lsr #8           @ r2<- A(+)
+    beq     common_errNullObject        @ object was null
+    and     r2, r2, #15
+    GET_VREG(r0, r2)                    @ r0<- fp[A]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    str     r0, [r3, r1]                @ obj.field (always 32 bits)<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_VIRTUAL_QUICK: /* 0xf8 */
+/* File: armv5te/OP_INVOKE_VIRTUAL_QUICK.S */
+    /*
+     * Handle an optimized virtual method call.
+     *
+     * for: [opt] invoke-virtual-quick, invoke-virtual-quick/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r3, 2)                        @ r3<- FEDC or CCCC
+    FETCH(r1, 1)                        @ r1<- BBBB
+    .if     (!0)
+    and     r3, r3, #15                 @ r3<- C (or stays CCCC)
+    .endif
+    GET_VREG(r2, r3)                    @ r2<- vC ("this" ptr)
+    cmp     r2, #0                      @ is "this" null?
+    beq     common_errNullObject        @ null "this", throw exception
+    ldr     r2, [r2, #offObject_clazz]  @ r2<- thisPtr->clazz
+    ldr     r2, [r2, #offClassObject_vtable]    @ r2<- thisPtr->clazz->vtable
+    EXPORT_PC()                         @ invoke must export
+    ldr     r0, [r2, r1, lsl #2]        @ r3<- vtable[BBBB]
+    bl      common_invokeMethodNoRange @ continue on
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_VIRTUAL_QUICK_RANGE: /* 0xf9 */
+/* File: armv5te/OP_INVOKE_VIRTUAL_QUICK_RANGE.S */
+/* File: armv5te/OP_INVOKE_VIRTUAL_QUICK.S */
+    /*
+     * Handle an optimized virtual method call.
+     *
+     * for: [opt] invoke-virtual-quick, invoke-virtual-quick/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r3, 2)                        @ r3<- FEDC or CCCC
+    FETCH(r1, 1)                        @ r1<- BBBB
+    .if     (!1)
+    and     r3, r3, #15                 @ r3<- C (or stays CCCC)
+    .endif
+    GET_VREG(r2, r3)                    @ r2<- vC ("this" ptr)
+    cmp     r2, #0                      @ is "this" null?
+    beq     common_errNullObject        @ null "this", throw exception
+    ldr     r2, [r2, #offObject_clazz]  @ r2<- thisPtr->clazz
+    ldr     r2, [r2, #offClassObject_vtable]    @ r2<- thisPtr->clazz->vtable
+    EXPORT_PC()                         @ invoke must export
+    ldr     r0, [r2, r1, lsl #2]        @ r3<- vtable[BBBB]
+    bl      common_invokeMethodRange @ continue on
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_SUPER_QUICK: /* 0xfa */
+/* File: armv5te/OP_INVOKE_SUPER_QUICK.S */
+    /*
+     * Handle an optimized "super" method call.
+     *
+     * for: [opt] invoke-super-quick, invoke-super-quick/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    .if     (!0)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r2, [r2, #offMethod_clazz]  @ r2<- method->clazz
+    EXPORT_PC()                         @ must export for invoke
+    ldr     r2, [r2, #offClassObject_super]     @ r2<- method->clazz->super
+    GET_VREG(r3, r10)                   @ r3<- "this"
+    ldr     r2, [r2, #offClassObject_vtable]    @ r2<- ...clazz->super->vtable
+    cmp     r3, #0                      @ null "this" ref?
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- super->vtable[BBBB]
+    beq     common_errNullObject        @ "this" is null, throw exception
+    bl      common_invokeMethodNoRange @ continue on
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_SUPER_QUICK_RANGE: /* 0xfb */
+/* File: armv5te/OP_INVOKE_SUPER_QUICK_RANGE.S */
+/* File: armv5te/OP_INVOKE_SUPER_QUICK.S */
+    /*
+     * Handle an optimized "super" method call.
+     *
+     * for: [opt] invoke-super-quick, invoke-super-quick/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    .if     (!1)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r2, [r2, #offMethod_clazz]  @ r2<- method->clazz
+    EXPORT_PC()                         @ must export for invoke
+    ldr     r2, [r2, #offClassObject_super]     @ r2<- method->clazz->super
+    GET_VREG(r3, r10)                   @ r3<- "this"
+    ldr     r2, [r2, #offClassObject_vtable]    @ r2<- ...clazz->super->vtable
+    cmp     r3, #0                      @ null "this" ref?
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- super->vtable[BBBB]
+    beq     common_errNullObject        @ "this" is null, throw exception
+    bl      common_invokeMethodRange @ continue on
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FC: /* 0xfc */
+/* File: armv5te/OP_UNUSED_FC.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FD: /* 0xfd */
+/* File: armv5te/OP_UNUSED_FD.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FE: /* 0xfe */
+/* File: armv5te/OP_UNUSED_FE.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FF: /* 0xff */
+/* File: armv5te/OP_UNUSED_FF.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+
+    .balign 64
+    .size   dvmAsmInstructionStart, .-dvmAsmInstructionStart
+    .global dvmAsmInstructionEnd
+dvmAsmInstructionEnd:
+
+/*
+ * ===========================================================================
+ *  Sister implementations
+ * ===========================================================================
+ */
+    .global dvmAsmSisterStart
+    .type   dvmAsmSisterStart, %function
+    .text
+    .balign 4
+dvmAsmSisterStart:
+
+/* continuation for OP_CONST_STRING */
+
+    /*
+     * Continuation if the String has not yet been resolved.
+     *  r1: BBBB (String ref)
+     *  r9: target register
+     */
+.LOP_CONST_STRING_resolve:
+    EXPORT_PC()
+    ldr     r0, [rGLUE, #offGlue_method] @ r0<- glue->method
+    ldr     r0, [r0, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveString            @ r0<- String reference
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yup, handle the exception
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_CONST_STRING_JUMBO */
+
+    /*
+     * Continuation if the String has not yet been resolved.
+     *  r1: BBBBBBBB (String ref)
+     *  r9: target register
+     */
+.LOP_CONST_STRING_JUMBO_resolve:
+    EXPORT_PC()
+    ldr     r0, [rGLUE, #offGlue_method] @ r0<- glue->method
+    ldr     r0, [r0, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveString            @ r0<- String reference
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yup, handle the exception
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_CONST_CLASS */
+
+    /*
+     * Continuation if the Class has not yet been resolved.
+     *  r1: BBBB (Class ref)
+     *  r9: target register
+     */
+.LOP_CONST_CLASS_resolve:
+    EXPORT_PC()
+    ldr     r0, [rGLUE, #offGlue_method] @ r0<- glue->method
+    mov     r2, #1                      @ r2<- true
+    ldr     r0, [r0, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- Class reference
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yup, handle the exception
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_CHECK_CAST */
+
+    /*
+     * Trivial test failed, need to perform full check.  This is common.
+     *  r0 holds obj->clazz
+     *  r1 holds class resolved from BBBB
+     *  r9 holds object
+     */
+.LOP_CHECK_CAST_fullcheck:
+    bl      dvmInstanceofNonTrivial     @ r0<- boolean result
+    cmp     r0, #0                      @ failed?
+    bne     .LOP_CHECK_CAST_okay            @ no, success
+
+    @ A cast has failed.  We need to throw a ClassCastException with the
+    @ class of the object that failed to be cast.
+    EXPORT_PC()                         @ about to throw
+    ldr     r3, [r9, #offObject_clazz]  @ r3<- obj->clazz
+    ldr     r0, .LstrClassCastExceptionPtr
+    ldr     r1, [r3, #offClassObject_descriptor] @ r1<- obj->clazz->descriptor
+    bl      dvmThrowExceptionWithClassMessage
+    b       common_exceptionThrown
+
+    /*
+     * Resolution required.  This is the least-likely path.
+     *
+     *  r2 holds BBBB
+     *  r9 holds object
+     */
+.LOP_CHECK_CAST_resolve:
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    mov     r1, r2                      @ r1<- BBBB
+    mov     r2, #0                      @ r2<- false
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- resolved ClassObject ptr
+    cmp     r0, #0                      @ got null?
+    beq     common_exceptionThrown      @ yes, handle exception
+    mov     r1, r0                      @ r1<- class resolved from BBB
+    ldr     r0, [r9, #offObject_clazz]  @ r0<- obj->clazz
+    b       .LOP_CHECK_CAST_resolved        @ pick up where we left off
+
+.LstrClassCastExceptionPtr:
+    .word   .LstrClassCastException
+
+
+/* continuation for OP_INSTANCE_OF */
+
+    /*
+     * Trivial test failed, need to perform full check.  This is common.
+     *  r0 holds obj->clazz
+     *  r1 holds class resolved from BBBB
+     *  r9 holds A
+     */
+.LOP_INSTANCE_OF_fullcheck:
+    bl      dvmInstanceofNonTrivial     @ r0<- boolean result
+    @ fall through to OP_INSTANCE_OF_store
+
+    /*
+     * r0 holds boolean result
+     * r9 holds A
+     */
+.LOP_INSTANCE_OF_store:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r0, r9)                    @ vA<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+    /*
+     * Trivial test succeeded, save and bail.
+     *  r9 holds A
+     */
+.LOP_INSTANCE_OF_trivial:
+    mov     r0, #1                      @ indicate success
+    @ could b OP_INSTANCE_OF_store, but copying is faster and cheaper
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r0, r9)                    @ vA<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+    /*
+     * Resolution required.  This is the least-likely path.
+     *
+     *  r3 holds BBBB
+     *  r9 holds A
+     */
+.LOP_INSTANCE_OF_resolve:
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [rGLUE, #offGlue_method]    @ r0<- glue->method
+    mov     r1, r3                      @ r1<- BBBB
+    mov     r2, #1                      @ r2<- true
+    ldr     r0, [r0, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- resolved ClassObject ptr
+    cmp     r0, #0                      @ got null?
+    beq     common_exceptionThrown      @ yes, handle exception
+    mov     r1, r0                      @ r1<- class resolved from BBB
+    mov     r3, rINST, lsr #12          @ r3<- B
+    GET_VREG(r0, r3)                    @ r0<- vB (object)
+    ldr     r0, [r0, #offObject_clazz]  @ r0<- obj->clazz
+    b       .LOP_INSTANCE_OF_resolved        @ pick up where we left off
+
+
+/* continuation for OP_NEW_INSTANCE */
+
+    .balign 32                          @ minimize cache lines
+.LOP_NEW_INSTANCE_finish: @ r0=class
+    bl      dvmAllocObject              @ r0<- new object
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yes, handle the exception
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r3)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+    /*
+     * Class initialization required.
+     *
+     *  r0 holds class object
+     */
+.LOP_NEW_INSTANCE_needinit:
+    mov     r9, r0                      @ save r0
+    bl      dvmInitClass                @ initialize class
+    cmp     r0, #0                      @ check boolean result
+    mov     r0, r9                      @ restore r0
+    bne     .LOP_NEW_INSTANCE_initialized     @ success, continue
+    b       common_exceptionThrown      @ failed, deal with init exception
+
+    /*
+     * Resolution required.  This is the least-likely path.
+     *
+     *  r1 holds BBBB
+     */
+.LOP_NEW_INSTANCE_resolve:
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    mov     r2, #0                      @ r2<- false
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- resolved ClassObject ptr
+    cmp     r0, #0                      @ got null?
+    bne     .LOP_NEW_INSTANCE_resolved        @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+    /*
+     * We can't instantiate an abstract class or interface, so throw an
+     * InstantiationError with the class descriptor as the message.
+     *
+     *  r0 holds class object
+     */
+.LOP_NEW_INSTANCE_abstract:
+    ldr     r1, [r0, #offClassObject_descriptor]
+    ldr     r0, .LstrInstantiationErrorPtr
+    bl      dvmThrowExceptionWithClassMessage
+    b       common_exceptionThrown
+
+.LstrInstantiationErrorPtr:
+    .word   .LstrInstantiationError
+
+
+/* continuation for OP_NEW_ARRAY */
+
+
+    /*
+     * Resolve class.  (This is an uncommon case.)
+     *
+     *  r1 holds array length
+     *  r2 holds class ref CCCC
+     */
+.LOP_NEW_ARRAY_resolve:
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    mov     r9, r1                      @ r9<- length (save)
+    mov     r1, r2                      @ r1<- CCCC
+    mov     r2, #0                      @ r2<- false
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- call(clazz, ref)
+    cmp     r0, #0                      @ got null?
+    mov     r1, r9                      @ r1<- length (restore)
+    beq     common_exceptionThrown      @ yes, handle exception
+    @ fall through to OP_NEW_ARRAY_finish
+
+    /*
+     * Finish allocation.
+     *
+     *  r0 holds class
+     *  r1 holds array length
+     */
+.LOP_NEW_ARRAY_finish:
+    mov     r2, #ALLOC_DONT_TRACK       @ don't track in local refs table
+    bl      dvmAllocArrayByClass        @ r0<- call(clazz, length, flags)
+    cmp     r0, #0                      @ failed?
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    beq     common_exceptionThrown      @ yes, handle the exception
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ vA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_FILLED_NEW_ARRAY */
+
+    /*
+     * On entry:
+     *  r0 holds array class
+     *  r10 holds AA or BA
+     */
+.LOP_FILLED_NEW_ARRAY_continue:
+    ldr     r3, [r0, #offClassObject_descriptor] @ r3<- arrayClass->descriptor
+    mov     r2, #ALLOC_DONT_TRACK       @ r2<- alloc flags
+    ldrb    r3, [r3, #1]                @ r3<- descriptor[1]
+    .if     0
+    mov     r1, r10                     @ r1<- AA (length)
+    .else
+    mov     r1, r10, lsr #4             @ r1<- B (length)
+    .endif
+    cmp     r3, #'I'                    @ array of ints?
+    cmpne   r3, #'L'                    @ array of objects?
+    cmpne   r3, #'['                    @ array of arrays?
+    mov     r9, r1                      @ save length in r9
+    bne     .LOP_FILLED_NEW_ARRAY_notimpl         @ no, not handled yet
+    bl      dvmAllocArrayByClass        @ r0<- call(arClass, length, flags)
+    cmp     r0, #0                      @ null return?
+    beq     common_exceptionThrown      @ alloc failed, handle exception
+
+    FETCH(r1, 2)                        @ r1<- FEDC or CCCC
+    str     r0, [rGLUE, #offGlue_retval]    @ retval.l <- new array
+    add     r0, r0, #offArrayObject_contents @ r0<- newArray->contents
+    subs    r9, r9, #1                  @ length--, check for neg
+    FETCH_ADVANCE_INST(3)               @ advance to next instr, load rINST
+    bmi     2f                          @ was zero, bail
+
+    @ copy values from registers into the array
+    @ r0=array, r1=CCCC/FEDC, r9=length (from AA or B), r10=AA/BA
+    .if     0
+    add     r2, rFP, r1, lsl #2         @ r2<- &fp[CCCC]
+1:  ldr     r3, [r2], #4                @ r3<- *r2++
+    subs    r9, r9, #1                  @ count--
+    str     r3, [r0], #4                @ *contents++ = vX
+    bpl     1b
+    @ continue at 2
+    .else
+    cmp     r9, #4                      @ length was initially 5?
+    and     r2, r10, #15                @ r2<- A
+    bne     1f                          @ <= 4 args, branch
+    GET_VREG(r3, r2)                    @ r3<- vA
+    sub     r9, r9, #1                  @ count--
+    str     r3, [r0, #16]               @ contents[4] = vA
+1:  and     r2, r1, #15                 @ r2<- F/E/D/C
+    GET_VREG(r3, r2)                    @ r3<- vF/vE/vD/vC
+    mov     r1, r1, lsr #4              @ r1<- next reg in low 4
+    subs    r9, r9, #1                  @ count--
+    str     r3, [r0], #4                @ *contents++ = vX
+    bpl     1b
+    @ continue at 2
+    .endif
+
+2:
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    GOTO_OPCODE(ip)                     @ execute it
+
+    /*
+     * Throw an exception indicating that we have not implemented this
+     * mode of filled-new-array.
+     */
+.LOP_FILLED_NEW_ARRAY_notimpl:
+    ldr     r0, .L_strInternalError
+    ldr     r1, .L_strFilledNewArrayNotImpl
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+    .if     (!0)                 @ define in one or the other, not both
+.L_strFilledNewArrayNotImpl:
+    .word   .LstrFilledNewArrayNotImpl
+.L_strInternalError:
+    .word   .LstrInternalError
+    .endif
+
+
+/* continuation for OP_FILLED_NEW_ARRAY_RANGE */
+
+    /*
+     * On entry:
+     *  r0 holds array class
+     *  r10 holds AA or BA
+     */
+.LOP_FILLED_NEW_ARRAY_RANGE_continue:
+    ldr     r3, [r0, #offClassObject_descriptor] @ r3<- arrayClass->descriptor
+    mov     r2, #ALLOC_DONT_TRACK       @ r2<- alloc flags
+    ldrb    r3, [r3, #1]                @ r3<- descriptor[1]
+    .if     1
+    mov     r1, r10                     @ r1<- AA (length)
+    .else
+    mov     r1, r10, lsr #4             @ r1<- B (length)
+    .endif
+    cmp     r3, #'I'                    @ array of ints?
+    cmpne   r3, #'L'                    @ array of objects?
+    cmpne   r3, #'['                    @ array of arrays?
+    mov     r9, r1                      @ save length in r9
+    bne     .LOP_FILLED_NEW_ARRAY_RANGE_notimpl         @ no, not handled yet
+    bl      dvmAllocArrayByClass        @ r0<- call(arClass, length, flags)
+    cmp     r0, #0                      @ null return?
+    beq     common_exceptionThrown      @ alloc failed, handle exception
+
+    FETCH(r1, 2)                        @ r1<- FEDC or CCCC
+    str     r0, [rGLUE, #offGlue_retval]    @ retval.l <- new array
+    add     r0, r0, #offArrayObject_contents @ r0<- newArray->contents
+    subs    r9, r9, #1                  @ length--, check for neg
+    FETCH_ADVANCE_INST(3)               @ advance to next instr, load rINST
+    bmi     2f                          @ was zero, bail
+
+    @ copy values from registers into the array
+    @ r0=array, r1=CCCC/FEDC, r9=length (from AA or B), r10=AA/BA
+    .if     1
+    add     r2, rFP, r1, lsl #2         @ r2<- &fp[CCCC]
+1:  ldr     r3, [r2], #4                @ r3<- *r2++
+    subs    r9, r9, #1                  @ count--
+    str     r3, [r0], #4                @ *contents++ = vX
+    bpl     1b
+    @ continue at 2
+    .else
+    cmp     r9, #4                      @ length was initially 5?
+    and     r2, r10, #15                @ r2<- A
+    bne     1f                          @ <= 4 args, branch
+    GET_VREG(r3, r2)                    @ r3<- vA
+    sub     r9, r9, #1                  @ count--
+    str     r3, [r0, #16]               @ contents[4] = vA
+1:  and     r2, r1, #15                 @ r2<- F/E/D/C
+    GET_VREG(r3, r2)                    @ r3<- vF/vE/vD/vC
+    mov     r1, r1, lsr #4              @ r1<- next reg in low 4
+    subs    r9, r9, #1                  @ count--
+    str     r3, [r0], #4                @ *contents++ = vX
+    bpl     1b
+    @ continue at 2
+    .endif
+
+2:
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    GOTO_OPCODE(ip)                     @ execute it
+
+    /*
+     * Throw an exception indicating that we have not implemented this
+     * mode of filled-new-array.
+     */
+.LOP_FILLED_NEW_ARRAY_RANGE_notimpl:
+    ldr     r0, .L_strInternalError
+    ldr     r1, .L_strFilledNewArrayNotImpl
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+    .if     (!1)                 @ define in one or the other, not both
+.L_strFilledNewArrayNotImpl:
+    .word   .LstrFilledNewArrayNotImpl
+.L_strInternalError:
+    .word   .LstrInternalError
+    .endif
+
+
+/* continuation for OP_CMPL_FLOAT */
+
+    @ Test for NaN with a second comparison.  EABI forbids testing bit
+    @ patterns, and we can't represent 0x7fc00000 in immediate form, so
+    @ make the library call.
+.LOP_CMPL_FLOAT_gt_or_nan:
+    mov     r1, r9                      @ reverse order
+    mov     r0, r10
+    bl      __aeabi_cfcmple             @ r0<- Z set if eq, C clear if <
+    @bleq    common_abort
+    movcc   r1, #1                      @ (greater than) r1<- 1
+    bcc     .LOP_CMPL_FLOAT_finish
+    mvn     r1, #0                            @ r1<- 1 or -1 for NaN
+    b       .LOP_CMPL_FLOAT_finish
+
+
+#if 0       /* "clasic" form */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    GET_VREG(r9, r2)                    @ r9<- vBB
+    GET_VREG(r10, r3)                   @ r10<- vCC
+    mov     r0, r9                      @ r0<- vBB
+    mov     r1, r10                     @ r1<- vCC
+    bl      __aeabi_fcmpeq              @ r0<- (vBB == vCC)
+    cmp     r0, #0                      @ equal?
+    movne   r1, #0                      @ yes, result is 0
+    bne     OP_CMPL_FLOAT_finish
+    mov     r0, r9                      @ r0<- vBB
+    mov     r1, r10                     @ r1<- vCC
+    bl      __aeabi_fcmplt              @ r0<- (vBB < vCC)
+    cmp     r0, #0                      @ less than?
+    b       OP_CMPL_FLOAT_continue
+@%break
+
+OP_CMPL_FLOAT_continue:
+    mvnne   r1, #0                      @ yes, result is -1
+    bne     OP_CMPL_FLOAT_finish
+    mov     r0, r9                      @ r0<- vBB
+    mov     r1, r10                     @ r1<- vCC
+    bl      __aeabi_fcmpgt              @ r0<- (vBB > vCC)
+    cmp     r0, #0                      @ greater than?
+    beq     OP_CMPL_FLOAT_nan               @ no, must be NaN
+    mov     r1, #1                      @ yes, result is 1
+    @ fall through to _finish
+
+OP_CMPL_FLOAT_finish:
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r3)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+    /*
+     * This is expected to be uncommon, so we double-branch (once to here,
+     * again back to _finish).
+     */
+OP_CMPL_FLOAT_nan:
+    mvn     r1, #0                            @ r1<- 1 or -1 for NaN
+    b       OP_CMPL_FLOAT_finish
+
+#endif
+
+
+/* continuation for OP_CMPG_FLOAT */
+
+    @ Test for NaN with a second comparison.  EABI forbids testing bit
+    @ patterns, and we can't represent 0x7fc00000 in immediate form, so
+    @ make the library call.
+.LOP_CMPG_FLOAT_gt_or_nan:
+    mov     r1, r9                      @ reverse order
+    mov     r0, r10
+    bl      __aeabi_cfcmple             @ r0<- Z set if eq, C clear if <
+    @bleq    common_abort
+    movcc   r1, #1                      @ (greater than) r1<- 1
+    bcc     .LOP_CMPG_FLOAT_finish
+    mov     r1, #1                            @ r1<- 1 or -1 for NaN
+    b       .LOP_CMPG_FLOAT_finish
+
+
+#if 0       /* "clasic" form */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    GET_VREG(r9, r2)                    @ r9<- vBB
+    GET_VREG(r10, r3)                   @ r10<- vCC
+    mov     r0, r9                      @ r0<- vBB
+    mov     r1, r10                     @ r1<- vCC
+    bl      __aeabi_fcmpeq              @ r0<- (vBB == vCC)
+    cmp     r0, #0                      @ equal?
+    movne   r1, #0                      @ yes, result is 0
+    bne     OP_CMPG_FLOAT_finish
+    mov     r0, r9                      @ r0<- vBB
+    mov     r1, r10                     @ r1<- vCC
+    bl      __aeabi_fcmplt              @ r0<- (vBB < vCC)
+    cmp     r0, #0                      @ less than?
+    b       OP_CMPG_FLOAT_continue
+@%break
+
+OP_CMPG_FLOAT_continue:
+    mvnne   r1, #0                      @ yes, result is -1
+    bne     OP_CMPG_FLOAT_finish
+    mov     r0, r9                      @ r0<- vBB
+    mov     r1, r10                     @ r1<- vCC
+    bl      __aeabi_fcmpgt              @ r0<- (vBB > vCC)
+    cmp     r0, #0                      @ greater than?
+    beq     OP_CMPG_FLOAT_nan               @ no, must be NaN
+    mov     r1, #1                      @ yes, result is 1
+    @ fall through to _finish
+
+OP_CMPG_FLOAT_finish:
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r3)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+    /*
+     * This is expected to be uncommon, so we double-branch (once to here,
+     * again back to _finish).
+     */
+OP_CMPG_FLOAT_nan:
+    mov     r1, #1                            @ r1<- 1 or -1 for NaN
+    b       OP_CMPG_FLOAT_finish
+
+#endif
+
+
+/* continuation for OP_CMPL_DOUBLE */
+
+    @ Test for NaN with a second comparison.  EABI forbids testing bit
+    @ patterns, and we can't represent 0x7fc00000 in immediate form, so
+    @ make the library call.
+.LOP_CMPL_DOUBLE_gt_or_nan:
+    ldmia   r10, {r0-r1}                @ reverse order
+    ldmia   r9, {r2-r3}
+    bl      __aeabi_cdcmple             @ r0<- Z set if eq, C clear if <
+    @bleq    common_abort
+    movcc   r1, #1                      @ (greater than) r1<- 1
+    bcc     .LOP_CMPL_DOUBLE_finish
+    mvn     r1, #0                            @ r1<- 1 or -1 for NaN
+    b       .LOP_CMPL_DOUBLE_finish
+
+
+/* continuation for OP_CMPG_DOUBLE */
+
+    @ Test for NaN with a second comparison.  EABI forbids testing bit
+    @ patterns, and we can't represent 0x7fc00000 in immediate form, so
+    @ make the library call.
+.LOP_CMPG_DOUBLE_gt_or_nan:
+    ldmia   r10, {r0-r1}                @ reverse order
+    ldmia   r9, {r2-r3}
+    bl      __aeabi_cdcmple             @ r0<- Z set if eq, C clear if <
+    @bleq    common_abort
+    movcc   r1, #1                      @ (greater than) r1<- 1
+    bcc     .LOP_CMPG_DOUBLE_finish
+    mov     r1, #1                            @ r1<- 1 or -1 for NaN
+    b       .LOP_CMPG_DOUBLE_finish
+
+
+/* continuation for OP_CMP_LONG */
+
+.LOP_CMP_LONG_less:
+    mvn     r1, #0                      @ r1<- -1
+    @ Want to cond code the next mov so we can avoid branch, but don't see it;
+    @ instead, we just replicate the tail end.
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r9)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+.LOP_CMP_LONG_greater:
+    mov     r1, #1                      @ r1<- 1
+    @ fall through to _finish
+
+.LOP_CMP_LONG_finish:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r9)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_AGET_WIDE */
+
+.LOP_AGET_WIDE_finish:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldrd    r2, [r0, #offArrayObject_contents]  @ r2/r3<- vBB[vCC]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r2-r3}                 @ vAA/vAA+1<- r2/r3
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_APUT_WIDE */
+
+.LOP_APUT_WIDE_finish:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldmia   r9, {r2-r3}                 @ r2/r3<- vAA/vAA+1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strd    r2, [r0, #offArrayObject_contents]  @ r2/r3<- vBB[vCC]
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_APUT_OBJECT */
+    /*
+     * On entry:
+     *  r1 = vBB (arrayObj)
+     *  r9 = vAA (obj)
+     *  r10 = offset into array (vBB + vCC * width)
+     */
+.LOP_APUT_OBJECT_finish:
+    cmp     r9, #0                      @ storing null reference?
+    beq     .LOP_APUT_OBJECT_skip_check      @ yes, skip type checks
+    ldr     r0, [r9, #offObject_clazz]  @ r0<- obj->clazz
+    ldr     r1, [r1, #offObject_clazz]  @ r1<- arrayObj->clazz
+    bl      dvmCanPutArrayElement       @ test object type vs. array type
+    cmp     r0, #0                      @ okay?
+    beq     common_errArrayStore        @ no
+.LOP_APUT_OBJECT_skip_check:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r9, [r10, #offArrayObject_contents] @ vBB[vCC]<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_finish:
+    @bl      common_squeak0
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    ldr   r0, [r9, r3]                @ r0<- obj.field (8/16/32 bits)
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET_WIDE */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_WIDE_finish:
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    ldrd    r0, [r9, r3]                @ r0/r1<- obj.field (64-bit align ok)
+    and     r2, r2, #15                 @ r2<- A
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    add     r3, rFP, r2, lsl #2         @ r3<- &fp[A]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r3, {r0-r1}                 @ fp[A]<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET_OBJECT */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_OBJECT_finish:
+    @bl      common_squeak0
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    ldr   r0, [r9, r3]                @ r0<- obj.field (8/16/32 bits)
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET_BOOLEAN */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_BOOLEAN_finish:
+    @bl      common_squeak1
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    ldr   r0, [r9, r3]                @ r0<- obj.field (8/16/32 bits)
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET_BYTE */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_BYTE_finish:
+    @bl      common_squeak2
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    ldr   r0, [r9, r3]                @ r0<- obj.field (8/16/32 bits)
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET_CHAR */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_CHAR_finish:
+    @bl      common_squeak3
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    ldr   r0, [r9, r3]                @ r0<- obj.field (8/16/32 bits)
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET_SHORT */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_SHORT_finish:
+    @bl      common_squeak4
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    ldr   r0, [r9, r3]                @ r0<- obj.field (8/16/32 bits)
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_finish:
+    @bl      common_squeak0
+    mov     r1, rINST, lsr #8           @ r1<- A+
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    and     r1, r1, #15                 @ r1<- A
+    cmp     r9, #0                      @ check object for null
+    GET_VREG(r0, r1)                    @ r0<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT_WIDE */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_WIDE_finish:
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    cmp     r9, #0                      @ check object for null
+    and     r2, r2, #15                 @ r2<- A
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    add     r2, rFP, r2, lsl #2         @ r3<- &fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldmia   r2, {r0-r1}                 @ r0/r1<- fp[A]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strd    r0, [r9, r3]                @ obj.field (64 bits, aligned)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT_OBJECT */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_OBJECT_finish:
+    @bl      common_squeak0
+    mov     r1, rINST, lsr #8           @ r1<- A+
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    and     r1, r1, #15                 @ r1<- A
+    cmp     r9, #0                      @ check object for null
+    GET_VREG(r0, r1)                    @ r0<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT_BOOLEAN */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_BOOLEAN_finish:
+    @bl      common_squeak1
+    mov     r1, rINST, lsr #8           @ r1<- A+
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    and     r1, r1, #15                 @ r1<- A
+    cmp     r9, #0                      @ check object for null
+    GET_VREG(r0, r1)                    @ r0<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT_BYTE */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_BYTE_finish:
+    @bl      common_squeak2
+    mov     r1, rINST, lsr #8           @ r1<- A+
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    and     r1, r1, #15                 @ r1<- A
+    cmp     r9, #0                      @ check object for null
+    GET_VREG(r0, r1)                    @ r0<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT_CHAR */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_CHAR_finish:
+    @bl      common_squeak3
+    mov     r1, rINST, lsr #8           @ r1<- A+
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    and     r1, r1, #15                 @ r1<- A
+    cmp     r9, #0                      @ check object for null
+    GET_VREG(r0, r1)                    @ r0<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT_SHORT */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_SHORT_finish:
+    @bl      common_squeak4
+    mov     r1, rINST, lsr #8           @ r1<- A+
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    and     r1, r1, #15                 @ r1<- A
+    cmp     r9, #0                      @ check object for null
+    GET_VREG(r0, r1)                    @ r0<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_SGET */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SGET_WIDE */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_WIDE_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_WIDE_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SGET_OBJECT */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_OBJECT_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_OBJECT_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SGET_BOOLEAN */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_BOOLEAN_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_BOOLEAN_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SGET_BYTE */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_BYTE_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_BYTE_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SGET_CHAR */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_CHAR_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_CHAR_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SGET_SHORT */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_SHORT_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_SHORT_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SPUT_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT_WIDE */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     *  r9: &fp[AA]
+     */
+.LOP_SPUT_WIDE_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_WIDE_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT_OBJECT */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SPUT_OBJECT_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_OBJECT_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT_BOOLEAN */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SPUT_BOOLEAN_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_BOOLEAN_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT_BYTE */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SPUT_BYTE_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_BYTE_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT_CHAR */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SPUT_CHAR_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_CHAR_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT_SHORT */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SPUT_SHORT_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_SHORT_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_INVOKE_VIRTUAL */
+
+    /*
+     * At this point:
+     *  r0 = resolved base method
+     *  r10 = C or CCCC (index of first arg, which is the "this" ptr)
+     */
+.LOP_INVOKE_VIRTUAL_continue:
+    GET_VREG(r1, r10)                   @ r1<- "this" ptr
+    ldrh    r2, [r0, #offMethod_methodIndex]    @ r2<- baseMethod->methodIndex
+    cmp     r1, #0                      @ is "this" null?
+    beq     common_errNullObject        @ null "this", throw exception
+    ldr     r3, [r1, #offObject_clazz]  @ r1<- thisPtr->clazz
+    ldr     r3, [r3, #offClassObject_vtable]    @ r3<- thisPtr->clazz->vtable
+    ldr     r0, [r3, r2, lsl #2]        @ r3<- vtable[methodIndex]
+    bl      common_invokeMethodNoRange @ continue on
+
+
+/* continuation for OP_INVOKE_SUPER */
+
+    /*
+     * At this point:
+     *  r0 = resolved base method
+     *  r9 = method->clazz
+     */
+.LOP_INVOKE_SUPER_continue:
+    ldr     r1, [r9, #offClassObject_super]     @ r1<- method->clazz->super
+    ldrh    r2, [r0, #offMethod_methodIndex]    @ r2<- baseMethod->methodIndex
+    ldr     r3, [r1, #offClassObject_vtableCount]   @ r3<- super->vtableCount
+    EXPORT_PC()                         @ must export for invoke
+    cmp     r2, r3                      @ compare (methodIndex, vtableCount)
+    bcs     .LOP_INVOKE_SUPER_nsm             @ method not present in superclass
+    ldr     r1, [r1, #offClassObject_vtable]    @ r1<- ...clazz->super->vtable
+    ldr     r0, [r1, r2, lsl #2]        @ r3<- vtable[methodIndex]
+    bl      common_invokeMethodNoRange @ continue on
+
+.LOP_INVOKE_SUPER_resolve:
+    mov     r0, r9                      @ r0<- method->clazz
+    mov     r2, #METHOD_VIRTUAL         @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    bne     .LOP_INVOKE_SUPER_continue        @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+    /*
+     * Throw a NoSuchMethodError with the method name as the message.
+     *  r0 = resolved base method
+     */
+.LOP_INVOKE_SUPER_nsm:
+    ldr     r1, [r0, #offMethod_name]   @ r1<- method name
+    b       common_errNoSuchMethod
+
+
+/* continuation for OP_INVOKE_DIRECT */
+
+    /*
+     * On entry:
+     *  r1 = reference (BBBB or CCCC)
+     *  r10 = "this" register
+     */
+.LOP_INVOKE_DIRECT_resolve:
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    mov     r2, #METHOD_DIRECT          @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    GET_VREG(r2, r10)                   @ r2<- "this" ptr (reload)
+    bne     .LOP_INVOKE_DIRECT_finish          @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+
+/* continuation for OP_INVOKE_VIRTUAL_RANGE */
+
+    /*
+     * At this point:
+     *  r0 = resolved base method
+     *  r10 = C or CCCC (index of first arg, which is the "this" ptr)
+     */
+.LOP_INVOKE_VIRTUAL_RANGE_continue:
+    GET_VREG(r1, r10)                   @ r1<- "this" ptr
+    ldrh    r2, [r0, #offMethod_methodIndex]    @ r2<- baseMethod->methodIndex
+    cmp     r1, #0                      @ is "this" null?
+    beq     common_errNullObject        @ null "this", throw exception
+    ldr     r3, [r1, #offObject_clazz]  @ r1<- thisPtr->clazz
+    ldr     r3, [r3, #offClassObject_vtable]    @ r3<- thisPtr->clazz->vtable
+    ldr     r0, [r3, r2, lsl #2]        @ r3<- vtable[methodIndex]
+    bl      common_invokeMethodRange @ continue on
+
+
+/* continuation for OP_INVOKE_SUPER_RANGE */
+
+    /*
+     * At this point:
+     *  r0 = resolved base method
+     *  r9 = method->clazz
+     */
+.LOP_INVOKE_SUPER_RANGE_continue:
+    ldr     r1, [r9, #offClassObject_super]     @ r1<- method->clazz->super
+    ldrh    r2, [r0, #offMethod_methodIndex]    @ r2<- baseMethod->methodIndex
+    ldr     r3, [r1, #offClassObject_vtableCount]   @ r3<- super->vtableCount
+    EXPORT_PC()                         @ must export for invoke
+    cmp     r2, r3                      @ compare (methodIndex, vtableCount)
+    bcs     .LOP_INVOKE_SUPER_RANGE_nsm             @ method not present in superclass
+    ldr     r1, [r1, #offClassObject_vtable]    @ r1<- ...clazz->super->vtable
+    ldr     r0, [r1, r2, lsl #2]        @ r3<- vtable[methodIndex]
+    bl      common_invokeMethodRange @ continue on
+
+.LOP_INVOKE_SUPER_RANGE_resolve:
+    mov     r0, r9                      @ r0<- method->clazz
+    mov     r2, #METHOD_VIRTUAL         @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    bne     .LOP_INVOKE_SUPER_RANGE_continue        @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+    /*
+     * Throw a NoSuchMethodError with the method name as the message.
+     *  r0 = resolved base method
+     */
+.LOP_INVOKE_SUPER_RANGE_nsm:
+    ldr     r1, [r0, #offMethod_name]   @ r1<- method name
+    b       common_errNoSuchMethod
+
+
+/* continuation for OP_INVOKE_DIRECT_RANGE */
+
+    /*
+     * On entry:
+     *  r1 = reference (BBBB or CCCC)
+     *  r10 = "this" register
+     */
+.LOP_INVOKE_DIRECT_RANGE_resolve:
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    mov     r2, #METHOD_DIRECT          @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    GET_VREG(r2, r10)                   @ r2<- "this" ptr (reload)
+    bne     .LOP_INVOKE_DIRECT_RANGE_finish          @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+
+/* continuation for OP_FLOAT_TO_LONG */
+/*
+ * Convert the float in r0 to a long in r0/r1.
+ *
+ * We have to clip values to long min/max per the specification.  The
+ * expected common case is a "reasonable" value that converts directly
+ * to modest integer.  The EABI convert function isn't doing this for us.
+ */
+f2l_doconv:
+    stmfd   sp!, {r4, lr}
+    mov     r1, #0x5f000000             @ (float)maxlong
+    mov     r4, r0
+    bl      __aeabi_fcmpge              @ is arg >= maxlong?
+    cmp     r0, #0                      @ nonzero == yes
+    mvnne   r0, #0                      @ return maxlong (7fffffff)
+    mvnne   r1, #0x80000000
+    ldmnefd sp!, {r4, pc}
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, #0xdf000000             @ (float)minlong
+    bl      __aeabi_fcmple              @ is arg <= minlong?
+    cmp     r0, #0                      @ nonzero == yes
+    movne   r0, #0                      @ return minlong (80000000)
+    movne   r1, #0x80000000
+    ldmnefd sp!, {r4, pc}
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r4
+    bl      __aeabi_fcmpeq              @ is arg == self?
+    cmp     r0, #0                      @ zero == no
+    moveq   r1, #0                      @ return zero for NaN
+    ldmeqfd sp!, {r4, pc}
+
+    mov     r0, r4                      @ recover arg
+    bl      __aeabi_f2lz                @ convert float to long
+    ldmfd   sp!, {r4, pc}
+
+
+/* continuation for OP_DOUBLE_TO_LONG */
+/*
+ * Convert the double in r0/r1 to a long in r0/r1.
+ *
+ * We have to clip values to long min/max per the specification.  The
+ * expected common case is a "reasonable" value that converts directly
+ * to modest integer.  The EABI convert function isn't doing this for us.
+ */
+d2l_doconv:
+    stmfd   sp!, {r4, r5, lr}           @ save regs
+    ldr     r3, .LOP_DOUBLE_TO_LONG_max         @ (double)maxlong, hi
+    sub     sp, sp, #4                  @ align for EABI
+    mov     r2, #0                      @ (double)maxlong, lo
+    mov     r4, r0                      @ save r0
+    mov     r5, r1                      @  and r1
+    bl      __aeabi_dcmpge              @ is arg >= maxlong?
+    cmp     r0, #0                      @ nonzero == yes
+    mvnne   r0, #0                      @ return maxlong (7fffffffffffffff)
+    mvnne   r1, #0x80000000
+    bne     1f
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r5
+    ldr     r3, .LOP_DOUBLE_TO_LONG_min         @ (double)minlong, hi
+    mov     r2, #0                      @ (double)minlong, lo
+    bl      __aeabi_dcmple              @ is arg <= minlong?
+    cmp     r0, #0                      @ nonzero == yes
+    movne   r0, #0                      @ return minlong (8000000000000000)
+    movne   r1, #0x80000000
+    bne     1f
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r5
+    mov     r2, r4                      @ compare against self
+    mov     r3, r5
+    bl      __aeabi_dcmpeq              @ is arg == self?
+    cmp     r0, #0                      @ zero == no
+    moveq   r1, #0                      @ return zero for NaN
+    beq     1f
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r5
+    bl      __aeabi_d2lz                @ convert double to long
+
+1:
+    add     sp, sp, #4
+    ldmfd   sp!, {r4, r5, pc}
+
+.LOP_DOUBLE_TO_LONG_max:
+    .word   0x43e00000                  @ maxlong, as a double (high word)
+.LOP_DOUBLE_TO_LONG_min:
+    .word   0xc3e00000                  @ minlong, as a double (high word)
+
+
+/* continuation for OP_MUL_LONG */
+
+.LOP_MUL_LONG_finish:
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r0, {r9-r10}                @ vAA/vAA+1<- r9/r10
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_SHL_LONG */
+
+.LOP_SHL_LONG_finish:
+    mov     r0, r0, asl r2              @  r0<- r0 << r2
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_SHR_LONG */
+
+.LOP_SHR_LONG_finish:
+    mov     r1, r1, asr r2              @  r1<- r1 >> r2
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_USHR_LONG */
+
+.LOP_USHR_LONG_finish:
+    mov     r1, r1, lsr r2              @  r1<- r1 >>> r2
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_SHL_LONG_2ADDR */
+
+.LOP_SHL_LONG_2ADDR_finish:
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_SHR_LONG_2ADDR */
+
+.LOP_SHR_LONG_2ADDR_finish:
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_USHR_LONG_2ADDR */
+
+.LOP_USHR_LONG_2ADDR_finish:
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_EXECUTE_INLINE */
+
+    /*
+     * Extract args, call function.
+     *  r0 = #of args (0-4)
+     *  r10 = call index
+     *  lr = return addr, above  [DO NOT bl out of here w/o preserving LR]
+     *
+     * Other ideas:
+     * - Use a jump table from the main piece to jump directly into the
+     *   AND/LDR pairs.  Costs a data load, saves a branch.
+     * - Have five separate pieces that do the loading, so we can work the
+     *   interleave a little better.  Increases code size.
+     */
+.LOP_EXECUTE_INLINE_continue:
+    rsb     r0, r0, #4                  @ r0<- 4-r0
+    FETCH(r9, 2)                        @ r9<- FEDC
+    add     pc, pc, r0, lsl #3          @ computed goto, 2 instrs each
+    bl      common_abort                @ (skipped due to ARM prefetch)
+4:  and     ip, r9, #0xf000             @ isolate F
+    ldr     r3, [rFP, ip, lsr #10]      @ r3<- vF (shift right 12, left 2)
+3:  and     ip, r9, #0x0f00             @ isolate E
+    ldr     r2, [rFP, ip, lsr #6]       @ r2<- vE
+2:  and     ip, r9, #0x00f0             @ isolate D
+    ldr     r1, [rFP, ip, lsr #2]       @ r1<- vD
+1:  and     ip, r9, #0x000f             @ isolate C
+    ldr     r0, [rFP, ip, lsl #2]       @ r0<- vC
+0:
+    ldr     r9, .LOP_EXECUTE_INLINE_table       @ table of InlineOperation
+    LDR_PC  "[r9, r10, lsl #4]"         @ sizeof=16, "func" is first entry
+    @ (not reached)
+
+.LOP_EXECUTE_INLINE_table:
+    .word   gDvmInlineOpsTable
+
+
+    .size   dvmAsmSisterStart, .-dvmAsmSisterStart
+    .global dvmAsmSisterEnd
+dvmAsmSisterEnd:
+
+/* File: armv5te/footer.S */
+/*
+ * ===========================================================================
+ *  Common subroutines and data
+ * ===========================================================================
+ */
+
+    .text
+    .align  2
+
+/*
+ * Common code when a backward branch is taken.
+ *
+ * On entry:
+ *  r9 is PC adjustment *in bytes*
+ */
+common_backwardBranch:
+    mov     r0, #kInterpEntryInstr
+    bl      common_periodicChecks
+    FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/*
+ * Need to see if the thread needs to be suspended or debugger/profiler
+ * activity has begun.
+ *
+ * TODO: if JDWP isn't running, zero out pDebuggerActive pointer so we don't
+ * have to do the second ldr.
+ *
+ * TODO: reduce this so we're just checking a single location.
+ *
+ * On entry:
+ *  r0 is reentry type, e.g. kInterpEntryInstr
+ *  r9 is trampoline PC adjustment *in bytes*
+ */
+common_periodicChecks:
+    ldr     r3, [rGLUE, #offGlue_pSelfSuspendCount] @ r3<- &suspendCount
+
+#if defined(WITH_DEBUGGER)
+    ldr     r1, [rGLUE, #offGlue_pDebuggerActive]   @ r1<- &debuggerActive
+#endif
+#if defined(WITH_PROFILER)
+    ldr     r2, [rGLUE, #offGlue_pActiveProfilers]  @ r2<- &activeProfilers
+#endif
+
+    ldr     r3, [r3]                    @ r3<- suspendCount (int)
+
+#if defined(WITH_DEBUGGER)
+    ldrb    r1, [r1]                    @ r1<- debuggerActive (boolean)
+#endif
+#if defined (WITH_PROFILER)
+    ldr     r2, [r2]                    @ r2<- activeProfilers (int)
+#endif
+
+    cmp     r3, #0                      @ suspend pending?
+    bne     2f                          @ yes, check suspend
+
+#if defined(WITH_DEBUGGER) || defined(WITH_PROFILER)
+# if defined(WITH_DEBUGGER) && defined(WITH_PROFILER)
+    orrs    r1, r1, r2                  @ r1<- r1 | r2
+    cmp     r1, #0                      @ debugger attached or profiler started?
+# elif defined(WITH_DEBUGGER)
+    cmp     r1, #0                      @ debugger attached?
+# elif defined(WITH_PROFILER)
+    cmp     r2, #0                      @ profiler started?
+# endif
+    bne     3f                          @ debugger/profiler, switch interp
+#endif
+
+    bx      lr                          @ nothing to do, return
+
+2:  @ check suspend
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- glue->self
+    b       dvmCheckSuspendPending      @ suspend if necessary, then return
+
+3:  @ debugger/profiler enabled, bail out
+    add     rPC, rPC, r9                @ update rPC
+    str     r0, [rGLUE, #offGlue_entryPoint]
+    mov     r1, #1                      @ "want switch" = true
+    b       common_gotoBail
+
+
+/*
+ * The equivalent of "goto bail", this calls through the "bail handler".
+ *
+ * State registers will be saved to the "glue" area before bailing.
+ *
+ * On entry:
+ *  r1 is "bool changeInterp", indicating if we want to switch to the
+ *     other interpreter or just bail all the way out
+ */
+common_gotoBail:
+    SAVE_PC_FP_TO_GLUE()                @ export state to "glue"
+    mov     r0, rGLUE                   @ r0<- glue ptr
+    b       dvmMterpStdBail             @ call(glue, changeInterp)
+
+    @add     r1, r1, #1                  @ using (boolean+1)
+    @add     r0, rGLUE, #offGlue_jmpBuf  @ r0<- &glue->jmpBuf
+    @bl      _longjmp                    @ does not return
+    @bl      common_abort
+
+
+/*
+ * Common code for method invocation with range.
+ *
+ * On entry:
+ *  r0 is "Method* methodToCall", the method we're trying to call
+ */
+common_invokeMethodRange:
+.LinvokeNewRange:
+    @ prepare to copy args to "outs" area of current frame
+    movs    r2, rINST, lsr #8           @ r2<- AA (arg count) -- test for zero
+    SAVEAREA_FROM_FP(r10, rFP)          @ r10<- stack save area
+    beq     .LinvokeArgsDone            @ if no args, skip the rest
+    FETCH(r1, 2)                        @ r1<- CCCC
+
+    @ r0=methodToCall, r1=CCCC, r2=count, r10=outs
+    @ (very few methods have > 10 args; could unroll for common cases)
+    add     r3, rFP, r1, lsl #2         @ r3<- &fp[CCCC]
+    sub     r10, r10, r2, lsl #2        @ r10<- "outs" area, for call args
+1:  ldr     r1, [r3], #4                @ val = *fp++
+    subs    r2, r2, #1                  @ count--
+    str     r1, [r10], #4               @ *outs++ = val
+    bne     1b                          @ ...while count != 0
+    b       .LinvokeArgsDone
+
+/*
+ * Common code for method invocation without range.
+ *
+ * On entry:
+ *  r0 is "Method* methodToCall", the method we're trying to call
+ */
+common_invokeMethodNoRange:
+.LinvokeNewNoRange:
+    @ prepare to copy args to "outs" area of current frame
+    movs    r2, rINST, lsr #12          @ r2<- B (arg count) -- test for zero
+    SAVEAREA_FROM_FP(r10, rFP)          @ r10<- stack save area
+    beq     .LinvokeArgsDone            @ if no args, skip the rest
+    FETCH(r1, 2)                        @ r1<- GFED
+
+    @ r0=methodToCall, r1=GFED, r2=count, r10=outs
+.LinvokeNonRange:
+    rsb     r2, r2, #5                  @ r2<- 5-r2
+    add     pc, pc, r2, lsl #4          @ computed goto, 4 instrs each
+    bl      common_abort                @ (skipped due to ARM prefetch)
+5:  and     ip, rINST, #0x0f00          @ isolate A
+    ldr     r3, [rFP, ip, lsr #6]       @ r3<- vA (shift right 8, left 2)
+    mov     r0, r0                      @ nop
+    str     r3, [r10, #-4]!             @ *--outs = vA
+4:  and     ip, r1, #0xf000             @ isolate G
+    ldr     r3, [rFP, ip, lsr #10]      @ r3<- vG (shift right 12, left 2)
+    mov     r0, r0                      @ nop
+    str     r3, [r10, #-4]!             @ *--outs = vG
+3:  and     ip, r1, #0x0f00             @ isolate F
+    ldr     r3, [rFP, ip, lsr #6]       @ r3<- vF
+    mov     r0, r0                      @ nop
+    str     r3, [r10, #-4]!             @ *--outs = vF
+2:  and     ip, r1, #0x00f0             @ isolate E
+    ldr     r3, [rFP, ip, lsr #2]       @ r3<- vE
+    mov     r0, r0                      @ nop
+    str     r3, [r10, #-4]!             @ *--outs = vE
+1:  and     ip, r1, #0x000f             @ isolate D
+    ldr     r3, [rFP, ip, lsl #2]       @ r3<- vD
+    mov     r0, r0                      @ nop
+    str     r3, [r10, #-4]!             @ *--outs = vD
+0:  @ fall through to .LinvokeArgsDone
+
+.LinvokeArgsDone: @ r0=methodToCall
+    @ find space for the new stack frame, check for overflow
+    SAVEAREA_FROM_FP(r1, rFP)           @ r1<- stack save area
+    ldrh    r2, [r0, #offMethod_registersSize]  @ r2<- methodToCall->regsSize
+    ldrh    r3, [r0, #offMethod_outsSize]   @ r3<- methodToCall->outsSize
+    sub     r1, r1, r2, lsl #2          @ r1<- newFp (old savearea - regsSize)
+    SAVEAREA_FROM_FP(r10, r1)           @ r10<- newSaveArea
+@    bl      common_dumpRegs
+    ldr     r9, [rGLUE, #offGlue_interpStackEnd]    @ r9<- interpStackEnd
+    sub     r3, r10, r3, lsl #2         @ r3<- bottom (newsave - outsSize)
+    cmp     r3, r9                      @ bottom < interpStackEnd?
+    blt     .LstackOverflow             @ yes, this frame will overflow stack
+
+    @ set up newSaveArea
+#ifdef EASY_GDB
+    SAVEAREA_FROM_FP(ip, rFP)           @ ip<- stack save area
+    str     ip, [r10, #offStackSaveArea_prevSave]
+#endif
+    str     rFP, [r10, #offStackSaveArea_prevFrame]
+    str     rPC, [r10, #offStackSaveArea_savedPc]
+    str     r0, [r10, #offStackSaveArea_method]
+
+    ldr     r3, [r0, #offMethod_accessFlags] @ r3<- methodToCall->accessFlags
+    tst     r3, #ACC_NATIVE
+    bne     .LinvokeNative
+
+    /*
+    stmfd   sp!, {r0-r3}
+    bl      common_printNewline
+    mov     r0, rFP
+    mov     r1, #0
+    bl      dvmDumpFp
+    ldmfd   sp!, {r0-r3}
+    stmfd   sp!, {r0-r3}
+    mov     r0, r1
+    mov     r1, r10
+    bl      dvmDumpFp
+    bl      common_printNewline
+    ldmfd   sp!, {r0-r3}
+    */
+
+    @ Update "glue" values for the new method
+    @ r0=methodToCall, r1=newFp
+    ldr     r3, [r0, #offMethod_clazz]      @ r3<- method->clazz
+    str     r0, [rGLUE, #offGlue_method]    @ glue->method = methodToCall
+    ldr     r3, [r3, #offClassObject_pDvmDex] @ r3<- method->clazz->pDvmDex
+    ldr     rPC, [r0, #offMethod_insns]     @ rPC<- method->insns
+    str     r3, [rGLUE, #offGlue_methodClassDex] @ glue->methodClassDex = ...
+    ldr     r2, [rGLUE, #offGlue_self]      @ r2<- glue->self
+    FETCH_INST()                            @ load rINST from rPC
+    mov     rFP, r1                         @ fp = newFp
+    GET_INST_OPCODE(ip)                     @ extract opcode from rINST
+    str     r1, [r2, #offThread_curFrame]   @ self->curFrame = newFp
+    GOTO_OPCODE(ip)                         @ jump to next instruction
+
+.LinvokeNative:
+    @ Prep for the native call
+    @ r0=methodToCall, r1=newFp, r10=newSaveArea
+    ldr     r3, [rGLUE, #offGlue_self]      @ r3<- glue->self
+    ldr     r9, [r3, #offThread_jniLocal_nextEntry] @ r9<- thread->refNext
+    str     r1, [r3, #offThread_curFrame]   @ self->curFrame = newFp
+    str     r9, [r10, #offStackSaveArea_localRefTop] @newFp->localRefTop=refNext
+    mov     r9, r3                      @ r9<- glue->self (preserve)
+
+    mov     r2, r0                      @ r2<- methodToCall
+    mov     r0, r1                      @ r0<- newFp (points to args)
+    add     r1, rGLUE, #offGlue_retval  @ r1<- &retval
+
+#ifdef ASSIST_DEBUGGER
+    /* insert fake function header to help gdb find the stack frame */
+    b       .Lskip
+    .type   dalvik_mterp, %function
+dalvik_mterp:
+    .fnstart
+    MTERP_ENTRY1
+    MTERP_ENTRY2
+.Lskip:
+#endif
+
+    @mov     lr, pc                      @ set return addr
+    @ldr     pc, [r2, #offMethod_nativeFunc] @ pc<- methodToCall->nativeFunc
+    LDR_PC_LR "[r2, #offMethod_nativeFunc]"
+
+    @ native return; r9=self, r10=newSaveArea
+    @ equivalent to dvmPopJniLocals
+    ldr     r0, [r10, #offStackSaveArea_localRefTop] @ r0<- newSave->localRefTop
+    ldr     r1, [r9, #offThread_exception] @ check for exception
+    str     rFP, [r9, #offThread_curFrame]  @ self->curFrame = fp
+    cmp     r1, #0                      @ null?
+    str     r0, [r9, #offThread_jniLocal_nextEntry] @ self->refNext<- r0
+    bne     common_exceptionThrown      @ no, handle exception
+
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+.LstackOverflow:
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- self
+    bl      dvmHandleStackOverflow
+    b       common_exceptionThrown
+#ifdef ASSIST_DEBUGGER
+    .fnend
+#endif
+
+
+    /*
+     * Common code for method invocation, calling through "glue code".
+     *
+     * TODO: now that we have range and non-range invoke handlers, this
+     *       needs to be split into two.  Maybe just create entry points
+     *       that set r9 and jump here?
+     *
+     * On entry:
+     *  r0 is "Method* methodToCall", the method we're trying to call
+     *  r9 is "bool methodCallRange", indicating if this is a /range variant
+     */
+     .if    0
+.LinvokeOld:
+    sub     sp, sp, #8                  @ space for args + pad
+    FETCH(ip, 2)                        @ ip<- FEDC or CCCC
+    mov     r2, r0                      @ A2<- methodToCall
+    mov     r0, rGLUE                   @ A0<- glue
+    SAVE_PC_FP_TO_GLUE()                @ export state to "glue"
+    mov     r1, r9                      @ A1<- methodCallRange
+    mov     r3, rINST, lsr #8           @ A3<- AA
+    str     ip, [sp, #0]                @ A4<- ip
+    bl      dvmMterp_invokeMethod       @ call the C invokeMethod
+    add     sp, sp, #8                  @ remove arg area
+    b       common_resumeAfterGlueCall  @ continue to next instruction
+    .endif
+
+
+
+/*
+ * Common code for handling a return instruction.
+ *
+ * This does not return.
+ */
+common_returnFromMethod:
+.LreturnNew:
+    mov     r0, #kInterpEntryReturn
+    mov     r9, #0
+    bl      common_periodicChecks
+
+    SAVEAREA_FROM_FP(r0, rFP)           @ r0<- saveArea (old)
+    ldr     rFP, [r0, #offStackSaveArea_prevFrame] @ fp = saveArea->prevFrame
+    ldr     r2, [rFP, #(offStackSaveArea_method - sizeofStackSaveArea)]
+                                        @ r2<- method we're returning to
+    cmp     r2, #0                      @ is this a break frame?
+    mov     r1, #0                      @ "want switch" = false
+    beq     common_gotoBail             @ break frame, bail out completely
+
+    ldr     rPC, [r0, #offStackSaveArea_savedPc] @ pc = saveArea->savedPc
+    ldr     r3, [rGLUE, #offGlue_self]      @ r3<- glue->self
+    str     r2, [rGLUE, #offGlue_method]    @ glue->method = newSave->method
+    str     rFP, [r3, #offThread_curFrame]  @ self->curFrame = fp
+    ldr     r1, [r2, #offMethod_clazz]      @ r1<- method->clazz
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    ldr     r1, [r1, #offClassObject_pDvmDex]   @ r1<- method->clazz->pDvmDex
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [rGLUE, #offGlue_methodClassDex]
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+    /*
+     * Return handling, calls through "glue code".
+     */
+     .if    0
+.LreturnOld:
+    SAVE_PC_FP_TO_GLUE()                @ export state
+    mov     r0, rGLUE                   @ arg to function
+    bl      dvmMterp_returnFromMethod
+    b       common_resumeAfterGlueCall
+    .endif
+
+
+/*
+ * Somebody has thrown an exception.  Handle it.
+ *
+ * If the exception processing code returns to us (instead of falling
+ * out of the interpreter), continue with whatever the next instruction
+ * now happens to be.
+ *
+ * This does not return.
+ */
+common_exceptionThrown:
+.LexceptionNew:
+    mov     r0, #kInterpEntryThrow
+    mov     r9, #0
+    bl      common_periodicChecks
+
+    ldr     r10, [rGLUE, #offGlue_self] @ r10<- glue->self
+    ldr     r9, [r10, #offThread_exception] @ r9<- self->exception
+    mov     r1, r10                     @ r1<- self
+    mov     r0, r9                      @ r0<- exception
+    bl      dvmAddTrackedAlloc          @ don't let the exception be GCed
+    mov     r3, #0                      @ r3<- NULL
+    str     r3, [r10, #offThread_exception] @ self->exception = NULL
+
+    /* set up args and a local for "&fp" */
+    /* (str sp, [sp, #-4]!  would be perfect here, but is discouraged) */
+    str     rFP, [sp, #-4]!             @ *--sp = fp
+    mov     ip, sp                      @ ip<- &fp
+    mov     r3, #0                      @ r3<- false
+    str     ip, [sp, #-4]!              @ *--sp = &fp
+    ldr     r1, [rGLUE, #offGlue_method] @ r1<- glue->method
+    mov     r0, r10                     @ r0<- self
+    ldr     r1, [r1, #offMethod_insns]  @ r1<- method->insns
+    mov     r2, r9                      @ r2<- exception
+    sub     r1, rPC, r1                 @ r1<- pc - method->insns
+    mov     r1, r1, asr #1              @ r1<- offset in code units
+
+    /* call, r0 gets catchRelPc (a code-unit offset) */
+    bl      dvmFindCatchBlock           @ call(self, relPc, exc, scan?, &fp)
+
+    /* fix earlier stack overflow if necessary; may trash rFP */
+    ldrb    r1, [r10, #offThread_stackOverflowed]
+    cmp     r1, #0                      @ did we overflow earlier?
+    beq     1f                          @ no, skip ahead
+    mov     rFP, r0                     @ save relPc result in rFP
+    mov     r0, r10                     @ r0<- self
+    bl      dvmCleanupStackOverflow     @ call(self)
+    mov     r0, rFP                     @ restore result
+1:
+
+    /* update frame pointer and check result from dvmFindCatchBlock */
+    ldr     rFP, [sp, #4]               @ retrieve the updated rFP
+    cmp     r0, #0                      @ is catchRelPc < 0?
+    add     sp, sp, #8                  @ restore stack
+    bmi     .LnotCaughtLocally
+
+    /* adjust locals to match self->curFrame and updated PC */
+    SAVEAREA_FROM_FP(r1, rFP)           @ r1<- new save area
+    ldr     r1, [r1, #offStackSaveArea_method] @ r1<- new method
+    str     r1, [rGLUE, #offGlue_method]    @ glue->method = new method
+    ldr     r2, [r1, #offMethod_clazz]      @ r2<- method->clazz
+    ldr     r3, [r1, #offMethod_insns]      @ r3<- method->insns
+    ldr     r2, [r2, #offClassObject_pDvmDex] @ r2<- method->clazz->pDvmDex
+    add     rPC, r3, r0, asl #1             @ rPC<- method->insns + catchRelPc
+    str     r2, [rGLUE, #offGlue_methodClassDex] @ glue->pDvmDex = meth...
+
+    /* release the tracked alloc on the exception */
+    mov     r0, r9                      @ r0<- exception
+    mov     r1, r10                     @ r1<- self
+    bl      dvmReleaseTrackedAlloc      @ release the exception
+
+    /* restore the exception if the handler wants it */
+    FETCH_INST()                        @ load rINST from rPC
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    cmp     ip, #OP_MOVE_EXCEPTION      @ is it "move-exception"?
+    streq   r9, [r10, #offThread_exception] @ yes, restore the exception
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+.LnotCaughtLocally: @ r9=exception, r10=self
+    /* fix stack overflow if necessary */
+    ldrb    r1, [r10, #offThread_stackOverflowed]
+    cmp     r1, #0                      @ did we overflow earlier?
+    movne   r0, r10                     @ if yes: r0<- self
+    blne    dvmCleanupStackOverflow     @ if yes: call(self)
+
+    @ may want to show "not caught locally" debug messages here
+#if DVM_SHOW_EXCEPTION >= 2
+    /* call __android_log_print(prio, tag, format, ...) */
+    /* "Exception %s from %s:%d not caught locally" */
+    @ dvmLineNumFromPC(method, pc - method->insns)
+    ldr     r0, [rGLUE, #offGlue_method]
+    ldr     r1, [r0, #offMethod_insns]
+    sub     r1, rPC, r1
+    asr     r1, r1, #1
+    bl      dvmLineNumFromPC
+    str     r0, [sp, #-4]!
+    @ dvmGetMethodSourceFile(method)
+    ldr     r0, [rGLUE, #offGlue_method]
+    bl      dvmGetMethodSourceFile
+    str     r0, [sp, #-4]!
+    @ exception->clazz->descriptor
+    ldr     r3, [r9, #offObject_clazz]
+    ldr     r3, [r3, #offClassObject_descriptor]
+    @
+    ldr     r2, strExceptionNotCaughtLocally
+    ldr     r1, strLogTag
+    mov     r0, #3                      @ LOG_DEBUG
+    bl      __android_log_print
+#endif
+    str     r9, [r10, #offThread_exception] @ restore exception
+    mov     r0, r9                      @ r0<- exception
+    mov     r1, r10                     @ r1<- self
+    bl      dvmReleaseTrackedAlloc      @ release the exception
+    mov     r1, #0                      @ "want switch" = false
+    b       common_gotoBail             @ bail out
+
+
+    /*
+     * Exception handling, calls through "glue code".
+     */
+    .if     0
+.LexceptionOld:
+    SAVE_PC_FP_TO_GLUE()                @ export state
+    mov     r0, rGLUE                   @ arg to function
+    bl      dvmMterp_exceptionThrown
+    b       common_resumeAfterGlueCall
+    .endif
+
+
+/*
+ * After returning from a "glued" function, pull out the updated
+ * values and start executing at the next instruction.
+ */
+common_resumeAfterGlueCall:
+    LOAD_PC_FP_FROM_GLUE()              @ pull rPC and rFP out of glue
+    FETCH_INST()                        @ load rINST from rPC
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/*
+ * Invalid array index.
+ */
+common_errArrayIndex:
+    EXPORT_PC()
+    ldr     r0, strArrayIndexException
+    mov     r1, #0
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+/*
+ * Invalid array value.
+ */
+common_errArrayStore:
+    EXPORT_PC()
+    ldr     r0, strArrayStoreException
+    mov     r1, #0
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+/*
+ * Integer divide or mod by zero.
+ */
+common_errDivideByZero:
+    EXPORT_PC()
+    ldr     r0, strArithmeticException
+    ldr     r1, strDivideByZero
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+/*
+ * Attempt to allocate an array with a negative size.
+ */
+common_errNegativeArraySize:
+    EXPORT_PC()
+    ldr     r0, strNegativeArraySizeException
+    mov     r1, #0
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+/*
+ * Invocation of a non-existent method.
+ */
+common_errNoSuchMethod:
+    EXPORT_PC()
+    ldr     r0, strNoSuchMethodError
+    mov     r1, #0
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+/*
+ * We encountered a null object when we weren't expecting one.  We
+ * export the PC, throw a NullPointerException, and goto the exception
+ * processing code.
+ */
+common_errNullObject:
+    EXPORT_PC()
+    ldr     r0, strNullPointerException
+    mov     r1, #0
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+/*
+ * For debugging, cause an immediate fault.  The source address will
+ * be in lr (use a bl instruction to jump here).
+ */
+common_abort:
+    ldr     pc, .LdeadFood
+.LdeadFood:
+    .word   0xdeadf00d
+
+/*
+ * Spit out a "we were here", preserving all registers.  (The attempt
+ * to save ip won't work, but we need to save an even number of
+ * registers for EABI 64-bit stack alignment.)
+ */
+    .macro  SQUEAK num
+common_squeak\num:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    ldr     r0, strSqueak
+    mov     r1, #\num
+    bl      printf
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+    .endm
+
+    SQUEAK  0
+    SQUEAK  1
+    SQUEAK  2
+    SQUEAK  3
+    SQUEAK  4
+    SQUEAK  5
+
+/*
+ * Spit out the number in r0, preserving registers.
+ */
+common_printNum:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    mov     r1, r0
+    ldr     r0, strSqueak
+    bl      printf
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+
+/*
+ * Print a newline, preserving registers.
+ */
+common_printNewline:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    ldr     r0, strNewline
+    bl      printf
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+
+    /*
+     * Print the 32-bit quantity in r0 as a hex value, preserving registers.
+     */
+common_printHex:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    mov     r1, r0
+    ldr     r0, strPrintHex
+    bl      printf
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+
+/*
+ * Print the 64-bit quantity in r0-r1, preserving registers.
+ */
+common_printLong:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    mov     r3, r1
+    mov     r2, r0
+    ldr     r0, strPrintLong
+    bl      printf
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+
+/*
+ * Print full method info.  Pass the Method* in r0.  Preserves regs.
+ */
+common_printMethod:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bl      dvmMterpPrintMethod
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+
+/*
+ * Call a C helper function that dumps regs and possibly some
+ * additional info.  Requires the C function to be compiled in.
+ */
+    .if     0
+common_dumpRegs:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bl      dvmMterpDumpArmRegs
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+    .endif
+
+
+/*
+ * String references, must be close to the code that uses them.
+ */
+    .align  2
+strArithmeticException:
+    .word   .LstrArithmeticException
+strArrayIndexException:
+    .word   .LstrArrayIndexException
+strArrayStoreException:
+    .word   .LstrArrayStoreException
+strDivideByZero:
+    .word   .LstrDivideByZero
+strNegativeArraySizeException:
+    .word   .LstrNegativeArraySizeException
+strNoSuchMethodError:
+    .word   .LstrNoSuchMethodError
+strNullPointerException:
+    .word   .LstrNullPointerException
+
+strLogTag:
+    .word   .LstrLogTag
+strExceptionNotCaughtLocally:
+    .word   .LstrExceptionNotCaughtLocally
+
+strNewline:
+    .word   .LstrNewline
+strSqueak:
+    .word   .LstrSqueak
+strPrintHex:
+    .word   .LstrPrintHex
+strPrintLong:
+    .word   .LstrPrintLong
+
+/*
+ * Zero-terminated ASCII string data.
+ *
+ * On ARM we have two choices: do like gcc does, and LDR from a .word
+ * with the address, or use an ADR pseudo-op to get the address
+ * directly.  ADR saves 4 bytes and an indirection, but it's using a
+ * PC-relative addressing mode and hence has a limited range, which
+ * makes it not work well with mergeable string sections.
+ */
+    .section .rodata.str1.4,"aMS",%progbits,1
+
+.LstrBadEntryPoint:
+    .asciz  "Bad entry point %d\n"
+.LstrArithmeticException:
+    .asciz  "Ljava/lang/ArithmeticException;"
+.LstrArrayIndexException:
+    .asciz  "Ljava/lang/ArrayIndexOutOfBoundsException;"
+.LstrArrayStoreException:
+    .asciz  "Ljava/lang/ArrayStoreException;"
+.LstrClassCastException:
+    .asciz  "Ljava/lang/ClassCastException;"
+.LstrDivideByZero:
+    .asciz  "divide by zero"
+.LstrFilledNewArrayNotImpl:
+    .asciz  "filled-new-array only implemented for objects and 'int'"
+.LstrInternalError:
+    .asciz  "Ljava/lang/InternalError;"
+.LstrInstantiationError:
+    .asciz  "Ljava/lang/InstantiationError;"
+.LstrNegativeArraySizeException:
+    .asciz  "Ljava/lang/NegativeArraySizeException;"
+.LstrNoSuchMethodError:
+    .asciz  "Ljava/lang/NoSuchMethodError;"
+.LstrNullPointerException:
+    .asciz  "Ljava/lang/NullPointerException;"
+
+.LstrLogTag:
+    .asciz  "mterp"
+.LstrExceptionNotCaughtLocally:
+    .asciz  "Exception %s from %s:%d not caught locally\n"
+
+.LstrNewline:
+    .asciz  "\n"
+.LstrSqueak:
+    .asciz  "<%d>"
+.LstrPrintHex:
+    .asciz  "<0x%x>"
+.LstrPrintLong:
+    .asciz  "<%lld>"
+
+
diff --git a/vm/mterp/out/InterpAsm-x86.S b/vm/mterp/out/InterpAsm-x86.S
new file mode 100644
index 0000000..a80e59e
--- /dev/null
+++ b/vm/mterp/out/InterpAsm-x86.S
@@ -0,0 +1,8844 @@
+/*
+ * This file was generated automatically by gen-mterp.py for 'x86'.
+ *
+ * --> DO NOT EDIT <--
+ */
+
+/* File: x86/header.S */
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ * 32-bit x86 definitions and declarations.
+ */
+
+/*
+386 ABI general notes:
+
+Caller save set:
+   eax, edx, ecx, st(0)-st(7)
+Callee save set:
+   ebx, esi, edi, ebp
+Return regs:
+   32-bit in eax
+   64-bit in edx:eax (low-order 32 in eax)
+   fp on top of fp stack st(0)
+
+Parameters passed on stack, pushed right-to-left.  On entry to target, first
+parm is at 4(%esp).  Traditional entry code is:
+
+functEntry:
+    push    %ebp             # save old frame pointer
+    mov     %ebp,%esp        # establish new frame pointer
+    sub     FrameSize,%esp   # Allocate storage for spill, locals & outs
+
+Once past the prologue, arguments are referenced at ((argno + 2)*4)(%ebp)
+
+Alignment of stack not strictly required, but should be for performance.  We'll
+align frame sizes to 16-byte multiples.
+
+If we're not doing variable stack allocation (alloca), the frame pointer can be
+eliminated and all arg references adjusted to be esp relative.
+
+Mterp notes:
+
+Some key interpreter variables will be assigned to registers.  Note that each
+will also have an associated spill location (mostly used useful for those assigned
+to callee save registers).
+
+  nick     reg   purpose
+  rPC      edx   interpreted program counter, used for fetching instructions
+  rFP      esi   interpreted frame pointer, used for accessing locals and args
+  rIBASE   edi   Base pointer for instruction dispatch computed goto
+  rINST    bx    first 16-bit code of current instruction
+  rOPCODE  bl    opcode portion of instruction word
+  rINST_HI bh    high byte of instruction word, usually contains src/tgt reg names
+
+Notes:
+   o High order 16 bits of ebx must be zero on entry to handler
+   o rPC, rFP, rIBASE, rINST/rOPCODE valid on handler entry and exit
+   o eax and ecx are scratch, rINST/ebx sometimes scratch
+   o rPC is in the caller save set, and will be killed across external calls. Don't
+     forget to SPILL/UNSPILL it around call points
+
+*/
+
+#define rPC      %edx
+#define rFP      %esi
+#define rIBASE   %edi
+#define rINST_FULL %ebx
+#define rINST    %bx
+#define rINST_HI %bh
+#define rINST_LO %bl
+#define rOPCODE  %bl
+
+
+/* Frame diagram while executing dvmMterpStdRun, high to low addresses */
+#define IN_ARG0        (  8)
+#define CALLER_RP      (  4)
+#define PREV_FP        (  0) /* <- dvmMterpStdRun ebp */
+/* Spill offsets relative to %ebp */
+#define EDI_SPILL      ( -4)
+#define ESI_SPILL      ( -8)
+#define EDX_SPILL      (-12) /* <- esp following dmMterpStdRun header */
+#define rPC_SPILL      (-16)
+#define rFP_SPILL      (-20)
+#define rGLUE_SPILL    (-24)
+#define rIBASE_SPILL   (-28)
+#define rINST_FULL_SPILL    (-32)
+#define TMP_SPILL      (-36)
+#define LOCAL0_OFFSET  (-40)
+#define LOCAL1_OFFSET  (-44)
+#define LOCAL2_OFFSET  (-48)
+#define LOCAL3_OFFSET  (-52)
+/* Out Arg offsets, relative to %sp */
+#define OUT_ARG4       ( 16)
+#define OUT_ARG3       ( 12)
+#define OUT_ARG2       (  8)
+#define OUT_ARG1       (  4)
+#define OUT_ARG0       (  0)  /* <- dvmMterpStdRun esp */
+
+#define SPILL(reg) movl reg##,reg##_SPILL(%ebp)
+#define UNSPILL(reg) movl reg##_SPILL(%ebp),reg
+#define SPILL_TMP(reg) movl reg,TMP_SPILL(%ebp)
+#define UNSPILL_TMP(reg) movl TMP_SPILL(%ebp),reg
+
+
+/* save/restore the PC and/or FP from the glue struct */
+#define LOAD_PC_FROM_GLUE(_glu)     movl    offGlue_pc(_glu),rPC
+#define SAVE_PC_TO_GLUE(_glu)       movl    rPC,offGlue_pc(_glu)
+#define LOAD_FP_FROM_GLUE(_glu)     movl    offGlue_fp(_glu),rFP
+#define SAVE_FP_TO_GLUE(_glu)       movl    rFP,offGlue_fp(_glu)
+
+#define GET_GLUE(_reg)     movl   rGLUE_SPILL(%ebp),_reg
+
+/* The interpreter assumes a properly aligned stack on entry, and
+ * will preserve 16-byte alignment.
+ */
+
+/*
+ * "export" the PC to the interpreted stack frame, f/b/o future exception
+ * objects.  Must * be done *before* something calls dvmThrowException.
+ *
+ * In C this is "SAVEAREA_FROM_FP(fp)->xtra.currentPc = pc", i.e.
+ * fp - sizeof(StackSaveArea) + offsetof(SaveArea, xtra.currentPc)
+ *
+ * It's okay to do this more than once.
+ */
+#define EXPORT_PC() \
+    movl     rPC, (-sizeofStackSaveArea + offStackSaveArea_currentPc)(rFP)
+
+/*
+ * Given a frame pointer, find the stack save area.
+ *
+ * In C this is "((StackSaveArea*)(_fp) -1)".
+ */
+#define SAVEAREA_FROM_FP(_reg, _fpreg) \
+    leal    -sizeofStackSaveArea(_fpreg),_reg
+
+/*
+ * Fetch the next instruction from rPC into rINST.  Does not advance rPC.
+ */
+#define FETCH_INST()            movzwl    (rPC),rINST_FULL
+
+/*
+ * Fetch the nth instruction word from rPC into rINST.  Does not advance
+ * rPC, and _count is in words
+ */
+#define FETCH_INST_WORD(_count)  movzwl  _count*2(rPC),rINST_FULL
+
+/*
+ * Fetch instruction word indexed (used for branching).
+ * Index is in instruction word units.
+ */
+#define FETCH_INST_INDEXED(_reg) movzwl  (rPC,_reg,2),rINST_FULL
+
+/*
+ * Extract the opcode of the instruction in rINST
+ */
+#define EXTRACT_OPCODE(_reg)   movzx rOPCODE,_reg
+
+/*
+ * Advance rPC by instruction count
+ */
+#define ADVANCE_PC(_count)    leal  2*_count(rPC),rPC
+
+/*
+ * Advance rPC by branch offset in register
+ */
+#define ADVANCE_PC_INDEXED(_reg) leal (rPC,_reg,2),rPC
+
+/*
+ * Note: assumes opcode previously fetched and in rINST, and
+ *       %eax is killable at this point.
+ */
+#if 1
+.macro GOTO_NEXT
+    /* For computed next version */
+     movzx    rOPCODE,%eax
+     sall     $6,%eax
+     addl     rIBASE,%eax
+     jmp      *%eax
+.endm
+#else
+   /* For jump table version */
+.macro GOTO_NEXT
+     movzx   rOPCODE,%eax
+     jmp     *(rIBASE,%eax,4)
+.endm
+#endif
+
+/*
+ * Get/set the 32-bit value from a Dalvik register.
+ */
+#define GET_VREG(_reg, _vreg)   movl     (rFP,_vreg,4),_reg
+#define SET_VREG(_reg, _vreg)   movl     _reg,(rFP,_vreg,4)
+#define GET_VREG_WORD(_reg, _vreg, _offset)   movl     4*(_offset)(rFP,_vreg,4),_reg
+#define SET_VREG_WORD(_reg, _vreg, _offset)   movl     _reg,4*(_offset)(rFP,_vreg,4)
+
+/*
+ * This is a #include, not a %include, because we want the C pre-processor
+ * to expand the macros into assembler assignment statements.
+ */
+#include "../common/asm-constants.h"
+
+
+
+    .global dvmAsmInstructionStart
+    .type   dvmAsmInstructionStart, %function
+dvmAsmInstructionStart = .L_OP_NOP
+    .text
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NOP: /* 0x00 */
+/* File: x86/OP_NOP.S */
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE: /* 0x01 */
+/* File: x86/OP_MOVE.S */
+    /* for move, move-object, long-to-int */
+    /* op vA, vB */
+    movzbl rINST_HI,%eax         # eax<- BA
+    andb   $0xf,%al             # eax<- A
+    shrl   $12,rINST_FULL       # rINST_FULL<- B
+    GET_VREG(%ecx,rINST_FULL)
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    SET_VREG(%ecx,%eax)          # fp[A]<-fp[B]
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_FROM16: /* 0x02 */
+/* File: x86/OP_MOVE_FROM16.S */
+    /* for: move/from16, move-object/from16 */
+    /* op vAA, vBBBB */
+    movzx    rINST_HI,%eax            # eax <= AA
+    movw     2(rPC),rINST             # rINST <= BBBB
+    GET_VREG (%ecx,rINST_FULL)        # ecx<- fp[BBBB]
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG (%ecx,%eax)              # fp[AA]<- ecx]
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_16: /* 0x03 */
+/* File: x86/OP_MOVE_16.S */
+    /* for: move/16, move-object/16 */
+    /* op vAAAA, vBBBB */
+    movzwl    4(rPC),%ecx              # ecx<- BBBB
+    movzwl    2(rPC),%eax              # eax<- AAAA
+    GET_VREG(%ecx,%ecx)
+    FETCH_INST_WORD(3)
+    ADVANCE_PC(3)
+    SET_VREG(%ecx,%eax)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_WIDE: /* 0x04 */
+/* File: x86/OP_MOVE_WIDE.S */
+    /* move-wide vA, vB */
+    /* NOTE: regs can overlap, e.g. "move v6,v7" or "move v7,v6" */
+    movzbl    rINST_HI,%ecx                # ecx <- BA
+    sarl      $12,rINST_FULL              # rinst_FULL<- B
+    GET_VREG_WORD(%eax,rINST_FULL,0)       # eax<- v[B+0]
+    GET_VREG_WORD(rINST_FULL,rINST_FULL,1) # rINST_FULL<- v[B+1]
+    andb      $0xf,%cl                    # ecx <- A
+    SET_VREG_WORD(rINST_FULL,%ecx,1)       # v[A+1]<- rINST_FULL
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    SET_VREG_WORD(%eax,%ecx,0)             # v[A+0]<- eax
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_WIDE_FROM16: /* 0x05 */
+/* File: x86/OP_MOVE_WIDE_FROM16.S */
+    /* move-wide/from16 vAA, vBBBB */
+    /* NOTE: regs can overlap, e.g. "move v6,v7" or "move v7,v6" */
+    movzwl    2(rPC),%ecx              # ecx<- BBBB
+    movzbl    rINST_HI,%eax            # eax<- AAAA
+    GET_VREG_WORD(rINST_FULL,%ecx,0)   # rINST_FULL<- v[BBBB+0]
+    GET_VREG_WORD(%ecx,%ecx,1)         # ecx<- v[BBBB+1]
+    SET_VREG_WORD(rINST_FULL,%eax,0)   # v[AAAA+0]<- rINST_FULL
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG_WORD(%ecx,%eax,1)         # v[AAAA+1]<- eax
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_WIDE_16: /* 0x06 */
+/* File: x86/OP_MOVE_WIDE_16.S */
+    /* move-wide/16 vAAAA, vBBBB */
+    /* NOTE: regs can overlap, e.g. "move v6,v7" or "move v7,v6" */
+    movzwl    4(rPC),%ecx            # ecx<- BBBB
+    movzwl    2(rPC),%eax            # eax<- AAAA
+    GET_VREG_WORD(rINST_FULL,%ecx,0) # rINST_WORD<- v[BBBB+0]
+    GET_VREG_WORD(%ecx,%ecx,1)       # ecx<- v[BBBB+1]
+    SET_VREG_WORD(rINST_FULL,%eax,0) # v[AAAA+0]<- rINST_FULL
+    FETCH_INST_WORD(3)
+    ADVANCE_PC(3)
+    SET_VREG_WORD(%ecx,%eax,1)       # v[AAAA+1]<- ecx
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_OBJECT: /* 0x07 */
+/* File: x86/OP_MOVE_OBJECT.S */
+/* File: x86/OP_MOVE.S */
+    /* for move, move-object, long-to-int */
+    /* op vA, vB */
+    movzbl rINST_HI,%eax         # eax<- BA
+    andb   $0xf,%al             # eax<- A
+    shrl   $12,rINST_FULL       # rINST_FULL<- B
+    GET_VREG(%ecx,rINST_FULL)
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    SET_VREG(%ecx,%eax)          # fp[A]<-fp[B]
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_OBJECT_FROM16: /* 0x08 */
+/* File: x86/OP_MOVE_OBJECT_FROM16.S */
+/* File: x86/OP_MOVE_FROM16.S */
+    /* for: move/from16, move-object/from16 */
+    /* op vAA, vBBBB */
+    movzx    rINST_HI,%eax            # eax <= AA
+    movw     2(rPC),rINST             # rINST <= BBBB
+    GET_VREG (%ecx,rINST_FULL)        # ecx<- fp[BBBB]
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG (%ecx,%eax)              # fp[AA]<- ecx]
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_OBJECT_16: /* 0x09 */
+/* File: x86/OP_MOVE_OBJECT_16.S */
+/* File: x86/OP_MOVE_16.S */
+    /* for: move/16, move-object/16 */
+    /* op vAAAA, vBBBB */
+    movzwl    4(rPC),%ecx              # ecx<- BBBB
+    movzwl    2(rPC),%eax              # eax<- AAAA
+    GET_VREG(%ecx,%ecx)
+    FETCH_INST_WORD(3)
+    ADVANCE_PC(3)
+    SET_VREG(%ecx,%eax)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_RESULT: /* 0x0a */
+/* File: x86/OP_MOVE_RESULT.S */
+    /* for: move-result, move-result-object */
+    /* op vAA */
+    GET_GLUE(%eax)                         # eax<- rGLUE
+    movzx    rINST_HI,%ecx                 # ecx<- AA
+    movl     offGlue_retval(%eax),%eax     # eax<- glue->retval.l
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    SET_VREG (%eax,%ecx)                   # fp[AA]<- retval.l
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_RESULT_WIDE: /* 0x0b */
+/* File: x86/OP_MOVE_RESULT_WIDE.S */
+    /* move-result-wide vAA */
+    GET_GLUE(%ecx)
+    movzbl  rINST_HI,rINST_FULL         # rINST_FULL<- AA
+    movl    offGlue_retval(%ecx),%eax
+    movl    4+offGlue_retval(%ecx),%ecx
+    SET_VREG_WORD(%eax,rINST_FULL,0)    # v[AA+0] <- eax
+    SET_VREG_WORD(%ecx,rINST_FULL,1)    # v[AA+1] <- ecx
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_RESULT_OBJECT: /* 0x0c */
+/* File: x86/OP_MOVE_RESULT_OBJECT.S */
+/* File: x86/OP_MOVE_RESULT.S */
+    /* for: move-result, move-result-object */
+    /* op vAA */
+    GET_GLUE(%eax)                         # eax<- rGLUE
+    movzx    rINST_HI,%ecx                 # ecx<- AA
+    movl     offGlue_retval(%eax),%eax     # eax<- glue->retval.l
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    SET_VREG (%eax,%ecx)                   # fp[AA]<- retval.l
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_EXCEPTION: /* 0x0d */
+/* File: x86/OP_MOVE_EXCEPTION.S */
+    /* move-exception vAA */
+    GET_GLUE(%ecx)
+    movzbl  rINST_HI,rINST_FULL        # rINST_FULL<- AA
+    movl    offGlue_self(%ecx),%ecx    # ecx<- glue->self
+    movl    offThread_exception(%ecx),%eax # eax<- dvmGetException bypass
+    SET_VREG(%eax,rINST_FULL)          # fp[AA]<- exception object
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    movl    $0,offThread_exception(%ecx) # dvmClearException bypass
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RETURN_VOID: /* 0x0e */
+/* File: x86/OP_RETURN_VOID.S */
+    jmp       common_returnFromMethod
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RETURN: /* 0x0f */
+/* File: x86/OP_RETURN.S */
+    /*
+     * Return a 32-bit value.  Copies the return value into the "glue"
+     * structure, then jumps to the return handler.
+     *
+     * for: return, return-object
+     */
+    /* op vAA */
+    GET_GLUE(%ecx)
+    movzbl  rINST_HI,rINST_FULL         # rINST_FULL<- AA
+    GET_VREG(%eax,rINST_FULL)           # eax<- vAA
+    movl    %eax,offGlue_retval(%ecx)   # retval.i <- AA
+    jmp     common_returnFromMethod
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RETURN_WIDE: /* 0x10 */
+/* File: x86/OP_RETURN_WIDE.S */
+    /*
+     * Return a 64-bit value.  Copies the return value into the "glue"
+     * structure, then jumps to the return handler.
+     */
+    /* return-wide vAA */
+    GET_GLUE(%ecx)
+    movzbl  rINST_HI,rINST_FULL            # rINST_FULL<- AA
+    GET_VREG_WORD(%eax,rINST_FULL,0)       # eax<- v[AA+0]
+    GET_VREG_WORD(rINST_FULL,rINST_FULL,1) # rINST_FULL<- v[AA+1]
+    movl    %eax,offGlue_retval(%ecx)
+    movl    rINST_FULL,4+offGlue_retval(%ecx)
+    jmp     common_returnFromMethod
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RETURN_OBJECT: /* 0x11 */
+/* File: x86/OP_RETURN_OBJECT.S */
+/* File: x86/OP_RETURN.S */
+    /*
+     * Return a 32-bit value.  Copies the return value into the "glue"
+     * structure, then jumps to the return handler.
+     *
+     * for: return, return-object
+     */
+    /* op vAA */
+    GET_GLUE(%ecx)
+    movzbl  rINST_HI,rINST_FULL         # rINST_FULL<- AA
+    GET_VREG(%eax,rINST_FULL)           # eax<- vAA
+    movl    %eax,offGlue_retval(%ecx)   # retval.i <- AA
+    jmp     common_returnFromMethod
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_4: /* 0x12 */
+/* File: x86/OP_CONST_4.S */
+    /* const/4 vA, #+B */
+    movsx   rINST_HI,%eax              # eax<-ssssssBx
+    movl    $0xf,%ecx
+    andl    %eax,%ecx                  # ecx<- A
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    sarl    $4,%eax
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_16: /* 0x13 */
+/* File: x86/OP_CONST_16.S */
+    /* const/16 vAA, #+BBBB */
+    movswl  2(rPC),%ecx                # ecx<- ssssBBBB
+    movzx   rINST_HI,%eax              # eax<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG(%ecx,%eax)                # vAA<- ssssBBBB
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST: /* 0x14 */
+/* File: x86/OP_CONST.S */
+    /* const vAA, #+BBBBbbbb */
+    movzbl    rINST_HI,%ecx           # ecx<- AA
+    movl      2(rPC),%eax               # grab all 32 bits at once
+    FETCH_INST_WORD(3)
+    ADVANCE_PC(3)
+    SET_VREG(%eax,%ecx)                 # vAA<- eax
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_HIGH16: /* 0x15 */
+/* File: x86/OP_CONST_HIGH16.S */
+    /* const/high16 vAA, #+BBBB0000 */
+    movzwl     2(rPC),%eax                # eax<- 0000BBBB
+    movzbl     rINST_HI,%ecx              # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    sall       $16,%eax                  # eax<- BBBB0000
+    SET_VREG(%eax,%ecx)                   # vAA<- eax
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_WIDE_16: /* 0x16 */
+/* File: x86/OP_CONST_WIDE_16.S */
+    /* const-wide/16 vAA, #+BBBB */
+    movswl    2(rPC),%eax               # eax<- ssssBBBB
+    SPILL(rPC)
+    movzbl    rINST_HI,%ecx             # ecx<- AA
+    FETCH_INST_WORD(2)
+    cltd                                # rPC:eax<- ssssssssssssBBBB
+    SET_VREG_WORD(rPC,%ecx,1)           # store msw
+    UNSPILL(rPC)
+    SET_VREG_WORD(%eax,%ecx,0)          # store lsw
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_WIDE_32: /* 0x17 */
+/* File: x86/OP_CONST_WIDE_32.S */
+    /* const-wide/32 vAA, #+BBBBbbbb */
+    movl     2(rPC),%eax               # eax<- BBBBbbbb
+    SPILL(rPC)
+    movzbl    rINST_HI,%ecx             # ecx<- AA
+    FETCH_INST_WORD(3)
+    cltd                                # rPC:eax<- ssssssssssssBBBB
+    SET_VREG_WORD(rPC,%ecx,1)           # store msw
+    UNSPILL(rPC)
+    SET_VREG_WORD(%eax,%ecx,0)          # store lsw
+    ADVANCE_PC(3)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_WIDE: /* 0x18 */
+/* File: x86/OP_CONST_WIDE.S */
+    /* const-wide vAA, #+HHHHhhhhBBBBbbbb */
+    movl      2(rPC),%eax         # eax<- lsw
+    movzbl    rINST_HI,%ecx       # ecx <- AA
+    movl      6(rPC),rINST_FULL   # rINST_FULL<- msw
+    leal      (rFP,%ecx,4),%ecx   # dst addr
+    movl      rINST_FULL,4(%ecx)
+    FETCH_INST_WORD(5)
+    movl      %eax,(%ecx)
+    ADVANCE_PC(5)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_WIDE_HIGH16: /* 0x19 */
+/* File: x86/OP_CONST_WIDE_HIGH16.S */
+    /* const-wide/high16 vAA, #+BBBB000000000000 */
+    movzwl     2(rPC),%eax                # eax<- 0000BBBB
+    movzbl     rINST_HI,%ecx              # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    sall       $16,%eax                  # eax<- BBBB0000
+    SET_VREG_WORD(%eax,%ecx,1)            # v[AA+1]<- eax
+    xorl       %eax,%eax
+    SET_VREG_WORD(%eax,%ecx,0)            # v[AA+0]<- eax
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_STRING: /* 0x1a */
+/* File: x86/OP_CONST_STRING.S */
+
+    /* const/string vAA, String@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax              # eax<- BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx# ecx<- glue->methodClassDex
+    movzbl    rINST_HI,rINST_FULL      # rINST_FULL<- AA
+    movl      offDvmDex_pResStrings(%ecx),%ecx # ecx<- dvmDex->pResStrings
+    movl      (%ecx,%eax,4),%eax       # eax<- rResString[BBBB]
+    movl      rINST_FULL,%ecx
+    FETCH_INST_WORD(2)
+    testl     %eax,%eax                # resolved yet?
+    je        .LOP_CONST_STRING_resolve
+    SET_VREG(%eax,%ecx)                # vAA<- rResString[BBBB]
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_STRING_JUMBO: /* 0x1b */
+/* File: x86/OP_CONST_STRING_JUMBO.S */
+
+    /* const/string vAA, String@BBBBBBBB */
+    GET_GLUE(%ecx)
+    movl      2(rPC),%eax              # eax<- BBBBBBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx# ecx<- glue->methodClassDex
+    movzbl    rINST_HI,rINST_FULL      # rINST_FULL<- AA
+    movl      offDvmDex_pResStrings(%ecx),%ecx # ecx<- dvmDex->pResStrings
+    movl      (%ecx,%eax,4),%eax       # eax<- rResString[BBBB]
+    movl      rINST_FULL,%ecx
+    FETCH_INST_WORD(3)
+    testl     %eax,%eax                # resolved yet?
+    je        .LOP_CONST_STRING_JUMBO_resolve
+    SET_VREG(%eax,%ecx)                # vAA<- rResString[BBBB]
+    ADVANCE_PC(3)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_CLASS: /* 0x1c */
+/* File: x86/OP_CONST_CLASS.S */
+
+    /* const/class vAA, Class@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax              # eax<- BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx# ecx<- glue->methodClassDex
+    movzbl    rINST_HI,rINST_FULL      # rINST_FULL<- AA
+    movl      offDvmDex_pResClasses(%ecx),%ecx # ecx<- dvmDex->pResClasses
+    movl      (%ecx,%eax,4),%eax       # eax<- rResClasses[BBBB]
+    movl      rINST_FULL,%ecx
+    FETCH_INST_WORD(2)
+    testl     %eax,%eax                # resolved yet?
+    je        .LOP_CONST_CLASS_resolve
+    SET_VREG(%eax,%ecx)                # vAA<- rResClasses[BBBB]
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MONITOR_ENTER: /* 0x1d */
+/* File: x86/OP_MONITOR_ENTER.S */
+    /*
+     * Synchronize on an object.
+     */
+    /* monitor-enter vAA */
+    GET_GLUE(%ecx)
+    movzbl  rINST_HI,rINST_FULL         # rINST_FULL<- AA
+    GET_VREG(%eax,rINST_FULL)           # eax<- vAA
+    movl    offGlue_self(%ecx),%ecx     # ecx<- glue->self
+    FETCH_INST_WORD(1)
+    testl   %eax,%eax                   # null object?
+#ifdef WITH_MONITOR_TRACKING
+    EXPORT_PC()
+#endif
+    jne     .LOP_MONITOR_ENTER_continue
+    jmp     common_errNullObject
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MONITOR_EXIT: /* 0x1e */
+/* File: x86/OP_MONITOR_EXIT.S */
+    /*
+     * Unlock an object.
+     *
+     * Exceptions that occur when unlocking a monitor need to appear as
+     * if they happened at the following instruction.  See the Dalvik
+     * instruction spec.
+     */
+    /* monitor-exit vAA */
+    movzbl  rINST_HI,rINST_FULL         # rINST_FULL<- AA
+    GET_VREG(%eax,rINST_FULL)
+    GET_GLUE(%ecx)
+    EXPORT_PC()
+    testl   %eax,%eax                   # null object?
+    je      common_errNullObject        # go if so
+    movl    offGlue_self(%ecx),%ecx     # ecx<- glue->self
+    movl    %eax,OUT_ARG1(%esp)
+    SPILL(rPC)
+    movl    %ecx,OUT_ARG0(%esp)
+    jmp     .LOP_MONITOR_EXIT_continue
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CHECK_CAST: /* 0x1f */
+/* File: x86/OP_CHECK_CAST.S */
+    /*
+     * Check to see if a cast from one class to another is allowed.
+     */
+    /* check-cast vAA, class@BBBB */
+    GET_GLUE(%ecx)
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    GET_VREG(rINST_FULL,rINST_FULL)     # rINST_FULL<- vAA (object)
+    movzwl    2(rPC),%eax               # eax<- BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx # ecx<- pDvmDex
+    testl     rINST_FULL,rINST_FULL     # is oject null?
+    movl      offDvmDex_pResClasses(%ecx),%ecx # ecx<- pDvmDex->pResClasses
+    je        .LOP_CHECK_CAST_okay          # null obj, cast always succeeds
+    movl      (%ecx,%eax,4),%eax        # eax<- resolved class
+    movl      offObject_clazz(rINST_FULL),%ecx # ecx<- obj->clazz
+    testl     %eax,%eax                 # have we resolved this before?
+    je        .LOP_CHECK_CAST_resolve       # no, go do it now
+.LOP_CHECK_CAST_resolved:
+    cmpl      %eax,%ecx                 # same class (trivial success)?
+    jne       .LOP_CHECK_CAST_fullcheck     # no, do full check
+.LOP_CHECK_CAST_okay:
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INSTANCE_OF: /* 0x20 */
+/* File: x86/OP_INSTANCE_OF.S */
+    /*
+     * Check to see if an object reference is an instance of a class.
+     *
+     * Most common situation is a non-null object, being compared against
+     * an already-resolved class.
+     */
+    /* instance-of vA, vB, class@CCCC */
+    movzbl  rINST_HI,%eax               # eax<- BA
+    sarl    $4,%eax                    # eax<- B
+    GET_VREG(%eax,%eax)                 # eax<- vB (obj)
+    GET_GLUE(%ecx)
+    testl   %eax,%eax                   # object null?
+    movl    offGlue_methodClassDex(%ecx),%ecx  # ecx<- pDvmDex
+    SPILL(rPC)
+    je      .LOP_INSTANCE_OF_store           # null obj, not instance, store it
+    movzwl  2(rPC),rPC                  # rPC<- CCCC
+    movl    offDvmDex_pResClasses(%ecx),%ecx # ecx<- pDvmDex->pResClasses
+    movl    (%ecx,rPC,4),%ecx           # ecx<- resolved class
+    movl    offObject_clazz(%eax),%eax  # eax<- obj->clazz
+    testl   %ecx,%ecx                   # have we resolved this before?
+    je      .LOP_INSTANCE_OF_resolve         # not resolved, do it now
+.LOP_INSTANCE_OF_resolved:  # eax<- obj->clazz, ecx<- resolved class
+    cmpl    %eax,%ecx                   # same class (trivial success)?
+    je      .LOP_INSTANCE_OF_trivial         # yes, trivial finish
+    jmp     .LOP_INSTANCE_OF_fullcheck       # no, do full check
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ARRAY_LENGTH: /* 0x21 */
+/* File: x86/OP_ARRAY_LENGTH.S */
+    /*
+     * Return the length of an array.
+     */
+   movzbl   rINST_HI,%eax              # eax<- BA
+   sarl     $12,rINST_FULL            # rINST_FULL<- B
+   GET_VREG(%ecx,rINST_FULL)           # ecx<- vB (object ref)
+   andb     $0xf,%al                  # eax<- A
+   testl    %ecx,%ecx                  # is null?
+   je       common_errNullObject
+   FETCH_INST_WORD(1)
+   movl     offArrayObject_length(%ecx),%ecx
+   ADVANCE_PC(1)
+   SET_VREG(%ecx,%eax)
+   GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEW_INSTANCE: /* 0x22 */
+/* File: x86/OP_NEW_INSTANCE.S */
+    /*
+     * Create a new instance of a class.
+     */
+    /* new-instance vAA, class@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax               # eax<- BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx  # ecx<- pDvmDex
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    movl      offDvmDex_pResClasses(%ecx),%ecx # ecx<- pDvmDex->pResClasses
+    EXPORT_PC()
+    movl      (%ecx,%eax,4),%ecx        # ecx<- resolved class
+    SPILL(rPC)
+    testl     %ecx,%ecx                 # resolved?
+    je        .LOP_NEW_INSTANCE_resolve       # no, go do it
+.LOP_NEW_INSTANCE_resolved:  # on entry, ecx<- class
+    cmpb      $CLASS_INITIALIZED,offClassObject_status(%ecx)
+    je        .LOP_NEW_INSTANCE_initialized
+    jmp       .LOP_NEW_INSTANCE_needinit
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEW_ARRAY: /* 0x23 */
+/* File: x86/OP_NEW_ARRAY.S */
+    /*
+     * Allocate an array of objects, specified with the array class
+     * and a count.
+     *
+     * The verifier guarantees that this is an array class, so we don't
+     * check for it here.
+     */
+    /* new-array vA, vB, class@CCCC */
+    GET_GLUE(%ecx)
+    EXPORT_PC()
+    movl    offGlue_methodClassDex(%ecx),%ecx # ecx<- pDvmDex
+    movzwl  2(rPC),%eax                       # eax<- CCCC
+    movl    offDvmDex_pResClasses(%ecx),%ecx  # ecx<- pDvmDex->pResClasses
+    movl    (%ecx,%eax,4),%ecx                # ecx<- resolved class
+    movzbl  rINST_HI,%eax
+    sarl    $4,%eax                          # eax<- B
+    GET_VREG(%eax,%eax)                       # eax<- vB (array length)
+    movzbl  rINST_HI,rINST_FULL
+    andb    $0xf,rINST_LO                    # rINST_FULL<- A
+    testl   %eax,%eax
+    js      common_errNegativeArraySize       # bail
+    testl   %ecx,%ecx                         # already resolved?
+    jne     .LOP_NEW_ARRAY_finish                # yes, fast path
+    jmp     .LOP_NEW_ARRAY_resolve               # resolve now
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FILLED_NEW_ARRAY: /* 0x24 */
+/* File: x86/OP_FILLED_NEW_ARRAY.S */
+    /*
+     * Create a new array with elements filled from registers.
+     *
+     * for: filled-new-array, filled-new-array/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, type@BBBB */
+    GET_GLUE(%eax)
+    movzbl  rINST_HI,rINST_FULL               # rINST_FULL<- AA or BA
+    movl    offGlue_methodClassDex(%eax),%eax # eax<- pDvmDex
+    movzwl  2(rPC),%ecx                       # ecx<- BBBB
+    movl    offDvmDex_pResClasses(%eax),%eax  # eax<- pDvmDex->pResClasses
+    SPILL(rPC)
+    movl    (%eax,%ecx,4),%eax                # eax<- resolved class
+    EXPORT_PC()
+    testl   %eax,%eax                         # already resolved?
+    jne     .LOP_FILLED_NEW_ARRAY_continue              # yes, continue
+    # less frequent path, so we'll redo some work
+    GET_GLUE(%eax)
+    movl    $0,OUT_ARG2(%esp)                # arg2<- false
+    movl    %ecx,OUT_ARG1(%esp)               # arg1<- BBBB
+    movl    offGlue_method(%eax),%eax         # eax<- glue->method
+    jmp     .LOP_FILLED_NEW_ARRAY_more
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FILLED_NEW_ARRAY_RANGE: /* 0x25 */
+/* File: x86/OP_FILLED_NEW_ARRAY_RANGE.S */
+/* File: x86/OP_FILLED_NEW_ARRAY.S */
+    /*
+     * Create a new array with elements filled from registers.
+     *
+     * for: filled-new-array, filled-new-array/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, type@BBBB */
+    GET_GLUE(%eax)
+    movzbl  rINST_HI,rINST_FULL               # rINST_FULL<- AA or BA
+    movl    offGlue_methodClassDex(%eax),%eax # eax<- pDvmDex
+    movzwl  2(rPC),%ecx                       # ecx<- BBBB
+    movl    offDvmDex_pResClasses(%eax),%eax  # eax<- pDvmDex->pResClasses
+    SPILL(rPC)
+    movl    (%eax,%ecx,4),%eax                # eax<- resolved class
+    EXPORT_PC()
+    testl   %eax,%eax                         # already resolved?
+    jne     .LOP_FILLED_NEW_ARRAY_RANGE_continue              # yes, continue
+    # less frequent path, so we'll redo some work
+    GET_GLUE(%eax)
+    movl    $0,OUT_ARG2(%esp)                # arg2<- false
+    movl    %ecx,OUT_ARG1(%esp)               # arg1<- BBBB
+    movl    offGlue_method(%eax),%eax         # eax<- glue->method
+    jmp     .LOP_FILLED_NEW_ARRAY_RANGE_more
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FILL_ARRAY_DATA: /* 0x26 */
+/* File: x86/OP_FILL_ARRAY_DATA.S */
+    /* fill-array-data vAA, +BBBBBBBB */
+    movl    2(rPC),%ecx                # ecx<- BBBBbbbb
+    movzbl  rINST_HI,rINST_FULL        # rINST_FULL<- AA
+    leal    (rPC,%ecx,2),%ecx          # ecx<- PC + BBBBbbbb*2
+    GET_VREG(%eax,rINST_FULL)
+    SPILL(rPC)
+    EXPORT_PC()
+    movl    %eax,OUT_ARG0(%esp)
+    movl    %ecx,OUT_ARG1(%esp)
+    call    dvmInterpHandleFillArrayData
+    UNSPILL(rPC)
+    FETCH_INST_WORD(3)
+    testl   %eax,%eax                   # exception thrown?
+    je      common_exceptionThrown
+    ADVANCE_PC(3)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_THROW: /* 0x27 */
+/* File: x86/OP_THROW.S */
+    /*
+     * Throw an exception object in the current thread.
+     */
+    /* throw vAA */
+    GET_GLUE(%ecx)
+    movzbl   rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    GET_VREG(%eax,rINST_FULL)          # eax<- exception object
+    movl     offGlue_self(%ecx),%ecx   # ecx<- glue->self
+    testl    %eax,%eax                 # null object?
+    je       common_errNullObject
+    movl     %eax,offThread_exception(%ecx) # thread->exception<- obj
+    jmp      common_exceptionThrown
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_GOTO: /* 0x28 */
+/* File: x86/OP_GOTO.S */
+    /*
+     * Unconditional branch, 8-bit offset.
+     *
+     * The branch distance is a signed code-unit offset, which we need to
+     * double to get a byte offset.
+     */
+    /* goto +AA */
+    movsbl  rINST_HI,rINST_FULL         # ebx<- ssssssAA
+    testl   rINST_FULL,rINST_FULL       # test for <0
+    js      common_backwardBranch
+    movl    rINST_FULL,%eax
+    FETCH_INST_INDEXED(%eax)
+    ADVANCE_PC_INDEXED(%eax)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_GOTO_16: /* 0x29 */
+/* File: x86/OP_GOTO_16.S */
+    /*
+     * Unconditional branch, 16-bit offset.
+     *
+     * The branch distance is a signed code-unit offset
+     */
+    /* goto/16 +AAAA */
+    movswl  2(rPC),rINST_FULL           # rINST_FULL<- ssssAAAA
+    testl   rINST_FULL,rINST_FULL       # test for <0
+    js      common_backwardBranch
+    movl    rINST_FULL,%eax
+    FETCH_INST_INDEXED(%eax)
+    ADVANCE_PC_INDEXED(%eax)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_GOTO_32: /* 0x2a */
+/* File: x86/OP_GOTO_32.S */
+    /*
+     * Unconditional branch, 32-bit offset.
+     *
+     * The branch distance is a signed code-unit offset.
+     *
+     * Unlike most opcodes, this one is allowed to branch to itself, so
+     * our "backward branch" test must be "<=0" instead of "<0".
+     */
+    /* goto/32 AAAAAAAA */
+    movl    2(rPC),rINST_FULL           # rINST_FULL<- AAAAAAAA
+    cmpl    $0,rINST_FULL              # test for <= 0
+    jle     common_backwardBranch
+    movl    rINST_FULL,%eax
+    FETCH_INST_INDEXED(%eax)
+    ADVANCE_PC_INDEXED(%eax)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_PACKED_SWITCH: /* 0x2b */
+/* File: x86/OP_PACKED_SWITCH.S */
+    /*
+     * Handle a packed-switch or sparse-switch instruction.  In both cases
+     * we decode it and hand it off to a helper function.
+     *
+     * We don't really expect backward branches in a switch statement, but
+     * they're perfectly legal, so we check for them here.
+     *
+     * for: packed-switch, sparse-switch
+     */
+    /* op vAA, +BBBB */
+    movzbl  rINST_HI,rINST_FULL         # rINST_FULL<- AA
+    movl    2(rPC),%ecx                 # ecx<- BBBBbbbb
+    GET_VREG(%eax,rINST_FULL)           # eax<- vAA
+    leal    (rPC,%ecx,2),%ecx           # ecx<- PC + BBBBbbbb*2
+    movl    %eax,OUT_ARG1(%esp)         # ARG1<- vAA
+    movl    %ecx,OUT_ARG0(%esp)         # ARG0<- switchData
+    SPILL(rPC)
+    call    dvmInterpHandlePackedSwitch
+    UNSPILL(rPC)
+    testl   %eax,%eax
+    movl    %eax,rINST_FULL             # set up word offset
+    jle     common_backwardBranch       # check on special actions
+    ADVANCE_PC_INDEXED(rINST_FULL)
+    FETCH_INST()
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPARSE_SWITCH: /* 0x2c */
+/* File: x86/OP_SPARSE_SWITCH.S */
+/* File: x86/OP_PACKED_SWITCH.S */
+    /*
+     * Handle a packed-switch or sparse-switch instruction.  In both cases
+     * we decode it and hand it off to a helper function.
+     *
+     * We don't really expect backward branches in a switch statement, but
+     * they're perfectly legal, so we check for them here.
+     *
+     * for: packed-switch, sparse-switch
+     */
+    /* op vAA, +BBBB */
+    movzbl  rINST_HI,rINST_FULL         # rINST_FULL<- AA
+    movl    2(rPC),%ecx                 # ecx<- BBBBbbbb
+    GET_VREG(%eax,rINST_FULL)           # eax<- vAA
+    leal    (rPC,%ecx,2),%ecx           # ecx<- PC + BBBBbbbb*2
+    movl    %eax,OUT_ARG1(%esp)         # ARG1<- vAA
+    movl    %ecx,OUT_ARG0(%esp)         # ARG0<- switchData
+    SPILL(rPC)
+    call    dvmInterpHandleSparseSwitch
+    UNSPILL(rPC)
+    testl   %eax,%eax
+    movl    %eax,rINST_FULL             # set up word offset
+    jle     common_backwardBranch       # check on special actions
+    ADVANCE_PC_INDEXED(rINST_FULL)
+    FETCH_INST()
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMPL_FLOAT: /* 0x2d */
+/* File: x86/OP_CMPL_FLOAT.S */
+/* File: x86/OP_CMPG_DOUBLE.S */
+    /* float/double_cmp[gl] vAA, vBB, vCC */
+    movzbl    3(rPC),%eax             # eax<- CC
+    movzbl    2(rPC),%ecx             # ecx<- BB
+    .if 0
+    fldl     (rFP,%eax,4)
+    fldl     (rFP,%ecx,4)
+    .else
+    flds     (rFP,%eax,4)
+    flds     (rFP,%ecx,4)
+    .endif
+    movzbl   rINST_HI,rINST_FULL
+    xorl     %ecx,%ecx
+    fucompp     # z if equal, p set if NaN, c set if st0 < st1
+    fnstsw   %ax
+    sahf
+    movl      rINST_FULL,%eax
+    FETCH_INST_WORD(2)
+    jp       .LOP_CMPL_FLOAT_isNaN
+    je       .LOP_CMPL_FLOAT_finish
+    sbbl     %ecx,%ecx
+    jb       .LOP_CMPL_FLOAT_finish
+    incl     %ecx
+.LOP_CMPL_FLOAT_finish:
+    SET_VREG(%ecx,%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMPG_FLOAT: /* 0x2e */
+/* File: x86/OP_CMPG_FLOAT.S */
+/* File: x86/OP_CMPG_DOUBLE.S */
+    /* float/double_cmp[gl] vAA, vBB, vCC */
+    movzbl    3(rPC),%eax             # eax<- CC
+    movzbl    2(rPC),%ecx             # ecx<- BB
+    .if 0
+    fldl     (rFP,%eax,4)
+    fldl     (rFP,%ecx,4)
+    .else
+    flds     (rFP,%eax,4)
+    flds     (rFP,%ecx,4)
+    .endif
+    movzbl   rINST_HI,rINST_FULL
+    xorl     %ecx,%ecx
+    fucompp     # z if equal, p set if NaN, c set if st0 < st1
+    fnstsw   %ax
+    sahf
+    movl      rINST_FULL,%eax
+    FETCH_INST_WORD(2)
+    jp       .LOP_CMPG_FLOAT_isNaN
+    je       .LOP_CMPG_FLOAT_finish
+    sbbl     %ecx,%ecx
+    jb       .LOP_CMPG_FLOAT_finish
+    incl     %ecx
+.LOP_CMPG_FLOAT_finish:
+    SET_VREG(%ecx,%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMPL_DOUBLE: /* 0x2f */
+/* File: x86/OP_CMPL_DOUBLE.S */
+/* File: x86/OP_CMPG_DOUBLE.S */
+    /* float/double_cmp[gl] vAA, vBB, vCC */
+    movzbl    3(rPC),%eax             # eax<- CC
+    movzbl    2(rPC),%ecx             # ecx<- BB
+    .if 1
+    fldl     (rFP,%eax,4)
+    fldl     (rFP,%ecx,4)
+    .else
+    flds     (rFP,%eax,4)
+    flds     (rFP,%ecx,4)
+    .endif
+    movzbl   rINST_HI,rINST_FULL
+    xorl     %ecx,%ecx
+    fucompp     # z if equal, p set if NaN, c set if st0 < st1
+    fnstsw   %ax
+    sahf
+    movl      rINST_FULL,%eax
+    FETCH_INST_WORD(2)
+    jp       .LOP_CMPL_DOUBLE_isNaN
+    je       .LOP_CMPL_DOUBLE_finish
+    sbbl     %ecx,%ecx
+    jb       .LOP_CMPL_DOUBLE_finish
+    incl     %ecx
+.LOP_CMPL_DOUBLE_finish:
+    SET_VREG(%ecx,%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMPG_DOUBLE: /* 0x30 */
+/* File: x86/OP_CMPG_DOUBLE.S */
+    /* float/double_cmp[gl] vAA, vBB, vCC */
+    movzbl    3(rPC),%eax             # eax<- CC
+    movzbl    2(rPC),%ecx             # ecx<- BB
+    .if 1
+    fldl     (rFP,%eax,4)
+    fldl     (rFP,%ecx,4)
+    .else
+    flds     (rFP,%eax,4)
+    flds     (rFP,%ecx,4)
+    .endif
+    movzbl   rINST_HI,rINST_FULL
+    xorl     %ecx,%ecx
+    fucompp     # z if equal, p set if NaN, c set if st0 < st1
+    fnstsw   %ax
+    sahf
+    movl      rINST_FULL,%eax
+    FETCH_INST_WORD(2)
+    jp       .LOP_CMPG_DOUBLE_isNaN
+    je       .LOP_CMPG_DOUBLE_finish
+    sbbl     %ecx,%ecx
+    jb       .LOP_CMPG_DOUBLE_finish
+    incl     %ecx
+.LOP_CMPG_DOUBLE_finish:
+    SET_VREG(%ecx,%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMP_LONG: /* 0x31 */
+/* File: x86/OP_CMP_LONG.S */
+    /*
+     * Compare two 64-bit values.  Puts 0, 1, or -1 into the destination
+     * register based on the results of the comparison.
+     */
+    /* cmp-long vAA, vBB, vCC */
+    movzbl    2(rPC),%ecx              # ecx<- BB
+    SPILL(rPC)
+    movzbl    3(rPC),rPC               # rPC<- CC
+    GET_VREG_WORD(%eax,%ecx,1)         # eax<- v[BB+1]
+    GET_VREG_WORD(%ecx,%ecx,0)         # ecx<- v[BB+0]
+    movzbl    rINST_HI,rINST_FULL      # rINST_FULL<- AA
+    cmpl      4(rFP,rPC,4),%eax
+    jl        .LOP_CMP_LONG_smaller
+    jg        .LOP_CMP_LONG_bigger
+    sub       (rFP,rPC,4),%ecx
+    ja        .LOP_CMP_LONG_bigger
+    jb        .LOP_CMP_LONG_smaller
+    UNSPILL(rPC)
+    jmp       .LOP_CMP_LONG_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_EQ: /* 0x32 */
+/* File: x86/OP_IF_EQ.S */
+/* File: x86/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    movzx    rINST_HI,%ecx              # ecx <- A+
+    andb     $0xf,%cl                  # ecx <- A
+    GET_VREG(%eax,%ecx)                 # eax <- vA
+    sarl     $12,rINST_FULL            # rINST_FULL<- B
+    cmpl     (rFP,rINST_FULL,4),%eax    # compare (vA, vB)
+    movswl   2(rPC),rINST_FULL          # Get signed branch offset
+    movl     $2,%eax                   # assume not taken
+    jne   1f
+    testl    rINST_FULL,rINST_FULL
+    js       common_backwardBranch
+    movl     rINST_FULL,%eax
+1:
+    FETCH_INST_INDEXED(%eax)
+    ADVANCE_PC_INDEXED(%eax)
+    GOTO_NEXT
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_NE: /* 0x33 */
+/* File: x86/OP_IF_NE.S */
+/* File: x86/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    movzx    rINST_HI,%ecx              # ecx <- A+
+    andb     $0xf,%cl                  # ecx <- A
+    GET_VREG(%eax,%ecx)                 # eax <- vA
+    sarl     $12,rINST_FULL            # rINST_FULL<- B
+    cmpl     (rFP,rINST_FULL,4),%eax    # compare (vA, vB)
+    movswl   2(rPC),rINST_FULL          # Get signed branch offset
+    movl     $2,%eax                   # assume not taken
+    je   1f
+    testl    rINST_FULL,rINST_FULL
+    js       common_backwardBranch
+    movl     rINST_FULL,%eax
+1:
+    FETCH_INST_INDEXED(%eax)
+    ADVANCE_PC_INDEXED(%eax)
+    GOTO_NEXT
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_LT: /* 0x34 */
+/* File: x86/OP_IF_LT.S */
+/* File: x86/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    movzx    rINST_HI,%ecx              # ecx <- A+
+    andb     $0xf,%cl                  # ecx <- A
+    GET_VREG(%eax,%ecx)                 # eax <- vA
+    sarl     $12,rINST_FULL            # rINST_FULL<- B
+    cmpl     (rFP,rINST_FULL,4),%eax    # compare (vA, vB)
+    movswl   2(rPC),rINST_FULL          # Get signed branch offset
+    movl     $2,%eax                   # assume not taken
+    jge   1f
+    testl    rINST_FULL,rINST_FULL
+    js       common_backwardBranch
+    movl     rINST_FULL,%eax
+1:
+    FETCH_INST_INDEXED(%eax)
+    ADVANCE_PC_INDEXED(%eax)
+    GOTO_NEXT
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_GE: /* 0x35 */
+/* File: x86/OP_IF_GE.S */
+/* File: x86/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    movzx    rINST_HI,%ecx              # ecx <- A+
+    andb     $0xf,%cl                  # ecx <- A
+    GET_VREG(%eax,%ecx)                 # eax <- vA
+    sarl     $12,rINST_FULL            # rINST_FULL<- B
+    cmpl     (rFP,rINST_FULL,4),%eax    # compare (vA, vB)
+    movswl   2(rPC),rINST_FULL          # Get signed branch offset
+    movl     $2,%eax                   # assume not taken
+    jl   1f
+    testl    rINST_FULL,rINST_FULL
+    js       common_backwardBranch
+    movl     rINST_FULL,%eax
+1:
+    FETCH_INST_INDEXED(%eax)
+    ADVANCE_PC_INDEXED(%eax)
+    GOTO_NEXT
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_GT: /* 0x36 */
+/* File: x86/OP_IF_GT.S */
+/* File: x86/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    movzx    rINST_HI,%ecx              # ecx <- A+
+    andb     $0xf,%cl                  # ecx <- A
+    GET_VREG(%eax,%ecx)                 # eax <- vA
+    sarl     $12,rINST_FULL            # rINST_FULL<- B
+    cmpl     (rFP,rINST_FULL,4),%eax    # compare (vA, vB)
+    movswl   2(rPC),rINST_FULL          # Get signed branch offset
+    movl     $2,%eax                   # assume not taken
+    jle   1f
+    testl    rINST_FULL,rINST_FULL
+    js       common_backwardBranch
+    movl     rINST_FULL,%eax
+1:
+    FETCH_INST_INDEXED(%eax)
+    ADVANCE_PC_INDEXED(%eax)
+    GOTO_NEXT
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_LE: /* 0x37 */
+/* File: x86/OP_IF_LE.S */
+/* File: x86/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    movzx    rINST_HI,%ecx              # ecx <- A+
+    andb     $0xf,%cl                  # ecx <- A
+    GET_VREG(%eax,%ecx)                 # eax <- vA
+    sarl     $12,rINST_FULL            # rINST_FULL<- B
+    cmpl     (rFP,rINST_FULL,4),%eax    # compare (vA, vB)
+    movswl   2(rPC),rINST_FULL          # Get signed branch offset
+    movl     $2,%eax                   # assume not taken
+    jg   1f
+    testl    rINST_FULL,rINST_FULL
+    js       common_backwardBranch
+    movl     rINST_FULL,%eax
+1:
+    FETCH_INST_INDEXED(%eax)
+    ADVANCE_PC_INDEXED(%eax)
+    GOTO_NEXT
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_EQZ: /* 0x38 */
+/* File: x86/OP_IF_EQZ.S */
+/* File: x86/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    movzx    rINST_HI,%ecx             # ecx <- AA
+    cmpl     $0,(rFP,%ecx,4)          # compare (vA, 0)
+    movswl   2(rPC),rINST_FULL         # fetch signed displacement
+    movl     $2,%eax                  # assume branch not taken
+    jne   1f
+    testl    rINST_FULL,rINST_FULL
+    js       common_backwardBranch
+    movl     rINST_FULL,%eax
+1:
+    FETCH_INST_INDEXED(%eax)
+    ADVANCE_PC_INDEXED(%eax)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_NEZ: /* 0x39 */
+/* File: x86/OP_IF_NEZ.S */
+/* File: x86/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    movzx    rINST_HI,%ecx             # ecx <- AA
+    cmpl     $0,(rFP,%ecx,4)          # compare (vA, 0)
+    movswl   2(rPC),rINST_FULL         # fetch signed displacement
+    movl     $2,%eax                  # assume branch not taken
+    je   1f
+    testl    rINST_FULL,rINST_FULL
+    js       common_backwardBranch
+    movl     rINST_FULL,%eax
+1:
+    FETCH_INST_INDEXED(%eax)
+    ADVANCE_PC_INDEXED(%eax)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_LTZ: /* 0x3a */
+/* File: x86/OP_IF_LTZ.S */
+/* File: x86/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    movzx    rINST_HI,%ecx             # ecx <- AA
+    cmpl     $0,(rFP,%ecx,4)          # compare (vA, 0)
+    movswl   2(rPC),rINST_FULL         # fetch signed displacement
+    movl     $2,%eax                  # assume branch not taken
+    jge   1f
+    testl    rINST_FULL,rINST_FULL
+    js       common_backwardBranch
+    movl     rINST_FULL,%eax
+1:
+    FETCH_INST_INDEXED(%eax)
+    ADVANCE_PC_INDEXED(%eax)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_GEZ: /* 0x3b */
+/* File: x86/OP_IF_GEZ.S */
+/* File: x86/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    movzx    rINST_HI,%ecx             # ecx <- AA
+    cmpl     $0,(rFP,%ecx,4)          # compare (vA, 0)
+    movswl   2(rPC),rINST_FULL         # fetch signed displacement
+    movl     $2,%eax                  # assume branch not taken
+    jl   1f
+    testl    rINST_FULL,rINST_FULL
+    js       common_backwardBranch
+    movl     rINST_FULL,%eax
+1:
+    FETCH_INST_INDEXED(%eax)
+    ADVANCE_PC_INDEXED(%eax)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_GTZ: /* 0x3c */
+/* File: x86/OP_IF_GTZ.S */
+/* File: x86/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    movzx    rINST_HI,%ecx             # ecx <- AA
+    cmpl     $0,(rFP,%ecx,4)          # compare (vA, 0)
+    movswl   2(rPC),rINST_FULL         # fetch signed displacement
+    movl     $2,%eax                  # assume branch not taken
+    jle   1f
+    testl    rINST_FULL,rINST_FULL
+    js       common_backwardBranch
+    movl     rINST_FULL,%eax
+1:
+    FETCH_INST_INDEXED(%eax)
+    ADVANCE_PC_INDEXED(%eax)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_LEZ: /* 0x3d */
+/* File: x86/OP_IF_LEZ.S */
+/* File: x86/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    movzx    rINST_HI,%ecx             # ecx <- AA
+    cmpl     $0,(rFP,%ecx,4)          # compare (vA, 0)
+    movswl   2(rPC),rINST_FULL         # fetch signed displacement
+    movl     $2,%eax                  # assume branch not taken
+    jg   1f
+    testl    rINST_FULL,rINST_FULL
+    js       common_backwardBranch
+    movl     rINST_FULL,%eax
+1:
+    FETCH_INST_INDEXED(%eax)
+    ADVANCE_PC_INDEXED(%eax)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_3E: /* 0x3e */
+/* File: x86/OP_UNUSED_3E.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_3F: /* 0x3f */
+/* File: x86/OP_UNUSED_3F.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_40: /* 0x40 */
+/* File: x86/OP_UNUSED_40.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_41: /* 0x41 */
+/* File: x86/OP_UNUSED_41.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_42: /* 0x42 */
+/* File: x86/OP_UNUSED_42.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_43: /* 0x43 */
+/* File: x86/OP_UNUSED_43.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET: /* 0x44 */
+/* File: x86/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    GET_VREG(%eax,%eax)                 # eax<- vBB (array object)
+    GET_VREG(%ecx,%ecx)                 # ecs<- vCC (requested index)
+    testl     %eax,%eax                 # null array object?
+    je        common_errNullObject      # bail if so
+    cmpl      offArrayObject_length(%eax),%ecx
+    jae       common_errArrayIndex      # index >= length, bail
+    movl     offArrayObject_contents(%eax,%ecx,4),%eax
+    movl      rINST_FULL,%ecx
+    FETCH_INST_WORD(2)
+    SET_VREG(%eax,%ecx)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_WIDE: /* 0x45 */
+/* File: x86/OP_AGET_WIDE.S */
+    /*
+     * Array get, 64 bits.  vAA <- vBB[vCC].
+     *
+     */
+    /* op vAA, vBB, vCC */
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    GET_VREG(%eax,%eax)                 # eax<- vBB (array object)
+    GET_VREG(%ecx,%ecx)                 # ecs<- vCC (requested index)
+    testl     %eax,%eax                 # null array object?
+    je        common_errNullObject      # bail if so
+    cmpl      offArrayObject_length(%eax),%ecx
+    jb        .LOP_AGET_WIDE_finish        # index < length, OK
+    jmp       common_errArrayIndex      # index >= length, bail
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_OBJECT: /* 0x46 */
+/* File: x86/OP_AGET_OBJECT.S */
+/* File: x86/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    GET_VREG(%eax,%eax)                 # eax<- vBB (array object)
+    GET_VREG(%ecx,%ecx)                 # ecs<- vCC (requested index)
+    testl     %eax,%eax                 # null array object?
+    je        common_errNullObject      # bail if so
+    cmpl      offArrayObject_length(%eax),%ecx
+    jae       common_errArrayIndex      # index >= length, bail
+    movl     offArrayObject_contents(%eax,%ecx,4),%eax
+    movl      rINST_FULL,%ecx
+    FETCH_INST_WORD(2)
+    SET_VREG(%eax,%ecx)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_BOOLEAN: /* 0x47 */
+/* File: x86/OP_AGET_BOOLEAN.S */
+/* File: x86/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    GET_VREG(%eax,%eax)                 # eax<- vBB (array object)
+    GET_VREG(%ecx,%ecx)                 # ecs<- vCC (requested index)
+    testl     %eax,%eax                 # null array object?
+    je        common_errNullObject      # bail if so
+    cmpl      offArrayObject_length(%eax),%ecx
+    jae       common_errArrayIndex      # index >= length, bail
+    movzbl     offArrayObject_contents(%eax,%ecx,1),%eax
+    movl      rINST_FULL,%ecx
+    FETCH_INST_WORD(2)
+    SET_VREG(%eax,%ecx)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_BYTE: /* 0x48 */
+/* File: x86/OP_AGET_BYTE.S */
+/* File: x86/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    GET_VREG(%eax,%eax)                 # eax<- vBB (array object)
+    GET_VREG(%ecx,%ecx)                 # ecs<- vCC (requested index)
+    testl     %eax,%eax                 # null array object?
+    je        common_errNullObject      # bail if so
+    cmpl      offArrayObject_length(%eax),%ecx
+    jae       common_errArrayIndex      # index >= length, bail
+    movsbl     offArrayObject_contents(%eax,%ecx,1),%eax
+    movl      rINST_FULL,%ecx
+    FETCH_INST_WORD(2)
+    SET_VREG(%eax,%ecx)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_CHAR: /* 0x49 */
+/* File: x86/OP_AGET_CHAR.S */
+/* File: x86/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    GET_VREG(%eax,%eax)                 # eax<- vBB (array object)
+    GET_VREG(%ecx,%ecx)                 # ecs<- vCC (requested index)
+    testl     %eax,%eax                 # null array object?
+    je        common_errNullObject      # bail if so
+    cmpl      offArrayObject_length(%eax),%ecx
+    jae       common_errArrayIndex      # index >= length, bail
+    movzwl     offArrayObject_contents(%eax,%ecx,2),%eax
+    movl      rINST_FULL,%ecx
+    FETCH_INST_WORD(2)
+    SET_VREG(%eax,%ecx)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_SHORT: /* 0x4a */
+/* File: x86/OP_AGET_SHORT.S */
+/* File: x86/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    GET_VREG(%eax,%eax)                 # eax<- vBB (array object)
+    GET_VREG(%ecx,%ecx)                 # ecs<- vCC (requested index)
+    testl     %eax,%eax                 # null array object?
+    je        common_errNullObject      # bail if so
+    cmpl      offArrayObject_length(%eax),%ecx
+    jae       common_errArrayIndex      # index >= length, bail
+    movswl     offArrayObject_contents(%eax,%ecx,2),%eax
+    movl      rINST_FULL,%ecx
+    FETCH_INST_WORD(2)
+    SET_VREG(%eax,%ecx)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT: /* 0x4b */
+/* File: x86/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA
+     *
+     * for: aput, aput-object, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    GET_VREG(%eax,%eax)                 # eax<- vBB (array object)
+    GET_VREG(%ecx,%ecx)                 # ecs<- vCC (requested index)
+    testl     %eax,%eax                 # null array object?
+    je        common_errNullObject      # bail if so
+    cmpl      offArrayObject_length(%eax),%ecx
+    jae       common_errArrayIndex      # index >= length, bail
+    leal      offArrayObject_contents(%eax,%ecx,4),%eax
+    GET_VREG(%ecx,rINST_FULL)
+    FETCH_INST_WORD(2)
+    movl     %ecx,(%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_WIDE: /* 0x4c */
+/* File: x86/OP_APUT_WIDE.S */
+    /*
+     * Array put, 64 bits.  vBB[vCC]<-vAA.
+     *
+     */
+    /* op vAA, vBB, vCC */
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    GET_VREG(%eax,%eax)                 # eax<- vBB (array object)
+    GET_VREG(%ecx,%ecx)                 # ecs<- vCC (requested index)
+    testl     %eax,%eax                 # null array object?
+    je        common_errNullObject      # bail if so
+    cmpl      offArrayObject_length(%eax),%ecx
+    jb        .LOP_APUT_WIDE_finish        # index < length, OK
+    jmp       common_errArrayIndex      # index >= length, bail
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_OBJECT: /* 0x4d */
+/* File: x86/OP_APUT_OBJECT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA
+     *
+     * for: aput, aput-object, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    GET_VREG(%eax,%eax)                 # eax<- vBB (array object)
+    GET_VREG(%ecx,%ecx)                 # ecs<- vCC (requested index)
+    GET_VREG(rINST_FULL,rINST_FULL)     # rINST_FULL<- vAA
+    testl     %eax,%eax                 # null array object?
+    je        common_errNullObject      # bail if so
+    cmpl      offArrayObject_length(%eax),%ecx
+    jb        .LOP_APUT_OBJECT_continue
+    jmp       common_errArrayIndex      # index >= length, bail
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_BOOLEAN: /* 0x4e */
+/* File: x86/OP_APUT_BOOLEAN.S */
+/* File: x86/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA
+     *
+     * for: aput, aput-object, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    GET_VREG(%eax,%eax)                 # eax<- vBB (array object)
+    GET_VREG(%ecx,%ecx)                 # ecs<- vCC (requested index)
+    testl     %eax,%eax                 # null array object?
+    je        common_errNullObject      # bail if so
+    cmpl      offArrayObject_length(%eax),%ecx
+    jae       common_errArrayIndex      # index >= length, bail
+    leal      offArrayObject_contents(%eax,%ecx,1),%eax
+    GET_VREG(%ecx,rINST_FULL)
+    FETCH_INST_WORD(2)
+    movb     %cl,(%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_BYTE: /* 0x4f */
+/* File: x86/OP_APUT_BYTE.S */
+/* File: x86/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA
+     *
+     * for: aput, aput-object, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    GET_VREG(%eax,%eax)                 # eax<- vBB (array object)
+    GET_VREG(%ecx,%ecx)                 # ecs<- vCC (requested index)
+    testl     %eax,%eax                 # null array object?
+    je        common_errNullObject      # bail if so
+    cmpl      offArrayObject_length(%eax),%ecx
+    jae       common_errArrayIndex      # index >= length, bail
+    leal      offArrayObject_contents(%eax,%ecx,1),%eax
+    GET_VREG(%ecx,rINST_FULL)
+    FETCH_INST_WORD(2)
+    movb     %cl,(%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_CHAR: /* 0x50 */
+/* File: x86/OP_APUT_CHAR.S */
+/* File: x86/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA
+     *
+     * for: aput, aput-object, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    GET_VREG(%eax,%eax)                 # eax<- vBB (array object)
+    GET_VREG(%ecx,%ecx)                 # ecs<- vCC (requested index)
+    testl     %eax,%eax                 # null array object?
+    je        common_errNullObject      # bail if so
+    cmpl      offArrayObject_length(%eax),%ecx
+    jae       common_errArrayIndex      # index >= length, bail
+    leal      offArrayObject_contents(%eax,%ecx,2),%eax
+    GET_VREG(%ecx,rINST_FULL)
+    FETCH_INST_WORD(2)
+    movw     %cx,(%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_SHORT: /* 0x51 */
+/* File: x86/OP_APUT_SHORT.S */
+/* File: x86/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA
+     *
+     * for: aput, aput-object, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    GET_VREG(%eax,%eax)                 # eax<- vBB (array object)
+    GET_VREG(%ecx,%ecx)                 # ecs<- vCC (requested index)
+    testl     %eax,%eax                 # null array object?
+    je        common_errNullObject      # bail if so
+    cmpl      offArrayObject_length(%eax),%ecx
+    jae       common_errArrayIndex      # index >= length, bail
+    leal      offArrayObject_contents(%eax,%ecx,2),%eax
+    GET_VREG(%ecx,rINST_FULL)
+    FETCH_INST_WORD(2)
+    movw     %cx,(%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET: /* 0x52 */
+/* File: x86/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    GET_GLUE(%ecx)
+    SPILL(rIBASE)                                 # need another reg
+    movzwl  2(rPC),rIBASE                         # rIBASE<- 0000CCCC
+    movl    offGlue_methodClassDex(%ecx),%eax     # eax<- DvmDex
+    movzbl  rINST_HI,%ecx                         # ecx<- BA
+    sarl    $4,%ecx                              # ecx<- B
+    movl    offDvmDex_pResFields(%eax),%eax       # eax<- pDvmDex->pResFields
+    movzbl  rINST_HI,rINST_FULL                   # rINST_FULL<- BA
+    andb    $0xf,rINST_LO                        # rINST_FULL<- A
+    GET_VREG(%ecx,%ecx)                           # ecx<- fp[B], the object ptr
+    movl    (%eax,rIBASE,4),%eax                  # resolved entry
+    testl   %eax,%eax                             # is resolved entry null?
+    jne     .LOP_IGET_finish                    # no, already resolved
+    movl    rIBASE,OUT_ARG1(%esp)                 # needed by dvmResolveInstField
+    GET_GLUE(rIBASE)
+    jmp     .LOP_IGET_resolve
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_WIDE: /* 0x53 */
+/* File: x86/OP_IGET_WIDE.S */
+    /*
+     * 64-bit instance field get.
+     *
+     */
+    /* op vA, vB, field@CCCC */
+    GET_GLUE(%ecx)
+    SPILL(rIBASE)                                 # need another reg
+    movzwl  2(rPC),rIBASE                         # rIBASE<- 0000CCCC
+    movl    offGlue_methodClassDex(%ecx),%eax     # eax<- DvmDex
+    movzbl  rINST_HI,%ecx                         # ecx<- BA
+    sarl    $4,%ecx                              # ecx<- B
+    movl    offDvmDex_pResFields(%eax),%eax       # eax<- pDvmDex->pResFields
+    movzbl  rINST_HI,rINST_FULL                   # rINST_FULL<- BA
+    andb    $0xf,rINST_LO                        # rINST_FULL<- A
+    GET_VREG(%ecx,%ecx)                           # ecx<- fp[B], the object ptr
+    movl    (%eax,rIBASE,4),%eax                  # resolved entry
+    testl   %eax,%eax                             # is resolved entry null?
+    jne     .LOP_IGET_WIDE_finish                    # no, already resolved
+    movl    rIBASE,OUT_ARG1(%esp)                 # needed by dvmResolveInstField
+    GET_GLUE(rIBASE)
+    jmp     .LOP_IGET_WIDE_resolve
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_OBJECT: /* 0x54 */
+/* File: x86/OP_IGET_OBJECT.S */
+/* File: x86/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    GET_GLUE(%ecx)
+    SPILL(rIBASE)                                 # need another reg
+    movzwl  2(rPC),rIBASE                         # rIBASE<- 0000CCCC
+    movl    offGlue_methodClassDex(%ecx),%eax     # eax<- DvmDex
+    movzbl  rINST_HI,%ecx                         # ecx<- BA
+    sarl    $4,%ecx                              # ecx<- B
+    movl    offDvmDex_pResFields(%eax),%eax       # eax<- pDvmDex->pResFields
+    movzbl  rINST_HI,rINST_FULL                   # rINST_FULL<- BA
+    andb    $0xf,rINST_LO                        # rINST_FULL<- A
+    GET_VREG(%ecx,%ecx)                           # ecx<- fp[B], the object ptr
+    movl    (%eax,rIBASE,4),%eax                  # resolved entry
+    testl   %eax,%eax                             # is resolved entry null?
+    jne     .LOP_IGET_OBJECT_finish                    # no, already resolved
+    movl    rIBASE,OUT_ARG1(%esp)                 # needed by dvmResolveInstField
+    GET_GLUE(rIBASE)
+    jmp     .LOP_IGET_OBJECT_resolve
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_BOOLEAN: /* 0x55 */
+/* File: x86/OP_IGET_BOOLEAN.S */
+/* File: x86/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    GET_GLUE(%ecx)
+    SPILL(rIBASE)                                 # need another reg
+    movzwl  2(rPC),rIBASE                         # rIBASE<- 0000CCCC
+    movl    offGlue_methodClassDex(%ecx),%eax     # eax<- DvmDex
+    movzbl  rINST_HI,%ecx                         # ecx<- BA
+    sarl    $4,%ecx                              # ecx<- B
+    movl    offDvmDex_pResFields(%eax),%eax       # eax<- pDvmDex->pResFields
+    movzbl  rINST_HI,rINST_FULL                   # rINST_FULL<- BA
+    andb    $0xf,rINST_LO                        # rINST_FULL<- A
+    GET_VREG(%ecx,%ecx)                           # ecx<- fp[B], the object ptr
+    movl    (%eax,rIBASE,4),%eax                  # resolved entry
+    testl   %eax,%eax                             # is resolved entry null?
+    jne     .LOP_IGET_BOOLEAN_finish                    # no, already resolved
+    movl    rIBASE,OUT_ARG1(%esp)                 # needed by dvmResolveInstField
+    GET_GLUE(rIBASE)
+    jmp     .LOP_IGET_BOOLEAN_resolve
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_BYTE: /* 0x56 */
+/* File: x86/OP_IGET_BYTE.S */
+/* File: x86/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    GET_GLUE(%ecx)
+    SPILL(rIBASE)                                 # need another reg
+    movzwl  2(rPC),rIBASE                         # rIBASE<- 0000CCCC
+    movl    offGlue_methodClassDex(%ecx),%eax     # eax<- DvmDex
+    movzbl  rINST_HI,%ecx                         # ecx<- BA
+    sarl    $4,%ecx                              # ecx<- B
+    movl    offDvmDex_pResFields(%eax),%eax       # eax<- pDvmDex->pResFields
+    movzbl  rINST_HI,rINST_FULL                   # rINST_FULL<- BA
+    andb    $0xf,rINST_LO                        # rINST_FULL<- A
+    GET_VREG(%ecx,%ecx)                           # ecx<- fp[B], the object ptr
+    movl    (%eax,rIBASE,4),%eax                  # resolved entry
+    testl   %eax,%eax                             # is resolved entry null?
+    jne     .LOP_IGET_BYTE_finish                    # no, already resolved
+    movl    rIBASE,OUT_ARG1(%esp)                 # needed by dvmResolveInstField
+    GET_GLUE(rIBASE)
+    jmp     .LOP_IGET_BYTE_resolve
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_CHAR: /* 0x57 */
+/* File: x86/OP_IGET_CHAR.S */
+/* File: x86/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    GET_GLUE(%ecx)
+    SPILL(rIBASE)                                 # need another reg
+    movzwl  2(rPC),rIBASE                         # rIBASE<- 0000CCCC
+    movl    offGlue_methodClassDex(%ecx),%eax     # eax<- DvmDex
+    movzbl  rINST_HI,%ecx                         # ecx<- BA
+    sarl    $4,%ecx                              # ecx<- B
+    movl    offDvmDex_pResFields(%eax),%eax       # eax<- pDvmDex->pResFields
+    movzbl  rINST_HI,rINST_FULL                   # rINST_FULL<- BA
+    andb    $0xf,rINST_LO                        # rINST_FULL<- A
+    GET_VREG(%ecx,%ecx)                           # ecx<- fp[B], the object ptr
+    movl    (%eax,rIBASE,4),%eax                  # resolved entry
+    testl   %eax,%eax                             # is resolved entry null?
+    jne     .LOP_IGET_CHAR_finish                    # no, already resolved
+    movl    rIBASE,OUT_ARG1(%esp)                 # needed by dvmResolveInstField
+    GET_GLUE(rIBASE)
+    jmp     .LOP_IGET_CHAR_resolve
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_SHORT: /* 0x58 */
+/* File: x86/OP_IGET_SHORT.S */
+/* File: x86/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    GET_GLUE(%ecx)
+    SPILL(rIBASE)                                 # need another reg
+    movzwl  2(rPC),rIBASE                         # rIBASE<- 0000CCCC
+    movl    offGlue_methodClassDex(%ecx),%eax     # eax<- DvmDex
+    movzbl  rINST_HI,%ecx                         # ecx<- BA
+    sarl    $4,%ecx                              # ecx<- B
+    movl    offDvmDex_pResFields(%eax),%eax       # eax<- pDvmDex->pResFields
+    movzbl  rINST_HI,rINST_FULL                   # rINST_FULL<- BA
+    andb    $0xf,rINST_LO                        # rINST_FULL<- A
+    GET_VREG(%ecx,%ecx)                           # ecx<- fp[B], the object ptr
+    movl    (%eax,rIBASE,4),%eax                  # resolved entry
+    testl   %eax,%eax                             # is resolved entry null?
+    jne     .LOP_IGET_SHORT_finish                    # no, already resolved
+    movl    rIBASE,OUT_ARG1(%esp)                 # needed by dvmResolveInstField
+    GET_GLUE(rIBASE)
+    jmp     .LOP_IGET_SHORT_resolve
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT: /* 0x59 */
+/* File: x86/OP_IPUT.S */
+
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    GET_GLUE(%ecx)
+    SPILL(rIBASE)                                 # need another reg
+    movzwl  2(rPC),rIBASE                         # rIBASE<- 0000CCCC
+    movl    offGlue_methodClassDex(%ecx),%eax     # eax<- DvmDex
+    movzbl  rINST_HI,%ecx                         # ecx<- BA
+    sarl    $4,%ecx                              # ecx<- B
+    movl    offDvmDex_pResFields(%eax),%eax       # eax<- pDvmDex->pResFields
+    movzbl  rINST_HI,rINST_FULL                   # rINST_FULL<- BA
+    andb    $0xf,rINST_LO                        # rINST_FULL<- A
+    GET_VREG(%ecx,%ecx)                           # ecx<- fp[B], the object ptr
+    movl    (%eax,rIBASE,4),%eax                  # resolved entry
+    testl   %eax,%eax                             # is resolved entry null?
+    jne     .LOP_IPUT_finish                    # no, already resolved
+    movl    rIBASE,OUT_ARG1(%esp)
+    GET_GLUE(rIBASE)
+    jmp     .LOP_IPUT_resolve
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_WIDE: /* 0x5a */
+/* File: x86/OP_IPUT_WIDE.S */
+    /*
+     * 64-bit instance field put.
+     *
+     */
+    /* op vA, vB, field@CCCC */
+    GET_GLUE(%ecx)
+    SPILL(rIBASE)                                 # need another reg
+    movzwl  2(rPC),rIBASE                         # rIBASE<- 0000CCCC
+    movl    offGlue_methodClassDex(%ecx),%eax     # eax<- DvmDex
+    movzbl  rINST_HI,%ecx                         # ecx<- BA
+    sarl    $4,%ecx                              # ecx<- B
+    movl    offDvmDex_pResFields(%eax),%eax       # eax<- pDvmDex->pResFields
+    movzbl  rINST_HI,rINST_FULL                   # rINST_FULL<- BA
+    andb    $0xf,rINST_LO                        # rINST_FULL<- A
+    GET_VREG(%ecx,%ecx)                           # ecx<- fp[B], the object ptr
+    movl    (%eax,rIBASE,4),%eax                  # resolved entry
+    testl   %eax,%eax                             # is resolved entry null?
+    jne     .LOP_IPUT_WIDE_finish                    # no, already resolved
+    movl    rIBASE,OUT_ARG1(%esp)
+    GET_GLUE(rIBASE)
+    jmp     .LOP_IPUT_WIDE_resolve
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_OBJECT: /* 0x5b */
+/* File: x86/OP_IPUT_OBJECT.S */
+/* File: x86/OP_IPUT.S */
+
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    GET_GLUE(%ecx)
+    SPILL(rIBASE)                                 # need another reg
+    movzwl  2(rPC),rIBASE                         # rIBASE<- 0000CCCC
+    movl    offGlue_methodClassDex(%ecx),%eax     # eax<- DvmDex
+    movzbl  rINST_HI,%ecx                         # ecx<- BA
+    sarl    $4,%ecx                              # ecx<- B
+    movl    offDvmDex_pResFields(%eax),%eax       # eax<- pDvmDex->pResFields
+    movzbl  rINST_HI,rINST_FULL                   # rINST_FULL<- BA
+    andb    $0xf,rINST_LO                        # rINST_FULL<- A
+    GET_VREG(%ecx,%ecx)                           # ecx<- fp[B], the object ptr
+    movl    (%eax,rIBASE,4),%eax                  # resolved entry
+    testl   %eax,%eax                             # is resolved entry null?
+    jne     .LOP_IPUT_OBJECT_finish                    # no, already resolved
+    movl    rIBASE,OUT_ARG1(%esp)
+    GET_GLUE(rIBASE)
+    jmp     .LOP_IPUT_OBJECT_resolve
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_BOOLEAN: /* 0x5c */
+/* File: x86/OP_IPUT_BOOLEAN.S */
+/* File: x86/OP_IPUT.S */
+
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    GET_GLUE(%ecx)
+    SPILL(rIBASE)                                 # need another reg
+    movzwl  2(rPC),rIBASE                         # rIBASE<- 0000CCCC
+    movl    offGlue_methodClassDex(%ecx),%eax     # eax<- DvmDex
+    movzbl  rINST_HI,%ecx                         # ecx<- BA
+    sarl    $4,%ecx                              # ecx<- B
+    movl    offDvmDex_pResFields(%eax),%eax       # eax<- pDvmDex->pResFields
+    movzbl  rINST_HI,rINST_FULL                   # rINST_FULL<- BA
+    andb    $0xf,rINST_LO                        # rINST_FULL<- A
+    GET_VREG(%ecx,%ecx)                           # ecx<- fp[B], the object ptr
+    movl    (%eax,rIBASE,4),%eax                  # resolved entry
+    testl   %eax,%eax                             # is resolved entry null?
+    jne     .LOP_IPUT_BOOLEAN_finish                    # no, already resolved
+    movl    rIBASE,OUT_ARG1(%esp)
+    GET_GLUE(rIBASE)
+    jmp     .LOP_IPUT_BOOLEAN_resolve
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_BYTE: /* 0x5d */
+/* File: x86/OP_IPUT_BYTE.S */
+/* File: x86/OP_IPUT.S */
+
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    GET_GLUE(%ecx)
+    SPILL(rIBASE)                                 # need another reg
+    movzwl  2(rPC),rIBASE                         # rIBASE<- 0000CCCC
+    movl    offGlue_methodClassDex(%ecx),%eax     # eax<- DvmDex
+    movzbl  rINST_HI,%ecx                         # ecx<- BA
+    sarl    $4,%ecx                              # ecx<- B
+    movl    offDvmDex_pResFields(%eax),%eax       # eax<- pDvmDex->pResFields
+    movzbl  rINST_HI,rINST_FULL                   # rINST_FULL<- BA
+    andb    $0xf,rINST_LO                        # rINST_FULL<- A
+    GET_VREG(%ecx,%ecx)                           # ecx<- fp[B], the object ptr
+    movl    (%eax,rIBASE,4),%eax                  # resolved entry
+    testl   %eax,%eax                             # is resolved entry null?
+    jne     .LOP_IPUT_BYTE_finish                    # no, already resolved
+    movl    rIBASE,OUT_ARG1(%esp)
+    GET_GLUE(rIBASE)
+    jmp     .LOP_IPUT_BYTE_resolve
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_CHAR: /* 0x5e */
+/* File: x86/OP_IPUT_CHAR.S */
+/* File: x86/OP_IPUT.S */
+
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    GET_GLUE(%ecx)
+    SPILL(rIBASE)                                 # need another reg
+    movzwl  2(rPC),rIBASE                         # rIBASE<- 0000CCCC
+    movl    offGlue_methodClassDex(%ecx),%eax     # eax<- DvmDex
+    movzbl  rINST_HI,%ecx                         # ecx<- BA
+    sarl    $4,%ecx                              # ecx<- B
+    movl    offDvmDex_pResFields(%eax),%eax       # eax<- pDvmDex->pResFields
+    movzbl  rINST_HI,rINST_FULL                   # rINST_FULL<- BA
+    andb    $0xf,rINST_LO                        # rINST_FULL<- A
+    GET_VREG(%ecx,%ecx)                           # ecx<- fp[B], the object ptr
+    movl    (%eax,rIBASE,4),%eax                  # resolved entry
+    testl   %eax,%eax                             # is resolved entry null?
+    jne     .LOP_IPUT_CHAR_finish                    # no, already resolved
+    movl    rIBASE,OUT_ARG1(%esp)
+    GET_GLUE(rIBASE)
+    jmp     .LOP_IPUT_CHAR_resolve
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_SHORT: /* 0x5f */
+/* File: x86/OP_IPUT_SHORT.S */
+/* File: x86/OP_IPUT.S */
+
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    GET_GLUE(%ecx)
+    SPILL(rIBASE)                                 # need another reg
+    movzwl  2(rPC),rIBASE                         # rIBASE<- 0000CCCC
+    movl    offGlue_methodClassDex(%ecx),%eax     # eax<- DvmDex
+    movzbl  rINST_HI,%ecx                         # ecx<- BA
+    sarl    $4,%ecx                              # ecx<- B
+    movl    offDvmDex_pResFields(%eax),%eax       # eax<- pDvmDex->pResFields
+    movzbl  rINST_HI,rINST_FULL                   # rINST_FULL<- BA
+    andb    $0xf,rINST_LO                        # rINST_FULL<- A
+    GET_VREG(%ecx,%ecx)                           # ecx<- fp[B], the object ptr
+    movl    (%eax,rIBASE,4),%eax                  # resolved entry
+    testl   %eax,%eax                             # is resolved entry null?
+    jne     .LOP_IPUT_SHORT_finish                    # no, already resolved
+    movl    rIBASE,OUT_ARG1(%esp)
+    GET_GLUE(rIBASE)
+    jmp     .LOP_IPUT_SHORT_resolve
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET: /* 0x60 */
+/* File: x86/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax                        # eax<- field ref BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx  # ecx<- DvmDex
+    movl      offDvmDex_pResFields(%ecx),%ecx    # ecx<- dvmDex->pResFields
+    movl      (%ecx,%eax,4),%eax                 # eax<- resolved StaticField ptr
+    testl     %eax,%eax                          # resolved entry null?
+    je        .LOP_SGET_resolve                # if not, make it so
+.LOP_SGET_finish:     # field ptr in eax
+    movl      offStaticField_value(%eax),%eax
+    movzbl    rINST_HI,%ecx                      # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_WIDE: /* 0x61 */
+/* File: x86/OP_SGET_WIDE.S */
+    /*
+     * 64-bit SGET handler.
+     *
+     */
+    /* sget-wide vAA, field@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax                        # eax<- field ref BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx  # ecx<- DvmDex
+    movl      offDvmDex_pResFields(%ecx),%ecx    # ecx<- dvmDex->pResFields
+    movl      (%ecx,%eax,4),%eax                 # eax<- resolved StaticField ptr
+    testl     %eax,%eax                          # resolved entry null?
+    je        .LOP_SGET_WIDE_resolve                # if not, make it so
+.LOP_SGET_WIDE_finish:     # field ptr in eax
+    movl      offStaticField_value(%eax),%ecx    # ecx<- lsw
+    movl      4+offStaticField_value(%eax),%eax  # eax<- msw
+    movzbl    rINST_HI,rINST_FULL                # rINST_FULL<- AA
+    SET_VREG_WORD(%ecx,rINST_FULL,0)
+    SET_VREG_WORD(%eax,rINST_FULL,1)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_OBJECT: /* 0x62 */
+/* File: x86/OP_SGET_OBJECT.S */
+/* File: x86/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax                        # eax<- field ref BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx  # ecx<- DvmDex
+    movl      offDvmDex_pResFields(%ecx),%ecx    # ecx<- dvmDex->pResFields
+    movl      (%ecx,%eax,4),%eax                 # eax<- resolved StaticField ptr
+    testl     %eax,%eax                          # resolved entry null?
+    je        .LOP_SGET_OBJECT_resolve                # if not, make it so
+.LOP_SGET_OBJECT_finish:     # field ptr in eax
+    movl      offStaticField_value(%eax),%eax
+    movzbl    rINST_HI,%ecx                      # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_BOOLEAN: /* 0x63 */
+/* File: x86/OP_SGET_BOOLEAN.S */
+/* File: x86/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax                        # eax<- field ref BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx  # ecx<- DvmDex
+    movl      offDvmDex_pResFields(%ecx),%ecx    # ecx<- dvmDex->pResFields
+    movl      (%ecx,%eax,4),%eax                 # eax<- resolved StaticField ptr
+    testl     %eax,%eax                          # resolved entry null?
+    je        .LOP_SGET_BOOLEAN_resolve                # if not, make it so
+.LOP_SGET_BOOLEAN_finish:     # field ptr in eax
+    movl      offStaticField_value(%eax),%eax
+    movzbl    rINST_HI,%ecx                      # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_BYTE: /* 0x64 */
+/* File: x86/OP_SGET_BYTE.S */
+/* File: x86/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax                        # eax<- field ref BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx  # ecx<- DvmDex
+    movl      offDvmDex_pResFields(%ecx),%ecx    # ecx<- dvmDex->pResFields
+    movl      (%ecx,%eax,4),%eax                 # eax<- resolved StaticField ptr
+    testl     %eax,%eax                          # resolved entry null?
+    je        .LOP_SGET_BYTE_resolve                # if not, make it so
+.LOP_SGET_BYTE_finish:     # field ptr in eax
+    movl      offStaticField_value(%eax),%eax
+    movzbl    rINST_HI,%ecx                      # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_CHAR: /* 0x65 */
+/* File: x86/OP_SGET_CHAR.S */
+/* File: x86/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax                        # eax<- field ref BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx  # ecx<- DvmDex
+    movl      offDvmDex_pResFields(%ecx),%ecx    # ecx<- dvmDex->pResFields
+    movl      (%ecx,%eax,4),%eax                 # eax<- resolved StaticField ptr
+    testl     %eax,%eax                          # resolved entry null?
+    je        .LOP_SGET_CHAR_resolve                # if not, make it so
+.LOP_SGET_CHAR_finish:     # field ptr in eax
+    movl      offStaticField_value(%eax),%eax
+    movzbl    rINST_HI,%ecx                      # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_SHORT: /* 0x66 */
+/* File: x86/OP_SGET_SHORT.S */
+/* File: x86/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax                        # eax<- field ref BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx  # ecx<- DvmDex
+    movl      offDvmDex_pResFields(%ecx),%ecx    # ecx<- dvmDex->pResFields
+    movl      (%ecx,%eax,4),%eax                 # eax<- resolved StaticField ptr
+    testl     %eax,%eax                          # resolved entry null?
+    je        .LOP_SGET_SHORT_resolve                # if not, make it so
+.LOP_SGET_SHORT_finish:     # field ptr in eax
+    movl      offStaticField_value(%eax),%eax
+    movzbl    rINST_HI,%ecx                      # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT: /* 0x67 */
+/* File: x86/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax                        # eax<- field ref BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx  # ecx<- DvmDex
+    movl      offDvmDex_pResFields(%ecx),%ecx    # ecx<- dvmDex->pResFields
+    movl      (%ecx,%eax,4),%eax                 # eax<- resolved StaticField ptr
+    testl     %eax,%eax                          # resolved entry null?
+    je        .LOP_SPUT_resolve                # if not, make it so
+.LOP_SPUT_finish:     # field ptr in eax
+    movzbl    rINST_HI,%ecx                      # ecx<- AA
+    GET_VREG(%ecx,%ecx)
+    FETCH_INST_WORD(2)
+    movl      %ecx,offStaticField_value(%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_WIDE: /* 0x68 */
+/* File: x86/OP_SPUT_WIDE.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax                        # eax<- field ref BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx  # ecx<- DvmDex
+    movl      offDvmDex_pResFields(%ecx),%ecx    # ecx<- dvmDex->pResFields
+    movl      (%ecx,%eax,4),%eax                 # eax<- resolved StaticField ptr
+    testl     %eax,%eax                          # resolved entry null?
+    je        .LOP_SPUT_WIDE_resolve                # if not, make it so
+.LOP_SPUT_WIDE_finish:     # field ptr in eax
+    movzbl    rINST_HI,%ecx                      # ecx<- AA
+    GET_VREG_WORD(rINST_FULL,%ecx,0)             # rINST_FULL<- lsw
+    GET_VREG_WORD(%ecx,%ecx,1)                   # ecx<- msw
+    movl      rINST_FULL,offStaticField_value(%eax)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    movl      %ecx,4+offStaticField_value(%eax)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_OBJECT: /* 0x69 */
+/* File: x86/OP_SPUT_OBJECT.S */
+/* File: x86/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax                        # eax<- field ref BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx  # ecx<- DvmDex
+    movl      offDvmDex_pResFields(%ecx),%ecx    # ecx<- dvmDex->pResFields
+    movl      (%ecx,%eax,4),%eax                 # eax<- resolved StaticField ptr
+    testl     %eax,%eax                          # resolved entry null?
+    je        .LOP_SPUT_OBJECT_resolve                # if not, make it so
+.LOP_SPUT_OBJECT_finish:     # field ptr in eax
+    movzbl    rINST_HI,%ecx                      # ecx<- AA
+    GET_VREG(%ecx,%ecx)
+    FETCH_INST_WORD(2)
+    movl      %ecx,offStaticField_value(%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_BOOLEAN: /* 0x6a */
+/* File: x86/OP_SPUT_BOOLEAN.S */
+/* File: x86/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax                        # eax<- field ref BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx  # ecx<- DvmDex
+    movl      offDvmDex_pResFields(%ecx),%ecx    # ecx<- dvmDex->pResFields
+    movl      (%ecx,%eax,4),%eax                 # eax<- resolved StaticField ptr
+    testl     %eax,%eax                          # resolved entry null?
+    je        .LOP_SPUT_BOOLEAN_resolve                # if not, make it so
+.LOP_SPUT_BOOLEAN_finish:     # field ptr in eax
+    movzbl    rINST_HI,%ecx                      # ecx<- AA
+    GET_VREG(%ecx,%ecx)
+    FETCH_INST_WORD(2)
+    movl      %ecx,offStaticField_value(%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_BYTE: /* 0x6b */
+/* File: x86/OP_SPUT_BYTE.S */
+/* File: x86/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax                        # eax<- field ref BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx  # ecx<- DvmDex
+    movl      offDvmDex_pResFields(%ecx),%ecx    # ecx<- dvmDex->pResFields
+    movl      (%ecx,%eax,4),%eax                 # eax<- resolved StaticField ptr
+    testl     %eax,%eax                          # resolved entry null?
+    je        .LOP_SPUT_BYTE_resolve                # if not, make it so
+.LOP_SPUT_BYTE_finish:     # field ptr in eax
+    movzbl    rINST_HI,%ecx                      # ecx<- AA
+    GET_VREG(%ecx,%ecx)
+    FETCH_INST_WORD(2)
+    movl      %ecx,offStaticField_value(%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_CHAR: /* 0x6c */
+/* File: x86/OP_SPUT_CHAR.S */
+/* File: x86/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax                        # eax<- field ref BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx  # ecx<- DvmDex
+    movl      offDvmDex_pResFields(%ecx),%ecx    # ecx<- dvmDex->pResFields
+    movl      (%ecx,%eax,4),%eax                 # eax<- resolved StaticField ptr
+    testl     %eax,%eax                          # resolved entry null?
+    je        .LOP_SPUT_CHAR_resolve                # if not, make it so
+.LOP_SPUT_CHAR_finish:     # field ptr in eax
+    movzbl    rINST_HI,%ecx                      # ecx<- AA
+    GET_VREG(%ecx,%ecx)
+    FETCH_INST_WORD(2)
+    movl      %ecx,offStaticField_value(%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_SHORT: /* 0x6d */
+/* File: x86/OP_SPUT_SHORT.S */
+/* File: x86/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax                        # eax<- field ref BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx  # ecx<- DvmDex
+    movl      offDvmDex_pResFields(%ecx),%ecx    # ecx<- dvmDex->pResFields
+    movl      (%ecx,%eax,4),%eax                 # eax<- resolved StaticField ptr
+    testl     %eax,%eax                          # resolved entry null?
+    je        .LOP_SPUT_SHORT_resolve                # if not, make it so
+.LOP_SPUT_SHORT_finish:     # field ptr in eax
+    movzbl    rINST_HI,%ecx                      # ecx<- AA
+    GET_VREG(%ecx,%ecx)
+    FETCH_INST_WORD(2)
+    movl      %ecx,offStaticField_value(%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_VIRTUAL: /* 0x6e */
+/* File: x86/OP_INVOKE_VIRTUAL.S */
+
+    /*
+     * Handle a virtual method call.
+     *
+     * for: invoke-virtual, invoke-virtual/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    GET_GLUE(%eax)
+    movzwl    2(rPC),%ecx                 # ecx<- BBBB
+    movl      offGlue_methodClassDex(%eax),%eax  # eax<- pDvmDex
+    EXPORT_PC()
+    movl      offDvmDex_pResMethods(%eax),%eax   # eax<- pDvmDex->pResMethods
+    movl      (%eax,%ecx,4),%eax          # eax<- resolved baseMethod
+    testl     %eax,%eax                   # already resolved?
+    jne       .LOP_INVOKE_VIRTUAL_continue        # yes, continue
+    GET_GLUE(%eax)
+    movl      %ecx,OUT_ARG1(%esp)         # arg1<- ref
+    movl      offGlue_method(%eax),%eax   # eax<- glue->method
+    SPILL(rPC)
+    jmp       .LOP_INVOKE_VIRTUAL_more
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_SUPER: /* 0x6f */
+/* File: x86/OP_INVOKE_SUPER.S */
+    /*
+     * Handle a "super" method call.
+     *
+     * for: invoke-super, invoke-super/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    GET_GLUE(rINST_FULL)
+    movzwl    2(rPC),%eax               # eax<- BBBB
+    movl      offGlue_methodClassDex(rINST_FULL),%ecx # ecx<- pDvmDex
+    EXPORT_PC()
+    movl      offDvmDex_pResMethods(%ecx),%ecx # ecx<- pDvmDex->pResMethods
+    movl      (%ecx,%eax,4),%ecx        # ecx<- resolved baseMethod
+    movl      offGlue_method(rINST_FULL),%eax # eax<- method
+    movzwl    4(rPC),rINST_FULL         # rINST_FULL<- GFED or CCCC
+    .if       (!0)
+    andl      $0xf,rINST_FULL          # rINST_FULL<- D (or stays CCCC)
+    .endif
+    GET_VREG(rINST_FULL,rINST_FULL)     # rINST_FULL<- "this" ptr
+    testl     rINST_FULL,rINST_FULL     # null "this"?
+    je        common_errNullObject      # yes, throw
+    movl      offMethod_clazz(%eax),%eax # eax<- method->clazz
+    testl     %ecx,%ecx                 # already resolved?
+    jne       .LOP_INVOKE_SUPER_continue      # yes - go on
+    jmp       .LOP_INVOKE_SUPER_resolve
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_DIRECT: /* 0x70 */
+/* File: x86/OP_INVOKE_DIRECT.S */
+    /*
+     * Handle a direct method call.
+     *
+     * (We could defer the "is 'this' pointer null" test to the common
+     * method invocation code, and use a flag to indicate that static
+     * calls don't count.  If we do this as part of copying the arguments
+     * out we could avoiding loading the first arg twice.)
+     *
+     * for: invoke-direct, invoke-direct/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax              # eax<- BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx # ecx<- pDvmDex
+    EXPORT_PC()
+    SPILL(rPC)
+    movl      offDvmDex_pResMethods(%ecx),%ecx  # ecx<- pDvmDex->pResMethods
+    movzwl    4(rPC),rPC               # rPC<- GFED or CCCC
+    movl      (%ecx,%eax,4),%eax       # eax<- resolved methodToCall
+    .if       (!0)
+    andl      $0xf,rPC                # rPC<- D (or stays CCCC)
+    .endif
+    testl     %eax,%eax                # already resolved?
+    GET_VREG(%ecx,rPC)                 # ecx<- "this" ptr
+    je        .LOP_INVOKE_DIRECT_resolve      # not resolved, do it now
+.LOP_INVOKE_DIRECT_finish:
+    UNSPILL(rPC)
+    testl     %ecx,%ecx                # null "this"?
+    movl      $0,%ecx
+    #jne       common_invokeMethodNoRange  # no, continue on
+    jne       common_invokeOld          # no, continue on, eax<- method, ecx<- methodCallRange
+    jmp       common_errNullObject
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_STATIC: /* 0x71 */
+/* File: x86/OP_INVOKE_STATIC.S */
+    /*
+     * Handle a static method call.
+     *
+     * for: invoke-static, invoke-static/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax               # eax<- BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx # ecx<- pDvmDex
+    EXPORT_PC()
+    movl      offDvmDex_pResMethods(%ecx),%ecx  # ecx<- pDvmDex->pResMethods
+    movl      (%ecx,%eax,4),%eax        # eax<- resolved methodToCall
+    movl      $0,%ecx           # needed by common_invokeOld - revisit
+    testl     %eax,%eax
+    #jne       common_invokeMethodNoRange
+    jne       common_invokeOld
+    GET_GLUE(%ecx)
+    movl      offGlue_method(%ecx),%ecx # ecx<- glue->method
+    movzwl    2(rPC),%eax
+    movl      offMethod_clazz(%ecx),%ecx# ecx<- method->clazz
+    movl      %eax,OUT_ARG1(%esp)       # arg1<- BBBB
+    movl      %ecx,OUT_ARG0(%esp)       # arg0<- clazz
+    jmp       .LOP_INVOKE_STATIC_continue
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_INTERFACE: /* 0x72 */
+/* File: x86/OP_INVOKE_INTERFACE.S */
+    /*
+     * Handle an interface method call.
+     *
+     * for: invoke-interface, invoke-interface/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    movzwl     4(rPC),%eax              # eax<- FEDC or CCCC
+    GET_GLUE(%ecx)
+    .if        (!0)
+    andl       $0xf,%eax               # eax<- C (or stays CCCC)
+    .endif
+    GET_VREG(%eax,%eax)                 # eax<- "this"
+    EXPORT_PC()
+    testl      %eax,%eax                # null this?
+    je         common_errNullObject     # yes, fail
+    movl       offObject_clazz(%eax),%eax# eax<- thisPtr->clazz
+    movl       %eax,OUT_ARG0(%esp)                 # arg0<- class
+    movl       offGlue_methodClassDex(%ecx),%eax   # eax<- methodClassDex
+    movl       offGlue_method(%ecx),%ecx           # ecx<- method
+    movl       %eax,OUT_ARG3(%esp)                 # arg3<- dex
+    movzwl     2(rPC),%eax                         # eax<- BBBB
+    movl       %ecx,OUT_ARG2(%esp)                 # arg2<- method
+    movl       %eax,OUT_ARG1(%esp)                 # arg1<- BBBB
+    SPILL(rPC)
+    jmp        .LOP_INVOKE_INTERFACE_continue
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_73: /* 0x73 */
+/* File: x86/OP_UNUSED_73.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_VIRTUAL_RANGE: /* 0x74 */
+/* File: x86/OP_INVOKE_VIRTUAL_RANGE.S */
+/* File: x86/OP_INVOKE_VIRTUAL.S */
+
+    /*
+     * Handle a virtual method call.
+     *
+     * for: invoke-virtual, invoke-virtual/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    GET_GLUE(%eax)
+    movzwl    2(rPC),%ecx                 # ecx<- BBBB
+    movl      offGlue_methodClassDex(%eax),%eax  # eax<- pDvmDex
+    EXPORT_PC()
+    movl      offDvmDex_pResMethods(%eax),%eax   # eax<- pDvmDex->pResMethods
+    movl      (%eax,%ecx,4),%eax          # eax<- resolved baseMethod
+    testl     %eax,%eax                   # already resolved?
+    jne       .LOP_INVOKE_VIRTUAL_RANGE_continue        # yes, continue
+    GET_GLUE(%eax)
+    movl      %ecx,OUT_ARG1(%esp)         # arg1<- ref
+    movl      offGlue_method(%eax),%eax   # eax<- glue->method
+    SPILL(rPC)
+    jmp       .LOP_INVOKE_VIRTUAL_RANGE_more
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_SUPER_RANGE: /* 0x75 */
+/* File: x86/OP_INVOKE_SUPER_RANGE.S */
+/* File: x86/OP_INVOKE_SUPER.S */
+    /*
+     * Handle a "super" method call.
+     *
+     * for: invoke-super, invoke-super/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    GET_GLUE(rINST_FULL)
+    movzwl    2(rPC),%eax               # eax<- BBBB
+    movl      offGlue_methodClassDex(rINST_FULL),%ecx # ecx<- pDvmDex
+    EXPORT_PC()
+    movl      offDvmDex_pResMethods(%ecx),%ecx # ecx<- pDvmDex->pResMethods
+    movl      (%ecx,%eax,4),%ecx        # ecx<- resolved baseMethod
+    movl      offGlue_method(rINST_FULL),%eax # eax<- method
+    movzwl    4(rPC),rINST_FULL         # rINST_FULL<- GFED or CCCC
+    .if       (!1)
+    andl      $0xf,rINST_FULL          # rINST_FULL<- D (or stays CCCC)
+    .endif
+    GET_VREG(rINST_FULL,rINST_FULL)     # rINST_FULL<- "this" ptr
+    testl     rINST_FULL,rINST_FULL     # null "this"?
+    je        common_errNullObject      # yes, throw
+    movl      offMethod_clazz(%eax),%eax # eax<- method->clazz
+    testl     %ecx,%ecx                 # already resolved?
+    jne       .LOP_INVOKE_SUPER_RANGE_continue      # yes - go on
+    jmp       .LOP_INVOKE_SUPER_RANGE_resolve
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_DIRECT_RANGE: /* 0x76 */
+/* File: x86/OP_INVOKE_DIRECT_RANGE.S */
+/* File: x86/OP_INVOKE_DIRECT.S */
+    /*
+     * Handle a direct method call.
+     *
+     * (We could defer the "is 'this' pointer null" test to the common
+     * method invocation code, and use a flag to indicate that static
+     * calls don't count.  If we do this as part of copying the arguments
+     * out we could avoiding loading the first arg twice.)
+     *
+     * for: invoke-direct, invoke-direct/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax              # eax<- BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx # ecx<- pDvmDex
+    EXPORT_PC()
+    SPILL(rPC)
+    movl      offDvmDex_pResMethods(%ecx),%ecx  # ecx<- pDvmDex->pResMethods
+    movzwl    4(rPC),rPC               # rPC<- GFED or CCCC
+    movl      (%ecx,%eax,4),%eax       # eax<- resolved methodToCall
+    .if       (!1)
+    andl      $0xf,rPC                # rPC<- D (or stays CCCC)
+    .endif
+    testl     %eax,%eax                # already resolved?
+    GET_VREG(%ecx,rPC)                 # ecx<- "this" ptr
+    je        .LOP_INVOKE_DIRECT_RANGE_resolve      # not resolved, do it now
+.LOP_INVOKE_DIRECT_RANGE_finish:
+    UNSPILL(rPC)
+    testl     %ecx,%ecx                # null "this"?
+    movl      $1,%ecx
+    #jne       common_invokeMethodRange  # no, continue on
+    jne       common_invokeOld          # no, continue on, eax<- method, ecx<- methodCallRange
+    jmp       common_errNullObject
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_STATIC_RANGE: /* 0x77 */
+/* File: x86/OP_INVOKE_STATIC_RANGE.S */
+/* File: x86/OP_INVOKE_STATIC.S */
+    /*
+     * Handle a static method call.
+     *
+     * for: invoke-static, invoke-static/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    2(rPC),%eax               # eax<- BBBB
+    movl      offGlue_methodClassDex(%ecx),%ecx # ecx<- pDvmDex
+    EXPORT_PC()
+    movl      offDvmDex_pResMethods(%ecx),%ecx  # ecx<- pDvmDex->pResMethods
+    movl      (%ecx,%eax,4),%eax        # eax<- resolved methodToCall
+    movl      $1,%ecx           # needed by common_invokeOld - revisit
+    testl     %eax,%eax
+    #jne       common_invokeMethodRange
+    jne       common_invokeOld
+    GET_GLUE(%ecx)
+    movl      offGlue_method(%ecx),%ecx # ecx<- glue->method
+    movzwl    2(rPC),%eax
+    movl      offMethod_clazz(%ecx),%ecx# ecx<- method->clazz
+    movl      %eax,OUT_ARG1(%esp)       # arg1<- BBBB
+    movl      %ecx,OUT_ARG0(%esp)       # arg0<- clazz
+    jmp       .LOP_INVOKE_STATIC_RANGE_continue
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_INTERFACE_RANGE: /* 0x78 */
+/* File: x86/OP_INVOKE_INTERFACE_RANGE.S */
+/* File: x86/OP_INVOKE_INTERFACE.S */
+    /*
+     * Handle an interface method call.
+     *
+     * for: invoke-interface, invoke-interface/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    movzwl     4(rPC),%eax              # eax<- FEDC or CCCC
+    GET_GLUE(%ecx)
+    .if        (!1)
+    andl       $0xf,%eax               # eax<- C (or stays CCCC)
+    .endif
+    GET_VREG(%eax,%eax)                 # eax<- "this"
+    EXPORT_PC()
+    testl      %eax,%eax                # null this?
+    je         common_errNullObject     # yes, fail
+    movl       offObject_clazz(%eax),%eax# eax<- thisPtr->clazz
+    movl       %eax,OUT_ARG0(%esp)                 # arg0<- class
+    movl       offGlue_methodClassDex(%ecx),%eax   # eax<- methodClassDex
+    movl       offGlue_method(%ecx),%ecx           # ecx<- method
+    movl       %eax,OUT_ARG3(%esp)                 # arg3<- dex
+    movzwl     2(rPC),%eax                         # eax<- BBBB
+    movl       %ecx,OUT_ARG2(%esp)                 # arg2<- method
+    movl       %eax,OUT_ARG1(%esp)                 # arg1<- BBBB
+    SPILL(rPC)
+    jmp        .LOP_INVOKE_INTERFACE_RANGE_continue
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_79: /* 0x79 */
+/* File: x86/OP_UNUSED_79.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_7A: /* 0x7a */
+/* File: x86/OP_UNUSED_7A.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEG_INT: /* 0x7b */
+/* File: x86/OP_NEG_INT.S */
+/* File: x86/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op eax".
+     */
+    /* unop vA, vB */
+    movzbl   rINST_HI,%ecx           # ecx<- A+
+    sarl     $12,rINST_FULL         # rINST_FULL<- B
+    GET_VREG(%eax,rINST_FULL)        # eax<- vB
+    andb     $0xf,%cl               # ecx<- A
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    
+    
+    negl %eax
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NOT_INT: /* 0x7c */
+/* File: x86/OP_NOT_INT.S */
+/* File: x86/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op eax".
+     */
+    /* unop vA, vB */
+    movzbl   rINST_HI,%ecx           # ecx<- A+
+    sarl     $12,rINST_FULL         # rINST_FULL<- B
+    GET_VREG(%eax,rINST_FULL)        # eax<- vB
+    andb     $0xf,%cl               # ecx<- A
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    
+    
+    notl %eax
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEG_LONG: /* 0x7d */
+/* File: x86/OP_NEG_LONG.S */
+    /* unop vA, vB */
+    movzbl    rINST_HI,%ecx            # ecx<- BA
+    sarl      $4,%ecx                 # ecx<- B
+    movzbl    rINST_HI,rINST_FULL      # ecx<- BA
+    andb      $0xf,rINST_LO           # rINST_FULL<- A
+    GET_VREG_WORD(%eax,%ecx,0)         # eax<- v[B+0]
+    GET_VREG_WORD(%ecx,%ecx,1)         # ecx<- v[B+1]
+    negl      %eax
+    adcl      $0,%ecx
+    negl      %ecx
+    SET_VREG_WORD(%eax,rINST_FULL,0)   # v[A+0]<- eax
+    SET_VREG_WORD(%ecx,rINST_FULL,1)   # v[A+1]<- ecx
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NOT_LONG: /* 0x7e */
+/* File: x86/OP_NOT_LONG.S */
+    /* unop vA, vB */
+    movzbl    rINST_HI,%ecx            # ecx<- BA
+    sarl      $4,%ecx                 # ecx<- B
+    movzbl    rINST_HI,rINST_FULL      # ecx<- BA
+    andb      $0xf,rINST_LO           # rINST_FULL<- A
+    GET_VREG_WORD(%eax,%ecx,0)         # eax<- v[B+0]
+    GET_VREG_WORD(%ecx,%ecx,1)         # ecx<- v[B+1]
+    notl      %eax
+    notl      %ecx
+    SET_VREG_WORD(%eax,rINST_FULL,0)   # v[A+0]<- eax
+    SET_VREG_WORD(%ecx,rINST_FULL,1)   # v[A+1]<- ecx
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEG_FLOAT: /* 0x7f */
+/* File: x86/OP_NEG_FLOAT.S */
+/* File: x86/fpcvt.S */
+    /*
+     * Generic 32-bit FP conversion operation.
+     */
+    /* unop vA, vB */
+    movzbl   rINST_HI,%ecx           # ecx<- A+
+    sarl     $12,rINST_FULL         # rINST_FULL<- B
+    flds    (rFP,rINST_FULL,4)      # %st0<- vB
+    andb     $0xf,%cl               # ecx<- A
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    fchs
+    fstps  (rFP,%ecx,4)             # vA<- %st0
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEG_DOUBLE: /* 0x80 */
+/* File: x86/OP_NEG_DOUBLE.S */
+/* File: x86/fpcvt.S */
+    /*
+     * Generic 32-bit FP conversion operation.
+     */
+    /* unop vA, vB */
+    movzbl   rINST_HI,%ecx           # ecx<- A+
+    sarl     $12,rINST_FULL         # rINST_FULL<- B
+    fldl    (rFP,rINST_FULL,4)      # %st0<- vB
+    andb     $0xf,%cl               # ecx<- A
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    fchs
+    fstpl  (rFP,%ecx,4)             # vA<- %st0
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_LONG: /* 0x81 */
+/* File: x86/OP_INT_TO_LONG.S */
+    /* int to long vA, vB */
+    movzbl  rINST_HI,%ecx               # ecx<- +A
+    sarl    $12,rINST_FULL             # rINST_FULL<- B
+    GET_VREG(%eax,rINST_FULL)           # eax<- vB
+    SPILL(rPC)                          # will step on edx later
+    andb    $0xf,%cl                   # ecx<- A
+    cltd                                # edx:eax<- sssssssBBBBBBBB
+    SET_VREG_WORD(%edx,%ecx,1)          # v[A+1]<- edx/rPC
+    UNSPILL(rPC)
+    SET_VREG_WORD(%eax,%ecx,0)          # v[A+0]<- %eax
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_FLOAT: /* 0x82 */
+/* File: x86/OP_INT_TO_FLOAT.S */
+/* File: x86/fpcvt.S */
+    /*
+     * Generic 32-bit FP conversion operation.
+     */
+    /* unop vA, vB */
+    movzbl   rINST_HI,%ecx           # ecx<- A+
+    sarl     $12,rINST_FULL         # rINST_FULL<- B
+    fildl    (rFP,rINST_FULL,4)      # %st0<- vB
+    andb     $0xf,%cl               # ecx<- A
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    
+    fstps  (rFP,%ecx,4)             # vA<- %st0
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_DOUBLE: /* 0x83 */
+/* File: x86/OP_INT_TO_DOUBLE.S */
+/* File: x86/fpcvt.S */
+    /*
+     * Generic 32-bit FP conversion operation.
+     */
+    /* unop vA, vB */
+    movzbl   rINST_HI,%ecx           # ecx<- A+
+    sarl     $12,rINST_FULL         # rINST_FULL<- B
+    fildl    (rFP,rINST_FULL,4)      # %st0<- vB
+    andb     $0xf,%cl               # ecx<- A
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    
+    fstpl  (rFP,%ecx,4)             # vA<- %st0
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_LONG_TO_INT: /* 0x84 */
+/* File: x86/OP_LONG_TO_INT.S */
+/* we ignore the high word, making this equivalent to a 32-bit reg move */
+/* File: x86/OP_MOVE.S */
+    /* for move, move-object, long-to-int */
+    /* op vA, vB */
+    movzbl rINST_HI,%eax         # eax<- BA
+    andb   $0xf,%al             # eax<- A
+    shrl   $12,rINST_FULL       # rINST_FULL<- B
+    GET_VREG(%ecx,rINST_FULL)
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    SET_VREG(%ecx,%eax)          # fp[A]<-fp[B]
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_LONG_TO_FLOAT: /* 0x85 */
+/* File: x86/OP_LONG_TO_FLOAT.S */
+/* File: x86/fpcvt.S */
+    /*
+     * Generic 32-bit FP conversion operation.
+     */
+    /* unop vA, vB */
+    movzbl   rINST_HI,%ecx           # ecx<- A+
+    sarl     $12,rINST_FULL         # rINST_FULL<- B
+    fildll    (rFP,rINST_FULL,4)      # %st0<- vB
+    andb     $0xf,%cl               # ecx<- A
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    
+    fstps  (rFP,%ecx,4)             # vA<- %st0
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_LONG_TO_DOUBLE: /* 0x86 */
+/* File: x86/OP_LONG_TO_DOUBLE.S */
+/* File: x86/fpcvt.S */
+    /*
+     * Generic 32-bit FP conversion operation.
+     */
+    /* unop vA, vB */
+    movzbl   rINST_HI,%ecx           # ecx<- A+
+    sarl     $12,rINST_FULL         # rINST_FULL<- B
+    fildll    (rFP,rINST_FULL,4)      # %st0<- vB
+    andb     $0xf,%cl               # ecx<- A
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    
+    fstpl  (rFP,%ecx,4)             # vA<- %st0
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FLOAT_TO_INT: /* 0x87 */
+/* File: x86/OP_FLOAT_TO_INT.S */
+/* File: x86/cvtfp_int.S */
+/* On fp to int conversions, Java requires that
+ * if the result > maxint, it should be clamped to maxint.  If it is less
+ * than minint, it should be clamped to minint.  If it is a nan, the result
+ * should be zero.  Further, the rounding mode is to truncate.  This model
+ * differs from what is delivered normally via the x86 fpu, so we have
+ * to play some games.
+ */
+    /* float/double to int/long vA, vB */
+    movzbl    rINST_HI,%ecx           # ecx<- A+
+    sarl      $12,rINST_FULL         # rINST_FULL<- B
+    .if 0
+    fldl     (rFP,rINST_FULL,4)       # %st0<- vB
+    .else
+    flds     (rFP,rINST_FULL,4)       # %st0<- vB
+    .endif
+    ftst
+    fnstcw   LOCAL0_OFFSET(%ebp)      # remember original rounding mode
+    movzwl   LOCAL0_OFFSET(%ebp),%eax
+    movb     $0xc,%ah
+    movw     %ax,LOCAL0_OFFSET+2(%ebp)
+    fldcw    LOCAL0_OFFSET+2(%ebp)      # set "to zero" rounding mode
+    FETCH_INST_WORD(1)
+    andb     $0xf,%cl                # ecx<- A
+    .if 0
+    fistpll  (rFP,%ecx,4)             # convert and store
+    .else
+    fistpl   (rFP,%ecx,4)             # convert and store
+    .endif
+    fldcw    LOCAL0_OFFSET(%ebp)      # restore previous rounding mode
+    jmp      .LOP_FLOAT_TO_INT_continue
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FLOAT_TO_LONG: /* 0x88 */
+/* File: x86/OP_FLOAT_TO_LONG.S */
+/* File: x86/cvtfp_int.S */
+/* On fp to int conversions, Java requires that
+ * if the result > maxint, it should be clamped to maxint.  If it is less
+ * than minint, it should be clamped to minint.  If it is a nan, the result
+ * should be zero.  Further, the rounding mode is to truncate.  This model
+ * differs from what is delivered normally via the x86 fpu, so we have
+ * to play some games.
+ */
+    /* float/double to int/long vA, vB */
+    movzbl    rINST_HI,%ecx           # ecx<- A+
+    sarl      $12,rINST_FULL         # rINST_FULL<- B
+    .if 0
+    fldl     (rFP,rINST_FULL,4)       # %st0<- vB
+    .else
+    flds     (rFP,rINST_FULL,4)       # %st0<- vB
+    .endif
+    ftst
+    fnstcw   LOCAL0_OFFSET(%ebp)      # remember original rounding mode
+    movzwl   LOCAL0_OFFSET(%ebp),%eax
+    movb     $0xc,%ah
+    movw     %ax,LOCAL0_OFFSET+2(%ebp)
+    fldcw    LOCAL0_OFFSET+2(%ebp)      # set "to zero" rounding mode
+    FETCH_INST_WORD(1)
+    andb     $0xf,%cl                # ecx<- A
+    .if 1
+    fistpll  (rFP,%ecx,4)             # convert and store
+    .else
+    fistpl   (rFP,%ecx,4)             # convert and store
+    .endif
+    fldcw    LOCAL0_OFFSET(%ebp)      # restore previous rounding mode
+    jmp      .LOP_FLOAT_TO_LONG_continue
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FLOAT_TO_DOUBLE: /* 0x89 */
+/* File: x86/OP_FLOAT_TO_DOUBLE.S */
+/* File: x86/fpcvt.S */
+    /*
+     * Generic 32-bit FP conversion operation.
+     */
+    /* unop vA, vB */
+    movzbl   rINST_HI,%ecx           # ecx<- A+
+    sarl     $12,rINST_FULL         # rINST_FULL<- B
+    flds    (rFP,rINST_FULL,4)      # %st0<- vB
+    andb     $0xf,%cl               # ecx<- A
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    
+    fstpl  (rFP,%ecx,4)             # vA<- %st0
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DOUBLE_TO_INT: /* 0x8a */
+/* File: x86/OP_DOUBLE_TO_INT.S */
+/* File: x86/cvtfp_int.S */
+/* On fp to int conversions, Java requires that
+ * if the result > maxint, it should be clamped to maxint.  If it is less
+ * than minint, it should be clamped to minint.  If it is a nan, the result
+ * should be zero.  Further, the rounding mode is to truncate.  This model
+ * differs from what is delivered normally via the x86 fpu, so we have
+ * to play some games.
+ */
+    /* float/double to int/long vA, vB */
+    movzbl    rINST_HI,%ecx           # ecx<- A+
+    sarl      $12,rINST_FULL         # rINST_FULL<- B
+    .if 1
+    fldl     (rFP,rINST_FULL,4)       # %st0<- vB
+    .else
+    flds     (rFP,rINST_FULL,4)       # %st0<- vB
+    .endif
+    ftst
+    fnstcw   LOCAL0_OFFSET(%ebp)      # remember original rounding mode
+    movzwl   LOCAL0_OFFSET(%ebp),%eax
+    movb     $0xc,%ah
+    movw     %ax,LOCAL0_OFFSET+2(%ebp)
+    fldcw    LOCAL0_OFFSET+2(%ebp)      # set "to zero" rounding mode
+    FETCH_INST_WORD(1)
+    andb     $0xf,%cl                # ecx<- A
+    .if 0
+    fistpll  (rFP,%ecx,4)             # convert and store
+    .else
+    fistpl   (rFP,%ecx,4)             # convert and store
+    .endif
+    fldcw    LOCAL0_OFFSET(%ebp)      # restore previous rounding mode
+    jmp      .LOP_DOUBLE_TO_INT_continue
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DOUBLE_TO_LONG: /* 0x8b */
+/* File: x86/OP_DOUBLE_TO_LONG.S */
+/* File: x86/cvtfp_int.S */
+/* On fp to int conversions, Java requires that
+ * if the result > maxint, it should be clamped to maxint.  If it is less
+ * than minint, it should be clamped to minint.  If it is a nan, the result
+ * should be zero.  Further, the rounding mode is to truncate.  This model
+ * differs from what is delivered normally via the x86 fpu, so we have
+ * to play some games.
+ */
+    /* float/double to int/long vA, vB */
+    movzbl    rINST_HI,%ecx           # ecx<- A+
+    sarl      $12,rINST_FULL         # rINST_FULL<- B
+    .if 1
+    fldl     (rFP,rINST_FULL,4)       # %st0<- vB
+    .else
+    flds     (rFP,rINST_FULL,4)       # %st0<- vB
+    .endif
+    ftst
+    fnstcw   LOCAL0_OFFSET(%ebp)      # remember original rounding mode
+    movzwl   LOCAL0_OFFSET(%ebp),%eax
+    movb     $0xc,%ah
+    movw     %ax,LOCAL0_OFFSET+2(%ebp)
+    fldcw    LOCAL0_OFFSET+2(%ebp)      # set "to zero" rounding mode
+    FETCH_INST_WORD(1)
+    andb     $0xf,%cl                # ecx<- A
+    .if 1
+    fistpll  (rFP,%ecx,4)             # convert and store
+    .else
+    fistpl   (rFP,%ecx,4)             # convert and store
+    .endif
+    fldcw    LOCAL0_OFFSET(%ebp)      # restore previous rounding mode
+    jmp      .LOP_DOUBLE_TO_LONG_continue
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DOUBLE_TO_FLOAT: /* 0x8c */
+/* File: x86/OP_DOUBLE_TO_FLOAT.S */
+/* File: x86/fpcvt.S */
+    /*
+     * Generic 32-bit FP conversion operation.
+     */
+    /* unop vA, vB */
+    movzbl   rINST_HI,%ecx           # ecx<- A+
+    sarl     $12,rINST_FULL         # rINST_FULL<- B
+    fldl    (rFP,rINST_FULL,4)      # %st0<- vB
+    andb     $0xf,%cl               # ecx<- A
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    
+    fstps  (rFP,%ecx,4)             # vA<- %st0
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_BYTE: /* 0x8d */
+/* File: x86/OP_INT_TO_BYTE.S */
+/* File: x86/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op eax".
+     */
+    /* unop vA, vB */
+    movzbl   rINST_HI,%ecx           # ecx<- A+
+    sarl     $12,rINST_FULL         # rINST_FULL<- B
+    GET_VREG(%eax,rINST_FULL)        # eax<- vB
+    andb     $0xf,%cl               # ecx<- A
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    
+    
+    movsbl %al,%eax
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_CHAR: /* 0x8e */
+/* File: x86/OP_INT_TO_CHAR.S */
+/* File: x86/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op eax".
+     */
+    /* unop vA, vB */
+    movzbl   rINST_HI,%ecx           # ecx<- A+
+    sarl     $12,rINST_FULL         # rINST_FULL<- B
+    GET_VREG(%eax,rINST_FULL)        # eax<- vB
+    andb     $0xf,%cl               # ecx<- A
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    
+    
+    movzwl %ax,%eax
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_SHORT: /* 0x8f */
+/* File: x86/OP_INT_TO_SHORT.S */
+/* File: x86/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op eax".
+     */
+    /* unop vA, vB */
+    movzbl   rINST_HI,%ecx           # ecx<- A+
+    sarl     $12,rINST_FULL         # rINST_FULL<- B
+    GET_VREG(%eax,rINST_FULL)        # eax<- vB
+    andb     $0xf,%cl               # ecx<- A
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    
+    
+    movswl %ax,%eax
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_INT: /* 0x90 */
+/* File: x86/OP_ADD_INT.S */
+/* File: x86/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = eax op (rFP,%ecx,4)".
+     * This could be an x86 instruction or a function call.  (If the result
+     * comes back in a register other than eax, you can override "result".)
+     *
+     * For: add-int, sub-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int
+     */
+    /* binop vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- BB
+    movzbl   3(rPC),%ecx            # ecx<- CC
+    GET_VREG(%eax,%eax)             # eax<- vBB
+    addl (rFP,%ecx,4),%eax              # ex: addl    (rFP,%ecx,4),%eax
+    movzbl   rINST_HI,%ecx         # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_INT: /* 0x91 */
+/* File: x86/OP_SUB_INT.S */
+/* File: x86/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = eax op (rFP,%ecx,4)".
+     * This could be an x86 instruction or a function call.  (If the result
+     * comes back in a register other than eax, you can override "result".)
+     *
+     * For: add-int, sub-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int
+     */
+    /* binop vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- BB
+    movzbl   3(rPC),%ecx            # ecx<- CC
+    GET_VREG(%eax,%eax)             # eax<- vBB
+    subl   (rFP,%ecx,4),%eax              # ex: addl    (rFP,%ecx,4),%eax
+    movzbl   rINST_HI,%ecx         # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_INT: /* 0x92 */
+/* File: x86/OP_MUL_INT.S */
+    /*
+     * 32-bit binary multiplication.
+     */
+    /* mul vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- BB
+    movzbl   3(rPC),%ecx            # ecx<- CC
+    SPILL(rPC)
+    GET_VREG(%eax,%eax)             # eax<- vBB
+    imull    (rFP,%ecx,4),%eax      # trashes rPC/edx
+    UNSPILL(rPC)
+    movzbl   rINST_HI,%ecx          # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_INT: /* 0x93 */
+/* File: x86/OP_DIV_INT.S */
+/* File: x86/bindiv.S */
+
+    /*
+     * 32-bit binary div/rem operation.  Handles special case of op0=minint and
+     * op1=-1.
+     */
+    /* binop vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- BB
+    movzbl   3(rPC),%ecx            # ecx<- CC
+    GET_VREG(%eax,%eax)             # eax<- vBB
+    GET_VREG(%ecx,%ecx)             # eax<- vBB
+    SPILL(rPC)
+    cmpl     $0,%ecx
+    je       common_errDivideByZero
+    cmpl     $-1,%ecx
+    jne      .LOP_DIV_INT_continue_div
+    cmpl     $0x80000000,%eax
+    jne      .LOP_DIV_INT_continue_div
+    movl     $0x80000000,%eax
+    jmp      .LOP_DIV_INT_finish_div
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_INT: /* 0x94 */
+/* File: x86/OP_REM_INT.S */
+/* File: x86/bindiv.S */
+
+    /*
+     * 32-bit binary div/rem operation.  Handles special case of op0=minint and
+     * op1=-1.
+     */
+    /* binop vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- BB
+    movzbl   3(rPC),%ecx            # ecx<- CC
+    GET_VREG(%eax,%eax)             # eax<- vBB
+    GET_VREG(%ecx,%ecx)             # eax<- vBB
+    SPILL(rPC)
+    cmpl     $0,%ecx
+    je       common_errDivideByZero
+    cmpl     $-1,%ecx
+    jne      .LOP_REM_INT_continue_div
+    cmpl     $0x80000000,%eax
+    jne      .LOP_REM_INT_continue_div
+    movl     $0,%edx
+    jmp      .LOP_REM_INT_finish_div
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_INT: /* 0x95 */
+/* File: x86/OP_AND_INT.S */
+/* File: x86/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = eax op (rFP,%ecx,4)".
+     * This could be an x86 instruction or a function call.  (If the result
+     * comes back in a register other than eax, you can override "result".)
+     *
+     * For: add-int, sub-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int
+     */
+    /* binop vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- BB
+    movzbl   3(rPC),%ecx            # ecx<- CC
+    GET_VREG(%eax,%eax)             # eax<- vBB
+    andl   (rFP,%ecx,4),%eax              # ex: addl    (rFP,%ecx,4),%eax
+    movzbl   rINST_HI,%ecx         # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_INT: /* 0x96 */
+/* File: x86/OP_OR_INT.S */
+/* File: x86/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = eax op (rFP,%ecx,4)".
+     * This could be an x86 instruction or a function call.  (If the result
+     * comes back in a register other than eax, you can override "result".)
+     *
+     * For: add-int, sub-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int
+     */
+    /* binop vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- BB
+    movzbl   3(rPC),%ecx            # ecx<- CC
+    GET_VREG(%eax,%eax)             # eax<- vBB
+    orl   (rFP,%ecx,4),%eax              # ex: addl    (rFP,%ecx,4),%eax
+    movzbl   rINST_HI,%ecx         # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_INT: /* 0x97 */
+/* File: x86/OP_XOR_INT.S */
+/* File: x86/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = eax op (rFP,%ecx,4)".
+     * This could be an x86 instruction or a function call.  (If the result
+     * comes back in a register other than eax, you can override "result".)
+     *
+     * For: add-int, sub-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int
+     */
+    /* binop vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- BB
+    movzbl   3(rPC),%ecx            # ecx<- CC
+    GET_VREG(%eax,%eax)             # eax<- vBB
+    xorl   (rFP,%ecx,4),%eax              # ex: addl    (rFP,%ecx,4),%eax
+    movzbl   rINST_HI,%ecx         # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_INT: /* 0x98 */
+/* File: x86/OP_SHL_INT.S */
+/* File: x86/binop1.S */
+    /*
+     * Generic 32-bit binary operation in which both operands loaded to
+     * registers (op0 in eax, op1 in ecx).
+     */
+    /* binop vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- BB
+    movzbl   3(rPC),%ecx            # ecx<- CC
+    GET_VREG(%eax,%eax)             # eax<- vBB
+    GET_VREG(%ecx,%ecx)             # eax<- vBB
+    sall    %cl,%eax                          # ex: addl    %ecx,%eax
+    movzbl   rINST_HI,%ecx          # tmp<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_INT: /* 0x99 */
+/* File: x86/OP_SHR_INT.S */
+/* File: x86/binop1.S */
+    /*
+     * Generic 32-bit binary operation in which both operands loaded to
+     * registers (op0 in eax, op1 in ecx).
+     */
+    /* binop vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- BB
+    movzbl   3(rPC),%ecx            # ecx<- CC
+    GET_VREG(%eax,%eax)             # eax<- vBB
+    GET_VREG(%ecx,%ecx)             # eax<- vBB
+    sarl    %cl,%eax                          # ex: addl    %ecx,%eax
+    movzbl   rINST_HI,%ecx          # tmp<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_INT: /* 0x9a */
+/* File: x86/OP_USHR_INT.S */
+/* File: x86/binop1.S */
+    /*
+     * Generic 32-bit binary operation in which both operands loaded to
+     * registers (op0 in eax, op1 in ecx).
+     */
+    /* binop vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- BB
+    movzbl   3(rPC),%ecx            # ecx<- CC
+    GET_VREG(%eax,%eax)             # eax<- vBB
+    GET_VREG(%ecx,%ecx)             # eax<- vBB
+    shrl    %cl,%eax                          # ex: addl    %ecx,%eax
+    movzbl   rINST_HI,%ecx          # tmp<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    SET_VREG(%eax,%ecx)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_LONG: /* 0x9b */
+/* File: x86/OP_ADD_LONG.S */
+/* File: x86/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.
+     */
+    /* binop vAA, vBB, vCC */
+
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    SPILL(rPC)
+    GET_VREG_WORD(rPC,%eax,0)           # rPC<- v[BB+0]
+    GET_VREG_WORD(%eax,%eax,1)          # eax<- v[BB+1]
+    addl (rFP,%ecx,4),rPC         # ex: addl   (rFP,%ecx,4),rPC
+    adcl 4(rFP,%ecx,4),%eax         # ex: adcl   4(rFP,%ecx,4),%eax
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    SET_VREG_WORD(rPC,rINST_FULL,0)     # v[AA+0] <- rPC
+    UNSPILL(rPC)
+    SET_VREG_WORD(%eax,rINST_FULL,1)    # v[AA+1] <- eax
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_LONG: /* 0x9c */
+/* File: x86/OP_SUB_LONG.S */
+/* File: x86/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.
+     */
+    /* binop vAA, vBB, vCC */
+
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    SPILL(rPC)
+    GET_VREG_WORD(rPC,%eax,0)           # rPC<- v[BB+0]
+    GET_VREG_WORD(%eax,%eax,1)          # eax<- v[BB+1]
+    subl (rFP,%ecx,4),rPC         # ex: addl   (rFP,%ecx,4),rPC
+    sbbl 4(rFP,%ecx,4),%eax         # ex: adcl   4(rFP,%ecx,4),%eax
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    SET_VREG_WORD(rPC,rINST_FULL,0)     # v[AA+0] <- rPC
+    UNSPILL(rPC)
+    SET_VREG_WORD(%eax,rINST_FULL,1)    # v[AA+1] <- eax
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_LONG: /* 0x9d */
+/* File: x86/OP_MUL_LONG.S */
+    /*
+     * Signed 64-bit integer multiply.
+     *
+     * We could definately use more free registers for
+     * this code.  We must spill rPC (edx) because it
+     * is used by imul.  We'll also spill rINST (ebx),
+     * giving us eax, ebc, ecx and edx as computational
+     * temps.  On top of that, we'll spill rIBASE (edi)
+     * for use as the vB pointer and rFP (esi) for use
+     * as the vC pointer.  Yuck.
+     */
+    /* mul-long vAA, vBB, vCC */
+    movzbl    2(rPC),%eax              # eax<- B
+    movzbl    3(rPC),%ecx              # ecx<- C
+    SPILL(rPC)
+    SPILL(rIBASE)
+    SPILL(rFP)
+    SPILL(rINST_FULL)
+    leal      (rFP,%eax,4),rIBASE      # rIBASE<- &v[B]
+    leal      (rFP,%ecx,4),rFP         # rFP<- &v[C]
+    movl      4(rIBASE),%ecx      # ecx<- Bmsw
+    imull     (rFP),%ecx          # ecx<- (Bmsw*Clsw)
+    movl      4(rFP),%eax         # eax<- Cmsw
+    imull     (rIBASE),%eax       # eax<- (Cmsw*Blsw)
+    addl      %eax,%ecx           # ecx<- (Bmsw*Clsw)+(Cmsw*Blsw)
+    movl      (rFP),%eax          # eax<- Clsw
+    mull      (rIBASE)            # eax<- (Clsw*Alsw)
+    UNSPILL(rINST_FULL)
+    UNSPILL(rFP)
+    jmp       .LOP_MUL_LONG_continue
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_LONG: /* 0x9e */
+/* File: x86/OP_DIV_LONG.S */
+    /* div vAA, vBB, vCC */
+    movzbl    3(rPC),%eax              # eax<- CC
+    movzbl    2(rPC),%ecx              # ecx<- BB
+    SPILL(rPC)
+    GET_VREG_WORD(rPC,%eax,0)
+    GET_VREG_WORD(%eax,%eax,1)
+    movl     rPC,OUT_ARG2(%esp)
+    testl    %eax,%eax
+    je       .LOP_DIV_LONG_check_zero
+    cmpl     $-1,%eax
+    je       .LOP_DIV_LONG_check_neg1
+.LOP_DIV_LONG_notSpecial:
+    GET_VREG_WORD(rPC,%ecx,0)
+    GET_VREG_WORD(%ecx,%ecx,1)
+.LOP_DIV_LONG_notSpecial1:
+    movl     %eax,OUT_ARG3(%esp)
+    movl     rPC,OUT_ARG0(%esp)
+    movl     %ecx,OUT_ARG1(%esp)
+    jmp      .LOP_DIV_LONG_continue
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_LONG: /* 0x9f */
+/* File: x86/OP_REM_LONG.S */
+/* File: x86/OP_DIV_LONG.S */
+    /* div vAA, vBB, vCC */
+    movzbl    3(rPC),%eax              # eax<- CC
+    movzbl    2(rPC),%ecx              # ecx<- BB
+    SPILL(rPC)
+    GET_VREG_WORD(rPC,%eax,0)
+    GET_VREG_WORD(%eax,%eax,1)
+    movl     rPC,OUT_ARG2(%esp)
+    testl    %eax,%eax
+    je       .LOP_REM_LONG_check_zero
+    cmpl     $-1,%eax
+    je       .LOP_REM_LONG_check_neg1
+.LOP_REM_LONG_notSpecial:
+    GET_VREG_WORD(rPC,%ecx,0)
+    GET_VREG_WORD(%ecx,%ecx,1)
+.LOP_REM_LONG_notSpecial1:
+    movl     %eax,OUT_ARG3(%esp)
+    movl     rPC,OUT_ARG0(%esp)
+    movl     %ecx,OUT_ARG1(%esp)
+    jmp      .LOP_REM_LONG_continue
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_LONG: /* 0xa0 */
+/* File: x86/OP_AND_LONG.S */
+/* File: x86/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.
+     */
+    /* binop vAA, vBB, vCC */
+
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    SPILL(rPC)
+    GET_VREG_WORD(rPC,%eax,0)           # rPC<- v[BB+0]
+    GET_VREG_WORD(%eax,%eax,1)          # eax<- v[BB+1]
+    andl (rFP,%ecx,4),rPC         # ex: addl   (rFP,%ecx,4),rPC
+    andl 4(rFP,%ecx,4),%eax         # ex: adcl   4(rFP,%ecx,4),%eax
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    SET_VREG_WORD(rPC,rINST_FULL,0)     # v[AA+0] <- rPC
+    UNSPILL(rPC)
+    SET_VREG_WORD(%eax,rINST_FULL,1)    # v[AA+1] <- eax
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_LONG: /* 0xa1 */
+/* File: x86/OP_OR_LONG.S */
+/* File: x86/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.
+     */
+    /* binop vAA, vBB, vCC */
+
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    SPILL(rPC)
+    GET_VREG_WORD(rPC,%eax,0)           # rPC<- v[BB+0]
+    GET_VREG_WORD(%eax,%eax,1)          # eax<- v[BB+1]
+    orl (rFP,%ecx,4),rPC         # ex: addl   (rFP,%ecx,4),rPC
+    orl 4(rFP,%ecx,4),%eax         # ex: adcl   4(rFP,%ecx,4),%eax
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    SET_VREG_WORD(rPC,rINST_FULL,0)     # v[AA+0] <- rPC
+    UNSPILL(rPC)
+    SET_VREG_WORD(%eax,rINST_FULL,1)    # v[AA+1] <- eax
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_LONG: /* 0xa2 */
+/* File: x86/OP_XOR_LONG.S */
+/* File: x86/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.
+     */
+    /* binop vAA, vBB, vCC */
+
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    SPILL(rPC)
+    GET_VREG_WORD(rPC,%eax,0)           # rPC<- v[BB+0]
+    GET_VREG_WORD(%eax,%eax,1)          # eax<- v[BB+1]
+    xorl (rFP,%ecx,4),rPC         # ex: addl   (rFP,%ecx,4),rPC
+    xorl 4(rFP,%ecx,4),%eax         # ex: adcl   4(rFP,%ecx,4),%eax
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- AA
+    SET_VREG_WORD(rPC,rINST_FULL,0)     # v[AA+0] <- rPC
+    UNSPILL(rPC)
+    SET_VREG_WORD(%eax,rINST_FULL,1)    # v[AA+1] <- eax
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_LONG: /* 0xa3 */
+/* File: x86/OP_SHL_LONG.S */
+    /*
+     * Long integer shift.  This is different from the generic 32/64-bit
+     * binary operations because vAA/vBB are 64-bit but vCC (the shift
+     * distance) is 32-bit.  Also, Dalvik requires us to mask off the low
+     * 6 bits of the shift distance.  x86 shifts automatically mask off
+     * the low 5 bits of %cl, so have to handle the 64 > shiftcount > 31
+     * case specially.
+     */
+    /* shl-long vAA, vBB, vCC */
+    /* ecx gets shift count */
+    /* Need to spill edx */
+    /* rINST gets AA */
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    SPILL(rPC)                          # spill edx
+    GET_VREG_WORD(%edx,%eax,1)          # ecx<- v[BB+1]
+    GET_VREG  (%ecx,%ecx)               # ecx<- vCC
+    GET_VREG_WORD(%eax,%eax,0)          # eax<- v[BB+0]
+    shldl     %eax,%edx
+    sall      %cl,%eax
+    testb     $32,%cl
+    je        2f
+    movl      %eax,%edx
+    xorl      %eax,%eax
+2:
+    movzbl    rINST_HI,%ecx
+    SET_VREG_WORD(%edx,%ecx,1)         # v[AA+1]<- %edx
+    UNSPILL(rPC)
+    FETCH_INST_WORD(2)
+    jmp       .LOP_SHL_LONG_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_LONG: /* 0xa4 */
+/* File: x86/OP_SHR_LONG.S */
+    /*
+     * Long integer shift.  This is different from the generic 32/64-bit
+     * binary operations because vAA/vBB are 64-bit but vCC (the shift
+     * distance) is 32-bit.  Also, Dalvik requires us to mask off the low
+     * 6 bits of the shift distance.  x86 shifts automatically mask off
+     * the low 5 bits of %cl, so have to handle the 64 > shiftcount > 31
+     * case specially.
+     */
+    /* shr-long vAA, vBB, vCC */
+    /* ecx gets shift count */
+    /* Need to spill edx */
+    /* rINST gets AA */
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    SPILL(rPC)                          # spill edx
+    GET_VREG_WORD(%edx,%eax,1)          # edx<- v[BB+1]
+    GET_VREG  (%ecx,%ecx)               # ecx<- vCC
+    GET_VREG_WORD(%eax,%eax,0)          # eax<- v[BB+0]
+    shrdl     %edx,%eax
+    sarl      %cl,%edx
+    testb     $32,%cl
+    je        2f
+    movl      %edx,%eax
+    sarl      $31,%edx
+2:
+    movzbl    rINST_HI,%ecx
+    SET_VREG_WORD(%edx,%ecx,1)         # v[AA+1]<- edx
+    UNSPILL(rPC)
+    FETCH_INST_WORD(2)
+    jmp       .LOP_SHR_LONG_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_LONG: /* 0xa5 */
+/* File: x86/OP_USHR_LONG.S */
+    /*
+     * Long integer shift.  This is different from the generic 32/64-bit
+     * binary operations because vAA/vBB are 64-bit but vCC (the shift
+     * distance) is 32-bit.  Also, Dalvik requires us to mask off the low
+     * 6 bits of the shift distance.  x86 shifts automatically mask off
+     * the low 5 bits of %cl, so have to handle the 64 > shiftcount > 31
+     * case specially.
+     */
+    /* shr-long vAA, vBB, vCC */
+    /* ecx gets shift count */
+    /* Need to spill edx */
+    /* rINST gets AA */
+    movzbl    2(rPC),%eax               # eax<- BB
+    movzbl    3(rPC),%ecx               # ecx<- CC
+    SPILL(rPC)                          # spill edx
+    GET_VREG_WORD(%edx,%eax,1)          # edx<- v[BB+1]
+    GET_VREG  (%ecx,%ecx)               # ecx<- vCC
+    GET_VREG_WORD(%eax,%eax,0)          # eax<- v[BB+0]
+    shrdl     %edx,%eax
+    shrl      %cl,%edx
+    testb     $32,%cl
+    je        2f
+    movl      %edx,%eax
+    xorl      %edx,%edx
+2:
+    movzbl    rINST_HI,%ecx
+    SET_VREG_WORD(%edx,%ecx,1)         # v[BB+1]<- edx
+    UNSPILL(rPC)
+    jmp       .LOP_USHR_LONG_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_FLOAT: /* 0xa6 */
+/* File: x86/OP_ADD_FLOAT.S */
+/* File: x86/binflop.S */
+    /*
+     * Generic 32-bit binary float operation.
+     *
+     * For: add-fp, sub-fp, mul-fp, div-fp
+     */
+    /* binop vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- CC
+    movzbl   3(rPC),%ecx            # ecx<- BB
+    flds    (rFP,%eax,4)           # vCC to fp stack
+    fadds   (rFP,%ecx,4)           # ex: faddp
+    movzbl   rINST_HI,%ecx         # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    fstps   (rFP,%ecx,4)           # %st to vAA
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_FLOAT: /* 0xa7 */
+/* File: x86/OP_SUB_FLOAT.S */
+/* File: x86/binflop.S */
+    /*
+     * Generic 32-bit binary float operation.
+     *
+     * For: add-fp, sub-fp, mul-fp, div-fp
+     */
+    /* binop vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- CC
+    movzbl   3(rPC),%ecx            # ecx<- BB
+    flds    (rFP,%eax,4)           # vCC to fp stack
+    fsubs   (rFP,%ecx,4)           # ex: faddp
+    movzbl   rINST_HI,%ecx         # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    fstps   (rFP,%ecx,4)           # %st to vAA
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_FLOAT: /* 0xa8 */
+/* File: x86/OP_MUL_FLOAT.S */
+/* File: x86/binflop.S */
+    /*
+     * Generic 32-bit binary float operation.
+     *
+     * For: add-fp, sub-fp, mul-fp, div-fp
+     */
+    /* binop vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- CC
+    movzbl   3(rPC),%ecx            # ecx<- BB
+    flds    (rFP,%eax,4)           # vCC to fp stack
+    fmuls   (rFP,%ecx,4)           # ex: faddp
+    movzbl   rINST_HI,%ecx         # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    fstps   (rFP,%ecx,4)           # %st to vAA
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_FLOAT: /* 0xa9 */
+/* File: x86/OP_DIV_FLOAT.S */
+/* File: x86/binflop.S */
+    /*
+     * Generic 32-bit binary float operation.
+     *
+     * For: add-fp, sub-fp, mul-fp, div-fp
+     */
+    /* binop vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- CC
+    movzbl   3(rPC),%ecx            # ecx<- BB
+    flds    (rFP,%eax,4)           # vCC to fp stack
+    fdivs   (rFP,%ecx,4)           # ex: faddp
+    movzbl   rINST_HI,%ecx         # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    fstps   (rFP,%ecx,4)           # %st to vAA
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_FLOAT: /* 0xaa */
+/* File: x86/OP_REM_FLOAT.S */
+    /* rem_float vAA, vBB, vCC */
+    movzbl   3(rPC),%ecx            # ecx<- BB
+    movzbl   2(rPC),%eax            # eax<- CC
+    flds     (rFP,%ecx,4)           # vCC to fp stack
+    flds     (rFP,%eax,4)           # vCC to fp stack
+    movzbl   rINST_HI,%ecx         # ecx<- AA
+    FETCH_INST_WORD(2)
+1:
+    fprem
+    fstsw     %ax
+    sahf
+    jp        1b
+    fstp      %st(1)
+    ADVANCE_PC(2)
+    fstps    (rFP,%ecx,4)           # %st to vAA
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_DOUBLE: /* 0xab */
+/* File: x86/OP_ADD_DOUBLE.S */
+/* File: x86/binflop.S */
+    /*
+     * Generic 32-bit binary float operation.
+     *
+     * For: add-fp, sub-fp, mul-fp, div-fp
+     */
+    /* binop vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- CC
+    movzbl   3(rPC),%ecx            # ecx<- BB
+    fldl    (rFP,%eax,4)           # vCC to fp stack
+    faddl   (rFP,%ecx,4)           # ex: faddp
+    movzbl   rINST_HI,%ecx         # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    fstpl   (rFP,%ecx,4)           # %st to vAA
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_DOUBLE: /* 0xac */
+/* File: x86/OP_SUB_DOUBLE.S */
+/* File: x86/binflop.S */
+    /*
+     * Generic 32-bit binary float operation.
+     *
+     * For: add-fp, sub-fp, mul-fp, div-fp
+     */
+    /* binop vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- CC
+    movzbl   3(rPC),%ecx            # ecx<- BB
+    fldl    (rFP,%eax,4)           # vCC to fp stack
+    fsubl   (rFP,%ecx,4)           # ex: faddp
+    movzbl   rINST_HI,%ecx         # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    fstpl   (rFP,%ecx,4)           # %st to vAA
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_DOUBLE: /* 0xad */
+/* File: x86/OP_MUL_DOUBLE.S */
+/* File: x86/binflop.S */
+    /*
+     * Generic 32-bit binary float operation.
+     *
+     * For: add-fp, sub-fp, mul-fp, div-fp
+     */
+    /* binop vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- CC
+    movzbl   3(rPC),%ecx            # ecx<- BB
+    fldl    (rFP,%eax,4)           # vCC to fp stack
+    fmull   (rFP,%ecx,4)           # ex: faddp
+    movzbl   rINST_HI,%ecx         # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    fstpl   (rFP,%ecx,4)           # %st to vAA
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_DOUBLE: /* 0xae */
+/* File: x86/OP_DIV_DOUBLE.S */
+/* File: x86/binflop.S */
+    /*
+     * Generic 32-bit binary float operation.
+     *
+     * For: add-fp, sub-fp, mul-fp, div-fp
+     */
+    /* binop vAA, vBB, vCC */
+    movzbl   2(rPC),%eax            # eax<- CC
+    movzbl   3(rPC),%ecx            # ecx<- BB
+    fldl    (rFP,%eax,4)           # vCC to fp stack
+    fdivl   (rFP,%ecx,4)           # ex: faddp
+    movzbl   rINST_HI,%ecx         # ecx<- AA
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    fstpl   (rFP,%ecx,4)           # %st to vAA
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_DOUBLE: /* 0xaf */
+/* File: x86/OP_REM_DOUBLE.S */
+    /* rem_float vAA, vBB, vCC */
+    movzbl   3(rPC),%ecx            # ecx<- BB
+    movzbl   2(rPC),%eax            # eax<- CC
+    fldl     (rFP,%ecx,4)           # vCC to fp stack
+    fldl     (rFP,%eax,4)           # vCC to fp stack
+    movzbl   rINST_HI,%ecx          # ecx<- AA
+    FETCH_INST_WORD(2)
+1:
+    fprem
+    fstsw     %ax
+    sahf
+    jp        1b
+    fstp      %st(1)
+    ADVANCE_PC(2)
+    fstpl    (rFP,%ecx,4)           # %st to vAA
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_INT_2ADDR: /* 0xb0 */
+/* File: x86/OP_ADD_INT_2ADDR.S */
+/* File: x86/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    movzx   rINST_HI,%ecx               # ecx<- A+
+    sarl    $12,rINST_FULL             # rINST_FULL<- B
+    GET_VREG(%eax,rINST_FULL)           # eax<- vB
+    FETCH_INST_WORD(1)
+    andb    $0xf,%cl                   # ecx<- A
+    addl     %eax,(rFP,%ecx,4)                              # for ex: addl   %eax,(rFP,%ecx,4)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_INT_2ADDR: /* 0xb1 */
+/* File: x86/OP_SUB_INT_2ADDR.S */
+/* File: x86/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    movzx   rINST_HI,%ecx               # ecx<- A+
+    sarl    $12,rINST_FULL             # rINST_FULL<- B
+    GET_VREG(%eax,rINST_FULL)           # eax<- vB
+    FETCH_INST_WORD(1)
+    andb    $0xf,%cl                   # ecx<- A
+    subl     %eax,(rFP,%ecx,4)                              # for ex: addl   %eax,(rFP,%ecx,4)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_INT_2ADDR: /* 0xb2 */
+/* File: x86/OP_MUL_INT_2ADDR.S */
+    /* mul vA, vB */
+    movzx   rINST_HI,%ecx               # ecx<- A+
+    sarl    $12,rINST_FULL             # rINST_FULL<- B
+    GET_VREG(%eax,rINST_FULL)           # eax<- vB
+    andb    $0xf,%cl                   # ecx<- A
+    SPILL(rPC)
+    imull   (rFP,%ecx,4),%eax
+    UNSPILL(rPC)
+    SET_VREG(%eax,%ecx)
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_INT_2ADDR: /* 0xb3 */
+/* File: x86/OP_DIV_INT_2ADDR.S */
+/* File: x86/bindiv2addr.S */
+    /*
+     * 32-bit binary div/rem operation.  Handles special case of op0=minint and
+     * op1=-1.
+     */
+    /* div/rem/2addr vA, vB */
+    movzx    rINST_HI,%ecx          # eax<- BA
+    sarl     $4,%ecx               # ecx<- B
+    GET_VREG(%ecx,%ecx)             # eax<- vBB
+    movzbl   rINST_HI,rINST_FULL    # rINST_FULL<- BA
+    andb     $0xf,rINST_LO         # rINST_FULL<- A
+    GET_VREG(%eax,rINST_FULL)       # eax<- vBB
+    SPILL(rPC)
+    cmpl     $0,%ecx
+    je       common_errDivideByZero
+    cmpl     $-1,%ecx
+    jne      .LOP_DIV_INT_2ADDR_continue_div2addr
+    cmpl     $0x80000000,%eax
+    jne      .LOP_DIV_INT_2ADDR_continue_div2addr
+    movl     $0x80000000,%eax
+    jmp      .LOP_DIV_INT_2ADDR_finish_div2addr
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_INT_2ADDR: /* 0xb4 */
+/* File: x86/OP_REM_INT_2ADDR.S */
+/* File: x86/bindiv2addr.S */
+    /*
+     * 32-bit binary div/rem operation.  Handles special case of op0=minint and
+     * op1=-1.
+     */
+    /* div/rem/2addr vA, vB */
+    movzx    rINST_HI,%ecx          # eax<- BA
+    sarl     $4,%ecx               # ecx<- B
+    GET_VREG(%ecx,%ecx)             # eax<- vBB
+    movzbl   rINST_HI,rINST_FULL    # rINST_FULL<- BA
+    andb     $0xf,rINST_LO         # rINST_FULL<- A
+    GET_VREG(%eax,rINST_FULL)       # eax<- vBB
+    SPILL(rPC)
+    cmpl     $0,%ecx
+    je       common_errDivideByZero
+    cmpl     $-1,%ecx
+    jne      .LOP_REM_INT_2ADDR_continue_div2addr
+    cmpl     $0x80000000,%eax
+    jne      .LOP_REM_INT_2ADDR_continue_div2addr
+    movl     $0,%edx
+    jmp      .LOP_REM_INT_2ADDR_finish_div2addr
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_INT_2ADDR: /* 0xb5 */
+/* File: x86/OP_AND_INT_2ADDR.S */
+/* File: x86/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    movzx   rINST_HI,%ecx               # ecx<- A+
+    sarl    $12,rINST_FULL             # rINST_FULL<- B
+    GET_VREG(%eax,rINST_FULL)           # eax<- vB
+    FETCH_INST_WORD(1)
+    andb    $0xf,%cl                   # ecx<- A
+    andl     %eax,(rFP,%ecx,4)                              # for ex: addl   %eax,(rFP,%ecx,4)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_INT_2ADDR: /* 0xb6 */
+/* File: x86/OP_OR_INT_2ADDR.S */
+/* File: x86/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    movzx   rINST_HI,%ecx               # ecx<- A+
+    sarl    $12,rINST_FULL             # rINST_FULL<- B
+    GET_VREG(%eax,rINST_FULL)           # eax<- vB
+    FETCH_INST_WORD(1)
+    andb    $0xf,%cl                   # ecx<- A
+    orl     %eax,(rFP,%ecx,4)                              # for ex: addl   %eax,(rFP,%ecx,4)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_INT_2ADDR: /* 0xb7 */
+/* File: x86/OP_XOR_INT_2ADDR.S */
+/* File: x86/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    movzx   rINST_HI,%ecx               # ecx<- A+
+    sarl    $12,rINST_FULL             # rINST_FULL<- B
+    GET_VREG(%eax,rINST_FULL)           # eax<- vB
+    FETCH_INST_WORD(1)
+    andb    $0xf,%cl                   # ecx<- A
+    xorl     %eax,(rFP,%ecx,4)                              # for ex: addl   %eax,(rFP,%ecx,4)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_INT_2ADDR: /* 0xb8 */
+/* File: x86/OP_SHL_INT_2ADDR.S */
+/* File: x86/shop2addr.S */
+    /*
+     * Generic 32-bit "shift/2addr" operation.
+     */
+    /* shift/2addr vA, vB */
+    movzx    rINST_HI,%ecx          # eax<- BA
+    sarl     $4,%ecx               # ecx<- B
+    GET_VREG(%ecx,%ecx)             # eax<- vBB
+    movzbl   rINST_HI,rINST_FULL    # rINST_FULL<- BA
+    andb     $0xf,rINST_LO         # rINST_FULL<- A
+    GET_VREG(%eax,rINST_FULL)       # eax<- vAA
+    sall    %cl,%eax                          # ex: sarl %cl,%eax
+    SET_VREG(%eax,rINST_FULL)
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_INT_2ADDR: /* 0xb9 */
+/* File: x86/OP_SHR_INT_2ADDR.S */
+/* File: x86/shop2addr.S */
+    /*
+     * Generic 32-bit "shift/2addr" operation.
+     */
+    /* shift/2addr vA, vB */
+    movzx    rINST_HI,%ecx          # eax<- BA
+    sarl     $4,%ecx               # ecx<- B
+    GET_VREG(%ecx,%ecx)             # eax<- vBB
+    movzbl   rINST_HI,rINST_FULL    # rINST_FULL<- BA
+    andb     $0xf,rINST_LO         # rINST_FULL<- A
+    GET_VREG(%eax,rINST_FULL)       # eax<- vAA
+    sarl    %cl,%eax                          # ex: sarl %cl,%eax
+    SET_VREG(%eax,rINST_FULL)
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_INT_2ADDR: /* 0xba */
+/* File: x86/OP_USHR_INT_2ADDR.S */
+/* File: x86/shop2addr.S */
+    /*
+     * Generic 32-bit "shift/2addr" operation.
+     */
+    /* shift/2addr vA, vB */
+    movzx    rINST_HI,%ecx          # eax<- BA
+    sarl     $4,%ecx               # ecx<- B
+    GET_VREG(%ecx,%ecx)             # eax<- vBB
+    movzbl   rINST_HI,rINST_FULL    # rINST_FULL<- BA
+    andb     $0xf,rINST_LO         # rINST_FULL<- A
+    GET_VREG(%eax,rINST_FULL)       # eax<- vAA
+    shrl    %cl,%eax                          # ex: sarl %cl,%eax
+    SET_VREG(%eax,rINST_FULL)
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_LONG_2ADDR: /* 0xbb */
+/* File: x86/OP_ADD_LONG_2ADDR.S */
+/* File: x86/binopWide2addr.S */
+    /*
+     * Generic 64-bit binary operation.
+     */
+    /* binop/2addr vA, vB */
+    movzbl    rINST_HI,%ecx             # ecx<- BA
+    sarl      $4,%ecx                  # ecx<- B
+    GET_VREG_WORD(%eax,%ecx,0)          # eax<- v[B+0]
+    GET_VREG_WORD(%ecx,%ecx,1)          # eax<- v[B+1]
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- BA
+    andb      $0xF,rINST_LO            # rINST_FULL<- A
+    addl %eax,(rFP,rINST_FULL,4)         # example: addl   %eax,(rFP,rINST_FULL,4)
+    adcl %ecx,4(rFP,rINST_FULL,4)         # example: adcl   %ecx,4(rFP,rINST_FULL,4)
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_LONG_2ADDR: /* 0xbc */
+/* File: x86/OP_SUB_LONG_2ADDR.S */
+/* File: x86/binopWide2addr.S */
+    /*
+     * Generic 64-bit binary operation.
+     */
+    /* binop/2addr vA, vB */
+    movzbl    rINST_HI,%ecx             # ecx<- BA
+    sarl      $4,%ecx                  # ecx<- B
+    GET_VREG_WORD(%eax,%ecx,0)          # eax<- v[B+0]
+    GET_VREG_WORD(%ecx,%ecx,1)          # eax<- v[B+1]
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- BA
+    andb      $0xF,rINST_LO            # rINST_FULL<- A
+    subl %eax,(rFP,rINST_FULL,4)         # example: addl   %eax,(rFP,rINST_FULL,4)
+    sbbl %ecx,4(rFP,rINST_FULL,4)         # example: adcl   %ecx,4(rFP,rINST_FULL,4)
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_LONG_2ADDR: /* 0xbd */
+/* File: x86/OP_MUL_LONG_2ADDR.S */
+    /*
+     * Signed 64-bit integer multiply, 2-addr version
+     *
+     * We could definately use more free registers for
+     * this code.  We must spill rPC (edx) because it
+     * is used by imul.  We'll also spill rINST (ebx),
+     * giving us eax, ebc, ecx and edx as computational
+     * temps.  On top of that, we'll spill rIBASE (edi)
+     * for use as the vA pointer and rFP (esi) for use
+     * as the vB pointer.  Yuck.
+     */
+    /* mul-long/2addr vA, vB */
+    movzbl    rINST_HI,%eax            # eax<- BA
+    andb      $0xf,%al                # eax<- A
+    sarl      $12,rINST_FULL          # rINST_FULL<- B
+    SPILL(rPC)
+    SPILL(rIBASE)
+    SPILL(rFP)
+    leal      (rFP,%eax,4),rIBASE      # rIBASE<- &v[A]
+    leal      (rFP,rINST_FULL,4),rFP   # rFP<- &v[B]
+    movl      4(rIBASE),%ecx      # ecx<- Amsw
+    imull     (rFP),%ecx          # ecx<- (Amsw*Blsw)
+    movl      4(rFP),%eax         # eax<- Bmsw
+    imull     (rIBASE),%eax       # eax<- (Bmsw*Alsw)
+    addl      %eax,%ecx           # ecx<- (Amsw*Blsw)+(Bmsw*Alsw)
+    movl      (rFP),%eax          # eax<- Blsw
+    mull      (rIBASE)            # eax<- (Blsw*Alsw)
+    jmp       .LOP_MUL_LONG_2ADDR_continue
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_LONG_2ADDR: /* 0xbe */
+/* File: x86/OP_DIV_LONG_2ADDR.S */
+    /* div/2addr vA, vB */
+    movzbl    rINST_HI,%eax
+    shrl      $4,%eax                  # eax<- B
+    movzbl    rINST_HI,rINST_FULL
+    andb      $0xf,rINST_LO            # rINST_FULL<- A
+    SPILL(rPC)
+    GET_VREG_WORD(rPC,%eax,0)
+    GET_VREG_WORD(%eax,%eax,1)
+    movl     rPC,OUT_ARG2(%esp)
+    testl    %eax,%eax
+    je       .LOP_DIV_LONG_2ADDR_check_zero
+    cmpl     $-1,%eax
+    je       .LOP_DIV_LONG_2ADDR_check_neg1
+.LOP_DIV_LONG_2ADDR_notSpecial:
+    GET_VREG_WORD(rPC,rINST_FULL,0)
+    GET_VREG_WORD(%ecx,rINST_FULL,1)
+.LOP_DIV_LONG_2ADDR_notSpecial1:
+    jmp      .LOP_DIV_LONG_2ADDR_continue
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_LONG_2ADDR: /* 0xbf */
+/* File: x86/OP_REM_LONG_2ADDR.S */
+/* File: x86/OP_DIV_LONG_2ADDR.S */
+    /* div/2addr vA, vB */
+    movzbl    rINST_HI,%eax
+    shrl      $4,%eax                  # eax<- B
+    movzbl    rINST_HI,rINST_FULL
+    andb      $0xf,rINST_LO            # rINST_FULL<- A
+    SPILL(rPC)
+    GET_VREG_WORD(rPC,%eax,0)
+    GET_VREG_WORD(%eax,%eax,1)
+    movl     rPC,OUT_ARG2(%esp)
+    testl    %eax,%eax
+    je       .LOP_REM_LONG_2ADDR_check_zero
+    cmpl     $-1,%eax
+    je       .LOP_REM_LONG_2ADDR_check_neg1
+.LOP_REM_LONG_2ADDR_notSpecial:
+    GET_VREG_WORD(rPC,rINST_FULL,0)
+    GET_VREG_WORD(%ecx,rINST_FULL,1)
+.LOP_REM_LONG_2ADDR_notSpecial1:
+    jmp      .LOP_REM_LONG_2ADDR_continue
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_LONG_2ADDR: /* 0xc0 */
+/* File: x86/OP_AND_LONG_2ADDR.S */
+/* File: x86/binopWide2addr.S */
+    /*
+     * Generic 64-bit binary operation.
+     */
+    /* binop/2addr vA, vB */
+    movzbl    rINST_HI,%ecx             # ecx<- BA
+    sarl      $4,%ecx                  # ecx<- B
+    GET_VREG_WORD(%eax,%ecx,0)          # eax<- v[B+0]
+    GET_VREG_WORD(%ecx,%ecx,1)          # eax<- v[B+1]
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- BA
+    andb      $0xF,rINST_LO            # rINST_FULL<- A
+    andl %eax,(rFP,rINST_FULL,4)         # example: addl   %eax,(rFP,rINST_FULL,4)
+    andl %ecx,4(rFP,rINST_FULL,4)         # example: adcl   %ecx,4(rFP,rINST_FULL,4)
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_LONG_2ADDR: /* 0xc1 */
+/* File: x86/OP_OR_LONG_2ADDR.S */
+/* File: x86/binopWide2addr.S */
+    /*
+     * Generic 64-bit binary operation.
+     */
+    /* binop/2addr vA, vB */
+    movzbl    rINST_HI,%ecx             # ecx<- BA
+    sarl      $4,%ecx                  # ecx<- B
+    GET_VREG_WORD(%eax,%ecx,0)          # eax<- v[B+0]
+    GET_VREG_WORD(%ecx,%ecx,1)          # eax<- v[B+1]
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- BA
+    andb      $0xF,rINST_LO            # rINST_FULL<- A
+    orl %eax,(rFP,rINST_FULL,4)         # example: addl   %eax,(rFP,rINST_FULL,4)
+    orl %ecx,4(rFP,rINST_FULL,4)         # example: adcl   %ecx,4(rFP,rINST_FULL,4)
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_LONG_2ADDR: /* 0xc2 */
+/* File: x86/OP_XOR_LONG_2ADDR.S */
+/* File: x86/binopWide2addr.S */
+    /*
+     * Generic 64-bit binary operation.
+     */
+    /* binop/2addr vA, vB */
+    movzbl    rINST_HI,%ecx             # ecx<- BA
+    sarl      $4,%ecx                  # ecx<- B
+    GET_VREG_WORD(%eax,%ecx,0)          # eax<- v[B+0]
+    GET_VREG_WORD(%ecx,%ecx,1)          # eax<- v[B+1]
+    movzbl    rINST_HI,rINST_FULL       # rINST_FULL<- BA
+    andb      $0xF,rINST_LO            # rINST_FULL<- A
+    xorl %eax,(rFP,rINST_FULL,4)         # example: addl   %eax,(rFP,rINST_FULL,4)
+    xorl %ecx,4(rFP,rINST_FULL,4)         # example: adcl   %ecx,4(rFP,rINST_FULL,4)
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_LONG_2ADDR: /* 0xc3 */
+/* File: x86/OP_SHL_LONG_2ADDR.S */
+    /*
+     * Long integer shift, 2addr version.  vA is 64-bit value/result, vB is
+     * 32-bit shift distance.
+     */
+    /* shl-long/2addr vA, vB */
+    /* ecx gets shift count */
+    /* Need to spill edx */
+    /* rINST gets AA */
+    movzbl    rINST_HI,%ecx             # ecx<- BA
+    movzbl    rINST_HI,rINST_FULL       # rINST_HI<- BA
+    andb      $0xf,rINST_LO            # rINST_FULL<- A
+    GET_VREG_WORD(%eax,rINST_FULL,0)    # eax<- v[AA+0]
+    sarl      $4,%ecx                  # ecx<- B
+    SPILL(rPC)
+    GET_VREG_WORD(%edx,rINST_FULL,1)    # edx<- v[AA+1]
+    GET_VREG(%ecx,%ecx)                 # ecx<- vBB
+    shldl     %eax,%edx
+    sall      %cl,%eax
+    testb     $32,%cl
+    je        2f
+    movl      %eax,%edx
+    xorl      %eax,%eax
+2:
+    SET_VREG_WORD(%edx,rINST_FULL,1)   # v[AA+1]<- edx
+    UNSPILL(rPC)
+    jmp       .LOP_SHL_LONG_2ADDR_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_LONG_2ADDR: /* 0xc4 */
+/* File: x86/OP_SHR_LONG_2ADDR.S */
+    /*
+     * Long integer shift, 2addr version.  vA is 64-bit value/result, vB is
+     * 32-bit shift distance.
+     */
+    /* shl-long/2addr vA, vB */
+    /* ecx gets shift count */
+    /* Need to spill edx */
+    /* rINST gets AA */
+    movzbl    rINST_HI,%ecx             # ecx<- BA
+    movzbl    rINST_HI,rINST_FULL       # rINST_HI<- BA
+    andb      $0xf,rINST_LO            # rINST_FULL<- A
+    GET_VREG_WORD(%eax,rINST_FULL,0)    # eax<- v[AA+0]
+    sarl      $4,%ecx                  # ecx<- B
+    SPILL(rPC)
+    GET_VREG_WORD(%edx,rINST_FULL,1)    # edx<- v[AA+1]
+    GET_VREG(%ecx,%ecx)                 # ecx<- vBB
+    shrdl     %edx,%eax
+    sarl      %cl,%edx
+    testb     $32,%cl
+    je        2f
+    movl      %edx,%eax
+    sarl      $31,%edx
+2:
+    SET_VREG_WORD(%edx,rINST_FULL,1)   # v[AA+1]<- edx
+    UNSPILL(rPC)
+    jmp       .LOP_SHR_LONG_2ADDR_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_LONG_2ADDR: /* 0xc5 */
+/* File: x86/OP_USHR_LONG_2ADDR.S */
+    /*
+     * Long integer shift, 2addr version.  vA is 64-bit value/result, vB is
+     * 32-bit shift distance.
+     */
+    /* shl-long/2addr vA, vB */
+    /* ecx gets shift count */
+    /* Need to spill edx */
+    /* rINST gets AA */
+    movzbl    rINST_HI,%ecx             # ecx<- BA
+    movzbl    rINST_HI,rINST_FULL       # rINST_HI<- BA
+    andb      $0xf,rINST_LO            # rINST_FULL<- A
+    GET_VREG_WORD(%eax,rINST_FULL,0)    # eax<- v[AA+0]
+    sarl      $4,%ecx                  # ecx<- B
+    SPILL(rPC)
+    GET_VREG_WORD(%edx,rINST_FULL,1)    # edx<- v[AA+1]
+    GET_VREG(%ecx,%ecx)                 # ecx<- vBB
+    shrdl     %edx,%eax
+    shrl      %cl,%edx
+    testb     $32,%cl
+    je        2f
+    movl      %edx,%eax
+    xorl      %edx,%edx
+2:
+    SET_VREG_WORD(%edx,rINST_FULL,1)   # v[AA+1]<- edx
+    UNSPILL(rPC)
+    jmp       .LOP_USHR_LONG_2ADDR_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_FLOAT_2ADDR: /* 0xc6 */
+/* File: x86/OP_ADD_FLOAT_2ADDR.S */
+/* File: x86/binflop2addr.S */
+    /*
+     * Generic 32-bit binary float operation.
+     *
+     * For: add-fp, sub-fp, mul-fp, div-fp
+     */
+
+    /* binop/2addr vA, vB */
+    movzx   rINST_HI,%ecx               # ecx<- A+
+    andb    $0xf,%cl                   # ecx<- A
+    flds    (rFP,%ecx,4)               # vAA to fp stack
+    sarl    $12,rINST_FULL             # rINST_FULL<- B
+    fadds   (rFP,rINST_FULL,4)           # ex: faddp
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    fstps    (rFP,%ecx,4)              # %st to vA
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_FLOAT_2ADDR: /* 0xc7 */
+/* File: x86/OP_SUB_FLOAT_2ADDR.S */
+/* File: x86/binflop2addr.S */
+    /*
+     * Generic 32-bit binary float operation.
+     *
+     * For: add-fp, sub-fp, mul-fp, div-fp
+     */
+
+    /* binop/2addr vA, vB */
+    movzx   rINST_HI,%ecx               # ecx<- A+
+    andb    $0xf,%cl                   # ecx<- A
+    flds    (rFP,%ecx,4)               # vAA to fp stack
+    sarl    $12,rINST_FULL             # rINST_FULL<- B
+    fsubs   (rFP,rINST_FULL,4)           # ex: faddp
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    fstps    (rFP,%ecx,4)              # %st to vA
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_FLOAT_2ADDR: /* 0xc8 */
+/* File: x86/OP_MUL_FLOAT_2ADDR.S */
+/* File: x86/binflop2addr.S */
+    /*
+     * Generic 32-bit binary float operation.
+     *
+     * For: add-fp, sub-fp, mul-fp, div-fp
+     */
+
+    /* binop/2addr vA, vB */
+    movzx   rINST_HI,%ecx               # ecx<- A+
+    andb    $0xf,%cl                   # ecx<- A
+    flds    (rFP,%ecx,4)               # vAA to fp stack
+    sarl    $12,rINST_FULL             # rINST_FULL<- B
+    fmuls   (rFP,rINST_FULL,4)           # ex: faddp
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    fstps    (rFP,%ecx,4)              # %st to vA
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_FLOAT_2ADDR: /* 0xc9 */
+/* File: x86/OP_DIV_FLOAT_2ADDR.S */
+/* File: x86/binflop2addr.S */
+    /*
+     * Generic 32-bit binary float operation.
+     *
+     * For: add-fp, sub-fp, mul-fp, div-fp
+     */
+
+    /* binop/2addr vA, vB */
+    movzx   rINST_HI,%ecx               # ecx<- A+
+    andb    $0xf,%cl                   # ecx<- A
+    flds    (rFP,%ecx,4)               # vAA to fp stack
+    sarl    $12,rINST_FULL             # rINST_FULL<- B
+    fdivs   (rFP,rINST_FULL,4)           # ex: faddp
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    fstps    (rFP,%ecx,4)              # %st to vA
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_FLOAT_2ADDR: /* 0xca */
+/* File: x86/OP_REM_FLOAT_2ADDR.S */
+    /* rem_float/2addr vA, vB */
+    movzx   rINST_HI,%ecx               # ecx<- A+
+    sarl    $12,rINST_FULL             # rINST_FULL<- B
+    flds     (rFP,rINST_FULL,4)         # vBB to fp stack
+    andb    $0xf,%cl                   # ecx<- A
+    flds     (rFP,%ecx,4)               # vAA to fp stack
+    FETCH_INST_WORD(1)
+1:
+    fprem
+    fstsw     %ax
+    sahf
+    jp        1b
+    fstp      %st(1)
+    ADVANCE_PC(1)
+    fstps    (rFP,%ecx,4)               # %st to vA
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_DOUBLE_2ADDR: /* 0xcb */
+/* File: x86/OP_ADD_DOUBLE_2ADDR.S */
+/* File: x86/binflop2addr.S */
+    /*
+     * Generic 32-bit binary float operation.
+     *
+     * For: add-fp, sub-fp, mul-fp, div-fp
+     */
+
+    /* binop/2addr vA, vB */
+    movzx   rINST_HI,%ecx               # ecx<- A+
+    andb    $0xf,%cl                   # ecx<- A
+    fldl    (rFP,%ecx,4)               # vAA to fp stack
+    sarl    $12,rINST_FULL             # rINST_FULL<- B
+    faddl   (rFP,rINST_FULL,4)           # ex: faddp
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    fstpl    (rFP,%ecx,4)              # %st to vA
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_DOUBLE_2ADDR: /* 0xcc */
+/* File: x86/OP_SUB_DOUBLE_2ADDR.S */
+/* File: x86/binflop2addr.S */
+    /*
+     * Generic 32-bit binary float operation.
+     *
+     * For: add-fp, sub-fp, mul-fp, div-fp
+     */
+
+    /* binop/2addr vA, vB */
+    movzx   rINST_HI,%ecx               # ecx<- A+
+    andb    $0xf,%cl                   # ecx<- A
+    fldl    (rFP,%ecx,4)               # vAA to fp stack
+    sarl    $12,rINST_FULL             # rINST_FULL<- B
+    fsubl   (rFP,rINST_FULL,4)           # ex: faddp
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    fstpl    (rFP,%ecx,4)              # %st to vA
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_DOUBLE_2ADDR: /* 0xcd */
+/* File: x86/OP_MUL_DOUBLE_2ADDR.S */
+/* File: x86/binflop2addr.S */
+    /*
+     * Generic 32-bit binary float operation.
+     *
+     * For: add-fp, sub-fp, mul-fp, div-fp
+     */
+
+    /* binop/2addr vA, vB */
+    movzx   rINST_HI,%ecx               # ecx<- A+
+    andb    $0xf,%cl                   # ecx<- A
+    fldl    (rFP,%ecx,4)               # vAA to fp stack
+    sarl    $12,rINST_FULL             # rINST_FULL<- B
+    fmull   (rFP,rINST_FULL,4)           # ex: faddp
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    fstpl    (rFP,%ecx,4)              # %st to vA
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_DOUBLE_2ADDR: /* 0xce */
+/* File: x86/OP_DIV_DOUBLE_2ADDR.S */
+/* File: x86/binflop2addr.S */
+    /*
+     * Generic 32-bit binary float operation.
+     *
+     * For: add-fp, sub-fp, mul-fp, div-fp
+     */
+
+    /* binop/2addr vA, vB */
+    movzx   rINST_HI,%ecx               # ecx<- A+
+    andb    $0xf,%cl                   # ecx<- A
+    fldl    (rFP,%ecx,4)               # vAA to fp stack
+    sarl    $12,rINST_FULL             # rINST_FULL<- B
+    fdivl   (rFP,rINST_FULL,4)           # ex: faddp
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    fstpl    (rFP,%ecx,4)              # %st to vA
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_DOUBLE_2ADDR: /* 0xcf */
+/* File: x86/OP_REM_DOUBLE_2ADDR.S */
+    /* rem_float/2addr vA, vB */
+    movzx   rINST_HI,%ecx               # ecx<- A+
+    sarl    $12,rINST_FULL             # rINST_FULL<- B
+    fldl     (rFP,rINST_FULL,4)         # vBB to fp stack
+    andb    $0xf,%cl                   # ecx<- A
+    fldl     (rFP,%ecx,4)               # vAA to fp stack
+    FETCH_INST_WORD(1)
+1:
+    fprem
+    fstsw     %ax
+    sahf
+    jp        1b
+    fstp      %st(1)
+    ADVANCE_PC(1)
+    fstpl    (rFP,%ecx,4)               # %st to vA
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_INT_LIT16: /* 0xd0 */
+/* File: x86/OP_ADD_INT_LIT16.S */
+/* File: x86/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = eax op ecx".
+     * This could be an x86 instruction or a function call.  (If the result
+     * comes back in a register other than eax, you can override "result".)
+     *
+     * For: add-int/lit16, rsub-int,
+     *      and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    movzbl   rINST_HI,%eax              # eax<- 000000BA
+    sarl     $4,%eax                   # eax<- B
+    GET_VREG(%eax,%eax)                 # eax<- vB
+    movswl   2(rPC),%ecx                # ecx<- ssssCCCC
+    movzbl   rINST_HI,rINST_FULL        # rINST_FULL<- BA
+    andb     $0xf,rINST_LO             # rINST_FULL<- A
+    addl %ecx,%eax                              # for example: addl %ecx, %eax
+    SET_VREG(%eax,rINST_FULL)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RSUB_INT: /* 0xd1 */
+/* File: x86/OP_RSUB_INT.S */
+/* File: x86/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = eax op ecx".
+     * This could be an x86 instruction or a function call.  (If the result
+     * comes back in a register other than eax, you can override "result".)
+     *
+     * For: add-int/lit16, rsub-int,
+     *      and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    movzbl   rINST_HI,%eax              # eax<- 000000BA
+    sarl     $4,%eax                   # eax<- B
+    GET_VREG(%eax,%eax)                 # eax<- vB
+    movswl   2(rPC),%ecx                # ecx<- ssssCCCC
+    movzbl   rINST_HI,rINST_FULL        # rINST_FULL<- BA
+    andb     $0xf,rINST_LO             # rINST_FULL<- A
+    subl %eax,%ecx                              # for example: addl %ecx, %eax
+    SET_VREG(%ecx,rINST_FULL)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_INT_LIT16: /* 0xd2 */
+/* File: x86/OP_MUL_INT_LIT16.S */
+    /* mul/lit16 vA, vB, #+CCCC */
+    /* Need A in rINST_FULL, ssssCCCC in ecx, vB in eax */
+    movzbl   rINST_HI,%eax              # eax<- 000000BA
+    sarl     $4,%eax                   # eax<- B
+    GET_VREG(%eax,%eax)                 # eax<- vB
+    movswl   2(rPC),%ecx                # ecx<- ssssCCCC
+    SPILL(rPC)
+    movzbl   rINST_HI,rINST_FULL        # rINST_FULL<- BA
+    andb     $0xf,rINST_LO             # rINST_FULL<- A
+    imull     %ecx,%eax                 # trashes rPC
+    UNSPILL(rPC)
+    SET_VREG(%eax,rINST_FULL)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_INT_LIT16: /* 0xd3 */
+/* File: x86/OP_DIV_INT_LIT16.S */
+/* File: x86/bindivLit16.S */
+    /*
+     * 32-bit binary div/rem operation.  Handles special case of op0=minint and
+     * op1=-1.
+     */
+    /* div/rem/lit16 vA, vB, #+CCCC */
+    /* Need A in rINST_FULL, ssssCCCC in ecx, vB in eax */
+    movzbl   rINST_HI,%eax              # eax<- 000000BA
+    sarl     $4,%eax                   # eax<- B
+    GET_VREG(%eax,%eax)                 # eax<- vB
+    movswl   2(rPC),%ecx                # ecx<- ssssCCCC
+    movzbl   rINST_HI,rINST_FULL        # rINST_FULL<- BA
+    andb     $0xf,rINST_LO             # rINST_FULL<- A
+    SPILL(rPC)
+    cmpl     $0,%ecx
+    je       common_errDivideByZero
+    cmpl     $-1,%ecx
+    jne      .LOP_DIV_INT_LIT16_continue_div
+    cmpl     $0x80000000,%eax
+    jne      .LOP_DIV_INT_LIT16_continue_div
+    movl     $0x80000000,%eax
+    jmp      .LOP_DIV_INT_LIT16_finish_div
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_INT_LIT16: /* 0xd4 */
+/* File: x86/OP_REM_INT_LIT16.S */
+/* File: x86/bindivLit16.S */
+    /*
+     * 32-bit binary div/rem operation.  Handles special case of op0=minint and
+     * op1=-1.
+     */
+    /* div/rem/lit16 vA, vB, #+CCCC */
+    /* Need A in rINST_FULL, ssssCCCC in ecx, vB in eax */
+    movzbl   rINST_HI,%eax              # eax<- 000000BA
+    sarl     $4,%eax                   # eax<- B
+    GET_VREG(%eax,%eax)                 # eax<- vB
+    movswl   2(rPC),%ecx                # ecx<- ssssCCCC
+    movzbl   rINST_HI,rINST_FULL        # rINST_FULL<- BA
+    andb     $0xf,rINST_LO             # rINST_FULL<- A
+    SPILL(rPC)
+    cmpl     $0,%ecx
+    je       common_errDivideByZero
+    cmpl     $-1,%ecx
+    jne      .LOP_REM_INT_LIT16_continue_div
+    cmpl     $0x80000000,%eax
+    jne      .LOP_REM_INT_LIT16_continue_div
+    movl     $0,%edx
+    jmp      .LOP_REM_INT_LIT16_finish_div
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_INT_LIT16: /* 0xd5 */
+/* File: x86/OP_AND_INT_LIT16.S */
+/* File: x86/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = eax op ecx".
+     * This could be an x86 instruction or a function call.  (If the result
+     * comes back in a register other than eax, you can override "result".)
+     *
+     * For: add-int/lit16, rsub-int,
+     *      and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    movzbl   rINST_HI,%eax              # eax<- 000000BA
+    sarl     $4,%eax                   # eax<- B
+    GET_VREG(%eax,%eax)                 # eax<- vB
+    movswl   2(rPC),%ecx                # ecx<- ssssCCCC
+    movzbl   rINST_HI,rINST_FULL        # rINST_FULL<- BA
+    andb     $0xf,rINST_LO             # rINST_FULL<- A
+    andl %ecx,%eax                              # for example: addl %ecx, %eax
+    SET_VREG(%eax,rINST_FULL)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_INT_LIT16: /* 0xd6 */
+/* File: x86/OP_OR_INT_LIT16.S */
+/* File: x86/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = eax op ecx".
+     * This could be an x86 instruction or a function call.  (If the result
+     * comes back in a register other than eax, you can override "result".)
+     *
+     * For: add-int/lit16, rsub-int,
+     *      and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    movzbl   rINST_HI,%eax              # eax<- 000000BA
+    sarl     $4,%eax                   # eax<- B
+    GET_VREG(%eax,%eax)                 # eax<- vB
+    movswl   2(rPC),%ecx                # ecx<- ssssCCCC
+    movzbl   rINST_HI,rINST_FULL        # rINST_FULL<- BA
+    andb     $0xf,rINST_LO             # rINST_FULL<- A
+    orl     %ecx,%eax                              # for example: addl %ecx, %eax
+    SET_VREG(%eax,rINST_FULL)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_INT_LIT16: /* 0xd7 */
+/* File: x86/OP_XOR_INT_LIT16.S */
+/* File: x86/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = eax op ecx".
+     * This could be an x86 instruction or a function call.  (If the result
+     * comes back in a register other than eax, you can override "result".)
+     *
+     * For: add-int/lit16, rsub-int,
+     *      and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    movzbl   rINST_HI,%eax              # eax<- 000000BA
+    sarl     $4,%eax                   # eax<- B
+    GET_VREG(%eax,%eax)                 # eax<- vB
+    movswl   2(rPC),%ecx                # ecx<- ssssCCCC
+    movzbl   rINST_HI,rINST_FULL        # rINST_FULL<- BA
+    andb     $0xf,rINST_LO             # rINST_FULL<- A
+    xor    %ecx,%eax                              # for example: addl %ecx, %eax
+    SET_VREG(%eax,rINST_FULL)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_INT_LIT8: /* 0xd8 */
+/* File: x86/OP_ADD_INT_LIT8.S */
+/* File: x86/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = eax op ecx".
+     * This could be an x86 instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * For: add-int/lit8, rsub-int/lit8
+     *      and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    movzbl    2(rPC),%eax              # eax<- BB
+    movsbl    3(rPC),%ecx              # ecx<- ssssssCC
+    movzx     rINST_HI,rINST_FULL      # rINST_FULL<- AA
+    GET_VREG  (%eax,%eax)              # eax<- rBB
+    addl %ecx,%eax                             # ex: addl %ecx,%eax
+    SET_VREG  (%eax,rINST_FULL)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RSUB_INT_LIT8: /* 0xd9 */
+/* File: x86/OP_RSUB_INT_LIT8.S */
+/* File: x86/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = eax op ecx".
+     * This could be an x86 instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * For: add-int/lit8, rsub-int/lit8
+     *      and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    movzbl    2(rPC),%eax              # eax<- BB
+    movsbl    3(rPC),%ecx              # ecx<- ssssssCC
+    movzx     rINST_HI,rINST_FULL      # rINST_FULL<- AA
+    GET_VREG  (%eax,%eax)              # eax<- rBB
+    subl  %eax,%ecx                             # ex: addl %ecx,%eax
+    SET_VREG  (%ecx,rINST_FULL)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_INT_LIT8: /* 0xda */
+/* File: x86/OP_MUL_INT_LIT8.S */
+    /* mul/lit8 vAA, vBB, #+CC */
+    movzbl    2(rPC),%eax              # eax<- BB
+    movsbl    3(rPC),%ecx              # ecx<- ssssssCC
+    SPILL(rPC)
+    movzx     rINST_HI,rINST_FULL      # rINST_FULL<- AA
+    GET_VREG  (%eax,%eax)              # eax<- rBB
+    imull     %ecx,%eax                # trashes rPC
+    UNSPILL(rPC)
+    SET_VREG  (%eax,rINST_FULL)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_INT_LIT8: /* 0xdb */
+/* File: x86/OP_DIV_INT_LIT8.S */
+/* File: x86/bindivLit8.S */
+    /*
+     * 32-bit div/rem "lit8" binary operation.  Handles special case of
+     * op0=minint & op1=-1
+     */
+    /* div/rem/lit8 vAA, vBB, #+CC */
+    movzbl    2(rPC),%eax              # eax<- BB
+    movsbl    3(rPC),%ecx              # ecx<- ssssssCC
+    GET_VREG  (%eax,%eax)              # eax<- rBB
+    movzx     rINST_HI,rINST_FULL      # rINST_FULL<- AA
+    SPILL(rPC)
+    cmpl     $0,%ecx
+    je       common_errDivideByZero
+    cmpl     $0x80000000,%eax
+    jne      .LOP_DIV_INT_LIT8_continue_div
+    cmpl     $-1,%ecx
+    jne      .LOP_DIV_INT_LIT8_continue_div
+    movl     $0x80000000,%eax
+    jmp      .LOP_DIV_INT_LIT8_finish_div
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_INT_LIT8: /* 0xdc */
+/* File: x86/OP_REM_INT_LIT8.S */
+/* File: x86/bindivLit8.S */
+    /*
+     * 32-bit div/rem "lit8" binary operation.  Handles special case of
+     * op0=minint & op1=-1
+     */
+    /* div/rem/lit8 vAA, vBB, #+CC */
+    movzbl    2(rPC),%eax              # eax<- BB
+    movsbl    3(rPC),%ecx              # ecx<- ssssssCC
+    GET_VREG  (%eax,%eax)              # eax<- rBB
+    movzx     rINST_HI,rINST_FULL      # rINST_FULL<- AA
+    SPILL(rPC)
+    cmpl     $0,%ecx
+    je       common_errDivideByZero
+    cmpl     $0x80000000,%eax
+    jne      .LOP_REM_INT_LIT8_continue_div
+    cmpl     $-1,%ecx
+    jne      .LOP_REM_INT_LIT8_continue_div
+    movl     $0,%edx
+    jmp      .LOP_REM_INT_LIT8_finish_div
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_INT_LIT8: /* 0xdd */
+/* File: x86/OP_AND_INT_LIT8.S */
+/* File: x86/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = eax op ecx".
+     * This could be an x86 instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * For: add-int/lit8, rsub-int/lit8
+     *      and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    movzbl    2(rPC),%eax              # eax<- BB
+    movsbl    3(rPC),%ecx              # ecx<- ssssssCC
+    movzx     rINST_HI,rINST_FULL      # rINST_FULL<- AA
+    GET_VREG  (%eax,%eax)              # eax<- rBB
+    andl %ecx,%eax                             # ex: addl %ecx,%eax
+    SET_VREG  (%eax,rINST_FULL)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_INT_LIT8: /* 0xde */
+/* File: x86/OP_OR_INT_LIT8.S */
+/* File: x86/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = eax op ecx".
+     * This could be an x86 instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * For: add-int/lit8, rsub-int/lit8
+     *      and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    movzbl    2(rPC),%eax              # eax<- BB
+    movsbl    3(rPC),%ecx              # ecx<- ssssssCC
+    movzx     rINST_HI,rINST_FULL      # rINST_FULL<- AA
+    GET_VREG  (%eax,%eax)              # eax<- rBB
+    orl     %ecx,%eax                             # ex: addl %ecx,%eax
+    SET_VREG  (%eax,rINST_FULL)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_INT_LIT8: /* 0xdf */
+/* File: x86/OP_XOR_INT_LIT8.S */
+/* File: x86/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = eax op ecx".
+     * This could be an x86 instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * For: add-int/lit8, rsub-int/lit8
+     *      and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    movzbl    2(rPC),%eax              # eax<- BB
+    movsbl    3(rPC),%ecx              # ecx<- ssssssCC
+    movzx     rINST_HI,rINST_FULL      # rINST_FULL<- AA
+    GET_VREG  (%eax,%eax)              # eax<- rBB
+    xor    %ecx,%eax                             # ex: addl %ecx,%eax
+    SET_VREG  (%eax,rINST_FULL)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_INT_LIT8: /* 0xe0 */
+/* File: x86/OP_SHL_INT_LIT8.S */
+/* File: x86/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = eax op ecx".
+     * This could be an x86 instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * For: add-int/lit8, rsub-int/lit8
+     *      and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    movzbl    2(rPC),%eax              # eax<- BB
+    movsbl    3(rPC),%ecx              # ecx<- ssssssCC
+    movzx     rINST_HI,rINST_FULL      # rINST_FULL<- AA
+    GET_VREG  (%eax,%eax)              # eax<- rBB
+    sall  %cl,%eax                             # ex: addl %ecx,%eax
+    SET_VREG  (%eax,rINST_FULL)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_INT_LIT8: /* 0xe1 */
+/* File: x86/OP_SHR_INT_LIT8.S */
+/* File: x86/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = eax op ecx".
+     * This could be an x86 instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * For: add-int/lit8, rsub-int/lit8
+     *      and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    movzbl    2(rPC),%eax              # eax<- BB
+    movsbl    3(rPC),%ecx              # ecx<- ssssssCC
+    movzx     rINST_HI,rINST_FULL      # rINST_FULL<- AA
+    GET_VREG  (%eax,%eax)              # eax<- rBB
+    sarl    %cl,%eax                             # ex: addl %ecx,%eax
+    SET_VREG  (%eax,rINST_FULL)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_INT_LIT8: /* 0xe2 */
+/* File: x86/OP_USHR_INT_LIT8.S */
+/* File: x86/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = eax op ecx".
+     * This could be an x86 instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * For: add-int/lit8, rsub-int/lit8
+     *      and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    movzbl    2(rPC),%eax              # eax<- BB
+    movsbl    3(rPC),%ecx              # ecx<- ssssssCC
+    movzx     rINST_HI,rINST_FULL      # rINST_FULL<- AA
+    GET_VREG  (%eax,%eax)              # eax<- rBB
+    shrl     %cl,%eax                             # ex: addl %ecx,%eax
+    SET_VREG  (%eax,rINST_FULL)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E3: /* 0xe3 */
+/* File: x86/OP_UNUSED_E3.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E4: /* 0xe4 */
+/* File: x86/OP_UNUSED_E4.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E5: /* 0xe5 */
+/* File: x86/OP_UNUSED_E5.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E6: /* 0xe6 */
+/* File: x86/OP_UNUSED_E6.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E7: /* 0xe7 */
+/* File: x86/OP_UNUSED_E7.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E8: /* 0xe8 */
+/* File: x86/OP_UNUSED_E8.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E9: /* 0xe9 */
+/* File: x86/OP_UNUSED_E9.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EA: /* 0xea */
+/* File: x86/OP_UNUSED_EA.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EB: /* 0xeb */
+/* File: x86/OP_UNUSED_EB.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EC: /* 0xec */
+/* File: x86/OP_UNUSED_EC.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_ED: /* 0xed */
+/* File: x86/OP_UNUSED_ED.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_EXECUTE_INLINE: /* 0xee */
+/* File: x86/OP_EXECUTE_INLINE.S */
+    /*
+     * Execute a "native inline" instruction.
+     *
+     * We will be calling through a function table:
+     *
+     * (*gDvmInlineOpsTable[opIndex].func)(arg0, arg1, arg2, arg3, pResult)
+     *
+     */
+    /* [opt] execute-inline vAA, {vC, vD, vE, vF}, inline@BBBB */
+    GET_GLUE(%ecx)
+    EXPORT_PC()
+    movzwl    2(rPC),%eax               # eax<- BBBB
+    leal      offGlue_retval(%ecx),%ecx # ecx<- & glue->retval
+    movl      %ecx,OUT_ARG4(%esp)
+    sarl      $12,rINST_FULL           # rINST_FULL<- arg count (0-4)
+    SPILL(rPC)
+    call      .LOP_EXECUTE_INLINE_continue      # make call; will return after
+    UNSPILL(rPC)
+    testl     %eax,%eax                 # successful?
+    FETCH_INST_WORD(3)
+    je        common_exceptionThrown    # no, handle exception
+    ADVANCE_PC(3)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EF: /* 0xef */
+/* File: x86/OP_UNUSED_EF.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_DIRECT_EMPTY: /* 0xf0 */
+/* File: x86/OP_INVOKE_DIRECT_EMPTY.S */
+    /*
+     * invoke-direct-empty is a no-op in a "standard" interpreter.
+     */
+    FETCH_INST_WORD(3)
+    ADVANCE_PC(3)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_F1: /* 0xf1 */
+/* File: x86/OP_UNUSED_F1.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_QUICK: /* 0xf2 */
+/* File: x86/OP_IGET_QUICK.S */
+    /* For: iget-quick, iget-object-quick */
+    /* op vA, vB, offset@CCCC */
+    movzbl    rINST_HI,%ecx             # ecx<- BA
+    sarl      $4,%ecx                  # ecx<- B
+    GET_VREG(%ecx,%ecx)                 # vB (object we're operating on)
+    movzwl    2(rPC),%eax               # eax<- field byte offset
+    cmpl      $0,%ecx                  # is object null?
+    je        common_errNullObject
+    movl      (%ecx,%eax,1),%eax
+    movzbl    rINST_HI,%ecx
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    andb      $0xf,%cl                 # rINST_FULL<- A
+    SET_VREG  (%eax,%ecx)               # fp[A]<- result
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_WIDE_QUICK: /* 0xf3 */
+/* File: x86/OP_IGET_WIDE_QUICK.S */
+    /* For: iget-wide-quick */
+    /* op vA, vB, offset@CCCC */
+    movzbl    rINST_HI,%ecx             # ecx<- BA
+    sarl      $4,%ecx                  # ecx<- B
+    GET_VREG(%ecx,%ecx)                 # vB (object we're operating on)
+    movzwl    2(rPC),%eax               # eax<- field byte offset
+    cmpl      $0,%ecx                  # is object null?
+    je        common_errNullObject
+    leal      (%ecx,%eax,1),%eax        # eax<- address of 64-bit source
+    movl      (%eax),%ecx               # ecx<- lsw
+    movl      4(%eax),%eax               # eax<- msw
+    movzbl    rINST_HI,rINST_FULL
+    andb      $0xf,rINST_LO            # rINST_FULL<- A
+    SET_VREG_WORD(%ecx,rINST_FULL,0)    # v[A+0]<- lsw
+    SET_VREG_WORD(%eax,rINST_FULL,1)    # v[A+1]<- msw
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_OBJECT_QUICK: /* 0xf4 */
+/* File: x86/OP_IGET_OBJECT_QUICK.S */
+/* File: x86/OP_IGET_QUICK.S */
+    /* For: iget-quick, iget-object-quick */
+    /* op vA, vB, offset@CCCC */
+    movzbl    rINST_HI,%ecx             # ecx<- BA
+    sarl      $4,%ecx                  # ecx<- B
+    GET_VREG(%ecx,%ecx)                 # vB (object we're operating on)
+    movzwl    2(rPC),%eax               # eax<- field byte offset
+    cmpl      $0,%ecx                  # is object null?
+    je        common_errNullObject
+    movl      (%ecx,%eax,1),%eax
+    movzbl    rINST_HI,%ecx
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    andb      $0xf,%cl                 # rINST_FULL<- A
+    SET_VREG  (%eax,%ecx)               # fp[A]<- result
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_QUICK: /* 0xf5 */
+/* File: x86/OP_IPUT_QUICK.S */
+    /* For: iput-quick, iput-object-quick */
+    /* op vA, vB, offset@CCCC */
+    movzbl    rINST_HI,%ecx             # ecx<- BA
+    sarl      $4,%ecx                  # ecx<- B
+    GET_VREG(%ecx,%ecx)                 # vB (object we're operating on)
+    movzbl    rINST_HI,rINST_FULL
+    andb      $0xf,rINST_LO            # rINST_FULL<- A
+    GET_VREG(rINST_FULL,rINST_FULL)     # rINST_FULL<- v[A]
+    movzwl    2(rPC),%eax               # eax<- field byte offset
+    testl     %ecx,%ecx                  # is object null?
+    je        common_errNullObject
+    movl      rINST_FULL,(%ecx,%eax,1)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_WIDE_QUICK: /* 0xf6 */
+/* File: x86/OP_IPUT_WIDE_QUICK.S */
+    /* For: iput-wide-quick */
+    /* op vA, vB, offset@CCCC */
+    movzbl    rINST_HI,%ecx             # ecx<- BA
+    sarl      $4,%ecx                  # ecx<- B
+    GET_VREG(%ecx,%ecx)                 # vB (object we're operating on)
+    movzwl    2(rPC),%eax               # eax<- field byte offset
+    testl      %ecx,%ecx                # is object null?
+    je        common_errNullObject
+    leal      (%ecx,%eax,1),%ecx        # ecx<- Address of 64-bit target
+    movzbl    rINST_HI,rINST_FULL
+    andb      $0xf,rINST_LO            # rINST_FULL<- A
+    GET_VREG_WORD(%eax,rINST_FULL,0)    # eax<- lsw
+    GET_VREG_WORD(rINST_FULL,rINST_FULL,1) # rINST_FULL<- msw
+    movl      %eax,(%ecx)
+    movl      rINST_FULL,4(%ecx)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_OBJECT_QUICK: /* 0xf7 */
+/* File: x86/OP_IPUT_OBJECT_QUICK.S */
+/* File: x86/OP_IPUT_QUICK.S */
+    /* For: iput-quick, iput-object-quick */
+    /* op vA, vB, offset@CCCC */
+    movzbl    rINST_HI,%ecx             # ecx<- BA
+    sarl      $4,%ecx                  # ecx<- B
+    GET_VREG(%ecx,%ecx)                 # vB (object we're operating on)
+    movzbl    rINST_HI,rINST_FULL
+    andb      $0xf,rINST_LO            # rINST_FULL<- A
+    GET_VREG(rINST_FULL,rINST_FULL)     # rINST_FULL<- v[A]
+    movzwl    2(rPC),%eax               # eax<- field byte offset
+    testl     %ecx,%ecx                  # is object null?
+    je        common_errNullObject
+    movl      rINST_FULL,(%ecx,%eax,1)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_VIRTUAL_QUICK: /* 0xf8 */
+/* File: x86/OP_INVOKE_VIRTUAL_QUICK.S */
+    /*
+     * Handle an optimized virtual method call.
+     *
+     * for: [opt] invoke-virtual-quick, invoke-virtual-quick/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    movzwl    4(rPC),%eax               # eax<- FEDC or CCCC
+    movzwl    2(rPC),%ecx               # ecx<- BBBB
+    .if     (!0)
+    andl      $0xf,%eax                # eax<- C (or stays CCCC)
+    .endif
+    GET_VREG(%eax,%eax)                 # eax<- vC ("this" ptr)
+    testl     %eax,%eax                 # null?
+    je        common_errNullObject      # yep, throw exception
+    movl      offObject_clazz(%eax),%eax # eax<- thisPtr->clazz
+    movl      offClassObject_vtable(%eax),%eax # eax<- thisPtr->clazz->vtable
+    EXPORT_PC()                         # might throw later - get ready
+    movl      (%eax,%ecx,4),%eax        # eax<- vtable[BBBB]
+    movl      $0,%ecx           # pass range flag
+    #jmp       common_invokeMethodNoRange
+    jmp       common_invokeOld
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_VIRTUAL_QUICK_RANGE: /* 0xf9 */
+/* File: x86/OP_INVOKE_VIRTUAL_QUICK_RANGE.S */
+/* File: x86/OP_INVOKE_VIRTUAL_QUICK.S */
+    /*
+     * Handle an optimized virtual method call.
+     *
+     * for: [opt] invoke-virtual-quick, invoke-virtual-quick/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    movzwl    4(rPC),%eax               # eax<- FEDC or CCCC
+    movzwl    2(rPC),%ecx               # ecx<- BBBB
+    .if     (!1)
+    andl      $0xf,%eax                # eax<- C (or stays CCCC)
+    .endif
+    GET_VREG(%eax,%eax)                 # eax<- vC ("this" ptr)
+    testl     %eax,%eax                 # null?
+    je        common_errNullObject      # yep, throw exception
+    movl      offObject_clazz(%eax),%eax # eax<- thisPtr->clazz
+    movl      offClassObject_vtable(%eax),%eax # eax<- thisPtr->clazz->vtable
+    EXPORT_PC()                         # might throw later - get ready
+    movl      (%eax,%ecx,4),%eax        # eax<- vtable[BBBB]
+    movl      $1,%ecx           # pass range flag
+    #jmp       common_invokeMethodRange
+    jmp       common_invokeOld
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_SUPER_QUICK: /* 0xfa */
+/* File: x86/OP_INVOKE_SUPER_QUICK.S */
+    /*
+     * Handle an optimized "super" method call.
+     *
+     * for: [opt] invoke-super-quick, invoke-super-quick/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    4(rPC),%eax               # eax<- GFED or CCCC
+    movl      offGlue_method(%ecx),%ecx # ecx<- current method
+    .if       (!0)
+    andl      $0xf,%eax                # eax<- D (or stays CCCC)
+    .endif
+    movl      offMethod_clazz(%ecx),%ecx # ecx<- method->clazz
+    GET_VREG(%eax,%eax)                 # eax<- "this"
+    movl      offClassObject_super(%ecx),%ecx # ecx<- method->clazz->super
+    testl     %eax,%eax                 # null "this"?
+    je        common_errNullObject      # "this" is null, throw exception
+    movzwl    2(rPC),%eax               # eax<- BBBB
+    movl      offClassObject_vtable(%ecx),%ecx # ecx<- vtable
+    EXPORT_PC()
+    movl      (%ecx,%eax,4),%eax        # eax<- super->vtable[BBBB]
+    movl      $0,%ecx           # ecx<- range flag
+    #jmp       common_invokeMethodNoRange
+    jmp       common_invokeOld
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_SUPER_QUICK_RANGE: /* 0xfb */
+/* File: x86/OP_INVOKE_SUPER_QUICK_RANGE.S */
+/* File: x86/OP_INVOKE_SUPER_QUICK.S */
+    /*
+     * Handle an optimized "super" method call.
+     *
+     * for: [opt] invoke-super-quick, invoke-super-quick/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    GET_GLUE(%ecx)
+    movzwl    4(rPC),%eax               # eax<- GFED or CCCC
+    movl      offGlue_method(%ecx),%ecx # ecx<- current method
+    .if       (!1)
+    andl      $0xf,%eax                # eax<- D (or stays CCCC)
+    .endif
+    movl      offMethod_clazz(%ecx),%ecx # ecx<- method->clazz
+    GET_VREG(%eax,%eax)                 # eax<- "this"
+    movl      offClassObject_super(%ecx),%ecx # ecx<- method->clazz->super
+    testl     %eax,%eax                 # null "this"?
+    je        common_errNullObject      # "this" is null, throw exception
+    movzwl    2(rPC),%eax               # eax<- BBBB
+    movl      offClassObject_vtable(%ecx),%ecx # ecx<- vtable
+    EXPORT_PC()
+    movl      (%ecx,%eax,4),%eax        # eax<- super->vtable[BBBB]
+    movl      $1,%ecx           # ecx<- range flag
+    #jmp       common_invokeMethodRange
+    jmp       common_invokeOld
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FC: /* 0xfc */
+/* File: x86/OP_UNUSED_FC.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FD: /* 0xfd */
+/* File: x86/OP_UNUSED_FD.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FE: /* 0xfe */
+/* File: x86/OP_UNUSED_FE.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FF: /* 0xff */
+/* File: x86/OP_UNUSED_FF.S */
+/* File: x86/unused.S */
+    jmp     common_abort
+
+
+
+    .balign 64
+    .size   dvmAsmInstructionStart, .-dvmAsmInstructionStart
+    .global dvmAsmInstructionEnd
+dvmAsmInstructionEnd:
+
+/*
+ * ===========================================================================
+ *  Sister implementations
+ * ===========================================================================
+ */
+    .global dvmAsmSisterStart
+    .type   dvmAsmSisterStart, %function
+    .text
+    .balign 4
+dvmAsmSisterStart:
+
+/* continuation for OP_CONST_STRING */
+
+/* This is the less common path, so we'll redo some work
+   here rather than force spills on the common path */
+.LOP_CONST_STRING_resolve:
+    GET_GLUE(%eax)
+    movl     %ecx,rINST_FULL           # rINST_FULL<- AA
+    EXPORT_PC()
+    movl     offGlue_method(%eax),%eax # eax<- glue->method
+    movzwl   2(rPC),%ecx               # ecx<- BBBB
+    movl     offMethod_clazz(%eax),%eax
+    SPILL(rPC)
+    movl     %ecx,OUT_ARG1(%esp)
+    movl     %eax,OUT_ARG0(%esp)
+    call     dvmResolveString          # go resolve
+    UNSPILL(rPC)
+    testl    %eax,%eax                 # failed?
+    je       common_exceptionThrown
+    SET_VREG(%eax,rINST_FULL)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* continuation for OP_CONST_STRING_JUMBO */
+
+/* This is the less common path, so we'll redo some work
+   here rather than force spills on the common path */
+.LOP_CONST_STRING_JUMBO_resolve:
+    GET_GLUE(%eax)
+    movl     %ecx,rINST_FULL           # rINST_FULL<- AA
+    EXPORT_PC()
+    movl     offGlue_method(%eax),%eax # eax<- glue->method
+    movl     2(rPC),%ecx               # ecx<- BBBBBBBB
+    movl     offMethod_clazz(%eax),%eax
+    SPILL(rPC)
+    movl     %ecx,OUT_ARG1(%esp)
+    movl     %eax,OUT_ARG0(%esp)
+    call     dvmResolveString          # go resolve
+    UNSPILL(rPC)
+    testl    %eax,%eax                 # failed?
+    je       common_exceptionThrown
+    SET_VREG(%eax,rINST_FULL)
+    FETCH_INST_WORD(3)
+    ADVANCE_PC(3)
+    GOTO_NEXT
+
+
+/* continuation for OP_CONST_CLASS */
+
+/* This is the less common path, so we'll redo some work
+   here rather than force spills on the common path */
+.LOP_CONST_CLASS_resolve:
+    GET_GLUE(%eax)
+    movl     %ecx,rINST_FULL           # rINST_FULL<- AA
+    EXPORT_PC()
+    movl     offGlue_method(%eax),%eax # eax<- glue->method
+    movl     $1,OUT_ARG2(%esp)        # true
+    movzwl   2(rPC),%ecx               # ecx<- BBBB
+    movl     offMethod_clazz(%eax),%eax
+    SPILL(rPC)
+    movl     %ecx,OUT_ARG1(%esp)
+    movl     %eax,OUT_ARG0(%esp)
+    call     dvmResolveClass           # go resolve
+    UNSPILL(rPC)
+    testl    %eax,%eax                 # failed?
+    je       common_exceptionThrown
+    SET_VREG(%eax,rINST_FULL)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* continuation for OP_MONITOR_ENTER */
+
+.LOP_MONITOR_ENTER_continue:
+    SPILL(rPC)                          # have to - caller save
+    movl    %ecx,OUT_ARG0(%esp)
+    movl    %eax,OUT_ARG1(%esp)
+    call    dvmLockObject               # dvmLockObject(self,object)
+    UNSPILL(rPC)
+#ifdef WITH_DEADLOCK_PREDICTION
+    GET_GLUE(%ecx)
+    movl    offGlueSelf(%ecx),%ecx      # ecx<- glue->self
+    movl    offThread_exception(%ecx),%eax
+    testl   %eax,%eax
+    jne     common_exceptionThrown
+#endif
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+/* continuation for OP_MONITOR_EXIT */
+
+.LOP_MONITOR_EXIT_continue:
+    call    dvmUnlockObject             # unlock(self,obj)
+    UNSPILL(rPC)
+    FETCH_INST_WORD(1)
+    testl   %eax,%eax                   # success?
+    je      common_exceptionThrown      # no, exception pending
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+/* continuation for OP_CHECK_CAST */
+
+    /*
+     * Trivial test failed, need to perform full check.  This is common.
+     *  ecx holds obj->clazz
+     *  eax holds class resolved from BBBB
+     *  rINST_FULL holds object
+     */
+.LOP_CHECK_CAST_fullcheck:
+    movl    %eax,OUT_ARG1(%esp)
+    movl    %ecx,OUT_ARG0(%esp)
+    SPILL(rPC)
+    call    dvmInstanceofNonTrivial     # eax<- boolean result
+    UNSPILL(rPC)
+    testl   %eax,%eax                   # failed?
+    jne     .LOP_CHECK_CAST_okay            # no, success
+
+    # A cast has failed.  We need to throw a ClassCastException with the
+    # class of the object that failed to be cast.
+    EXPORT_PC()
+    movl    offObject_clazz(rINST_FULL),%ecx  # ecx<- obj->clazz
+    movl    $.LstrClassCastException,%eax
+    movl    offClassObject_descriptor(%ecx),%ecx
+    movl    %eax,OUT_ARG0(%esp)     # arg0<- message
+    movl    %ecx,OUT_ARG1(%esp)     # arg1<- obj->clazz->descriptor
+    SPILL(rPC)
+    call    dvmThrowExceptionWithClassMessage
+    UNSPILL(rPC)
+    jmp     common_exceptionThrown
+
+    /*
+     * Resolution required.  This is the least-likely path, and we're
+     * going to have to recreate some data.
+     *
+     *  rINST_FULL holds object
+     */
+.LOP_CHECK_CAST_resolve:
+    GET_GLUE(%ecx)
+    EXPORT_PC()
+    movzwl  2(rPC),%eax                # eax<- BBBB
+    movl    offGlue_method(%ecx),%ecx  # ecx<- glue->method
+    movl    %eax,OUT_ARG1(%esp)        # arg1<- BBBB
+    movl    offMethod_clazz(%ecx),%ecx # ecx<- metho->clazz
+    movl    $0,OUT_ARG2(%esp)         # arg2<- false
+    movl    %ecx,OUT_ARG0(%esp)        # arg0<- method->clazz
+    SPILL(rPC)
+    call    dvmResolveClass            # eax<- resolved ClassObject ptr
+    UNSPILL(rPC)
+    testl   %eax,%eax                  # got null?
+    je      common_exceptionThrown     # yes, handle exception
+    movl    offObject_clazz(rINST_FULL),%ecx  # ecx<- obj->clazz
+    jmp     .LOP_CHECK_CAST_resolved       # pick up where we left off
+
+/* continuation for OP_INSTANCE_OF */
+
+    /*
+     * Trivial test failed, need to perform full check.  This is common.
+     *  eax holds obj->clazz
+     *  ecx holds class resolved from BBBB
+     *  rINST_HI has BA
+     *  rPC already spilled
+     */
+.LOP_INSTANCE_OF_fullcheck:
+    movl    %eax,OUT_ARG0(%esp)
+    movl    %ecx,OUT_ARG1(%esp)
+    call    dvmInstanceofNonTrivial     # eax<- boolean result
+    # fall through to OP_INSTANCE_OF_store
+
+    /*
+     * eax holds boolean result
+     * rINST_HI holds BA
+     */
+.LOP_INSTANCE_OF_store:
+    UNSPILL(rPC)
+    movzbl  rINST_HI,%ecx               # ecx<- BA
+    FETCH_INST_WORD(2)
+    andb    $0xf,%cl                   # ecl<- A
+    ADVANCE_PC(2)
+    SET_VREG(%eax,%ecx)                 # vA<- eax
+    GOTO_NEXT
+
+    /*
+     * Trivial test succeeded, save and bail.
+     *  r9 holds A
+     */
+.LOP_INSTANCE_OF_trivial:
+    UNSPILL(rPC)
+    movzbl  rINST_HI,%ecx               # ecx<- BA
+    FETCH_INST_WORD(2)
+    andb    $0xf,%cl                   # ecl<- A
+    ADVANCE_PC(2)
+    movl    $1,%eax
+    SET_VREG(%eax,%ecx)                  # vA<- true
+    GOTO_NEXT
+
+    /*
+     * Resolution required.  This is the least-likely path.
+     *
+     *  rPC holds BBBB
+     *  rINST_HI holds BA
+     */
+.LOP_INSTANCE_OF_resolve:
+    movl    rPC,OUT_ARG1(%esp)          # arg1<- BBBB
+    GET_GLUE(%ecx)
+    UNSPILL(rPC)
+    movl    offGlue_method(%ecx),%ecx
+    movl    $1,OUT_ARG2(%esp)          # arg2<- true
+    movl    offMethod_clazz(%ecx),%ecx  # ecx<- method->clazz
+    EXPORT_PC()
+    movl    %ecx,OUT_ARG0(%esp)         # arg0<- method->clazz
+    call    dvmResolveClass             # eax<- resolved ClassObject ptr
+    UNSPILL(rPC)
+    testl   %eax,%eax                   # success?
+    je      common_exceptionThrown      # no, handle exception
+/* Now, we need to sync up with fast path.  We need eax to
+ * hold the obj->clazz, and ecx to hold the resolved class
+ */
+    movl    %eax,%ecx                   # ecx<- resolved class
+    movzbl  rINST_HI,%eax               # eax<- BA
+    sarl    $4,%eax                    # eax<- B
+    GET_VREG(%eax,%eax)                 # eax<- vB (obj)
+    movl    offObject_clazz(%eax),%eax  # eax<- obj->clazz
+    jmp     .LOP_INSTANCE_OF_resolved
+
+
+/* continuation for OP_NEW_INSTANCE */
+
+.LOP_NEW_INSTANCE_initialized:  # on entry, ecx<- class
+    testl     $(ACC_INTERFACE|ACC_ABSTRACT),offClassObject_accessFlags(%ecx)
+    movl      $ALLOC_DONT_TRACK,OUT_ARG1(%esp)
+    jne       .LOP_NEW_INSTANCE_abstract
+.LOP_NEW_INSTANCE_finish: # ecx=class
+    movl     %ecx,OUT_ARG0(%esp)
+    call     dvmAllocObject             # eax<- new object
+    UNSPILL(rPC)
+    movl     rINST_FULL,%ecx
+    FETCH_INST_WORD(2)
+    testl    %eax,%eax                  # success?
+    je       common_exceptionThrown     # no, bail out
+    SET_VREG(%eax,%ecx)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+    /*
+     * Class initialization required.
+     *
+     *  ecx holds class object
+     */
+.LOP_NEW_INSTANCE_needinit:
+    SPILL_TMP(%ecx)                     # save object
+    movl    %ecx,OUT_ARG0(%esp)
+    call    dvmInitClass                # initialize class
+    UNSPILL_TMP(%ecx)                   # restore object
+    testl   %eax,%eax                   # success?
+    jne     .LOP_NEW_INSTANCE_initialized     # success, continue
+    UNSPILL(rPC)                        # failed, restore PC
+    jmp     common_exceptionThrown      # go deal with init exception
+
+    /*
+     * Resolution required.  This is the least-likely path.
+     *
+     */
+.LOP_NEW_INSTANCE_resolve:
+    GET_GLUE(%ecx)
+    movzwl  2(rPC),%eax
+    movl    offGlue_method(%ecx),%ecx   # ecx<- glue->method
+    movl    %eax,OUT_ARG1(%esp)
+    movl    offMethod_clazz(%ecx),%ecx  # ecx<- method->clazz
+    movl    $0,OUT_ARG2(%esp)
+    movl    %ecx,OUT_ARG0(%esp)
+    call    dvmResolveClass             # call(clazz,off,flags)
+    movl    %eax,%ecx                   # ecx<- resolved ClassObject ptr
+    testl   %ecx,%ecx                   # success?
+    jne     .LOP_NEW_INSTANCE_resolved        # good to go
+    UNSPILL(rPC)
+    jmp     common_exceptionThrown      # no, handle exception
+
+    /*
+     * We can't instantiate an abstract class or interface, so throw an
+     * InstantiationError with the class descriptor as the message.
+     *
+     *  ecx holds class object
+     */
+.LOP_NEW_INSTANCE_abstract:
+    movl    offClassObject_descriptor(%ecx),%eax
+    movl    $.LstrInstantiationError,OUT_ARG0(%esp)
+    movl    %eax,OUT_ARG1(%esp)
+    call    dvmThrowExceptionWithClassMessage
+    UNSPILL(rPC)
+    jmp     common_exceptionThrown
+
+
+
+
+/* continuation for OP_NEW_ARRAY */
+
+    /*
+     * Resolve class.  (This is an uncommon case.)
+     *  ecx holds class (null here)
+     *  eax holds array length (vB)
+     */
+.LOP_NEW_ARRAY_resolve:
+    GET_GLUE(%ecx)
+    SPILL_TMP(%eax)                    # save array length
+    movl    offGlue_method(%ecx),%ecx  # ecx<- glue->method
+    movzwl  2(rPC),%eax                # eax<- CCCC
+    movl    offMethod_clazz(%ecx),%ecx # ecx<- method->clazz
+    movl    %eax,OUT_ARG1(%esp)
+    movl    $0,OUT_ARG2(%esp)
+    movl    %ecx,OUT_ARG0(%esp)
+    SPILL(rPC)
+    call    dvmResolveClass            # eax<- call(clazz,ref,flag)
+    UNSPILL(rPC)
+    movl    %eax,%ecx
+    UNSPILL_TMP(%eax)
+    testl   %ecx,%ecx                  # successful resolution?
+    je      common_exceptionThrown     # no, bail.
+# fall through to OP_NEW_ARRAY_finish
+
+    /*
+     * Finish allocation
+     *
+     * ecx holds class
+     * eax holds array length (vB)
+     */
+.LOP_NEW_ARRAY_finish:
+    movl    %ecx,OUT_ARG0(%esp)
+    movl    %eax,OUT_ARG1(%esp)
+    movl    $ALLOC_DONT_TRACK,OUT_ARG2(%esp)
+    SPILL(rPC)
+    call    dvmAllocArrayByClass    # eax<- call(clazz,length,flags)
+    UNSPILL(rPC)
+    testl   %eax,%eax               # failed?
+    je      common_exceptionThrown  # yup - go handle
+    movl    rINST_FULL,%ecx
+    FETCH_INST_WORD(2)
+    SET_VREG(%eax,%ecx)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* continuation for OP_FILLED_NEW_ARRAY */
+
+.LOP_FILLED_NEW_ARRAY_more:
+    movl    offMethod_clazz(%eax),%eax        # eax<- method->clazz
+    movl    %eax,OUT_ARG0(%esp)               # arg0<- clazz
+    call    dvmResolveClass                   # eax<- call(clazz,ref,flag)
+    UNSPILL(rPC)
+    testl   %eax,%eax                         # null?
+    je      common_exceptionThrown            # yes, handle it
+
+       # note: fall through to .LOP_FILLED_NEW_ARRAY_continue
+
+    /*
+     * On entry:
+     *    eax holds array class [r0]
+     *    rINST_FULL holds AA or BB [r10]
+     *    ecx is scratch
+     *    rPC is valid, but has been spilled
+     */
+.LOP_FILLED_NEW_ARRAY_continue:
+    movl    offClassObject_descriptor(%eax),%ecx  # ecx<- arrayClass->descriptor
+    movl    $ALLOC_DONT_TRACK,OUT_ARG2(%esp)     # arg2<- flags
+    movzbl  1(%ecx),%ecx                          # ecx<- descriptor[1]
+    movl    %eax,OUT_ARG0(%esp)                   # arg0<- arrayClass
+    cmpb    $'I',%cl                             # supported?
+    je      1f
+    cmpb    $'L',%cl
+    je      1f
+    cmpb    $'[',%cl
+    jne      .LOP_FILLED_NEW_ARRAY_notimpl                  # no, not handled yet
+1:
+    .if      (!0)
+    SPILL_TMP(rINST_FULL)                         # save copy, need "B" later
+    sarl    $4,rINST_FULL
+    .endif
+    movl    rINST_FULL,OUT_ARG1(%esp)             # arg1<- A or AA (length)
+    call    dvmAllocArrayByClass                  # eax<- call(arrayClass, length, flags)
+    UNSPILL(rPC)
+    GET_GLUE(%ecx)
+    testl   %eax,%eax                             # alloc successful?
+    je      common_exceptionThrown                # no, handle exception
+    movl    %eax,offGlue_retval(%ecx)             # retval.l<- new array
+    movzwl  4(rPC),%ecx                           # ecx<- FEDC or CCCC
+    leal    offArrayObject_contents(%eax),%eax    # eax<- newArray->contents
+
+/* at this point:
+ *     eax is pointer to tgt
+ *     rINST_FULL is length
+ *     ecx is FEDC or CCCC
+ *     TMP_SPILL is BA
+ *     rPC is valid, but spilled
+ *  We now need to copy values from registers into the array
+ */
+
+    .if 0
+    # set up src pointer
+    SPILL(rFP)     # esi
+    SPILL(rIBASE)   # edi
+    movl    %eax,%edi         # set up dst ptr
+    leal    (rFP,%ecx,4),%esi # set up src ptr
+    movl    rINST_FULL,%ecx   # load count register
+    FETCH_INST_WORD(3)
+    rep
+    movsd
+    UNSPILL(rIBASE)
+    UNSPILL(rFP)
+    .else
+    testl  rINST_FULL,rINST_FULL
+    je     4f
+    UNSPILL_TMP(rPC)
+    andl   $0x0f,rPC            # rPC<- 0000000A
+    sall   $16,rPC              # rPC<- 000A0000
+    orl    %ecx,rPC              # rpc<- 000AFEDC
+3:
+    movl   $0xf,%ecx
+    andl   rPC,%ecx           # ecx<- next reg to load
+    GET_VREG(%ecx,%ecx)
+    shrl   $4,rPC
+    leal   4(%eax),%eax
+    movl   %ecx,-4(%eax)
+    sub    $1,rINST_FULL
+    jne    3b
+4:
+    UNSPILL(rPC)
+    FETCH_INST_WORD(3)
+    .endif
+
+    ADVANCE_PC(3)
+    GOTO_NEXT
+
+
+    /*
+     * Throw an exception indicating that we have not implemented this
+     * mode of filled-new-array.
+     */
+.LOP_FILLED_NEW_ARRAY_notimpl:
+    movl    $.LstrInternalError,%eax
+    movl    %eax,OUT_ARG0(%esp)
+    movl    $.LstrFilledNewArrayNotImpl,%eax
+    movl    %eax,OUT_ARG1(%esp)
+    call    dvmThrowException
+    UNSPILL(rPC)
+    jmp     common_exceptionThrown
+
+
+/* continuation for OP_FILLED_NEW_ARRAY_RANGE */
+
+.LOP_FILLED_NEW_ARRAY_RANGE_more:
+    movl    offMethod_clazz(%eax),%eax        # eax<- method->clazz
+    movl    %eax,OUT_ARG0(%esp)               # arg0<- clazz
+    call    dvmResolveClass                   # eax<- call(clazz,ref,flag)
+    UNSPILL(rPC)
+    testl   %eax,%eax                         # null?
+    je      common_exceptionThrown            # yes, handle it
+
+       # note: fall through to .LOP_FILLED_NEW_ARRAY_RANGE_continue
+
+    /*
+     * On entry:
+     *    eax holds array class [r0]
+     *    rINST_FULL holds AA or BB [r10]
+     *    ecx is scratch
+     *    rPC is valid, but has been spilled
+     */
+.LOP_FILLED_NEW_ARRAY_RANGE_continue:
+    movl    offClassObject_descriptor(%eax),%ecx  # ecx<- arrayClass->descriptor
+    movl    $ALLOC_DONT_TRACK,OUT_ARG2(%esp)     # arg2<- flags
+    movzbl  1(%ecx),%ecx                          # ecx<- descriptor[1]
+    movl    %eax,OUT_ARG0(%esp)                   # arg0<- arrayClass
+    cmpb    $'I',%cl                             # supported?
+    je      1f
+    cmpb    $'L',%cl
+    je      1f
+    cmpb    $'[',%cl
+    jne      .LOP_FILLED_NEW_ARRAY_RANGE_notimpl                  # no, not handled yet
+1:
+    .if      (!1)
+    SPILL_TMP(rINST_FULL)                         # save copy, need "B" later
+    sarl    $4,rINST_FULL
+    .endif
+    movl    rINST_FULL,OUT_ARG1(%esp)             # arg1<- A or AA (length)
+    call    dvmAllocArrayByClass                  # eax<- call(arrayClass, length, flags)
+    UNSPILL(rPC)
+    GET_GLUE(%ecx)
+    testl   %eax,%eax                             # alloc successful?
+    je      common_exceptionThrown                # no, handle exception
+    movl    %eax,offGlue_retval(%ecx)             # retval.l<- new array
+    movzwl  4(rPC),%ecx                           # ecx<- FEDC or CCCC
+    leal    offArrayObject_contents(%eax),%eax    # eax<- newArray->contents
+
+/* at this point:
+ *     eax is pointer to tgt
+ *     rINST_FULL is length
+ *     ecx is FEDC or CCCC
+ *     TMP_SPILL is BA
+ *     rPC is valid, but spilled
+ *  We now need to copy values from registers into the array
+ */
+
+    .if 1
+    # set up src pointer
+    SPILL(rFP)     # esi
+    SPILL(rIBASE)   # edi
+    movl    %eax,%edi         # set up dst ptr
+    leal    (rFP,%ecx,4),%esi # set up src ptr
+    movl    rINST_FULL,%ecx   # load count register
+    FETCH_INST_WORD(3)
+    rep
+    movsd
+    UNSPILL(rIBASE)
+    UNSPILL(rFP)
+    .else
+    testl  rINST_FULL,rINST_FULL
+    je     4f
+    UNSPILL_TMP(rPC)
+    andl   $0x0f,rPC            # rPC<- 0000000A
+    sall   $16,rPC              # rPC<- 000A0000
+    orl    %ecx,rPC              # rpc<- 000AFEDC
+3:
+    movl   $0xf,%ecx
+    andl   rPC,%ecx           # ecx<- next reg to load
+    GET_VREG(%ecx,%ecx)
+    shrl   $4,rPC
+    leal   4(%eax),%eax
+    movl   %ecx,-4(%eax)
+    sub    $1,rINST_FULL
+    jne    3b
+4:
+    UNSPILL(rPC)
+    FETCH_INST_WORD(3)
+    .endif
+
+    ADVANCE_PC(3)
+    GOTO_NEXT
+
+
+    /*
+     * Throw an exception indicating that we have not implemented this
+     * mode of filled-new-array.
+     */
+.LOP_FILLED_NEW_ARRAY_RANGE_notimpl:
+    movl    $.LstrInternalError,%eax
+    movl    %eax,OUT_ARG0(%esp)
+    movl    $.LstrFilledNewArrayNotImpl,%eax
+    movl    %eax,OUT_ARG1(%esp)
+    call    dvmThrowException
+    UNSPILL(rPC)
+    jmp     common_exceptionThrown
+
+
+/* continuation for OP_CMPL_FLOAT */
+
+.LOP_CMPL_FLOAT_isNaN:
+    movl      $-1,%ecx
+    jmp       .LOP_CMPL_FLOAT_finish
+
+/* continuation for OP_CMPG_FLOAT */
+
+.LOP_CMPG_FLOAT_isNaN:
+    movl      $1,%ecx
+    jmp       .LOP_CMPG_FLOAT_finish
+
+/* continuation for OP_CMPL_DOUBLE */
+
+.LOP_CMPL_DOUBLE_isNaN:
+    movl      $-1,%ecx
+    jmp       .LOP_CMPL_DOUBLE_finish
+
+/* continuation for OP_CMPG_DOUBLE */
+
+.LOP_CMPG_DOUBLE_isNaN:
+    movl      $1,%ecx
+    jmp       .LOP_CMPG_DOUBLE_finish
+
+/* continuation for OP_CMP_LONG */
+
+.LOP_CMP_LONG_bigger:
+    UNSPILL(rPC)
+    movl      $1,%ecx
+    jmp       .LOP_CMP_LONG_finish
+.LOP_CMP_LONG_smaller:
+    UNSPILL(rPC)
+    movl      $-1,%ecx
+.LOP_CMP_LONG_finish:
+    SET_VREG(%ecx,rINST_FULL)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* continuation for OP_AGET_WIDE */
+
+.LOP_AGET_WIDE_finish:
+    leal      offArrayObject_contents(%eax,%ecx,8),%eax
+    movl      (%eax),%ecx
+    movl      4(%eax),%eax
+    SET_VREG_WORD(%ecx,rINST_FULL,0)
+    SET_VREG_WORD(%eax,rINST_FULL,1)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* continuation for OP_APUT_WIDE */
+
+.LOP_APUT_WIDE_finish:
+    leal      offArrayObject_contents(%eax,%ecx,8),%eax
+    GET_VREG_WORD(%ecx,rINST_FULL,0)
+    GET_VREG_WORD(rINST_FULL,rINST_FULL,1)
+    movl      rINST_FULL,4(%eax)
+    FETCH_INST_WORD(2)
+    movl      %ecx,(%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* continuation for OP_APUT_OBJECT */
+
+    /* On entry:
+     *   eax<- array object
+     *   ecx<- index
+     *   rINST_FULL<- vAA
+     */
+.LOP_APUT_OBJECT_continue:
+    leal      offArrayObject_contents(%eax,%ecx,4),%ecx
+    testl     rINST_FULL,rINST_FULL     # storing null reference?
+    je        .LOP_APUT_OBJECT_skip_check
+    SPILL(rPC)
+    SPILL_TMP(%ecx)
+    movl      offObject_clazz(%eax),%eax # eax<- arrayObj->clazz
+    movl      offObject_clazz(rINST_FULL),%ecx # ecx<- obj->clazz
+    movl      %eax,OUT_ARG1(%esp)
+    movl      %ecx,OUT_ARG0(%esp)
+    call      dvmCanPutArrayElement     # test object type vs. array type
+    UNSPILL(rPC)
+    UNSPILL_TMP(%ecx)
+    testl     %eax,%eax
+    je        common_errArrayStore
+
+.LOP_APUT_OBJECT_skip_check:
+    movl      rINST_FULL,(%ecx)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* continuation for OP_IGET */
+
+
+.LOP_IGET_resolve:
+    EXPORT_PC()
+    SPILL(rPC)
+    movl    offGlue_method(rIBASE),rPC            # rPC<- current method
+    UNSPILL(rIBASE)
+    movl    offMethod_clazz(rPC),rPC              # rPC<- method->clazz
+    SPILL_TMP(%ecx)                               # save object pointer across call
+    movl    rPC,OUT_ARG0(%esp)                    # pass in method->clazz
+    call    dvmResolveInstField                   #  ... to dvmResolveInstField
+    UNSPILL_TMP(%ecx)
+    UNSPILL(rPC)
+    testl   %eax,%eax                             #  ... which returns InstrField ptr
+    jne     .LOP_IGET_finish
+    jmp     common_exceptionThrown
+
+.LOP_IGET_finish:
+    /*
+     * Currently:
+     *   eax holds resolved field
+     *   ecx holds object
+     *   rIBASE is scratch, but needs to be unspilled
+     *   rINST_FULL holds A
+     */
+    movl    offInstField_byteOffset(%eax),%eax   # eax<- byte offset of field
+    UNSPILL(rIBASE)
+    testl   %ecx,%ecx                            # object null?
+    je      common_errNullObject                 # object was null
+    movl   (%ecx,%eax,1),%ecx                   # ecx<- obj.field (8/16/32 bits)
+    movl    rINST_FULL,%eax                      # eax<- A
+    FETCH_INST_WORD(2)
+    SET_VREG(%ecx,%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* continuation for OP_IGET_WIDE */
+
+
+.LOP_IGET_WIDE_resolve:
+    EXPORT_PC()
+    SPILL(rPC)
+    movl    offGlue_method(rIBASE),rPC            # rPC<- current method
+    UNSPILL(rIBASE)
+    movl    offMethod_clazz(rPC),rPC              # rPC<- method->clazz
+    SPILL_TMP(%ecx)                               # save object pointer across call
+    movl    rPC,OUT_ARG0(%esp)                    # pass in method->clazz
+    call    dvmResolveInstField                   #  ... to dvmResolveInstField
+    UNSPILL_TMP(%ecx)
+    UNSPILL(rPC)
+    testl   %eax,%eax                             #  ... which returns InstrField ptr
+    jne     .LOP_IGET_WIDE_finish
+    jmp     common_exceptionThrown
+
+.LOP_IGET_WIDE_finish:
+    /*
+     * Currently:
+     *   eax holds resolved field
+     *   ecx holds object
+     *   rIBASE is scratch, but needs to be unspilled
+     *   rINST_FULL holds A
+     */
+    movl    offInstField_byteOffset(%eax),%eax   # eax<- byte offset of field
+    UNSPILL(rIBASE)
+    testl   %ecx,%ecx                            # object null?
+    je      common_errNullObject                 # object was null
+    leal    (%ecx,%eax,1),%eax                   # eax<- address of field
+    movl    (%eax),%ecx                          # ecx<- lsw
+    movl    4(%eax),%eax                         # eax<- msw
+    SET_VREG_WORD(%ecx,rINST_FULL,0)
+    SET_VREG_WORD(%eax,rINST_FULL,1)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* continuation for OP_IGET_OBJECT */
+
+
+.LOP_IGET_OBJECT_resolve:
+    EXPORT_PC()
+    SPILL(rPC)
+    movl    offGlue_method(rIBASE),rPC            # rPC<- current method
+    UNSPILL(rIBASE)
+    movl    offMethod_clazz(rPC),rPC              # rPC<- method->clazz
+    SPILL_TMP(%ecx)                               # save object pointer across call
+    movl    rPC,OUT_ARG0(%esp)                    # pass in method->clazz
+    call    dvmResolveInstField                   #  ... to dvmResolveInstField
+    UNSPILL_TMP(%ecx)
+    UNSPILL(rPC)
+    testl   %eax,%eax                             #  ... which returns InstrField ptr
+    jne     .LOP_IGET_OBJECT_finish
+    jmp     common_exceptionThrown
+
+.LOP_IGET_OBJECT_finish:
+    /*
+     * Currently:
+     *   eax holds resolved field
+     *   ecx holds object
+     *   rIBASE is scratch, but needs to be unspilled
+     *   rINST_FULL holds A
+     */
+    movl    offInstField_byteOffset(%eax),%eax   # eax<- byte offset of field
+    UNSPILL(rIBASE)
+    testl   %ecx,%ecx                            # object null?
+    je      common_errNullObject                 # object was null
+    movl   (%ecx,%eax,1),%ecx                   # ecx<- obj.field (8/16/32 bits)
+    movl    rINST_FULL,%eax                      # eax<- A
+    FETCH_INST_WORD(2)
+    SET_VREG(%ecx,%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* continuation for OP_IGET_BOOLEAN */
+
+
+.LOP_IGET_BOOLEAN_resolve:
+    EXPORT_PC()
+    SPILL(rPC)
+    movl    offGlue_method(rIBASE),rPC            # rPC<- current method
+    UNSPILL(rIBASE)
+    movl    offMethod_clazz(rPC),rPC              # rPC<- method->clazz
+    SPILL_TMP(%ecx)                               # save object pointer across call
+    movl    rPC,OUT_ARG0(%esp)                    # pass in method->clazz
+    call    dvmResolveInstField                   #  ... to dvmResolveInstField
+    UNSPILL_TMP(%ecx)
+    UNSPILL(rPC)
+    testl   %eax,%eax                             #  ... which returns InstrField ptr
+    jne     .LOP_IGET_BOOLEAN_finish
+    jmp     common_exceptionThrown
+
+.LOP_IGET_BOOLEAN_finish:
+    /*
+     * Currently:
+     *   eax holds resolved field
+     *   ecx holds object
+     *   rIBASE is scratch, but needs to be unspilled
+     *   rINST_FULL holds A
+     */
+    movl    offInstField_byteOffset(%eax),%eax   # eax<- byte offset of field
+    UNSPILL(rIBASE)
+    testl   %ecx,%ecx                            # object null?
+    je      common_errNullObject                 # object was null
+    movzbl   (%ecx,%eax,1),%ecx                   # ecx<- obj.field (8/16/32 bits)
+    movl    rINST_FULL,%eax                      # eax<- A
+    FETCH_INST_WORD(2)
+    SET_VREG(%ecx,%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* continuation for OP_IGET_BYTE */
+
+
+.LOP_IGET_BYTE_resolve:
+    EXPORT_PC()
+    SPILL(rPC)
+    movl    offGlue_method(rIBASE),rPC            # rPC<- current method
+    UNSPILL(rIBASE)
+    movl    offMethod_clazz(rPC),rPC              # rPC<- method->clazz
+    SPILL_TMP(%ecx)                               # save object pointer across call
+    movl    rPC,OUT_ARG0(%esp)                    # pass in method->clazz
+    call    dvmResolveInstField                   #  ... to dvmResolveInstField
+    UNSPILL_TMP(%ecx)
+    UNSPILL(rPC)
+    testl   %eax,%eax                             #  ... which returns InstrField ptr
+    jne     .LOP_IGET_BYTE_finish
+    jmp     common_exceptionThrown
+
+.LOP_IGET_BYTE_finish:
+    /*
+     * Currently:
+     *   eax holds resolved field
+     *   ecx holds object
+     *   rIBASE is scratch, but needs to be unspilled
+     *   rINST_FULL holds A
+     */
+    movl    offInstField_byteOffset(%eax),%eax   # eax<- byte offset of field
+    UNSPILL(rIBASE)
+    testl   %ecx,%ecx                            # object null?
+    je      common_errNullObject                 # object was null
+    movsbl   (%ecx,%eax,1),%ecx                   # ecx<- obj.field (8/16/32 bits)
+    movl    rINST_FULL,%eax                      # eax<- A
+    FETCH_INST_WORD(2)
+    SET_VREG(%ecx,%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* continuation for OP_IGET_CHAR */
+
+
+.LOP_IGET_CHAR_resolve:
+    EXPORT_PC()
+    SPILL(rPC)
+    movl    offGlue_method(rIBASE),rPC            # rPC<- current method
+    UNSPILL(rIBASE)
+    movl    offMethod_clazz(rPC),rPC              # rPC<- method->clazz
+    SPILL_TMP(%ecx)                               # save object pointer across call
+    movl    rPC,OUT_ARG0(%esp)                    # pass in method->clazz
+    call    dvmResolveInstField                   #  ... to dvmResolveInstField
+    UNSPILL_TMP(%ecx)
+    UNSPILL(rPC)
+    testl   %eax,%eax                             #  ... which returns InstrField ptr
+    jne     .LOP_IGET_CHAR_finish
+    jmp     common_exceptionThrown
+
+.LOP_IGET_CHAR_finish:
+    /*
+     * Currently:
+     *   eax holds resolved field
+     *   ecx holds object
+     *   rIBASE is scratch, but needs to be unspilled
+     *   rINST_FULL holds A
+     */
+    movl    offInstField_byteOffset(%eax),%eax   # eax<- byte offset of field
+    UNSPILL(rIBASE)
+    testl   %ecx,%ecx                            # object null?
+    je      common_errNullObject                 # object was null
+    movzwl   (%ecx,%eax,1),%ecx                   # ecx<- obj.field (8/16/32 bits)
+    movl    rINST_FULL,%eax                      # eax<- A
+    FETCH_INST_WORD(2)
+    SET_VREG(%ecx,%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* continuation for OP_IGET_SHORT */
+
+
+.LOP_IGET_SHORT_resolve:
+    EXPORT_PC()
+    SPILL(rPC)
+    movl    offGlue_method(rIBASE),rPC            # rPC<- current method
+    UNSPILL(rIBASE)
+    movl    offMethod_clazz(rPC),rPC              # rPC<- method->clazz
+    SPILL_TMP(%ecx)                               # save object pointer across call
+    movl    rPC,OUT_ARG0(%esp)                    # pass in method->clazz
+    call    dvmResolveInstField                   #  ... to dvmResolveInstField
+    UNSPILL_TMP(%ecx)
+    UNSPILL(rPC)
+    testl   %eax,%eax                             #  ... which returns InstrField ptr
+    jne     .LOP_IGET_SHORT_finish
+    jmp     common_exceptionThrown
+
+.LOP_IGET_SHORT_finish:
+    /*
+     * Currently:
+     *   eax holds resolved field
+     *   ecx holds object
+     *   rIBASE is scratch, but needs to be unspilled
+     *   rINST_FULL holds A
+     */
+    movl    offInstField_byteOffset(%eax),%eax   # eax<- byte offset of field
+    UNSPILL(rIBASE)
+    testl   %ecx,%ecx                            # object null?
+    je      common_errNullObject                 # object was null
+    movswl   (%ecx,%eax,1),%ecx                   # ecx<- obj.field (8/16/32 bits)
+    movl    rINST_FULL,%eax                      # eax<- A
+    FETCH_INST_WORD(2)
+    SET_VREG(%ecx,%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* continuation for OP_IPUT */
+
+
+.LOP_IPUT_resolve:
+    EXPORT_PC()
+    SPILL(rPC)
+    movl    offGlue_method(rIBASE),rPC            # rPC<- current method
+    UNSPILL(rIBASE)
+    movl    offMethod_clazz(rPC),rPC              # rPC<- method->clazz
+    SPILL_TMP(%ecx)                               # save object pointer across call
+    movl    rPC,OUT_ARG0(%esp)                    # pass in method->clazz
+    call    dvmResolveInstField                   #  ... to dvmResolveInstField
+    UNSPILL_TMP(%ecx)
+    UNSPILL(rPC)
+    testl   %eax,%eax                             #  ... which returns InstrField ptr
+    jne     .LOP_IPUT_finish
+    jmp     common_exceptionThrown
+
+.LOP_IPUT_finish:
+    /*
+     * Currently:
+     *   eax holds resolved field
+     *   ecx holds object
+     *   rIBASE is scratch, but needs to be unspilled
+     *   rINST_FULL holds A
+     */
+    GET_VREG(rINST_FULL,rINST_FULL)              # rINST_FULL<- v[A]
+    movl    offInstField_byteOffset(%eax),%eax   # eax<- byte offset of field
+    UNSPILL(rIBASE)
+    testl   %ecx,%ecx                            # object null?
+    je      common_errNullObject                 # object was null
+    movl   rINST_FULL,(%ecx,%eax,1)            # obj.field <- v[A](8/16/32 bits)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* continuation for OP_IPUT_WIDE */
+
+
+.LOP_IPUT_WIDE_resolve:
+    EXPORT_PC()
+    SPILL(rPC)
+    movl    offGlue_method(rIBASE),rPC            # rPC<- current method
+    UNSPILL(rIBASE)
+    movl    offMethod_clazz(rPC),rPC              # rPC<- method->clazz
+    SPILL_TMP(%ecx)                               # save object pointer across call
+    movl    rPC,OUT_ARG0(%esp)                    # pass in method->clazz
+    call    dvmResolveInstField                   #  ... to dvmResolveInstField
+    UNSPILL_TMP(%ecx)
+    UNSPILL(rPC)
+    testl   %eax,%eax                             #  ... which returns InstrField ptr
+    jne     .LOP_IPUT_WIDE_finish
+    jmp     common_exceptionThrown
+
+.LOP_IPUT_WIDE_finish:
+    /*
+     * Currently:
+     *   eax holds resolved field
+     *   ecx holds object
+     *   rIBASE is scratch, but needs to be unspilled
+     *   rINST_FULL holds A
+     */
+    movl    offInstField_byteOffset(%eax),%eax   # eax<- byte offset of field
+    UNSPILL(rIBASE)
+    testl   %ecx,%ecx                            # object null?
+    je      common_errNullObject                 # object was null
+    leal    (%ecx,%eax,1),%eax                   # eax<- address of field
+    GET_VREG_WORD(%ecx,rINST_FULL,0)             # ecx<- lsw
+    GET_VREG_WORD(rINST_FULL,rINST_FULL,1)       # rINST_FULL<- msw
+    movl    rINST_FULL,4(%eax)
+    FETCH_INST_WORD(2)
+    movl    %ecx,(%eax)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* continuation for OP_IPUT_OBJECT */
+
+
+.LOP_IPUT_OBJECT_resolve:
+    EXPORT_PC()
+    SPILL(rPC)
+    movl    offGlue_method(rIBASE),rPC            # rPC<- current method
+    UNSPILL(rIBASE)
+    movl    offMethod_clazz(rPC),rPC              # rPC<- method->clazz
+    SPILL_TMP(%ecx)                               # save object pointer across call
+    movl    rPC,OUT_ARG0(%esp)                    # pass in method->clazz
+    call    dvmResolveInstField                   #  ... to dvmResolveInstField
+    UNSPILL_TMP(%ecx)
+    UNSPILL(rPC)
+    testl   %eax,%eax                             #  ... which returns InstrField ptr
+    jne     .LOP_IPUT_OBJECT_finish
+    jmp     common_exceptionThrown
+
+.LOP_IPUT_OBJECT_finish:
+    /*
+     * Currently:
+     *   eax holds resolved field
+     *   ecx holds object
+     *   rIBASE is scratch, but needs to be unspilled
+     *   rINST_FULL holds A
+     */
+    GET_VREG(rINST_FULL,rINST_FULL)              # rINST_FULL<- v[A]
+    movl    offInstField_byteOffset(%eax),%eax   # eax<- byte offset of field
+    UNSPILL(rIBASE)
+    testl   %ecx,%ecx                            # object null?
+    je      common_errNullObject                 # object was null
+    movl   rINST_FULL,(%ecx,%eax,1)            # obj.field <- v[A](8/16/32 bits)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* continuation for OP_IPUT_BOOLEAN */
+
+
+.LOP_IPUT_BOOLEAN_resolve:
+    EXPORT_PC()
+    SPILL(rPC)
+    movl    offGlue_method(rIBASE),rPC            # rPC<- current method
+    UNSPILL(rIBASE)
+    movl    offMethod_clazz(rPC),rPC              # rPC<- method->clazz
+    SPILL_TMP(%ecx)                               # save object pointer across call
+    movl    rPC,OUT_ARG0(%esp)                    # pass in method->clazz
+    call    dvmResolveInstField                   #  ... to dvmResolveInstField
+    UNSPILL_TMP(%ecx)
+    UNSPILL(rPC)
+    testl   %eax,%eax                             #  ... which returns InstrField ptr
+    jne     .LOP_IPUT_BOOLEAN_finish
+    jmp     common_exceptionThrown
+
+.LOP_IPUT_BOOLEAN_finish:
+    /*
+     * Currently:
+     *   eax holds resolved field
+     *   ecx holds object
+     *   rIBASE is scratch, but needs to be unspilled
+     *   rINST_FULL holds A
+     */
+    GET_VREG(rINST_FULL,rINST_FULL)              # rINST_FULL<- v[A]
+    movl    offInstField_byteOffset(%eax),%eax   # eax<- byte offset of field
+    UNSPILL(rIBASE)
+    testl   %ecx,%ecx                            # object null?
+    je      common_errNullObject                 # object was null
+    movb   rINST_LO,(%ecx,%eax,1)            # obj.field <- v[A](8/16/32 bits)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* continuation for OP_IPUT_BYTE */
+
+
+.LOP_IPUT_BYTE_resolve:
+    EXPORT_PC()
+    SPILL(rPC)
+    movl    offGlue_method(rIBASE),rPC            # rPC<- current method
+    UNSPILL(rIBASE)
+    movl    offMethod_clazz(rPC),rPC              # rPC<- method->clazz
+    SPILL_TMP(%ecx)                               # save object pointer across call
+    movl    rPC,OUT_ARG0(%esp)                    # pass in method->clazz
+    call    dvmResolveInstField                   #  ... to dvmResolveInstField
+    UNSPILL_TMP(%ecx)
+    UNSPILL(rPC)
+    testl   %eax,%eax                             #  ... which returns InstrField ptr
+    jne     .LOP_IPUT_BYTE_finish
+    jmp     common_exceptionThrown
+
+.LOP_IPUT_BYTE_finish:
+    /*
+     * Currently:
+     *   eax holds resolved field
+     *   ecx holds object
+     *   rIBASE is scratch, but needs to be unspilled
+     *   rINST_FULL holds A
+     */
+    GET_VREG(rINST_FULL,rINST_FULL)              # rINST_FULL<- v[A]
+    movl    offInstField_byteOffset(%eax),%eax   # eax<- byte offset of field
+    UNSPILL(rIBASE)
+    testl   %ecx,%ecx                            # object null?
+    je      common_errNullObject                 # object was null
+    movb   rINST_LO,(%ecx,%eax,1)            # obj.field <- v[A](8/16/32 bits)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* continuation for OP_IPUT_CHAR */
+
+
+.LOP_IPUT_CHAR_resolve:
+    EXPORT_PC()
+    SPILL(rPC)
+    movl    offGlue_method(rIBASE),rPC            # rPC<- current method
+    UNSPILL(rIBASE)
+    movl    offMethod_clazz(rPC),rPC              # rPC<- method->clazz
+    SPILL_TMP(%ecx)                               # save object pointer across call
+    movl    rPC,OUT_ARG0(%esp)                    # pass in method->clazz
+    call    dvmResolveInstField                   #  ... to dvmResolveInstField
+    UNSPILL_TMP(%ecx)
+    UNSPILL(rPC)
+    testl   %eax,%eax                             #  ... which returns InstrField ptr
+    jne     .LOP_IPUT_CHAR_finish
+    jmp     common_exceptionThrown
+
+.LOP_IPUT_CHAR_finish:
+    /*
+     * Currently:
+     *   eax holds resolved field
+     *   ecx holds object
+     *   rIBASE is scratch, but needs to be unspilled
+     *   rINST_FULL holds A
+     */
+    GET_VREG(rINST_FULL,rINST_FULL)              # rINST_FULL<- v[A]
+    movl    offInstField_byteOffset(%eax),%eax   # eax<- byte offset of field
+    UNSPILL(rIBASE)
+    testl   %ecx,%ecx                            # object null?
+    je      common_errNullObject                 # object was null
+    movw   rINST,(%ecx,%eax,1)            # obj.field <- v[A](8/16/32 bits)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* continuation for OP_IPUT_SHORT */
+
+
+.LOP_IPUT_SHORT_resolve:
+    EXPORT_PC()
+    SPILL(rPC)
+    movl    offGlue_method(rIBASE),rPC            # rPC<- current method
+    UNSPILL(rIBASE)
+    movl    offMethod_clazz(rPC),rPC              # rPC<- method->clazz
+    SPILL_TMP(%ecx)                               # save object pointer across call
+    movl    rPC,OUT_ARG0(%esp)                    # pass in method->clazz
+    call    dvmResolveInstField                   #  ... to dvmResolveInstField
+    UNSPILL_TMP(%ecx)
+    UNSPILL(rPC)
+    testl   %eax,%eax                             #  ... which returns InstrField ptr
+    jne     .LOP_IPUT_SHORT_finish
+    jmp     common_exceptionThrown
+
+.LOP_IPUT_SHORT_finish:
+    /*
+     * Currently:
+     *   eax holds resolved field
+     *   ecx holds object
+     *   rIBASE is scratch, but needs to be unspilled
+     *   rINST_FULL holds A
+     */
+    GET_VREG(rINST_FULL,rINST_FULL)              # rINST_FULL<- v[A]
+    movl    offInstField_byteOffset(%eax),%eax   # eax<- byte offset of field
+    UNSPILL(rIBASE)
+    testl   %ecx,%ecx                            # object null?
+    je      common_errNullObject                 # object was null
+    movw   rINST,(%ecx,%eax,1)            # obj.field <- v[A](8/16/32 bits)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* continuation for OP_SGET */
+
+    /*
+     * Go resolve the field
+     */
+.LOP_SGET_resolve:
+    GET_GLUE(%ecx)
+    movzwl   2(rPC),%eax                        # eax<- field ref BBBB
+    movl     offGlue_method(%ecx),%ecx          # ecx<- current method
+    EXPORT_PC()                                 # could throw, need to export
+    movl     offMethod_clazz(%ecx),%ecx         # ecx<- method->clazz
+    SPILL(rPC)
+    movl     %eax,OUT_ARG1(%esp)
+    movl     %ecx,OUT_ARG0(%esp)
+    call     dvmResolveStaticField              # eax<- resolved StaticField ptr
+    UNSPILL(rPC)
+    testl    %eax,%eax
+    jne      .LOP_SGET_finish                 # success, continue
+    jmp      common_exceptionThrown             # no, handle exception
+
+
+/* continuation for OP_SGET_WIDE */
+
+    /*
+     * Go resolve the field
+     */
+.LOP_SGET_WIDE_resolve:
+    GET_GLUE(%ecx)
+    movzwl   2(rPC),%eax                        # eax<- field ref BBBB
+    movl     offGlue_method(%ecx),%ecx          # ecx<- current method
+    EXPORT_PC()                                 # could throw, need to export
+    movl     offMethod_clazz(%ecx),%ecx         # ecx<- method->clazz
+    SPILL(rPC)
+    movl     %eax,OUT_ARG1(%esp)
+    movl     %ecx,OUT_ARG0(%esp)
+    call     dvmResolveStaticField              # eax<- resolved StaticField ptr
+    UNSPILL(rPC)
+    testl    %eax,%eax
+    jne      .LOP_SGET_WIDE_finish                 # success, continue
+    jmp      common_exceptionThrown             # no, handle exception
+
+
+/* continuation for OP_SGET_OBJECT */
+
+    /*
+     * Go resolve the field
+     */
+.LOP_SGET_OBJECT_resolve:
+    GET_GLUE(%ecx)
+    movzwl   2(rPC),%eax                        # eax<- field ref BBBB
+    movl     offGlue_method(%ecx),%ecx          # ecx<- current method
+    EXPORT_PC()                                 # could throw, need to export
+    movl     offMethod_clazz(%ecx),%ecx         # ecx<- method->clazz
+    SPILL(rPC)
+    movl     %eax,OUT_ARG1(%esp)
+    movl     %ecx,OUT_ARG0(%esp)
+    call     dvmResolveStaticField              # eax<- resolved StaticField ptr
+    UNSPILL(rPC)
+    testl    %eax,%eax
+    jne      .LOP_SGET_OBJECT_finish                 # success, continue
+    jmp      common_exceptionThrown             # no, handle exception
+
+
+/* continuation for OP_SGET_BOOLEAN */
+
+    /*
+     * Go resolve the field
+     */
+.LOP_SGET_BOOLEAN_resolve:
+    GET_GLUE(%ecx)
+    movzwl   2(rPC),%eax                        # eax<- field ref BBBB
+    movl     offGlue_method(%ecx),%ecx          # ecx<- current method
+    EXPORT_PC()                                 # could throw, need to export
+    movl     offMethod_clazz(%ecx),%ecx         # ecx<- method->clazz
+    SPILL(rPC)
+    movl     %eax,OUT_ARG1(%esp)
+    movl     %ecx,OUT_ARG0(%esp)
+    call     dvmResolveStaticField              # eax<- resolved StaticField ptr
+    UNSPILL(rPC)
+    testl    %eax,%eax
+    jne      .LOP_SGET_BOOLEAN_finish                 # success, continue
+    jmp      common_exceptionThrown             # no, handle exception
+
+
+/* continuation for OP_SGET_BYTE */
+
+    /*
+     * Go resolve the field
+     */
+.LOP_SGET_BYTE_resolve:
+    GET_GLUE(%ecx)
+    movzwl   2(rPC),%eax                        # eax<- field ref BBBB
+    movl     offGlue_method(%ecx),%ecx          # ecx<- current method
+    EXPORT_PC()                                 # could throw, need to export
+    movl     offMethod_clazz(%ecx),%ecx         # ecx<- method->clazz
+    SPILL(rPC)
+    movl     %eax,OUT_ARG1(%esp)
+    movl     %ecx,OUT_ARG0(%esp)
+    call     dvmResolveStaticField              # eax<- resolved StaticField ptr
+    UNSPILL(rPC)
+    testl    %eax,%eax
+    jne      .LOP_SGET_BYTE_finish                 # success, continue
+    jmp      common_exceptionThrown             # no, handle exception
+
+
+/* continuation for OP_SGET_CHAR */
+
+    /*
+     * Go resolve the field
+     */
+.LOP_SGET_CHAR_resolve:
+    GET_GLUE(%ecx)
+    movzwl   2(rPC),%eax                        # eax<- field ref BBBB
+    movl     offGlue_method(%ecx),%ecx          # ecx<- current method
+    EXPORT_PC()                                 # could throw, need to export
+    movl     offMethod_clazz(%ecx),%ecx         # ecx<- method->clazz
+    SPILL(rPC)
+    movl     %eax,OUT_ARG1(%esp)
+    movl     %ecx,OUT_ARG0(%esp)
+    call     dvmResolveStaticField              # eax<- resolved StaticField ptr
+    UNSPILL(rPC)
+    testl    %eax,%eax
+    jne      .LOP_SGET_CHAR_finish                 # success, continue
+    jmp      common_exceptionThrown             # no, handle exception
+
+
+/* continuation for OP_SGET_SHORT */
+
+    /*
+     * Go resolve the field
+     */
+.LOP_SGET_SHORT_resolve:
+    GET_GLUE(%ecx)
+    movzwl   2(rPC),%eax                        # eax<- field ref BBBB
+    movl     offGlue_method(%ecx),%ecx          # ecx<- current method
+    EXPORT_PC()                                 # could throw, need to export
+    movl     offMethod_clazz(%ecx),%ecx         # ecx<- method->clazz
+    SPILL(rPC)
+    movl     %eax,OUT_ARG1(%esp)
+    movl     %ecx,OUT_ARG0(%esp)
+    call     dvmResolveStaticField              # eax<- resolved StaticField ptr
+    UNSPILL(rPC)
+    testl    %eax,%eax
+    jne      .LOP_SGET_SHORT_finish                 # success, continue
+    jmp      common_exceptionThrown             # no, handle exception
+
+
+/* continuation for OP_SPUT */
+
+    /*
+     * Go resolve the field
+     */
+.LOP_SPUT_resolve:
+    GET_GLUE(%ecx)
+    movzwl   2(rPC),%eax                        # eax<- field ref BBBB
+    movl     offGlue_method(%ecx),%ecx          # ecx<- current method
+    EXPORT_PC()                                 # could throw, need to export
+    movl     offMethod_clazz(%ecx),%ecx         # ecx<- method->clazz
+    SPILL(rPC)
+    movl     %eax,OUT_ARG1(%esp)
+    movl     %ecx,OUT_ARG0(%esp)
+    call     dvmResolveStaticField              # eax<- resolved StaticField ptr
+    UNSPILL(rPC)
+    testl    %eax,%eax
+    jne      .LOP_SPUT_finish                 # success, continue
+    jmp      common_exceptionThrown             # no, handle exception
+
+
+/* continuation for OP_SPUT_WIDE */
+
+    /*
+     * Go resolve the field
+     */
+.LOP_SPUT_WIDE_resolve:
+    GET_GLUE(%ecx)
+    movzwl   2(rPC),%eax                        # eax<- field ref BBBB
+    movl     offGlue_method(%ecx),%ecx          # ecx<- current method
+    EXPORT_PC()                                 # could throw, need to export
+    movl     offMethod_clazz(%ecx),%ecx         # ecx<- method->clazz
+    SPILL(rPC)
+    movl     %eax,OUT_ARG1(%esp)
+    movl     %ecx,OUT_ARG0(%esp)
+    call     dvmResolveStaticField              # eax<- resolved StaticField ptr
+    UNSPILL(rPC)
+    testl    %eax,%eax
+    jne      .LOP_SPUT_WIDE_finish                 # success, continue
+    jmp      common_exceptionThrown             # no, handle exception
+
+
+/* continuation for OP_SPUT_OBJECT */
+
+    /*
+     * Go resolve the field
+     */
+.LOP_SPUT_OBJECT_resolve:
+    GET_GLUE(%ecx)
+    movzwl   2(rPC),%eax                        # eax<- field ref BBBB
+    movl     offGlue_method(%ecx),%ecx          # ecx<- current method
+    EXPORT_PC()                                 # could throw, need to export
+    movl     offMethod_clazz(%ecx),%ecx         # ecx<- method->clazz
+    SPILL(rPC)
+    movl     %eax,OUT_ARG1(%esp)
+    movl     %ecx,OUT_ARG0(%esp)
+    call     dvmResolveStaticField              # eax<- resolved StaticField ptr
+    UNSPILL(rPC)
+    testl    %eax,%eax
+    jne      .LOP_SPUT_OBJECT_finish                 # success, continue
+    jmp      common_exceptionThrown             # no, handle exception
+
+
+/* continuation for OP_SPUT_BOOLEAN */
+
+    /*
+     * Go resolve the field
+     */
+.LOP_SPUT_BOOLEAN_resolve:
+    GET_GLUE(%ecx)
+    movzwl   2(rPC),%eax                        # eax<- field ref BBBB
+    movl     offGlue_method(%ecx),%ecx          # ecx<- current method
+    EXPORT_PC()                                 # could throw, need to export
+    movl     offMethod_clazz(%ecx),%ecx         # ecx<- method->clazz
+    SPILL(rPC)
+    movl     %eax,OUT_ARG1(%esp)
+    movl     %ecx,OUT_ARG0(%esp)
+    call     dvmResolveStaticField              # eax<- resolved StaticField ptr
+    UNSPILL(rPC)
+    testl    %eax,%eax
+    jne      .LOP_SPUT_BOOLEAN_finish                 # success, continue
+    jmp      common_exceptionThrown             # no, handle exception
+
+
+/* continuation for OP_SPUT_BYTE */
+
+    /*
+     * Go resolve the field
+     */
+.LOP_SPUT_BYTE_resolve:
+    GET_GLUE(%ecx)
+    movzwl   2(rPC),%eax                        # eax<- field ref BBBB
+    movl     offGlue_method(%ecx),%ecx          # ecx<- current method
+    EXPORT_PC()                                 # could throw, need to export
+    movl     offMethod_clazz(%ecx),%ecx         # ecx<- method->clazz
+    SPILL(rPC)
+    movl     %eax,OUT_ARG1(%esp)
+    movl     %ecx,OUT_ARG0(%esp)
+    call     dvmResolveStaticField              # eax<- resolved StaticField ptr
+    UNSPILL(rPC)
+    testl    %eax,%eax
+    jne      .LOP_SPUT_BYTE_finish                 # success, continue
+    jmp      common_exceptionThrown             # no, handle exception
+
+
+/* continuation for OP_SPUT_CHAR */
+
+    /*
+     * Go resolve the field
+     */
+.LOP_SPUT_CHAR_resolve:
+    GET_GLUE(%ecx)
+    movzwl   2(rPC),%eax                        # eax<- field ref BBBB
+    movl     offGlue_method(%ecx),%ecx          # ecx<- current method
+    EXPORT_PC()                                 # could throw, need to export
+    movl     offMethod_clazz(%ecx),%ecx         # ecx<- method->clazz
+    SPILL(rPC)
+    movl     %eax,OUT_ARG1(%esp)
+    movl     %ecx,OUT_ARG0(%esp)
+    call     dvmResolveStaticField              # eax<- resolved StaticField ptr
+    UNSPILL(rPC)
+    testl    %eax,%eax
+    jne      .LOP_SPUT_CHAR_finish                 # success, continue
+    jmp      common_exceptionThrown             # no, handle exception
+
+
+/* continuation for OP_SPUT_SHORT */
+
+    /*
+     * Go resolve the field
+     */
+.LOP_SPUT_SHORT_resolve:
+    GET_GLUE(%ecx)
+    movzwl   2(rPC),%eax                        # eax<- field ref BBBB
+    movl     offGlue_method(%ecx),%ecx          # ecx<- current method
+    EXPORT_PC()                                 # could throw, need to export
+    movl     offMethod_clazz(%ecx),%ecx         # ecx<- method->clazz
+    SPILL(rPC)
+    movl     %eax,OUT_ARG1(%esp)
+    movl     %ecx,OUT_ARG0(%esp)
+    call     dvmResolveStaticField              # eax<- resolved StaticField ptr
+    UNSPILL(rPC)
+    testl    %eax,%eax
+    jne      .LOP_SPUT_SHORT_finish                 # success, continue
+    jmp      common_exceptionThrown             # no, handle exception
+
+
+/* continuation for OP_INVOKE_VIRTUAL */
+
+
+.LOP_INVOKE_VIRTUAL_more:
+    movl      offMethod_clazz(%eax),%eax  # ecx<- method->clazz
+    movl      %eax,OUT_ARG0(%esp)         # arg0<- clazz
+    movl      $METHOD_VIRTUAL,OUT_ARG2(%esp) # arg2<- flags
+    call      dvmResolveMethod            # eax<- call(clazz, ref, flags)
+    UNSPILL(rPC)
+    testl     %eax,%eax                   # got null?
+    jne       .LOP_INVOKE_VIRTUAL_continue        # no, continue
+    jmp       common_exceptionThrown      # yes, handle exception
+
+    /* At this point:
+     *   eax = resolved base method
+     *   ecx = scratch
+     */
+.LOP_INVOKE_VIRTUAL_continue:
+    movzwl    4(rPC),%ecx               # ecx<- GFED or CCCC
+    .if       (!0)
+    andl      $0xf,%ecx                # ecx<- D (or stays CCCC)
+    .endif
+    GET_VREG(%ecx,%ecx)                 # ecx<- "this"
+    movzwl    offMethod_methodIndex(%eax),%eax  # eax<- baseMethod->methodIndex
+    testl     %ecx,%ecx                 # null this?
+    je        common_errNullObject      # go if so
+    movl      offObject_clazz(%ecx),%ecx  # ecx<- thisPtr->clazz
+    movl      offClassObject_vtable(%ecx),%ecx # ecx<- thisPtr->clazz->vtable
+    movl      (%ecx,%eax,4),%eax        # eax<- vtable[methodIndex]
+    movl      $0,%ecx           # needed for common_invokeOld
+    #jmp       common_invokeMethodNoRange
+    jmp       common_invokeOld
+
+
+/* continuation for OP_INVOKE_SUPER */
+
+    /*
+     * At this point:
+     *  ecx = resolved base method [r0]
+     *  eax = method->clazz [r9]
+     */
+.LOP_INVOKE_SUPER_continue:
+    movl    offClassObject_super(%eax),%eax   # eax<- method->clazz->super
+    movzwl  offMethod_methodIndex(%ecx),%ecx  # ecx<- baseMthod->methodIndex
+    cmpl    offClassObject_vtableCount(%eax),%ecx # compare(methodIndex,vtableCount)
+    jae     .LOP_INVOKE_SUPER_nsm           # method not present in superclass
+    movl    offClassObject_vtable(%eax),%eax   # eax<- ...clazz->super->vtable
+    movl    (%eax,%ecx,4),%eax        # eax<- vtable[methodIndex]
+    movl    $0,%ecx
+    #jmp     common_invokeMethodNoRange
+    jmp     common_invokeOld
+
+    /* At this point:
+     * ecx = null (needs to be resolved base method)
+     * eax = method->clazz
+    */
+.LOP_INVOKE_SUPER_resolve:
+    SPILL_TMP(%eax)                     # method->clazz
+    movl    %eax,OUT_ARG0(%esp)         # arg0<- method->clazz
+    movzwl  2(rPC),%ecx                 # ecx<- BBBB
+    movl    $METHOD_VIRTUAL,OUT_ARG2(%esp)  # arg2<- resolver method type
+    movl    %ecx,OUT_ARG1(%esp)         # arg1<- ref
+    SPILL(rPC)
+    call    dvmResolveMethod            # eax<- call(clazz, ref, flags)
+    UNSPILL(rPC)
+    testl   %eax,%eax                   # got null?
+    movl    %eax,%ecx                   # ecx<- resolved base method
+    UNSPILL_TMP(%eax)                   # restore method->clazz
+    jne     .LOP_INVOKE_SUPER_continue        # good to go - continue
+    jmp     common_exceptionThrown      # handle exception
+
+    /*
+     * Throw a NoSuchMethodError with the method name as the message.
+     *  ecx = resolved base method
+     */
+.LOP_INVOKE_SUPER_nsm:
+    movl    offMethod_name(%ecx),%eax
+    mov     %eax,OUT_ARG1(%esp)
+    jmp     common_errNoSuchMethod
+
+/* continuation for OP_INVOKE_DIRECT */
+
+    /*
+     * On entry:
+     *   TMP_SPILL  <- "this" register
+     * Things a bit ugly on this path, but it's the less
+     * frequent one.  We'll have to do some reloading.
+     */
+.LOP_INVOKE_DIRECT_resolve:
+     SPILL_TMP(%ecx)
+     GET_GLUE(%ecx)
+     UNSPILL(rPC)
+     movl     offGlue_method(%ecx),%ecx  # ecx<- glue->method
+     movzwl   2(rPC),%eax      # reference (BBBB or CCCC)
+     movl     offMethod_clazz(%ecx),%ecx # ecx<- method->clazz
+     movl     $METHOD_DIRECT,OUT_ARG2(%esp)
+     movl     %eax,OUT_ARG1(%esp)
+     movl     %ecx,OUT_ARG0(%esp)
+     call     dvmResolveMethod # eax<- call(clazz, ref, flags)
+     UNSPILL_TMP(%ecx)
+     testl    %eax,%eax
+     jne      .LOP_INVOKE_DIRECT_finish
+     UNSPILL(rPC)
+     jmp      common_exceptionThrown
+
+/* continuation for OP_INVOKE_STATIC */
+
+.LOP_INVOKE_STATIC_continue:
+    movl      $METHOD_STATIC,%eax
+    movl      %eax,OUT_ARG2(%esp)       # arg2<- flags
+    SPILL(rPC)
+    call      dvmResolveMethod          # call(clazz,ref,flags)
+    UNSPILL(rPC)
+    movl      $0,%ecx
+    testl     %eax,%eax                 # got null?
+    #jne       common_invokeMethodNoRange
+    jne       common_invokeOld
+    jmp       common_exceptionThrown
+
+
+/* continuation for OP_INVOKE_INTERFACE */
+
+.LOP_INVOKE_INTERFACE_continue:
+    call       dvmFindInterfaceMethodInCache # eax<- call(class, ref, method, dex)
+    UNSPILL(rPC)
+    testl      %eax,%eax
+    je         common_exceptionThrown
+    movl       $0,%ecx
+    #jmp        common_invokeMethodNoRange
+    jmp        common_invokeOld
+
+
+/* continuation for OP_INVOKE_VIRTUAL_RANGE */
+
+
+.LOP_INVOKE_VIRTUAL_RANGE_more:
+    movl      offMethod_clazz(%eax),%eax  # ecx<- method->clazz
+    movl      %eax,OUT_ARG0(%esp)         # arg0<- clazz
+    movl      $METHOD_VIRTUAL,OUT_ARG2(%esp) # arg2<- flags
+    call      dvmResolveMethod            # eax<- call(clazz, ref, flags)
+    UNSPILL(rPC)
+    testl     %eax,%eax                   # got null?
+    jne       .LOP_INVOKE_VIRTUAL_RANGE_continue        # no, continue
+    jmp       common_exceptionThrown      # yes, handle exception
+
+    /* At this point:
+     *   eax = resolved base method
+     *   ecx = scratch
+     */
+.LOP_INVOKE_VIRTUAL_RANGE_continue:
+    movzwl    4(rPC),%ecx               # ecx<- GFED or CCCC
+    .if       (!1)
+    andl      $0xf,%ecx                # ecx<- D (or stays CCCC)
+    .endif
+    GET_VREG(%ecx,%ecx)                 # ecx<- "this"
+    movzwl    offMethod_methodIndex(%eax),%eax  # eax<- baseMethod->methodIndex
+    testl     %ecx,%ecx                 # null this?
+    je        common_errNullObject      # go if so
+    movl      offObject_clazz(%ecx),%ecx  # ecx<- thisPtr->clazz
+    movl      offClassObject_vtable(%ecx),%ecx # ecx<- thisPtr->clazz->vtable
+    movl      (%ecx,%eax,4),%eax        # eax<- vtable[methodIndex]
+    movl      $1,%ecx           # needed for common_invokeOld
+    #jmp       common_invokeMethodRange
+    jmp       common_invokeOld
+
+
+/* continuation for OP_INVOKE_SUPER_RANGE */
+
+    /*
+     * At this point:
+     *  ecx = resolved base method [r0]
+     *  eax = method->clazz [r9]
+     */
+.LOP_INVOKE_SUPER_RANGE_continue:
+    movl    offClassObject_super(%eax),%eax   # eax<- method->clazz->super
+    movzwl  offMethod_methodIndex(%ecx),%ecx  # ecx<- baseMthod->methodIndex
+    cmpl    offClassObject_vtableCount(%eax),%ecx # compare(methodIndex,vtableCount)
+    jae     .LOP_INVOKE_SUPER_RANGE_nsm           # method not present in superclass
+    movl    offClassObject_vtable(%eax),%eax   # eax<- ...clazz->super->vtable
+    movl    (%eax,%ecx,4),%eax        # eax<- vtable[methodIndex]
+    movl    $1,%ecx
+    #jmp     common_invokeMethodRange
+    jmp     common_invokeOld
+
+    /* At this point:
+     * ecx = null (needs to be resolved base method)
+     * eax = method->clazz
+    */
+.LOP_INVOKE_SUPER_RANGE_resolve:
+    SPILL_TMP(%eax)                     # method->clazz
+    movl    %eax,OUT_ARG0(%esp)         # arg0<- method->clazz
+    movzwl  2(rPC),%ecx                 # ecx<- BBBB
+    movl    $METHOD_VIRTUAL,OUT_ARG2(%esp)  # arg2<- resolver method type
+    movl    %ecx,OUT_ARG1(%esp)         # arg1<- ref
+    SPILL(rPC)
+    call    dvmResolveMethod            # eax<- call(clazz, ref, flags)
+    UNSPILL(rPC)
+    testl   %eax,%eax                   # got null?
+    movl    %eax,%ecx                   # ecx<- resolved base method
+    UNSPILL_TMP(%eax)                   # restore method->clazz
+    jne     .LOP_INVOKE_SUPER_RANGE_continue        # good to go - continue
+    jmp     common_exceptionThrown      # handle exception
+
+    /*
+     * Throw a NoSuchMethodError with the method name as the message.
+     *  ecx = resolved base method
+     */
+.LOP_INVOKE_SUPER_RANGE_nsm:
+    movl    offMethod_name(%ecx),%eax
+    mov     %eax,OUT_ARG1(%esp)
+    jmp     common_errNoSuchMethod
+
+/* continuation for OP_INVOKE_DIRECT_RANGE */
+
+    /*
+     * On entry:
+     *   TMP_SPILL  <- "this" register
+     * Things a bit ugly on this path, but it's the less
+     * frequent one.  We'll have to do some reloading.
+     */
+.LOP_INVOKE_DIRECT_RANGE_resolve:
+     SPILL_TMP(%ecx)
+     GET_GLUE(%ecx)
+     UNSPILL(rPC)
+     movl     offGlue_method(%ecx),%ecx  # ecx<- glue->method
+     movzwl   2(rPC),%eax      # reference (BBBB or CCCC)
+     movl     offMethod_clazz(%ecx),%ecx # ecx<- method->clazz
+     movl     $METHOD_DIRECT,OUT_ARG2(%esp)
+     movl     %eax,OUT_ARG1(%esp)
+     movl     %ecx,OUT_ARG0(%esp)
+     call     dvmResolveMethod # eax<- call(clazz, ref, flags)
+     UNSPILL_TMP(%ecx)
+     testl    %eax,%eax
+     jne      .LOP_INVOKE_DIRECT_RANGE_finish
+     UNSPILL(rPC)
+     jmp      common_exceptionThrown
+
+/* continuation for OP_INVOKE_STATIC_RANGE */
+
+.LOP_INVOKE_STATIC_RANGE_continue:
+    movl      $METHOD_STATIC,%eax
+    movl      %eax,OUT_ARG2(%esp)       # arg2<- flags
+    SPILL(rPC)
+    call      dvmResolveMethod          # call(clazz,ref,flags)
+    UNSPILL(rPC)
+    movl      $1,%ecx
+    testl     %eax,%eax                 # got null?
+    #jne       common_invokeMethodRange
+    jne       common_invokeOld
+    jmp       common_exceptionThrown
+
+
+/* continuation for OP_INVOKE_INTERFACE_RANGE */
+
+.LOP_INVOKE_INTERFACE_RANGE_continue:
+    call       dvmFindInterfaceMethodInCache # eax<- call(class, ref, method, dex)
+    UNSPILL(rPC)
+    testl      %eax,%eax
+    je         common_exceptionThrown
+    movl       $1,%ecx
+    #jmp        common_invokeMethodRange
+    jmp        common_invokeOld
+
+
+/* continuation for OP_FLOAT_TO_INT */
+
+
+.LOP_FLOAT_TO_INT_continue:
+    .if 0
+    movl     $0x80000000,%eax
+    xorl     4(rFP,%ecx,4),%eax
+    orl      (rFP,%ecx,4),%eax
+    .else
+    cmpl     $0x80000000,(rFP,%ecx,4)
+    .endif
+    je       .LOP_FLOAT_TO_INT_special_case # fix up result
+
+.LOP_FLOAT_TO_INT_finish:
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+.LOP_FLOAT_TO_INT_special_case:
+    fnstsw   %ax
+    sahf
+    jp       .LOP_FLOAT_TO_INT_isNaN
+    adcl     $-1,(rFP,%ecx,4)
+    .if 0
+    adcl     $-1,4(rFP,%ecx,4)
+    .endif
+   jmp       .LOP_FLOAT_TO_INT_finish
+.LOP_FLOAT_TO_INT_isNaN:
+    movl      $0,(rFP,%ecx,4)
+    .if 0
+    movl      $0,4(rFP,%ecx,4)
+    .endif
+    jmp       .LOP_FLOAT_TO_INT_finish
+
+/* continuation for OP_FLOAT_TO_LONG */
+
+
+.LOP_FLOAT_TO_LONG_continue:
+    .if 1
+    movl     $0x80000000,%eax
+    xorl     4(rFP,%ecx,4),%eax
+    orl      (rFP,%ecx,4),%eax
+    .else
+    cmpl     $0x80000000,(rFP,%ecx,4)
+    .endif
+    je       .LOP_FLOAT_TO_LONG_special_case # fix up result
+
+.LOP_FLOAT_TO_LONG_finish:
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+.LOP_FLOAT_TO_LONG_special_case:
+    fnstsw   %ax
+    sahf
+    jp       .LOP_FLOAT_TO_LONG_isNaN
+    adcl     $-1,(rFP,%ecx,4)
+    .if 1
+    adcl     $-1,4(rFP,%ecx,4)
+    .endif
+   jmp       .LOP_FLOAT_TO_LONG_finish
+.LOP_FLOAT_TO_LONG_isNaN:
+    movl      $0,(rFP,%ecx,4)
+    .if 1
+    movl      $0,4(rFP,%ecx,4)
+    .endif
+    jmp       .LOP_FLOAT_TO_LONG_finish
+
+/* continuation for OP_DOUBLE_TO_INT */
+
+
+.LOP_DOUBLE_TO_INT_continue:
+    .if 0
+    movl     $0x80000000,%eax
+    xorl     4(rFP,%ecx,4),%eax
+    orl      (rFP,%ecx,4),%eax
+    .else
+    cmpl     $0x80000000,(rFP,%ecx,4)
+    .endif
+    je       .LOP_DOUBLE_TO_INT_special_case # fix up result
+
+.LOP_DOUBLE_TO_INT_finish:
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+.LOP_DOUBLE_TO_INT_special_case:
+    fnstsw   %ax
+    sahf
+    jp       .LOP_DOUBLE_TO_INT_isNaN
+    adcl     $-1,(rFP,%ecx,4)
+    .if 0
+    adcl     $-1,4(rFP,%ecx,4)
+    .endif
+   jmp       .LOP_DOUBLE_TO_INT_finish
+.LOP_DOUBLE_TO_INT_isNaN:
+    movl      $0,(rFP,%ecx,4)
+    .if 0
+    movl      $0,4(rFP,%ecx,4)
+    .endif
+    jmp       .LOP_DOUBLE_TO_INT_finish
+
+/* continuation for OP_DOUBLE_TO_LONG */
+
+
+.LOP_DOUBLE_TO_LONG_continue:
+    .if 1
+    movl     $0x80000000,%eax
+    xorl     4(rFP,%ecx,4),%eax
+    orl      (rFP,%ecx,4),%eax
+    .else
+    cmpl     $0x80000000,(rFP,%ecx,4)
+    .endif
+    je       .LOP_DOUBLE_TO_LONG_special_case # fix up result
+
+.LOP_DOUBLE_TO_LONG_finish:
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+.LOP_DOUBLE_TO_LONG_special_case:
+    fnstsw   %ax
+    sahf
+    jp       .LOP_DOUBLE_TO_LONG_isNaN
+    adcl     $-1,(rFP,%ecx,4)
+    .if 1
+    adcl     $-1,4(rFP,%ecx,4)
+    .endif
+   jmp       .LOP_DOUBLE_TO_LONG_finish
+.LOP_DOUBLE_TO_LONG_isNaN:
+    movl      $0,(rFP,%ecx,4)
+    .if 1
+    movl      $0,4(rFP,%ecx,4)
+    .endif
+    jmp       .LOP_DOUBLE_TO_LONG_finish
+
+/* continuation for OP_DIV_INT */
+.LOP_DIV_INT_continue_div:
+    cltd
+    idivl   %ecx
+.LOP_DIV_INT_finish_div:
+    movzbl   rINST_HI,%ecx         # ecl<- AA
+    SET_VREG(%eax,%ecx)
+    UNSPILL(rPC)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* continuation for OP_REM_INT */
+.LOP_REM_INT_continue_div:
+    cltd
+    idivl   %ecx
+.LOP_REM_INT_finish_div:
+    movzbl   rINST_HI,%ecx         # ecl<- AA
+    SET_VREG(%edx,%ecx)
+    UNSPILL(rPC)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* continuation for OP_MUL_LONG */
+
+.LOP_MUL_LONG_continue:
+    leal      (%ecx,%edx),%edx    # full result now in %edx:%eax
+    movzbl    rINST_HI,%ecx       # ecx<- A
+    movl      %edx,4(rFP,%ecx,4)  # v[B+1]<- %edx
+    UNSPILL(rPC)                  # restore rPC/%edx
+    FETCH_INST_WORD(2)
+    UNSPILL(rIBASE)
+    movl      %eax,(rFP,%ecx,4)   # v[B]<- %eax
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+
+/* continuation for OP_DIV_LONG */
+
+.LOP_DIV_LONG_continue:
+    call     __divdi3
+.LOP_DIV_LONG_finish:
+    movzbl   rINST_HI,%ecx
+    SET_VREG_WORD(rPC,%ecx,1)
+    UNSPILL(rPC)
+    SET_VREG_WORD(%eax,%ecx,0)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+.LOP_DIV_LONG_check_zero:
+    testl   rPC,rPC
+    jne     .LOP_DIV_LONG_notSpecial
+    UNSPILL(rPC)
+    jmp     common_errDivideByZero
+.LOP_DIV_LONG_check_neg1:
+    testl   rPC,%eax
+    jne     .LOP_DIV_LONG_notSpecial
+    GET_VREG_WORD(rPC,%ecx,0)
+    GET_VREG_WORD(%ecx,%ecx,1)
+    testl    rPC,rPC
+    jne      .LOP_DIV_LONG_notSpecial1
+    cmpl     $0x80000000,%ecx
+    jne      .LOP_DIV_LONG_notSpecial1
+    /* minint / -1, return minint on div, 0 on rem */
+    xorl     %eax,%eax
+    movl     $0x80000000,%edx
+    jmp      .LOP_DIV_LONG_finish
+
+
+
+
+/* continuation for OP_REM_LONG */
+
+.LOP_REM_LONG_continue:
+    call     __moddi3
+.LOP_REM_LONG_finish:
+    movzbl   rINST_HI,%ecx
+    SET_VREG_WORD(rPC,%ecx,1)
+    UNSPILL(rPC)
+    SET_VREG_WORD(%eax,%ecx,0)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+.LOP_REM_LONG_check_zero:
+    testl   rPC,rPC
+    jne     .LOP_REM_LONG_notSpecial
+    UNSPILL(rPC)
+    jmp     common_errDivideByZero
+.LOP_REM_LONG_check_neg1:
+    testl   rPC,%eax
+    jne     .LOP_REM_LONG_notSpecial
+    GET_VREG_WORD(rPC,%ecx,0)
+    GET_VREG_WORD(%ecx,%ecx,1)
+    testl    rPC,rPC
+    jne      .LOP_REM_LONG_notSpecial1
+    cmpl     $0x80000000,%ecx
+    jne      .LOP_REM_LONG_notSpecial1
+    /* minint / -1, return minint on div, 0 on rem */
+    xorl     %eax,%eax
+    movl     $0,%edx
+    jmp      .LOP_REM_LONG_finish
+
+
+
+
+/* continuation for OP_SHL_LONG */
+
+.LOP_SHL_LONG_finish:
+    SET_VREG_WORD(%eax,%ecx,0)         # v[AA+0]<- %eax
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* continuation for OP_SHR_LONG */
+
+
+.LOP_SHR_LONG_finish:
+    SET_VREG_WORD(%eax,%ecx,0)         # v[AA+0]<- eax
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* continuation for OP_USHR_LONG */
+
+
+.LOP_USHR_LONG_finish:
+    SET_VREG_WORD(%eax,%ecx,0)        # v[BB+0]<- eax
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* continuation for OP_DIV_INT_2ADDR */
+.LOP_DIV_INT_2ADDR_continue_div2addr:
+    cltd
+    idivl   %ecx
+.LOP_DIV_INT_2ADDR_finish_div2addr:
+    SET_VREG(%eax,rINST_FULL)
+    UNSPILL(rPC)
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+/* continuation for OP_REM_INT_2ADDR */
+.LOP_REM_INT_2ADDR_continue_div2addr:
+    cltd
+    idivl   %ecx
+.LOP_REM_INT_2ADDR_finish_div2addr:
+    SET_VREG(%edx,rINST_FULL)
+    UNSPILL(rPC)
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+/* continuation for OP_MUL_LONG_2ADDR */
+
+.LOP_MUL_LONG_2ADDR_continue:
+    leal      (%ecx,%edx),%edx    # full result now in %edx:%eax
+    movl      %edx,4(rIBASE)      # v[A+1]<- %edx
+    UNSPILL(rPC)                  # restore rPC/%edx
+    FETCH_INST_WORD(1)
+    movl      %eax,(rIBASE)       # v[A]<- %eax
+    UNSPILL(rFP)
+    UNSPILL(rIBASE)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+
+/* continuation for OP_DIV_LONG_2ADDR */
+
+.LOP_DIV_LONG_2ADDR_continue:
+    movl     %eax,OUT_ARG3(%esp)
+    movl     rPC,OUT_ARG0(%esp)
+    movl     %ecx,OUT_ARG1(%esp)
+    call     __divdi3
+.LOP_DIV_LONG_2ADDR_finish:
+    movl     rINST_FULL,%ecx
+    SET_VREG_WORD(rPC,%ecx,1)
+    UNSPILL(rPC)
+    SET_VREG_WORD(%eax,%ecx,0)
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+.LOP_DIV_LONG_2ADDR_check_zero:
+    testl   rPC,rPC
+    jne     .LOP_DIV_LONG_2ADDR_notSpecial
+    UNSPILL(rPC)
+    jmp     common_errDivideByZero
+.LOP_DIV_LONG_2ADDR_check_neg1:
+    testl   rPC,%eax
+    jne     .LOP_DIV_LONG_2ADDR_notSpecial
+    GET_VREG_WORD(rPC,rINST_FULL,0)
+    GET_VREG_WORD(%ecx,rINST_FULL,1)
+    testl    rPC,rPC
+    jne      .LOP_DIV_LONG_2ADDR_notSpecial1
+    cmpl     $0x80000000,%ecx
+    jne      .LOP_DIV_LONG_2ADDR_notSpecial1
+    /* minint / -1, return minint on div, 0 on rem */
+    xorl     %eax,%eax
+    movl     $0x80000000,%edx
+    jmp      .LOP_DIV_LONG_2ADDR_finish
+
+
+/* continuation for OP_REM_LONG_2ADDR */
+
+.LOP_REM_LONG_2ADDR_continue:
+    movl     %eax,OUT_ARG3(%esp)
+    movl     rPC,OUT_ARG0(%esp)
+    movl     %ecx,OUT_ARG1(%esp)
+    call     __moddi3
+.LOP_REM_LONG_2ADDR_finish:
+    movl     rINST_FULL,%ecx
+    SET_VREG_WORD(rPC,%ecx,1)
+    UNSPILL(rPC)
+    SET_VREG_WORD(%eax,%ecx,0)
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+.LOP_REM_LONG_2ADDR_check_zero:
+    testl   rPC,rPC
+    jne     .LOP_REM_LONG_2ADDR_notSpecial
+    UNSPILL(rPC)
+    jmp     common_errDivideByZero
+.LOP_REM_LONG_2ADDR_check_neg1:
+    testl   rPC,%eax
+    jne     .LOP_REM_LONG_2ADDR_notSpecial
+    GET_VREG_WORD(rPC,rINST_FULL,0)
+    GET_VREG_WORD(%ecx,rINST_FULL,1)
+    testl    rPC,rPC
+    jne      .LOP_REM_LONG_2ADDR_notSpecial1
+    cmpl     $0x80000000,%ecx
+    jne      .LOP_REM_LONG_2ADDR_notSpecial1
+    /* minint / -1, return minint on div, 0 on rem */
+    xorl     %eax,%eax
+    movl     $0,%edx
+    jmp      .LOP_REM_LONG_2ADDR_finish
+
+
+/* continuation for OP_SHL_LONG_2ADDR */
+
+
+.LOP_SHL_LONG_2ADDR_finish:
+    SET_VREG_WORD(%eax,rINST_FULL,0)  # v[AA+0]<- eax
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+/* continuation for OP_SHR_LONG_2ADDR */
+
+
+.LOP_SHR_LONG_2ADDR_finish:
+    SET_VREG_WORD(%eax,rINST_FULL,0)  # v[AA+0]<- eax
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+/* continuation for OP_USHR_LONG_2ADDR */
+
+
+.LOP_USHR_LONG_2ADDR_finish:
+    SET_VREG_WORD(%eax,rINST_FULL,0)   # v[AA+0]<- eax
+    FETCH_INST_WORD(1)
+    ADVANCE_PC(1)
+    GOTO_NEXT
+
+/* continuation for OP_DIV_INT_LIT16 */
+.LOP_DIV_INT_LIT16_continue_div:
+    cltd
+    idivl   %ecx
+.LOP_DIV_INT_LIT16_finish_div:
+    SET_VREG(%eax,rINST_FULL)
+    UNSPILL(rPC)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* continuation for OP_REM_INT_LIT16 */
+.LOP_REM_INT_LIT16_continue_div:
+    cltd
+    idivl   %ecx
+.LOP_REM_INT_LIT16_finish_div:
+    SET_VREG(%edx,rINST_FULL)
+    UNSPILL(rPC)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* continuation for OP_DIV_INT_LIT8 */
+.LOP_DIV_INT_LIT8_continue_div:
+    cltd
+    idivl   %ecx
+.LOP_DIV_INT_LIT8_finish_div:
+    SET_VREG(%eax,rINST_FULL)
+    UNSPILL(rPC)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* continuation for OP_REM_INT_LIT8 */
+.LOP_REM_INT_LIT8_continue_div:
+    cltd
+    idivl   %ecx
+.LOP_REM_INT_LIT8_finish_div:
+    SET_VREG(%edx,rINST_FULL)
+    UNSPILL(rPC)
+    FETCH_INST_WORD(2)
+    ADVANCE_PC(2)
+    GOTO_NEXT
+
+/* continuation for OP_EXECUTE_INLINE */
+
+.LOP_EXECUTE_INLINE_continue:
+    /*
+     * Extract args, call function.
+     *  ecx = #of args (0-4)
+     *  eax = call index
+     *  @esp = return addr
+     *  esp is -4 from normal
+     *
+     *  Go ahead and load all 4 args, even if not used.
+     */
+    movzwl    4(rPC),rPC
+
+    movl      $0xf,%ecx
+    andl      rPC,%ecx
+    GET_VREG(%ecx,%ecx)
+    sarl      $4,rPC
+    movl      %ecx,4+OUT_ARG0(%esp)
+
+    movl      $0xf,%ecx
+    andl      rPC,%ecx
+    GET_VREG(%ecx,%ecx)
+    sarl      $4,rPC
+    movl      %ecx,4+OUT_ARG1(%esp)
+
+    movl      $0xf,%ecx
+    andl      rPC,%ecx
+    GET_VREG(%ecx,%ecx)
+    sarl      $4,rPC
+    movl      %ecx,4+OUT_ARG2(%esp)
+
+    movl      $0xf,%ecx
+    andl      rPC,%ecx
+    GET_VREG(%ecx,%ecx)
+    sarl      $4,rPC
+    movl      %ecx,4+OUT_ARG3(%esp)
+
+    sall      $4,%eax      # index *= sizeof(table entry)
+    jmp       *gDvmInlineOpsTable(%eax)
+    # will return to caller of .LOP_EXECUTE_INLINE_continue
+
+
+    .size   dvmAsmSisterStart, .-dvmAsmSisterStart
+    .global dvmAsmSisterEnd
+dvmAsmSisterEnd:
+
+/* File: x86/entry.S */
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+
+    .text
+    .global dvmMterpStdRun
+    .type   dvmMterpStdRun, %function
+/*
+ * bool dvmMterpStdRun(MterpGlue* glue)
+ *
+ * Interpreter entry point.  Returns changeInterp.
+ *
+ */
+dvmMterpStdRun:
+    push    %ebp
+    movl    %esp,%ebp
+    push    %edi
+    push    %esi
+    push    %ebx
+
+/* at this point, stack is misaligned by 1 word
+   We're allocating spill space for 6 words, plus
+   outgoing argument (5 words) and local variables
+   (4 words) - 15 words or 60 bytes total. See
+   diagram in header.S
+*/
+    subl   $60,%esp
+
+/* Set up "named" registers */
+    movl    IN_ARG0(%ebp),%ecx
+    movl    %ecx,rGLUE_SPILL(%ebp)
+    LOAD_PC_FROM_GLUE(%ecx)
+    LOAD_FP_FROM_GLUE(%ecx)
+    movl    $dvmAsmInstructionStart,rIBASE
+
+/* Remember %esp for future "longjmp" */
+    movl    %esp,offGlue_bailPtr(%ecx)
+
+/* How to start? */
+    movb    offGlue_entryPoint(%ecx),%al
+
+/* Normal start? */
+    cmpb    $kInterpEntryInstr,%al
+    jne     .Lnot_instr
+
+   /* Normal case: start executing the instruction at rPC */
+    FETCH_INST()
+    GOTO_NEXT
+
+.Lnot_instr:
+    /* Reset to normal case */
+    movb   $kInterpEntryInstr,offGlue_entryPoint(%ecx)
+    cmpb   $kInterpEntryReturn,%al
+    je     common_returnFromMethod
+    cmpb   $kInterpEntryThrow,%al
+    je     common_exceptionThrown
+    movzx  %al,%eax
+    movl   %eax,OUT_ARG1(%esp)
+    movl   $.LstrBadEntryPoint,OUT_ARG0(%esp)
+    call   printf
+    call   dvmAbort
+    /* Not reached */
+
+
+    .global dvmMterpStdBail
+    .type   dvmMterpStdBail, %function
+/*
+ * void dvmMterpStdBail(MterpGlue* glue, bool changeInterp)
+ *
+ * Restore the stack pointer and PC from the save point established on entry.
+ * This is essentially the same as a longjmp, but should be cheaper.  The
+ * last instruction causes us to return to whoever called dvmMterpStdRun.
+ *
+ * We're not going to build a standard frame here, so the arg accesses will
+ * look a little strange.
+ *
+ * On entry:
+ *  esp+4 (arg0)  MterpGlue* glue
+ *  esp+8 (arg1)  bool changeInterp
+ */
+dvmMterpStdBail:
+    movl    4(%esp),%ecx                 # grab glue
+    movl    8(%esp),%eax                 # changeInterp to return reg
+    movl    offGlue_bailPtr(%ecx),%esp   # Stack back to normal
+    addl    $60,%esp                    # Strip dvmMterpStdRun's frame
+    pop     %ebx
+    pop     %esi
+    pop     %edi
+    pop     %ebp
+    ret                                  # return to dvmMterpStdRun's caller
+
+
+/*
+ * Strings
+ */
+    .section    .rodata
+.LstrBadEntryPoint:
+    .asciz  "Bad entry point %d\n"
+
+/* File: x86/footer.S */
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ * Common subroutines and data.
+ */
+
+/*
+ * Common code when a backwards branch is taken
+ *
+ * On entry:
+ *   ebx (a.k.a. rINST_FULL) -> PC adjustment in 16-bit words
+ */
+common_backwardBranch:
+    GET_GLUE(%ecx)
+    call   common_periodicChecks      # Note: expects rPC to be preserved
+    ADVANCE_PC_INDEXED(rINST_FULL)
+    FETCH_INST()
+    GOTO_NEXT
+
+/*
+ * Common invoke code (old-style).
+ * TUNING:  Rewrite along lines of new armv5 code?
+ *
+ * On entry:
+ *   eax = Method* methodToCall
+ *   ecx = bool methodCallRange
+ *   rINST trashed, must reload
+ */
+common_invokeOld:
+    movl     %ecx,OUT_ARG1(%esp)     # arg1<- methodCallRange
+    GET_GLUE(%ecx)
+    movzwl  (rPC),rINST_FULL         # recover rINST
+    movl     %eax,OUT_ARG2(%esp)     # arg2<- method
+    movzwl   4(rPC),%eax             # eax<- GFED or CCCC
+    SAVE_PC_TO_GLUE(%ecx)
+    SAVE_FP_TO_GLUE(%ecx)
+    movzbl   rINST_HI,rINST_FULL
+    movl     rINST_FULL,OUT_ARG3(%esp)# arg3<- AA
+    movl     %ecx,OUT_ARG0(%esp)     # arg0<- GLUE
+    movl     %eax,OUT_ARG4(%esp)     # arg4<- GFED/CCCC
+    call     dvmMterp_invokeMethod
+    jmp      common_resumeAfterGlueCall
+
+
+/*
+ * Do we need the thread to be suspended or have debugger/profiling activity?
+ *
+ * On entry:
+ *   ebx  -> PC adjustment in 16-bit words (must be preserved)
+ *   ecx  -> GLUE pointer
+ *
+ * Note: A call will normally kill %eax, rPC/%edx and %ecx.  To
+ *       streamline the normal case, this routine will preserve rPC and
+ *       %ecx in addition to the normal caller save regs.  The save/restore
+ *       is a bit ugly, but will happen in the relatively uncommon path.
+ * TUNING: Might be worthwhile to inline this.
+ * TODO: Basic-block style Jit will need a hook here as well.  Fold it into
+ *       the suspendCount check so we can get both in 1 shot.
+ */
+common_periodicChecks:
+    movl    offGlue_pSelfSuspendCount(%ecx),%eax    # eax <- &suspendCount
+    cmpl    $0,(%eax)
+    jne     1f
+
+#if defined(WITH_DEBUGGER) || defined(WITH_PROFILER)
+#if defined(WITH_DEBUGGER)
+    movl   offGlue_pDebuggerActive(%ecx),%eax      # eax <- &DebuggerActive
+#endif
+#if defined(WITH_PROFILER)
+    movl   offGlue_pActiveProfilers(%ecx),%ecx     # ecx <- &ActiveProfilers
+#endif
+#if defined(WITH_DEBUGGER) && defined(WITH_PROFILER)
+    movzbl (%eax),%eax             # eax <- debuggerActive (boolean)
+    orl    (%ecx),%eax             # eax <- debuggerActive || activeProfilers
+#elif defined(WITH_DEBUGGER)
+    movzbl (%eax),%eax             # eax <- debuggerActive (boolean)
+#elif defined(WITH_PROFILER)
+    movl   (%ecx),%eax             # eax <= activeProfilers
+#endif
+    GET_GLUE(%ecx)                 # restore rGLUE
+    testl  %eax,%eax
+    jne    3f                      # one or both active - switch interp
+#endif
+
+    ret
+
+    /* Check for suspend */
+1:
+    /*  At this point, the return pointer to the caller of
+     *  common_periodicChecks is on the top of stack.  We need to preserve
+     *  rPC(edx) and GLUE(ecx).  We'll spill rPC, and reload GLUE.
+     *  The outgoing profile is:
+     *      bool dvmCheckSuspendPending(Thread* self)
+     *  Because we reached here via a call, go ahead and build a new frame.
+     */
+    movl    offGlue_self(%ecx),%eax      # eax<- glue->self
+    SPILL(rPC)                      # save edx
+    push    %ebp
+    movl    %esp,%ebp
+    subl    $24,%esp
+    movl    %eax,OUT_ARG0(%esp)
+    call    dvmCheckSuspendPending
+    addl    $24,%esp
+    pop     %ebp
+    UNSPILL(rPC)
+    GET_GLUE(%ecx)
+    ret
+
+    /* Switch interpreters */
+    /* Note: %ebx contains the 16-bit word offset to be applied to rPC to
+     * "complete" the interpretation of backwards branches.  In effect, we
+     * are completing the interpretation of the branch instruction here,
+     * and the new interpreter will resume interpretation at the branch
+     * target. However, a switch request recognized during the handling
+     * of a return from method instruction results in an immediate abort,
+     * and the new interpreter will resume by re-interpreting the return
+     * instruction.
+     */
+3:
+    leal    (rPC,%ebx,2),rPC       # adjust pc to show target
+    GET_GLUE(%ecx)                 # bail expect GLUE already loaded
+    movl    $1,rINST_FULL         # set changeInterp to true
+    jmp     common_gotoBail
+
+
+/*
+ * Common code for handling a return instruction
+ */
+common_returnFromMethod:
+    GET_GLUE(%ecx)
+    /* Set entry mode in case we bail */
+    movb    $kInterpEntryReturn,offGlue_entryPoint(%ecx)
+    xorl    rINST_FULL,rINST_FULL   # zero offset in case we switch interps
+    call    common_periodicChecks   # Note: expects %ecx to be preserved
+
+    SAVEAREA_FROM_FP(%eax,rFP)                    # eax<- saveArea (old)
+    movl    offStackSaveArea_prevFrame(%eax),rFP  # rFP<- prevFrame
+    movl    (offStackSaveArea_method-sizeofStackSaveArea)(rFP),rINST_FULL
+    cmpl    $0,rINST_FULL                        # break?
+    je      common_gotoBail    # break frame, bail out completely
+
+    movl    offStackSaveArea_savedPc(%eax),rPC    # pc<- saveArea->savedPC
+    movl    offGlue_self(%ecx),%eax               # eax<- self
+    movl    rINST_FULL,offGlue_method(%ecx)  # glue->method = newSave->meethod
+    movl    rFP,offThread_curFrame(%eax)     # self->curFrame = fp
+    movl    offMethod_clazz(rINST_FULL),%eax # eax<- method->clazz
+    FETCH_INST_WORD(3)
+    movl    offClassObject_pDvmDex(%eax),%eax # eax<- method->clazz->pDvmDex
+    ADVANCE_PC(3)
+    movl    %eax,offGlue_methodClassDex(%ecx)
+    /* not bailing - restore entry mode to default */
+    movb    $kInterpEntryInstr,offGlue_entryPoint(%ecx)
+    GOTO_NEXT
+
+/*
+ * Prepare to strip the current frame and "longjump" back to caller of
+ * dvmMterpStdRun.
+ *
+ * on entry:
+ *    rINST_FULL holds changeInterp
+ *    ecx holds glue pointer
+ *
+ * expected profile: dvmMterpStdBail(MterpGlue *glue, bool changeInterp)
+ */
+common_gotoBail:
+    SAVE_PC_TO_GLUE(%ecx)                # export state to glue
+    SAVE_FP_TO_GLUE(%ecx)
+    movl   %ecx,OUT_ARG0(%esp)           # glue in arg0
+    movl   rINST_FULL,OUT_ARG1(%esp)     # changeInterp in arg1
+    call    dvmMterpStdBail              # bail out....
+
+
+/*
+ * After returning from a "glued" function, pull out the updated values
+ * and start executing at the next instruction.
+ */
+ common_resumeAfterGlueCall:
+     GET_GLUE(%ecx)
+     LOAD_PC_FROM_GLUE(%ecx)
+     LOAD_FP_FROM_GLUE(%ecx)
+     FETCH_INST()
+     GOTO_NEXT
+
+/*
+ * Integer divide or mod by zero
+ */
+common_errDivideByZero:
+    EXPORT_PC()
+    movl    $.LstrArithmeticException,%eax
+    movl    %eax,OUT_ARG0(%esp)
+    movl    $.LstrDivideByZero,%eax
+    movl    %eax,OUT_ARG1(%esp)
+    SPILL(rPC)
+    call    dvmThrowException
+    UNSPILL(rPC)
+    jmp     common_exceptionThrown
+
+/*
+ * Attempt to allocate an array with a negative size.
+ */
+common_errNegativeArraySize:
+    EXPORT_PC()
+    movl    $.LstrNegativeArraySizeException,%eax
+    movl    %eax,OUT_ARG0(%esp)
+    xorl    %eax,%eax
+    movl    %eax,OUT_ARG1(%esp)
+    SPILL(rPC)
+    call    dvmThrowException
+    UNSPILL(rPC)
+    jmp     common_exceptionThrown
+
+/*
+ * Attempt to allocate an array with a negative size.
+ */
+common_errNoSuchMethod:
+
+    EXPORT_PC()
+    movl    $.LstrNoSuchMethodError,%eax
+    movl    %eax,OUT_ARG0(%esp)
+    xorl    %eax,%eax
+    movl    %eax,OUT_ARG1(%esp)
+    SPILL(rPC)
+    call    dvmThrowException
+    UNSPILL(rPC)
+    jmp     common_exceptionThrown
+
+/*
+ * Hit a null object when we weren't expecting one.  Export the PC, throw a
+ * NullPointerException and goto the exception processing code.
+ */
+common_errNullObject:
+    EXPORT_PC()
+    movl    $.LstrNullPointerException,%eax
+    movl    %eax,OUT_ARG0(%esp)
+    xorl    %eax,%eax
+    movl    %eax,OUT_ARG1(%esp)
+    SPILL(rPC)
+    call    dvmThrowException
+    UNSPILL(rPC)
+    jmp     common_exceptionThrown
+
+/*
+ * Array index exceeds max.
+ */
+common_errArrayIndex:
+    EXPORT_PC()
+    movl    $.LstrArrayIndexException,%eax
+    movl    %eax,OUT_ARG0(%esp)
+    xorl    %eax,%eax
+    movl    %eax,OUT_ARG1(%esp)
+    SPILL(rPC)
+    call    dvmThrowException
+    UNSPILL(rPC)
+    jmp     common_exceptionThrown
+/*
+ * Invalid array value.
+ */
+common_errArrayStore:
+    EXPORT_PC()
+    movl    $.LstrArrayStoreException,%eax
+    movl    %eax,OUT_ARG0(%esp)
+    xorl    %eax,%eax
+    movl    %eax,OUT_ARG1(%esp)
+    SPILL(rPC)
+    call    dvmThrowException
+    UNSPILL(rPC)
+    jmp     common_exceptionThrown
+
+/*
+ * Somebody has thrown an exception.  Handle it.
+ *
+ * If the exception processing code returns to us (instead of falling
+ * out of the interpreter), continue with whatever the next instruction
+ * now happens to be.
+ *
+ * This does not return.
+ */
+common_exceptionThrown:
+    GET_GLUE(%ecx)
+    SAVE_PC_TO_GLUE(%ecx)
+    SAVE_FP_TO_GLUE(%ecx)
+    movl    %ecx,OUT_ARG0(%esp)
+    call    dvmMterp_exceptionThrown
+    jmp     common_resumeAfterGlueCall
+
+common_abort:
+    movl    $0xdeadf00d,%eax
+    call     *%eax
+
+
+/*
+ * Strings
+ */
+
+    .section     .rodata
+.LstrNullPointerException:
+    .asciz    "Ljava/lang/NullPointerException;"
+.LstrArithmeticException:
+    .asciz  "Ljava/lang/ArithmeticException;"
+.LstrDivideByZero:
+    .asciz  "divide by zero"
+.LstrArrayIndexException:
+    .asciz  "Ljava/lang/ArrayIndexOutOfBoundsException;"
+.LstrArrayStoreException:
+    .asciz  "Ljava/lang/ArrayStoreException;"
+.LstrNegativeArraySizeException:
+    .asciz  "Ljava/lang/NegativeArraySizeException;"
+.LstrInstantiationError:
+    .asciz  "Ljava/lang/InstantiationError;"
+.LstrClassCastException:
+    .asciz  "Ljava/lang/ClassCastException;"
+.LstrNoSuchMethodError:
+    .asciz  "Ljava/lang/NoSuchMethodError;"
+.LstrInternalError:
+    .asciz  "Ljava/lang/InternalError;"
+.LstrFilledNewArrayNotImpl:
+    .asciz  "filled-new-array only implemented for 'int'"
+
+
diff --git a/vm/mterp/out/InterpC-allstubs.c b/vm/mterp/out/InterpC-allstubs.c
new file mode 100644
index 0000000..635a873
--- /dev/null
+++ b/vm/mterp/out/InterpC-allstubs.c
@@ -0,0 +1,3969 @@
+/*
+ * This file was generated automatically by gen-mterp.py for 'allstubs'.
+ *
+ * --> DO NOT EDIT <--
+ */
+
+/* File: c/header.c */
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/* common includes */
+#include "Dalvik.h"
+#include "interp/InterpDefs.h"
+#include "mterp/Mterp.h"
+#include <math.h>                   // needed for fmod, fmodf
+
+/*
+ * Configuration defines.  These affect the C implementations, i.e. the
+ * portable interpreter(s) and C stubs.
+ *
+ * Some defines are controlled by the Makefile, e.g.:
+ *   WITH_PROFILER
+ *   WITH_DEBUGGER
+ *   WITH_INSTR_CHECKS
+ *   WITH_TRACKREF_CHECKS
+ *   EASY_GDB
+ *   NDEBUG
+ *
+ * If THREADED_INTERP is not defined, we use a classic "while true / switch"
+ * interpreter.  If it is defined, then the tail end of each instruction
+ * handler fetches the next instruction and jumps directly to the handler.
+ * This increases the size of the "Std" interpreter by about 10%, but
+ * provides a speedup of about the same magnitude.
+ *
+ * There's a "hybrid" approach that uses a goto table instead of a switch
+ * statement, avoiding the "is the opcode in range" tests required for switch.
+ * The performance is close to the threaded version, and without the 10%
+ * size increase, but the benchmark results are off enough that it's not
+ * worth adding as a third option.
+ */
+#define THREADED_INTERP             /* threaded vs. while-loop interpreter */
+
+#ifdef WITH_INSTR_CHECKS            /* instruction-level paranoia */
+# define CHECK_BRANCH_OFFSETS
+# define CHECK_REGISTER_INDICES
+#endif
+
+/*
+ * ARM EABI requires 64-bit alignment for access to 64-bit data types.  We
+ * can't just use pointers to copy 64-bit values out of our interpreted
+ * register set, because gcc will generate ldrd/strd.
+ *
+ * The __UNION version copies data in and out of a union.  The __MEMCPY
+ * version uses a memcpy() call to do the transfer; gcc is smart enough to
+ * not actually call memcpy().  The __UNION version is very bad on ARM;
+ * it only uses one more instruction than __MEMCPY, but for some reason
+ * gcc thinks it needs separate storage for every instance of the union.
+ * On top of that, it feels the need to zero them out at the start of the
+ * method.  Net result is we zero out ~700 bytes of stack space at the top
+ * of the interpreter using ARM STM instructions.
+ */
+#if defined(__ARM_EABI__)
+//# define NO_UNALIGN_64__UNION
+# define NO_UNALIGN_64__MEMCPY
+#endif
+
+//#define LOG_INSTR                   /* verbose debugging */
+/* set and adjust ANDROID_LOG_TAGS='*:i jdwp:i dalvikvm:i dalvikvmi:i' */
+
+/*
+ * Keep a tally of accesses to fields.  Currently only works if full DEX
+ * optimization is disabled.
+ */
+#ifdef PROFILE_FIELD_ACCESS
+# define UPDATE_FIELD_GET(_field) { (_field)->gets++; }
+# define UPDATE_FIELD_PUT(_field) { (_field)->puts++; }
+#else
+# define UPDATE_FIELD_GET(_field) ((void)0)
+# define UPDATE_FIELD_PUT(_field) ((void)0)
+#endif
+
+/*
+ * Adjust the program counter.  "_offset" is a signed int, in 16-bit units.
+ *
+ * Assumes the existence of "const u2* pc" and "const u2* curMethod->insns".
+ *
+ * We don't advance the program counter until we finish an instruction or
+ * branch, because we do want to have to unroll the PC if there's an
+ * exception.
+ */
+#ifdef CHECK_BRANCH_OFFSETS
+# define ADJUST_PC(_offset) do {                                            \
+        int myoff = _offset;        /* deref only once */                   \
+        if (pc + myoff < curMethod->insns ||                                \
+            pc + myoff >= curMethod->insns + dvmGetMethodInsnsSize(curMethod)) \
+        {                                                                   \
+            char* desc;                                                     \
+            desc = dexProtoCopyMethodDescriptor(&curMethod->prototype);     \
+            LOGE("Invalid branch %d at 0x%04x in %s.%s %s\n",               \
+                myoff, (int) (pc - curMethod->insns),                       \
+                curMethod->clazz->descriptor, curMethod->name, desc);       \
+            free(desc);                                                     \
+            dvmAbort();                                                     \
+        }                                                                   \
+        pc += myoff;                                                        \
+    } while (false)
+#else
+# define ADJUST_PC(_offset) (pc += _offset)
+#endif
+
+/*
+ * If enabled, log instructions as we execute them.
+ */
+#ifdef LOG_INSTR
+# define ILOGD(...) ILOG(LOG_DEBUG, __VA_ARGS__)
+# define ILOGV(...) ILOG(LOG_VERBOSE, __VA_ARGS__)
+# define ILOG(_level, ...) do {                                             \
+        char debugStrBuf[128];                                              \
+        snprintf(debugStrBuf, sizeof(debugStrBuf), __VA_ARGS__);            \
+        if (curMethod != NULL)                                                 \
+            LOG(_level, LOG_TAG"i", "%-2d|%04x%s\n",                        \
+                self->threadId, (int)(pc - curMethod->insns), debugStrBuf); \
+        else                                                                \
+            LOG(_level, LOG_TAG"i", "%-2d|####%s\n",                        \
+                self->threadId, debugStrBuf);                               \
+    } while(false)
+void dvmDumpRegs(const Method* method, const u4* framePtr, bool inOnly);
+# define DUMP_REGS(_meth, _frame, _inOnly) dvmDumpRegs(_meth, _frame, _inOnly)
+static const char kSpacing[] = "            ";
+#else
+# define ILOGD(...) ((void)0)
+# define ILOGV(...) ((void)0)
+# define DUMP_REGS(_meth, _frame, _inOnly) ((void)0)
+#endif
+
+/* get a long from an array of u4 */
+static inline s8 getLongFromArray(const u4* ptr, int idx)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { s8 ll; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.parts[0] = ptr[0];
+    conv.parts[1] = ptr[1];
+    return conv.ll;
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    s8 val;
+    memcpy(&val, &ptr[idx], 8);
+    return val;
+#else
+    return *((s8*) &ptr[idx]);
+#endif
+}
+
+/* store a long into an array of u4 */
+static inline void putLongToArray(u4* ptr, int idx, s8 val)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { s8 ll; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.ll = val;
+    ptr[0] = conv.parts[0];
+    ptr[1] = conv.parts[1];
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    memcpy(&ptr[idx], &val, 8);
+#else
+    *((s8*) &ptr[idx]) = val;
+#endif
+}
+
+/* get a double from an array of u4 */
+static inline double getDoubleFromArray(const u4* ptr, int idx)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { double d; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.parts[0] = ptr[0];
+    conv.parts[1] = ptr[1];
+    return conv.d;
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    double dval;
+    memcpy(&dval, &ptr[idx], 8);
+    return dval;
+#else
+    return *((double*) &ptr[idx]);
+#endif
+}
+
+/* store a double into an array of u4 */
+static inline void putDoubleToArray(u4* ptr, int idx, double dval)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { double d; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.d = dval;
+    ptr[0] = conv.parts[0];
+    ptr[1] = conv.parts[1];
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    memcpy(&ptr[idx], &dval, 8);
+#else
+    *((double*) &ptr[idx]) = dval;
+#endif
+}
+
+/*
+ * If enabled, validate the register number on every access.  Otherwise,
+ * just do an array access.
+ *
+ * Assumes the existence of "u4* fp".
+ *
+ * "_idx" may be referenced more than once.
+ */
+#ifdef CHECK_REGISTER_INDICES
+# define GET_REGISTER(_idx) \
+    ( (_idx) < curMethod->registersSize ? \
+        (fp[(_idx)]) : (assert(!"bad reg"),1969) )
+# define SET_REGISTER(_idx, _val) \
+    ( (_idx) < curMethod->registersSize ? \
+        (fp[(_idx)] = (u4)(_val)) : (assert(!"bad reg"),1969) )
+# define GET_REGISTER_AS_OBJECT(_idx)       ((Object *)GET_REGISTER(_idx))
+# define SET_REGISTER_AS_OBJECT(_idx, _val) SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_INT(_idx) ((s4) GET_REGISTER(_idx))
+# define SET_REGISTER_INT(_idx, _val) SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_WIDE(_idx) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        getLongFromArray(fp, (_idx)) : (assert(!"bad reg"),1969) )
+# define SET_REGISTER_WIDE(_idx, _val) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        putLongToArray(fp, (_idx), (_val)) : (assert(!"bad reg"),1969) )
+# define GET_REGISTER_FLOAT(_idx) \
+    ( (_idx) < curMethod->registersSize ? \
+        (*((float*) &fp[(_idx)])) : (assert(!"bad reg"),1969.0f) )
+# define SET_REGISTER_FLOAT(_idx, _val) \
+    ( (_idx) < curMethod->registersSize ? \
+        (*((float*) &fp[(_idx)]) = (_val)) : (assert(!"bad reg"),1969.0f) )
+# define GET_REGISTER_DOUBLE(_idx) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        getDoubleFromArray(fp, (_idx)) : (assert(!"bad reg"),1969.0) )
+# define SET_REGISTER_DOUBLE(_idx, _val) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        putDoubleToArray(fp, (_idx), (_val)) : (assert(!"bad reg"),1969.0) )
+#else
+# define GET_REGISTER(_idx)                 (fp[(_idx)])
+# define SET_REGISTER(_idx, _val)           (fp[(_idx)] = (_val))
+# define GET_REGISTER_AS_OBJECT(_idx)       ((Object*) fp[(_idx)])
+# define SET_REGISTER_AS_OBJECT(_idx, _val) (fp[(_idx)] = (u4)(_val))
+# define GET_REGISTER_INT(_idx)             ((s4)GET_REGISTER(_idx))
+# define SET_REGISTER_INT(_idx, _val)       SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_WIDE(_idx)            getLongFromArray(fp, (_idx))
+# define SET_REGISTER_WIDE(_idx, _val)      putLongToArray(fp, (_idx), (_val))
+# define GET_REGISTER_FLOAT(_idx)           (*((float*) &fp[(_idx)]))
+# define SET_REGISTER_FLOAT(_idx, _val)     (*((float*) &fp[(_idx)]) = (_val))
+# define GET_REGISTER_DOUBLE(_idx)          getDoubleFromArray(fp, (_idx))
+# define SET_REGISTER_DOUBLE(_idx, _val)    putDoubleToArray(fp, (_idx), (_val))
+#endif
+
+/*
+ * Get 16 bits from the specified offset of the program counter.  We always
+ * want to load 16 bits at a time from the instruction stream -- it's more
+ * efficient than 8 and won't have the alignment problems that 32 might.
+ *
+ * Assumes existence of "const u2* pc".
+ */
+#define FETCH(_offset)     (pc[(_offset)])
+
+/*
+ * Extract instruction byte from 16-bit fetch (_inst is a u2).
+ */
+#define INST_INST(_inst)    ((_inst) & 0xff)
+
+/*
+ * Extract the "vA, vB" 4-bit registers from the instruction word (_inst is u2).
+ */
+#define INST_A(_inst)       (((_inst) >> 8) & 0x0f)
+#define INST_B(_inst)       ((_inst) >> 12)
+
+/*
+ * Get the 8-bit "vAA" 8-bit register index from the instruction word.
+ * (_inst is u2)
+ */
+#define INST_AA(_inst)      ((_inst) >> 8)
+
+/*
+ * The current PC must be available to Throwable constructors, e.g.
+ * those created by dvmThrowException(), so that the exception stack
+ * trace can be generated correctly.  If we don't do this, the offset
+ * within the current method won't be shown correctly.  See the notes
+ * in Exception.c.
+ *
+ * Assumes existence of "u4* fp" and "const u2* pc".
+ */
+#define EXPORT_PC()         (SAVEAREA_FROM_FP(fp)->xtra.currentPc = pc)
+
+/*
+ * Determine if we need to switch to a different interpreter.  "_current"
+ * is either INTERP_STD or INTERP_DBG.  It should be fixed for a given
+ * interpreter generation file, which should remove the outer conditional
+ * from the following.
+ *
+ * If we're building without debug and profiling support, we never switch.
+ */
+#if defined(WITH_PROFILER) || defined(WITH_DEBUGGER)
+# define NEED_INTERP_SWITCH(_current) (                                     \
+    (_current == INTERP_STD) ?                                              \
+        dvmDebuggerOrProfilerActive() : !dvmDebuggerOrProfilerActive() )
+#else
+# define NEED_INTERP_SWITCH(_current) (false)
+#endif
+
+/*
+ * Look up an interface on a class using the cache.
+ */
+INLINE Method* dvmFindInterfaceMethodInCache(ClassObject* thisClass,
+    u4 methodIdx, const Method* method, DvmDex* methodClassDex)
+{
+#define ATOMIC_CACHE_CALC \
+    dvmInterpFindInterfaceMethod(thisClass, methodIdx, method, methodClassDex)
+
+    return (Method*) ATOMIC_CACHE_LOOKUP(methodClassDex->pInterfaceCache,
+                DEX_INTERFACE_CACHE_SIZE, thisClass, methodIdx);
+
+#undef ATOMIC_CACHE_CALC
+}
+
+/*
+ * Check to see if "obj" is NULL.  If so, throw an exception.  Assumes the
+ * pc has already been exported to the stack.
+ *
+ * Perform additional checks on debug builds.
+ *
+ * Use this to check for NULL when the instruction handler calls into
+ * something that could throw an exception (so we have already called
+ * EXPORT_PC at the top).
+ */
+static inline bool checkForNull(Object* obj)
+{
+    if (obj == NULL) {
+        dvmThrowException("Ljava/lang/NullPointerException;", NULL);
+        return false;
+    }
+#ifdef WITH_EXTRA_OBJECT_VALIDATION
+    if (!dvmIsValidObject(obj)) {
+        LOGE("Invalid object %p\n", obj);
+        dvmAbort();
+    }
+#endif
+#ifndef NDEBUG
+    if (obj->clazz == NULL || ((u4) obj->clazz) <= 65536) {
+        /* probable heap corruption */
+        LOGE("Invalid object class %p (in %p)\n", obj->clazz, obj);
+        dvmAbort();
+    }
+#endif
+    return true;
+}
+
+/*
+ * Check to see if "obj" is NULL.  If so, export the PC into the stack
+ * frame and throw an exception.
+ *
+ * Perform additional checks on debug builds.
+ *
+ * Use this to check for NULL when the instruction handler doesn't do
+ * anything else that can throw an exception.
+ */
+static inline bool checkForNullExportPC(Object* obj, u4* fp, const u2* pc)
+{
+    if (obj == NULL) {
+        EXPORT_PC();
+        dvmThrowException("Ljava/lang/NullPointerException;", NULL);
+        return false;
+    }
+#ifdef WITH_EXTRA_OBJECT_VALIDATION
+    if (!dvmIsValidObject(obj)) {
+        LOGE("Invalid object %p\n", obj);
+        dvmAbort();
+    }
+#endif
+#ifndef NDEBUG
+    if (obj->clazz == NULL || ((u4) obj->clazz) <= 65536) {
+        /* probable heap corruption */
+        LOGE("Invalid object class %p (in %p)\n", obj->clazz, obj);
+        dvmAbort();
+    }
+#endif
+    return true;
+}
+
+
+/* File: cstubs/stubdefs.c */
+/* this is a standard (no debug support) interpreter */
+#define INTERP_TYPE INTERP_STD
+#define CHECK_DEBUG_AND_PROF() ((void)0)
+# define CHECK_TRACKED_REFS() ((void)0)
+
+/*
+ * In the C mterp stubs, "goto" is a function call followed immediately
+ * by a return.
+ */
+
+#define GOTO_TARGET_DECL(_target, ...)                                      \
+    void dvmMterp_##_target(MterpGlue* glue, ## __VA_ARGS__);
+
+#define GOTO_TARGET(_target, ...)                                           \
+    void dvmMterp_##_target(MterpGlue* glue, ## __VA_ARGS__) {              \
+        u2 ref, vsrc1, vsrc2, vdst;                                         \
+        u2 inst = FETCH(0);                                                 \
+        const Method* methodToCall;                                         \
+        StackSaveArea* debugSaveArea;
+
+#define GOTO_TARGET_END }
+
+/*
+ * Redefine what used to be local variable accesses into MterpGlue struct
+ * references.  (These are undefined down in "footer.c".)
+ */
+#define retval                  glue->retval
+#define pc                      glue->pc
+#define fp                      glue->fp
+#define curMethod               glue->method
+#define methodClassDex          glue->methodClassDex
+#define self                    glue->self
+#define debugTrackedRefStart    glue->debugTrackedRefStart
+
+/* ugh */
+#define STUB_HACK(x) x
+
+
+/*
+ * Opcode handler framing macros.  Here, each opcode is a separate function
+ * that takes a "glue" argument and returns void.  We can't declare
+ * these "static" because they may be called from an assembly stub.
+ */
+#define HANDLE_OPCODE(_op)                                                  \
+    void dvmMterp_##_op(MterpGlue* glue) {                                  \
+        u2 ref, vsrc1, vsrc2, vdst;                                         \
+        u2 inst = FETCH(0);
+
+#define OP_END }
+
+/*
+ * Like the "portable" FINISH, but don't reload "inst", and return to caller
+ * when done.
+ */
+#define FINISH(_offset) {                                                   \
+        ADJUST_PC(_offset);                                                 \
+        CHECK_DEBUG_AND_PROF();                                             \
+        CHECK_TRACKED_REFS();                                               \
+        return;                                                             \
+    }
+
+
+/*
+ * The "goto label" statements turn into function calls followed by
+ * return statements.  Some of the functions take arguments, which in the
+ * portable interpreter are handled by assigning values to globals.
+ */
+
+#define GOTO_exceptionThrown()                                              \
+    do {                                                                    \
+        dvmMterp_exceptionThrown(glue);                                     \
+        return;                                                             \
+    } while(false)
+
+#define GOTO_returnFromMethod()                                             \
+    do {                                                                    \
+        dvmMterp_returnFromMethod(glue);                                    \
+        return;                                                             \
+    } while(false)
+
+#define GOTO_invoke(_target, _methodCallRange)                              \
+    do {                                                                    \
+        dvmMterp_##_target(glue, _methodCallRange);                         \
+        return;                                                             \
+    } while(false)
+
+#define GOTO_invokeMethod(_methodCallRange, _methodToCall, _vsrc1, _vdst)   \
+    do {                                                                    \
+        dvmMterp_invokeMethod(glue, _methodCallRange, _methodToCall,        \
+            _vsrc1, _vdst);                                                 \
+        return;                                                             \
+    } while(false)
+
+/*
+ * As a special case, "goto bail" turns into a longjmp.  Use "bail_switch"
+ * if we need to switch to the other interpreter upon our return.
+ */
+#define GOTO_bail()                                                         \
+    dvmMterpStdBail(glue, false);
+#define GOTO_bail_switch()                                                  \
+    dvmMterpStdBail(glue, true);
+
+/*
+ * Periodically check for thread suspension.
+ *
+ * While we're at it, see if a debugger has attached or the profiler has
+ * started.  If so, switch to a different "goto" table.
+ */
+#define PERIODIC_CHECKS(_entryPoint, _pcadj) {                              \
+        dvmCheckSuspendQuick(self);                                         \
+        if (NEED_INTERP_SWITCH(INTERP_TYPE)) {                              \
+            ADJUST_PC(_pcadj);                                              \
+            glue->entryPoint = _entryPoint;                                 \
+            LOGVV("threadid=%d: switch to STD ep=%d adj=%d\n",              \
+                glue->self->threadId, (_entryPoint), (_pcadj));             \
+            GOTO_bail_switch();                                             \
+        }                                                                   \
+    }
+
+
+/* File: c/opcommon.c */
+/* forward declarations of goto targets */
+GOTO_TARGET_DECL(filledNewArray, bool methodCallRange);
+GOTO_TARGET_DECL(invokeVirtual, bool methodCallRange);
+GOTO_TARGET_DECL(invokeSuper, bool methodCallRange);
+GOTO_TARGET_DECL(invokeInterface, bool methodCallRange);
+GOTO_TARGET_DECL(invokeDirect, bool methodCallRange);
+GOTO_TARGET_DECL(invokeStatic, bool methodCallRange);
+GOTO_TARGET_DECL(invokeVirtualQuick, bool methodCallRange);
+GOTO_TARGET_DECL(invokeSuperQuick, bool methodCallRange);
+GOTO_TARGET_DECL(invokeMethod, bool methodCallRange, const Method* methodToCall,
+    u2 count, u2 regs);
+GOTO_TARGET_DECL(returnFromMethod);
+GOTO_TARGET_DECL(exceptionThrown);
+
+/*
+ * ===========================================================================
+ *
+ * What follows are opcode definitions shared between multiple opcodes with
+ * minor substitutions handled by the C pre-processor.  These should probably
+ * use the mterp substitution mechanism instead, with the code here moved
+ * into common fragment files (like the asm "binop.S"), although it's hard
+ * to give up the C preprocessor in favor of the much simpler text subst.
+ *
+ * ===========================================================================
+ */
+
+#define HANDLE_NUMCONV(_opcode, _opname, _fromtype, _totype)                \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        SET_REGISTER##_totype(vdst,                                         \
+            GET_REGISTER##_fromtype(vsrc1));                                \
+        FINISH(1);
+
+#define HANDLE_FLOAT_TO_INT(_opcode, _opname, _fromvtype, _fromrtype,       \
+        _tovtype, _tortype)                                                 \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+    {                                                                       \
+        /* spec defines specific handling for +/- inf and NaN values */     \
+        _fromvtype val;                                                     \
+        _tovtype intMin, intMax, result;                                    \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        val = GET_REGISTER##_fromrtype(vsrc1);                              \
+        intMin = (_tovtype) 1 << (sizeof(_tovtype) * 8 -1);                 \
+        intMax = ~intMin;                                                   \
+        result = (_tovtype) val;                                            \
+        if (val >= intMax)          /* +inf */                              \
+            result = intMax;                                                \
+        else if (val <= intMin)     /* -inf */                              \
+            result = intMin;                                                \
+        else if (val != val)        /* NaN */                               \
+            result = 0;                                                     \
+        else                                                                \
+            result = (_tovtype) val;                                        \
+        SET_REGISTER##_tortype(vdst, result);                               \
+    }                                                                       \
+    FINISH(1);
+
+#define HANDLE_INT_TO_SMALL(_opcode, _opname, _type)                        \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|int-to-%s v%d,v%d", (_opname), vdst, vsrc1);                \
+        SET_REGISTER(vdst, (_type) GET_REGISTER(vsrc1));                    \
+        FINISH(1);
+
+/* NOTE: the comparison result is always a signed 4-byte integer */
+#define HANDLE_OP_CMPX(_opcode, _opname, _varType, _type, _nanVal)          \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        int result;                                                         \
+        u2 regs;                                                            \
+        _varType val1, val2;                                                \
+        vdst = INST_AA(inst);                                               \
+        regs = FETCH(1);                                                    \
+        vsrc1 = regs & 0xff;                                                \
+        vsrc2 = regs >> 8;                                                  \
+        ILOGV("|cmp%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);         \
+        val1 = GET_REGISTER##_type(vsrc1);                                  \
+        val2 = GET_REGISTER##_type(vsrc2);                                  \
+        if (val1 == val2)                                                   \
+            result = 0;                                                     \
+        else if (val1 < val2)                                               \
+            result = -1;                                                    \
+        else if (val1 > val2)                                               \
+            result = 1;                                                     \
+        else                                                                \
+            result = (_nanVal);                                             \
+        ILOGV("+ result=%d\n", result);                                     \
+        SET_REGISTER(vdst, result);                                         \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_IF_XX(_opcode, _opname, _cmp)                             \
+    HANDLE_OPCODE(_opcode /*vA, vB, +CCCC*/)                                \
+        vsrc1 = INST_A(inst);                                               \
+        vsrc2 = INST_B(inst);                                               \
+        if ((s4) GET_REGISTER(vsrc1) _cmp (s4) GET_REGISTER(vsrc2)) {       \
+            int branchOffset = (s2)FETCH(1);    /* sign-extended */         \
+            ILOGV("|if-%s v%d,v%d,+0x%04x", (_opname), vsrc1, vsrc2,        \
+                branchOffset);                                              \
+            ILOGV("> branch taken");                                        \
+            if (branchOffset < 0)                                           \
+                PERIODIC_CHECKS(kInterpEntryInstr, branchOffset);           \
+            FINISH(branchOffset);                                           \
+        } else {                                                            \
+            ILOGV("|if-%s v%d,v%d,-", (_opname), vsrc1, vsrc2);             \
+            FINISH(2);                                                      \
+        }
+
+#define HANDLE_OP_IF_XXZ(_opcode, _opname, _cmp)                            \
+    HANDLE_OPCODE(_opcode /*vAA, +BBBB*/)                                   \
+        vsrc1 = INST_AA(inst);                                              \
+        if ((s4) GET_REGISTER(vsrc1) _cmp 0) {                              \
+            int branchOffset = (s2)FETCH(1);    /* sign-extended */         \
+            ILOGV("|if-%s v%d,+0x%04x", (_opname), vsrc1, branchOffset);    \
+            ILOGV("> branch taken");                                        \
+            if (branchOffset < 0)                                           \
+                PERIODIC_CHECKS(kInterpEntryInstr, branchOffset);           \
+            FINISH(branchOffset);                                           \
+        } else {                                                            \
+            ILOGV("|if-%s v%d,-", (_opname), vsrc1);                        \
+            FINISH(2);                                                      \
+        }
+
+#define HANDLE_UNOP(_opcode, _opname, _pfx, _sfx, _type)                    \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        SET_REGISTER##_type(vdst, _pfx GET_REGISTER##_type(vsrc1) _sfx);    \
+        FINISH(1);
+
+#define HANDLE_OP_X_INT(_opcode, _opname, _op, _chkdiv)                     \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-int v%d,v%d", (_opname), vdst, vsrc1);                   \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            secondVal = GET_REGISTER(vsrc2);                                \
+            if (secondVal == 0) {                                           \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && secondVal == -1) {            \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            /* non-div/rem case */                                          \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vsrc1) _op (s4) GET_REGISTER(vsrc2));     \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_INT(_opcode, _opname, _cast, _op)                     \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-int v%d,v%d", (_opname), vdst, vsrc1);                   \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vsrc1) _op (GET_REGISTER(vsrc2) & 0x1f));    \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_INT_LIT16(_opcode, _opname, _op, _chkdiv)               \
+    HANDLE_OPCODE(_opcode /*vA, vB, #+CCCC*/)                               \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        vsrc2 = FETCH(1);                                                   \
+        ILOGV("|%s-int/lit16 v%d,v%d,#+0x%04x",                             \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, result;                                            \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            if ((s2) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                      \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && ((s2) vsrc2) == -1) {         \
+                /* won't generate /lit16 instr for this; check anyway */    \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op (s2) vsrc2;                           \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            /* non-div/rem case */                                          \
+            SET_REGISTER(vdst, GET_REGISTER(vsrc1) _op (s2) vsrc2);         \
+        }                                                                   \
+        FINISH(2);
+
+#define HANDLE_OP_X_INT_LIT8(_opcode, _opname, _op, _chkdiv)                \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, #+CC*/)                               \
+    {                                                                       \
+        u2 litInfo;                                                         \
+        vdst = INST_AA(inst);                                               \
+        litInfo = FETCH(1);                                                 \
+        vsrc1 = litInfo & 0xff;                                             \
+        vsrc2 = litInfo >> 8;       /* constant */                          \
+        ILOGV("|%s-int/lit8 v%d,v%d,#+0x%02x",                              \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, result;                                            \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            if ((s1) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && ((s1) vsrc2) == -1) {         \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op ((s1) vsrc2);                         \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vsrc1) _op (s1) vsrc2);                   \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_INT_LIT8(_opcode, _opname, _cast, _op)                \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, #+CC*/)                               \
+    {                                                                       \
+        u2 litInfo;                                                         \
+        vdst = INST_AA(inst);                                               \
+        litInfo = FETCH(1);                                                 \
+        vsrc1 = litInfo & 0xff;                                             \
+        vsrc2 = litInfo >> 8;       /* constant */                          \
+        ILOGV("|%s-int/lit8 v%d,v%d,#+0x%02x",                              \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vsrc1) _op (vsrc2 & 0x1f));                  \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_INT_2ADDR(_opcode, _opname, _op, _chkdiv)               \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-int-2addr v%d,v%d", (_opname), vdst, vsrc1);             \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER(vdst);                                  \
+            secondVal = GET_REGISTER(vsrc1);                                \
+            if (secondVal == 0) {                                           \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && secondVal == -1) {            \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vdst) _op (s4) GET_REGISTER(vsrc1));      \
+        }                                                                   \
+        FINISH(1);
+
+#define HANDLE_OP_SHX_INT_2ADDR(_opcode, _opname, _cast, _op)               \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-int-2addr v%d,v%d", (_opname), vdst, vsrc1);             \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vdst) _op (GET_REGISTER(vsrc1) & 0x1f));     \
+        FINISH(1);
+
+#define HANDLE_OP_X_LONG(_opcode, _opname, _op, _chkdiv)                    \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-long v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);       \
+        if (_chkdiv != 0) {                                                 \
+            s8 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER_WIDE(vsrc1);                            \
+            secondVal = GET_REGISTER_WIDE(vsrc2);                           \
+            if (secondVal == 0LL) {                                         \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u8)firstVal == 0x8000000000000000ULL &&                    \
+                secondVal == -1LL)                                          \
+            {                                                               \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER_WIDE(vdst, result);                                \
+        } else {                                                            \
+            SET_REGISTER_WIDE(vdst,                                         \
+                (s8) GET_REGISTER_WIDE(vsrc1) _op (s8) GET_REGISTER_WIDE(vsrc2)); \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_LONG(_opcode, _opname, _cast, _op)                    \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-long v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);       \
+        SET_REGISTER_WIDE(vdst,                                             \
+            _cast GET_REGISTER_WIDE(vsrc1) _op (GET_REGISTER(vsrc2) & 0x3f)); \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_LONG_2ADDR(_opcode, _opname, _op, _chkdiv)              \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-long-2addr v%d,v%d", (_opname), vdst, vsrc1);            \
+        if (_chkdiv != 0) {                                                 \
+            s8 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER_WIDE(vdst);                             \
+            secondVal = GET_REGISTER_WIDE(vsrc1);                           \
+            if (secondVal == 0LL) {                                         \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u8)firstVal == 0x8000000000000000ULL &&                    \
+                secondVal == -1LL)                                          \
+            {                                                               \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER_WIDE(vdst, result);                                \
+        } else {                                                            \
+            SET_REGISTER_WIDE(vdst,                                         \
+                (s8) GET_REGISTER_WIDE(vdst) _op (s8)GET_REGISTER_WIDE(vsrc1));\
+        }                                                                   \
+        FINISH(1);
+
+#define HANDLE_OP_SHX_LONG_2ADDR(_opcode, _opname, _cast, _op)              \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-long-2addr v%d,v%d", (_opname), vdst, vsrc1);            \
+        SET_REGISTER_WIDE(vdst,                                             \
+            _cast GET_REGISTER_WIDE(vdst) _op (GET_REGISTER(vsrc1) & 0x3f)); \
+        FINISH(1);
+
+#define HANDLE_OP_X_FLOAT(_opcode, _opname, _op)                            \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-float v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);      \
+        SET_REGISTER_FLOAT(vdst,                                            \
+            GET_REGISTER_FLOAT(vsrc1) _op GET_REGISTER_FLOAT(vsrc2));       \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_DOUBLE(_opcode, _opname, _op)                           \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-double v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);     \
+        SET_REGISTER_DOUBLE(vdst,                                           \
+            GET_REGISTER_DOUBLE(vsrc1) _op GET_REGISTER_DOUBLE(vsrc2));     \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_FLOAT_2ADDR(_opcode, _opname, _op)                      \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-float-2addr v%d,v%d", (_opname), vdst, vsrc1);           \
+        SET_REGISTER_FLOAT(vdst,                                            \
+            GET_REGISTER_FLOAT(vdst) _op GET_REGISTER_FLOAT(vsrc1));        \
+        FINISH(1);
+
+#define HANDLE_OP_X_DOUBLE_2ADDR(_opcode, _opname, _op)                     \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-double-2addr v%d,v%d", (_opname), vdst, vsrc1);          \
+        SET_REGISTER_DOUBLE(vdst,                                           \
+            GET_REGISTER_DOUBLE(vdst) _op GET_REGISTER_DOUBLE(vsrc1));      \
+        FINISH(1);
+
+#define HANDLE_OP_AGET(_opcode, _opname, _type, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        ArrayObject* arrayObj;                                              \
+        u2 arrayInfo;                                                       \
+        EXPORT_PC();                                                        \
+        vdst = INST_AA(inst);                                               \
+        arrayInfo = FETCH(1);                                               \
+        vsrc1 = arrayInfo & 0xff;    /* array ptr */                        \
+        vsrc2 = arrayInfo >> 8;      /* index */                            \
+        ILOGV("|aget%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);        \
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);                      \
+        if (!checkForNull((Object*) arrayObj))                              \
+            GOTO_exceptionThrown();                                         \
+        if (GET_REGISTER(vsrc2) >= arrayObj->length) {                      \
+            LOGV("Invalid array access: %p %d (len=%d)\n",                  \
+                arrayObj, vsrc2, arrayObj->length);                         \
+            dvmThrowException("Ljava/lang/ArrayIndexOutOfBoundsException;", \
+                NULL);                                                      \
+            GOTO_exceptionThrown();                                         \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst,                                        \
+            ((_type*) arrayObj->contents)[GET_REGISTER(vsrc2)]);            \
+        ILOGV("+ AGET[%d]=0x%x", GET_REGISTER(vsrc2), GET_REGISTER(vdst));  \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_APUT(_opcode, _opname, _type, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        ArrayObject* arrayObj;                                              \
+        u2 arrayInfo;                                                       \
+        EXPORT_PC();                                                        \
+        vdst = INST_AA(inst);       /* AA: source value */                  \
+        arrayInfo = FETCH(1);                                               \
+        vsrc1 = arrayInfo & 0xff;   /* BB: array ptr */                     \
+        vsrc2 = arrayInfo >> 8;     /* CC: index */                         \
+        ILOGV("|aput%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);        \
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);                      \
+        if (!checkForNull((Object*) arrayObj))                              \
+            GOTO_exceptionThrown();                                         \
+        if (GET_REGISTER(vsrc2) >= arrayObj->length) {                      \
+            dvmThrowException("Ljava/lang/ArrayIndexOutOfBoundsException;", \
+                NULL);                                                      \
+            GOTO_exceptionThrown();                                         \
+        }                                                                   \
+        ILOGV("+ APUT[%d]=0x%08x", GET_REGISTER(vsrc2), GET_REGISTER(vdst));\
+        ((_type*) arrayObj->contents)[GET_REGISTER(vsrc2)] =                \
+            GET_REGISTER##_regsize(vdst);                                   \
+    }                                                                       \
+    FINISH(2);
+
+/*
+ * It's possible to get a bad value out of a field with sub-32-bit stores
+ * because the -quick versions always operate on 32 bits.  Consider:
+ *   short foo = -1  (sets a 32-bit register to 0xffffffff)
+ *   iput-quick foo  (writes all 32 bits to the field)
+ *   short bar = 1   (sets a 32-bit register to 0x00000001)
+ *   iput-short      (writes the low 16 bits to the field)
+ *   iget-quick foo  (reads all 32 bits from the field, yielding 0xffff0001)
+ * This can only happen when optimized and non-optimized code has interleaved
+ * access to the same field.  This is unlikely but possible.
+ *
+ * The easiest way to fix this is to always read/write 32 bits at a time.  On
+ * a device with a 16-bit data bus this is sub-optimal.  (The alternative
+ * approach is to have sub-int versions of iget-quick, but now we're wasting
+ * Dalvik instruction space and making it less likely that handler code will
+ * already be in the CPU i-cache.)
+ */
+#define HANDLE_IGET_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        InstField* ifield;                                                  \
+        Object* obj;                                                        \
+        EXPORT_PC();                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|iget%s v%d,v%d,field@0x%04x", (_opname), vdst, vsrc1, ref); \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNull(obj))                                             \
+            GOTO_exceptionThrown();                                         \
+        ifield = (InstField*) dvmDexGetResolvedField(methodClassDex, ref);  \
+        if (ifield == NULL) {                                               \
+            ifield = dvmResolveInstField(curMethod->clazz, ref);            \
+            if (ifield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst,                                        \
+            dvmGetField##_ftype(obj, ifield->byteOffset));                  \
+        ILOGV("+ IGET '%s'=0x%08llx", ifield->field.name,                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+        UPDATE_FIELD_GET(&ifield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IGET_X_QUICK(_opcode, _opname, _ftype, _regsize)             \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        Object* obj;                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field offset */                          \
+        ILOGV("|iget%s-quick v%d,v%d,field@+%u",                            \
+            (_opname), vdst, vsrc1, ref);                                   \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNullExportPC(obj, fp, pc))                             \
+            GOTO_exceptionThrown();                                         \
+        SET_REGISTER##_regsize(vdst, dvmGetField##_ftype(obj, ref));        \
+        ILOGV("+ IGETQ %d=0x%08llx", ref,                                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IPUT_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        InstField* ifield;                                                  \
+        Object* obj;                                                        \
+        EXPORT_PC();                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|iput%s v%d,v%d,field@0x%04x", (_opname), vdst, vsrc1, ref); \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNull(obj))                                             \
+            GOTO_exceptionThrown();                                         \
+        ifield = (InstField*) dvmDexGetResolvedField(methodClassDex, ref);  \
+        if (ifield == NULL) {                                               \
+            ifield = dvmResolveInstField(curMethod->clazz, ref);            \
+            if (ifield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        dvmSetField##_ftype(obj, ifield->byteOffset,                        \
+            GET_REGISTER##_regsize(vdst));                                  \
+        ILOGV("+ IPUT '%s'=0x%08llx", ifield->field.name,                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+        UPDATE_FIELD_PUT(&ifield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IPUT_X_QUICK(_opcode, _opname, _ftype, _regsize)             \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        Object* obj;                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field offset */                          \
+        ILOGV("|iput%s-quick v%d,v%d,field@0x%04x",                         \
+            (_opname), vdst, vsrc1, ref);                                   \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNullExportPC(obj, fp, pc))                             \
+            GOTO_exceptionThrown();                                         \
+        dvmSetField##_ftype(obj, ref, GET_REGISTER##_regsize(vdst));        \
+        ILOGV("+ IPUTQ %d=0x%08llx", ref,                                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_SGET_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, field@BBBB*/)                              \
+    {                                                                       \
+        StaticField* sfield;                                                \
+        vdst = INST_AA(inst);                                               \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|sget%s v%d,sfield@0x%04x", (_opname), vdst, ref);           \
+        sfield = (StaticField*)dvmDexGetResolvedField(methodClassDex, ref); \
+        if (sfield == NULL) {                                               \
+            EXPORT_PC();                                                    \
+            sfield = dvmResolveStaticField(curMethod->clazz, ref);          \
+            if (sfield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst, dvmGetStaticField##_ftype(sfield));    \
+        ILOGV("+ SGET '%s'=0x%08llx",                                       \
+            sfield->field.name, (u8)GET_REGISTER##_regsize(vdst));          \
+        UPDATE_FIELD_GET(&sfield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_SPUT_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, field@BBBB*/)                              \
+    {                                                                       \
+        StaticField* sfield;                                                \
+        vdst = INST_AA(inst);                                               \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|sput%s v%d,sfield@0x%04x", (_opname), vdst, ref);           \
+        sfield = (StaticField*)dvmDexGetResolvedField(methodClassDex, ref); \
+        if (sfield == NULL) {                                               \
+            EXPORT_PC();                                                    \
+            sfield = dvmResolveStaticField(curMethod->clazz, ref);          \
+            if (sfield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        dvmSetStaticField##_ftype(sfield, GET_REGISTER##_regsize(vdst));    \
+        ILOGV("+ SPUT '%s'=0x%08llx",                                       \
+            sfield->field.name, (u8)GET_REGISTER##_regsize(vdst));          \
+        UPDATE_FIELD_PUT(&sfield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+
+/* File: c/OP_NOP.c */
+HANDLE_OPCODE(OP_NOP)
+    FINISH(1);
+OP_END
+
+/* File: c/OP_MOVE.c */
+HANDLE_OPCODE(OP_MOVE /*vA, vB*/)
+    vdst = INST_A(inst);
+    vsrc1 = INST_B(inst);
+    ILOGV("|move%s v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(1);
+OP_END
+
+/* File: c/OP_MOVE_FROM16.c */
+HANDLE_OPCODE(OP_MOVE_FROM16 /*vAA, vBBBB*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|move%s/from16 v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE_FROM16) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(2);
+OP_END
+
+/* File: c/OP_MOVE_16.c */
+HANDLE_OPCODE(OP_MOVE_16 /*vAAAA, vBBBB*/)
+    vdst = FETCH(1);
+    vsrc1 = FETCH(2);
+    ILOGV("|move%s/16 v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE_16) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(3);
+OP_END
+
+/* File: c/OP_MOVE_WIDE.c */
+HANDLE_OPCODE(OP_MOVE_WIDE /*vA, vB*/)
+    /* IMPORTANT: must correctly handle overlapping registers, e.g. both
+     * "move-wide v6, v7" and "move-wide v7, v6" */
+    vdst = INST_A(inst);
+    vsrc1 = INST_B(inst);
+    ILOGV("|move-wide v%d,v%d %s(v%d=0x%08llx)", vdst, vsrc1,
+        kSpacing+5, vdst, GET_REGISTER_WIDE(vsrc1));
+    SET_REGISTER_WIDE(vdst, GET_REGISTER_WIDE(vsrc1));
+    FINISH(1);
+OP_END
+
+/* File: c/OP_MOVE_WIDE_FROM16.c */
+HANDLE_OPCODE(OP_MOVE_WIDE_FROM16 /*vAA, vBBBB*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|move-wide/from16 v%d,v%d  (v%d=0x%08llx)", vdst, vsrc1,
+        vdst, GET_REGISTER_WIDE(vsrc1));
+    SET_REGISTER_WIDE(vdst, GET_REGISTER_WIDE(vsrc1));
+    FINISH(2);
+OP_END
+
+/* File: c/OP_MOVE_WIDE_16.c */
+HANDLE_OPCODE(OP_MOVE_WIDE_16 /*vAAAA, vBBBB*/)
+    vdst = FETCH(1);
+    vsrc1 = FETCH(2);
+    ILOGV("|move-wide/16 v%d,v%d %s(v%d=0x%08llx)", vdst, vsrc1,
+        kSpacing+8, vdst, GET_REGISTER_WIDE(vsrc1));
+    SET_REGISTER_WIDE(vdst, GET_REGISTER_WIDE(vsrc1));
+    FINISH(3);
+OP_END
+
+/* File: c/OP_MOVE_OBJECT.c */
+/* File: c/OP_MOVE.c */
+HANDLE_OPCODE(OP_MOVE_OBJECT /*vA, vB*/)
+    vdst = INST_A(inst);
+    vsrc1 = INST_B(inst);
+    ILOGV("|move%s v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(1);
+OP_END
+
+
+/* File: c/OP_MOVE_OBJECT_FROM16.c */
+/* File: c/OP_MOVE_FROM16.c */
+HANDLE_OPCODE(OP_MOVE_OBJECT_FROM16 /*vAA, vBBBB*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|move%s/from16 v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE_FROM16) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(2);
+OP_END
+
+
+/* File: c/OP_MOVE_OBJECT_16.c */
+/* File: c/OP_MOVE_16.c */
+HANDLE_OPCODE(OP_MOVE_OBJECT_16 /*vAAAA, vBBBB*/)
+    vdst = FETCH(1);
+    vsrc1 = FETCH(2);
+    ILOGV("|move%s/16 v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE_16) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(3);
+OP_END
+
+
+/* File: c/OP_MOVE_RESULT.c */
+HANDLE_OPCODE(OP_MOVE_RESULT /*vAA*/)
+    vdst = INST_AA(inst);
+    ILOGV("|move-result%s v%d %s(v%d=0x%08x)",
+         (INST_INST(inst) == OP_MOVE_RESULT) ? "" : "-object",
+         vdst, kSpacing+4, vdst,retval.i);
+    SET_REGISTER(vdst, retval.i);
+    FINISH(1);
+OP_END
+
+/* File: c/OP_MOVE_RESULT_WIDE.c */
+HANDLE_OPCODE(OP_MOVE_RESULT_WIDE /*vAA*/)
+    vdst = INST_AA(inst);
+    ILOGV("|move-result-wide v%d %s(0x%08llx)", vdst, kSpacing, retval.j);
+    SET_REGISTER_WIDE(vdst, retval.j);
+    FINISH(1);
+OP_END
+
+/* File: c/OP_MOVE_RESULT_OBJECT.c */
+/* File: c/OP_MOVE_RESULT.c */
+HANDLE_OPCODE(OP_MOVE_RESULT_OBJECT /*vAA*/)
+    vdst = INST_AA(inst);
+    ILOGV("|move-result%s v%d %s(v%d=0x%08x)",
+         (INST_INST(inst) == OP_MOVE_RESULT) ? "" : "-object",
+         vdst, kSpacing+4, vdst,retval.i);
+    SET_REGISTER(vdst, retval.i);
+    FINISH(1);
+OP_END
+
+
+/* File: c/OP_MOVE_EXCEPTION.c */
+HANDLE_OPCODE(OP_MOVE_EXCEPTION /*vAA*/)
+    vdst = INST_AA(inst);
+    ILOGV("|move-exception v%d", vdst);
+    assert(self->exception != NULL);
+    SET_REGISTER(vdst, (u4)self->exception);
+    dvmClearException(self);
+    FINISH(1);
+OP_END
+
+/* File: c/OP_RETURN_VOID.c */
+HANDLE_OPCODE(OP_RETURN_VOID /**/)
+    ILOGV("|return-void");
+#ifndef NDEBUG
+    retval.j = 0xababababULL;    // placate valgrind
+#endif
+    GOTO_returnFromMethod();
+OP_END
+
+/* File: c/OP_RETURN.c */
+HANDLE_OPCODE(OP_RETURN /*vAA*/)
+    vsrc1 = INST_AA(inst);
+    ILOGV("|return%s v%d",
+        (INST_INST(inst) == OP_RETURN) ? "" : "-object", vsrc1);
+    retval.i = GET_REGISTER(vsrc1);
+    GOTO_returnFromMethod();
+OP_END
+
+/* File: c/OP_RETURN_WIDE.c */
+HANDLE_OPCODE(OP_RETURN_WIDE /*vAA*/)
+    vsrc1 = INST_AA(inst);
+    ILOGV("|return-wide v%d", vsrc1);
+    retval.j = GET_REGISTER_WIDE(vsrc1);
+    GOTO_returnFromMethod();
+OP_END
+
+/* File: c/OP_RETURN_OBJECT.c */
+/* File: c/OP_RETURN.c */
+HANDLE_OPCODE(OP_RETURN_OBJECT /*vAA*/)
+    vsrc1 = INST_AA(inst);
+    ILOGV("|return%s v%d",
+        (INST_INST(inst) == OP_RETURN) ? "" : "-object", vsrc1);
+    retval.i = GET_REGISTER(vsrc1);
+    GOTO_returnFromMethod();
+OP_END
+
+
+/* File: c/OP_CONST_4.c */
+HANDLE_OPCODE(OP_CONST_4 /*vA, #+B*/)
+    {
+        s4 tmp;
+
+        vdst = INST_A(inst);
+        tmp = (s4) (INST_B(inst) << 28) >> 28;  // sign extend 4-bit value
+        ILOGV("|const/4 v%d,#0x%02x", vdst, (s4)tmp);
+        SET_REGISTER(vdst, tmp);
+    }
+    FINISH(1);
+OP_END
+
+/* File: c/OP_CONST_16.c */
+HANDLE_OPCODE(OP_CONST_16 /*vAA, #+BBBB*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|const/16 v%d,#0x%04x", vdst, (s2)vsrc1);
+    SET_REGISTER(vdst, (s2) vsrc1);
+    FINISH(2);
+OP_END
+
+/* File: c/OP_CONST.c */
+HANDLE_OPCODE(OP_CONST /*vAA, #+BBBBBBBB*/)
+    {
+        u4 tmp;
+
+        vdst = INST_AA(inst);
+        tmp = FETCH(1);
+        tmp |= (u4)FETCH(2) << 16;
+        ILOGV("|const v%d,#0x%08x", vdst, tmp);
+        SET_REGISTER(vdst, tmp);
+    }
+    FINISH(3);
+OP_END
+
+/* File: c/OP_CONST_HIGH16.c */
+HANDLE_OPCODE(OP_CONST_HIGH16 /*vAA, #+BBBB0000*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|const/high16 v%d,#0x%04x0000", vdst, vsrc1);
+    SET_REGISTER(vdst, vsrc1 << 16);
+    FINISH(2);
+OP_END
+
+/* File: c/OP_CONST_WIDE_16.c */
+HANDLE_OPCODE(OP_CONST_WIDE_16 /*vAA, #+BBBB*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|const-wide/16 v%d,#0x%04x", vdst, (s2)vsrc1);
+    SET_REGISTER_WIDE(vdst, (s2)vsrc1);
+    FINISH(2);
+OP_END
+
+/* File: c/OP_CONST_WIDE_32.c */
+HANDLE_OPCODE(OP_CONST_WIDE_32 /*vAA, #+BBBBBBBB*/)
+    {
+        u4 tmp;
+
+        vdst = INST_AA(inst);
+        tmp = FETCH(1);
+        tmp |= (u4)FETCH(2) << 16;
+        ILOGV("|const-wide/32 v%d,#0x%08x", vdst, tmp);
+        SET_REGISTER_WIDE(vdst, (s4) tmp);
+    }
+    FINISH(3);
+OP_END
+
+/* File: c/OP_CONST_WIDE.c */
+HANDLE_OPCODE(OP_CONST_WIDE /*vAA, #+BBBBBBBBBBBBBBBB*/)
+    {
+        u8 tmp;
+
+        vdst = INST_AA(inst);
+        tmp = FETCH(1);
+        tmp |= (u8)FETCH(2) << 16;
+        tmp |= (u8)FETCH(3) << 32;
+        tmp |= (u8)FETCH(4) << 48;
+        ILOGV("|const-wide v%d,#0x%08llx", vdst, tmp);
+        SET_REGISTER_WIDE(vdst, tmp);
+    }
+    FINISH(5);
+OP_END
+
+/* File: c/OP_CONST_WIDE_HIGH16.c */
+HANDLE_OPCODE(OP_CONST_WIDE_HIGH16 /*vAA, #+BBBB000000000000*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|const-wide/high16 v%d,#0x%04x000000000000", vdst, vsrc1);
+    SET_REGISTER_WIDE(vdst, ((u8) vsrc1) << 48);
+    FINISH(2);
+OP_END
+
+/* File: c/OP_CONST_STRING.c */
+HANDLE_OPCODE(OP_CONST_STRING /*vAA, string@BBBB*/)
+    {
+        StringObject* strObj;
+
+        vdst = INST_AA(inst);
+        ref = FETCH(1);
+        ILOGV("|const-string v%d string@0x%04x", vdst, ref);
+        strObj = dvmDexGetResolvedString(methodClassDex, ref);
+        if (strObj == NULL) {
+            EXPORT_PC();
+            strObj = dvmResolveString(curMethod->clazz, ref);
+            if (strObj == NULL)
+                GOTO_exceptionThrown();
+        }
+        SET_REGISTER(vdst, (u4) strObj);
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_CONST_STRING_JUMBO.c */
+HANDLE_OPCODE(OP_CONST_STRING_JUMBO /*vAA, string@BBBBBBBB*/)
+    {
+        StringObject* strObj;
+        u4 tmp;
+
+        vdst = INST_AA(inst);
+        tmp = FETCH(1);
+        tmp |= (u4)FETCH(2) << 16;
+        ILOGV("|const-string/jumbo v%d string@0x%08x", vdst, tmp);
+        strObj = dvmDexGetResolvedString(methodClassDex, tmp);
+        if (strObj == NULL) {
+            EXPORT_PC();
+            strObj = dvmResolveString(curMethod->clazz, tmp);
+            if (strObj == NULL)
+                GOTO_exceptionThrown();
+        }
+        SET_REGISTER(vdst, (u4) strObj);
+    }
+    FINISH(3);
+OP_END
+
+/* File: c/OP_CONST_CLASS.c */
+HANDLE_OPCODE(OP_CONST_CLASS /*vAA, class@BBBB*/)
+    {
+        ClassObject* clazz;
+
+        vdst = INST_AA(inst);
+        ref = FETCH(1);
+        ILOGV("|const-class v%d class@0x%04x", vdst, ref);
+        clazz = dvmDexGetResolvedClass(methodClassDex, ref);
+        if (clazz == NULL) {
+            EXPORT_PC();
+            clazz = dvmResolveClass(curMethod->clazz, ref, true);
+            if (clazz == NULL)
+                GOTO_exceptionThrown();
+        }
+        SET_REGISTER(vdst, (u4) clazz);
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_MONITOR_ENTER.c */
+HANDLE_OPCODE(OP_MONITOR_ENTER /*vAA*/)
+    {
+        Object* obj;
+
+        vsrc1 = INST_AA(inst);
+        ILOGV("|monitor-enter v%d %s(0x%08x)",
+            vsrc1, kSpacing+6, GET_REGISTER(vsrc1));
+        obj = (Object*)GET_REGISTER(vsrc1);
+        if (!checkForNullExportPC(obj, fp, pc))
+            GOTO_exceptionThrown();
+        ILOGV("+ locking %p %s\n", obj, obj->clazz->descriptor);
+#ifdef WITH_MONITOR_TRACKING
+        EXPORT_PC();        /* need for stack trace */
+#endif
+        dvmLockObject(self, obj);
+#ifdef WITH_DEADLOCK_PREDICTION
+        if (dvmCheckException(self))
+            GOTO_exceptionThrown();
+#endif
+    }
+    FINISH(1);
+OP_END
+
+/* File: c/OP_MONITOR_EXIT.c */
+HANDLE_OPCODE(OP_MONITOR_EXIT /*vAA*/)
+    {
+        Object* obj;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);
+        ILOGV("|monitor-exit v%d %s(0x%08x)",
+            vsrc1, kSpacing+5, GET_REGISTER(vsrc1));
+        obj = (Object*)GET_REGISTER(vsrc1);
+        if (!checkForNull(obj)) {
+            /*
+             * The exception needs to be processed at the *following*
+             * instruction, not the current instruction (see the Dalvik
+             * spec).  Because we're jumping to an exception handler,
+             * we're not actually at risk of skipping an instruction
+             * by doing so.
+             */
+            ADJUST_PC(1);           /* monitor-exit width is 1 */
+            GOTO_exceptionThrown();
+        }
+        ILOGV("+ unlocking %p %s\n", obj, obj->clazz->descriptor);
+        if (!dvmUnlockObject(self, obj)) {
+            assert(dvmCheckException(self));
+            ADJUST_PC(1);
+            GOTO_exceptionThrown();
+        }
+    }
+    FINISH(1);
+OP_END
+
+/* File: c/OP_CHECK_CAST.c */
+HANDLE_OPCODE(OP_CHECK_CAST /*vAA, class@BBBB*/)
+    {
+        ClassObject* clazz;
+        Object* obj;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);
+        ref = FETCH(1);         /* class to check against */
+        ILOGV("|check-cast v%d,class@0x%04x", vsrc1, ref);
+
+        obj = (Object*)GET_REGISTER(vsrc1);
+        if (obj != NULL) {
+#if defined(WITH_EXTRA_OBJECT_VALIDATION)
+            if (!checkForNull(obj))
+                GOTO_exceptionThrown();
+#endif
+            clazz = dvmDexGetResolvedClass(methodClassDex, ref);
+            if (clazz == NULL) {
+                clazz = dvmResolveClass(curMethod->clazz, ref, false);
+                if (clazz == NULL)
+                    GOTO_exceptionThrown();
+            }
+            if (!dvmInstanceof(obj->clazz, clazz)) {
+                dvmThrowExceptionWithClassMessage(
+                    "Ljava/lang/ClassCastException;", obj->clazz->descriptor);
+                GOTO_exceptionThrown();
+            }
+        }
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_INSTANCE_OF.c */
+HANDLE_OPCODE(OP_INSTANCE_OF /*vA, vB, class@CCCC*/)
+    {
+        ClassObject* clazz;
+        Object* obj;
+
+        vdst = INST_A(inst);
+        vsrc1 = INST_B(inst);   /* object to check */
+        ref = FETCH(1);         /* class to check against */
+        ILOGV("|instance-of v%d,v%d,class@0x%04x", vdst, vsrc1, ref);
+
+        obj = (Object*)GET_REGISTER(vsrc1);
+        if (obj == NULL) {
+            SET_REGISTER(vdst, 0);
+        } else {
+#if defined(WITH_EXTRA_OBJECT_VALIDATION)
+            if (!checkForNullExportPC(obj, fp, pc))
+                GOTO_exceptionThrown();
+#endif
+            clazz = dvmDexGetResolvedClass(methodClassDex, ref);
+            if (clazz == NULL) {
+                EXPORT_PC();
+                clazz = dvmResolveClass(curMethod->clazz, ref, true);
+                if (clazz == NULL)
+                    GOTO_exceptionThrown();
+            }
+            SET_REGISTER(vdst, dvmInstanceof(obj->clazz, clazz));
+        }
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_ARRAY_LENGTH.c */
+HANDLE_OPCODE(OP_ARRAY_LENGTH /*vA, vB*/)
+    {
+        ArrayObject* arrayObj;
+
+        vdst = INST_A(inst);
+        vsrc1 = INST_B(inst);
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);
+        ILOGV("|array-length v%d,v%d  (%p)", vdst, vsrc1, arrayObj);
+        if (!checkForNullExportPC((Object*) arrayObj, fp, pc))
+            GOTO_exceptionThrown();
+        /* verifier guarantees this is an array reference */
+        SET_REGISTER(vdst, arrayObj->length);
+    }
+    FINISH(1);
+OP_END
+
+/* File: c/OP_NEW_INSTANCE.c */
+HANDLE_OPCODE(OP_NEW_INSTANCE /*vAA, class@BBBB*/)
+    {
+        ClassObject* clazz;
+        Object* newObj;
+
+        EXPORT_PC();
+
+        vdst = INST_AA(inst);
+        ref = FETCH(1);
+        ILOGV("|new-instance v%d,class@0x%04x", vdst, ref);
+        clazz = dvmDexGetResolvedClass(methodClassDex, ref);
+        if (clazz == NULL) {
+            clazz = dvmResolveClass(curMethod->clazz, ref, false);
+            if (clazz == NULL)
+                GOTO_exceptionThrown();
+        }
+
+        if (!dvmIsClassInitialized(clazz) && !dvmInitClass(clazz))
+            GOTO_exceptionThrown();
+
+        /*
+         * Note: the verifier can ensure that this never happens, allowing us
+         * to remove the check.  However, the spec requires we throw the
+         * exception at runtime, not verify time, so the verifier would
+         * need to replace the new-instance call with a magic "throw
+         * InstantiationError" instruction.
+         *
+         * Since this relies on the verifier, which is optional, we would
+         * also need a "new-instance-quick" instruction to identify instances
+         * that don't require the check.
+         */
+        if (dvmIsInterfaceClass(clazz) || dvmIsAbstractClass(clazz)) {
+            dvmThrowExceptionWithClassMessage("Ljava/lang/InstantiationError;",
+                clazz->descriptor);
+            GOTO_exceptionThrown();
+        }
+        newObj = dvmAllocObject(clazz, ALLOC_DONT_TRACK);
+        if (newObj == NULL)
+            GOTO_exceptionThrown();
+        SET_REGISTER(vdst, (u4) newObj);
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_NEW_ARRAY.c */
+HANDLE_OPCODE(OP_NEW_ARRAY /*vA, vB, class@CCCC*/)
+    {
+        ClassObject* arrayClass;
+        ArrayObject* newArray;
+        s4 length;
+
+        EXPORT_PC();
+
+        vdst = INST_A(inst);
+        vsrc1 = INST_B(inst);       /* length reg */
+        ref = FETCH(1);
+        ILOGV("|new-array v%d,v%d,class@0x%04x  (%d elements)",
+            vdst, vsrc1, ref, (s4) GET_REGISTER(vsrc1));
+        length = (s4) GET_REGISTER(vsrc1);
+        if (length < 0) {
+            dvmThrowException("Ljava/lang/NegativeArraySizeException;", NULL);
+            GOTO_exceptionThrown();
+        }
+        arrayClass = dvmDexGetResolvedClass(methodClassDex, ref);
+        if (arrayClass == NULL) {
+            arrayClass = dvmResolveClass(curMethod->clazz, ref, false);
+            if (arrayClass == NULL)
+                GOTO_exceptionThrown();
+        }
+        /* verifier guarantees this is an array class */
+        assert(dvmIsArrayClass(arrayClass));
+        assert(dvmIsClassInitialized(arrayClass));
+
+        newArray = dvmAllocArrayByClass(arrayClass, length, ALLOC_DONT_TRACK);
+        if (newArray == NULL)
+            GOTO_exceptionThrown();
+        SET_REGISTER(vdst, (u4) newArray);
+    }
+    FINISH(2);
+OP_END
+
+
+/* File: c/OP_FILLED_NEW_ARRAY.c */
+HANDLE_OPCODE(OP_FILLED_NEW_ARRAY /*vB, {vD, vE, vF, vG, vA}, class@CCCC*/)
+    GOTO_invoke(filledNewArray, false);
+OP_END
+
+/* File: c/OP_FILLED_NEW_ARRAY_RANGE.c */
+HANDLE_OPCODE(OP_FILLED_NEW_ARRAY_RANGE /*{vCCCC..v(CCCC+AA-1)}, class@BBBB*/)
+    GOTO_invoke(filledNewArray, true);
+OP_END
+
+/* File: c/OP_FILL_ARRAY_DATA.c */
+HANDLE_OPCODE(OP_FILL_ARRAY_DATA)   /*vAA, +BBBBBBBB*/
+    {
+        const u2* arrayData;
+        s4 offset;
+        ArrayObject* arrayObj;
+
+        EXPORT_PC();
+        vsrc1 = INST_AA(inst);
+        offset = FETCH(1) | (((s4) FETCH(2)) << 16);
+        ILOGV("|fill-array-data v%d +0x%04x", vsrc1, offset);
+        arrayData = pc + offset;       // offset in 16-bit units
+#ifndef NDEBUG
+        if (arrayData < curMethod->insns ||
+            arrayData >= curMethod->insns + dvmGetMethodInsnsSize(curMethod))
+        {
+            /* should have been caught in verifier */
+            dvmThrowException("Ljava/lang/InternalError;", 
+                              "bad fill array data");
+            GOTO_exceptionThrown();
+        }
+#endif
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);
+        if (!dvmInterpHandleFillArrayData(arrayObj, arrayData)) {
+            GOTO_exceptionThrown();
+        }
+        FINISH(3);
+    }
+OP_END
+
+/* File: c/OP_THROW.c */
+HANDLE_OPCODE(OP_THROW /*vAA*/)
+    {
+        Object* obj;
+
+        vsrc1 = INST_AA(inst);
+        ILOGV("|throw v%d  (%p)", vsrc1, (void*)GET_REGISTER(vsrc1));
+        obj = (Object*) GET_REGISTER(vsrc1);
+        if (!checkForNullExportPC(obj, fp, pc)) {
+            /* will throw a null pointer exception */
+            LOGVV("Bad exception\n");
+        } else {
+            /* use the requested exception */
+            dvmSetException(self, obj);
+        }
+        GOTO_exceptionThrown();
+    }
+OP_END
+
+/* File: c/OP_GOTO.c */
+HANDLE_OPCODE(OP_GOTO /*+AA*/)
+    vdst = INST_AA(inst);
+    if ((s1)vdst < 0)
+        ILOGV("|goto -0x%02x", -((s1)vdst));
+    else
+        ILOGV("|goto +0x%02x", ((s1)vdst));
+    ILOGV("> branch taken");
+    if ((s1)vdst < 0)
+        PERIODIC_CHECKS(kInterpEntryInstr, (s1)vdst);
+    FINISH((s1)vdst);
+OP_END
+
+/* File: c/OP_GOTO_16.c */
+HANDLE_OPCODE(OP_GOTO_16 /*+AAAA*/)
+    {
+        s4 offset = (s2) FETCH(1);          /* sign-extend next code unit */
+
+        if (offset < 0)
+            ILOGV("|goto/16 -0x%04x", -offset);
+        else
+            ILOGV("|goto/16 +0x%04x", offset);
+        ILOGV("> branch taken");
+        if (offset < 0)
+            PERIODIC_CHECKS(kInterpEntryInstr, offset);
+        FINISH(offset);
+    }
+OP_END
+
+/* File: c/OP_GOTO_32.c */
+HANDLE_OPCODE(OP_GOTO_32 /*+AAAAAAAA*/)
+    {
+        s4 offset = FETCH(1);               /* low-order 16 bits */
+        offset |= ((s4) FETCH(2)) << 16;    /* high-order 16 bits */
+
+        if (offset < 0)
+            ILOGV("|goto/32 -0x%08x", -offset);
+        else
+            ILOGV("|goto/32 +0x%08x", offset);
+        ILOGV("> branch taken");
+        if (offset <= 0)    /* allowed to branch to self */
+            PERIODIC_CHECKS(kInterpEntryInstr, offset);
+        FINISH(offset);
+    }
+OP_END
+
+/* File: c/OP_PACKED_SWITCH.c */
+HANDLE_OPCODE(OP_PACKED_SWITCH /*vAA, +BBBB*/)
+    {
+        const u2* switchData;
+        u4 testVal;
+        s4 offset;
+
+        vsrc1 = INST_AA(inst);
+        offset = FETCH(1) | (((s4) FETCH(2)) << 16);
+        ILOGV("|packed-switch v%d +0x%04x", vsrc1, vsrc2);
+        switchData = pc + offset;       // offset in 16-bit units
+#ifndef NDEBUG
+        if (switchData < curMethod->insns ||
+            switchData >= curMethod->insns + dvmGetMethodInsnsSize(curMethod))
+        {
+            /* should have been caught in verifier */
+            EXPORT_PC();
+            dvmThrowException("Ljava/lang/InternalError;", "bad packed switch");
+            GOTO_exceptionThrown();
+        }
+#endif
+        testVal = GET_REGISTER(vsrc1);
+
+        offset = dvmInterpHandlePackedSwitch(switchData, testVal);
+        ILOGV("> branch taken (0x%04x)\n", offset);
+        if (offset <= 0)  /* uncommon */
+            PERIODIC_CHECKS(kInterpEntryInstr, offset);
+        FINISH(offset);
+    }
+OP_END
+
+/* File: c/OP_SPARSE_SWITCH.c */
+HANDLE_OPCODE(OP_SPARSE_SWITCH /*vAA, +BBBB*/)
+    {
+        const u2* switchData;
+        u4 testVal;
+        s4 offset;
+
+        vsrc1 = INST_AA(inst);
+        offset = FETCH(1) | (((s4) FETCH(2)) << 16);
+        ILOGV("|sparse-switch v%d +0x%04x", vsrc1, vsrc2);
+        switchData = pc + offset;       // offset in 16-bit units
+#ifndef NDEBUG
+        if (switchData < curMethod->insns ||
+            switchData >= curMethod->insns + dvmGetMethodInsnsSize(curMethod))
+        {
+            /* should have been caught in verifier */
+            EXPORT_PC();
+            dvmThrowException("Ljava/lang/InternalError;", "bad sparse switch");
+            GOTO_exceptionThrown();
+        }
+#endif
+        testVal = GET_REGISTER(vsrc1);
+
+        offset = dvmInterpHandleSparseSwitch(switchData, testVal);
+        ILOGV("> branch taken (0x%04x)\n", offset);
+        if (offset <= 0)  /* uncommon */
+            PERIODIC_CHECKS(kInterpEntryInstr, offset);
+        FINISH(offset);
+    }
+OP_END
+
+/* File: c/OP_CMPL_FLOAT.c */
+HANDLE_OP_CMPX(OP_CMPL_FLOAT, "l-float", float, _FLOAT, -1)
+OP_END
+
+/* File: c/OP_CMPG_FLOAT.c */
+HANDLE_OP_CMPX(OP_CMPG_FLOAT, "g-float", float, _FLOAT, 1)
+OP_END
+
+/* File: c/OP_CMPL_DOUBLE.c */
+HANDLE_OP_CMPX(OP_CMPL_DOUBLE, "l-double", double, _DOUBLE, -1)
+OP_END
+
+/* File: c/OP_CMPG_DOUBLE.c */
+HANDLE_OP_CMPX(OP_CMPG_DOUBLE, "g-double", double, _DOUBLE, 1)
+OP_END
+
+/* File: c/OP_CMP_LONG.c */
+HANDLE_OP_CMPX(OP_CMP_LONG, "-long", s8, _WIDE, 0)
+OP_END
+
+/* File: c/OP_IF_EQ.c */
+HANDLE_OP_IF_XX(OP_IF_EQ, "eq", ==)
+OP_END
+
+/* File: c/OP_IF_NE.c */
+HANDLE_OP_IF_XX(OP_IF_NE, "ne", !=)
+OP_END
+
+/* File: c/OP_IF_LT.c */
+HANDLE_OP_IF_XX(OP_IF_LT, "lt", <)
+OP_END
+
+/* File: c/OP_IF_GE.c */
+HANDLE_OP_IF_XX(OP_IF_GE, "ge", >=)
+OP_END
+
+/* File: c/OP_IF_GT.c */
+HANDLE_OP_IF_XX(OP_IF_GT, "gt", >)
+OP_END
+
+/* File: c/OP_IF_LE.c */
+HANDLE_OP_IF_XX(OP_IF_LE, "le", <=)
+OP_END
+
+/* File: c/OP_IF_EQZ.c */
+HANDLE_OP_IF_XXZ(OP_IF_EQZ, "eqz", ==)
+OP_END
+
+/* File: c/OP_IF_NEZ.c */
+HANDLE_OP_IF_XXZ(OP_IF_NEZ, "nez", !=)
+OP_END
+
+/* File: c/OP_IF_LTZ.c */
+HANDLE_OP_IF_XXZ(OP_IF_LTZ, "ltz", <)
+OP_END
+
+/* File: c/OP_IF_GEZ.c */
+HANDLE_OP_IF_XXZ(OP_IF_GEZ, "gez", >=)
+OP_END
+
+/* File: c/OP_IF_GTZ.c */
+HANDLE_OP_IF_XXZ(OP_IF_GTZ, "gtz", >)
+OP_END
+
+/* File: c/OP_IF_LEZ.c */
+HANDLE_OP_IF_XXZ(OP_IF_LEZ, "lez", <=)
+OP_END
+
+/* File: c/OP_UNUSED_3E.c */
+HANDLE_OPCODE(OP_UNUSED_3E)
+OP_END
+
+/* File: c/OP_UNUSED_3F.c */
+HANDLE_OPCODE(OP_UNUSED_3F)
+OP_END
+
+/* File: c/OP_UNUSED_40.c */
+HANDLE_OPCODE(OP_UNUSED_40)
+OP_END
+
+/* File: c/OP_UNUSED_41.c */
+HANDLE_OPCODE(OP_UNUSED_41)
+OP_END
+
+/* File: c/OP_UNUSED_42.c */
+HANDLE_OPCODE(OP_UNUSED_42)
+OP_END
+
+/* File: c/OP_UNUSED_43.c */
+HANDLE_OPCODE(OP_UNUSED_43)
+OP_END
+
+/* File: c/OP_AGET.c */
+HANDLE_OP_AGET(OP_AGET, "", u4, )
+OP_END
+
+/* File: c/OP_AGET_WIDE.c */
+HANDLE_OP_AGET(OP_AGET_WIDE, "-wide", s8, _WIDE)
+OP_END
+
+/* File: c/OP_AGET_OBJECT.c */
+HANDLE_OP_AGET(OP_AGET_OBJECT, "-object", u4, )
+OP_END
+
+/* File: c/OP_AGET_BOOLEAN.c */
+HANDLE_OP_AGET(OP_AGET_BOOLEAN, "-boolean", u1, )
+OP_END
+
+/* File: c/OP_AGET_BYTE.c */
+HANDLE_OP_AGET(OP_AGET_BYTE, "-byte", s1, )
+OP_END
+
+/* File: c/OP_AGET_CHAR.c */
+HANDLE_OP_AGET(OP_AGET_CHAR, "-char", u2, )
+OP_END
+
+/* File: c/OP_AGET_SHORT.c */
+HANDLE_OP_AGET(OP_AGET_SHORT, "-short", s2, )
+OP_END
+
+/* File: c/OP_APUT.c */
+HANDLE_OP_APUT(OP_APUT, "", u4, )
+OP_END
+
+/* File: c/OP_APUT_WIDE.c */
+HANDLE_OP_APUT(OP_APUT_WIDE, "-wide", s8, _WIDE)
+OP_END
+
+/* File: c/OP_APUT_OBJECT.c */
+HANDLE_OPCODE(OP_APUT_OBJECT /*vAA, vBB, vCC*/)
+    {
+        ArrayObject* arrayObj;
+        Object* obj;
+        u2 arrayInfo;
+        EXPORT_PC();
+        vdst = INST_AA(inst);       /* AA: source value */
+        arrayInfo = FETCH(1);
+        vsrc1 = arrayInfo & 0xff;   /* BB: array ptr */
+        vsrc2 = arrayInfo >> 8;     /* CC: index */
+        ILOGV("|aput%s v%d,v%d,v%d", "-object", vdst, vsrc1, vsrc2);
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);
+        if (!checkForNull((Object*) arrayObj))
+            GOTO_exceptionThrown();
+        if (GET_REGISTER(vsrc2) >= arrayObj->length) {
+            dvmThrowException("Ljava/lang/ArrayIndexOutOfBoundsException;",
+                NULL);
+            GOTO_exceptionThrown();
+        }
+        obj = (Object*) GET_REGISTER(vdst);
+        if (obj != NULL) {
+            if (!checkForNull(obj))
+                GOTO_exceptionThrown();
+            if (!dvmCanPutArrayElement(obj->clazz, arrayObj->obj.clazz)) {
+                LOGV("Can't put a '%s'(%p) into array type='%s'(%p)\n",
+                    obj->clazz->descriptor, obj,
+                    arrayObj->obj.clazz->descriptor, arrayObj);
+                //dvmDumpClass(obj->clazz);
+                //dvmDumpClass(arrayObj->obj.clazz);
+                dvmThrowException("Ljava/lang/ArrayStoreException;", NULL);
+                GOTO_exceptionThrown();
+            }
+        }
+        ILOGV("+ APUT[%d]=0x%08x", GET_REGISTER(vsrc2), GET_REGISTER(vdst));
+        ((u4*) arrayObj->contents)[GET_REGISTER(vsrc2)] =
+            GET_REGISTER(vdst);
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_APUT_BOOLEAN.c */
+HANDLE_OP_APUT(OP_APUT_BOOLEAN, "-boolean", u1, )
+OP_END
+
+/* File: c/OP_APUT_BYTE.c */
+HANDLE_OP_APUT(OP_APUT_BYTE, "-byte", s1, )
+OP_END
+
+/* File: c/OP_APUT_CHAR.c */
+HANDLE_OP_APUT(OP_APUT_CHAR, "-char", u2, )
+OP_END
+
+/* File: c/OP_APUT_SHORT.c */
+HANDLE_OP_APUT(OP_APUT_SHORT, "-short", s2, )
+OP_END
+
+/* File: c/OP_IGET.c */
+HANDLE_IGET_X(OP_IGET,                  "", Int, )
+OP_END
+
+/* File: c/OP_IGET_WIDE.c */
+HANDLE_IGET_X(OP_IGET_WIDE,             "-wide", Long, _WIDE)
+OP_END
+
+/* File: c/OP_IGET_OBJECT.c */
+HANDLE_IGET_X(OP_IGET_OBJECT,           "-object", Object, _AS_OBJECT)
+OP_END
+
+/* File: c/OP_IGET_BOOLEAN.c */
+HANDLE_IGET_X(OP_IGET_BOOLEAN,          "", Int, )
+OP_END
+
+/* File: c/OP_IGET_BYTE.c */
+HANDLE_IGET_X(OP_IGET_BYTE,             "", Int, )
+OP_END
+
+/* File: c/OP_IGET_CHAR.c */
+HANDLE_IGET_X(OP_IGET_CHAR,             "", Int, )
+OP_END
+
+/* File: c/OP_IGET_SHORT.c */
+HANDLE_IGET_X(OP_IGET_SHORT,            "", Int, )
+OP_END
+
+/* File: c/OP_IPUT.c */
+HANDLE_IPUT_X(OP_IPUT,                  "", Int, )
+OP_END
+
+/* File: c/OP_IPUT_WIDE.c */
+HANDLE_IPUT_X(OP_IPUT_WIDE,             "-wide", Long, _WIDE)
+OP_END
+
+/* File: c/OP_IPUT_OBJECT.c */
+/*
+ * The VM spec says we should verify that the reference being stored into
+ * the field is assignment compatible.  In practice, many popular VMs don't
+ * do this because it slows down a very common operation.  It's not so bad
+ * for us, since "dexopt" quickens it whenever possible, but it's still an
+ * issue.
+ *
+ * To make this spec-complaint, we'd need to add a ClassObject pointer to
+ * the Field struct, resolve the field's type descriptor at link or class
+ * init time, and then verify the type here.
+ */
+HANDLE_IPUT_X(OP_IPUT_OBJECT,           "-object", Object, _AS_OBJECT)
+OP_END
+
+/* File: c/OP_IPUT_BOOLEAN.c */
+HANDLE_IPUT_X(OP_IPUT_BOOLEAN,          "", Int, )
+OP_END
+
+/* File: c/OP_IPUT_BYTE.c */
+HANDLE_IPUT_X(OP_IPUT_BYTE,             "", Int, )
+OP_END
+
+/* File: c/OP_IPUT_CHAR.c */
+HANDLE_IPUT_X(OP_IPUT_CHAR,             "", Int, )
+OP_END
+
+/* File: c/OP_IPUT_SHORT.c */
+HANDLE_IPUT_X(OP_IPUT_SHORT,            "", Int, )
+OP_END
+
+/* File: c/OP_SGET.c */
+HANDLE_SGET_X(OP_SGET,                  "", Int, )
+OP_END
+
+/* File: c/OP_SGET_WIDE.c */
+HANDLE_SGET_X(OP_SGET_WIDE,             "-wide", Long, _WIDE)
+OP_END
+
+/* File: c/OP_SGET_OBJECT.c */
+HANDLE_SGET_X(OP_SGET_OBJECT,           "-object", Object, _AS_OBJECT)
+OP_END
+
+/* File: c/OP_SGET_BOOLEAN.c */
+HANDLE_SGET_X(OP_SGET_BOOLEAN,          "", Int, )
+OP_END
+
+/* File: c/OP_SGET_BYTE.c */
+HANDLE_SGET_X(OP_SGET_BYTE,             "", Int, )
+OP_END
+
+/* File: c/OP_SGET_CHAR.c */
+HANDLE_SGET_X(OP_SGET_CHAR,             "", Int, )
+OP_END
+
+/* File: c/OP_SGET_SHORT.c */
+HANDLE_SGET_X(OP_SGET_SHORT,            "", Int, )
+OP_END
+
+/* File: c/OP_SPUT.c */
+HANDLE_SPUT_X(OP_SPUT,                  "", Int, )
+OP_END
+
+/* File: c/OP_SPUT_WIDE.c */
+HANDLE_SPUT_X(OP_SPUT_WIDE,             "-wide", Long, _WIDE)
+OP_END
+
+/* File: c/OP_SPUT_OBJECT.c */
+HANDLE_SPUT_X(OP_SPUT_OBJECT,           "-object", Object, _AS_OBJECT)
+OP_END
+
+/* File: c/OP_SPUT_BOOLEAN.c */
+HANDLE_SPUT_X(OP_SPUT_BOOLEAN,          "", Int, )
+OP_END
+
+/* File: c/OP_SPUT_BYTE.c */
+HANDLE_SPUT_X(OP_SPUT_BYTE,             "", Int, )
+OP_END
+
+/* File: c/OP_SPUT_CHAR.c */
+HANDLE_SPUT_X(OP_SPUT_CHAR,             "", Int, )
+OP_END
+
+/* File: c/OP_SPUT_SHORT.c */
+HANDLE_SPUT_X(OP_SPUT_SHORT,            "", Int, )
+OP_END
+
+/* File: c/OP_INVOKE_VIRTUAL.c */
+HANDLE_OPCODE(OP_INVOKE_VIRTUAL /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeVirtual, false);
+OP_END
+
+/* File: c/OP_INVOKE_SUPER.c */
+HANDLE_OPCODE(OP_INVOKE_SUPER /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeSuper, false);
+OP_END
+
+/* File: c/OP_INVOKE_DIRECT.c */
+HANDLE_OPCODE(OP_INVOKE_DIRECT /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeDirect, false);
+OP_END
+
+/* File: c/OP_INVOKE_STATIC.c */
+HANDLE_OPCODE(OP_INVOKE_STATIC /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeStatic, false);
+OP_END
+
+/* File: c/OP_INVOKE_INTERFACE.c */
+HANDLE_OPCODE(OP_INVOKE_INTERFACE /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeInterface, false);
+OP_END
+
+/* File: c/OP_UNUSED_73.c */
+HANDLE_OPCODE(OP_UNUSED_73)
+OP_END
+
+/* File: c/OP_INVOKE_VIRTUAL_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_VIRTUAL_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeVirtual, true);
+OP_END
+
+/* File: c/OP_INVOKE_SUPER_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_SUPER_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeSuper, true);
+OP_END
+
+/* File: c/OP_INVOKE_DIRECT_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_DIRECT_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeDirect, true);
+OP_END
+
+/* File: c/OP_INVOKE_STATIC_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_STATIC_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeStatic, true);
+OP_END
+
+/* File: c/OP_INVOKE_INTERFACE_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_INTERFACE_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeInterface, true);
+OP_END
+
+/* File: c/OP_UNUSED_79.c */
+HANDLE_OPCODE(OP_UNUSED_79)
+OP_END
+
+/* File: c/OP_UNUSED_7A.c */
+HANDLE_OPCODE(OP_UNUSED_7A)
+OP_END
+
+/* File: c/OP_NEG_INT.c */
+HANDLE_UNOP(OP_NEG_INT, "neg-int", -, , )
+OP_END
+
+/* File: c/OP_NOT_INT.c */
+HANDLE_UNOP(OP_NOT_INT, "not-int", , ^ 0xffffffff, )
+OP_END
+
+/* File: c/OP_NEG_LONG.c */
+HANDLE_UNOP(OP_NEG_LONG, "neg-long", -, , _WIDE)
+OP_END
+
+/* File: c/OP_NOT_LONG.c */
+HANDLE_UNOP(OP_NOT_LONG, "not-long", , ^ 0xffffffffffffffffULL, _WIDE)
+OP_END
+
+/* File: c/OP_NEG_FLOAT.c */
+HANDLE_UNOP(OP_NEG_FLOAT, "neg-float", -, , _FLOAT)
+OP_END
+
+/* File: c/OP_NEG_DOUBLE.c */
+HANDLE_UNOP(OP_NEG_DOUBLE, "neg-double", -, , _DOUBLE)
+OP_END
+
+/* File: c/OP_INT_TO_LONG.c */
+HANDLE_NUMCONV(OP_INT_TO_LONG,          "int-to-long", _INT, _WIDE)
+OP_END
+
+/* File: c/OP_INT_TO_FLOAT.c */
+HANDLE_NUMCONV(OP_INT_TO_FLOAT,         "int-to-float", _INT, _FLOAT)
+OP_END
+
+/* File: c/OP_INT_TO_DOUBLE.c */
+HANDLE_NUMCONV(OP_INT_TO_DOUBLE,        "int-to-double", _INT, _DOUBLE)
+OP_END
+
+/* File: c/OP_LONG_TO_INT.c */
+HANDLE_NUMCONV(OP_LONG_TO_INT,          "long-to-int", _WIDE, _INT)
+OP_END
+
+/* File: c/OP_LONG_TO_FLOAT.c */
+HANDLE_NUMCONV(OP_LONG_TO_FLOAT,        "long-to-float", _WIDE, _FLOAT)
+OP_END
+
+/* File: c/OP_LONG_TO_DOUBLE.c */
+HANDLE_NUMCONV(OP_LONG_TO_DOUBLE,       "long-to-double", _WIDE, _DOUBLE)
+OP_END
+
+/* File: c/OP_FLOAT_TO_INT.c */
+HANDLE_FLOAT_TO_INT(OP_FLOAT_TO_INT,    "float-to-int",
+    float, _FLOAT, s4, _INT)
+OP_END
+
+/* File: c/OP_FLOAT_TO_LONG.c */
+HANDLE_FLOAT_TO_INT(OP_FLOAT_TO_LONG,   "float-to-long",
+    float, _FLOAT, s8, _WIDE)
+OP_END
+
+/* File: c/OP_FLOAT_TO_DOUBLE.c */
+HANDLE_NUMCONV(OP_FLOAT_TO_DOUBLE,      "float-to-double", _FLOAT, _DOUBLE)
+OP_END
+
+/* File: c/OP_DOUBLE_TO_INT.c */
+HANDLE_FLOAT_TO_INT(OP_DOUBLE_TO_INT,   "double-to-int",
+    double, _DOUBLE, s4, _INT)
+OP_END
+
+/* File: c/OP_DOUBLE_TO_LONG.c */
+HANDLE_FLOAT_TO_INT(OP_DOUBLE_TO_LONG,  "double-to-long",
+    double, _DOUBLE, s8, _WIDE)
+OP_END
+
+/* File: c/OP_DOUBLE_TO_FLOAT.c */
+HANDLE_NUMCONV(OP_DOUBLE_TO_FLOAT,      "double-to-float", _DOUBLE, _FLOAT)
+OP_END
+
+/* File: c/OP_INT_TO_BYTE.c */
+HANDLE_INT_TO_SMALL(OP_INT_TO_BYTE,     "byte", s1)
+OP_END
+
+/* File: c/OP_INT_TO_CHAR.c */
+HANDLE_INT_TO_SMALL(OP_INT_TO_CHAR,     "char", u2)
+OP_END
+
+/* File: c/OP_INT_TO_SHORT.c */
+HANDLE_INT_TO_SMALL(OP_INT_TO_SHORT,    "short", s2)    /* want sign bit */
+OP_END
+
+/* File: c/OP_ADD_INT.c */
+HANDLE_OP_X_INT(OP_ADD_INT, "add", +, 0)
+OP_END
+
+/* File: c/OP_SUB_INT.c */
+HANDLE_OP_X_INT(OP_SUB_INT, "sub", -, 0)
+OP_END
+
+/* File: c/OP_MUL_INT.c */
+HANDLE_OP_X_INT(OP_MUL_INT, "mul", *, 0)
+OP_END
+
+/* File: c/OP_DIV_INT.c */
+HANDLE_OP_X_INT(OP_DIV_INT, "div", /, 1)
+OP_END
+
+/* File: c/OP_REM_INT.c */
+HANDLE_OP_X_INT(OP_REM_INT, "rem", %, 2)
+OP_END
+
+/* File: c/OP_AND_INT.c */
+HANDLE_OP_X_INT(OP_AND_INT, "and", &, 0)
+OP_END
+
+/* File: c/OP_OR_INT.c */
+HANDLE_OP_X_INT(OP_OR_INT,  "or",  |, 0)
+OP_END
+
+/* File: c/OP_XOR_INT.c */
+HANDLE_OP_X_INT(OP_XOR_INT, "xor", ^, 0)
+OP_END
+
+/* File: c/OP_SHL_INT.c */
+HANDLE_OP_SHX_INT(OP_SHL_INT, "shl", (s4), <<)
+OP_END
+
+/* File: c/OP_SHR_INT.c */
+HANDLE_OP_SHX_INT(OP_SHR_INT, "shr", (s4), >>)
+OP_END
+
+/* File: c/OP_USHR_INT.c */
+HANDLE_OP_SHX_INT(OP_USHR_INT, "ushr", (u4), >>)
+OP_END
+
+/* File: c/OP_ADD_LONG.c */
+HANDLE_OP_X_LONG(OP_ADD_LONG, "add", +, 0)
+OP_END
+
+/* File: c/OP_SUB_LONG.c */
+HANDLE_OP_X_LONG(OP_SUB_LONG, "sub", -, 0)
+OP_END
+
+/* File: c/OP_MUL_LONG.c */
+HANDLE_OP_X_LONG(OP_MUL_LONG, "mul", *, 0)
+OP_END
+
+/* File: c/OP_DIV_LONG.c */
+HANDLE_OP_X_LONG(OP_DIV_LONG, "div", /, 1)
+OP_END
+
+/* File: c/OP_REM_LONG.c */
+HANDLE_OP_X_LONG(OP_REM_LONG, "rem", %, 2)
+OP_END
+
+/* File: c/OP_AND_LONG.c */
+HANDLE_OP_X_LONG(OP_AND_LONG, "and", &, 0)
+OP_END
+
+/* File: c/OP_OR_LONG.c */
+HANDLE_OP_X_LONG(OP_OR_LONG,  "or", |, 0)
+OP_END
+
+/* File: c/OP_XOR_LONG.c */
+HANDLE_OP_X_LONG(OP_XOR_LONG, "xor", ^, 0)
+OP_END
+
+/* File: c/OP_SHL_LONG.c */
+HANDLE_OP_SHX_LONG(OP_SHL_LONG, "shl", (s8), <<)
+OP_END
+
+/* File: c/OP_SHR_LONG.c */
+HANDLE_OP_SHX_LONG(OP_SHR_LONG, "shr", (s8), >>)
+OP_END
+
+/* File: c/OP_USHR_LONG.c */
+HANDLE_OP_SHX_LONG(OP_USHR_LONG, "ushr", (u8), >>)
+OP_END
+
+/* File: c/OP_ADD_FLOAT.c */
+HANDLE_OP_X_FLOAT(OP_ADD_FLOAT, "add", +)
+OP_END
+
+/* File: c/OP_SUB_FLOAT.c */
+HANDLE_OP_X_FLOAT(OP_SUB_FLOAT, "sub", -)
+OP_END
+
+/* File: c/OP_MUL_FLOAT.c */
+HANDLE_OP_X_FLOAT(OP_MUL_FLOAT, "mul", *)
+OP_END
+
+/* File: c/OP_DIV_FLOAT.c */
+HANDLE_OP_X_FLOAT(OP_DIV_FLOAT, "div", /)
+OP_END
+
+/* File: c/OP_REM_FLOAT.c */
+HANDLE_OPCODE(OP_REM_FLOAT /*vAA, vBB, vCC*/)
+    {
+        u2 srcRegs;
+        vdst = INST_AA(inst);
+        srcRegs = FETCH(1);
+        vsrc1 = srcRegs & 0xff;
+        vsrc2 = srcRegs >> 8;
+        ILOGV("|%s-float v%d,v%d,v%d", "mod", vdst, vsrc1, vsrc2);
+        SET_REGISTER_FLOAT(vdst,
+            fmodf(GET_REGISTER_FLOAT(vsrc1), GET_REGISTER_FLOAT(vsrc2)));
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_ADD_DOUBLE.c */
+HANDLE_OP_X_DOUBLE(OP_ADD_DOUBLE, "add", +)
+OP_END
+
+/* File: c/OP_SUB_DOUBLE.c */
+HANDLE_OP_X_DOUBLE(OP_SUB_DOUBLE, "sub", -)
+OP_END
+
+/* File: c/OP_MUL_DOUBLE.c */
+HANDLE_OP_X_DOUBLE(OP_MUL_DOUBLE, "mul", *)
+OP_END
+
+/* File: c/OP_DIV_DOUBLE.c */
+HANDLE_OP_X_DOUBLE(OP_DIV_DOUBLE, "div", /)
+OP_END
+
+/* File: c/OP_REM_DOUBLE.c */
+HANDLE_OPCODE(OP_REM_DOUBLE /*vAA, vBB, vCC*/)
+    {
+        u2 srcRegs;
+        vdst = INST_AA(inst);
+        srcRegs = FETCH(1);
+        vsrc1 = srcRegs & 0xff;
+        vsrc2 = srcRegs >> 8;
+        ILOGV("|%s-double v%d,v%d,v%d", "mod", vdst, vsrc1, vsrc2);
+        SET_REGISTER_DOUBLE(vdst,
+            fmod(GET_REGISTER_DOUBLE(vsrc1), GET_REGISTER_DOUBLE(vsrc2)));
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_ADD_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_ADD_INT_2ADDR, "add", +, 0)
+OP_END
+
+/* File: c/OP_SUB_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_SUB_INT_2ADDR, "sub", -, 0)
+OP_END
+
+/* File: c/OP_MUL_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_MUL_INT_2ADDR, "mul", *, 0)
+OP_END
+
+/* File: c/OP_DIV_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_DIV_INT_2ADDR, "div", /, 1)
+OP_END
+
+/* File: c/OP_REM_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_REM_INT_2ADDR, "rem", %, 2)
+OP_END
+
+/* File: c/OP_AND_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_AND_INT_2ADDR, "and", &, 0)
+OP_END
+
+/* File: c/OP_OR_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_OR_INT_2ADDR,  "or", |, 0)
+OP_END
+
+/* File: c/OP_XOR_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_XOR_INT_2ADDR, "xor", ^, 0)
+OP_END
+
+/* File: c/OP_SHL_INT_2ADDR.c */
+HANDLE_OP_SHX_INT_2ADDR(OP_SHL_INT_2ADDR, "shl", (s4), <<)
+OP_END
+
+/* File: c/OP_SHR_INT_2ADDR.c */
+HANDLE_OP_SHX_INT_2ADDR(OP_SHR_INT_2ADDR, "shr", (s4), >>)
+OP_END
+
+/* File: c/OP_USHR_INT_2ADDR.c */
+HANDLE_OP_SHX_INT_2ADDR(OP_USHR_INT_2ADDR, "ushr", (u4), >>)
+OP_END
+
+/* File: c/OP_ADD_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_ADD_LONG_2ADDR, "add", +, 0)
+OP_END
+
+/* File: c/OP_SUB_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_SUB_LONG_2ADDR, "sub", -, 0)
+OP_END
+
+/* File: c/OP_MUL_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_MUL_LONG_2ADDR, "mul", *, 0)
+OP_END
+
+/* File: c/OP_DIV_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_DIV_LONG_2ADDR, "div", /, 1)
+OP_END
+
+/* File: c/OP_REM_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_REM_LONG_2ADDR, "rem", %, 2)
+OP_END
+
+/* File: c/OP_AND_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_AND_LONG_2ADDR, "and", &, 0)
+OP_END
+
+/* File: c/OP_OR_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_OR_LONG_2ADDR,  "or", |, 0)
+OP_END
+
+/* File: c/OP_XOR_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_XOR_LONG_2ADDR, "xor", ^, 0)
+OP_END
+
+/* File: c/OP_SHL_LONG_2ADDR.c */
+HANDLE_OP_SHX_LONG_2ADDR(OP_SHL_LONG_2ADDR, "shl", (s8), <<)
+OP_END
+
+/* File: c/OP_SHR_LONG_2ADDR.c */
+HANDLE_OP_SHX_LONG_2ADDR(OP_SHR_LONG_2ADDR, "shr", (s8), >>)
+OP_END
+
+/* File: c/OP_USHR_LONG_2ADDR.c */
+HANDLE_OP_SHX_LONG_2ADDR(OP_USHR_LONG_2ADDR, "ushr", (u8), >>)
+OP_END
+
+/* File: c/OP_ADD_FLOAT_2ADDR.c */
+HANDLE_OP_X_FLOAT_2ADDR(OP_ADD_FLOAT_2ADDR, "add", +)
+OP_END
+
+/* File: c/OP_SUB_FLOAT_2ADDR.c */
+HANDLE_OP_X_FLOAT_2ADDR(OP_SUB_FLOAT_2ADDR, "sub", -)
+OP_END
+
+/* File: c/OP_MUL_FLOAT_2ADDR.c */
+HANDLE_OP_X_FLOAT_2ADDR(OP_MUL_FLOAT_2ADDR, "mul", *)
+OP_END
+
+/* File: c/OP_DIV_FLOAT_2ADDR.c */
+HANDLE_OP_X_FLOAT_2ADDR(OP_DIV_FLOAT_2ADDR, "div", /)
+OP_END
+
+/* File: c/OP_REM_FLOAT_2ADDR.c */
+HANDLE_OPCODE(OP_REM_FLOAT_2ADDR /*vA, vB*/)
+    vdst = INST_A(inst);
+    vsrc1 = INST_B(inst);
+    ILOGV("|%s-float-2addr v%d,v%d", "mod", vdst, vsrc1);
+    SET_REGISTER_FLOAT(vdst,
+        fmodf(GET_REGISTER_FLOAT(vdst), GET_REGISTER_FLOAT(vsrc1)));
+    FINISH(1);
+OP_END
+
+/* File: c/OP_ADD_DOUBLE_2ADDR.c */
+HANDLE_OP_X_DOUBLE_2ADDR(OP_ADD_DOUBLE_2ADDR, "add", +)
+OP_END
+
+/* File: c/OP_SUB_DOUBLE_2ADDR.c */
+HANDLE_OP_X_DOUBLE_2ADDR(OP_SUB_DOUBLE_2ADDR, "sub", -)
+OP_END
+
+/* File: c/OP_MUL_DOUBLE_2ADDR.c */
+HANDLE_OP_X_DOUBLE_2ADDR(OP_MUL_DOUBLE_2ADDR, "mul", *)
+OP_END
+
+/* File: c/OP_DIV_DOUBLE_2ADDR.c */
+HANDLE_OP_X_DOUBLE_2ADDR(OP_DIV_DOUBLE_2ADDR, "div", /)
+OP_END
+
+/* File: c/OP_REM_DOUBLE_2ADDR.c */
+HANDLE_OPCODE(OP_REM_DOUBLE_2ADDR /*vA, vB*/)
+    vdst = INST_A(inst);
+    vsrc1 = INST_B(inst);
+    ILOGV("|%s-double-2addr v%d,v%d", "mod", vdst, vsrc1);
+    SET_REGISTER_DOUBLE(vdst,
+        fmod(GET_REGISTER_DOUBLE(vdst), GET_REGISTER_DOUBLE(vsrc1)));
+    FINISH(1);
+OP_END
+
+/* File: c/OP_ADD_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_ADD_INT_LIT16, "add", +, 0)
+OP_END
+
+/* File: c/OP_RSUB_INT.c */
+HANDLE_OPCODE(OP_RSUB_INT /*vA, vB, #+CCCC*/)
+    {
+        vdst = INST_A(inst);
+        vsrc1 = INST_B(inst);
+        vsrc2 = FETCH(1);
+        ILOGV("|rsub-int v%d,v%d,#+0x%04x", vdst, vsrc1, vsrc2);
+        SET_REGISTER(vdst, (s2) vsrc2 - (s4) GET_REGISTER(vsrc1));
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_MUL_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_MUL_INT_LIT16, "mul", *, 0)
+OP_END
+
+/* File: c/OP_DIV_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_DIV_INT_LIT16, "div", /, 1)
+OP_END
+
+/* File: c/OP_REM_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_REM_INT_LIT16, "rem", %, 2)
+OP_END
+
+/* File: c/OP_AND_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_AND_INT_LIT16, "and", &, 0)
+OP_END
+
+/* File: c/OP_OR_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_OR_INT_LIT16,  "or",  |, 0)
+OP_END
+
+/* File: c/OP_XOR_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_XOR_INT_LIT16, "xor", ^, 0)
+OP_END
+
+/* File: c/OP_ADD_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_ADD_INT_LIT8,   "add", +, 0)
+OP_END
+
+/* File: c/OP_RSUB_INT_LIT8.c */
+HANDLE_OPCODE(OP_RSUB_INT_LIT8 /*vAA, vBB, #+CC*/)
+    {
+        u2 litInfo;
+        vdst = INST_AA(inst);
+        litInfo = FETCH(1);
+        vsrc1 = litInfo & 0xff;
+        vsrc2 = litInfo >> 8;
+        ILOGV("|%s-int/lit8 v%d,v%d,#+0x%02x", "rsub", vdst, vsrc1, vsrc2);
+        SET_REGISTER(vdst, (s1) vsrc2 - (s4) GET_REGISTER(vsrc1));
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_MUL_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_MUL_INT_LIT8,   "mul", *, 0)
+OP_END
+
+/* File: c/OP_DIV_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_DIV_INT_LIT8,   "div", /, 1)
+OP_END
+
+/* File: c/OP_REM_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_REM_INT_LIT8,   "rem", %, 2)
+OP_END
+
+/* File: c/OP_AND_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_AND_INT_LIT8,   "and", &, 0)
+OP_END
+
+/* File: c/OP_OR_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_OR_INT_LIT8,    "or",  |, 0)
+OP_END
+
+/* File: c/OP_XOR_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_XOR_INT_LIT8,   "xor", ^, 0)
+OP_END
+
+/* File: c/OP_SHL_INT_LIT8.c */
+HANDLE_OP_SHX_INT_LIT8(OP_SHL_INT_LIT8,   "shl", (s4), <<)
+OP_END
+
+/* File: c/OP_SHR_INT_LIT8.c */
+HANDLE_OP_SHX_INT_LIT8(OP_SHR_INT_LIT8,   "shr", (s4), >>)
+OP_END
+
+/* File: c/OP_USHR_INT_LIT8.c */
+HANDLE_OP_SHX_INT_LIT8(OP_USHR_INT_LIT8,  "ushr", (u4), >>)
+OP_END
+
+/* File: c/OP_UNUSED_E3.c */
+HANDLE_OPCODE(OP_UNUSED_E3)
+OP_END
+
+/* File: c/OP_UNUSED_E4.c */
+HANDLE_OPCODE(OP_UNUSED_E4)
+OP_END
+
+/* File: c/OP_UNUSED_E5.c */
+HANDLE_OPCODE(OP_UNUSED_E5)
+OP_END
+
+/* File: c/OP_UNUSED_E6.c */
+HANDLE_OPCODE(OP_UNUSED_E6)
+OP_END
+
+/* File: c/OP_UNUSED_E7.c */
+HANDLE_OPCODE(OP_UNUSED_E7)
+OP_END
+
+/* File: c/OP_UNUSED_E8.c */
+HANDLE_OPCODE(OP_UNUSED_E8)
+OP_END
+
+/* File: c/OP_UNUSED_E9.c */
+HANDLE_OPCODE(OP_UNUSED_E9)
+OP_END
+
+/* File: c/OP_UNUSED_EA.c */
+HANDLE_OPCODE(OP_UNUSED_EA)
+OP_END
+
+/* File: c/OP_UNUSED_EB.c */
+HANDLE_OPCODE(OP_UNUSED_EB)
+OP_END
+
+/* File: c/OP_UNUSED_EC.c */
+HANDLE_OPCODE(OP_UNUSED_EC)
+OP_END
+
+/* File: c/OP_UNUSED_ED.c */
+HANDLE_OPCODE(OP_UNUSED_ED)
+OP_END
+
+/* File: c/OP_EXECUTE_INLINE.c */
+HANDLE_OPCODE(OP_EXECUTE_INLINE /*vB, {vD, vE, vF, vG}, inline@CCCC*/)
+    {
+        /*
+         * This has the same form as other method calls, but we ignore
+         * the 5th argument (vA).  This is chiefly because the first four
+         * arguments to a function on ARM are in registers.
+         *
+         * We only set the arguments that are actually used, leaving
+         * the rest uninitialized.  We're assuming that, if the method
+         * needs them, they'll be specified in the call.
+         *
+         * This annoys gcc when optimizations are enabled, causing a
+         * "may be used uninitialized" warning.  We can quiet the warnings
+         * for a slight penalty (5%: 373ns vs. 393ns on empty method).  Note
+         * that valgrind is perfectly happy with this arrangement, because
+         * the uninitialiezd values are never actually used.
+         */
+        u4 arg0, arg1, arg2, arg3;
+        //arg0 = arg1 = arg2 = arg3 = 0;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_B(inst);       /* #of args */
+        ref = FETCH(1);             /* inline call "ref" */
+        vdst = FETCH(2);            /* 0-4 register indices */
+        ILOGV("|execute-inline args=%d @%d {regs=0x%04x}",
+            vsrc1, ref, vdst);
+
+        assert((vdst >> 16) == 0);  // 16-bit type -or- high 16 bits clear
+        assert(vsrc1 <= 4);
+
+        switch (vsrc1) {
+        case 4:
+            arg3 = GET_REGISTER(vdst >> 12);
+            /* fall through */
+        case 3:
+            arg2 = GET_REGISTER((vdst & 0x0f00) >> 8);
+            /* fall through */
+        case 2:
+            arg1 = GET_REGISTER((vdst & 0x00f0) >> 4);
+            /* fall through */
+        case 1:
+            arg0 = GET_REGISTER(vdst & 0x0f);
+            /* fall through */
+        default:        // case 0
+            ;
+        }
+
+#if INTERP_TYPE == INTERP_DBG
+        if (!dvmPerformInlineOp4Dbg(arg0, arg1, arg2, arg3, &retval, ref))
+            GOTO_exceptionThrown();
+#else
+        if (!dvmPerformInlineOp4Std(arg0, arg1, arg2, arg3, &retval, ref))
+            GOTO_exceptionThrown();
+#endif
+    }
+    FINISH(3);
+OP_END
+
+/* File: c/OP_UNUSED_EF.c */
+HANDLE_OPCODE(OP_UNUSED_EF)
+OP_END
+
+/* File: c/OP_INVOKE_DIRECT_EMPTY.c */
+HANDLE_OPCODE(OP_INVOKE_DIRECT_EMPTY /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+#if INTERP_TYPE != INTERP_DBG
+    //LOGI("Ignoring empty\n");
+    FINISH(3);
+#else
+    if (!gDvm.debuggerActive) {
+        //LOGI("Skipping empty\n");
+        FINISH(3);      // don't want it to show up in profiler output
+    } else {
+        //LOGI("Running empty\n");
+        /* fall through to OP_INVOKE_DIRECT */
+        GOTO_invoke(invokeDirect, false);
+    }
+#endif
+OP_END
+
+/* File: c/OP_UNUSED_F1.c */
+HANDLE_OPCODE(OP_UNUSED_F1)
+OP_END
+
+/* File: c/OP_IGET_QUICK.c */
+HANDLE_IGET_X_QUICK(OP_IGET_QUICK,          "", Int, )
+OP_END
+
+/* File: c/OP_IGET_WIDE_QUICK.c */
+HANDLE_IGET_X_QUICK(OP_IGET_WIDE_QUICK,     "-wide", Long, _WIDE)
+OP_END
+
+/* File: c/OP_IGET_OBJECT_QUICK.c */
+HANDLE_IGET_X_QUICK(OP_IGET_OBJECT_QUICK,   "-object", Object, _AS_OBJECT)
+OP_END
+
+/* File: c/OP_IPUT_QUICK.c */
+HANDLE_IPUT_X_QUICK(OP_IPUT_QUICK,          "", Int, )
+OP_END
+
+/* File: c/OP_IPUT_WIDE_QUICK.c */
+HANDLE_IPUT_X_QUICK(OP_IPUT_WIDE_QUICK,     "-wide", Long, _WIDE)
+OP_END
+
+/* File: c/OP_IPUT_OBJECT_QUICK.c */
+HANDLE_IPUT_X_QUICK(OP_IPUT_OBJECT_QUICK,   "-object", Object, _AS_OBJECT)
+OP_END
+
+/* File: c/OP_INVOKE_VIRTUAL_QUICK.c */
+HANDLE_OPCODE(OP_INVOKE_VIRTUAL_QUICK /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeVirtualQuick, false);
+OP_END
+
+/* File: c/OP_INVOKE_VIRTUAL_QUICK_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_VIRTUAL_QUICK_RANGE/*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeVirtualQuick, true);
+OP_END
+
+/* File: c/OP_INVOKE_SUPER_QUICK.c */
+HANDLE_OPCODE(OP_INVOKE_SUPER_QUICK /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeSuperQuick, false);
+OP_END
+
+/* File: c/OP_INVOKE_SUPER_QUICK_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_SUPER_QUICK_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeSuperQuick, true);
+OP_END
+
+/* File: c/OP_UNUSED_FC.c */
+HANDLE_OPCODE(OP_UNUSED_FC)
+OP_END
+
+/* File: c/OP_UNUSED_FD.c */
+HANDLE_OPCODE(OP_UNUSED_FD)
+OP_END
+
+/* File: c/OP_UNUSED_FE.c */
+HANDLE_OPCODE(OP_UNUSED_FE)
+OP_END
+
+/* File: c/OP_UNUSED_FF.c */
+HANDLE_OPCODE(OP_UNUSED_FF)
+    /*
+     * In portable interp, most unused opcodes will fall through to here.
+     */
+    LOGE("unknown opcode 0x%02x\n", INST_INST(inst));
+    dvmAbort();
+    FINISH(1);
+OP_END
+
+/* File: cstubs/entry.c */
+/*
+ * Handler function table, one entry per opcode.
+ */
+#undef H
+#define H(_op) dvmMterp_##_op
+DEFINE_GOTO_TABLE(gDvmMterpHandlers)
+
+#undef H
+#define H(_op) #_op
+DEFINE_GOTO_TABLE(gDvmMterpHandlerNames)
+
+#include <setjmp.h>
+
+/*
+ * C mterp entry point.  This just calls the various C fallbacks, making
+ * this a slow but portable interpeter.
+ *
+ * This is only used for the "allstubs" variant.
+ */
+bool dvmMterpStdRun(MterpGlue* glue)
+{
+    jmp_buf jmpBuf;
+    int changeInterp;
+
+    glue->bailPtr = &jmpBuf;
+
+    /*
+     * We want to return "changeInterp" as a boolean, but we can't return
+     * zero through longjmp, so we return (boolean+1).
+     */
+    changeInterp = setjmp(jmpBuf) -1;
+    if (changeInterp >= 0) {
+        Thread* threadSelf = dvmThreadSelf();
+        LOGVV("mterp threadid=%d returning %d\n",
+            threadSelf->threadId, changeInterp);
+        return changeInterp;
+    }
+
+    /*
+     * We may not be starting at a point where we're executing instructions.
+     * We need to pick up where the other interpreter left off.
+     *
+     * In some cases we need to call into a throw/return handler which
+     * will do some processing and then either return to us (updating "glue")
+     * or longjmp back out.
+     */
+    switch (glue->entryPoint) {
+    case kInterpEntryInstr:
+        /* just start at the start */
+        break;
+    case kInterpEntryReturn:
+        dvmMterp_returnFromMethod(glue);
+        break;
+    case kInterpEntryThrow:
+        dvmMterp_exceptionThrown(glue);
+        break;
+    default:
+        dvmAbort();
+    }
+
+    /* run until somebody longjmp()s out */
+    while (true) {
+        typedef void (*Handler)(MterpGlue* glue);
+
+        u2 inst = /*glue->*/pc[0];
+        Handler handler = (Handler) gDvmMterpHandlers[inst & 0xff];
+        LOGVV("handler %p %s\n",
+            handler, (const char*) gDvmMterpHandlerNames[inst & 0xff]);
+        (*handler)(glue);
+    }
+}
+
+/*
+ * C mterp exit point.  Call here to bail out of the interpreter.
+ */
+void dvmMterpStdBail(MterpGlue* glue, bool changeInterp)
+{
+    jmp_buf* pJmpBuf = glue->bailPtr;
+    longjmp(*pJmpBuf, ((int)changeInterp)+1);
+}
+
+
+/* File: c/gotoTargets.c */
+/*
+ * C footer.  This has some common code shared by the various targets.
+ */
+
+/*
+ * Everything from here on is a "goto target".  In the basic interpreter
+ * we jump into these targets and then jump directly to the handler for
+ * next instruction.  Here, these are subroutines that return to the caller.
+ */
+
+GOTO_TARGET(filledNewArray, bool methodCallRange)
+    {
+        ClassObject* arrayClass;
+        ArrayObject* newArray;
+        u4* contents;
+        char typeCh;
+        int i;
+        u4 arg5;
+
+        EXPORT_PC();
+
+        ref = FETCH(1);             /* class ref */
+        vdst = FETCH(2);            /* first 4 regs -or- range base */
+
+        if (methodCallRange) {
+            vsrc1 = INST_AA(inst);  /* #of elements */
+            arg5 = -1;              /* silence compiler warning */
+            ILOGV("|filled-new-array-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+        } else {
+            arg5 = INST_A(inst);
+            vsrc1 = INST_B(inst);   /* #of elements */
+            ILOGV("|filled-new-array args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1, ref, vdst, arg5);
+        }
+
+        /*
+         * Resolve the array class.
+         */
+        arrayClass = dvmDexGetResolvedClass(methodClassDex, ref);
+        if (arrayClass == NULL) {
+            arrayClass = dvmResolveClass(curMethod->clazz, ref, false);
+            if (arrayClass == NULL)
+                GOTO_exceptionThrown();
+        }
+        /*
+        if (!dvmIsArrayClass(arrayClass)) {
+            dvmThrowException("Ljava/lang/RuntimeError;",
+                "filled-new-array needs array class");
+            GOTO_exceptionThrown();
+        }
+        */
+        /* verifier guarantees this is an array class */
+        assert(dvmIsArrayClass(arrayClass));
+        assert(dvmIsClassInitialized(arrayClass));
+
+        /*
+         * Create an array of the specified type.
+         */
+        LOGVV("+++ filled-new-array type is '%s'\n", arrayClass->descriptor);
+        typeCh = arrayClass->descriptor[1];
+        if (typeCh == 'D' || typeCh == 'J') {
+            /* category 2 primitives not allowed */
+            dvmThrowException("Ljava/lang/RuntimeError;",
+                "bad filled array req");
+            GOTO_exceptionThrown();
+        } else if (typeCh != 'L' && typeCh != '[' && typeCh != 'I') {
+            /* TODO: requires multiple "fill in" loops with different widths */
+            LOGE("non-int primitives not implemented\n");
+            dvmThrowException("Ljava/lang/InternalError;",
+                "filled-new-array not implemented for anything but 'int'");
+            GOTO_exceptionThrown();
+        }
+
+        newArray = dvmAllocArrayByClass(arrayClass, vsrc1, ALLOC_DONT_TRACK);
+        if (newArray == NULL)
+            GOTO_exceptionThrown();
+
+        /*
+         * Fill in the elements.  It's legal for vsrc1 to be zero.
+         */
+        contents = (u4*) newArray->contents;
+        if (methodCallRange) {
+            for (i = 0; i < vsrc1; i++)
+                contents[i] = GET_REGISTER(vdst+i);
+        } else {
+            assert(vsrc1 <= 5);
+            if (vsrc1 == 5) {
+                contents[4] = GET_REGISTER(arg5);
+                vsrc1--;
+            }
+            for (i = 0; i < vsrc1; i++) {
+                contents[i] = GET_REGISTER(vdst & 0x0f);
+                vdst >>= 4;
+            }
+        }
+
+        retval.l = newArray;
+    }
+    FINISH(3);
+GOTO_TARGET_END
+
+
+GOTO_TARGET(invokeVirtual, bool methodCallRange)
+    {
+        Method* baseMethod;
+        Object* thisPtr;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* method ref */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        /*
+         * The object against which we are executing a method is always
+         * in the first argument.
+         */
+        if (methodCallRange) {
+            assert(vsrc1 > 0);
+            ILOGV("|invoke-virtual-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisPtr = (Object*) GET_REGISTER(vdst);
+        } else {
+            assert((vsrc1>>4) > 0);
+            ILOGV("|invoke-virtual args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisPtr = (Object*) GET_REGISTER(vdst & 0x0f);
+        }
+
+        if (!checkForNull(thisPtr))
+            GOTO_exceptionThrown();
+
+        /*
+         * Resolve the method.  This is the correct method for the static
+         * type of the object.  We also verify access permissions here.
+         */
+        baseMethod = dvmDexGetResolvedMethod(methodClassDex, ref);
+        if (baseMethod == NULL) {
+            baseMethod = dvmResolveMethod(curMethod->clazz, ref,METHOD_VIRTUAL);
+            if (baseMethod == NULL) {
+                ILOGV("+ unknown method or access denied\n");
+                GOTO_exceptionThrown();
+            }
+        }
+
+        /*
+         * Combine the object we found with the vtable offset in the
+         * method.
+         */
+        assert(baseMethod->methodIndex < thisPtr->clazz->vtableCount);
+        methodToCall = thisPtr->clazz->vtable[baseMethod->methodIndex];
+
+#if 0
+        if (dvmIsAbstractMethod(methodToCall)) {
+            /*
+             * This can happen if you create two classes, Base and Sub, where
+             * Sub is a sub-class of Base.  Declare a protected abstract
+             * method foo() in Base, and invoke foo() from a method in Base.
+             * Base is an "abstract base class" and is never instantiated
+             * directly.  Now, Override foo() in Sub, and use Sub.  This
+             * Works fine unless Sub stops providing an implementation of
+             * the method.
+             */
+            dvmThrowException("Ljava/lang/AbstractMethodError;",
+                "abstract method not implemented");
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(!dvmIsAbstractMethod(methodToCall) ||
+            methodToCall->nativeFunc != NULL);
+#endif
+
+        LOGVV("+++ base=%s.%s virtual[%d]=%s.%s\n",
+            baseMethod->clazz->descriptor, baseMethod->name,
+            (u4) baseMethod->methodIndex,
+            methodToCall->clazz->descriptor, methodToCall->name);
+        assert(methodToCall != NULL);
+
+#if 0
+        if (vsrc1 != methodToCall->insSize) {
+            LOGW("WRONG METHOD: base=%s.%s virtual[%d]=%s.%s\n",
+                baseMethod->clazz->descriptor, baseMethod->name,
+                (u4) baseMethod->methodIndex,
+                methodToCall->clazz->descriptor, methodToCall->name);
+            //dvmDumpClass(baseMethod->clazz);
+            //dvmDumpClass(methodToCall->clazz);
+            dvmDumpAllClasses(0);
+        }
+#endif
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeSuper, bool methodCallRange)
+    {
+        Method* baseMethod;
+        u2 thisReg;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* method ref */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        if (methodCallRange) {
+            ILOGV("|invoke-super-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisReg = vdst;
+        } else {
+            ILOGV("|invoke-super args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisReg = vdst & 0x0f;
+        }
+        /* impossible in well-formed code, but we must check nevertheless */
+        if (!checkForNull((Object*) GET_REGISTER(thisReg)))
+            GOTO_exceptionThrown();
+
+        /*
+         * Resolve the method.  This is the correct method for the static
+         * type of the object.  We also verify access permissions here.
+         * The first arg to dvmResolveMethod() is just the referring class
+         * (used for class loaders and such), so we don't want to pass
+         * the superclass into the resolution call.
+         */
+        baseMethod = dvmDexGetResolvedMethod(methodClassDex, ref);
+        if (baseMethod == NULL) {
+            baseMethod = dvmResolveMethod(curMethod->clazz, ref,METHOD_VIRTUAL);
+            if (baseMethod == NULL) {
+                ILOGV("+ unknown method or access denied\n");
+                GOTO_exceptionThrown();
+            }
+        }
+
+        /*
+         * Combine the object we found with the vtable offset in the
+         * method's class.
+         *
+         * We're using the current method's class' superclass, not the
+         * superclass of "this".  This is because we might be executing
+         * in a method inherited from a superclass, and we want to run
+         * in that class' superclass.
+         */
+        if (baseMethod->methodIndex >= curMethod->clazz->super->vtableCount) {
+            /*
+             * Method does not exist in the superclass.  Could happen if
+             * superclass gets updated.
+             */
+            dvmThrowException("Ljava/lang/NoSuchMethodError;",
+                baseMethod->name);
+            GOTO_exceptionThrown();
+        }
+        methodToCall = curMethod->clazz->super->vtable[baseMethod->methodIndex];
+#if 0
+        if (dvmIsAbstractMethod(methodToCall)) {
+            dvmThrowException("Ljava/lang/AbstractMethodError;",
+                "abstract method not implemented");
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(!dvmIsAbstractMethod(methodToCall) ||
+            methodToCall->nativeFunc != NULL);
+#endif
+        LOGVV("+++ base=%s.%s super-virtual=%s.%s\n",
+            baseMethod->clazz->descriptor, baseMethod->name,
+            methodToCall->clazz->descriptor, methodToCall->name);
+        assert(methodToCall != NULL);
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeInterface, bool methodCallRange)
+    {
+        Object* thisPtr;
+        ClassObject* thisClass;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* method ref */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        /*
+         * The object against which we are executing a method is always
+         * in the first argument.
+         */
+        if (methodCallRange) {
+            assert(vsrc1 > 0);
+            ILOGV("|invoke-interface-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisPtr = (Object*) GET_REGISTER(vdst);
+        } else {
+            assert((vsrc1>>4) > 0);
+            ILOGV("|invoke-interface args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisPtr = (Object*) GET_REGISTER(vdst & 0x0f);
+        }
+        if (!checkForNull(thisPtr))
+            GOTO_exceptionThrown();
+
+        thisClass = thisPtr->clazz;
+
+        /*
+         * Given a class and a method index, find the Method* with the
+         * actual code we want to execute.
+         */
+        methodToCall = dvmFindInterfaceMethodInCache(thisClass, ref, curMethod,
+                        methodClassDex);
+        if (methodToCall == NULL) {
+            assert(dvmCheckException(self));
+            GOTO_exceptionThrown();
+        }
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeDirect, bool methodCallRange)
+    {
+        u2 thisReg;
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* method ref */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        EXPORT_PC();
+
+        if (methodCallRange) {
+            ILOGV("|invoke-direct-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisReg = vdst;
+        } else {
+            ILOGV("|invoke-direct args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisReg = vdst & 0x0f;
+        }
+        if (!checkForNull((Object*) GET_REGISTER(thisReg)))
+            GOTO_exceptionThrown();
+
+        methodToCall = dvmDexGetResolvedMethod(methodClassDex, ref);
+        if (methodToCall == NULL) {
+            methodToCall = dvmResolveMethod(curMethod->clazz, ref,
+                            METHOD_DIRECT);
+            if (methodToCall == NULL) {
+                ILOGV("+ unknown direct method\n");     // should be impossible
+                GOTO_exceptionThrown();
+            }
+        }
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeStatic, bool methodCallRange)
+    vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+    ref = FETCH(1);             /* method ref */
+    vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+    EXPORT_PC();
+
+    if (methodCallRange)
+        ILOGV("|invoke-static-range args=%d @0x%04x {regs=v%d-v%d}",
+            vsrc1, ref, vdst, vdst+vsrc1-1);
+    else
+        ILOGV("|invoke-static args=%d @0x%04x {regs=0x%04x %x}",
+            vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+
+    methodToCall = dvmDexGetResolvedMethod(methodClassDex, ref);
+    if (methodToCall == NULL) {
+        methodToCall = dvmResolveMethod(curMethod->clazz, ref, METHOD_STATIC);
+        if (methodToCall == NULL) {
+            ILOGV("+ unknown method\n");
+            GOTO_exceptionThrown();
+        }
+    }
+    GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeVirtualQuick, bool methodCallRange)
+    {
+        Object* thisPtr;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* vtable index */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        /*
+         * The object against which we are executing a method is always
+         * in the first argument.
+         */
+        if (methodCallRange) {
+            assert(vsrc1 > 0);
+            ILOGV("|invoke-virtual-quick-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisPtr = (Object*) GET_REGISTER(vdst);
+        } else {
+            assert((vsrc1>>4) > 0);
+            ILOGV("|invoke-virtual-quick args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisPtr = (Object*) GET_REGISTER(vdst & 0x0f);
+        }
+
+        if (!checkForNull(thisPtr))
+            GOTO_exceptionThrown();
+
+        /*
+         * Combine the object we found with the vtable offset in the
+         * method.
+         */
+        assert(ref < thisPtr->clazz->vtableCount);
+        methodToCall = thisPtr->clazz->vtable[ref];
+
+#if 0
+        if (dvmIsAbstractMethod(methodToCall)) {
+            dvmThrowException("Ljava/lang/AbstractMethodError;",
+                "abstract method not implemented");
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(!dvmIsAbstractMethod(methodToCall) ||
+            methodToCall->nativeFunc != NULL);
+#endif
+
+        LOGVV("+++ virtual[%d]=%s.%s\n",
+            ref, methodToCall->clazz->descriptor, methodToCall->name);
+        assert(methodToCall != NULL);
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeSuperQuick, bool methodCallRange)
+    {
+        u2 thisReg;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* vtable index */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        if (methodCallRange) {
+            ILOGV("|invoke-super-quick-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisReg = vdst;
+        } else {
+            ILOGV("|invoke-super-quick args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisReg = vdst & 0x0f;
+        }
+        /* impossible in well-formed code, but we must check nevertheless */
+        if (!checkForNull((Object*) GET_REGISTER(thisReg)))
+            GOTO_exceptionThrown();
+
+#if 0   /* impossible in optimized + verified code */
+        if (ref >= curMethod->clazz->super->vtableCount) {
+            dvmThrowException("Ljava/lang/NoSuchMethodError;", NULL);
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(ref < curMethod->clazz->super->vtableCount);
+#endif
+
+        /*
+         * Combine the object we found with the vtable offset in the
+         * method's class.
+         *
+         * We're using the current method's class' superclass, not the
+         * superclass of "this".  This is because we might be executing
+         * in a method inherited from a superclass, and we want to run
+         * in the method's class' superclass.
+         */
+        methodToCall = curMethod->clazz->super->vtable[ref];
+
+#if 0
+        if (dvmIsAbstractMethod(methodToCall)) {
+            dvmThrowException("Ljava/lang/AbstractMethodError;",
+                "abstract method not implemented");
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(!dvmIsAbstractMethod(methodToCall) ||
+            methodToCall->nativeFunc != NULL);
+#endif
+        LOGVV("+++ super-virtual[%d]=%s.%s\n",
+            ref, methodToCall->clazz->descriptor, methodToCall->name);
+        assert(methodToCall != NULL);
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+
+
+    /*
+     * General handling for return-void, return, and return-wide.  Put the
+     * return value in "retval" before jumping here.
+     */
+GOTO_TARGET(returnFromMethod)
+    {
+        StackSaveArea* saveArea;
+
+        /*
+         * We must do this BEFORE we pop the previous stack frame off, so
+         * that the GC can see the return value (if any) in the local vars.
+         *
+         * Since this is now an interpreter switch point, we must do it before
+         * we do anything at all.
+         */
+        PERIODIC_CHECKS(kInterpEntryReturn, 0);
+
+        ILOGV("> retval=0x%llx (leaving %s.%s %s)",
+            retval.j, curMethod->clazz->descriptor, curMethod->name,
+            curMethod->signature);
+        //DUMP_REGS(curMethod, fp);
+
+        saveArea = SAVEAREA_FROM_FP(fp);
+
+#ifdef EASY_GDB
+        debugSaveArea = saveArea;
+#endif
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_PROFILER)
+        TRACE_METHOD_EXIT(self, curMethod);
+#endif
+
+        /* back up to previous frame and see if we hit a break */
+        fp = saveArea->prevFrame;
+        assert(fp != NULL);
+        if (dvmIsBreakFrame(fp)) {
+            /* bail without popping the method frame from stack */
+            LOGVV("+++ returned into break frame\n");
+            GOTO_bail();
+        }
+
+        /* update thread FP, and reset local variables */
+        self->curFrame = fp;
+        curMethod = SAVEAREA_FROM_FP(fp)->method;
+        //methodClass = curMethod->clazz;
+        methodClassDex = curMethod->clazz->pDvmDex;
+        pc = saveArea->savedPc;
+        ILOGD("> (return to %s.%s %s)", curMethod->clazz->descriptor,
+            curMethod->name, curMethod->signature);
+
+        /* use FINISH on the caller's invoke instruction */
+        //u2 invokeInstr = INST_INST(FETCH(0));
+        if (true /*invokeInstr >= OP_INVOKE_VIRTUAL &&
+            invokeInstr <= OP_INVOKE_INTERFACE*/)
+        {
+            FINISH(3);
+        } else {
+            //LOGE("Unknown invoke instr %02x at %d\n",
+            //    invokeInstr, (int) (pc - curMethod->insns));
+            assert(false);
+        }
+    }
+GOTO_TARGET_END
+
+
+    /*
+     * Jump here when the code throws an exception.
+     *
+     * By the time we get here, the Throwable has been created and the stack
+     * trace has been saved off.
+     */
+GOTO_TARGET(exceptionThrown)
+    {
+        Object* exception;
+        int catchRelPc;
+
+        /*
+         * Since this is now an interpreter switch point, we must do it before
+         * we do anything at all.
+         */
+        PERIODIC_CHECKS(kInterpEntryThrow, 0);
+
+        /*
+         * We save off the exception and clear the exception status.  While
+         * processing the exception we might need to load some Throwable
+         * classes, and we don't want class loader exceptions to get
+         * confused with this one.
+         */
+        assert(dvmCheckException(self));
+        exception = dvmGetException(self);
+        dvmAddTrackedAlloc(exception, self);
+        dvmClearException(self);
+
+        LOGV("Handling exception %s at %s:%d\n",
+            exception->clazz->descriptor, curMethod->name,
+            dvmLineNumFromPC(curMethod, pc - curMethod->insns));
+
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_DEBUGGER)
+        /*
+         * Tell the debugger about it.
+         *
+         * TODO: if the exception was thrown by interpreted code, control
+         * fell through native, and then back to us, we will report the
+         * exception at the point of the throw and again here.  We can avoid
+         * this by not reporting exceptions when we jump here directly from
+         * the native call code above, but then we won't report exceptions
+         * that were thrown *from* the JNI code (as opposed to *through* it).
+         *
+         * The correct solution is probably to ignore from-native exceptions
+         * here, and have the JNI exception code do the reporting to the
+         * debugger.
+         */
+        if (gDvm.debuggerActive) {
+            void* catchFrame;
+            catchRelPc = dvmFindCatchBlock(self, pc - curMethod->insns,
+                        exception, true, &catchFrame);
+            dvmDbgPostException(fp, pc - curMethod->insns, catchFrame,
+                catchRelPc, exception);
+        }
+#endif
+
+        /*
+         * We need to unroll to the catch block or the nearest "break"
+         * frame.
+         *
+         * A break frame could indicate that we have reached an intermediate
+         * native call, or have gone off the top of the stack and the thread
+         * needs to exit.  Either way, we return from here, leaving the
+         * exception raised.
+         *
+         * If we do find a catch block, we want to transfer execution to
+         * that point.
+         */
+        catchRelPc = dvmFindCatchBlock(self, pc - curMethod->insns,
+                    exception, false, (void*)&fp);
+
+        /*
+         * Restore the stack bounds after an overflow.  This isn't going to
+         * be correct in all circumstances, e.g. if JNI code devours the
+         * exception this won't happen until some other exception gets
+         * thrown.  If the code keeps pushing the stack bounds we'll end
+         * up aborting the VM.
+         *
+         * Note we want to do this *after* the call to dvmFindCatchBlock,
+         * because that may need extra stack space to resolve exception
+         * classes (e.g. through a class loader).
+         */
+        if (self->stackOverflowed)
+            dvmCleanupStackOverflow(self);
+
+        if (catchRelPc < 0) {
+            /* falling through to JNI code or off the bottom of the stack */
+#if DVM_SHOW_EXCEPTION >= 2
+            LOGD("Exception %s from %s:%d not caught locally\n",
+                exception->clazz->descriptor, dvmGetMethodSourceFile(curMethod),
+                dvmLineNumFromPC(curMethod, pc - curMethod->insns));
+#endif
+            dvmSetException(self, exception);
+            dvmReleaseTrackedAlloc(exception, self);
+            GOTO_bail();
+        }
+
+#if DVM_SHOW_EXCEPTION >= 3
+        {
+            const Method* catchMethod = SAVEAREA_FROM_FP(fp)->method;
+            LOGD("Exception %s thrown from %s:%d to %s:%d\n",
+                exception->clazz->descriptor, dvmGetMethodSourceFile(curMethod),
+                dvmLineNumFromPC(curMethod, pc - curMethod->insns),
+                dvmGetMethodSourceFile(catchMethod),
+                dvmLineNumFromPC(catchMethod, catchRelPc));
+        }
+#endif
+
+        /*
+         * Adjust local variables to match self->curFrame and the
+         * updated PC.
+         */
+        //fp = (u4*) self->curFrame;
+        curMethod = SAVEAREA_FROM_FP(fp)->method;
+        //methodClass = curMethod->clazz;
+        methodClassDex = curMethod->clazz->pDvmDex;
+        pc = curMethod->insns + catchRelPc;
+        ILOGV("> pc <-- %s.%s %s", curMethod->clazz->descriptor,
+            curMethod->name, curMethod->signature);
+        DUMP_REGS(curMethod, fp, false);            // show all regs
+
+        /*
+         * Restore the exception if the handler wants it.
+         *
+         * The Dalvik spec mandates that, if an exception handler wants to
+         * do something with the exception, the first instruction executed
+         * must be "move-exception".  We can pass the exception along
+         * through the thread struct, and let the move-exception instruction
+         * clear it for us.
+         *
+         * If the handler doesn't call move-exception, we don't want to
+         * finish here with an exception still pending.
+         */
+        if (INST_INST(FETCH(0)) == OP_MOVE_EXCEPTION)
+            dvmSetException(self, exception);
+
+        dvmReleaseTrackedAlloc(exception, self);
+        FINISH(0);
+    }
+GOTO_TARGET_END
+
+
+    /*
+     * General handling for invoke-{virtual,super,direct,static,interface},
+     * including "quick" variants.
+     *
+     * Set "methodToCall" to the Method we're calling, and "methodCallRange"
+     * depending on whether this is a "/range" instruction.
+     *
+     * For a range call:
+     *  "vsrc1" holds the argument count (8 bits)
+     *  "vdst" holds the first argument in the range
+     * For a non-range call:
+     *  "vsrc1" holds the argument count (4 bits) and the 5th argument index
+     *  "vdst" holds four 4-bit register indices
+     *
+     * The caller must EXPORT_PC before jumping here, because any method
+     * call can throw a stack overflow exception.
+     */
+GOTO_TARGET(invokeMethod, bool methodCallRange, const Method* _methodToCall,
+    u2 count, u2 regs)
+    {
+        STUB_HACK(vsrc1 = count; vdst = regs; methodToCall = _methodToCall;);
+
+        //printf("range=%d call=%p count=%d regs=0x%04x\n",
+        //    methodCallRange, methodToCall, count, regs);
+        //printf(" --> %s.%s %s\n", methodToCall->clazz->descriptor,
+        //    methodToCall->name, methodToCall->signature);
+
+        u4* outs;
+        int i;
+
+        /*
+         * Copy args.  This may corrupt vsrc1/vdst.
+         */
+        if (methodCallRange) {
+            // could use memcpy or a "Duff's device"; most functions have
+            // so few args it won't matter much
+            assert(vsrc1 <= curMethod->outsSize);
+            assert(vsrc1 == methodToCall->insSize);
+            outs = OUTS_FROM_FP(fp, vsrc1);
+            for (i = 0; i < vsrc1; i++)
+                outs[i] = GET_REGISTER(vdst+i);
+        } else {
+            u4 count = vsrc1 >> 4;
+
+            assert(count <= curMethod->outsSize);
+            assert(count == methodToCall->insSize);
+            assert(count <= 5);
+
+            outs = OUTS_FROM_FP(fp, count);
+#if 0
+            if (count == 5) {
+                outs[4] = GET_REGISTER(vsrc1 & 0x0f);
+                count--;
+            }
+            for (i = 0; i < (int) count; i++) {
+                outs[i] = GET_REGISTER(vdst & 0x0f);
+                vdst >>= 4;
+            }
+#else
+            // This version executes fewer instructions but is larger
+            // overall.  Seems to be a teensy bit faster.
+            assert((vdst >> 16) == 0);  // 16 bits -or- high 16 bits clear
+            switch (count) {
+            case 5:
+                outs[4] = GET_REGISTER(vsrc1 & 0x0f);
+            case 4:
+                outs[3] = GET_REGISTER(vdst >> 12);
+            case 3:
+                outs[2] = GET_REGISTER((vdst & 0x0f00) >> 8);
+            case 2:
+                outs[1] = GET_REGISTER((vdst & 0x00f0) >> 4);
+            case 1:
+                outs[0] = GET_REGISTER(vdst & 0x0f);
+            default:
+                ;
+            }
+#endif
+        }
+    }
+
+    /*
+     * (This was originally a "goto" target; I've kept it separate from the
+     * stuff above in case we want to refactor things again.)
+     *
+     * At this point, we have the arguments stored in the "outs" area of
+     * the current method's stack frame, and the method to call in
+     * "methodToCall".  Push a new stack frame.
+     */
+    {
+        StackSaveArea* newSaveArea;
+        u4* newFp;
+
+        ILOGV("> %s%s.%s %s",
+            dvmIsNativeMethod(methodToCall) ? "(NATIVE) " : "",
+            methodToCall->clazz->descriptor, methodToCall->name,
+            methodToCall->signature);
+
+        newFp = (u4*) SAVEAREA_FROM_FP(fp) - methodToCall->registersSize;
+        newSaveArea = SAVEAREA_FROM_FP(newFp);
+
+        /* verify that we have enough space */
+        if (true) {
+            u1* bottom;
+            bottom = (u1*) newSaveArea - methodToCall->outsSize * sizeof(u4);
+            if (bottom < self->interpStackEnd) {
+                /* stack overflow */
+                LOGV("Stack overflow on method call (start=%p end=%p newBot=%p size=%d '%s')\n",
+                    self->interpStackStart, self->interpStackEnd, bottom,
+                    self->interpStackSize, methodToCall->name);
+                dvmHandleStackOverflow(self);
+                assert(dvmCheckException(self));
+                GOTO_exceptionThrown();
+            }
+            //LOGD("+++ fp=%p newFp=%p newSave=%p bottom=%p\n",
+            //    fp, newFp, newSaveArea, bottom);
+        }
+
+#ifdef LOG_INSTR
+        if (methodToCall->registersSize > methodToCall->insSize) {
+            /*
+             * This makes valgrind quiet when we print registers that
+             * haven't been initialized.  Turn it off when the debug
+             * messages are disabled -- we want valgrind to report any
+             * used-before-initialized issues.
+             */
+            memset(newFp, 0xcc,
+                (methodToCall->registersSize - methodToCall->insSize) * 4);
+        }
+#endif
+
+#ifdef EASY_GDB
+        newSaveArea->prevSave = SAVEAREA_FROM_FP(fp);
+#endif
+        newSaveArea->prevFrame = fp;
+        newSaveArea->savedPc = pc;
+        newSaveArea->method = methodToCall;
+
+        if (!dvmIsNativeMethod(methodToCall)) {
+            /*
+             * "Call" interpreted code.  Reposition the PC, update the
+             * frame pointer and other local state, and continue.
+             */
+            curMethod = methodToCall;
+            methodClassDex = curMethod->clazz->pDvmDex;
+            pc = methodToCall->insns;
+            fp = self->curFrame = newFp;
+#ifdef EASY_GDB
+            debugSaveArea = SAVEAREA_FROM_FP(newFp);
+#endif
+#if INTERP_TYPE == INTERP_DBG
+            debugIsMethodEntry = true;              // profiling, debugging
+#endif
+            ILOGD("> pc <-- %s.%s %s", curMethod->clazz->descriptor,
+                curMethod->name, curMethod->signature);
+            DUMP_REGS(curMethod, fp, true);         // show input args
+            FINISH(0);                              // jump to method start
+        } else {
+            /* set this up for JNI locals, even if not a JNI native */
+            newSaveArea->xtra.localRefTop = self->jniLocalRefTable.nextEntry;
+
+            self->curFrame = newFp;
+
+            DUMP_REGS(methodToCall, newFp, true);   // show input args
+
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_DEBUGGER)
+            if (gDvm.debuggerActive) {
+                dvmDbgPostLocationEvent(methodToCall, -1,
+                    dvmGetThisPtr(curMethod, fp), DBG_METHOD_ENTRY);
+            }
+#endif
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_PROFILER)
+            TRACE_METHOD_ENTER(self, methodToCall);
+#endif
+
+            ILOGD("> native <-- %s.%s %s", methodToCall->clazz->descriptor,
+                methodToCall->name, methodToCall->signature);
+
+            /*
+             * Jump through native call bridge.  Because we leave no
+             * space for locals on native calls, "newFp" points directly
+             * to the method arguments.
+             */
+            (*methodToCall->nativeFunc)(newFp, &retval, methodToCall, self);
+
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_DEBUGGER)
+            if (gDvm.debuggerActive) {
+                dvmDbgPostLocationEvent(methodToCall, -1,
+                    dvmGetThisPtr(curMethod, fp), DBG_METHOD_EXIT);
+            }
+#endif
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_PROFILER)
+            TRACE_METHOD_EXIT(self, methodToCall);
+#endif
+
+            /* pop frame off */
+            dvmPopJniLocals(self, newSaveArea);
+            self->curFrame = fp;
+
+            /*
+             * If the native code threw an exception, or interpreted code
+             * invoked by the native call threw one and nobody has cleared
+             * it, jump to our local exception handling.
+             */
+            if (dvmCheckException(self)) {
+                LOGV("Exception thrown by/below native code\n");
+                GOTO_exceptionThrown();
+            }
+
+            ILOGD("> retval=0x%llx (leaving native)", retval.j);
+            ILOGD("> (return from native %s.%s to %s.%s %s)",
+                methodToCall->clazz->descriptor, methodToCall->name,
+                curMethod->clazz->descriptor, curMethod->name,
+                curMethod->signature);
+
+            //u2 invokeInstr = INST_INST(FETCH(0));
+            if (true /*invokeInstr >= OP_INVOKE_VIRTUAL &&
+                invokeInstr <= OP_INVOKE_INTERFACE*/)
+            {
+                FINISH(3);
+            } else {
+                //LOGE("Unknown invoke instr %02x at %d\n",
+                //    invokeInstr, (int) (pc - curMethod->insns));
+                assert(false);
+            }
+        }
+    }
+    assert(false);      // should not get here
+GOTO_TARGET_END
+
+
+/* File: cstubs/enddefs.c */
+
+/* undefine "magic" name remapping */
+#undef retval
+#undef pc
+#undef fp
+#undef curMethod
+#undef methodClassDex
+#undef self
+#undef debugTrackedRefStart
+
diff --git a/vm/mterp/out/InterpC-armv4.c b/vm/mterp/out/InterpC-armv4.c
new file mode 100644
index 0000000..2fcdcab
--- /dev/null
+++ b/vm/mterp/out/InterpC-armv4.c
@@ -0,0 +1,1266 @@
+/*
+ * This file was generated automatically by gen-mterp.py for 'armv4'.
+ *
+ * --> DO NOT EDIT <--
+ */
+
+/* File: c/header.c */
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/* common includes */
+#include "Dalvik.h"
+#include "interp/InterpDefs.h"
+#include "mterp/Mterp.h"
+#include <math.h>                   // needed for fmod, fmodf
+
+/*
+ * Configuration defines.  These affect the C implementations, i.e. the
+ * portable interpreter(s) and C stubs.
+ *
+ * Some defines are controlled by the Makefile, e.g.:
+ *   WITH_PROFILER
+ *   WITH_DEBUGGER
+ *   WITH_INSTR_CHECKS
+ *   WITH_TRACKREF_CHECKS
+ *   EASY_GDB
+ *   NDEBUG
+ *
+ * If THREADED_INTERP is not defined, we use a classic "while true / switch"
+ * interpreter.  If it is defined, then the tail end of each instruction
+ * handler fetches the next instruction and jumps directly to the handler.
+ * This increases the size of the "Std" interpreter by about 10%, but
+ * provides a speedup of about the same magnitude.
+ *
+ * There's a "hybrid" approach that uses a goto table instead of a switch
+ * statement, avoiding the "is the opcode in range" tests required for switch.
+ * The performance is close to the threaded version, and without the 10%
+ * size increase, but the benchmark results are off enough that it's not
+ * worth adding as a third option.
+ */
+#define THREADED_INTERP             /* threaded vs. while-loop interpreter */
+
+#ifdef WITH_INSTR_CHECKS            /* instruction-level paranoia */
+# define CHECK_BRANCH_OFFSETS
+# define CHECK_REGISTER_INDICES
+#endif
+
+/*
+ * ARM EABI requires 64-bit alignment for access to 64-bit data types.  We
+ * can't just use pointers to copy 64-bit values out of our interpreted
+ * register set, because gcc will generate ldrd/strd.
+ *
+ * The __UNION version copies data in and out of a union.  The __MEMCPY
+ * version uses a memcpy() call to do the transfer; gcc is smart enough to
+ * not actually call memcpy().  The __UNION version is very bad on ARM;
+ * it only uses one more instruction than __MEMCPY, but for some reason
+ * gcc thinks it needs separate storage for every instance of the union.
+ * On top of that, it feels the need to zero them out at the start of the
+ * method.  Net result is we zero out ~700 bytes of stack space at the top
+ * of the interpreter using ARM STM instructions.
+ */
+#if defined(__ARM_EABI__)
+//# define NO_UNALIGN_64__UNION
+# define NO_UNALIGN_64__MEMCPY
+#endif
+
+//#define LOG_INSTR                   /* verbose debugging */
+/* set and adjust ANDROID_LOG_TAGS='*:i jdwp:i dalvikvm:i dalvikvmi:i' */
+
+/*
+ * Keep a tally of accesses to fields.  Currently only works if full DEX
+ * optimization is disabled.
+ */
+#ifdef PROFILE_FIELD_ACCESS
+# define UPDATE_FIELD_GET(_field) { (_field)->gets++; }
+# define UPDATE_FIELD_PUT(_field) { (_field)->puts++; }
+#else
+# define UPDATE_FIELD_GET(_field) ((void)0)
+# define UPDATE_FIELD_PUT(_field) ((void)0)
+#endif
+
+/*
+ * Adjust the program counter.  "_offset" is a signed int, in 16-bit units.
+ *
+ * Assumes the existence of "const u2* pc" and "const u2* curMethod->insns".
+ *
+ * We don't advance the program counter until we finish an instruction or
+ * branch, because we do want to have to unroll the PC if there's an
+ * exception.
+ */
+#ifdef CHECK_BRANCH_OFFSETS
+# define ADJUST_PC(_offset) do {                                            \
+        int myoff = _offset;        /* deref only once */                   \
+        if (pc + myoff < curMethod->insns ||                                \
+            pc + myoff >= curMethod->insns + dvmGetMethodInsnsSize(curMethod)) \
+        {                                                                   \
+            char* desc;                                                     \
+            desc = dexProtoCopyMethodDescriptor(&curMethod->prototype);     \
+            LOGE("Invalid branch %d at 0x%04x in %s.%s %s\n",               \
+                myoff, (int) (pc - curMethod->insns),                       \
+                curMethod->clazz->descriptor, curMethod->name, desc);       \
+            free(desc);                                                     \
+            dvmAbort();                                                     \
+        }                                                                   \
+        pc += myoff;                                                        \
+    } while (false)
+#else
+# define ADJUST_PC(_offset) (pc += _offset)
+#endif
+
+/*
+ * If enabled, log instructions as we execute them.
+ */
+#ifdef LOG_INSTR
+# define ILOGD(...) ILOG(LOG_DEBUG, __VA_ARGS__)
+# define ILOGV(...) ILOG(LOG_VERBOSE, __VA_ARGS__)
+# define ILOG(_level, ...) do {                                             \
+        char debugStrBuf[128];                                              \
+        snprintf(debugStrBuf, sizeof(debugStrBuf), __VA_ARGS__);            \
+        if (curMethod != NULL)                                                 \
+            LOG(_level, LOG_TAG"i", "%-2d|%04x%s\n",                        \
+                self->threadId, (int)(pc - curMethod->insns), debugStrBuf); \
+        else                                                                \
+            LOG(_level, LOG_TAG"i", "%-2d|####%s\n",                        \
+                self->threadId, debugStrBuf);                               \
+    } while(false)
+void dvmDumpRegs(const Method* method, const u4* framePtr, bool inOnly);
+# define DUMP_REGS(_meth, _frame, _inOnly) dvmDumpRegs(_meth, _frame, _inOnly)
+static const char kSpacing[] = "            ";
+#else
+# define ILOGD(...) ((void)0)
+# define ILOGV(...) ((void)0)
+# define DUMP_REGS(_meth, _frame, _inOnly) ((void)0)
+#endif
+
+/* get a long from an array of u4 */
+static inline s8 getLongFromArray(const u4* ptr, int idx)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { s8 ll; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.parts[0] = ptr[0];
+    conv.parts[1] = ptr[1];
+    return conv.ll;
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    s8 val;
+    memcpy(&val, &ptr[idx], 8);
+    return val;
+#else
+    return *((s8*) &ptr[idx]);
+#endif
+}
+
+/* store a long into an array of u4 */
+static inline void putLongToArray(u4* ptr, int idx, s8 val)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { s8 ll; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.ll = val;
+    ptr[0] = conv.parts[0];
+    ptr[1] = conv.parts[1];
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    memcpy(&ptr[idx], &val, 8);
+#else
+    *((s8*) &ptr[idx]) = val;
+#endif
+}
+
+/* get a double from an array of u4 */
+static inline double getDoubleFromArray(const u4* ptr, int idx)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { double d; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.parts[0] = ptr[0];
+    conv.parts[1] = ptr[1];
+    return conv.d;
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    double dval;
+    memcpy(&dval, &ptr[idx], 8);
+    return dval;
+#else
+    return *((double*) &ptr[idx]);
+#endif
+}
+
+/* store a double into an array of u4 */
+static inline void putDoubleToArray(u4* ptr, int idx, double dval)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { double d; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.d = dval;
+    ptr[0] = conv.parts[0];
+    ptr[1] = conv.parts[1];
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    memcpy(&ptr[idx], &dval, 8);
+#else
+    *((double*) &ptr[idx]) = dval;
+#endif
+}
+
+/*
+ * If enabled, validate the register number on every access.  Otherwise,
+ * just do an array access.
+ *
+ * Assumes the existence of "u4* fp".
+ *
+ * "_idx" may be referenced more than once.
+ */
+#ifdef CHECK_REGISTER_INDICES
+# define GET_REGISTER(_idx) \
+    ( (_idx) < curMethod->registersSize ? \
+        (fp[(_idx)]) : (assert(!"bad reg"),1969) )
+# define SET_REGISTER(_idx, _val) \
+    ( (_idx) < curMethod->registersSize ? \
+        (fp[(_idx)] = (u4)(_val)) : (assert(!"bad reg"),1969) )
+# define GET_REGISTER_AS_OBJECT(_idx)       ((Object *)GET_REGISTER(_idx))
+# define SET_REGISTER_AS_OBJECT(_idx, _val) SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_INT(_idx) ((s4) GET_REGISTER(_idx))
+# define SET_REGISTER_INT(_idx, _val) SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_WIDE(_idx) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        getLongFromArray(fp, (_idx)) : (assert(!"bad reg"),1969) )
+# define SET_REGISTER_WIDE(_idx, _val) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        putLongToArray(fp, (_idx), (_val)) : (assert(!"bad reg"),1969) )
+# define GET_REGISTER_FLOAT(_idx) \
+    ( (_idx) < curMethod->registersSize ? \
+        (*((float*) &fp[(_idx)])) : (assert(!"bad reg"),1969.0f) )
+# define SET_REGISTER_FLOAT(_idx, _val) \
+    ( (_idx) < curMethod->registersSize ? \
+        (*((float*) &fp[(_idx)]) = (_val)) : (assert(!"bad reg"),1969.0f) )
+# define GET_REGISTER_DOUBLE(_idx) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        getDoubleFromArray(fp, (_idx)) : (assert(!"bad reg"),1969.0) )
+# define SET_REGISTER_DOUBLE(_idx, _val) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        putDoubleToArray(fp, (_idx), (_val)) : (assert(!"bad reg"),1969.0) )
+#else
+# define GET_REGISTER(_idx)                 (fp[(_idx)])
+# define SET_REGISTER(_idx, _val)           (fp[(_idx)] = (_val))
+# define GET_REGISTER_AS_OBJECT(_idx)       ((Object*) fp[(_idx)])
+# define SET_REGISTER_AS_OBJECT(_idx, _val) (fp[(_idx)] = (u4)(_val))
+# define GET_REGISTER_INT(_idx)             ((s4)GET_REGISTER(_idx))
+# define SET_REGISTER_INT(_idx, _val)       SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_WIDE(_idx)            getLongFromArray(fp, (_idx))
+# define SET_REGISTER_WIDE(_idx, _val)      putLongToArray(fp, (_idx), (_val))
+# define GET_REGISTER_FLOAT(_idx)           (*((float*) &fp[(_idx)]))
+# define SET_REGISTER_FLOAT(_idx, _val)     (*((float*) &fp[(_idx)]) = (_val))
+# define GET_REGISTER_DOUBLE(_idx)          getDoubleFromArray(fp, (_idx))
+# define SET_REGISTER_DOUBLE(_idx, _val)    putDoubleToArray(fp, (_idx), (_val))
+#endif
+
+/*
+ * Get 16 bits from the specified offset of the program counter.  We always
+ * want to load 16 bits at a time from the instruction stream -- it's more
+ * efficient than 8 and won't have the alignment problems that 32 might.
+ *
+ * Assumes existence of "const u2* pc".
+ */
+#define FETCH(_offset)     (pc[(_offset)])
+
+/*
+ * Extract instruction byte from 16-bit fetch (_inst is a u2).
+ */
+#define INST_INST(_inst)    ((_inst) & 0xff)
+
+/*
+ * Extract the "vA, vB" 4-bit registers from the instruction word (_inst is u2).
+ */
+#define INST_A(_inst)       (((_inst) >> 8) & 0x0f)
+#define INST_B(_inst)       ((_inst) >> 12)
+
+/*
+ * Get the 8-bit "vAA" 8-bit register index from the instruction word.
+ * (_inst is u2)
+ */
+#define INST_AA(_inst)      ((_inst) >> 8)
+
+/*
+ * The current PC must be available to Throwable constructors, e.g.
+ * those created by dvmThrowException(), so that the exception stack
+ * trace can be generated correctly.  If we don't do this, the offset
+ * within the current method won't be shown correctly.  See the notes
+ * in Exception.c.
+ *
+ * Assumes existence of "u4* fp" and "const u2* pc".
+ */
+#define EXPORT_PC()         (SAVEAREA_FROM_FP(fp)->xtra.currentPc = pc)
+
+/*
+ * Determine if we need to switch to a different interpreter.  "_current"
+ * is either INTERP_STD or INTERP_DBG.  It should be fixed for a given
+ * interpreter generation file, which should remove the outer conditional
+ * from the following.
+ *
+ * If we're building without debug and profiling support, we never switch.
+ */
+#if defined(WITH_PROFILER) || defined(WITH_DEBUGGER)
+# define NEED_INTERP_SWITCH(_current) (                                     \
+    (_current == INTERP_STD) ?                                              \
+        dvmDebuggerOrProfilerActive() : !dvmDebuggerOrProfilerActive() )
+#else
+# define NEED_INTERP_SWITCH(_current) (false)
+#endif
+
+/*
+ * Look up an interface on a class using the cache.
+ */
+INLINE Method* dvmFindInterfaceMethodInCache(ClassObject* thisClass,
+    u4 methodIdx, const Method* method, DvmDex* methodClassDex)
+{
+#define ATOMIC_CACHE_CALC \
+    dvmInterpFindInterfaceMethod(thisClass, methodIdx, method, methodClassDex)
+
+    return (Method*) ATOMIC_CACHE_LOOKUP(methodClassDex->pInterfaceCache,
+                DEX_INTERFACE_CACHE_SIZE, thisClass, methodIdx);
+
+#undef ATOMIC_CACHE_CALC
+}
+
+/*
+ * Check to see if "obj" is NULL.  If so, throw an exception.  Assumes the
+ * pc has already been exported to the stack.
+ *
+ * Perform additional checks on debug builds.
+ *
+ * Use this to check for NULL when the instruction handler calls into
+ * something that could throw an exception (so we have already called
+ * EXPORT_PC at the top).
+ */
+static inline bool checkForNull(Object* obj)
+{
+    if (obj == NULL) {
+        dvmThrowException("Ljava/lang/NullPointerException;", NULL);
+        return false;
+    }
+#ifdef WITH_EXTRA_OBJECT_VALIDATION
+    if (!dvmIsValidObject(obj)) {
+        LOGE("Invalid object %p\n", obj);
+        dvmAbort();
+    }
+#endif
+#ifndef NDEBUG
+    if (obj->clazz == NULL || ((u4) obj->clazz) <= 65536) {
+        /* probable heap corruption */
+        LOGE("Invalid object class %p (in %p)\n", obj->clazz, obj);
+        dvmAbort();
+    }
+#endif
+    return true;
+}
+
+/*
+ * Check to see if "obj" is NULL.  If so, export the PC into the stack
+ * frame and throw an exception.
+ *
+ * Perform additional checks on debug builds.
+ *
+ * Use this to check for NULL when the instruction handler doesn't do
+ * anything else that can throw an exception.
+ */
+static inline bool checkForNullExportPC(Object* obj, u4* fp, const u2* pc)
+{
+    if (obj == NULL) {
+        EXPORT_PC();
+        dvmThrowException("Ljava/lang/NullPointerException;", NULL);
+        return false;
+    }
+#ifdef WITH_EXTRA_OBJECT_VALIDATION
+    if (!dvmIsValidObject(obj)) {
+        LOGE("Invalid object %p\n", obj);
+        dvmAbort();
+    }
+#endif
+#ifndef NDEBUG
+    if (obj->clazz == NULL || ((u4) obj->clazz) <= 65536) {
+        /* probable heap corruption */
+        LOGE("Invalid object class %p (in %p)\n", obj->clazz, obj);
+        dvmAbort();
+    }
+#endif
+    return true;
+}
+
+
+/* File: cstubs/stubdefs.c */
+/* this is a standard (no debug support) interpreter */
+#define INTERP_TYPE INTERP_STD
+#define CHECK_DEBUG_AND_PROF() ((void)0)
+# define CHECK_TRACKED_REFS() ((void)0)
+
+/*
+ * In the C mterp stubs, "goto" is a function call followed immediately
+ * by a return.
+ */
+
+#define GOTO_TARGET_DECL(_target, ...)                                      \
+    void dvmMterp_##_target(MterpGlue* glue, ## __VA_ARGS__);
+
+#define GOTO_TARGET(_target, ...)                                           \
+    void dvmMterp_##_target(MterpGlue* glue, ## __VA_ARGS__) {              \
+        u2 ref, vsrc1, vsrc2, vdst;                                         \
+        u2 inst = FETCH(0);                                                 \
+        const Method* methodToCall;                                         \
+        StackSaveArea* debugSaveArea;
+
+#define GOTO_TARGET_END }
+
+/*
+ * Redefine what used to be local variable accesses into MterpGlue struct
+ * references.  (These are undefined down in "footer.c".)
+ */
+#define retval                  glue->retval
+#define pc                      glue->pc
+#define fp                      glue->fp
+#define curMethod               glue->method
+#define methodClassDex          glue->methodClassDex
+#define self                    glue->self
+#define debugTrackedRefStart    glue->debugTrackedRefStart
+
+/* ugh */
+#define STUB_HACK(x) x
+
+
+/*
+ * Opcode handler framing macros.  Here, each opcode is a separate function
+ * that takes a "glue" argument and returns void.  We can't declare
+ * these "static" because they may be called from an assembly stub.
+ */
+#define HANDLE_OPCODE(_op)                                                  \
+    void dvmMterp_##_op(MterpGlue* glue) {                                  \
+        u2 ref, vsrc1, vsrc2, vdst;                                         \
+        u2 inst = FETCH(0);
+
+#define OP_END }
+
+/*
+ * Like the "portable" FINISH, but don't reload "inst", and return to caller
+ * when done.
+ */
+#define FINISH(_offset) {                                                   \
+        ADJUST_PC(_offset);                                                 \
+        CHECK_DEBUG_AND_PROF();                                             \
+        CHECK_TRACKED_REFS();                                               \
+        return;                                                             \
+    }
+
+
+/*
+ * The "goto label" statements turn into function calls followed by
+ * return statements.  Some of the functions take arguments, which in the
+ * portable interpreter are handled by assigning values to globals.
+ */
+
+#define GOTO_exceptionThrown()                                              \
+    do {                                                                    \
+        dvmMterp_exceptionThrown(glue);                                     \
+        return;                                                             \
+    } while(false)
+
+#define GOTO_returnFromMethod()                                             \
+    do {                                                                    \
+        dvmMterp_returnFromMethod(glue);                                    \
+        return;                                                             \
+    } while(false)
+
+#define GOTO_invoke(_target, _methodCallRange)                              \
+    do {                                                                    \
+        dvmMterp_##_target(glue, _methodCallRange);                         \
+        return;                                                             \
+    } while(false)
+
+#define GOTO_invokeMethod(_methodCallRange, _methodToCall, _vsrc1, _vdst)   \
+    do {                                                                    \
+        dvmMterp_invokeMethod(glue, _methodCallRange, _methodToCall,        \
+            _vsrc1, _vdst);                                                 \
+        return;                                                             \
+    } while(false)
+
+/*
+ * As a special case, "goto bail" turns into a longjmp.  Use "bail_switch"
+ * if we need to switch to the other interpreter upon our return.
+ */
+#define GOTO_bail()                                                         \
+    dvmMterpStdBail(glue, false);
+#define GOTO_bail_switch()                                                  \
+    dvmMterpStdBail(glue, true);
+
+/*
+ * Periodically check for thread suspension.
+ *
+ * While we're at it, see if a debugger has attached or the profiler has
+ * started.  If so, switch to a different "goto" table.
+ */
+#define PERIODIC_CHECKS(_entryPoint, _pcadj) {                              \
+        dvmCheckSuspendQuick(self);                                         \
+        if (NEED_INTERP_SWITCH(INTERP_TYPE)) {                              \
+            ADJUST_PC(_pcadj);                                              \
+            glue->entryPoint = _entryPoint;                                 \
+            LOGVV("threadid=%d: switch to STD ep=%d adj=%d\n",              \
+                glue->self->threadId, (_entryPoint), (_pcadj));             \
+            GOTO_bail_switch();                                             \
+        }                                                                   \
+    }
+
+
+/* File: c/opcommon.c */
+/* forward declarations of goto targets */
+GOTO_TARGET_DECL(filledNewArray, bool methodCallRange);
+GOTO_TARGET_DECL(invokeVirtual, bool methodCallRange);
+GOTO_TARGET_DECL(invokeSuper, bool methodCallRange);
+GOTO_TARGET_DECL(invokeInterface, bool methodCallRange);
+GOTO_TARGET_DECL(invokeDirect, bool methodCallRange);
+GOTO_TARGET_DECL(invokeStatic, bool methodCallRange);
+GOTO_TARGET_DECL(invokeVirtualQuick, bool methodCallRange);
+GOTO_TARGET_DECL(invokeSuperQuick, bool methodCallRange);
+GOTO_TARGET_DECL(invokeMethod, bool methodCallRange, const Method* methodToCall,
+    u2 count, u2 regs);
+GOTO_TARGET_DECL(returnFromMethod);
+GOTO_TARGET_DECL(exceptionThrown);
+
+/*
+ * ===========================================================================
+ *
+ * What follows are opcode definitions shared between multiple opcodes with
+ * minor substitutions handled by the C pre-processor.  These should probably
+ * use the mterp substitution mechanism instead, with the code here moved
+ * into common fragment files (like the asm "binop.S"), although it's hard
+ * to give up the C preprocessor in favor of the much simpler text subst.
+ *
+ * ===========================================================================
+ */
+
+#define HANDLE_NUMCONV(_opcode, _opname, _fromtype, _totype)                \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        SET_REGISTER##_totype(vdst,                                         \
+            GET_REGISTER##_fromtype(vsrc1));                                \
+        FINISH(1);
+
+#define HANDLE_FLOAT_TO_INT(_opcode, _opname, _fromvtype, _fromrtype,       \
+        _tovtype, _tortype)                                                 \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+    {                                                                       \
+        /* spec defines specific handling for +/- inf and NaN values */     \
+        _fromvtype val;                                                     \
+        _tovtype intMin, intMax, result;                                    \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        val = GET_REGISTER##_fromrtype(vsrc1);                              \
+        intMin = (_tovtype) 1 << (sizeof(_tovtype) * 8 -1);                 \
+        intMax = ~intMin;                                                   \
+        result = (_tovtype) val;                                            \
+        if (val >= intMax)          /* +inf */                              \
+            result = intMax;                                                \
+        else if (val <= intMin)     /* -inf */                              \
+            result = intMin;                                                \
+        else if (val != val)        /* NaN */                               \
+            result = 0;                                                     \
+        else                                                                \
+            result = (_tovtype) val;                                        \
+        SET_REGISTER##_tortype(vdst, result);                               \
+    }                                                                       \
+    FINISH(1);
+
+#define HANDLE_INT_TO_SMALL(_opcode, _opname, _type)                        \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|int-to-%s v%d,v%d", (_opname), vdst, vsrc1);                \
+        SET_REGISTER(vdst, (_type) GET_REGISTER(vsrc1));                    \
+        FINISH(1);
+
+/* NOTE: the comparison result is always a signed 4-byte integer */
+#define HANDLE_OP_CMPX(_opcode, _opname, _varType, _type, _nanVal)          \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        int result;                                                         \
+        u2 regs;                                                            \
+        _varType val1, val2;                                                \
+        vdst = INST_AA(inst);                                               \
+        regs = FETCH(1);                                                    \
+        vsrc1 = regs & 0xff;                                                \
+        vsrc2 = regs >> 8;                                                  \
+        ILOGV("|cmp%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);         \
+        val1 = GET_REGISTER##_type(vsrc1);                                  \
+        val2 = GET_REGISTER##_type(vsrc2);                                  \
+        if (val1 == val2)                                                   \
+            result = 0;                                                     \
+        else if (val1 < val2)                                               \
+            result = -1;                                                    \
+        else if (val1 > val2)                                               \
+            result = 1;                                                     \
+        else                                                                \
+            result = (_nanVal);                                             \
+        ILOGV("+ result=%d\n", result);                                     \
+        SET_REGISTER(vdst, result);                                         \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_IF_XX(_opcode, _opname, _cmp)                             \
+    HANDLE_OPCODE(_opcode /*vA, vB, +CCCC*/)                                \
+        vsrc1 = INST_A(inst);                                               \
+        vsrc2 = INST_B(inst);                                               \
+        if ((s4) GET_REGISTER(vsrc1) _cmp (s4) GET_REGISTER(vsrc2)) {       \
+            int branchOffset = (s2)FETCH(1);    /* sign-extended */         \
+            ILOGV("|if-%s v%d,v%d,+0x%04x", (_opname), vsrc1, vsrc2,        \
+                branchOffset);                                              \
+            ILOGV("> branch taken");                                        \
+            if (branchOffset < 0)                                           \
+                PERIODIC_CHECKS(kInterpEntryInstr, branchOffset);           \
+            FINISH(branchOffset);                                           \
+        } else {                                                            \
+            ILOGV("|if-%s v%d,v%d,-", (_opname), vsrc1, vsrc2);             \
+            FINISH(2);                                                      \
+        }
+
+#define HANDLE_OP_IF_XXZ(_opcode, _opname, _cmp)                            \
+    HANDLE_OPCODE(_opcode /*vAA, +BBBB*/)                                   \
+        vsrc1 = INST_AA(inst);                                              \
+        if ((s4) GET_REGISTER(vsrc1) _cmp 0) {                              \
+            int branchOffset = (s2)FETCH(1);    /* sign-extended */         \
+            ILOGV("|if-%s v%d,+0x%04x", (_opname), vsrc1, branchOffset);    \
+            ILOGV("> branch taken");                                        \
+            if (branchOffset < 0)                                           \
+                PERIODIC_CHECKS(kInterpEntryInstr, branchOffset);           \
+            FINISH(branchOffset);                                           \
+        } else {                                                            \
+            ILOGV("|if-%s v%d,-", (_opname), vsrc1);                        \
+            FINISH(2);                                                      \
+        }
+
+#define HANDLE_UNOP(_opcode, _opname, _pfx, _sfx, _type)                    \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        SET_REGISTER##_type(vdst, _pfx GET_REGISTER##_type(vsrc1) _sfx);    \
+        FINISH(1);
+
+#define HANDLE_OP_X_INT(_opcode, _opname, _op, _chkdiv)                     \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-int v%d,v%d", (_opname), vdst, vsrc1);                   \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            secondVal = GET_REGISTER(vsrc2);                                \
+            if (secondVal == 0) {                                           \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && secondVal == -1) {            \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            /* non-div/rem case */                                          \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vsrc1) _op (s4) GET_REGISTER(vsrc2));     \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_INT(_opcode, _opname, _cast, _op)                     \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-int v%d,v%d", (_opname), vdst, vsrc1);                   \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vsrc1) _op (GET_REGISTER(vsrc2) & 0x1f));    \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_INT_LIT16(_opcode, _opname, _op, _chkdiv)               \
+    HANDLE_OPCODE(_opcode /*vA, vB, #+CCCC*/)                               \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        vsrc2 = FETCH(1);                                                   \
+        ILOGV("|%s-int/lit16 v%d,v%d,#+0x%04x",                             \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, result;                                            \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            if ((s2) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                      \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && ((s2) vsrc2) == -1) {         \
+                /* won't generate /lit16 instr for this; check anyway */    \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op (s2) vsrc2;                           \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            /* non-div/rem case */                                          \
+            SET_REGISTER(vdst, GET_REGISTER(vsrc1) _op (s2) vsrc2);         \
+        }                                                                   \
+        FINISH(2);
+
+#define HANDLE_OP_X_INT_LIT8(_opcode, _opname, _op, _chkdiv)                \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, #+CC*/)                               \
+    {                                                                       \
+        u2 litInfo;                                                         \
+        vdst = INST_AA(inst);                                               \
+        litInfo = FETCH(1);                                                 \
+        vsrc1 = litInfo & 0xff;                                             \
+        vsrc2 = litInfo >> 8;       /* constant */                          \
+        ILOGV("|%s-int/lit8 v%d,v%d,#+0x%02x",                              \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, result;                                            \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            if ((s1) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && ((s1) vsrc2) == -1) {         \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op ((s1) vsrc2);                         \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vsrc1) _op (s1) vsrc2);                   \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_INT_LIT8(_opcode, _opname, _cast, _op)                \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, #+CC*/)                               \
+    {                                                                       \
+        u2 litInfo;                                                         \
+        vdst = INST_AA(inst);                                               \
+        litInfo = FETCH(1);                                                 \
+        vsrc1 = litInfo & 0xff;                                             \
+        vsrc2 = litInfo >> 8;       /* constant */                          \
+        ILOGV("|%s-int/lit8 v%d,v%d,#+0x%02x",                              \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vsrc1) _op (vsrc2 & 0x1f));                  \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_INT_2ADDR(_opcode, _opname, _op, _chkdiv)               \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-int-2addr v%d,v%d", (_opname), vdst, vsrc1);             \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER(vdst);                                  \
+            secondVal = GET_REGISTER(vsrc1);                                \
+            if (secondVal == 0) {                                           \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && secondVal == -1) {            \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vdst) _op (s4) GET_REGISTER(vsrc1));      \
+        }                                                                   \
+        FINISH(1);
+
+#define HANDLE_OP_SHX_INT_2ADDR(_opcode, _opname, _cast, _op)               \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-int-2addr v%d,v%d", (_opname), vdst, vsrc1);             \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vdst) _op (GET_REGISTER(vsrc1) & 0x1f));     \
+        FINISH(1);
+
+#define HANDLE_OP_X_LONG(_opcode, _opname, _op, _chkdiv)                    \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-long v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);       \
+        if (_chkdiv != 0) {                                                 \
+            s8 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER_WIDE(vsrc1);                            \
+            secondVal = GET_REGISTER_WIDE(vsrc2);                           \
+            if (secondVal == 0LL) {                                         \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u8)firstVal == 0x8000000000000000ULL &&                    \
+                secondVal == -1LL)                                          \
+            {                                                               \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER_WIDE(vdst, result);                                \
+        } else {                                                            \
+            SET_REGISTER_WIDE(vdst,                                         \
+                (s8) GET_REGISTER_WIDE(vsrc1) _op (s8) GET_REGISTER_WIDE(vsrc2)); \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_LONG(_opcode, _opname, _cast, _op)                    \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-long v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);       \
+        SET_REGISTER_WIDE(vdst,                                             \
+            _cast GET_REGISTER_WIDE(vsrc1) _op (GET_REGISTER(vsrc2) & 0x3f)); \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_LONG_2ADDR(_opcode, _opname, _op, _chkdiv)              \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-long-2addr v%d,v%d", (_opname), vdst, vsrc1);            \
+        if (_chkdiv != 0) {                                                 \
+            s8 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER_WIDE(vdst);                             \
+            secondVal = GET_REGISTER_WIDE(vsrc1);                           \
+            if (secondVal == 0LL) {                                         \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u8)firstVal == 0x8000000000000000ULL &&                    \
+                secondVal == -1LL)                                          \
+            {                                                               \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER_WIDE(vdst, result);                                \
+        } else {                                                            \
+            SET_REGISTER_WIDE(vdst,                                         \
+                (s8) GET_REGISTER_WIDE(vdst) _op (s8)GET_REGISTER_WIDE(vsrc1));\
+        }                                                                   \
+        FINISH(1);
+
+#define HANDLE_OP_SHX_LONG_2ADDR(_opcode, _opname, _cast, _op)              \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-long-2addr v%d,v%d", (_opname), vdst, vsrc1);            \
+        SET_REGISTER_WIDE(vdst,                                             \
+            _cast GET_REGISTER_WIDE(vdst) _op (GET_REGISTER(vsrc1) & 0x3f)); \
+        FINISH(1);
+
+#define HANDLE_OP_X_FLOAT(_opcode, _opname, _op)                            \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-float v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);      \
+        SET_REGISTER_FLOAT(vdst,                                            \
+            GET_REGISTER_FLOAT(vsrc1) _op GET_REGISTER_FLOAT(vsrc2));       \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_DOUBLE(_opcode, _opname, _op)                           \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-double v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);     \
+        SET_REGISTER_DOUBLE(vdst,                                           \
+            GET_REGISTER_DOUBLE(vsrc1) _op GET_REGISTER_DOUBLE(vsrc2));     \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_FLOAT_2ADDR(_opcode, _opname, _op)                      \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-float-2addr v%d,v%d", (_opname), vdst, vsrc1);           \
+        SET_REGISTER_FLOAT(vdst,                                            \
+            GET_REGISTER_FLOAT(vdst) _op GET_REGISTER_FLOAT(vsrc1));        \
+        FINISH(1);
+
+#define HANDLE_OP_X_DOUBLE_2ADDR(_opcode, _opname, _op)                     \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-double-2addr v%d,v%d", (_opname), vdst, vsrc1);          \
+        SET_REGISTER_DOUBLE(vdst,                                           \
+            GET_REGISTER_DOUBLE(vdst) _op GET_REGISTER_DOUBLE(vsrc1));      \
+        FINISH(1);
+
+#define HANDLE_OP_AGET(_opcode, _opname, _type, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        ArrayObject* arrayObj;                                              \
+        u2 arrayInfo;                                                       \
+        EXPORT_PC();                                                        \
+        vdst = INST_AA(inst);                                               \
+        arrayInfo = FETCH(1);                                               \
+        vsrc1 = arrayInfo & 0xff;    /* array ptr */                        \
+        vsrc2 = arrayInfo >> 8;      /* index */                            \
+        ILOGV("|aget%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);        \
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);                      \
+        if (!checkForNull((Object*) arrayObj))                              \
+            GOTO_exceptionThrown();                                         \
+        if (GET_REGISTER(vsrc2) >= arrayObj->length) {                      \
+            LOGV("Invalid array access: %p %d (len=%d)\n",                  \
+                arrayObj, vsrc2, arrayObj->length);                         \
+            dvmThrowException("Ljava/lang/ArrayIndexOutOfBoundsException;", \
+                NULL);                                                      \
+            GOTO_exceptionThrown();                                         \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst,                                        \
+            ((_type*) arrayObj->contents)[GET_REGISTER(vsrc2)]);            \
+        ILOGV("+ AGET[%d]=0x%x", GET_REGISTER(vsrc2), GET_REGISTER(vdst));  \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_APUT(_opcode, _opname, _type, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        ArrayObject* arrayObj;                                              \
+        u2 arrayInfo;                                                       \
+        EXPORT_PC();                                                        \
+        vdst = INST_AA(inst);       /* AA: source value */                  \
+        arrayInfo = FETCH(1);                                               \
+        vsrc1 = arrayInfo & 0xff;   /* BB: array ptr */                     \
+        vsrc2 = arrayInfo >> 8;     /* CC: index */                         \
+        ILOGV("|aput%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);        \
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);                      \
+        if (!checkForNull((Object*) arrayObj))                              \
+            GOTO_exceptionThrown();                                         \
+        if (GET_REGISTER(vsrc2) >= arrayObj->length) {                      \
+            dvmThrowException("Ljava/lang/ArrayIndexOutOfBoundsException;", \
+                NULL);                                                      \
+            GOTO_exceptionThrown();                                         \
+        }                                                                   \
+        ILOGV("+ APUT[%d]=0x%08x", GET_REGISTER(vsrc2), GET_REGISTER(vdst));\
+        ((_type*) arrayObj->contents)[GET_REGISTER(vsrc2)] =                \
+            GET_REGISTER##_regsize(vdst);                                   \
+    }                                                                       \
+    FINISH(2);
+
+/*
+ * It's possible to get a bad value out of a field with sub-32-bit stores
+ * because the -quick versions always operate on 32 bits.  Consider:
+ *   short foo = -1  (sets a 32-bit register to 0xffffffff)
+ *   iput-quick foo  (writes all 32 bits to the field)
+ *   short bar = 1   (sets a 32-bit register to 0x00000001)
+ *   iput-short      (writes the low 16 bits to the field)
+ *   iget-quick foo  (reads all 32 bits from the field, yielding 0xffff0001)
+ * This can only happen when optimized and non-optimized code has interleaved
+ * access to the same field.  This is unlikely but possible.
+ *
+ * The easiest way to fix this is to always read/write 32 bits at a time.  On
+ * a device with a 16-bit data bus this is sub-optimal.  (The alternative
+ * approach is to have sub-int versions of iget-quick, but now we're wasting
+ * Dalvik instruction space and making it less likely that handler code will
+ * already be in the CPU i-cache.)
+ */
+#define HANDLE_IGET_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        InstField* ifield;                                                  \
+        Object* obj;                                                        \
+        EXPORT_PC();                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|iget%s v%d,v%d,field@0x%04x", (_opname), vdst, vsrc1, ref); \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNull(obj))                                             \
+            GOTO_exceptionThrown();                                         \
+        ifield = (InstField*) dvmDexGetResolvedField(methodClassDex, ref);  \
+        if (ifield == NULL) {                                               \
+            ifield = dvmResolveInstField(curMethod->clazz, ref);            \
+            if (ifield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst,                                        \
+            dvmGetField##_ftype(obj, ifield->byteOffset));                  \
+        ILOGV("+ IGET '%s'=0x%08llx", ifield->field.name,                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+        UPDATE_FIELD_GET(&ifield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IGET_X_QUICK(_opcode, _opname, _ftype, _regsize)             \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        Object* obj;                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field offset */                          \
+        ILOGV("|iget%s-quick v%d,v%d,field@+%u",                            \
+            (_opname), vdst, vsrc1, ref);                                   \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNullExportPC(obj, fp, pc))                             \
+            GOTO_exceptionThrown();                                         \
+        SET_REGISTER##_regsize(vdst, dvmGetField##_ftype(obj, ref));        \
+        ILOGV("+ IGETQ %d=0x%08llx", ref,                                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IPUT_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        InstField* ifield;                                                  \
+        Object* obj;                                                        \
+        EXPORT_PC();                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|iput%s v%d,v%d,field@0x%04x", (_opname), vdst, vsrc1, ref); \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNull(obj))                                             \
+            GOTO_exceptionThrown();                                         \
+        ifield = (InstField*) dvmDexGetResolvedField(methodClassDex, ref);  \
+        if (ifield == NULL) {                                               \
+            ifield = dvmResolveInstField(curMethod->clazz, ref);            \
+            if (ifield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        dvmSetField##_ftype(obj, ifield->byteOffset,                        \
+            GET_REGISTER##_regsize(vdst));                                  \
+        ILOGV("+ IPUT '%s'=0x%08llx", ifield->field.name,                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+        UPDATE_FIELD_PUT(&ifield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IPUT_X_QUICK(_opcode, _opname, _ftype, _regsize)             \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        Object* obj;                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field offset */                          \
+        ILOGV("|iput%s-quick v%d,v%d,field@0x%04x",                         \
+            (_opname), vdst, vsrc1, ref);                                   \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNullExportPC(obj, fp, pc))                             \
+            GOTO_exceptionThrown();                                         \
+        dvmSetField##_ftype(obj, ref, GET_REGISTER##_regsize(vdst));        \
+        ILOGV("+ IPUTQ %d=0x%08llx", ref,                                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_SGET_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, field@BBBB*/)                              \
+    {                                                                       \
+        StaticField* sfield;                                                \
+        vdst = INST_AA(inst);                                               \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|sget%s v%d,sfield@0x%04x", (_opname), vdst, ref);           \
+        sfield = (StaticField*)dvmDexGetResolvedField(methodClassDex, ref); \
+        if (sfield == NULL) {                                               \
+            EXPORT_PC();                                                    \
+            sfield = dvmResolveStaticField(curMethod->clazz, ref);          \
+            if (sfield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst, dvmGetStaticField##_ftype(sfield));    \
+        ILOGV("+ SGET '%s'=0x%08llx",                                       \
+            sfield->field.name, (u8)GET_REGISTER##_regsize(vdst));          \
+        UPDATE_FIELD_GET(&sfield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_SPUT_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, field@BBBB*/)                              \
+    {                                                                       \
+        StaticField* sfield;                                                \
+        vdst = INST_AA(inst);                                               \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|sput%s v%d,sfield@0x%04x", (_opname), vdst, ref);           \
+        sfield = (StaticField*)dvmDexGetResolvedField(methodClassDex, ref); \
+        if (sfield == NULL) {                                               \
+            EXPORT_PC();                                                    \
+            sfield = dvmResolveStaticField(curMethod->clazz, ref);          \
+            if (sfield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        dvmSetStaticField##_ftype(sfield, GET_REGISTER##_regsize(vdst));    \
+        ILOGV("+ SPUT '%s'=0x%08llx",                                       \
+            sfield->field.name, (u8)GET_REGISTER##_regsize(vdst));          \
+        UPDATE_FIELD_PUT(&sfield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+
+/* File: cstubs/enddefs.c */
+
+/* undefine "magic" name remapping */
+#undef retval
+#undef pc
+#undef fp
+#undef curMethod
+#undef methodClassDex
+#undef self
+#undef debugTrackedRefStart
+
+/* File: armv5te/debug.c */
+#include <inttypes.h>
+
+/*
+ * Dump the fixed-purpose ARM registers, along with some other info.
+ *
+ * This function MUST be compiled in ARM mode -- THUMB will yield bogus
+ * results.
+ *
+ * This will NOT preserve r0-r3/ip.
+ */
+void dvmMterpDumpArmRegs(uint32_t r0, uint32_t r1, uint32_t r2, uint32_t r3)
+{
+    register uint32_t rPC       asm("r4");
+    register uint32_t rFP       asm("r5");
+    register uint32_t rGLUE     asm("r6");
+    register uint32_t rIBASE    asm("r7");
+    register uint32_t rINST     asm("r8");
+    register uint32_t r9        asm("r9");
+    register uint32_t r10       asm("r10");
+
+    extern char dvmAsmInstructionStart[];
+
+    printf("REGS: r0=%08x r1=%08x r2=%08x r3=%08x\n", r0, r1, r2, r3);
+    printf("    : rPC=%08x rFP=%08x rGLUE=%08x rIBASE=%08x\n",
+        rPC, rFP, rGLUE, rIBASE);
+    printf("    : rINST=%08x r9=%08x r10=%08x\n", rINST, r9, r10);
+
+    MterpGlue* glue = (MterpGlue*) rGLUE;
+    const Method* method = glue->method;
+    printf("    + self is %p\n", dvmThreadSelf());
+    //printf("    + currently in %s.%s %s\n",
+    //    method->clazz->descriptor, method->name, method->signature);
+    //printf("    + dvmAsmInstructionStart = %p\n", dvmAsmInstructionStart);
+    //printf("    + next handler for 0x%02x = %p\n",
+    //    rINST & 0xff, dvmAsmInstructionStart + (rINST & 0xff) * 64);
+}
+
+/*
+ * Dump the StackSaveArea for the specified frame pointer.
+ */
+void dvmDumpFp(void* fp, StackSaveArea* otherSaveArea)
+{
+    StackSaveArea* saveArea = SAVEAREA_FROM_FP(fp);
+    printf("StackSaveArea for fp %p [%p/%p]:\n", fp, saveArea, otherSaveArea);
+#ifdef EASY_GDB
+    printf("  prevSave=%p, prevFrame=%p savedPc=%p meth=%p curPc=%p\n",
+        saveArea->prevSave, saveArea->prevFrame, saveArea->savedPc,
+        saveArea->method, saveArea->xtra.currentPc);
+#else
+    printf("  prevFrame=%p savedPc=%p meth=%p curPc=%p fp[0]=0x%08x\n",
+        saveArea->prevFrame, saveArea->savedPc,
+        saveArea->method, saveArea->xtra.currentPc,
+        *(u4*)fp);
+#endif
+}
+
+/*
+ * Does the bulk of the work for common_printMethod().
+ */
+void dvmMterpPrintMethod(Method* method)
+{
+    /*
+     * It is a direct (non-virtual) method if it is static, private,
+     * or a constructor.
+     */
+    bool isDirect = 
+        ((method->accessFlags & (ACC_STATIC|ACC_PRIVATE)) != 0) ||
+        (method->name[0] == '<');
+
+    char* desc = dexProtoCopyMethodDescriptor(&method->prototype);
+        
+    printf("<%c:%s.%s %s> ",
+            isDirect ? 'D' : 'V',
+            method->clazz->descriptor,
+            method->name,
+            desc);
+
+    free(desc);
+}
+
diff --git a/vm/mterp/out/InterpC-armv5te.c b/vm/mterp/out/InterpC-armv5te.c
new file mode 100644
index 0000000..47c8709
--- /dev/null
+++ b/vm/mterp/out/InterpC-armv5te.c
@@ -0,0 +1,1266 @@
+/*
+ * This file was generated automatically by gen-mterp.py for 'armv5te'.
+ *
+ * --> DO NOT EDIT <--
+ */
+
+/* File: c/header.c */
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/* common includes */
+#include "Dalvik.h"
+#include "interp/InterpDefs.h"
+#include "mterp/Mterp.h"
+#include <math.h>                   // needed for fmod, fmodf
+
+/*
+ * Configuration defines.  These affect the C implementations, i.e. the
+ * portable interpreter(s) and C stubs.
+ *
+ * Some defines are controlled by the Makefile, e.g.:
+ *   WITH_PROFILER
+ *   WITH_DEBUGGER
+ *   WITH_INSTR_CHECKS
+ *   WITH_TRACKREF_CHECKS
+ *   EASY_GDB
+ *   NDEBUG
+ *
+ * If THREADED_INTERP is not defined, we use a classic "while true / switch"
+ * interpreter.  If it is defined, then the tail end of each instruction
+ * handler fetches the next instruction and jumps directly to the handler.
+ * This increases the size of the "Std" interpreter by about 10%, but
+ * provides a speedup of about the same magnitude.
+ *
+ * There's a "hybrid" approach that uses a goto table instead of a switch
+ * statement, avoiding the "is the opcode in range" tests required for switch.
+ * The performance is close to the threaded version, and without the 10%
+ * size increase, but the benchmark results are off enough that it's not
+ * worth adding as a third option.
+ */
+#define THREADED_INTERP             /* threaded vs. while-loop interpreter */
+
+#ifdef WITH_INSTR_CHECKS            /* instruction-level paranoia */
+# define CHECK_BRANCH_OFFSETS
+# define CHECK_REGISTER_INDICES
+#endif
+
+/*
+ * ARM EABI requires 64-bit alignment for access to 64-bit data types.  We
+ * can't just use pointers to copy 64-bit values out of our interpreted
+ * register set, because gcc will generate ldrd/strd.
+ *
+ * The __UNION version copies data in and out of a union.  The __MEMCPY
+ * version uses a memcpy() call to do the transfer; gcc is smart enough to
+ * not actually call memcpy().  The __UNION version is very bad on ARM;
+ * it only uses one more instruction than __MEMCPY, but for some reason
+ * gcc thinks it needs separate storage for every instance of the union.
+ * On top of that, it feels the need to zero them out at the start of the
+ * method.  Net result is we zero out ~700 bytes of stack space at the top
+ * of the interpreter using ARM STM instructions.
+ */
+#if defined(__ARM_EABI__)
+//# define NO_UNALIGN_64__UNION
+# define NO_UNALIGN_64__MEMCPY
+#endif
+
+//#define LOG_INSTR                   /* verbose debugging */
+/* set and adjust ANDROID_LOG_TAGS='*:i jdwp:i dalvikvm:i dalvikvmi:i' */
+
+/*
+ * Keep a tally of accesses to fields.  Currently only works if full DEX
+ * optimization is disabled.
+ */
+#ifdef PROFILE_FIELD_ACCESS
+# define UPDATE_FIELD_GET(_field) { (_field)->gets++; }
+# define UPDATE_FIELD_PUT(_field) { (_field)->puts++; }
+#else
+# define UPDATE_FIELD_GET(_field) ((void)0)
+# define UPDATE_FIELD_PUT(_field) ((void)0)
+#endif
+
+/*
+ * Adjust the program counter.  "_offset" is a signed int, in 16-bit units.
+ *
+ * Assumes the existence of "const u2* pc" and "const u2* curMethod->insns".
+ *
+ * We don't advance the program counter until we finish an instruction or
+ * branch, because we do want to have to unroll the PC if there's an
+ * exception.
+ */
+#ifdef CHECK_BRANCH_OFFSETS
+# define ADJUST_PC(_offset) do {                                            \
+        int myoff = _offset;        /* deref only once */                   \
+        if (pc + myoff < curMethod->insns ||                                \
+            pc + myoff >= curMethod->insns + dvmGetMethodInsnsSize(curMethod)) \
+        {                                                                   \
+            char* desc;                                                     \
+            desc = dexProtoCopyMethodDescriptor(&curMethod->prototype);     \
+            LOGE("Invalid branch %d at 0x%04x in %s.%s %s\n",               \
+                myoff, (int) (pc - curMethod->insns),                       \
+                curMethod->clazz->descriptor, curMethod->name, desc);       \
+            free(desc);                                                     \
+            dvmAbort();                                                     \
+        }                                                                   \
+        pc += myoff;                                                        \
+    } while (false)
+#else
+# define ADJUST_PC(_offset) (pc += _offset)
+#endif
+
+/*
+ * If enabled, log instructions as we execute them.
+ */
+#ifdef LOG_INSTR
+# define ILOGD(...) ILOG(LOG_DEBUG, __VA_ARGS__)
+# define ILOGV(...) ILOG(LOG_VERBOSE, __VA_ARGS__)
+# define ILOG(_level, ...) do {                                             \
+        char debugStrBuf[128];                                              \
+        snprintf(debugStrBuf, sizeof(debugStrBuf), __VA_ARGS__);            \
+        if (curMethod != NULL)                                                 \
+            LOG(_level, LOG_TAG"i", "%-2d|%04x%s\n",                        \
+                self->threadId, (int)(pc - curMethod->insns), debugStrBuf); \
+        else                                                                \
+            LOG(_level, LOG_TAG"i", "%-2d|####%s\n",                        \
+                self->threadId, debugStrBuf);                               \
+    } while(false)
+void dvmDumpRegs(const Method* method, const u4* framePtr, bool inOnly);
+# define DUMP_REGS(_meth, _frame, _inOnly) dvmDumpRegs(_meth, _frame, _inOnly)
+static const char kSpacing[] = "            ";
+#else
+# define ILOGD(...) ((void)0)
+# define ILOGV(...) ((void)0)
+# define DUMP_REGS(_meth, _frame, _inOnly) ((void)0)
+#endif
+
+/* get a long from an array of u4 */
+static inline s8 getLongFromArray(const u4* ptr, int idx)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { s8 ll; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.parts[0] = ptr[0];
+    conv.parts[1] = ptr[1];
+    return conv.ll;
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    s8 val;
+    memcpy(&val, &ptr[idx], 8);
+    return val;
+#else
+    return *((s8*) &ptr[idx]);
+#endif
+}
+
+/* store a long into an array of u4 */
+static inline void putLongToArray(u4* ptr, int idx, s8 val)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { s8 ll; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.ll = val;
+    ptr[0] = conv.parts[0];
+    ptr[1] = conv.parts[1];
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    memcpy(&ptr[idx], &val, 8);
+#else
+    *((s8*) &ptr[idx]) = val;
+#endif
+}
+
+/* get a double from an array of u4 */
+static inline double getDoubleFromArray(const u4* ptr, int idx)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { double d; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.parts[0] = ptr[0];
+    conv.parts[1] = ptr[1];
+    return conv.d;
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    double dval;
+    memcpy(&dval, &ptr[idx], 8);
+    return dval;
+#else
+    return *((double*) &ptr[idx]);
+#endif
+}
+
+/* store a double into an array of u4 */
+static inline void putDoubleToArray(u4* ptr, int idx, double dval)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { double d; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.d = dval;
+    ptr[0] = conv.parts[0];
+    ptr[1] = conv.parts[1];
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    memcpy(&ptr[idx], &dval, 8);
+#else
+    *((double*) &ptr[idx]) = dval;
+#endif
+}
+
+/*
+ * If enabled, validate the register number on every access.  Otherwise,
+ * just do an array access.
+ *
+ * Assumes the existence of "u4* fp".
+ *
+ * "_idx" may be referenced more than once.
+ */
+#ifdef CHECK_REGISTER_INDICES
+# define GET_REGISTER(_idx) \
+    ( (_idx) < curMethod->registersSize ? \
+        (fp[(_idx)]) : (assert(!"bad reg"),1969) )
+# define SET_REGISTER(_idx, _val) \
+    ( (_idx) < curMethod->registersSize ? \
+        (fp[(_idx)] = (u4)(_val)) : (assert(!"bad reg"),1969) )
+# define GET_REGISTER_AS_OBJECT(_idx)       ((Object *)GET_REGISTER(_idx))
+# define SET_REGISTER_AS_OBJECT(_idx, _val) SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_INT(_idx) ((s4) GET_REGISTER(_idx))
+# define SET_REGISTER_INT(_idx, _val) SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_WIDE(_idx) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        getLongFromArray(fp, (_idx)) : (assert(!"bad reg"),1969) )
+# define SET_REGISTER_WIDE(_idx, _val) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        putLongToArray(fp, (_idx), (_val)) : (assert(!"bad reg"),1969) )
+# define GET_REGISTER_FLOAT(_idx) \
+    ( (_idx) < curMethod->registersSize ? \
+        (*((float*) &fp[(_idx)])) : (assert(!"bad reg"),1969.0f) )
+# define SET_REGISTER_FLOAT(_idx, _val) \
+    ( (_idx) < curMethod->registersSize ? \
+        (*((float*) &fp[(_idx)]) = (_val)) : (assert(!"bad reg"),1969.0f) )
+# define GET_REGISTER_DOUBLE(_idx) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        getDoubleFromArray(fp, (_idx)) : (assert(!"bad reg"),1969.0) )
+# define SET_REGISTER_DOUBLE(_idx, _val) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        putDoubleToArray(fp, (_idx), (_val)) : (assert(!"bad reg"),1969.0) )
+#else
+# define GET_REGISTER(_idx)                 (fp[(_idx)])
+# define SET_REGISTER(_idx, _val)           (fp[(_idx)] = (_val))
+# define GET_REGISTER_AS_OBJECT(_idx)       ((Object*) fp[(_idx)])
+# define SET_REGISTER_AS_OBJECT(_idx, _val) (fp[(_idx)] = (u4)(_val))
+# define GET_REGISTER_INT(_idx)             ((s4)GET_REGISTER(_idx))
+# define SET_REGISTER_INT(_idx, _val)       SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_WIDE(_idx)            getLongFromArray(fp, (_idx))
+# define SET_REGISTER_WIDE(_idx, _val)      putLongToArray(fp, (_idx), (_val))
+# define GET_REGISTER_FLOAT(_idx)           (*((float*) &fp[(_idx)]))
+# define SET_REGISTER_FLOAT(_idx, _val)     (*((float*) &fp[(_idx)]) = (_val))
+# define GET_REGISTER_DOUBLE(_idx)          getDoubleFromArray(fp, (_idx))
+# define SET_REGISTER_DOUBLE(_idx, _val)    putDoubleToArray(fp, (_idx), (_val))
+#endif
+
+/*
+ * Get 16 bits from the specified offset of the program counter.  We always
+ * want to load 16 bits at a time from the instruction stream -- it's more
+ * efficient than 8 and won't have the alignment problems that 32 might.
+ *
+ * Assumes existence of "const u2* pc".
+ */
+#define FETCH(_offset)     (pc[(_offset)])
+
+/*
+ * Extract instruction byte from 16-bit fetch (_inst is a u2).
+ */
+#define INST_INST(_inst)    ((_inst) & 0xff)
+
+/*
+ * Extract the "vA, vB" 4-bit registers from the instruction word (_inst is u2).
+ */
+#define INST_A(_inst)       (((_inst) >> 8) & 0x0f)
+#define INST_B(_inst)       ((_inst) >> 12)
+
+/*
+ * Get the 8-bit "vAA" 8-bit register index from the instruction word.
+ * (_inst is u2)
+ */
+#define INST_AA(_inst)      ((_inst) >> 8)
+
+/*
+ * The current PC must be available to Throwable constructors, e.g.
+ * those created by dvmThrowException(), so that the exception stack
+ * trace can be generated correctly.  If we don't do this, the offset
+ * within the current method won't be shown correctly.  See the notes
+ * in Exception.c.
+ *
+ * Assumes existence of "u4* fp" and "const u2* pc".
+ */
+#define EXPORT_PC()         (SAVEAREA_FROM_FP(fp)->xtra.currentPc = pc)
+
+/*
+ * Determine if we need to switch to a different interpreter.  "_current"
+ * is either INTERP_STD or INTERP_DBG.  It should be fixed for a given
+ * interpreter generation file, which should remove the outer conditional
+ * from the following.
+ *
+ * If we're building without debug and profiling support, we never switch.
+ */
+#if defined(WITH_PROFILER) || defined(WITH_DEBUGGER)
+# define NEED_INTERP_SWITCH(_current) (                                     \
+    (_current == INTERP_STD) ?                                              \
+        dvmDebuggerOrProfilerActive() : !dvmDebuggerOrProfilerActive() )
+#else
+# define NEED_INTERP_SWITCH(_current) (false)
+#endif
+
+/*
+ * Look up an interface on a class using the cache.
+ */
+INLINE Method* dvmFindInterfaceMethodInCache(ClassObject* thisClass,
+    u4 methodIdx, const Method* method, DvmDex* methodClassDex)
+{
+#define ATOMIC_CACHE_CALC \
+    dvmInterpFindInterfaceMethod(thisClass, methodIdx, method, methodClassDex)
+
+    return (Method*) ATOMIC_CACHE_LOOKUP(methodClassDex->pInterfaceCache,
+                DEX_INTERFACE_CACHE_SIZE, thisClass, methodIdx);
+
+#undef ATOMIC_CACHE_CALC
+}
+
+/*
+ * Check to see if "obj" is NULL.  If so, throw an exception.  Assumes the
+ * pc has already been exported to the stack.
+ *
+ * Perform additional checks on debug builds.
+ *
+ * Use this to check for NULL when the instruction handler calls into
+ * something that could throw an exception (so we have already called
+ * EXPORT_PC at the top).
+ */
+static inline bool checkForNull(Object* obj)
+{
+    if (obj == NULL) {
+        dvmThrowException("Ljava/lang/NullPointerException;", NULL);
+        return false;
+    }
+#ifdef WITH_EXTRA_OBJECT_VALIDATION
+    if (!dvmIsValidObject(obj)) {
+        LOGE("Invalid object %p\n", obj);
+        dvmAbort();
+    }
+#endif
+#ifndef NDEBUG
+    if (obj->clazz == NULL || ((u4) obj->clazz) <= 65536) {
+        /* probable heap corruption */
+        LOGE("Invalid object class %p (in %p)\n", obj->clazz, obj);
+        dvmAbort();
+    }
+#endif
+    return true;
+}
+
+/*
+ * Check to see if "obj" is NULL.  If so, export the PC into the stack
+ * frame and throw an exception.
+ *
+ * Perform additional checks on debug builds.
+ *
+ * Use this to check for NULL when the instruction handler doesn't do
+ * anything else that can throw an exception.
+ */
+static inline bool checkForNullExportPC(Object* obj, u4* fp, const u2* pc)
+{
+    if (obj == NULL) {
+        EXPORT_PC();
+        dvmThrowException("Ljava/lang/NullPointerException;", NULL);
+        return false;
+    }
+#ifdef WITH_EXTRA_OBJECT_VALIDATION
+    if (!dvmIsValidObject(obj)) {
+        LOGE("Invalid object %p\n", obj);
+        dvmAbort();
+    }
+#endif
+#ifndef NDEBUG
+    if (obj->clazz == NULL || ((u4) obj->clazz) <= 65536) {
+        /* probable heap corruption */
+        LOGE("Invalid object class %p (in %p)\n", obj->clazz, obj);
+        dvmAbort();
+    }
+#endif
+    return true;
+}
+
+
+/* File: cstubs/stubdefs.c */
+/* this is a standard (no debug support) interpreter */
+#define INTERP_TYPE INTERP_STD
+#define CHECK_DEBUG_AND_PROF() ((void)0)
+# define CHECK_TRACKED_REFS() ((void)0)
+
+/*
+ * In the C mterp stubs, "goto" is a function call followed immediately
+ * by a return.
+ */
+
+#define GOTO_TARGET_DECL(_target, ...)                                      \
+    void dvmMterp_##_target(MterpGlue* glue, ## __VA_ARGS__);
+
+#define GOTO_TARGET(_target, ...)                                           \
+    void dvmMterp_##_target(MterpGlue* glue, ## __VA_ARGS__) {              \
+        u2 ref, vsrc1, vsrc2, vdst;                                         \
+        u2 inst = FETCH(0);                                                 \
+        const Method* methodToCall;                                         \
+        StackSaveArea* debugSaveArea;
+
+#define GOTO_TARGET_END }
+
+/*
+ * Redefine what used to be local variable accesses into MterpGlue struct
+ * references.  (These are undefined down in "footer.c".)
+ */
+#define retval                  glue->retval
+#define pc                      glue->pc
+#define fp                      glue->fp
+#define curMethod               glue->method
+#define methodClassDex          glue->methodClassDex
+#define self                    glue->self
+#define debugTrackedRefStart    glue->debugTrackedRefStart
+
+/* ugh */
+#define STUB_HACK(x) x
+
+
+/*
+ * Opcode handler framing macros.  Here, each opcode is a separate function
+ * that takes a "glue" argument and returns void.  We can't declare
+ * these "static" because they may be called from an assembly stub.
+ */
+#define HANDLE_OPCODE(_op)                                                  \
+    void dvmMterp_##_op(MterpGlue* glue) {                                  \
+        u2 ref, vsrc1, vsrc2, vdst;                                         \
+        u2 inst = FETCH(0);
+
+#define OP_END }
+
+/*
+ * Like the "portable" FINISH, but don't reload "inst", and return to caller
+ * when done.
+ */
+#define FINISH(_offset) {                                                   \
+        ADJUST_PC(_offset);                                                 \
+        CHECK_DEBUG_AND_PROF();                                             \
+        CHECK_TRACKED_REFS();                                               \
+        return;                                                             \
+    }
+
+
+/*
+ * The "goto label" statements turn into function calls followed by
+ * return statements.  Some of the functions take arguments, which in the
+ * portable interpreter are handled by assigning values to globals.
+ */
+
+#define GOTO_exceptionThrown()                                              \
+    do {                                                                    \
+        dvmMterp_exceptionThrown(glue);                                     \
+        return;                                                             \
+    } while(false)
+
+#define GOTO_returnFromMethod()                                             \
+    do {                                                                    \
+        dvmMterp_returnFromMethod(glue);                                    \
+        return;                                                             \
+    } while(false)
+
+#define GOTO_invoke(_target, _methodCallRange)                              \
+    do {                                                                    \
+        dvmMterp_##_target(glue, _methodCallRange);                         \
+        return;                                                             \
+    } while(false)
+
+#define GOTO_invokeMethod(_methodCallRange, _methodToCall, _vsrc1, _vdst)   \
+    do {                                                                    \
+        dvmMterp_invokeMethod(glue, _methodCallRange, _methodToCall,        \
+            _vsrc1, _vdst);                                                 \
+        return;                                                             \
+    } while(false)
+
+/*
+ * As a special case, "goto bail" turns into a longjmp.  Use "bail_switch"
+ * if we need to switch to the other interpreter upon our return.
+ */
+#define GOTO_bail()                                                         \
+    dvmMterpStdBail(glue, false);
+#define GOTO_bail_switch()                                                  \
+    dvmMterpStdBail(glue, true);
+
+/*
+ * Periodically check for thread suspension.
+ *
+ * While we're at it, see if a debugger has attached or the profiler has
+ * started.  If so, switch to a different "goto" table.
+ */
+#define PERIODIC_CHECKS(_entryPoint, _pcadj) {                              \
+        dvmCheckSuspendQuick(self);                                         \
+        if (NEED_INTERP_SWITCH(INTERP_TYPE)) {                              \
+            ADJUST_PC(_pcadj);                                              \
+            glue->entryPoint = _entryPoint;                                 \
+            LOGVV("threadid=%d: switch to STD ep=%d adj=%d\n",              \
+                glue->self->threadId, (_entryPoint), (_pcadj));             \
+            GOTO_bail_switch();                                             \
+        }                                                                   \
+    }
+
+
+/* File: c/opcommon.c */
+/* forward declarations of goto targets */
+GOTO_TARGET_DECL(filledNewArray, bool methodCallRange);
+GOTO_TARGET_DECL(invokeVirtual, bool methodCallRange);
+GOTO_TARGET_DECL(invokeSuper, bool methodCallRange);
+GOTO_TARGET_DECL(invokeInterface, bool methodCallRange);
+GOTO_TARGET_DECL(invokeDirect, bool methodCallRange);
+GOTO_TARGET_DECL(invokeStatic, bool methodCallRange);
+GOTO_TARGET_DECL(invokeVirtualQuick, bool methodCallRange);
+GOTO_TARGET_DECL(invokeSuperQuick, bool methodCallRange);
+GOTO_TARGET_DECL(invokeMethod, bool methodCallRange, const Method* methodToCall,
+    u2 count, u2 regs);
+GOTO_TARGET_DECL(returnFromMethod);
+GOTO_TARGET_DECL(exceptionThrown);
+
+/*
+ * ===========================================================================
+ *
+ * What follows are opcode definitions shared between multiple opcodes with
+ * minor substitutions handled by the C pre-processor.  These should probably
+ * use the mterp substitution mechanism instead, with the code here moved
+ * into common fragment files (like the asm "binop.S"), although it's hard
+ * to give up the C preprocessor in favor of the much simpler text subst.
+ *
+ * ===========================================================================
+ */
+
+#define HANDLE_NUMCONV(_opcode, _opname, _fromtype, _totype)                \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        SET_REGISTER##_totype(vdst,                                         \
+            GET_REGISTER##_fromtype(vsrc1));                                \
+        FINISH(1);
+
+#define HANDLE_FLOAT_TO_INT(_opcode, _opname, _fromvtype, _fromrtype,       \
+        _tovtype, _tortype)                                                 \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+    {                                                                       \
+        /* spec defines specific handling for +/- inf and NaN values */     \
+        _fromvtype val;                                                     \
+        _tovtype intMin, intMax, result;                                    \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        val = GET_REGISTER##_fromrtype(vsrc1);                              \
+        intMin = (_tovtype) 1 << (sizeof(_tovtype) * 8 -1);                 \
+        intMax = ~intMin;                                                   \
+        result = (_tovtype) val;                                            \
+        if (val >= intMax)          /* +inf */                              \
+            result = intMax;                                                \
+        else if (val <= intMin)     /* -inf */                              \
+            result = intMin;                                                \
+        else if (val != val)        /* NaN */                               \
+            result = 0;                                                     \
+        else                                                                \
+            result = (_tovtype) val;                                        \
+        SET_REGISTER##_tortype(vdst, result);                               \
+    }                                                                       \
+    FINISH(1);
+
+#define HANDLE_INT_TO_SMALL(_opcode, _opname, _type)                        \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|int-to-%s v%d,v%d", (_opname), vdst, vsrc1);                \
+        SET_REGISTER(vdst, (_type) GET_REGISTER(vsrc1));                    \
+        FINISH(1);
+
+/* NOTE: the comparison result is always a signed 4-byte integer */
+#define HANDLE_OP_CMPX(_opcode, _opname, _varType, _type, _nanVal)          \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        int result;                                                         \
+        u2 regs;                                                            \
+        _varType val1, val2;                                                \
+        vdst = INST_AA(inst);                                               \
+        regs = FETCH(1);                                                    \
+        vsrc1 = regs & 0xff;                                                \
+        vsrc2 = regs >> 8;                                                  \
+        ILOGV("|cmp%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);         \
+        val1 = GET_REGISTER##_type(vsrc1);                                  \
+        val2 = GET_REGISTER##_type(vsrc2);                                  \
+        if (val1 == val2)                                                   \
+            result = 0;                                                     \
+        else if (val1 < val2)                                               \
+            result = -1;                                                    \
+        else if (val1 > val2)                                               \
+            result = 1;                                                     \
+        else                                                                \
+            result = (_nanVal);                                             \
+        ILOGV("+ result=%d\n", result);                                     \
+        SET_REGISTER(vdst, result);                                         \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_IF_XX(_opcode, _opname, _cmp)                             \
+    HANDLE_OPCODE(_opcode /*vA, vB, +CCCC*/)                                \
+        vsrc1 = INST_A(inst);                                               \
+        vsrc2 = INST_B(inst);                                               \
+        if ((s4) GET_REGISTER(vsrc1) _cmp (s4) GET_REGISTER(vsrc2)) {       \
+            int branchOffset = (s2)FETCH(1);    /* sign-extended */         \
+            ILOGV("|if-%s v%d,v%d,+0x%04x", (_opname), vsrc1, vsrc2,        \
+                branchOffset);                                              \
+            ILOGV("> branch taken");                                        \
+            if (branchOffset < 0)                                           \
+                PERIODIC_CHECKS(kInterpEntryInstr, branchOffset);           \
+            FINISH(branchOffset);                                           \
+        } else {                                                            \
+            ILOGV("|if-%s v%d,v%d,-", (_opname), vsrc1, vsrc2);             \
+            FINISH(2);                                                      \
+        }
+
+#define HANDLE_OP_IF_XXZ(_opcode, _opname, _cmp)                            \
+    HANDLE_OPCODE(_opcode /*vAA, +BBBB*/)                                   \
+        vsrc1 = INST_AA(inst);                                              \
+        if ((s4) GET_REGISTER(vsrc1) _cmp 0) {                              \
+            int branchOffset = (s2)FETCH(1);    /* sign-extended */         \
+            ILOGV("|if-%s v%d,+0x%04x", (_opname), vsrc1, branchOffset);    \
+            ILOGV("> branch taken");                                        \
+            if (branchOffset < 0)                                           \
+                PERIODIC_CHECKS(kInterpEntryInstr, branchOffset);           \
+            FINISH(branchOffset);                                           \
+        } else {                                                            \
+            ILOGV("|if-%s v%d,-", (_opname), vsrc1);                        \
+            FINISH(2);                                                      \
+        }
+
+#define HANDLE_UNOP(_opcode, _opname, _pfx, _sfx, _type)                    \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        SET_REGISTER##_type(vdst, _pfx GET_REGISTER##_type(vsrc1) _sfx);    \
+        FINISH(1);
+
+#define HANDLE_OP_X_INT(_opcode, _opname, _op, _chkdiv)                     \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-int v%d,v%d", (_opname), vdst, vsrc1);                   \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            secondVal = GET_REGISTER(vsrc2);                                \
+            if (secondVal == 0) {                                           \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && secondVal == -1) {            \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            /* non-div/rem case */                                          \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vsrc1) _op (s4) GET_REGISTER(vsrc2));     \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_INT(_opcode, _opname, _cast, _op)                     \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-int v%d,v%d", (_opname), vdst, vsrc1);                   \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vsrc1) _op (GET_REGISTER(vsrc2) & 0x1f));    \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_INT_LIT16(_opcode, _opname, _op, _chkdiv)               \
+    HANDLE_OPCODE(_opcode /*vA, vB, #+CCCC*/)                               \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        vsrc2 = FETCH(1);                                                   \
+        ILOGV("|%s-int/lit16 v%d,v%d,#+0x%04x",                             \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, result;                                            \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            if ((s2) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                      \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && ((s2) vsrc2) == -1) {         \
+                /* won't generate /lit16 instr for this; check anyway */    \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op (s2) vsrc2;                           \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            /* non-div/rem case */                                          \
+            SET_REGISTER(vdst, GET_REGISTER(vsrc1) _op (s2) vsrc2);         \
+        }                                                                   \
+        FINISH(2);
+
+#define HANDLE_OP_X_INT_LIT8(_opcode, _opname, _op, _chkdiv)                \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, #+CC*/)                               \
+    {                                                                       \
+        u2 litInfo;                                                         \
+        vdst = INST_AA(inst);                                               \
+        litInfo = FETCH(1);                                                 \
+        vsrc1 = litInfo & 0xff;                                             \
+        vsrc2 = litInfo >> 8;       /* constant */                          \
+        ILOGV("|%s-int/lit8 v%d,v%d,#+0x%02x",                              \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, result;                                            \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            if ((s1) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && ((s1) vsrc2) == -1) {         \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op ((s1) vsrc2);                         \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vsrc1) _op (s1) vsrc2);                   \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_INT_LIT8(_opcode, _opname, _cast, _op)                \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, #+CC*/)                               \
+    {                                                                       \
+        u2 litInfo;                                                         \
+        vdst = INST_AA(inst);                                               \
+        litInfo = FETCH(1);                                                 \
+        vsrc1 = litInfo & 0xff;                                             \
+        vsrc2 = litInfo >> 8;       /* constant */                          \
+        ILOGV("|%s-int/lit8 v%d,v%d,#+0x%02x",                              \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vsrc1) _op (vsrc2 & 0x1f));                  \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_INT_2ADDR(_opcode, _opname, _op, _chkdiv)               \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-int-2addr v%d,v%d", (_opname), vdst, vsrc1);             \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER(vdst);                                  \
+            secondVal = GET_REGISTER(vsrc1);                                \
+            if (secondVal == 0) {                                           \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && secondVal == -1) {            \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vdst) _op (s4) GET_REGISTER(vsrc1));      \
+        }                                                                   \
+        FINISH(1);
+
+#define HANDLE_OP_SHX_INT_2ADDR(_opcode, _opname, _cast, _op)               \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-int-2addr v%d,v%d", (_opname), vdst, vsrc1);             \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vdst) _op (GET_REGISTER(vsrc1) & 0x1f));     \
+        FINISH(1);
+
+#define HANDLE_OP_X_LONG(_opcode, _opname, _op, _chkdiv)                    \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-long v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);       \
+        if (_chkdiv != 0) {                                                 \
+            s8 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER_WIDE(vsrc1);                            \
+            secondVal = GET_REGISTER_WIDE(vsrc2);                           \
+            if (secondVal == 0LL) {                                         \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u8)firstVal == 0x8000000000000000ULL &&                    \
+                secondVal == -1LL)                                          \
+            {                                                               \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER_WIDE(vdst, result);                                \
+        } else {                                                            \
+            SET_REGISTER_WIDE(vdst,                                         \
+                (s8) GET_REGISTER_WIDE(vsrc1) _op (s8) GET_REGISTER_WIDE(vsrc2)); \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_LONG(_opcode, _opname, _cast, _op)                    \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-long v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);       \
+        SET_REGISTER_WIDE(vdst,                                             \
+            _cast GET_REGISTER_WIDE(vsrc1) _op (GET_REGISTER(vsrc2) & 0x3f)); \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_LONG_2ADDR(_opcode, _opname, _op, _chkdiv)              \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-long-2addr v%d,v%d", (_opname), vdst, vsrc1);            \
+        if (_chkdiv != 0) {                                                 \
+            s8 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER_WIDE(vdst);                             \
+            secondVal = GET_REGISTER_WIDE(vsrc1);                           \
+            if (secondVal == 0LL) {                                         \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u8)firstVal == 0x8000000000000000ULL &&                    \
+                secondVal == -1LL)                                          \
+            {                                                               \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER_WIDE(vdst, result);                                \
+        } else {                                                            \
+            SET_REGISTER_WIDE(vdst,                                         \
+                (s8) GET_REGISTER_WIDE(vdst) _op (s8)GET_REGISTER_WIDE(vsrc1));\
+        }                                                                   \
+        FINISH(1);
+
+#define HANDLE_OP_SHX_LONG_2ADDR(_opcode, _opname, _cast, _op)              \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-long-2addr v%d,v%d", (_opname), vdst, vsrc1);            \
+        SET_REGISTER_WIDE(vdst,                                             \
+            _cast GET_REGISTER_WIDE(vdst) _op (GET_REGISTER(vsrc1) & 0x3f)); \
+        FINISH(1);
+
+#define HANDLE_OP_X_FLOAT(_opcode, _opname, _op)                            \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-float v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);      \
+        SET_REGISTER_FLOAT(vdst,                                            \
+            GET_REGISTER_FLOAT(vsrc1) _op GET_REGISTER_FLOAT(vsrc2));       \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_DOUBLE(_opcode, _opname, _op)                           \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-double v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);     \
+        SET_REGISTER_DOUBLE(vdst,                                           \
+            GET_REGISTER_DOUBLE(vsrc1) _op GET_REGISTER_DOUBLE(vsrc2));     \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_FLOAT_2ADDR(_opcode, _opname, _op)                      \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-float-2addr v%d,v%d", (_opname), vdst, vsrc1);           \
+        SET_REGISTER_FLOAT(vdst,                                            \
+            GET_REGISTER_FLOAT(vdst) _op GET_REGISTER_FLOAT(vsrc1));        \
+        FINISH(1);
+
+#define HANDLE_OP_X_DOUBLE_2ADDR(_opcode, _opname, _op)                     \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-double-2addr v%d,v%d", (_opname), vdst, vsrc1);          \
+        SET_REGISTER_DOUBLE(vdst,                                           \
+            GET_REGISTER_DOUBLE(vdst) _op GET_REGISTER_DOUBLE(vsrc1));      \
+        FINISH(1);
+
+#define HANDLE_OP_AGET(_opcode, _opname, _type, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        ArrayObject* arrayObj;                                              \
+        u2 arrayInfo;                                                       \
+        EXPORT_PC();                                                        \
+        vdst = INST_AA(inst);                                               \
+        arrayInfo = FETCH(1);                                               \
+        vsrc1 = arrayInfo & 0xff;    /* array ptr */                        \
+        vsrc2 = arrayInfo >> 8;      /* index */                            \
+        ILOGV("|aget%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);        \
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);                      \
+        if (!checkForNull((Object*) arrayObj))                              \
+            GOTO_exceptionThrown();                                         \
+        if (GET_REGISTER(vsrc2) >= arrayObj->length) {                      \
+            LOGV("Invalid array access: %p %d (len=%d)\n",                  \
+                arrayObj, vsrc2, arrayObj->length);                         \
+            dvmThrowException("Ljava/lang/ArrayIndexOutOfBoundsException;", \
+                NULL);                                                      \
+            GOTO_exceptionThrown();                                         \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst,                                        \
+            ((_type*) arrayObj->contents)[GET_REGISTER(vsrc2)]);            \
+        ILOGV("+ AGET[%d]=0x%x", GET_REGISTER(vsrc2), GET_REGISTER(vdst));  \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_APUT(_opcode, _opname, _type, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        ArrayObject* arrayObj;                                              \
+        u2 arrayInfo;                                                       \
+        EXPORT_PC();                                                        \
+        vdst = INST_AA(inst);       /* AA: source value */                  \
+        arrayInfo = FETCH(1);                                               \
+        vsrc1 = arrayInfo & 0xff;   /* BB: array ptr */                     \
+        vsrc2 = arrayInfo >> 8;     /* CC: index */                         \
+        ILOGV("|aput%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);        \
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);                      \
+        if (!checkForNull((Object*) arrayObj))                              \
+            GOTO_exceptionThrown();                                         \
+        if (GET_REGISTER(vsrc2) >= arrayObj->length) {                      \
+            dvmThrowException("Ljava/lang/ArrayIndexOutOfBoundsException;", \
+                NULL);                                                      \
+            GOTO_exceptionThrown();                                         \
+        }                                                                   \
+        ILOGV("+ APUT[%d]=0x%08x", GET_REGISTER(vsrc2), GET_REGISTER(vdst));\
+        ((_type*) arrayObj->contents)[GET_REGISTER(vsrc2)] =                \
+            GET_REGISTER##_regsize(vdst);                                   \
+    }                                                                       \
+    FINISH(2);
+
+/*
+ * It's possible to get a bad value out of a field with sub-32-bit stores
+ * because the -quick versions always operate on 32 bits.  Consider:
+ *   short foo = -1  (sets a 32-bit register to 0xffffffff)
+ *   iput-quick foo  (writes all 32 bits to the field)
+ *   short bar = 1   (sets a 32-bit register to 0x00000001)
+ *   iput-short      (writes the low 16 bits to the field)
+ *   iget-quick foo  (reads all 32 bits from the field, yielding 0xffff0001)
+ * This can only happen when optimized and non-optimized code has interleaved
+ * access to the same field.  This is unlikely but possible.
+ *
+ * The easiest way to fix this is to always read/write 32 bits at a time.  On
+ * a device with a 16-bit data bus this is sub-optimal.  (The alternative
+ * approach is to have sub-int versions of iget-quick, but now we're wasting
+ * Dalvik instruction space and making it less likely that handler code will
+ * already be in the CPU i-cache.)
+ */
+#define HANDLE_IGET_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        InstField* ifield;                                                  \
+        Object* obj;                                                        \
+        EXPORT_PC();                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|iget%s v%d,v%d,field@0x%04x", (_opname), vdst, vsrc1, ref); \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNull(obj))                                             \
+            GOTO_exceptionThrown();                                         \
+        ifield = (InstField*) dvmDexGetResolvedField(methodClassDex, ref);  \
+        if (ifield == NULL) {                                               \
+            ifield = dvmResolveInstField(curMethod->clazz, ref);            \
+            if (ifield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst,                                        \
+            dvmGetField##_ftype(obj, ifield->byteOffset));                  \
+        ILOGV("+ IGET '%s'=0x%08llx", ifield->field.name,                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+        UPDATE_FIELD_GET(&ifield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IGET_X_QUICK(_opcode, _opname, _ftype, _regsize)             \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        Object* obj;                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field offset */                          \
+        ILOGV("|iget%s-quick v%d,v%d,field@+%u",                            \
+            (_opname), vdst, vsrc1, ref);                                   \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNullExportPC(obj, fp, pc))                             \
+            GOTO_exceptionThrown();                                         \
+        SET_REGISTER##_regsize(vdst, dvmGetField##_ftype(obj, ref));        \
+        ILOGV("+ IGETQ %d=0x%08llx", ref,                                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IPUT_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        InstField* ifield;                                                  \
+        Object* obj;                                                        \
+        EXPORT_PC();                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|iput%s v%d,v%d,field@0x%04x", (_opname), vdst, vsrc1, ref); \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNull(obj))                                             \
+            GOTO_exceptionThrown();                                         \
+        ifield = (InstField*) dvmDexGetResolvedField(methodClassDex, ref);  \
+        if (ifield == NULL) {                                               \
+            ifield = dvmResolveInstField(curMethod->clazz, ref);            \
+            if (ifield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        dvmSetField##_ftype(obj, ifield->byteOffset,                        \
+            GET_REGISTER##_regsize(vdst));                                  \
+        ILOGV("+ IPUT '%s'=0x%08llx", ifield->field.name,                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+        UPDATE_FIELD_PUT(&ifield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IPUT_X_QUICK(_opcode, _opname, _ftype, _regsize)             \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        Object* obj;                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field offset */                          \
+        ILOGV("|iput%s-quick v%d,v%d,field@0x%04x",                         \
+            (_opname), vdst, vsrc1, ref);                                   \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNullExportPC(obj, fp, pc))                             \
+            GOTO_exceptionThrown();                                         \
+        dvmSetField##_ftype(obj, ref, GET_REGISTER##_regsize(vdst));        \
+        ILOGV("+ IPUTQ %d=0x%08llx", ref,                                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_SGET_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, field@BBBB*/)                              \
+    {                                                                       \
+        StaticField* sfield;                                                \
+        vdst = INST_AA(inst);                                               \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|sget%s v%d,sfield@0x%04x", (_opname), vdst, ref);           \
+        sfield = (StaticField*)dvmDexGetResolvedField(methodClassDex, ref); \
+        if (sfield == NULL) {                                               \
+            EXPORT_PC();                                                    \
+            sfield = dvmResolveStaticField(curMethod->clazz, ref);          \
+            if (sfield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst, dvmGetStaticField##_ftype(sfield));    \
+        ILOGV("+ SGET '%s'=0x%08llx",                                       \
+            sfield->field.name, (u8)GET_REGISTER##_regsize(vdst));          \
+        UPDATE_FIELD_GET(&sfield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_SPUT_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, field@BBBB*/)                              \
+    {                                                                       \
+        StaticField* sfield;                                                \
+        vdst = INST_AA(inst);                                               \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|sput%s v%d,sfield@0x%04x", (_opname), vdst, ref);           \
+        sfield = (StaticField*)dvmDexGetResolvedField(methodClassDex, ref); \
+        if (sfield == NULL) {                                               \
+            EXPORT_PC();                                                    \
+            sfield = dvmResolveStaticField(curMethod->clazz, ref);          \
+            if (sfield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        dvmSetStaticField##_ftype(sfield, GET_REGISTER##_regsize(vdst));    \
+        ILOGV("+ SPUT '%s'=0x%08llx",                                       \
+            sfield->field.name, (u8)GET_REGISTER##_regsize(vdst));          \
+        UPDATE_FIELD_PUT(&sfield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+
+/* File: cstubs/enddefs.c */
+
+/* undefine "magic" name remapping */
+#undef retval
+#undef pc
+#undef fp
+#undef curMethod
+#undef methodClassDex
+#undef self
+#undef debugTrackedRefStart
+
+/* File: armv5te/debug.c */
+#include <inttypes.h>
+
+/*
+ * Dump the fixed-purpose ARM registers, along with some other info.
+ *
+ * This function MUST be compiled in ARM mode -- THUMB will yield bogus
+ * results.
+ *
+ * This will NOT preserve r0-r3/ip.
+ */
+void dvmMterpDumpArmRegs(uint32_t r0, uint32_t r1, uint32_t r2, uint32_t r3)
+{
+    register uint32_t rPC       asm("r4");
+    register uint32_t rFP       asm("r5");
+    register uint32_t rGLUE     asm("r6");
+    register uint32_t rIBASE    asm("r7");
+    register uint32_t rINST     asm("r8");
+    register uint32_t r9        asm("r9");
+    register uint32_t r10       asm("r10");
+
+    extern char dvmAsmInstructionStart[];
+
+    printf("REGS: r0=%08x r1=%08x r2=%08x r3=%08x\n", r0, r1, r2, r3);
+    printf("    : rPC=%08x rFP=%08x rGLUE=%08x rIBASE=%08x\n",
+        rPC, rFP, rGLUE, rIBASE);
+    printf("    : rINST=%08x r9=%08x r10=%08x\n", rINST, r9, r10);
+
+    MterpGlue* glue = (MterpGlue*) rGLUE;
+    const Method* method = glue->method;
+    printf("    + self is %p\n", dvmThreadSelf());
+    //printf("    + currently in %s.%s %s\n",
+    //    method->clazz->descriptor, method->name, method->signature);
+    //printf("    + dvmAsmInstructionStart = %p\n", dvmAsmInstructionStart);
+    //printf("    + next handler for 0x%02x = %p\n",
+    //    rINST & 0xff, dvmAsmInstructionStart + (rINST & 0xff) * 64);
+}
+
+/*
+ * Dump the StackSaveArea for the specified frame pointer.
+ */
+void dvmDumpFp(void* fp, StackSaveArea* otherSaveArea)
+{
+    StackSaveArea* saveArea = SAVEAREA_FROM_FP(fp);
+    printf("StackSaveArea for fp %p [%p/%p]:\n", fp, saveArea, otherSaveArea);
+#ifdef EASY_GDB
+    printf("  prevSave=%p, prevFrame=%p savedPc=%p meth=%p curPc=%p\n",
+        saveArea->prevSave, saveArea->prevFrame, saveArea->savedPc,
+        saveArea->method, saveArea->xtra.currentPc);
+#else
+    printf("  prevFrame=%p savedPc=%p meth=%p curPc=%p fp[0]=0x%08x\n",
+        saveArea->prevFrame, saveArea->savedPc,
+        saveArea->method, saveArea->xtra.currentPc,
+        *(u4*)fp);
+#endif
+}
+
+/*
+ * Does the bulk of the work for common_printMethod().
+ */
+void dvmMterpPrintMethod(Method* method)
+{
+    /*
+     * It is a direct (non-virtual) method if it is static, private,
+     * or a constructor.
+     */
+    bool isDirect = 
+        ((method->accessFlags & (ACC_STATIC|ACC_PRIVATE)) != 0) ||
+        (method->name[0] == '<');
+
+    char* desc = dexProtoCopyMethodDescriptor(&method->prototype);
+        
+    printf("<%c:%s.%s %s> ",
+            isDirect ? 'D' : 'V',
+            method->clazz->descriptor,
+            method->name,
+            desc);
+
+    free(desc);
+}
+
diff --git a/vm/mterp/out/InterpC-portdbg.c b/vm/mterp/out/InterpC-portdbg.c
new file mode 100644
index 0000000..d527cc0
--- /dev/null
+++ b/vm/mterp/out/InterpC-portdbg.c
@@ -0,0 +1,4264 @@
+/*
+ * This file was generated automatically by gen-mterp.py for 'portdbg'.
+ *
+ * --> DO NOT EDIT <--
+ */
+
+/* File: c/header.c */
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/* common includes */
+#include "Dalvik.h"
+#include "interp/InterpDefs.h"
+#include "mterp/Mterp.h"
+#include <math.h>                   // needed for fmod, fmodf
+
+/*
+ * Configuration defines.  These affect the C implementations, i.e. the
+ * portable interpreter(s) and C stubs.
+ *
+ * Some defines are controlled by the Makefile, e.g.:
+ *   WITH_PROFILER
+ *   WITH_DEBUGGER
+ *   WITH_INSTR_CHECKS
+ *   WITH_TRACKREF_CHECKS
+ *   EASY_GDB
+ *   NDEBUG
+ *
+ * If THREADED_INTERP is not defined, we use a classic "while true / switch"
+ * interpreter.  If it is defined, then the tail end of each instruction
+ * handler fetches the next instruction and jumps directly to the handler.
+ * This increases the size of the "Std" interpreter by about 10%, but
+ * provides a speedup of about the same magnitude.
+ *
+ * There's a "hybrid" approach that uses a goto table instead of a switch
+ * statement, avoiding the "is the opcode in range" tests required for switch.
+ * The performance is close to the threaded version, and without the 10%
+ * size increase, but the benchmark results are off enough that it's not
+ * worth adding as a third option.
+ */
+#define THREADED_INTERP             /* threaded vs. while-loop interpreter */
+
+#ifdef WITH_INSTR_CHECKS            /* instruction-level paranoia */
+# define CHECK_BRANCH_OFFSETS
+# define CHECK_REGISTER_INDICES
+#endif
+
+/*
+ * ARM EABI requires 64-bit alignment for access to 64-bit data types.  We
+ * can't just use pointers to copy 64-bit values out of our interpreted
+ * register set, because gcc will generate ldrd/strd.
+ *
+ * The __UNION version copies data in and out of a union.  The __MEMCPY
+ * version uses a memcpy() call to do the transfer; gcc is smart enough to
+ * not actually call memcpy().  The __UNION version is very bad on ARM;
+ * it only uses one more instruction than __MEMCPY, but for some reason
+ * gcc thinks it needs separate storage for every instance of the union.
+ * On top of that, it feels the need to zero them out at the start of the
+ * method.  Net result is we zero out ~700 bytes of stack space at the top
+ * of the interpreter using ARM STM instructions.
+ */
+#if defined(__ARM_EABI__)
+//# define NO_UNALIGN_64__UNION
+# define NO_UNALIGN_64__MEMCPY
+#endif
+
+//#define LOG_INSTR                   /* verbose debugging */
+/* set and adjust ANDROID_LOG_TAGS='*:i jdwp:i dalvikvm:i dalvikvmi:i' */
+
+/*
+ * Keep a tally of accesses to fields.  Currently only works if full DEX
+ * optimization is disabled.
+ */
+#ifdef PROFILE_FIELD_ACCESS
+# define UPDATE_FIELD_GET(_field) { (_field)->gets++; }
+# define UPDATE_FIELD_PUT(_field) { (_field)->puts++; }
+#else
+# define UPDATE_FIELD_GET(_field) ((void)0)
+# define UPDATE_FIELD_PUT(_field) ((void)0)
+#endif
+
+/*
+ * Adjust the program counter.  "_offset" is a signed int, in 16-bit units.
+ *
+ * Assumes the existence of "const u2* pc" and "const u2* curMethod->insns".
+ *
+ * We don't advance the program counter until we finish an instruction or
+ * branch, because we do want to have to unroll the PC if there's an
+ * exception.
+ */
+#ifdef CHECK_BRANCH_OFFSETS
+# define ADJUST_PC(_offset) do {                                            \
+        int myoff = _offset;        /* deref only once */                   \
+        if (pc + myoff < curMethod->insns ||                                \
+            pc + myoff >= curMethod->insns + dvmGetMethodInsnsSize(curMethod)) \
+        {                                                                   \
+            char* desc;                                                     \
+            desc = dexProtoCopyMethodDescriptor(&curMethod->prototype);     \
+            LOGE("Invalid branch %d at 0x%04x in %s.%s %s\n",               \
+                myoff, (int) (pc - curMethod->insns),                       \
+                curMethod->clazz->descriptor, curMethod->name, desc);       \
+            free(desc);                                                     \
+            dvmAbort();                                                     \
+        }                                                                   \
+        pc += myoff;                                                        \
+    } while (false)
+#else
+# define ADJUST_PC(_offset) (pc += _offset)
+#endif
+
+/*
+ * If enabled, log instructions as we execute them.
+ */
+#ifdef LOG_INSTR
+# define ILOGD(...) ILOG(LOG_DEBUG, __VA_ARGS__)
+# define ILOGV(...) ILOG(LOG_VERBOSE, __VA_ARGS__)
+# define ILOG(_level, ...) do {                                             \
+        char debugStrBuf[128];                                              \
+        snprintf(debugStrBuf, sizeof(debugStrBuf), __VA_ARGS__);            \
+        if (curMethod != NULL)                                                 \
+            LOG(_level, LOG_TAG"i", "%-2d|%04x%s\n",                        \
+                self->threadId, (int)(pc - curMethod->insns), debugStrBuf); \
+        else                                                                \
+            LOG(_level, LOG_TAG"i", "%-2d|####%s\n",                        \
+                self->threadId, debugStrBuf);                               \
+    } while(false)
+void dvmDumpRegs(const Method* method, const u4* framePtr, bool inOnly);
+# define DUMP_REGS(_meth, _frame, _inOnly) dvmDumpRegs(_meth, _frame, _inOnly)
+static const char kSpacing[] = "            ";
+#else
+# define ILOGD(...) ((void)0)
+# define ILOGV(...) ((void)0)
+# define DUMP_REGS(_meth, _frame, _inOnly) ((void)0)
+#endif
+
+/* get a long from an array of u4 */
+static inline s8 getLongFromArray(const u4* ptr, int idx)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { s8 ll; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.parts[0] = ptr[0];
+    conv.parts[1] = ptr[1];
+    return conv.ll;
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    s8 val;
+    memcpy(&val, &ptr[idx], 8);
+    return val;
+#else
+    return *((s8*) &ptr[idx]);
+#endif
+}
+
+/* store a long into an array of u4 */
+static inline void putLongToArray(u4* ptr, int idx, s8 val)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { s8 ll; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.ll = val;
+    ptr[0] = conv.parts[0];
+    ptr[1] = conv.parts[1];
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    memcpy(&ptr[idx], &val, 8);
+#else
+    *((s8*) &ptr[idx]) = val;
+#endif
+}
+
+/* get a double from an array of u4 */
+static inline double getDoubleFromArray(const u4* ptr, int idx)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { double d; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.parts[0] = ptr[0];
+    conv.parts[1] = ptr[1];
+    return conv.d;
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    double dval;
+    memcpy(&dval, &ptr[idx], 8);
+    return dval;
+#else
+    return *((double*) &ptr[idx]);
+#endif
+}
+
+/* store a double into an array of u4 */
+static inline void putDoubleToArray(u4* ptr, int idx, double dval)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { double d; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.d = dval;
+    ptr[0] = conv.parts[0];
+    ptr[1] = conv.parts[1];
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    memcpy(&ptr[idx], &dval, 8);
+#else
+    *((double*) &ptr[idx]) = dval;
+#endif
+}
+
+/*
+ * If enabled, validate the register number on every access.  Otherwise,
+ * just do an array access.
+ *
+ * Assumes the existence of "u4* fp".
+ *
+ * "_idx" may be referenced more than once.
+ */
+#ifdef CHECK_REGISTER_INDICES
+# define GET_REGISTER(_idx) \
+    ( (_idx) < curMethod->registersSize ? \
+        (fp[(_idx)]) : (assert(!"bad reg"),1969) )
+# define SET_REGISTER(_idx, _val) \
+    ( (_idx) < curMethod->registersSize ? \
+        (fp[(_idx)] = (u4)(_val)) : (assert(!"bad reg"),1969) )
+# define GET_REGISTER_AS_OBJECT(_idx)       ((Object *)GET_REGISTER(_idx))
+# define SET_REGISTER_AS_OBJECT(_idx, _val) SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_INT(_idx) ((s4) GET_REGISTER(_idx))
+# define SET_REGISTER_INT(_idx, _val) SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_WIDE(_idx) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        getLongFromArray(fp, (_idx)) : (assert(!"bad reg"),1969) )
+# define SET_REGISTER_WIDE(_idx, _val) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        putLongToArray(fp, (_idx), (_val)) : (assert(!"bad reg"),1969) )
+# define GET_REGISTER_FLOAT(_idx) \
+    ( (_idx) < curMethod->registersSize ? \
+        (*((float*) &fp[(_idx)])) : (assert(!"bad reg"),1969.0f) )
+# define SET_REGISTER_FLOAT(_idx, _val) \
+    ( (_idx) < curMethod->registersSize ? \
+        (*((float*) &fp[(_idx)]) = (_val)) : (assert(!"bad reg"),1969.0f) )
+# define GET_REGISTER_DOUBLE(_idx) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        getDoubleFromArray(fp, (_idx)) : (assert(!"bad reg"),1969.0) )
+# define SET_REGISTER_DOUBLE(_idx, _val) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        putDoubleToArray(fp, (_idx), (_val)) : (assert(!"bad reg"),1969.0) )
+#else
+# define GET_REGISTER(_idx)                 (fp[(_idx)])
+# define SET_REGISTER(_idx, _val)           (fp[(_idx)] = (_val))
+# define GET_REGISTER_AS_OBJECT(_idx)       ((Object*) fp[(_idx)])
+# define SET_REGISTER_AS_OBJECT(_idx, _val) (fp[(_idx)] = (u4)(_val))
+# define GET_REGISTER_INT(_idx)             ((s4)GET_REGISTER(_idx))
+# define SET_REGISTER_INT(_idx, _val)       SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_WIDE(_idx)            getLongFromArray(fp, (_idx))
+# define SET_REGISTER_WIDE(_idx, _val)      putLongToArray(fp, (_idx), (_val))
+# define GET_REGISTER_FLOAT(_idx)           (*((float*) &fp[(_idx)]))
+# define SET_REGISTER_FLOAT(_idx, _val)     (*((float*) &fp[(_idx)]) = (_val))
+# define GET_REGISTER_DOUBLE(_idx)          getDoubleFromArray(fp, (_idx))
+# define SET_REGISTER_DOUBLE(_idx, _val)    putDoubleToArray(fp, (_idx), (_val))
+#endif
+
+/*
+ * Get 16 bits from the specified offset of the program counter.  We always
+ * want to load 16 bits at a time from the instruction stream -- it's more
+ * efficient than 8 and won't have the alignment problems that 32 might.
+ *
+ * Assumes existence of "const u2* pc".
+ */
+#define FETCH(_offset)     (pc[(_offset)])
+
+/*
+ * Extract instruction byte from 16-bit fetch (_inst is a u2).
+ */
+#define INST_INST(_inst)    ((_inst) & 0xff)
+
+/*
+ * Extract the "vA, vB" 4-bit registers from the instruction word (_inst is u2).
+ */
+#define INST_A(_inst)       (((_inst) >> 8) & 0x0f)
+#define INST_B(_inst)       ((_inst) >> 12)
+
+/*
+ * Get the 8-bit "vAA" 8-bit register index from the instruction word.
+ * (_inst is u2)
+ */
+#define INST_AA(_inst)      ((_inst) >> 8)
+
+/*
+ * The current PC must be available to Throwable constructors, e.g.
+ * those created by dvmThrowException(), so that the exception stack
+ * trace can be generated correctly.  If we don't do this, the offset
+ * within the current method won't be shown correctly.  See the notes
+ * in Exception.c.
+ *
+ * Assumes existence of "u4* fp" and "const u2* pc".
+ */
+#define EXPORT_PC()         (SAVEAREA_FROM_FP(fp)->xtra.currentPc = pc)
+
+/*
+ * Determine if we need to switch to a different interpreter.  "_current"
+ * is either INTERP_STD or INTERP_DBG.  It should be fixed for a given
+ * interpreter generation file, which should remove the outer conditional
+ * from the following.
+ *
+ * If we're building without debug and profiling support, we never switch.
+ */
+#if defined(WITH_PROFILER) || defined(WITH_DEBUGGER)
+# define NEED_INTERP_SWITCH(_current) (                                     \
+    (_current == INTERP_STD) ?                                              \
+        dvmDebuggerOrProfilerActive() : !dvmDebuggerOrProfilerActive() )
+#else
+# define NEED_INTERP_SWITCH(_current) (false)
+#endif
+
+/*
+ * Look up an interface on a class using the cache.
+ */
+INLINE Method* dvmFindInterfaceMethodInCache(ClassObject* thisClass,
+    u4 methodIdx, const Method* method, DvmDex* methodClassDex)
+{
+#define ATOMIC_CACHE_CALC \
+    dvmInterpFindInterfaceMethod(thisClass, methodIdx, method, methodClassDex)
+
+    return (Method*) ATOMIC_CACHE_LOOKUP(methodClassDex->pInterfaceCache,
+                DEX_INTERFACE_CACHE_SIZE, thisClass, methodIdx);
+
+#undef ATOMIC_CACHE_CALC
+}
+
+/*
+ * Check to see if "obj" is NULL.  If so, throw an exception.  Assumes the
+ * pc has already been exported to the stack.
+ *
+ * Perform additional checks on debug builds.
+ *
+ * Use this to check for NULL when the instruction handler calls into
+ * something that could throw an exception (so we have already called
+ * EXPORT_PC at the top).
+ */
+static inline bool checkForNull(Object* obj)
+{
+    if (obj == NULL) {
+        dvmThrowException("Ljava/lang/NullPointerException;", NULL);
+        return false;
+    }
+#ifdef WITH_EXTRA_OBJECT_VALIDATION
+    if (!dvmIsValidObject(obj)) {
+        LOGE("Invalid object %p\n", obj);
+        dvmAbort();
+    }
+#endif
+#ifndef NDEBUG
+    if (obj->clazz == NULL || ((u4) obj->clazz) <= 65536) {
+        /* probable heap corruption */
+        LOGE("Invalid object class %p (in %p)\n", obj->clazz, obj);
+        dvmAbort();
+    }
+#endif
+    return true;
+}
+
+/*
+ * Check to see if "obj" is NULL.  If so, export the PC into the stack
+ * frame and throw an exception.
+ *
+ * Perform additional checks on debug builds.
+ *
+ * Use this to check for NULL when the instruction handler doesn't do
+ * anything else that can throw an exception.
+ */
+static inline bool checkForNullExportPC(Object* obj, u4* fp, const u2* pc)
+{
+    if (obj == NULL) {
+        EXPORT_PC();
+        dvmThrowException("Ljava/lang/NullPointerException;", NULL);
+        return false;
+    }
+#ifdef WITH_EXTRA_OBJECT_VALIDATION
+    if (!dvmIsValidObject(obj)) {
+        LOGE("Invalid object %p\n", obj);
+        dvmAbort();
+    }
+#endif
+#ifndef NDEBUG
+    if (obj->clazz == NULL || ((u4) obj->clazz) <= 65536) {
+        /* probable heap corruption */
+        LOGE("Invalid object class %p (in %p)\n", obj->clazz, obj);
+        dvmAbort();
+    }
+#endif
+    return true;
+}
+
+
+/* File: portable/portdbg.c */
+#define INTERP_FUNC_NAME dvmInterpretDbg
+#define INTERP_TYPE INTERP_DBG
+
+#define CHECK_DEBUG_AND_PROF() \
+    checkDebugAndProf(pc, fp, self, curMethod, &debugIsMethodEntry)
+
+/* File: portable/stubdefs.c */
+/*
+ * In the C mterp stubs, "goto" is a function call followed immediately
+ * by a return.
+ */
+
+#define GOTO_TARGET_DECL(_target, ...)
+
+#define GOTO_TARGET(_target, ...) _target:
+
+#define GOTO_TARGET_END
+
+/* ugh */
+#define STUB_HACK(x)
+
+/*
+ * Instruction framing.  For a switch-oriented implementation this is
+ * case/break, for a threaded implementation it's a goto label and an
+ * instruction fetch/computed goto.
+ *
+ * Assumes the existence of "const u2* pc" and (for threaded operation)
+ * "u2 inst".
+ */
+#ifdef THREADED_INTERP
+# define H(_op)             &&op_##_op
+# define HANDLE_OPCODE(_op) op_##_op:
+# define FINISH(_offset) {                                                  \
+        ADJUST_PC(_offset);                                                 \
+        inst = FETCH(0);                                                    \
+        CHECK_DEBUG_AND_PROF();                                             \
+        CHECK_TRACKED_REFS();                                               \
+        goto *handlerTable[INST_INST(inst)];                                \
+    }
+#else
+# define HANDLE_OPCODE(_op) case _op:
+# define FINISH(_offset)    { ADJUST_PC(_offset); break; }
+#endif
+
+#define OP_END
+
+#if defined(WITH_TRACKREF_CHECKS)
+# define CHECK_TRACKED_REFS() \
+    dvmInterpCheckTrackedRefs(self, curMethod, debugTrackedRefStart)
+#else
+# define CHECK_TRACKED_REFS() ((void)0)
+#endif
+
+
+/*
+ * The "goto" targets just turn into goto statements.  The "arguments" are
+ * passed through local variables.
+ */
+
+#define GOTO_exceptionThrown() goto exceptionThrown;
+
+#define GOTO_returnFromMethod() goto returnFromMethod;
+
+#define GOTO_invoke(_target, _methodCallRange)                              \
+    do {                                                                    \
+        methodCallRange = _methodCallRange;                                 \
+        goto _target;                                                       \
+    } while(false)
+
+/* for this, the "args" are already in the locals */
+#define GOTO_invokeMethod(_methodCallRange, _methodToCall, _vsrc1, _vdst) goto invokeMethod;
+
+#define GOTO_bail() goto bail;
+#define GOTO_bail_switch() goto bail_switch;
+
+/*
+ * Periodically check for thread suspension.
+ *
+ * While we're at it, see if a debugger has attached or the profiler has
+ * started.  If so, switch to a different "goto" table.
+ */
+#define PERIODIC_CHECKS(_entryPoint, _pcadj) {                              \
+        dvmCheckSuspendQuick(self);                                         \
+        if (NEED_INTERP_SWITCH(INTERP_TYPE)) {                              \
+            ADJUST_PC(_pcadj);                                              \
+            interpState->entryPoint = _entryPoint;                          \
+            LOGVV("threadid=%d: switch to %s ep=%d adj=%d\n",               \
+                self->threadId,                                             \
+                (interpState->nextMode == INTERP_STD) ? "STD" : "DBG",      \
+                (_entryPoint), (_pcadj));                                   \
+            GOTO_bail_switch();                                             \
+        }                                                                   \
+    }
+
+
+/* File: c/opcommon.c */
+/* forward declarations of goto targets */
+GOTO_TARGET_DECL(filledNewArray, bool methodCallRange);
+GOTO_TARGET_DECL(invokeVirtual, bool methodCallRange);
+GOTO_TARGET_DECL(invokeSuper, bool methodCallRange);
+GOTO_TARGET_DECL(invokeInterface, bool methodCallRange);
+GOTO_TARGET_DECL(invokeDirect, bool methodCallRange);
+GOTO_TARGET_DECL(invokeStatic, bool methodCallRange);
+GOTO_TARGET_DECL(invokeVirtualQuick, bool methodCallRange);
+GOTO_TARGET_DECL(invokeSuperQuick, bool methodCallRange);
+GOTO_TARGET_DECL(invokeMethod, bool methodCallRange, const Method* methodToCall,
+    u2 count, u2 regs);
+GOTO_TARGET_DECL(returnFromMethod);
+GOTO_TARGET_DECL(exceptionThrown);
+
+/*
+ * ===========================================================================
+ *
+ * What follows are opcode definitions shared between multiple opcodes with
+ * minor substitutions handled by the C pre-processor.  These should probably
+ * use the mterp substitution mechanism instead, with the code here moved
+ * into common fragment files (like the asm "binop.S"), although it's hard
+ * to give up the C preprocessor in favor of the much simpler text subst.
+ *
+ * ===========================================================================
+ */
+
+#define HANDLE_NUMCONV(_opcode, _opname, _fromtype, _totype)                \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        SET_REGISTER##_totype(vdst,                                         \
+            GET_REGISTER##_fromtype(vsrc1));                                \
+        FINISH(1);
+
+#define HANDLE_FLOAT_TO_INT(_opcode, _opname, _fromvtype, _fromrtype,       \
+        _tovtype, _tortype)                                                 \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+    {                                                                       \
+        /* spec defines specific handling for +/- inf and NaN values */     \
+        _fromvtype val;                                                     \
+        _tovtype intMin, intMax, result;                                    \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        val = GET_REGISTER##_fromrtype(vsrc1);                              \
+        intMin = (_tovtype) 1 << (sizeof(_tovtype) * 8 -1);                 \
+        intMax = ~intMin;                                                   \
+        result = (_tovtype) val;                                            \
+        if (val >= intMax)          /* +inf */                              \
+            result = intMax;                                                \
+        else if (val <= intMin)     /* -inf */                              \
+            result = intMin;                                                \
+        else if (val != val)        /* NaN */                               \
+            result = 0;                                                     \
+        else                                                                \
+            result = (_tovtype) val;                                        \
+        SET_REGISTER##_tortype(vdst, result);                               \
+    }                                                                       \
+    FINISH(1);
+
+#define HANDLE_INT_TO_SMALL(_opcode, _opname, _type)                        \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|int-to-%s v%d,v%d", (_opname), vdst, vsrc1);                \
+        SET_REGISTER(vdst, (_type) GET_REGISTER(vsrc1));                    \
+        FINISH(1);
+
+/* NOTE: the comparison result is always a signed 4-byte integer */
+#define HANDLE_OP_CMPX(_opcode, _opname, _varType, _type, _nanVal)          \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        int result;                                                         \
+        u2 regs;                                                            \
+        _varType val1, val2;                                                \
+        vdst = INST_AA(inst);                                               \
+        regs = FETCH(1);                                                    \
+        vsrc1 = regs & 0xff;                                                \
+        vsrc2 = regs >> 8;                                                  \
+        ILOGV("|cmp%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);         \
+        val1 = GET_REGISTER##_type(vsrc1);                                  \
+        val2 = GET_REGISTER##_type(vsrc2);                                  \
+        if (val1 == val2)                                                   \
+            result = 0;                                                     \
+        else if (val1 < val2)                                               \
+            result = -1;                                                    \
+        else if (val1 > val2)                                               \
+            result = 1;                                                     \
+        else                                                                \
+            result = (_nanVal);                                             \
+        ILOGV("+ result=%d\n", result);                                     \
+        SET_REGISTER(vdst, result);                                         \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_IF_XX(_opcode, _opname, _cmp)                             \
+    HANDLE_OPCODE(_opcode /*vA, vB, +CCCC*/)                                \
+        vsrc1 = INST_A(inst);                                               \
+        vsrc2 = INST_B(inst);                                               \
+        if ((s4) GET_REGISTER(vsrc1) _cmp (s4) GET_REGISTER(vsrc2)) {       \
+            int branchOffset = (s2)FETCH(1);    /* sign-extended */         \
+            ILOGV("|if-%s v%d,v%d,+0x%04x", (_opname), vsrc1, vsrc2,        \
+                branchOffset);                                              \
+            ILOGV("> branch taken");                                        \
+            if (branchOffset < 0)                                           \
+                PERIODIC_CHECKS(kInterpEntryInstr, branchOffset);           \
+            FINISH(branchOffset);                                           \
+        } else {                                                            \
+            ILOGV("|if-%s v%d,v%d,-", (_opname), vsrc1, vsrc2);             \
+            FINISH(2);                                                      \
+        }
+
+#define HANDLE_OP_IF_XXZ(_opcode, _opname, _cmp)                            \
+    HANDLE_OPCODE(_opcode /*vAA, +BBBB*/)                                   \
+        vsrc1 = INST_AA(inst);                                              \
+        if ((s4) GET_REGISTER(vsrc1) _cmp 0) {                              \
+            int branchOffset = (s2)FETCH(1);    /* sign-extended */         \
+            ILOGV("|if-%s v%d,+0x%04x", (_opname), vsrc1, branchOffset);    \
+            ILOGV("> branch taken");                                        \
+            if (branchOffset < 0)                                           \
+                PERIODIC_CHECKS(kInterpEntryInstr, branchOffset);           \
+            FINISH(branchOffset);                                           \
+        } else {                                                            \
+            ILOGV("|if-%s v%d,-", (_opname), vsrc1);                        \
+            FINISH(2);                                                      \
+        }
+
+#define HANDLE_UNOP(_opcode, _opname, _pfx, _sfx, _type)                    \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        SET_REGISTER##_type(vdst, _pfx GET_REGISTER##_type(vsrc1) _sfx);    \
+        FINISH(1);
+
+#define HANDLE_OP_X_INT(_opcode, _opname, _op, _chkdiv)                     \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-int v%d,v%d", (_opname), vdst, vsrc1);                   \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            secondVal = GET_REGISTER(vsrc2);                                \
+            if (secondVal == 0) {                                           \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && secondVal == -1) {            \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            /* non-div/rem case */                                          \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vsrc1) _op (s4) GET_REGISTER(vsrc2));     \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_INT(_opcode, _opname, _cast, _op)                     \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-int v%d,v%d", (_opname), vdst, vsrc1);                   \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vsrc1) _op (GET_REGISTER(vsrc2) & 0x1f));    \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_INT_LIT16(_opcode, _opname, _op, _chkdiv)               \
+    HANDLE_OPCODE(_opcode /*vA, vB, #+CCCC*/)                               \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        vsrc2 = FETCH(1);                                                   \
+        ILOGV("|%s-int/lit16 v%d,v%d,#+0x%04x",                             \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, result;                                            \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            if ((s2) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                      \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && ((s2) vsrc2) == -1) {         \
+                /* won't generate /lit16 instr for this; check anyway */    \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op (s2) vsrc2;                           \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            /* non-div/rem case */                                          \
+            SET_REGISTER(vdst, GET_REGISTER(vsrc1) _op (s2) vsrc2);         \
+        }                                                                   \
+        FINISH(2);
+
+#define HANDLE_OP_X_INT_LIT8(_opcode, _opname, _op, _chkdiv)                \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, #+CC*/)                               \
+    {                                                                       \
+        u2 litInfo;                                                         \
+        vdst = INST_AA(inst);                                               \
+        litInfo = FETCH(1);                                                 \
+        vsrc1 = litInfo & 0xff;                                             \
+        vsrc2 = litInfo >> 8;       /* constant */                          \
+        ILOGV("|%s-int/lit8 v%d,v%d,#+0x%02x",                              \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, result;                                            \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            if ((s1) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && ((s1) vsrc2) == -1) {         \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op ((s1) vsrc2);                         \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vsrc1) _op (s1) vsrc2);                   \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_INT_LIT8(_opcode, _opname, _cast, _op)                \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, #+CC*/)                               \
+    {                                                                       \
+        u2 litInfo;                                                         \
+        vdst = INST_AA(inst);                                               \
+        litInfo = FETCH(1);                                                 \
+        vsrc1 = litInfo & 0xff;                                             \
+        vsrc2 = litInfo >> 8;       /* constant */                          \
+        ILOGV("|%s-int/lit8 v%d,v%d,#+0x%02x",                              \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vsrc1) _op (vsrc2 & 0x1f));                  \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_INT_2ADDR(_opcode, _opname, _op, _chkdiv)               \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-int-2addr v%d,v%d", (_opname), vdst, vsrc1);             \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER(vdst);                                  \
+            secondVal = GET_REGISTER(vsrc1);                                \
+            if (secondVal == 0) {                                           \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && secondVal == -1) {            \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vdst) _op (s4) GET_REGISTER(vsrc1));      \
+        }                                                                   \
+        FINISH(1);
+
+#define HANDLE_OP_SHX_INT_2ADDR(_opcode, _opname, _cast, _op)               \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-int-2addr v%d,v%d", (_opname), vdst, vsrc1);             \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vdst) _op (GET_REGISTER(vsrc1) & 0x1f));     \
+        FINISH(1);
+
+#define HANDLE_OP_X_LONG(_opcode, _opname, _op, _chkdiv)                    \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-long v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);       \
+        if (_chkdiv != 0) {                                                 \
+            s8 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER_WIDE(vsrc1);                            \
+            secondVal = GET_REGISTER_WIDE(vsrc2);                           \
+            if (secondVal == 0LL) {                                         \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u8)firstVal == 0x8000000000000000ULL &&                    \
+                secondVal == -1LL)                                          \
+            {                                                               \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER_WIDE(vdst, result);                                \
+        } else {                                                            \
+            SET_REGISTER_WIDE(vdst,                                         \
+                (s8) GET_REGISTER_WIDE(vsrc1) _op (s8) GET_REGISTER_WIDE(vsrc2)); \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_LONG(_opcode, _opname, _cast, _op)                    \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-long v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);       \
+        SET_REGISTER_WIDE(vdst,                                             \
+            _cast GET_REGISTER_WIDE(vsrc1) _op (GET_REGISTER(vsrc2) & 0x3f)); \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_LONG_2ADDR(_opcode, _opname, _op, _chkdiv)              \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-long-2addr v%d,v%d", (_opname), vdst, vsrc1);            \
+        if (_chkdiv != 0) {                                                 \
+            s8 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER_WIDE(vdst);                             \
+            secondVal = GET_REGISTER_WIDE(vsrc1);                           \
+            if (secondVal == 0LL) {                                         \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u8)firstVal == 0x8000000000000000ULL &&                    \
+                secondVal == -1LL)                                          \
+            {                                                               \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER_WIDE(vdst, result);                                \
+        } else {                                                            \
+            SET_REGISTER_WIDE(vdst,                                         \
+                (s8) GET_REGISTER_WIDE(vdst) _op (s8)GET_REGISTER_WIDE(vsrc1));\
+        }                                                                   \
+        FINISH(1);
+
+#define HANDLE_OP_SHX_LONG_2ADDR(_opcode, _opname, _cast, _op)              \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-long-2addr v%d,v%d", (_opname), vdst, vsrc1);            \
+        SET_REGISTER_WIDE(vdst,                                             \
+            _cast GET_REGISTER_WIDE(vdst) _op (GET_REGISTER(vsrc1) & 0x3f)); \
+        FINISH(1);
+
+#define HANDLE_OP_X_FLOAT(_opcode, _opname, _op)                            \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-float v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);      \
+        SET_REGISTER_FLOAT(vdst,                                            \
+            GET_REGISTER_FLOAT(vsrc1) _op GET_REGISTER_FLOAT(vsrc2));       \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_DOUBLE(_opcode, _opname, _op)                           \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-double v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);     \
+        SET_REGISTER_DOUBLE(vdst,                                           \
+            GET_REGISTER_DOUBLE(vsrc1) _op GET_REGISTER_DOUBLE(vsrc2));     \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_FLOAT_2ADDR(_opcode, _opname, _op)                      \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-float-2addr v%d,v%d", (_opname), vdst, vsrc1);           \
+        SET_REGISTER_FLOAT(vdst,                                            \
+            GET_REGISTER_FLOAT(vdst) _op GET_REGISTER_FLOAT(vsrc1));        \
+        FINISH(1);
+
+#define HANDLE_OP_X_DOUBLE_2ADDR(_opcode, _opname, _op)                     \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-double-2addr v%d,v%d", (_opname), vdst, vsrc1);          \
+        SET_REGISTER_DOUBLE(vdst,                                           \
+            GET_REGISTER_DOUBLE(vdst) _op GET_REGISTER_DOUBLE(vsrc1));      \
+        FINISH(1);
+
+#define HANDLE_OP_AGET(_opcode, _opname, _type, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        ArrayObject* arrayObj;                                              \
+        u2 arrayInfo;                                                       \
+        EXPORT_PC();                                                        \
+        vdst = INST_AA(inst);                                               \
+        arrayInfo = FETCH(1);                                               \
+        vsrc1 = arrayInfo & 0xff;    /* array ptr */                        \
+        vsrc2 = arrayInfo >> 8;      /* index */                            \
+        ILOGV("|aget%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);        \
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);                      \
+        if (!checkForNull((Object*) arrayObj))                              \
+            GOTO_exceptionThrown();                                         \
+        if (GET_REGISTER(vsrc2) >= arrayObj->length) {                      \
+            LOGV("Invalid array access: %p %d (len=%d)\n",                  \
+                arrayObj, vsrc2, arrayObj->length);                         \
+            dvmThrowException("Ljava/lang/ArrayIndexOutOfBoundsException;", \
+                NULL);                                                      \
+            GOTO_exceptionThrown();                                         \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst,                                        \
+            ((_type*) arrayObj->contents)[GET_REGISTER(vsrc2)]);            \
+        ILOGV("+ AGET[%d]=0x%x", GET_REGISTER(vsrc2), GET_REGISTER(vdst));  \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_APUT(_opcode, _opname, _type, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        ArrayObject* arrayObj;                                              \
+        u2 arrayInfo;                                                       \
+        EXPORT_PC();                                                        \
+        vdst = INST_AA(inst);       /* AA: source value */                  \
+        arrayInfo = FETCH(1);                                               \
+        vsrc1 = arrayInfo & 0xff;   /* BB: array ptr */                     \
+        vsrc2 = arrayInfo >> 8;     /* CC: index */                         \
+        ILOGV("|aput%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);        \
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);                      \
+        if (!checkForNull((Object*) arrayObj))                              \
+            GOTO_exceptionThrown();                                         \
+        if (GET_REGISTER(vsrc2) >= arrayObj->length) {                      \
+            dvmThrowException("Ljava/lang/ArrayIndexOutOfBoundsException;", \
+                NULL);                                                      \
+            GOTO_exceptionThrown();                                         \
+        }                                                                   \
+        ILOGV("+ APUT[%d]=0x%08x", GET_REGISTER(vsrc2), GET_REGISTER(vdst));\
+        ((_type*) arrayObj->contents)[GET_REGISTER(vsrc2)] =                \
+            GET_REGISTER##_regsize(vdst);                                   \
+    }                                                                       \
+    FINISH(2);
+
+/*
+ * It's possible to get a bad value out of a field with sub-32-bit stores
+ * because the -quick versions always operate on 32 bits.  Consider:
+ *   short foo = -1  (sets a 32-bit register to 0xffffffff)
+ *   iput-quick foo  (writes all 32 bits to the field)
+ *   short bar = 1   (sets a 32-bit register to 0x00000001)
+ *   iput-short      (writes the low 16 bits to the field)
+ *   iget-quick foo  (reads all 32 bits from the field, yielding 0xffff0001)
+ * This can only happen when optimized and non-optimized code has interleaved
+ * access to the same field.  This is unlikely but possible.
+ *
+ * The easiest way to fix this is to always read/write 32 bits at a time.  On
+ * a device with a 16-bit data bus this is sub-optimal.  (The alternative
+ * approach is to have sub-int versions of iget-quick, but now we're wasting
+ * Dalvik instruction space and making it less likely that handler code will
+ * already be in the CPU i-cache.)
+ */
+#define HANDLE_IGET_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        InstField* ifield;                                                  \
+        Object* obj;                                                        \
+        EXPORT_PC();                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|iget%s v%d,v%d,field@0x%04x", (_opname), vdst, vsrc1, ref); \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNull(obj))                                             \
+            GOTO_exceptionThrown();                                         \
+        ifield = (InstField*) dvmDexGetResolvedField(methodClassDex, ref);  \
+        if (ifield == NULL) {                                               \
+            ifield = dvmResolveInstField(curMethod->clazz, ref);            \
+            if (ifield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst,                                        \
+            dvmGetField##_ftype(obj, ifield->byteOffset));                  \
+        ILOGV("+ IGET '%s'=0x%08llx", ifield->field.name,                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+        UPDATE_FIELD_GET(&ifield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IGET_X_QUICK(_opcode, _opname, _ftype, _regsize)             \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        Object* obj;                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field offset */                          \
+        ILOGV("|iget%s-quick v%d,v%d,field@+%u",                            \
+            (_opname), vdst, vsrc1, ref);                                   \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNullExportPC(obj, fp, pc))                             \
+            GOTO_exceptionThrown();                                         \
+        SET_REGISTER##_regsize(vdst, dvmGetField##_ftype(obj, ref));        \
+        ILOGV("+ IGETQ %d=0x%08llx", ref,                                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IPUT_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        InstField* ifield;                                                  \
+        Object* obj;                                                        \
+        EXPORT_PC();                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|iput%s v%d,v%d,field@0x%04x", (_opname), vdst, vsrc1, ref); \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNull(obj))                                             \
+            GOTO_exceptionThrown();                                         \
+        ifield = (InstField*) dvmDexGetResolvedField(methodClassDex, ref);  \
+        if (ifield == NULL) {                                               \
+            ifield = dvmResolveInstField(curMethod->clazz, ref);            \
+            if (ifield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        dvmSetField##_ftype(obj, ifield->byteOffset,                        \
+            GET_REGISTER##_regsize(vdst));                                  \
+        ILOGV("+ IPUT '%s'=0x%08llx", ifield->field.name,                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+        UPDATE_FIELD_PUT(&ifield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IPUT_X_QUICK(_opcode, _opname, _ftype, _regsize)             \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        Object* obj;                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field offset */                          \
+        ILOGV("|iput%s-quick v%d,v%d,field@0x%04x",                         \
+            (_opname), vdst, vsrc1, ref);                                   \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNullExportPC(obj, fp, pc))                             \
+            GOTO_exceptionThrown();                                         \
+        dvmSetField##_ftype(obj, ref, GET_REGISTER##_regsize(vdst));        \
+        ILOGV("+ IPUTQ %d=0x%08llx", ref,                                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_SGET_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, field@BBBB*/)                              \
+    {                                                                       \
+        StaticField* sfield;                                                \
+        vdst = INST_AA(inst);                                               \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|sget%s v%d,sfield@0x%04x", (_opname), vdst, ref);           \
+        sfield = (StaticField*)dvmDexGetResolvedField(methodClassDex, ref); \
+        if (sfield == NULL) {                                               \
+            EXPORT_PC();                                                    \
+            sfield = dvmResolveStaticField(curMethod->clazz, ref);          \
+            if (sfield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst, dvmGetStaticField##_ftype(sfield));    \
+        ILOGV("+ SGET '%s'=0x%08llx",                                       \
+            sfield->field.name, (u8)GET_REGISTER##_regsize(vdst));          \
+        UPDATE_FIELD_GET(&sfield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_SPUT_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, field@BBBB*/)                              \
+    {                                                                       \
+        StaticField* sfield;                                                \
+        vdst = INST_AA(inst);                                               \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|sput%s v%d,sfield@0x%04x", (_opname), vdst, ref);           \
+        sfield = (StaticField*)dvmDexGetResolvedField(methodClassDex, ref); \
+        if (sfield == NULL) {                                               \
+            EXPORT_PC();                                                    \
+            sfield = dvmResolveStaticField(curMethod->clazz, ref);          \
+            if (sfield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        dvmSetStaticField##_ftype(sfield, GET_REGISTER##_regsize(vdst));    \
+        ILOGV("+ SPUT '%s'=0x%08llx",                                       \
+            sfield->field.name, (u8)GET_REGISTER##_regsize(vdst));          \
+        UPDATE_FIELD_PUT(&sfield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+
+/* File: portable/debug.c */
+/* code in here is only included in portable-debug interpreter */
+
+/*
+ * Determine if an address is "interesting" to the debugger.  This allows
+ * us to avoid scanning the entire event list before every instruction.
+ *
+ * The "debugBreakAddr" table is global and not synchronized.
+ */
+static bool isInterestingAddr(const u2* pc)
+{
+    const u2** ptr = gDvm.debugBreakAddr;
+    int i;
+
+    for (i = 0; i < MAX_BREAKPOINTS; i++, ptr++) {
+        if (*ptr == pc) {
+            LOGV("BKP: hit on %p\n", pc);
+            return true;
+        }
+    }
+    return false;
+}
+
+/*
+ * Update the debugger on interesting events, such as hitting a breakpoint
+ * or a single-step point.  This is called from the top of the interpreter
+ * loop, before the current instruction is processed.
+ *
+ * Set "methodEntry" if we've just entered the method.  This detects
+ * method exit by checking to see if the next instruction is "return".
+ *
+ * This can't catch native method entry/exit, so we have to handle that
+ * at the point of invocation.  We also need to catch it in dvmCallMethod
+ * if we want to capture native->native calls made through JNI.
+ *
+ * Notes to self:
+ * - Don't want to switch to VMWAIT while posting events to the debugger.
+ *   Let the debugger code decide if we need to change state.
+ * - We may want to check for debugger-induced thread suspensions on
+ *   every instruction.  That would make a "suspend all" more responsive
+ *   and reduce the chances of multiple simultaneous events occurring.
+ *   However, it could change the behavior some.
+ *
+ * TODO: method entry/exit events are probably less common than location
+ * breakpoints.  We may be able to speed things up a bit if we don't query
+ * the event list unless we know there's at least one lurking within.
+ */
+static void updateDebugger(const Method* method, const u2* pc, const u4* fp,
+    bool methodEntry, Thread* self)
+{
+    int eventFlags = 0;
+
+    /*
+     * Update xtra.currentPc on every instruction.  We need to do this if
+     * there's a chance that we could get suspended.  This can happen if
+     * eventFlags != 0 here, or somebody manually requests a suspend
+     * (which gets handled at PERIOD_CHECKS time).  One place where this
+     * needs to be correct is in dvmAddSingleStep().
+     */
+    EXPORT_PC();
+
+    if (methodEntry)
+        eventFlags |= DBG_METHOD_ENTRY;
+
+    /*
+     * See if we have a breakpoint here.
+     *
+     * Depending on the "mods" associated with event(s) on this address,
+     * we may or may not actually send a message to the debugger.
+     *
+     * Checking method->debugBreakpointCount is slower on the device than
+     * just scanning the table (!).  We could probably work something out
+     * where we just check it on method entry/exit and remember the result,
+     * but that's more fragile and requires passing more stuff around.
+     */
+#ifdef WITH_DEBUGGER
+    if (method->debugBreakpointCount > 0 && isInterestingAddr(pc)) {
+        eventFlags |= DBG_BREAKPOINT;
+    }
+#endif
+
+    /*
+     * If the debugger is single-stepping one of our threads, check to
+     * see if we're that thread and we've reached a step point.
+     */
+    const StepControl* pCtrl = &gDvm.stepControl;
+    if (pCtrl->active && pCtrl->thread == self) {
+        int line, frameDepth;
+        bool doStop = false;
+        const char* msg = NULL;
+
+        assert(!dvmIsNativeMethod(method));
+
+        if (pCtrl->depth == SD_INTO) {
+            /*
+             * Step into method calls.  We break when the line number
+             * or method pointer changes.  If we're in SS_MIN mode, we
+             * always stop.
+             */
+            if (pCtrl->method != method) {
+                doStop = true;
+                msg = "new method";
+            } else if (pCtrl->size == SS_MIN) {
+                doStop = true;
+                msg = "new instruction";
+            } else if (!dvmAddressSetGet(
+                    pCtrl->pAddressSet, pc - method->insns)) {
+                doStop = true;
+                msg = "new line";
+            }
+        } else if (pCtrl->depth == SD_OVER) {
+            /*
+             * Step over method calls.  We break when the line number is
+             * different and the frame depth is <= the original frame
+             * depth.  (We can't just compare on the method, because we
+             * might get unrolled past it by an exception, and it's tricky
+             * to identify recursion.)
+             */
+            frameDepth = dvmComputeVagueFrameDepth(self, fp);
+            if (frameDepth < pCtrl->frameDepth) {
+                /* popped up one or more frames, always trigger */
+                doStop = true;
+                msg = "method pop";
+            } else if (frameDepth == pCtrl->frameDepth) {
+                /* same depth, see if we moved */
+                if (pCtrl->size == SS_MIN) {
+                    doStop = true;
+                    msg = "new instruction";
+                } else if (!dvmAddressSetGet(pCtrl->pAddressSet, 
+                            pc - method->insns)) {
+                    doStop = true;
+                    msg = "new line";
+                }
+            }
+        } else {
+            assert(pCtrl->depth == SD_OUT);
+            /*
+             * Return from the current method.  We break when the frame
+             * depth pops up.
+             *
+             * This differs from the "method exit" break in that it stops
+             * with the PC at the next instruction in the returned-to
+             * function, rather than the end of the returning function.
+             */
+            frameDepth = dvmComputeVagueFrameDepth(self, fp);
+            if (frameDepth < pCtrl->frameDepth) {
+                doStop = true;
+                msg = "method pop";
+            }
+        }
+
+        if (doStop) {
+            LOGV("#####S %s\n", msg);
+            eventFlags |= DBG_SINGLE_STEP;
+        }
+    }
+
+    /*
+     * Check to see if this is a "return" instruction.  JDWP says we should
+     * send the event *after* the code has been executed, but it also says
+     * the location we provide is the last instruction.  Since the "return"
+     * instruction has no interesting side effects, we should be safe.
+     * (We can't just move this down to the returnFromMethod label because
+     * we potentially need to combine it with other events.)
+     *
+     * We're also not supposed to generate a method exit event if the method
+     * terminates "with a thrown exception".
+     */
+    u2 inst = INST_INST(FETCH(0));
+    if (inst == OP_RETURN_VOID || inst == OP_RETURN || inst == OP_RETURN_WIDE ||
+        inst == OP_RETURN_OBJECT)
+    {
+        eventFlags |= DBG_METHOD_EXIT;
+    }
+
+    /*
+     * If there's something interesting going on, see if it matches one
+     * of the debugger filters.
+     */
+    if (eventFlags != 0) {
+        Object* thisPtr = dvmGetThisPtr(method, fp);
+        if (thisPtr != NULL && !dvmIsValidObject(thisPtr)) {
+            /*
+             * TODO: remove this check if we're confident that the "this"
+             * pointer is where it should be -- slows us down, especially
+             * during single-step.
+             */
+            char* desc = dexProtoCopyMethodDescriptor(&method->prototype);
+            LOGE("HEY: invalid 'this' ptr %p (%s.%s %s)\n", thisPtr,
+                method->clazz->descriptor, method->name, desc);
+            free(desc);
+            dvmAbort();
+        }
+        dvmDbgPostLocationEvent(method, pc - method->insns, thisPtr,
+            eventFlags);
+    }
+}
+
+/*
+ * Perform some operations at the "top" of the interpreter loop.
+ * This stuff is required to support debugging and profiling.
+ *
+ * Using" __attribute__((noinline))" seems to do more harm than good.  This
+ * is best when inlined due to the large number of parameters, most of
+ * which are local vars in the main interp loop.
+ */
+static void checkDebugAndProf(const u2* pc, const u4* fp, Thread* self,
+    const Method* method, bool* pIsMethodEntry)
+{
+    /* check to see if we've run off end of method */
+    assert(pc >= method->insns && pc <
+            method->insns + dvmGetMethodInsnsSize(method));
+
+#if 0
+    /*
+     * When we hit a specific method, enable verbose instruction logging.
+     * Sometimes it's helpful to use the debugger attach as a trigger too.
+     */
+    if (*pIsMethodEntry) {
+        static const char* cd = "Landroid/test/Arithmetic;";
+        static const char* mn = "shiftTest2";
+        static const char* sg = "()V";
+
+        if (/*gDvm.debuggerActive &&*/
+            strcmp(method->clazz->descriptor, cd) == 0 &&
+            strcmp(method->name, mn) == 0 &&
+            strcmp(method->signature, sg) == 0)
+        {
+            LOGW("Reached %s.%s, enabling verbose mode\n",
+                method->clazz->descriptor, method->name);
+            android_setMinPriority(LOG_TAG"i", ANDROID_LOG_VERBOSE);
+            dumpRegs(method, fp, true);
+        }
+
+        if (!gDvm.debuggerActive)
+            *pIsMethodEntry = false;
+    }
+#endif
+
+    /*
+     * If the debugger is attached, check for events.  If the profiler is
+     * enabled, update that too.
+     *
+     * This code is executed for every instruction we interpret, so for
+     * performance we use a couple of #ifdef blocks instead of runtime tests.
+     */
+#ifdef WITH_PROFILER
+    /* profiler and probably debugger */
+    bool isEntry = *pIsMethodEntry;
+    if (isEntry) {
+        *pIsMethodEntry = false;
+        TRACE_METHOD_ENTER(self, method);
+    }
+    if (gDvm.debuggerActive) {
+        updateDebugger(method, pc, fp, isEntry, self);
+    }
+    if (gDvm.instructionCountEnableCount != 0) {
+        /*
+         * Count up the #of executed instructions.  This isn't synchronized
+         * for thread-safety; if we need that we should make this
+         * thread-local and merge counts into the global area when threads
+         * exit (perhaps suspending all other threads GC-style and pulling
+         * the data out of them).
+         */
+        int inst = *pc & 0xff;
+        gDvm.executedInstrCounts[inst]++;
+    }
+#else
+    /* debugger only */
+    if (gDvm.debuggerActive) {
+        bool isEntry = *pIsMethodEntry;
+        updateDebugger(method, pc, fp, isEntry, self);
+        if (isEntry)
+            *pIsMethodEntry = false;
+    }
+#endif
+}
+
+
+/* File: portable/entry.c */
+/*
+ * Main interpreter loop.
+ *
+ * This was written with an ARM implementation in mind.
+ */
+bool INTERP_FUNC_NAME(Thread* self, InterpState* interpState)
+{
+#if defined(EASY_GDB)
+    StackSaveArea* debugSaveArea = SAVEAREA_FROM_FP(self->curFrame);
+#endif
+#if INTERP_TYPE == INTERP_DBG
+    bool debugIsMethodEntry = interpState->debugIsMethodEntry;
+#endif
+#if defined(WITH_TRACKREF_CHECKS)
+    int debugTrackedRefStart = interpState->debugTrackedRefStart;
+#endif
+    DvmDex* methodClassDex;     // curMethod->clazz->pDvmDex
+    JValue retval;
+
+    /* core state */
+    const Method* curMethod;    // method we're interpreting
+    const u2* pc;               // program counter
+    u4* fp;                     // frame pointer
+    u2 inst;                    // current instruction
+    /* instruction decoding */
+    u2 ref;                     // 16-bit quantity fetched directly
+    u2 vsrc1, vsrc2, vdst;      // usually used for register indexes
+    /* method call setup */
+    const Method* methodToCall;
+    bool methodCallRange;
+
+#if defined(THREADED_INTERP)
+    /* static computed goto table */
+    DEFINE_GOTO_TABLE(handlerTable);
+#endif
+
+    /* copy state in */
+    curMethod = interpState->method;
+    pc = interpState->pc;
+    fp = interpState->fp;
+    retval = interpState->retval;   /* only need for kInterpEntryReturn? */
+
+    methodClassDex = curMethod->clazz->pDvmDex;
+
+    LOGVV("threadid=%d: entry(%s) %s.%s pc=0x%x fp=%p ep=%d\n",
+        self->threadId, (interpState->nextMode == INTERP_STD) ? "STD" : "DBG",
+        curMethod->clazz->descriptor, curMethod->name, pc - curMethod->insns,
+        fp, interpState->entryPoint);
+
+    /*
+     * DEBUG: scramble this to ensure we're not relying on it.
+     */
+    methodToCall = (const Method*) -1;
+
+#if INTERP_TYPE == INTERP_DBG
+    if (debugIsMethodEntry) {
+        ILOGD("|-- Now interpreting %s.%s", curMethod->clazz->descriptor,
+                curMethod->name);
+        DUMP_REGS(curMethod, interpState->fp, false);
+    }
+#endif
+
+    switch (interpState->entryPoint) {
+    case kInterpEntryInstr:
+        /* just fall through to instruction loop or threaded kickstart */
+        break;
+    case kInterpEntryReturn:
+        goto returnFromMethod;
+    case kInterpEntryThrow:
+        goto exceptionThrown;
+    default:
+        dvmAbort();
+    }
+
+#ifdef THREADED_INTERP
+    FINISH(0);                  /* fetch and execute first instruction */
+#else
+    while (1) {
+        CHECK_DEBUG_AND_PROF(); /* service debugger and profiling */
+        CHECK_TRACKED_REFS();   /* check local reference tracking */
+
+        /* fetch the next 16 bits from the instruction stream */
+        inst = FETCH(0);
+
+        switch (INST_INST(inst)) {
+#endif
+
+/*--- start of opcodes ---*/
+
+/* File: c/OP_NOP.c */
+HANDLE_OPCODE(OP_NOP)
+    FINISH(1);
+OP_END
+
+/* File: c/OP_MOVE.c */
+HANDLE_OPCODE(OP_MOVE /*vA, vB*/)
+    vdst = INST_A(inst);
+    vsrc1 = INST_B(inst);
+    ILOGV("|move%s v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(1);
+OP_END
+
+/* File: c/OP_MOVE_FROM16.c */
+HANDLE_OPCODE(OP_MOVE_FROM16 /*vAA, vBBBB*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|move%s/from16 v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE_FROM16) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(2);
+OP_END
+
+/* File: c/OP_MOVE_16.c */
+HANDLE_OPCODE(OP_MOVE_16 /*vAAAA, vBBBB*/)
+    vdst = FETCH(1);
+    vsrc1 = FETCH(2);
+    ILOGV("|move%s/16 v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE_16) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(3);
+OP_END
+
+/* File: c/OP_MOVE_WIDE.c */
+HANDLE_OPCODE(OP_MOVE_WIDE /*vA, vB*/)
+    /* IMPORTANT: must correctly handle overlapping registers, e.g. both
+     * "move-wide v6, v7" and "move-wide v7, v6" */
+    vdst = INST_A(inst);
+    vsrc1 = INST_B(inst);
+    ILOGV("|move-wide v%d,v%d %s(v%d=0x%08llx)", vdst, vsrc1,
+        kSpacing+5, vdst, GET_REGISTER_WIDE(vsrc1));
+    SET_REGISTER_WIDE(vdst, GET_REGISTER_WIDE(vsrc1));
+    FINISH(1);
+OP_END
+
+/* File: c/OP_MOVE_WIDE_FROM16.c */
+HANDLE_OPCODE(OP_MOVE_WIDE_FROM16 /*vAA, vBBBB*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|move-wide/from16 v%d,v%d  (v%d=0x%08llx)", vdst, vsrc1,
+        vdst, GET_REGISTER_WIDE(vsrc1));
+    SET_REGISTER_WIDE(vdst, GET_REGISTER_WIDE(vsrc1));
+    FINISH(2);
+OP_END
+
+/* File: c/OP_MOVE_WIDE_16.c */
+HANDLE_OPCODE(OP_MOVE_WIDE_16 /*vAAAA, vBBBB*/)
+    vdst = FETCH(1);
+    vsrc1 = FETCH(2);
+    ILOGV("|move-wide/16 v%d,v%d %s(v%d=0x%08llx)", vdst, vsrc1,
+        kSpacing+8, vdst, GET_REGISTER_WIDE(vsrc1));
+    SET_REGISTER_WIDE(vdst, GET_REGISTER_WIDE(vsrc1));
+    FINISH(3);
+OP_END
+
+/* File: c/OP_MOVE_OBJECT.c */
+/* File: c/OP_MOVE.c */
+HANDLE_OPCODE(OP_MOVE_OBJECT /*vA, vB*/)
+    vdst = INST_A(inst);
+    vsrc1 = INST_B(inst);
+    ILOGV("|move%s v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(1);
+OP_END
+
+
+/* File: c/OP_MOVE_OBJECT_FROM16.c */
+/* File: c/OP_MOVE_FROM16.c */
+HANDLE_OPCODE(OP_MOVE_OBJECT_FROM16 /*vAA, vBBBB*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|move%s/from16 v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE_FROM16) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(2);
+OP_END
+
+
+/* File: c/OP_MOVE_OBJECT_16.c */
+/* File: c/OP_MOVE_16.c */
+HANDLE_OPCODE(OP_MOVE_OBJECT_16 /*vAAAA, vBBBB*/)
+    vdst = FETCH(1);
+    vsrc1 = FETCH(2);
+    ILOGV("|move%s/16 v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE_16) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(3);
+OP_END
+
+
+/* File: c/OP_MOVE_RESULT.c */
+HANDLE_OPCODE(OP_MOVE_RESULT /*vAA*/)
+    vdst = INST_AA(inst);
+    ILOGV("|move-result%s v%d %s(v%d=0x%08x)",
+         (INST_INST(inst) == OP_MOVE_RESULT) ? "" : "-object",
+         vdst, kSpacing+4, vdst,retval.i);
+    SET_REGISTER(vdst, retval.i);
+    FINISH(1);
+OP_END
+
+/* File: c/OP_MOVE_RESULT_WIDE.c */
+HANDLE_OPCODE(OP_MOVE_RESULT_WIDE /*vAA*/)
+    vdst = INST_AA(inst);
+    ILOGV("|move-result-wide v%d %s(0x%08llx)", vdst, kSpacing, retval.j);
+    SET_REGISTER_WIDE(vdst, retval.j);
+    FINISH(1);
+OP_END
+
+/* File: c/OP_MOVE_RESULT_OBJECT.c */
+/* File: c/OP_MOVE_RESULT.c */
+HANDLE_OPCODE(OP_MOVE_RESULT_OBJECT /*vAA*/)
+    vdst = INST_AA(inst);
+    ILOGV("|move-result%s v%d %s(v%d=0x%08x)",
+         (INST_INST(inst) == OP_MOVE_RESULT) ? "" : "-object",
+         vdst, kSpacing+4, vdst,retval.i);
+    SET_REGISTER(vdst, retval.i);
+    FINISH(1);
+OP_END
+
+
+/* File: c/OP_MOVE_EXCEPTION.c */
+HANDLE_OPCODE(OP_MOVE_EXCEPTION /*vAA*/)
+    vdst = INST_AA(inst);
+    ILOGV("|move-exception v%d", vdst);
+    assert(self->exception != NULL);
+    SET_REGISTER(vdst, (u4)self->exception);
+    dvmClearException(self);
+    FINISH(1);
+OP_END
+
+/* File: c/OP_RETURN_VOID.c */
+HANDLE_OPCODE(OP_RETURN_VOID /**/)
+    ILOGV("|return-void");
+#ifndef NDEBUG
+    retval.j = 0xababababULL;    // placate valgrind
+#endif
+    GOTO_returnFromMethod();
+OP_END
+
+/* File: c/OP_RETURN.c */
+HANDLE_OPCODE(OP_RETURN /*vAA*/)
+    vsrc1 = INST_AA(inst);
+    ILOGV("|return%s v%d",
+        (INST_INST(inst) == OP_RETURN) ? "" : "-object", vsrc1);
+    retval.i = GET_REGISTER(vsrc1);
+    GOTO_returnFromMethod();
+OP_END
+
+/* File: c/OP_RETURN_WIDE.c */
+HANDLE_OPCODE(OP_RETURN_WIDE /*vAA*/)
+    vsrc1 = INST_AA(inst);
+    ILOGV("|return-wide v%d", vsrc1);
+    retval.j = GET_REGISTER_WIDE(vsrc1);
+    GOTO_returnFromMethod();
+OP_END
+
+/* File: c/OP_RETURN_OBJECT.c */
+/* File: c/OP_RETURN.c */
+HANDLE_OPCODE(OP_RETURN_OBJECT /*vAA*/)
+    vsrc1 = INST_AA(inst);
+    ILOGV("|return%s v%d",
+        (INST_INST(inst) == OP_RETURN) ? "" : "-object", vsrc1);
+    retval.i = GET_REGISTER(vsrc1);
+    GOTO_returnFromMethod();
+OP_END
+
+
+/* File: c/OP_CONST_4.c */
+HANDLE_OPCODE(OP_CONST_4 /*vA, #+B*/)
+    {
+        s4 tmp;
+
+        vdst = INST_A(inst);
+        tmp = (s4) (INST_B(inst) << 28) >> 28;  // sign extend 4-bit value
+        ILOGV("|const/4 v%d,#0x%02x", vdst, (s4)tmp);
+        SET_REGISTER(vdst, tmp);
+    }
+    FINISH(1);
+OP_END
+
+/* File: c/OP_CONST_16.c */
+HANDLE_OPCODE(OP_CONST_16 /*vAA, #+BBBB*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|const/16 v%d,#0x%04x", vdst, (s2)vsrc1);
+    SET_REGISTER(vdst, (s2) vsrc1);
+    FINISH(2);
+OP_END
+
+/* File: c/OP_CONST.c */
+HANDLE_OPCODE(OP_CONST /*vAA, #+BBBBBBBB*/)
+    {
+        u4 tmp;
+
+        vdst = INST_AA(inst);
+        tmp = FETCH(1);
+        tmp |= (u4)FETCH(2) << 16;
+        ILOGV("|const v%d,#0x%08x", vdst, tmp);
+        SET_REGISTER(vdst, tmp);
+    }
+    FINISH(3);
+OP_END
+
+/* File: c/OP_CONST_HIGH16.c */
+HANDLE_OPCODE(OP_CONST_HIGH16 /*vAA, #+BBBB0000*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|const/high16 v%d,#0x%04x0000", vdst, vsrc1);
+    SET_REGISTER(vdst, vsrc1 << 16);
+    FINISH(2);
+OP_END
+
+/* File: c/OP_CONST_WIDE_16.c */
+HANDLE_OPCODE(OP_CONST_WIDE_16 /*vAA, #+BBBB*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|const-wide/16 v%d,#0x%04x", vdst, (s2)vsrc1);
+    SET_REGISTER_WIDE(vdst, (s2)vsrc1);
+    FINISH(2);
+OP_END
+
+/* File: c/OP_CONST_WIDE_32.c */
+HANDLE_OPCODE(OP_CONST_WIDE_32 /*vAA, #+BBBBBBBB*/)
+    {
+        u4 tmp;
+
+        vdst = INST_AA(inst);
+        tmp = FETCH(1);
+        tmp |= (u4)FETCH(2) << 16;
+        ILOGV("|const-wide/32 v%d,#0x%08x", vdst, tmp);
+        SET_REGISTER_WIDE(vdst, (s4) tmp);
+    }
+    FINISH(3);
+OP_END
+
+/* File: c/OP_CONST_WIDE.c */
+HANDLE_OPCODE(OP_CONST_WIDE /*vAA, #+BBBBBBBBBBBBBBBB*/)
+    {
+        u8 tmp;
+
+        vdst = INST_AA(inst);
+        tmp = FETCH(1);
+        tmp |= (u8)FETCH(2) << 16;
+        tmp |= (u8)FETCH(3) << 32;
+        tmp |= (u8)FETCH(4) << 48;
+        ILOGV("|const-wide v%d,#0x%08llx", vdst, tmp);
+        SET_REGISTER_WIDE(vdst, tmp);
+    }
+    FINISH(5);
+OP_END
+
+/* File: c/OP_CONST_WIDE_HIGH16.c */
+HANDLE_OPCODE(OP_CONST_WIDE_HIGH16 /*vAA, #+BBBB000000000000*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|const-wide/high16 v%d,#0x%04x000000000000", vdst, vsrc1);
+    SET_REGISTER_WIDE(vdst, ((u8) vsrc1) << 48);
+    FINISH(2);
+OP_END
+
+/* File: c/OP_CONST_STRING.c */
+HANDLE_OPCODE(OP_CONST_STRING /*vAA, string@BBBB*/)
+    {
+        StringObject* strObj;
+
+        vdst = INST_AA(inst);
+        ref = FETCH(1);
+        ILOGV("|const-string v%d string@0x%04x", vdst, ref);
+        strObj = dvmDexGetResolvedString(methodClassDex, ref);
+        if (strObj == NULL) {
+            EXPORT_PC();
+            strObj = dvmResolveString(curMethod->clazz, ref);
+            if (strObj == NULL)
+                GOTO_exceptionThrown();
+        }
+        SET_REGISTER(vdst, (u4) strObj);
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_CONST_STRING_JUMBO.c */
+HANDLE_OPCODE(OP_CONST_STRING_JUMBO /*vAA, string@BBBBBBBB*/)
+    {
+        StringObject* strObj;
+        u4 tmp;
+
+        vdst = INST_AA(inst);
+        tmp = FETCH(1);
+        tmp |= (u4)FETCH(2) << 16;
+        ILOGV("|const-string/jumbo v%d string@0x%08x", vdst, tmp);
+        strObj = dvmDexGetResolvedString(methodClassDex, tmp);
+        if (strObj == NULL) {
+            EXPORT_PC();
+            strObj = dvmResolveString(curMethod->clazz, tmp);
+            if (strObj == NULL)
+                GOTO_exceptionThrown();
+        }
+        SET_REGISTER(vdst, (u4) strObj);
+    }
+    FINISH(3);
+OP_END
+
+/* File: c/OP_CONST_CLASS.c */
+HANDLE_OPCODE(OP_CONST_CLASS /*vAA, class@BBBB*/)
+    {
+        ClassObject* clazz;
+
+        vdst = INST_AA(inst);
+        ref = FETCH(1);
+        ILOGV("|const-class v%d class@0x%04x", vdst, ref);
+        clazz = dvmDexGetResolvedClass(methodClassDex, ref);
+        if (clazz == NULL) {
+            EXPORT_PC();
+            clazz = dvmResolveClass(curMethod->clazz, ref, true);
+            if (clazz == NULL)
+                GOTO_exceptionThrown();
+        }
+        SET_REGISTER(vdst, (u4) clazz);
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_MONITOR_ENTER.c */
+HANDLE_OPCODE(OP_MONITOR_ENTER /*vAA*/)
+    {
+        Object* obj;
+
+        vsrc1 = INST_AA(inst);
+        ILOGV("|monitor-enter v%d %s(0x%08x)",
+            vsrc1, kSpacing+6, GET_REGISTER(vsrc1));
+        obj = (Object*)GET_REGISTER(vsrc1);
+        if (!checkForNullExportPC(obj, fp, pc))
+            GOTO_exceptionThrown();
+        ILOGV("+ locking %p %s\n", obj, obj->clazz->descriptor);
+#ifdef WITH_MONITOR_TRACKING
+        EXPORT_PC();        /* need for stack trace */
+#endif
+        dvmLockObject(self, obj);
+#ifdef WITH_DEADLOCK_PREDICTION
+        if (dvmCheckException(self))
+            GOTO_exceptionThrown();
+#endif
+    }
+    FINISH(1);
+OP_END
+
+/* File: c/OP_MONITOR_EXIT.c */
+HANDLE_OPCODE(OP_MONITOR_EXIT /*vAA*/)
+    {
+        Object* obj;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);
+        ILOGV("|monitor-exit v%d %s(0x%08x)",
+            vsrc1, kSpacing+5, GET_REGISTER(vsrc1));
+        obj = (Object*)GET_REGISTER(vsrc1);
+        if (!checkForNull(obj)) {
+            /*
+             * The exception needs to be processed at the *following*
+             * instruction, not the current instruction (see the Dalvik
+             * spec).  Because we're jumping to an exception handler,
+             * we're not actually at risk of skipping an instruction
+             * by doing so.
+             */
+            ADJUST_PC(1);           /* monitor-exit width is 1 */
+            GOTO_exceptionThrown();
+        }
+        ILOGV("+ unlocking %p %s\n", obj, obj->clazz->descriptor);
+        if (!dvmUnlockObject(self, obj)) {
+            assert(dvmCheckException(self));
+            ADJUST_PC(1);
+            GOTO_exceptionThrown();
+        }
+    }
+    FINISH(1);
+OP_END
+
+/* File: c/OP_CHECK_CAST.c */
+HANDLE_OPCODE(OP_CHECK_CAST /*vAA, class@BBBB*/)
+    {
+        ClassObject* clazz;
+        Object* obj;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);
+        ref = FETCH(1);         /* class to check against */
+        ILOGV("|check-cast v%d,class@0x%04x", vsrc1, ref);
+
+        obj = (Object*)GET_REGISTER(vsrc1);
+        if (obj != NULL) {
+#if defined(WITH_EXTRA_OBJECT_VALIDATION)
+            if (!checkForNull(obj))
+                GOTO_exceptionThrown();
+#endif
+            clazz = dvmDexGetResolvedClass(methodClassDex, ref);
+            if (clazz == NULL) {
+                clazz = dvmResolveClass(curMethod->clazz, ref, false);
+                if (clazz == NULL)
+                    GOTO_exceptionThrown();
+            }
+            if (!dvmInstanceof(obj->clazz, clazz)) {
+                dvmThrowExceptionWithClassMessage(
+                    "Ljava/lang/ClassCastException;", obj->clazz->descriptor);
+                GOTO_exceptionThrown();
+            }
+        }
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_INSTANCE_OF.c */
+HANDLE_OPCODE(OP_INSTANCE_OF /*vA, vB, class@CCCC*/)
+    {
+        ClassObject* clazz;
+        Object* obj;
+
+        vdst = INST_A(inst);
+        vsrc1 = INST_B(inst);   /* object to check */
+        ref = FETCH(1);         /* class to check against */
+        ILOGV("|instance-of v%d,v%d,class@0x%04x", vdst, vsrc1, ref);
+
+        obj = (Object*)GET_REGISTER(vsrc1);
+        if (obj == NULL) {
+            SET_REGISTER(vdst, 0);
+        } else {
+#if defined(WITH_EXTRA_OBJECT_VALIDATION)
+            if (!checkForNullExportPC(obj, fp, pc))
+                GOTO_exceptionThrown();
+#endif
+            clazz = dvmDexGetResolvedClass(methodClassDex, ref);
+            if (clazz == NULL) {
+                EXPORT_PC();
+                clazz = dvmResolveClass(curMethod->clazz, ref, true);
+                if (clazz == NULL)
+                    GOTO_exceptionThrown();
+            }
+            SET_REGISTER(vdst, dvmInstanceof(obj->clazz, clazz));
+        }
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_ARRAY_LENGTH.c */
+HANDLE_OPCODE(OP_ARRAY_LENGTH /*vA, vB*/)
+    {
+        ArrayObject* arrayObj;
+
+        vdst = INST_A(inst);
+        vsrc1 = INST_B(inst);
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);
+        ILOGV("|array-length v%d,v%d  (%p)", vdst, vsrc1, arrayObj);
+        if (!checkForNullExportPC((Object*) arrayObj, fp, pc))
+            GOTO_exceptionThrown();
+        /* verifier guarantees this is an array reference */
+        SET_REGISTER(vdst, arrayObj->length);
+    }
+    FINISH(1);
+OP_END
+
+/* File: c/OP_NEW_INSTANCE.c */
+HANDLE_OPCODE(OP_NEW_INSTANCE /*vAA, class@BBBB*/)
+    {
+        ClassObject* clazz;
+        Object* newObj;
+
+        EXPORT_PC();
+
+        vdst = INST_AA(inst);
+        ref = FETCH(1);
+        ILOGV("|new-instance v%d,class@0x%04x", vdst, ref);
+        clazz = dvmDexGetResolvedClass(methodClassDex, ref);
+        if (clazz == NULL) {
+            clazz = dvmResolveClass(curMethod->clazz, ref, false);
+            if (clazz == NULL)
+                GOTO_exceptionThrown();
+        }
+
+        if (!dvmIsClassInitialized(clazz) && !dvmInitClass(clazz))
+            GOTO_exceptionThrown();
+
+        /*
+         * Note: the verifier can ensure that this never happens, allowing us
+         * to remove the check.  However, the spec requires we throw the
+         * exception at runtime, not verify time, so the verifier would
+         * need to replace the new-instance call with a magic "throw
+         * InstantiationError" instruction.
+         *
+         * Since this relies on the verifier, which is optional, we would
+         * also need a "new-instance-quick" instruction to identify instances
+         * that don't require the check.
+         */
+        if (dvmIsInterfaceClass(clazz) || dvmIsAbstractClass(clazz)) {
+            dvmThrowExceptionWithClassMessage("Ljava/lang/InstantiationError;",
+                clazz->descriptor);
+            GOTO_exceptionThrown();
+        }
+        newObj = dvmAllocObject(clazz, ALLOC_DONT_TRACK);
+        if (newObj == NULL)
+            GOTO_exceptionThrown();
+        SET_REGISTER(vdst, (u4) newObj);
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_NEW_ARRAY.c */
+HANDLE_OPCODE(OP_NEW_ARRAY /*vA, vB, class@CCCC*/)
+    {
+        ClassObject* arrayClass;
+        ArrayObject* newArray;
+        s4 length;
+
+        EXPORT_PC();
+
+        vdst = INST_A(inst);
+        vsrc1 = INST_B(inst);       /* length reg */
+        ref = FETCH(1);
+        ILOGV("|new-array v%d,v%d,class@0x%04x  (%d elements)",
+            vdst, vsrc1, ref, (s4) GET_REGISTER(vsrc1));
+        length = (s4) GET_REGISTER(vsrc1);
+        if (length < 0) {
+            dvmThrowException("Ljava/lang/NegativeArraySizeException;", NULL);
+            GOTO_exceptionThrown();
+        }
+        arrayClass = dvmDexGetResolvedClass(methodClassDex, ref);
+        if (arrayClass == NULL) {
+            arrayClass = dvmResolveClass(curMethod->clazz, ref, false);
+            if (arrayClass == NULL)
+                GOTO_exceptionThrown();
+        }
+        /* verifier guarantees this is an array class */
+        assert(dvmIsArrayClass(arrayClass));
+        assert(dvmIsClassInitialized(arrayClass));
+
+        newArray = dvmAllocArrayByClass(arrayClass, length, ALLOC_DONT_TRACK);
+        if (newArray == NULL)
+            GOTO_exceptionThrown();
+        SET_REGISTER(vdst, (u4) newArray);
+    }
+    FINISH(2);
+OP_END
+
+
+/* File: c/OP_FILLED_NEW_ARRAY.c */
+HANDLE_OPCODE(OP_FILLED_NEW_ARRAY /*vB, {vD, vE, vF, vG, vA}, class@CCCC*/)
+    GOTO_invoke(filledNewArray, false);
+OP_END
+
+/* File: c/OP_FILLED_NEW_ARRAY_RANGE.c */
+HANDLE_OPCODE(OP_FILLED_NEW_ARRAY_RANGE /*{vCCCC..v(CCCC+AA-1)}, class@BBBB*/)
+    GOTO_invoke(filledNewArray, true);
+OP_END
+
+/* File: c/OP_FILL_ARRAY_DATA.c */
+HANDLE_OPCODE(OP_FILL_ARRAY_DATA)   /*vAA, +BBBBBBBB*/
+    {
+        const u2* arrayData;
+        s4 offset;
+        ArrayObject* arrayObj;
+
+        EXPORT_PC();
+        vsrc1 = INST_AA(inst);
+        offset = FETCH(1) | (((s4) FETCH(2)) << 16);
+        ILOGV("|fill-array-data v%d +0x%04x", vsrc1, offset);
+        arrayData = pc + offset;       // offset in 16-bit units
+#ifndef NDEBUG
+        if (arrayData < curMethod->insns ||
+            arrayData >= curMethod->insns + dvmGetMethodInsnsSize(curMethod))
+        {
+            /* should have been caught in verifier */
+            dvmThrowException("Ljava/lang/InternalError;", 
+                              "bad fill array data");
+            GOTO_exceptionThrown();
+        }
+#endif
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);
+        if (!dvmInterpHandleFillArrayData(arrayObj, arrayData)) {
+            GOTO_exceptionThrown();
+        }
+        FINISH(3);
+    }
+OP_END
+
+/* File: c/OP_THROW.c */
+HANDLE_OPCODE(OP_THROW /*vAA*/)
+    {
+        Object* obj;
+
+        vsrc1 = INST_AA(inst);
+        ILOGV("|throw v%d  (%p)", vsrc1, (void*)GET_REGISTER(vsrc1));
+        obj = (Object*) GET_REGISTER(vsrc1);
+        if (!checkForNullExportPC(obj, fp, pc)) {
+            /* will throw a null pointer exception */
+            LOGVV("Bad exception\n");
+        } else {
+            /* use the requested exception */
+            dvmSetException(self, obj);
+        }
+        GOTO_exceptionThrown();
+    }
+OP_END
+
+/* File: c/OP_GOTO.c */
+HANDLE_OPCODE(OP_GOTO /*+AA*/)
+    vdst = INST_AA(inst);
+    if ((s1)vdst < 0)
+        ILOGV("|goto -0x%02x", -((s1)vdst));
+    else
+        ILOGV("|goto +0x%02x", ((s1)vdst));
+    ILOGV("> branch taken");
+    if ((s1)vdst < 0)
+        PERIODIC_CHECKS(kInterpEntryInstr, (s1)vdst);
+    FINISH((s1)vdst);
+OP_END
+
+/* File: c/OP_GOTO_16.c */
+HANDLE_OPCODE(OP_GOTO_16 /*+AAAA*/)
+    {
+        s4 offset = (s2) FETCH(1);          /* sign-extend next code unit */
+
+        if (offset < 0)
+            ILOGV("|goto/16 -0x%04x", -offset);
+        else
+            ILOGV("|goto/16 +0x%04x", offset);
+        ILOGV("> branch taken");
+        if (offset < 0)
+            PERIODIC_CHECKS(kInterpEntryInstr, offset);
+        FINISH(offset);
+    }
+OP_END
+
+/* File: c/OP_GOTO_32.c */
+HANDLE_OPCODE(OP_GOTO_32 /*+AAAAAAAA*/)
+    {
+        s4 offset = FETCH(1);               /* low-order 16 bits */
+        offset |= ((s4) FETCH(2)) << 16;    /* high-order 16 bits */
+
+        if (offset < 0)
+            ILOGV("|goto/32 -0x%08x", -offset);
+        else
+            ILOGV("|goto/32 +0x%08x", offset);
+        ILOGV("> branch taken");
+        if (offset <= 0)    /* allowed to branch to self */
+            PERIODIC_CHECKS(kInterpEntryInstr, offset);
+        FINISH(offset);
+    }
+OP_END
+
+/* File: c/OP_PACKED_SWITCH.c */
+HANDLE_OPCODE(OP_PACKED_SWITCH /*vAA, +BBBB*/)
+    {
+        const u2* switchData;
+        u4 testVal;
+        s4 offset;
+
+        vsrc1 = INST_AA(inst);
+        offset = FETCH(1) | (((s4) FETCH(2)) << 16);
+        ILOGV("|packed-switch v%d +0x%04x", vsrc1, vsrc2);
+        switchData = pc + offset;       // offset in 16-bit units
+#ifndef NDEBUG
+        if (switchData < curMethod->insns ||
+            switchData >= curMethod->insns + dvmGetMethodInsnsSize(curMethod))
+        {
+            /* should have been caught in verifier */
+            EXPORT_PC();
+            dvmThrowException("Ljava/lang/InternalError;", "bad packed switch");
+            GOTO_exceptionThrown();
+        }
+#endif
+        testVal = GET_REGISTER(vsrc1);
+
+        offset = dvmInterpHandlePackedSwitch(switchData, testVal);
+        ILOGV("> branch taken (0x%04x)\n", offset);
+        if (offset <= 0)  /* uncommon */
+            PERIODIC_CHECKS(kInterpEntryInstr, offset);
+        FINISH(offset);
+    }
+OP_END
+
+/* File: c/OP_SPARSE_SWITCH.c */
+HANDLE_OPCODE(OP_SPARSE_SWITCH /*vAA, +BBBB*/)
+    {
+        const u2* switchData;
+        u4 testVal;
+        s4 offset;
+
+        vsrc1 = INST_AA(inst);
+        offset = FETCH(1) | (((s4) FETCH(2)) << 16);
+        ILOGV("|sparse-switch v%d +0x%04x", vsrc1, vsrc2);
+        switchData = pc + offset;       // offset in 16-bit units
+#ifndef NDEBUG
+        if (switchData < curMethod->insns ||
+            switchData >= curMethod->insns + dvmGetMethodInsnsSize(curMethod))
+        {
+            /* should have been caught in verifier */
+            EXPORT_PC();
+            dvmThrowException("Ljava/lang/InternalError;", "bad sparse switch");
+            GOTO_exceptionThrown();
+        }
+#endif
+        testVal = GET_REGISTER(vsrc1);
+
+        offset = dvmInterpHandleSparseSwitch(switchData, testVal);
+        ILOGV("> branch taken (0x%04x)\n", offset);
+        if (offset <= 0)  /* uncommon */
+            PERIODIC_CHECKS(kInterpEntryInstr, offset);
+        FINISH(offset);
+    }
+OP_END
+
+/* File: c/OP_CMPL_FLOAT.c */
+HANDLE_OP_CMPX(OP_CMPL_FLOAT, "l-float", float, _FLOAT, -1)
+OP_END
+
+/* File: c/OP_CMPG_FLOAT.c */
+HANDLE_OP_CMPX(OP_CMPG_FLOAT, "g-float", float, _FLOAT, 1)
+OP_END
+
+/* File: c/OP_CMPL_DOUBLE.c */
+HANDLE_OP_CMPX(OP_CMPL_DOUBLE, "l-double", double, _DOUBLE, -1)
+OP_END
+
+/* File: c/OP_CMPG_DOUBLE.c */
+HANDLE_OP_CMPX(OP_CMPG_DOUBLE, "g-double", double, _DOUBLE, 1)
+OP_END
+
+/* File: c/OP_CMP_LONG.c */
+HANDLE_OP_CMPX(OP_CMP_LONG, "-long", s8, _WIDE, 0)
+OP_END
+
+/* File: c/OP_IF_EQ.c */
+HANDLE_OP_IF_XX(OP_IF_EQ, "eq", ==)
+OP_END
+
+/* File: c/OP_IF_NE.c */
+HANDLE_OP_IF_XX(OP_IF_NE, "ne", !=)
+OP_END
+
+/* File: c/OP_IF_LT.c */
+HANDLE_OP_IF_XX(OP_IF_LT, "lt", <)
+OP_END
+
+/* File: c/OP_IF_GE.c */
+HANDLE_OP_IF_XX(OP_IF_GE, "ge", >=)
+OP_END
+
+/* File: c/OP_IF_GT.c */
+HANDLE_OP_IF_XX(OP_IF_GT, "gt", >)
+OP_END
+
+/* File: c/OP_IF_LE.c */
+HANDLE_OP_IF_XX(OP_IF_LE, "le", <=)
+OP_END
+
+/* File: c/OP_IF_EQZ.c */
+HANDLE_OP_IF_XXZ(OP_IF_EQZ, "eqz", ==)
+OP_END
+
+/* File: c/OP_IF_NEZ.c */
+HANDLE_OP_IF_XXZ(OP_IF_NEZ, "nez", !=)
+OP_END
+
+/* File: c/OP_IF_LTZ.c */
+HANDLE_OP_IF_XXZ(OP_IF_LTZ, "ltz", <)
+OP_END
+
+/* File: c/OP_IF_GEZ.c */
+HANDLE_OP_IF_XXZ(OP_IF_GEZ, "gez", >=)
+OP_END
+
+/* File: c/OP_IF_GTZ.c */
+HANDLE_OP_IF_XXZ(OP_IF_GTZ, "gtz", >)
+OP_END
+
+/* File: c/OP_IF_LEZ.c */
+HANDLE_OP_IF_XXZ(OP_IF_LEZ, "lez", <=)
+OP_END
+
+/* File: c/OP_UNUSED_3E.c */
+HANDLE_OPCODE(OP_UNUSED_3E)
+OP_END
+
+/* File: c/OP_UNUSED_3F.c */
+HANDLE_OPCODE(OP_UNUSED_3F)
+OP_END
+
+/* File: c/OP_UNUSED_40.c */
+HANDLE_OPCODE(OP_UNUSED_40)
+OP_END
+
+/* File: c/OP_UNUSED_41.c */
+HANDLE_OPCODE(OP_UNUSED_41)
+OP_END
+
+/* File: c/OP_UNUSED_42.c */
+HANDLE_OPCODE(OP_UNUSED_42)
+OP_END
+
+/* File: c/OP_UNUSED_43.c */
+HANDLE_OPCODE(OP_UNUSED_43)
+OP_END
+
+/* File: c/OP_AGET.c */
+HANDLE_OP_AGET(OP_AGET, "", u4, )
+OP_END
+
+/* File: c/OP_AGET_WIDE.c */
+HANDLE_OP_AGET(OP_AGET_WIDE, "-wide", s8, _WIDE)
+OP_END
+
+/* File: c/OP_AGET_OBJECT.c */
+HANDLE_OP_AGET(OP_AGET_OBJECT, "-object", u4, )
+OP_END
+
+/* File: c/OP_AGET_BOOLEAN.c */
+HANDLE_OP_AGET(OP_AGET_BOOLEAN, "-boolean", u1, )
+OP_END
+
+/* File: c/OP_AGET_BYTE.c */
+HANDLE_OP_AGET(OP_AGET_BYTE, "-byte", s1, )
+OP_END
+
+/* File: c/OP_AGET_CHAR.c */
+HANDLE_OP_AGET(OP_AGET_CHAR, "-char", u2, )
+OP_END
+
+/* File: c/OP_AGET_SHORT.c */
+HANDLE_OP_AGET(OP_AGET_SHORT, "-short", s2, )
+OP_END
+
+/* File: c/OP_APUT.c */
+HANDLE_OP_APUT(OP_APUT, "", u4, )
+OP_END
+
+/* File: c/OP_APUT_WIDE.c */
+HANDLE_OP_APUT(OP_APUT_WIDE, "-wide", s8, _WIDE)
+OP_END
+
+/* File: c/OP_APUT_OBJECT.c */
+HANDLE_OPCODE(OP_APUT_OBJECT /*vAA, vBB, vCC*/)
+    {
+        ArrayObject* arrayObj;
+        Object* obj;
+        u2 arrayInfo;
+        EXPORT_PC();
+        vdst = INST_AA(inst);       /* AA: source value */
+        arrayInfo = FETCH(1);
+        vsrc1 = arrayInfo & 0xff;   /* BB: array ptr */
+        vsrc2 = arrayInfo >> 8;     /* CC: index */
+        ILOGV("|aput%s v%d,v%d,v%d", "-object", vdst, vsrc1, vsrc2);
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);
+        if (!checkForNull((Object*) arrayObj))
+            GOTO_exceptionThrown();
+        if (GET_REGISTER(vsrc2) >= arrayObj->length) {
+            dvmThrowException("Ljava/lang/ArrayIndexOutOfBoundsException;",
+                NULL);
+            GOTO_exceptionThrown();
+        }
+        obj = (Object*) GET_REGISTER(vdst);
+        if (obj != NULL) {
+            if (!checkForNull(obj))
+                GOTO_exceptionThrown();
+            if (!dvmCanPutArrayElement(obj->clazz, arrayObj->obj.clazz)) {
+                LOGV("Can't put a '%s'(%p) into array type='%s'(%p)\n",
+                    obj->clazz->descriptor, obj,
+                    arrayObj->obj.clazz->descriptor, arrayObj);
+                //dvmDumpClass(obj->clazz);
+                //dvmDumpClass(arrayObj->obj.clazz);
+                dvmThrowException("Ljava/lang/ArrayStoreException;", NULL);
+                GOTO_exceptionThrown();
+            }
+        }
+        ILOGV("+ APUT[%d]=0x%08x", GET_REGISTER(vsrc2), GET_REGISTER(vdst));
+        ((u4*) arrayObj->contents)[GET_REGISTER(vsrc2)] =
+            GET_REGISTER(vdst);
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_APUT_BOOLEAN.c */
+HANDLE_OP_APUT(OP_APUT_BOOLEAN, "-boolean", u1, )
+OP_END
+
+/* File: c/OP_APUT_BYTE.c */
+HANDLE_OP_APUT(OP_APUT_BYTE, "-byte", s1, )
+OP_END
+
+/* File: c/OP_APUT_CHAR.c */
+HANDLE_OP_APUT(OP_APUT_CHAR, "-char", u2, )
+OP_END
+
+/* File: c/OP_APUT_SHORT.c */
+HANDLE_OP_APUT(OP_APUT_SHORT, "-short", s2, )
+OP_END
+
+/* File: c/OP_IGET.c */
+HANDLE_IGET_X(OP_IGET,                  "", Int, )
+OP_END
+
+/* File: c/OP_IGET_WIDE.c */
+HANDLE_IGET_X(OP_IGET_WIDE,             "-wide", Long, _WIDE)
+OP_END
+
+/* File: c/OP_IGET_OBJECT.c */
+HANDLE_IGET_X(OP_IGET_OBJECT,           "-object", Object, _AS_OBJECT)
+OP_END
+
+/* File: c/OP_IGET_BOOLEAN.c */
+HANDLE_IGET_X(OP_IGET_BOOLEAN,          "", Int, )
+OP_END
+
+/* File: c/OP_IGET_BYTE.c */
+HANDLE_IGET_X(OP_IGET_BYTE,             "", Int, )
+OP_END
+
+/* File: c/OP_IGET_CHAR.c */
+HANDLE_IGET_X(OP_IGET_CHAR,             "", Int, )
+OP_END
+
+/* File: c/OP_IGET_SHORT.c */
+HANDLE_IGET_X(OP_IGET_SHORT,            "", Int, )
+OP_END
+
+/* File: c/OP_IPUT.c */
+HANDLE_IPUT_X(OP_IPUT,                  "", Int, )
+OP_END
+
+/* File: c/OP_IPUT_WIDE.c */
+HANDLE_IPUT_X(OP_IPUT_WIDE,             "-wide", Long, _WIDE)
+OP_END
+
+/* File: c/OP_IPUT_OBJECT.c */
+/*
+ * The VM spec says we should verify that the reference being stored into
+ * the field is assignment compatible.  In practice, many popular VMs don't
+ * do this because it slows down a very common operation.  It's not so bad
+ * for us, since "dexopt" quickens it whenever possible, but it's still an
+ * issue.
+ *
+ * To make this spec-complaint, we'd need to add a ClassObject pointer to
+ * the Field struct, resolve the field's type descriptor at link or class
+ * init time, and then verify the type here.
+ */
+HANDLE_IPUT_X(OP_IPUT_OBJECT,           "-object", Object, _AS_OBJECT)
+OP_END
+
+/* File: c/OP_IPUT_BOOLEAN.c */
+HANDLE_IPUT_X(OP_IPUT_BOOLEAN,          "", Int, )
+OP_END
+
+/* File: c/OP_IPUT_BYTE.c */
+HANDLE_IPUT_X(OP_IPUT_BYTE,             "", Int, )
+OP_END
+
+/* File: c/OP_IPUT_CHAR.c */
+HANDLE_IPUT_X(OP_IPUT_CHAR,             "", Int, )
+OP_END
+
+/* File: c/OP_IPUT_SHORT.c */
+HANDLE_IPUT_X(OP_IPUT_SHORT,            "", Int, )
+OP_END
+
+/* File: c/OP_SGET.c */
+HANDLE_SGET_X(OP_SGET,                  "", Int, )
+OP_END
+
+/* File: c/OP_SGET_WIDE.c */
+HANDLE_SGET_X(OP_SGET_WIDE,             "-wide", Long, _WIDE)
+OP_END
+
+/* File: c/OP_SGET_OBJECT.c */
+HANDLE_SGET_X(OP_SGET_OBJECT,           "-object", Object, _AS_OBJECT)
+OP_END
+
+/* File: c/OP_SGET_BOOLEAN.c */
+HANDLE_SGET_X(OP_SGET_BOOLEAN,          "", Int, )
+OP_END
+
+/* File: c/OP_SGET_BYTE.c */
+HANDLE_SGET_X(OP_SGET_BYTE,             "", Int, )
+OP_END
+
+/* File: c/OP_SGET_CHAR.c */
+HANDLE_SGET_X(OP_SGET_CHAR,             "", Int, )
+OP_END
+
+/* File: c/OP_SGET_SHORT.c */
+HANDLE_SGET_X(OP_SGET_SHORT,            "", Int, )
+OP_END
+
+/* File: c/OP_SPUT.c */
+HANDLE_SPUT_X(OP_SPUT,                  "", Int, )
+OP_END
+
+/* File: c/OP_SPUT_WIDE.c */
+HANDLE_SPUT_X(OP_SPUT_WIDE,             "-wide", Long, _WIDE)
+OP_END
+
+/* File: c/OP_SPUT_OBJECT.c */
+HANDLE_SPUT_X(OP_SPUT_OBJECT,           "-object", Object, _AS_OBJECT)
+OP_END
+
+/* File: c/OP_SPUT_BOOLEAN.c */
+HANDLE_SPUT_X(OP_SPUT_BOOLEAN,          "", Int, )
+OP_END
+
+/* File: c/OP_SPUT_BYTE.c */
+HANDLE_SPUT_X(OP_SPUT_BYTE,             "", Int, )
+OP_END
+
+/* File: c/OP_SPUT_CHAR.c */
+HANDLE_SPUT_X(OP_SPUT_CHAR,             "", Int, )
+OP_END
+
+/* File: c/OP_SPUT_SHORT.c */
+HANDLE_SPUT_X(OP_SPUT_SHORT,            "", Int, )
+OP_END
+
+/* File: c/OP_INVOKE_VIRTUAL.c */
+HANDLE_OPCODE(OP_INVOKE_VIRTUAL /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeVirtual, false);
+OP_END
+
+/* File: c/OP_INVOKE_SUPER.c */
+HANDLE_OPCODE(OP_INVOKE_SUPER /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeSuper, false);
+OP_END
+
+/* File: c/OP_INVOKE_DIRECT.c */
+HANDLE_OPCODE(OP_INVOKE_DIRECT /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeDirect, false);
+OP_END
+
+/* File: c/OP_INVOKE_STATIC.c */
+HANDLE_OPCODE(OP_INVOKE_STATIC /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeStatic, false);
+OP_END
+
+/* File: c/OP_INVOKE_INTERFACE.c */
+HANDLE_OPCODE(OP_INVOKE_INTERFACE /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeInterface, false);
+OP_END
+
+/* File: c/OP_UNUSED_73.c */
+HANDLE_OPCODE(OP_UNUSED_73)
+OP_END
+
+/* File: c/OP_INVOKE_VIRTUAL_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_VIRTUAL_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeVirtual, true);
+OP_END
+
+/* File: c/OP_INVOKE_SUPER_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_SUPER_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeSuper, true);
+OP_END
+
+/* File: c/OP_INVOKE_DIRECT_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_DIRECT_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeDirect, true);
+OP_END
+
+/* File: c/OP_INVOKE_STATIC_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_STATIC_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeStatic, true);
+OP_END
+
+/* File: c/OP_INVOKE_INTERFACE_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_INTERFACE_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeInterface, true);
+OP_END
+
+/* File: c/OP_UNUSED_79.c */
+HANDLE_OPCODE(OP_UNUSED_79)
+OP_END
+
+/* File: c/OP_UNUSED_7A.c */
+HANDLE_OPCODE(OP_UNUSED_7A)
+OP_END
+
+/* File: c/OP_NEG_INT.c */
+HANDLE_UNOP(OP_NEG_INT, "neg-int", -, , )
+OP_END
+
+/* File: c/OP_NOT_INT.c */
+HANDLE_UNOP(OP_NOT_INT, "not-int", , ^ 0xffffffff, )
+OP_END
+
+/* File: c/OP_NEG_LONG.c */
+HANDLE_UNOP(OP_NEG_LONG, "neg-long", -, , _WIDE)
+OP_END
+
+/* File: c/OP_NOT_LONG.c */
+HANDLE_UNOP(OP_NOT_LONG, "not-long", , ^ 0xffffffffffffffffULL, _WIDE)
+OP_END
+
+/* File: c/OP_NEG_FLOAT.c */
+HANDLE_UNOP(OP_NEG_FLOAT, "neg-float", -, , _FLOAT)
+OP_END
+
+/* File: c/OP_NEG_DOUBLE.c */
+HANDLE_UNOP(OP_NEG_DOUBLE, "neg-double", -, , _DOUBLE)
+OP_END
+
+/* File: c/OP_INT_TO_LONG.c */
+HANDLE_NUMCONV(OP_INT_TO_LONG,          "int-to-long", _INT, _WIDE)
+OP_END
+
+/* File: c/OP_INT_TO_FLOAT.c */
+HANDLE_NUMCONV(OP_INT_TO_FLOAT,         "int-to-float", _INT, _FLOAT)
+OP_END
+
+/* File: c/OP_INT_TO_DOUBLE.c */
+HANDLE_NUMCONV(OP_INT_TO_DOUBLE,        "int-to-double", _INT, _DOUBLE)
+OP_END
+
+/* File: c/OP_LONG_TO_INT.c */
+HANDLE_NUMCONV(OP_LONG_TO_INT,          "long-to-int", _WIDE, _INT)
+OP_END
+
+/* File: c/OP_LONG_TO_FLOAT.c */
+HANDLE_NUMCONV(OP_LONG_TO_FLOAT,        "long-to-float", _WIDE, _FLOAT)
+OP_END
+
+/* File: c/OP_LONG_TO_DOUBLE.c */
+HANDLE_NUMCONV(OP_LONG_TO_DOUBLE,       "long-to-double", _WIDE, _DOUBLE)
+OP_END
+
+/* File: c/OP_FLOAT_TO_INT.c */
+HANDLE_FLOAT_TO_INT(OP_FLOAT_TO_INT,    "float-to-int",
+    float, _FLOAT, s4, _INT)
+OP_END
+
+/* File: c/OP_FLOAT_TO_LONG.c */
+HANDLE_FLOAT_TO_INT(OP_FLOAT_TO_LONG,   "float-to-long",
+    float, _FLOAT, s8, _WIDE)
+OP_END
+
+/* File: c/OP_FLOAT_TO_DOUBLE.c */
+HANDLE_NUMCONV(OP_FLOAT_TO_DOUBLE,      "float-to-double", _FLOAT, _DOUBLE)
+OP_END
+
+/* File: c/OP_DOUBLE_TO_INT.c */
+HANDLE_FLOAT_TO_INT(OP_DOUBLE_TO_INT,   "double-to-int",
+    double, _DOUBLE, s4, _INT)
+OP_END
+
+/* File: c/OP_DOUBLE_TO_LONG.c */
+HANDLE_FLOAT_TO_INT(OP_DOUBLE_TO_LONG,  "double-to-long",
+    double, _DOUBLE, s8, _WIDE)
+OP_END
+
+/* File: c/OP_DOUBLE_TO_FLOAT.c */
+HANDLE_NUMCONV(OP_DOUBLE_TO_FLOAT,      "double-to-float", _DOUBLE, _FLOAT)
+OP_END
+
+/* File: c/OP_INT_TO_BYTE.c */
+HANDLE_INT_TO_SMALL(OP_INT_TO_BYTE,     "byte", s1)
+OP_END
+
+/* File: c/OP_INT_TO_CHAR.c */
+HANDLE_INT_TO_SMALL(OP_INT_TO_CHAR,     "char", u2)
+OP_END
+
+/* File: c/OP_INT_TO_SHORT.c */
+HANDLE_INT_TO_SMALL(OP_INT_TO_SHORT,    "short", s2)    /* want sign bit */
+OP_END
+
+/* File: c/OP_ADD_INT.c */
+HANDLE_OP_X_INT(OP_ADD_INT, "add", +, 0)
+OP_END
+
+/* File: c/OP_SUB_INT.c */
+HANDLE_OP_X_INT(OP_SUB_INT, "sub", -, 0)
+OP_END
+
+/* File: c/OP_MUL_INT.c */
+HANDLE_OP_X_INT(OP_MUL_INT, "mul", *, 0)
+OP_END
+
+/* File: c/OP_DIV_INT.c */
+HANDLE_OP_X_INT(OP_DIV_INT, "div", /, 1)
+OP_END
+
+/* File: c/OP_REM_INT.c */
+HANDLE_OP_X_INT(OP_REM_INT, "rem", %, 2)
+OP_END
+
+/* File: c/OP_AND_INT.c */
+HANDLE_OP_X_INT(OP_AND_INT, "and", &, 0)
+OP_END
+
+/* File: c/OP_OR_INT.c */
+HANDLE_OP_X_INT(OP_OR_INT,  "or",  |, 0)
+OP_END
+
+/* File: c/OP_XOR_INT.c */
+HANDLE_OP_X_INT(OP_XOR_INT, "xor", ^, 0)
+OP_END
+
+/* File: c/OP_SHL_INT.c */
+HANDLE_OP_SHX_INT(OP_SHL_INT, "shl", (s4), <<)
+OP_END
+
+/* File: c/OP_SHR_INT.c */
+HANDLE_OP_SHX_INT(OP_SHR_INT, "shr", (s4), >>)
+OP_END
+
+/* File: c/OP_USHR_INT.c */
+HANDLE_OP_SHX_INT(OP_USHR_INT, "ushr", (u4), >>)
+OP_END
+
+/* File: c/OP_ADD_LONG.c */
+HANDLE_OP_X_LONG(OP_ADD_LONG, "add", +, 0)
+OP_END
+
+/* File: c/OP_SUB_LONG.c */
+HANDLE_OP_X_LONG(OP_SUB_LONG, "sub", -, 0)
+OP_END
+
+/* File: c/OP_MUL_LONG.c */
+HANDLE_OP_X_LONG(OP_MUL_LONG, "mul", *, 0)
+OP_END
+
+/* File: c/OP_DIV_LONG.c */
+HANDLE_OP_X_LONG(OP_DIV_LONG, "div", /, 1)
+OP_END
+
+/* File: c/OP_REM_LONG.c */
+HANDLE_OP_X_LONG(OP_REM_LONG, "rem", %, 2)
+OP_END
+
+/* File: c/OP_AND_LONG.c */
+HANDLE_OP_X_LONG(OP_AND_LONG, "and", &, 0)
+OP_END
+
+/* File: c/OP_OR_LONG.c */
+HANDLE_OP_X_LONG(OP_OR_LONG,  "or", |, 0)
+OP_END
+
+/* File: c/OP_XOR_LONG.c */
+HANDLE_OP_X_LONG(OP_XOR_LONG, "xor", ^, 0)
+OP_END
+
+/* File: c/OP_SHL_LONG.c */
+HANDLE_OP_SHX_LONG(OP_SHL_LONG, "shl", (s8), <<)
+OP_END
+
+/* File: c/OP_SHR_LONG.c */
+HANDLE_OP_SHX_LONG(OP_SHR_LONG, "shr", (s8), >>)
+OP_END
+
+/* File: c/OP_USHR_LONG.c */
+HANDLE_OP_SHX_LONG(OP_USHR_LONG, "ushr", (u8), >>)
+OP_END
+
+/* File: c/OP_ADD_FLOAT.c */
+HANDLE_OP_X_FLOAT(OP_ADD_FLOAT, "add", +)
+OP_END
+
+/* File: c/OP_SUB_FLOAT.c */
+HANDLE_OP_X_FLOAT(OP_SUB_FLOAT, "sub", -)
+OP_END
+
+/* File: c/OP_MUL_FLOAT.c */
+HANDLE_OP_X_FLOAT(OP_MUL_FLOAT, "mul", *)
+OP_END
+
+/* File: c/OP_DIV_FLOAT.c */
+HANDLE_OP_X_FLOAT(OP_DIV_FLOAT, "div", /)
+OP_END
+
+/* File: c/OP_REM_FLOAT.c */
+HANDLE_OPCODE(OP_REM_FLOAT /*vAA, vBB, vCC*/)
+    {
+        u2 srcRegs;
+        vdst = INST_AA(inst);
+        srcRegs = FETCH(1);
+        vsrc1 = srcRegs & 0xff;
+        vsrc2 = srcRegs >> 8;
+        ILOGV("|%s-float v%d,v%d,v%d", "mod", vdst, vsrc1, vsrc2);
+        SET_REGISTER_FLOAT(vdst,
+            fmodf(GET_REGISTER_FLOAT(vsrc1), GET_REGISTER_FLOAT(vsrc2)));
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_ADD_DOUBLE.c */
+HANDLE_OP_X_DOUBLE(OP_ADD_DOUBLE, "add", +)
+OP_END
+
+/* File: c/OP_SUB_DOUBLE.c */
+HANDLE_OP_X_DOUBLE(OP_SUB_DOUBLE, "sub", -)
+OP_END
+
+/* File: c/OP_MUL_DOUBLE.c */
+HANDLE_OP_X_DOUBLE(OP_MUL_DOUBLE, "mul", *)
+OP_END
+
+/* File: c/OP_DIV_DOUBLE.c */
+HANDLE_OP_X_DOUBLE(OP_DIV_DOUBLE, "div", /)
+OP_END
+
+/* File: c/OP_REM_DOUBLE.c */
+HANDLE_OPCODE(OP_REM_DOUBLE /*vAA, vBB, vCC*/)
+    {
+        u2 srcRegs;
+        vdst = INST_AA(inst);
+        srcRegs = FETCH(1);
+        vsrc1 = srcRegs & 0xff;
+        vsrc2 = srcRegs >> 8;
+        ILOGV("|%s-double v%d,v%d,v%d", "mod", vdst, vsrc1, vsrc2);
+        SET_REGISTER_DOUBLE(vdst,
+            fmod(GET_REGISTER_DOUBLE(vsrc1), GET_REGISTER_DOUBLE(vsrc2)));
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_ADD_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_ADD_INT_2ADDR, "add", +, 0)
+OP_END
+
+/* File: c/OP_SUB_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_SUB_INT_2ADDR, "sub", -, 0)
+OP_END
+
+/* File: c/OP_MUL_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_MUL_INT_2ADDR, "mul", *, 0)
+OP_END
+
+/* File: c/OP_DIV_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_DIV_INT_2ADDR, "div", /, 1)
+OP_END
+
+/* File: c/OP_REM_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_REM_INT_2ADDR, "rem", %, 2)
+OP_END
+
+/* File: c/OP_AND_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_AND_INT_2ADDR, "and", &, 0)
+OP_END
+
+/* File: c/OP_OR_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_OR_INT_2ADDR,  "or", |, 0)
+OP_END
+
+/* File: c/OP_XOR_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_XOR_INT_2ADDR, "xor", ^, 0)
+OP_END
+
+/* File: c/OP_SHL_INT_2ADDR.c */
+HANDLE_OP_SHX_INT_2ADDR(OP_SHL_INT_2ADDR, "shl", (s4), <<)
+OP_END
+
+/* File: c/OP_SHR_INT_2ADDR.c */
+HANDLE_OP_SHX_INT_2ADDR(OP_SHR_INT_2ADDR, "shr", (s4), >>)
+OP_END
+
+/* File: c/OP_USHR_INT_2ADDR.c */
+HANDLE_OP_SHX_INT_2ADDR(OP_USHR_INT_2ADDR, "ushr", (u4), >>)
+OP_END
+
+/* File: c/OP_ADD_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_ADD_LONG_2ADDR, "add", +, 0)
+OP_END
+
+/* File: c/OP_SUB_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_SUB_LONG_2ADDR, "sub", -, 0)
+OP_END
+
+/* File: c/OP_MUL_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_MUL_LONG_2ADDR, "mul", *, 0)
+OP_END
+
+/* File: c/OP_DIV_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_DIV_LONG_2ADDR, "div", /, 1)
+OP_END
+
+/* File: c/OP_REM_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_REM_LONG_2ADDR, "rem", %, 2)
+OP_END
+
+/* File: c/OP_AND_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_AND_LONG_2ADDR, "and", &, 0)
+OP_END
+
+/* File: c/OP_OR_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_OR_LONG_2ADDR,  "or", |, 0)
+OP_END
+
+/* File: c/OP_XOR_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_XOR_LONG_2ADDR, "xor", ^, 0)
+OP_END
+
+/* File: c/OP_SHL_LONG_2ADDR.c */
+HANDLE_OP_SHX_LONG_2ADDR(OP_SHL_LONG_2ADDR, "shl", (s8), <<)
+OP_END
+
+/* File: c/OP_SHR_LONG_2ADDR.c */
+HANDLE_OP_SHX_LONG_2ADDR(OP_SHR_LONG_2ADDR, "shr", (s8), >>)
+OP_END
+
+/* File: c/OP_USHR_LONG_2ADDR.c */
+HANDLE_OP_SHX_LONG_2ADDR(OP_USHR_LONG_2ADDR, "ushr", (u8), >>)
+OP_END
+
+/* File: c/OP_ADD_FLOAT_2ADDR.c */
+HANDLE_OP_X_FLOAT_2ADDR(OP_ADD_FLOAT_2ADDR, "add", +)
+OP_END
+
+/* File: c/OP_SUB_FLOAT_2ADDR.c */
+HANDLE_OP_X_FLOAT_2ADDR(OP_SUB_FLOAT_2ADDR, "sub", -)
+OP_END
+
+/* File: c/OP_MUL_FLOAT_2ADDR.c */
+HANDLE_OP_X_FLOAT_2ADDR(OP_MUL_FLOAT_2ADDR, "mul", *)
+OP_END
+
+/* File: c/OP_DIV_FLOAT_2ADDR.c */
+HANDLE_OP_X_FLOAT_2ADDR(OP_DIV_FLOAT_2ADDR, "div", /)
+OP_END
+
+/* File: c/OP_REM_FLOAT_2ADDR.c */
+HANDLE_OPCODE(OP_REM_FLOAT_2ADDR /*vA, vB*/)
+    vdst = INST_A(inst);
+    vsrc1 = INST_B(inst);
+    ILOGV("|%s-float-2addr v%d,v%d", "mod", vdst, vsrc1);
+    SET_REGISTER_FLOAT(vdst,
+        fmodf(GET_REGISTER_FLOAT(vdst), GET_REGISTER_FLOAT(vsrc1)));
+    FINISH(1);
+OP_END
+
+/* File: c/OP_ADD_DOUBLE_2ADDR.c */
+HANDLE_OP_X_DOUBLE_2ADDR(OP_ADD_DOUBLE_2ADDR, "add", +)
+OP_END
+
+/* File: c/OP_SUB_DOUBLE_2ADDR.c */
+HANDLE_OP_X_DOUBLE_2ADDR(OP_SUB_DOUBLE_2ADDR, "sub", -)
+OP_END
+
+/* File: c/OP_MUL_DOUBLE_2ADDR.c */
+HANDLE_OP_X_DOUBLE_2ADDR(OP_MUL_DOUBLE_2ADDR, "mul", *)
+OP_END
+
+/* File: c/OP_DIV_DOUBLE_2ADDR.c */
+HANDLE_OP_X_DOUBLE_2ADDR(OP_DIV_DOUBLE_2ADDR, "div", /)
+OP_END
+
+/* File: c/OP_REM_DOUBLE_2ADDR.c */
+HANDLE_OPCODE(OP_REM_DOUBLE_2ADDR /*vA, vB*/)
+    vdst = INST_A(inst);
+    vsrc1 = INST_B(inst);
+    ILOGV("|%s-double-2addr v%d,v%d", "mod", vdst, vsrc1);
+    SET_REGISTER_DOUBLE(vdst,
+        fmod(GET_REGISTER_DOUBLE(vdst), GET_REGISTER_DOUBLE(vsrc1)));
+    FINISH(1);
+OP_END
+
+/* File: c/OP_ADD_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_ADD_INT_LIT16, "add", +, 0)
+OP_END
+
+/* File: c/OP_RSUB_INT.c */
+HANDLE_OPCODE(OP_RSUB_INT /*vA, vB, #+CCCC*/)
+    {
+        vdst = INST_A(inst);
+        vsrc1 = INST_B(inst);
+        vsrc2 = FETCH(1);
+        ILOGV("|rsub-int v%d,v%d,#+0x%04x", vdst, vsrc1, vsrc2);
+        SET_REGISTER(vdst, (s2) vsrc2 - (s4) GET_REGISTER(vsrc1));
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_MUL_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_MUL_INT_LIT16, "mul", *, 0)
+OP_END
+
+/* File: c/OP_DIV_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_DIV_INT_LIT16, "div", /, 1)
+OP_END
+
+/* File: c/OP_REM_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_REM_INT_LIT16, "rem", %, 2)
+OP_END
+
+/* File: c/OP_AND_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_AND_INT_LIT16, "and", &, 0)
+OP_END
+
+/* File: c/OP_OR_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_OR_INT_LIT16,  "or",  |, 0)
+OP_END
+
+/* File: c/OP_XOR_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_XOR_INT_LIT16, "xor", ^, 0)
+OP_END
+
+/* File: c/OP_ADD_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_ADD_INT_LIT8,   "add", +, 0)
+OP_END
+
+/* File: c/OP_RSUB_INT_LIT8.c */
+HANDLE_OPCODE(OP_RSUB_INT_LIT8 /*vAA, vBB, #+CC*/)
+    {
+        u2 litInfo;
+        vdst = INST_AA(inst);
+        litInfo = FETCH(1);
+        vsrc1 = litInfo & 0xff;
+        vsrc2 = litInfo >> 8;
+        ILOGV("|%s-int/lit8 v%d,v%d,#+0x%02x", "rsub", vdst, vsrc1, vsrc2);
+        SET_REGISTER(vdst, (s1) vsrc2 - (s4) GET_REGISTER(vsrc1));
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_MUL_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_MUL_INT_LIT8,   "mul", *, 0)
+OP_END
+
+/* File: c/OP_DIV_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_DIV_INT_LIT8,   "div", /, 1)
+OP_END
+
+/* File: c/OP_REM_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_REM_INT_LIT8,   "rem", %, 2)
+OP_END
+
+/* File: c/OP_AND_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_AND_INT_LIT8,   "and", &, 0)
+OP_END
+
+/* File: c/OP_OR_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_OR_INT_LIT8,    "or",  |, 0)
+OP_END
+
+/* File: c/OP_XOR_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_XOR_INT_LIT8,   "xor", ^, 0)
+OP_END
+
+/* File: c/OP_SHL_INT_LIT8.c */
+HANDLE_OP_SHX_INT_LIT8(OP_SHL_INT_LIT8,   "shl", (s4), <<)
+OP_END
+
+/* File: c/OP_SHR_INT_LIT8.c */
+HANDLE_OP_SHX_INT_LIT8(OP_SHR_INT_LIT8,   "shr", (s4), >>)
+OP_END
+
+/* File: c/OP_USHR_INT_LIT8.c */
+HANDLE_OP_SHX_INT_LIT8(OP_USHR_INT_LIT8,  "ushr", (u4), >>)
+OP_END
+
+/* File: c/OP_UNUSED_E3.c */
+HANDLE_OPCODE(OP_UNUSED_E3)
+OP_END
+
+/* File: c/OP_UNUSED_E4.c */
+HANDLE_OPCODE(OP_UNUSED_E4)
+OP_END
+
+/* File: c/OP_UNUSED_E5.c */
+HANDLE_OPCODE(OP_UNUSED_E5)
+OP_END
+
+/* File: c/OP_UNUSED_E6.c */
+HANDLE_OPCODE(OP_UNUSED_E6)
+OP_END
+
+/* File: c/OP_UNUSED_E7.c */
+HANDLE_OPCODE(OP_UNUSED_E7)
+OP_END
+
+/* File: c/OP_UNUSED_E8.c */
+HANDLE_OPCODE(OP_UNUSED_E8)
+OP_END
+
+/* File: c/OP_UNUSED_E9.c */
+HANDLE_OPCODE(OP_UNUSED_E9)
+OP_END
+
+/* File: c/OP_UNUSED_EA.c */
+HANDLE_OPCODE(OP_UNUSED_EA)
+OP_END
+
+/* File: c/OP_UNUSED_EB.c */
+HANDLE_OPCODE(OP_UNUSED_EB)
+OP_END
+
+/* File: c/OP_UNUSED_EC.c */
+HANDLE_OPCODE(OP_UNUSED_EC)
+OP_END
+
+/* File: c/OP_UNUSED_ED.c */
+HANDLE_OPCODE(OP_UNUSED_ED)
+OP_END
+
+/* File: c/OP_EXECUTE_INLINE.c */
+HANDLE_OPCODE(OP_EXECUTE_INLINE /*vB, {vD, vE, vF, vG}, inline@CCCC*/)
+    {
+        /*
+         * This has the same form as other method calls, but we ignore
+         * the 5th argument (vA).  This is chiefly because the first four
+         * arguments to a function on ARM are in registers.
+         *
+         * We only set the arguments that are actually used, leaving
+         * the rest uninitialized.  We're assuming that, if the method
+         * needs them, they'll be specified in the call.
+         *
+         * This annoys gcc when optimizations are enabled, causing a
+         * "may be used uninitialized" warning.  We can quiet the warnings
+         * for a slight penalty (5%: 373ns vs. 393ns on empty method).  Note
+         * that valgrind is perfectly happy with this arrangement, because
+         * the uninitialiezd values are never actually used.
+         */
+        u4 arg0, arg1, arg2, arg3;
+        //arg0 = arg1 = arg2 = arg3 = 0;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_B(inst);       /* #of args */
+        ref = FETCH(1);             /* inline call "ref" */
+        vdst = FETCH(2);            /* 0-4 register indices */
+        ILOGV("|execute-inline args=%d @%d {regs=0x%04x}",
+            vsrc1, ref, vdst);
+
+        assert((vdst >> 16) == 0);  // 16-bit type -or- high 16 bits clear
+        assert(vsrc1 <= 4);
+
+        switch (vsrc1) {
+        case 4:
+            arg3 = GET_REGISTER(vdst >> 12);
+            /* fall through */
+        case 3:
+            arg2 = GET_REGISTER((vdst & 0x0f00) >> 8);
+            /* fall through */
+        case 2:
+            arg1 = GET_REGISTER((vdst & 0x00f0) >> 4);
+            /* fall through */
+        case 1:
+            arg0 = GET_REGISTER(vdst & 0x0f);
+            /* fall through */
+        default:        // case 0
+            ;
+        }
+
+#if INTERP_TYPE == INTERP_DBG
+        if (!dvmPerformInlineOp4Dbg(arg0, arg1, arg2, arg3, &retval, ref))
+            GOTO_exceptionThrown();
+#else
+        if (!dvmPerformInlineOp4Std(arg0, arg1, arg2, arg3, &retval, ref))
+            GOTO_exceptionThrown();
+#endif
+    }
+    FINISH(3);
+OP_END
+
+/* File: c/OP_UNUSED_EF.c */
+HANDLE_OPCODE(OP_UNUSED_EF)
+OP_END
+
+/* File: c/OP_INVOKE_DIRECT_EMPTY.c */
+HANDLE_OPCODE(OP_INVOKE_DIRECT_EMPTY /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+#if INTERP_TYPE != INTERP_DBG
+    //LOGI("Ignoring empty\n");
+    FINISH(3);
+#else
+    if (!gDvm.debuggerActive) {
+        //LOGI("Skipping empty\n");
+        FINISH(3);      // don't want it to show up in profiler output
+    } else {
+        //LOGI("Running empty\n");
+        /* fall through to OP_INVOKE_DIRECT */
+        GOTO_invoke(invokeDirect, false);
+    }
+#endif
+OP_END
+
+/* File: c/OP_UNUSED_F1.c */
+HANDLE_OPCODE(OP_UNUSED_F1)
+OP_END
+
+/* File: c/OP_IGET_QUICK.c */
+HANDLE_IGET_X_QUICK(OP_IGET_QUICK,          "", Int, )
+OP_END
+
+/* File: c/OP_IGET_WIDE_QUICK.c */
+HANDLE_IGET_X_QUICK(OP_IGET_WIDE_QUICK,     "-wide", Long, _WIDE)
+OP_END
+
+/* File: c/OP_IGET_OBJECT_QUICK.c */
+HANDLE_IGET_X_QUICK(OP_IGET_OBJECT_QUICK,   "-object", Object, _AS_OBJECT)
+OP_END
+
+/* File: c/OP_IPUT_QUICK.c */
+HANDLE_IPUT_X_QUICK(OP_IPUT_QUICK,          "", Int, )
+OP_END
+
+/* File: c/OP_IPUT_WIDE_QUICK.c */
+HANDLE_IPUT_X_QUICK(OP_IPUT_WIDE_QUICK,     "-wide", Long, _WIDE)
+OP_END
+
+/* File: c/OP_IPUT_OBJECT_QUICK.c */
+HANDLE_IPUT_X_QUICK(OP_IPUT_OBJECT_QUICK,   "-object", Object, _AS_OBJECT)
+OP_END
+
+/* File: c/OP_INVOKE_VIRTUAL_QUICK.c */
+HANDLE_OPCODE(OP_INVOKE_VIRTUAL_QUICK /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeVirtualQuick, false);
+OP_END
+
+/* File: c/OP_INVOKE_VIRTUAL_QUICK_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_VIRTUAL_QUICK_RANGE/*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeVirtualQuick, true);
+OP_END
+
+/* File: c/OP_INVOKE_SUPER_QUICK.c */
+HANDLE_OPCODE(OP_INVOKE_SUPER_QUICK /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeSuperQuick, false);
+OP_END
+
+/* File: c/OP_INVOKE_SUPER_QUICK_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_SUPER_QUICK_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeSuperQuick, true);
+OP_END
+
+/* File: c/OP_UNUSED_FC.c */
+HANDLE_OPCODE(OP_UNUSED_FC)
+OP_END
+
+/* File: c/OP_UNUSED_FD.c */
+HANDLE_OPCODE(OP_UNUSED_FD)
+OP_END
+
+/* File: c/OP_UNUSED_FE.c */
+HANDLE_OPCODE(OP_UNUSED_FE)
+OP_END
+
+/* File: c/OP_UNUSED_FF.c */
+HANDLE_OPCODE(OP_UNUSED_FF)
+    /*
+     * In portable interp, most unused opcodes will fall through to here.
+     */
+    LOGE("unknown opcode 0x%02x\n", INST_INST(inst));
+    dvmAbort();
+    FINISH(1);
+OP_END
+
+/* File: c/gotoTargets.c */
+/*
+ * C footer.  This has some common code shared by the various targets.
+ */
+
+/*
+ * Everything from here on is a "goto target".  In the basic interpreter
+ * we jump into these targets and then jump directly to the handler for
+ * next instruction.  Here, these are subroutines that return to the caller.
+ */
+
+GOTO_TARGET(filledNewArray, bool methodCallRange)
+    {
+        ClassObject* arrayClass;
+        ArrayObject* newArray;
+        u4* contents;
+        char typeCh;
+        int i;
+        u4 arg5;
+
+        EXPORT_PC();
+
+        ref = FETCH(1);             /* class ref */
+        vdst = FETCH(2);            /* first 4 regs -or- range base */
+
+        if (methodCallRange) {
+            vsrc1 = INST_AA(inst);  /* #of elements */
+            arg5 = -1;              /* silence compiler warning */
+            ILOGV("|filled-new-array-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+        } else {
+            arg5 = INST_A(inst);
+            vsrc1 = INST_B(inst);   /* #of elements */
+            ILOGV("|filled-new-array args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1, ref, vdst, arg5);
+        }
+
+        /*
+         * Resolve the array class.
+         */
+        arrayClass = dvmDexGetResolvedClass(methodClassDex, ref);
+        if (arrayClass == NULL) {
+            arrayClass = dvmResolveClass(curMethod->clazz, ref, false);
+            if (arrayClass == NULL)
+                GOTO_exceptionThrown();
+        }
+        /*
+        if (!dvmIsArrayClass(arrayClass)) {
+            dvmThrowException("Ljava/lang/RuntimeError;",
+                "filled-new-array needs array class");
+            GOTO_exceptionThrown();
+        }
+        */
+        /* verifier guarantees this is an array class */
+        assert(dvmIsArrayClass(arrayClass));
+        assert(dvmIsClassInitialized(arrayClass));
+
+        /*
+         * Create an array of the specified type.
+         */
+        LOGVV("+++ filled-new-array type is '%s'\n", arrayClass->descriptor);
+        typeCh = arrayClass->descriptor[1];
+        if (typeCh == 'D' || typeCh == 'J') {
+            /* category 2 primitives not allowed */
+            dvmThrowException("Ljava/lang/RuntimeError;",
+                "bad filled array req");
+            GOTO_exceptionThrown();
+        } else if (typeCh != 'L' && typeCh != '[' && typeCh != 'I') {
+            /* TODO: requires multiple "fill in" loops with different widths */
+            LOGE("non-int primitives not implemented\n");
+            dvmThrowException("Ljava/lang/InternalError;",
+                "filled-new-array not implemented for anything but 'int'");
+            GOTO_exceptionThrown();
+        }
+
+        newArray = dvmAllocArrayByClass(arrayClass, vsrc1, ALLOC_DONT_TRACK);
+        if (newArray == NULL)
+            GOTO_exceptionThrown();
+
+        /*
+         * Fill in the elements.  It's legal for vsrc1 to be zero.
+         */
+        contents = (u4*) newArray->contents;
+        if (methodCallRange) {
+            for (i = 0; i < vsrc1; i++)
+                contents[i] = GET_REGISTER(vdst+i);
+        } else {
+            assert(vsrc1 <= 5);
+            if (vsrc1 == 5) {
+                contents[4] = GET_REGISTER(arg5);
+                vsrc1--;
+            }
+            for (i = 0; i < vsrc1; i++) {
+                contents[i] = GET_REGISTER(vdst & 0x0f);
+                vdst >>= 4;
+            }
+        }
+
+        retval.l = newArray;
+    }
+    FINISH(3);
+GOTO_TARGET_END
+
+
+GOTO_TARGET(invokeVirtual, bool methodCallRange)
+    {
+        Method* baseMethod;
+        Object* thisPtr;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* method ref */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        /*
+         * The object against which we are executing a method is always
+         * in the first argument.
+         */
+        if (methodCallRange) {
+            assert(vsrc1 > 0);
+            ILOGV("|invoke-virtual-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisPtr = (Object*) GET_REGISTER(vdst);
+        } else {
+            assert((vsrc1>>4) > 0);
+            ILOGV("|invoke-virtual args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisPtr = (Object*) GET_REGISTER(vdst & 0x0f);
+        }
+
+        if (!checkForNull(thisPtr))
+            GOTO_exceptionThrown();
+
+        /*
+         * Resolve the method.  This is the correct method for the static
+         * type of the object.  We also verify access permissions here.
+         */
+        baseMethod = dvmDexGetResolvedMethod(methodClassDex, ref);
+        if (baseMethod == NULL) {
+            baseMethod = dvmResolveMethod(curMethod->clazz, ref,METHOD_VIRTUAL);
+            if (baseMethod == NULL) {
+                ILOGV("+ unknown method or access denied\n");
+                GOTO_exceptionThrown();
+            }
+        }
+
+        /*
+         * Combine the object we found with the vtable offset in the
+         * method.
+         */
+        assert(baseMethod->methodIndex < thisPtr->clazz->vtableCount);
+        methodToCall = thisPtr->clazz->vtable[baseMethod->methodIndex];
+
+#if 0
+        if (dvmIsAbstractMethod(methodToCall)) {
+            /*
+             * This can happen if you create two classes, Base and Sub, where
+             * Sub is a sub-class of Base.  Declare a protected abstract
+             * method foo() in Base, and invoke foo() from a method in Base.
+             * Base is an "abstract base class" and is never instantiated
+             * directly.  Now, Override foo() in Sub, and use Sub.  This
+             * Works fine unless Sub stops providing an implementation of
+             * the method.
+             */
+            dvmThrowException("Ljava/lang/AbstractMethodError;",
+                "abstract method not implemented");
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(!dvmIsAbstractMethod(methodToCall) ||
+            methodToCall->nativeFunc != NULL);
+#endif
+
+        LOGVV("+++ base=%s.%s virtual[%d]=%s.%s\n",
+            baseMethod->clazz->descriptor, baseMethod->name,
+            (u4) baseMethod->methodIndex,
+            methodToCall->clazz->descriptor, methodToCall->name);
+        assert(methodToCall != NULL);
+
+#if 0
+        if (vsrc1 != methodToCall->insSize) {
+            LOGW("WRONG METHOD: base=%s.%s virtual[%d]=%s.%s\n",
+                baseMethod->clazz->descriptor, baseMethod->name,
+                (u4) baseMethod->methodIndex,
+                methodToCall->clazz->descriptor, methodToCall->name);
+            //dvmDumpClass(baseMethod->clazz);
+            //dvmDumpClass(methodToCall->clazz);
+            dvmDumpAllClasses(0);
+        }
+#endif
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeSuper, bool methodCallRange)
+    {
+        Method* baseMethod;
+        u2 thisReg;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* method ref */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        if (methodCallRange) {
+            ILOGV("|invoke-super-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisReg = vdst;
+        } else {
+            ILOGV("|invoke-super args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisReg = vdst & 0x0f;
+        }
+        /* impossible in well-formed code, but we must check nevertheless */
+        if (!checkForNull((Object*) GET_REGISTER(thisReg)))
+            GOTO_exceptionThrown();
+
+        /*
+         * Resolve the method.  This is the correct method for the static
+         * type of the object.  We also verify access permissions here.
+         * The first arg to dvmResolveMethod() is just the referring class
+         * (used for class loaders and such), so we don't want to pass
+         * the superclass into the resolution call.
+         */
+        baseMethod = dvmDexGetResolvedMethod(methodClassDex, ref);
+        if (baseMethod == NULL) {
+            baseMethod = dvmResolveMethod(curMethod->clazz, ref,METHOD_VIRTUAL);
+            if (baseMethod == NULL) {
+                ILOGV("+ unknown method or access denied\n");
+                GOTO_exceptionThrown();
+            }
+        }
+
+        /*
+         * Combine the object we found with the vtable offset in the
+         * method's class.
+         *
+         * We're using the current method's class' superclass, not the
+         * superclass of "this".  This is because we might be executing
+         * in a method inherited from a superclass, and we want to run
+         * in that class' superclass.
+         */
+        if (baseMethod->methodIndex >= curMethod->clazz->super->vtableCount) {
+            /*
+             * Method does not exist in the superclass.  Could happen if
+             * superclass gets updated.
+             */
+            dvmThrowException("Ljava/lang/NoSuchMethodError;",
+                baseMethod->name);
+            GOTO_exceptionThrown();
+        }
+        methodToCall = curMethod->clazz->super->vtable[baseMethod->methodIndex];
+#if 0
+        if (dvmIsAbstractMethod(methodToCall)) {
+            dvmThrowException("Ljava/lang/AbstractMethodError;",
+                "abstract method not implemented");
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(!dvmIsAbstractMethod(methodToCall) ||
+            methodToCall->nativeFunc != NULL);
+#endif
+        LOGVV("+++ base=%s.%s super-virtual=%s.%s\n",
+            baseMethod->clazz->descriptor, baseMethod->name,
+            methodToCall->clazz->descriptor, methodToCall->name);
+        assert(methodToCall != NULL);
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeInterface, bool methodCallRange)
+    {
+        Object* thisPtr;
+        ClassObject* thisClass;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* method ref */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        /*
+         * The object against which we are executing a method is always
+         * in the first argument.
+         */
+        if (methodCallRange) {
+            assert(vsrc1 > 0);
+            ILOGV("|invoke-interface-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisPtr = (Object*) GET_REGISTER(vdst);
+        } else {
+            assert((vsrc1>>4) > 0);
+            ILOGV("|invoke-interface args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisPtr = (Object*) GET_REGISTER(vdst & 0x0f);
+        }
+        if (!checkForNull(thisPtr))
+            GOTO_exceptionThrown();
+
+        thisClass = thisPtr->clazz;
+
+        /*
+         * Given a class and a method index, find the Method* with the
+         * actual code we want to execute.
+         */
+        methodToCall = dvmFindInterfaceMethodInCache(thisClass, ref, curMethod,
+                        methodClassDex);
+        if (methodToCall == NULL) {
+            assert(dvmCheckException(self));
+            GOTO_exceptionThrown();
+        }
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeDirect, bool methodCallRange)
+    {
+        u2 thisReg;
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* method ref */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        EXPORT_PC();
+
+        if (methodCallRange) {
+            ILOGV("|invoke-direct-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisReg = vdst;
+        } else {
+            ILOGV("|invoke-direct args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisReg = vdst & 0x0f;
+        }
+        if (!checkForNull((Object*) GET_REGISTER(thisReg)))
+            GOTO_exceptionThrown();
+
+        methodToCall = dvmDexGetResolvedMethod(methodClassDex, ref);
+        if (methodToCall == NULL) {
+            methodToCall = dvmResolveMethod(curMethod->clazz, ref,
+                            METHOD_DIRECT);
+            if (methodToCall == NULL) {
+                ILOGV("+ unknown direct method\n");     // should be impossible
+                GOTO_exceptionThrown();
+            }
+        }
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeStatic, bool methodCallRange)
+    vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+    ref = FETCH(1);             /* method ref */
+    vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+    EXPORT_PC();
+
+    if (methodCallRange)
+        ILOGV("|invoke-static-range args=%d @0x%04x {regs=v%d-v%d}",
+            vsrc1, ref, vdst, vdst+vsrc1-1);
+    else
+        ILOGV("|invoke-static args=%d @0x%04x {regs=0x%04x %x}",
+            vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+
+    methodToCall = dvmDexGetResolvedMethod(methodClassDex, ref);
+    if (methodToCall == NULL) {
+        methodToCall = dvmResolveMethod(curMethod->clazz, ref, METHOD_STATIC);
+        if (methodToCall == NULL) {
+            ILOGV("+ unknown method\n");
+            GOTO_exceptionThrown();
+        }
+    }
+    GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeVirtualQuick, bool methodCallRange)
+    {
+        Object* thisPtr;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* vtable index */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        /*
+         * The object against which we are executing a method is always
+         * in the first argument.
+         */
+        if (methodCallRange) {
+            assert(vsrc1 > 0);
+            ILOGV("|invoke-virtual-quick-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisPtr = (Object*) GET_REGISTER(vdst);
+        } else {
+            assert((vsrc1>>4) > 0);
+            ILOGV("|invoke-virtual-quick args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisPtr = (Object*) GET_REGISTER(vdst & 0x0f);
+        }
+
+        if (!checkForNull(thisPtr))
+            GOTO_exceptionThrown();
+
+        /*
+         * Combine the object we found with the vtable offset in the
+         * method.
+         */
+        assert(ref < thisPtr->clazz->vtableCount);
+        methodToCall = thisPtr->clazz->vtable[ref];
+
+#if 0
+        if (dvmIsAbstractMethod(methodToCall)) {
+            dvmThrowException("Ljava/lang/AbstractMethodError;",
+                "abstract method not implemented");
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(!dvmIsAbstractMethod(methodToCall) ||
+            methodToCall->nativeFunc != NULL);
+#endif
+
+        LOGVV("+++ virtual[%d]=%s.%s\n",
+            ref, methodToCall->clazz->descriptor, methodToCall->name);
+        assert(methodToCall != NULL);
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeSuperQuick, bool methodCallRange)
+    {
+        u2 thisReg;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* vtable index */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        if (methodCallRange) {
+            ILOGV("|invoke-super-quick-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisReg = vdst;
+        } else {
+            ILOGV("|invoke-super-quick args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisReg = vdst & 0x0f;
+        }
+        /* impossible in well-formed code, but we must check nevertheless */
+        if (!checkForNull((Object*) GET_REGISTER(thisReg)))
+            GOTO_exceptionThrown();
+
+#if 0   /* impossible in optimized + verified code */
+        if (ref >= curMethod->clazz->super->vtableCount) {
+            dvmThrowException("Ljava/lang/NoSuchMethodError;", NULL);
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(ref < curMethod->clazz->super->vtableCount);
+#endif
+
+        /*
+         * Combine the object we found with the vtable offset in the
+         * method's class.
+         *
+         * We're using the current method's class' superclass, not the
+         * superclass of "this".  This is because we might be executing
+         * in a method inherited from a superclass, and we want to run
+         * in the method's class' superclass.
+         */
+        methodToCall = curMethod->clazz->super->vtable[ref];
+
+#if 0
+        if (dvmIsAbstractMethod(methodToCall)) {
+            dvmThrowException("Ljava/lang/AbstractMethodError;",
+                "abstract method not implemented");
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(!dvmIsAbstractMethod(methodToCall) ||
+            methodToCall->nativeFunc != NULL);
+#endif
+        LOGVV("+++ super-virtual[%d]=%s.%s\n",
+            ref, methodToCall->clazz->descriptor, methodToCall->name);
+        assert(methodToCall != NULL);
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+
+
+    /*
+     * General handling for return-void, return, and return-wide.  Put the
+     * return value in "retval" before jumping here.
+     */
+GOTO_TARGET(returnFromMethod)
+    {
+        StackSaveArea* saveArea;
+
+        /*
+         * We must do this BEFORE we pop the previous stack frame off, so
+         * that the GC can see the return value (if any) in the local vars.
+         *
+         * Since this is now an interpreter switch point, we must do it before
+         * we do anything at all.
+         */
+        PERIODIC_CHECKS(kInterpEntryReturn, 0);
+
+        ILOGV("> retval=0x%llx (leaving %s.%s %s)",
+            retval.j, curMethod->clazz->descriptor, curMethod->name,
+            curMethod->signature);
+        //DUMP_REGS(curMethod, fp);
+
+        saveArea = SAVEAREA_FROM_FP(fp);
+
+#ifdef EASY_GDB
+        debugSaveArea = saveArea;
+#endif
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_PROFILER)
+        TRACE_METHOD_EXIT(self, curMethod);
+#endif
+
+        /* back up to previous frame and see if we hit a break */
+        fp = saveArea->prevFrame;
+        assert(fp != NULL);
+        if (dvmIsBreakFrame(fp)) {
+            /* bail without popping the method frame from stack */
+            LOGVV("+++ returned into break frame\n");
+            GOTO_bail();
+        }
+
+        /* update thread FP, and reset local variables */
+        self->curFrame = fp;
+        curMethod = SAVEAREA_FROM_FP(fp)->method;
+        //methodClass = curMethod->clazz;
+        methodClassDex = curMethod->clazz->pDvmDex;
+        pc = saveArea->savedPc;
+        ILOGD("> (return to %s.%s %s)", curMethod->clazz->descriptor,
+            curMethod->name, curMethod->signature);
+
+        /* use FINISH on the caller's invoke instruction */
+        //u2 invokeInstr = INST_INST(FETCH(0));
+        if (true /*invokeInstr >= OP_INVOKE_VIRTUAL &&
+            invokeInstr <= OP_INVOKE_INTERFACE*/)
+        {
+            FINISH(3);
+        } else {
+            //LOGE("Unknown invoke instr %02x at %d\n",
+            //    invokeInstr, (int) (pc - curMethod->insns));
+            assert(false);
+        }
+    }
+GOTO_TARGET_END
+
+
+    /*
+     * Jump here when the code throws an exception.
+     *
+     * By the time we get here, the Throwable has been created and the stack
+     * trace has been saved off.
+     */
+GOTO_TARGET(exceptionThrown)
+    {
+        Object* exception;
+        int catchRelPc;
+
+        /*
+         * Since this is now an interpreter switch point, we must do it before
+         * we do anything at all.
+         */
+        PERIODIC_CHECKS(kInterpEntryThrow, 0);
+
+        /*
+         * We save off the exception and clear the exception status.  While
+         * processing the exception we might need to load some Throwable
+         * classes, and we don't want class loader exceptions to get
+         * confused with this one.
+         */
+        assert(dvmCheckException(self));
+        exception = dvmGetException(self);
+        dvmAddTrackedAlloc(exception, self);
+        dvmClearException(self);
+
+        LOGV("Handling exception %s at %s:%d\n",
+            exception->clazz->descriptor, curMethod->name,
+            dvmLineNumFromPC(curMethod, pc - curMethod->insns));
+
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_DEBUGGER)
+        /*
+         * Tell the debugger about it.
+         *
+         * TODO: if the exception was thrown by interpreted code, control
+         * fell through native, and then back to us, we will report the
+         * exception at the point of the throw and again here.  We can avoid
+         * this by not reporting exceptions when we jump here directly from
+         * the native call code above, but then we won't report exceptions
+         * that were thrown *from* the JNI code (as opposed to *through* it).
+         *
+         * The correct solution is probably to ignore from-native exceptions
+         * here, and have the JNI exception code do the reporting to the
+         * debugger.
+         */
+        if (gDvm.debuggerActive) {
+            void* catchFrame;
+            catchRelPc = dvmFindCatchBlock(self, pc - curMethod->insns,
+                        exception, true, &catchFrame);
+            dvmDbgPostException(fp, pc - curMethod->insns, catchFrame,
+                catchRelPc, exception);
+        }
+#endif
+
+        /*
+         * We need to unroll to the catch block or the nearest "break"
+         * frame.
+         *
+         * A break frame could indicate that we have reached an intermediate
+         * native call, or have gone off the top of the stack and the thread
+         * needs to exit.  Either way, we return from here, leaving the
+         * exception raised.
+         *
+         * If we do find a catch block, we want to transfer execution to
+         * that point.
+         */
+        catchRelPc = dvmFindCatchBlock(self, pc - curMethod->insns,
+                    exception, false, (void*)&fp);
+
+        /*
+         * Restore the stack bounds after an overflow.  This isn't going to
+         * be correct in all circumstances, e.g. if JNI code devours the
+         * exception this won't happen until some other exception gets
+         * thrown.  If the code keeps pushing the stack bounds we'll end
+         * up aborting the VM.
+         *
+         * Note we want to do this *after* the call to dvmFindCatchBlock,
+         * because that may need extra stack space to resolve exception
+         * classes (e.g. through a class loader).
+         */
+        if (self->stackOverflowed)
+            dvmCleanupStackOverflow(self);
+
+        if (catchRelPc < 0) {
+            /* falling through to JNI code or off the bottom of the stack */
+#if DVM_SHOW_EXCEPTION >= 2
+            LOGD("Exception %s from %s:%d not caught locally\n",
+                exception->clazz->descriptor, dvmGetMethodSourceFile(curMethod),
+                dvmLineNumFromPC(curMethod, pc - curMethod->insns));
+#endif
+            dvmSetException(self, exception);
+            dvmReleaseTrackedAlloc(exception, self);
+            GOTO_bail();
+        }
+
+#if DVM_SHOW_EXCEPTION >= 3
+        {
+            const Method* catchMethod = SAVEAREA_FROM_FP(fp)->method;
+            LOGD("Exception %s thrown from %s:%d to %s:%d\n",
+                exception->clazz->descriptor, dvmGetMethodSourceFile(curMethod),
+                dvmLineNumFromPC(curMethod, pc - curMethod->insns),
+                dvmGetMethodSourceFile(catchMethod),
+                dvmLineNumFromPC(catchMethod, catchRelPc));
+        }
+#endif
+
+        /*
+         * Adjust local variables to match self->curFrame and the
+         * updated PC.
+         */
+        //fp = (u4*) self->curFrame;
+        curMethod = SAVEAREA_FROM_FP(fp)->method;
+        //methodClass = curMethod->clazz;
+        methodClassDex = curMethod->clazz->pDvmDex;
+        pc = curMethod->insns + catchRelPc;
+        ILOGV("> pc <-- %s.%s %s", curMethod->clazz->descriptor,
+            curMethod->name, curMethod->signature);
+        DUMP_REGS(curMethod, fp, false);            // show all regs
+
+        /*
+         * Restore the exception if the handler wants it.
+         *
+         * The Dalvik spec mandates that, if an exception handler wants to
+         * do something with the exception, the first instruction executed
+         * must be "move-exception".  We can pass the exception along
+         * through the thread struct, and let the move-exception instruction
+         * clear it for us.
+         *
+         * If the handler doesn't call move-exception, we don't want to
+         * finish here with an exception still pending.
+         */
+        if (INST_INST(FETCH(0)) == OP_MOVE_EXCEPTION)
+            dvmSetException(self, exception);
+
+        dvmReleaseTrackedAlloc(exception, self);
+        FINISH(0);
+    }
+GOTO_TARGET_END
+
+
+    /*
+     * General handling for invoke-{virtual,super,direct,static,interface},
+     * including "quick" variants.
+     *
+     * Set "methodToCall" to the Method we're calling, and "methodCallRange"
+     * depending on whether this is a "/range" instruction.
+     *
+     * For a range call:
+     *  "vsrc1" holds the argument count (8 bits)
+     *  "vdst" holds the first argument in the range
+     * For a non-range call:
+     *  "vsrc1" holds the argument count (4 bits) and the 5th argument index
+     *  "vdst" holds four 4-bit register indices
+     *
+     * The caller must EXPORT_PC before jumping here, because any method
+     * call can throw a stack overflow exception.
+     */
+GOTO_TARGET(invokeMethod, bool methodCallRange, const Method* _methodToCall,
+    u2 count, u2 regs)
+    {
+        STUB_HACK(vsrc1 = count; vdst = regs; methodToCall = _methodToCall;);
+
+        //printf("range=%d call=%p count=%d regs=0x%04x\n",
+        //    methodCallRange, methodToCall, count, regs);
+        //printf(" --> %s.%s %s\n", methodToCall->clazz->descriptor,
+        //    methodToCall->name, methodToCall->signature);
+
+        u4* outs;
+        int i;
+
+        /*
+         * Copy args.  This may corrupt vsrc1/vdst.
+         */
+        if (methodCallRange) {
+            // could use memcpy or a "Duff's device"; most functions have
+            // so few args it won't matter much
+            assert(vsrc1 <= curMethod->outsSize);
+            assert(vsrc1 == methodToCall->insSize);
+            outs = OUTS_FROM_FP(fp, vsrc1);
+            for (i = 0; i < vsrc1; i++)
+                outs[i] = GET_REGISTER(vdst+i);
+        } else {
+            u4 count = vsrc1 >> 4;
+
+            assert(count <= curMethod->outsSize);
+            assert(count == methodToCall->insSize);
+            assert(count <= 5);
+
+            outs = OUTS_FROM_FP(fp, count);
+#if 0
+            if (count == 5) {
+                outs[4] = GET_REGISTER(vsrc1 & 0x0f);
+                count--;
+            }
+            for (i = 0; i < (int) count; i++) {
+                outs[i] = GET_REGISTER(vdst & 0x0f);
+                vdst >>= 4;
+            }
+#else
+            // This version executes fewer instructions but is larger
+            // overall.  Seems to be a teensy bit faster.
+            assert((vdst >> 16) == 0);  // 16 bits -or- high 16 bits clear
+            switch (count) {
+            case 5:
+                outs[4] = GET_REGISTER(vsrc1 & 0x0f);
+            case 4:
+                outs[3] = GET_REGISTER(vdst >> 12);
+            case 3:
+                outs[2] = GET_REGISTER((vdst & 0x0f00) >> 8);
+            case 2:
+                outs[1] = GET_REGISTER((vdst & 0x00f0) >> 4);
+            case 1:
+                outs[0] = GET_REGISTER(vdst & 0x0f);
+            default:
+                ;
+            }
+#endif
+        }
+    }
+
+    /*
+     * (This was originally a "goto" target; I've kept it separate from the
+     * stuff above in case we want to refactor things again.)
+     *
+     * At this point, we have the arguments stored in the "outs" area of
+     * the current method's stack frame, and the method to call in
+     * "methodToCall".  Push a new stack frame.
+     */
+    {
+        StackSaveArea* newSaveArea;
+        u4* newFp;
+
+        ILOGV("> %s%s.%s %s",
+            dvmIsNativeMethod(methodToCall) ? "(NATIVE) " : "",
+            methodToCall->clazz->descriptor, methodToCall->name,
+            methodToCall->signature);
+
+        newFp = (u4*) SAVEAREA_FROM_FP(fp) - methodToCall->registersSize;
+        newSaveArea = SAVEAREA_FROM_FP(newFp);
+
+        /* verify that we have enough space */
+        if (true) {
+            u1* bottom;
+            bottom = (u1*) newSaveArea - methodToCall->outsSize * sizeof(u4);
+            if (bottom < self->interpStackEnd) {
+                /* stack overflow */
+                LOGV("Stack overflow on method call (start=%p end=%p newBot=%p size=%d '%s')\n",
+                    self->interpStackStart, self->interpStackEnd, bottom,
+                    self->interpStackSize, methodToCall->name);
+                dvmHandleStackOverflow(self);
+                assert(dvmCheckException(self));
+                GOTO_exceptionThrown();
+            }
+            //LOGD("+++ fp=%p newFp=%p newSave=%p bottom=%p\n",
+            //    fp, newFp, newSaveArea, bottom);
+        }
+
+#ifdef LOG_INSTR
+        if (methodToCall->registersSize > methodToCall->insSize) {
+            /*
+             * This makes valgrind quiet when we print registers that
+             * haven't been initialized.  Turn it off when the debug
+             * messages are disabled -- we want valgrind to report any
+             * used-before-initialized issues.
+             */
+            memset(newFp, 0xcc,
+                (methodToCall->registersSize - methodToCall->insSize) * 4);
+        }
+#endif
+
+#ifdef EASY_GDB
+        newSaveArea->prevSave = SAVEAREA_FROM_FP(fp);
+#endif
+        newSaveArea->prevFrame = fp;
+        newSaveArea->savedPc = pc;
+        newSaveArea->method = methodToCall;
+
+        if (!dvmIsNativeMethod(methodToCall)) {
+            /*
+             * "Call" interpreted code.  Reposition the PC, update the
+             * frame pointer and other local state, and continue.
+             */
+            curMethod = methodToCall;
+            methodClassDex = curMethod->clazz->pDvmDex;
+            pc = methodToCall->insns;
+            fp = self->curFrame = newFp;
+#ifdef EASY_GDB
+            debugSaveArea = SAVEAREA_FROM_FP(newFp);
+#endif
+#if INTERP_TYPE == INTERP_DBG
+            debugIsMethodEntry = true;              // profiling, debugging
+#endif
+            ILOGD("> pc <-- %s.%s %s", curMethod->clazz->descriptor,
+                curMethod->name, curMethod->signature);
+            DUMP_REGS(curMethod, fp, true);         // show input args
+            FINISH(0);                              // jump to method start
+        } else {
+            /* set this up for JNI locals, even if not a JNI native */
+            newSaveArea->xtra.localRefTop = self->jniLocalRefTable.nextEntry;
+
+            self->curFrame = newFp;
+
+            DUMP_REGS(methodToCall, newFp, true);   // show input args
+
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_DEBUGGER)
+            if (gDvm.debuggerActive) {
+                dvmDbgPostLocationEvent(methodToCall, -1,
+                    dvmGetThisPtr(curMethod, fp), DBG_METHOD_ENTRY);
+            }
+#endif
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_PROFILER)
+            TRACE_METHOD_ENTER(self, methodToCall);
+#endif
+
+            ILOGD("> native <-- %s.%s %s", methodToCall->clazz->descriptor,
+                methodToCall->name, methodToCall->signature);
+
+            /*
+             * Jump through native call bridge.  Because we leave no
+             * space for locals on native calls, "newFp" points directly
+             * to the method arguments.
+             */
+            (*methodToCall->nativeFunc)(newFp, &retval, methodToCall, self);
+
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_DEBUGGER)
+            if (gDvm.debuggerActive) {
+                dvmDbgPostLocationEvent(methodToCall, -1,
+                    dvmGetThisPtr(curMethod, fp), DBG_METHOD_EXIT);
+            }
+#endif
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_PROFILER)
+            TRACE_METHOD_EXIT(self, methodToCall);
+#endif
+
+            /* pop frame off */
+            dvmPopJniLocals(self, newSaveArea);
+            self->curFrame = fp;
+
+            /*
+             * If the native code threw an exception, or interpreted code
+             * invoked by the native call threw one and nobody has cleared
+             * it, jump to our local exception handling.
+             */
+            if (dvmCheckException(self)) {
+                LOGV("Exception thrown by/below native code\n");
+                GOTO_exceptionThrown();
+            }
+
+            ILOGD("> retval=0x%llx (leaving native)", retval.j);
+            ILOGD("> (return from native %s.%s to %s.%s %s)",
+                methodToCall->clazz->descriptor, methodToCall->name,
+                curMethod->clazz->descriptor, curMethod->name,
+                curMethod->signature);
+
+            //u2 invokeInstr = INST_INST(FETCH(0));
+            if (true /*invokeInstr >= OP_INVOKE_VIRTUAL &&
+                invokeInstr <= OP_INVOKE_INTERFACE*/)
+            {
+                FINISH(3);
+            } else {
+                //LOGE("Unknown invoke instr %02x at %d\n",
+                //    invokeInstr, (int) (pc - curMethod->insns));
+                assert(false);
+            }
+        }
+    }
+    assert(false);      // should not get here
+GOTO_TARGET_END
+
+
+/* File: portable/enddefs.c */
+/*--- end of opcodes ---*/
+
+#ifndef THREADED_INTERP
+        } // end of "switch"
+    } // end of "while"
+#endif
+
+bail:
+    ILOGD("|-- Leaving interpreter loop");      // note "curMethod" may be NULL
+
+    interpState->retval = retval;
+    return false;
+
+bail_switch:
+    /*
+     * The standard interpreter currently doesn't set or care about the
+     * "debugIsMethodEntry" value, so setting this is only of use if we're
+     * switching between two "debug" interpreters, which we never do.
+     *
+     * TODO: figure out if preserving this makes any sense.
+     */
+#if defined(WITH_PROFILER) || defined(WITH_DEBUGGER)
+# if INTERP_TYPE == INTERP_DBG
+    interpState->debugIsMethodEntry = debugIsMethodEntry;
+# else
+    interpState->debugIsMethodEntry = false;
+# endif
+#endif
+
+    /* export state changes */
+    interpState->method = curMethod;
+    interpState->pc = pc;
+    interpState->fp = fp;
+    /* debugTrackedRefStart doesn't change */
+    interpState->retval = retval;   /* need for _entryPoint=ret */
+    interpState->nextMode =
+        (INTERP_TYPE == INTERP_STD) ? INTERP_DBG : INTERP_STD;
+    LOGVV(" meth='%s.%s' pc=0x%x fp=%p\n",
+        curMethod->clazz->descriptor, curMethod->name,
+        pc - curMethod->insns, fp);
+    return true;
+}
+
+
diff --git a/vm/mterp/out/InterpC-portstd.c b/vm/mterp/out/InterpC-portstd.c
new file mode 100644
index 0000000..64e5ccd
--- /dev/null
+++ b/vm/mterp/out/InterpC-portstd.c
@@ -0,0 +1,3984 @@
+/*
+ * This file was generated automatically by gen-mterp.py for 'portstd'.
+ *
+ * --> DO NOT EDIT <--
+ */
+
+/* File: c/header.c */
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/* common includes */
+#include "Dalvik.h"
+#include "interp/InterpDefs.h"
+#include "mterp/Mterp.h"
+#include <math.h>                   // needed for fmod, fmodf
+
+/*
+ * Configuration defines.  These affect the C implementations, i.e. the
+ * portable interpreter(s) and C stubs.
+ *
+ * Some defines are controlled by the Makefile, e.g.:
+ *   WITH_PROFILER
+ *   WITH_DEBUGGER
+ *   WITH_INSTR_CHECKS
+ *   WITH_TRACKREF_CHECKS
+ *   EASY_GDB
+ *   NDEBUG
+ *
+ * If THREADED_INTERP is not defined, we use a classic "while true / switch"
+ * interpreter.  If it is defined, then the tail end of each instruction
+ * handler fetches the next instruction and jumps directly to the handler.
+ * This increases the size of the "Std" interpreter by about 10%, but
+ * provides a speedup of about the same magnitude.
+ *
+ * There's a "hybrid" approach that uses a goto table instead of a switch
+ * statement, avoiding the "is the opcode in range" tests required for switch.
+ * The performance is close to the threaded version, and without the 10%
+ * size increase, but the benchmark results are off enough that it's not
+ * worth adding as a third option.
+ */
+#define THREADED_INTERP             /* threaded vs. while-loop interpreter */
+
+#ifdef WITH_INSTR_CHECKS            /* instruction-level paranoia */
+# define CHECK_BRANCH_OFFSETS
+# define CHECK_REGISTER_INDICES
+#endif
+
+/*
+ * ARM EABI requires 64-bit alignment for access to 64-bit data types.  We
+ * can't just use pointers to copy 64-bit values out of our interpreted
+ * register set, because gcc will generate ldrd/strd.
+ *
+ * The __UNION version copies data in and out of a union.  The __MEMCPY
+ * version uses a memcpy() call to do the transfer; gcc is smart enough to
+ * not actually call memcpy().  The __UNION version is very bad on ARM;
+ * it only uses one more instruction than __MEMCPY, but for some reason
+ * gcc thinks it needs separate storage for every instance of the union.
+ * On top of that, it feels the need to zero them out at the start of the
+ * method.  Net result is we zero out ~700 bytes of stack space at the top
+ * of the interpreter using ARM STM instructions.
+ */
+#if defined(__ARM_EABI__)
+//# define NO_UNALIGN_64__UNION
+# define NO_UNALIGN_64__MEMCPY
+#endif
+
+//#define LOG_INSTR                   /* verbose debugging */
+/* set and adjust ANDROID_LOG_TAGS='*:i jdwp:i dalvikvm:i dalvikvmi:i' */
+
+/*
+ * Keep a tally of accesses to fields.  Currently only works if full DEX
+ * optimization is disabled.
+ */
+#ifdef PROFILE_FIELD_ACCESS
+# define UPDATE_FIELD_GET(_field) { (_field)->gets++; }
+# define UPDATE_FIELD_PUT(_field) { (_field)->puts++; }
+#else
+# define UPDATE_FIELD_GET(_field) ((void)0)
+# define UPDATE_FIELD_PUT(_field) ((void)0)
+#endif
+
+/*
+ * Adjust the program counter.  "_offset" is a signed int, in 16-bit units.
+ *
+ * Assumes the existence of "const u2* pc" and "const u2* curMethod->insns".
+ *
+ * We don't advance the program counter until we finish an instruction or
+ * branch, because we do want to have to unroll the PC if there's an
+ * exception.
+ */
+#ifdef CHECK_BRANCH_OFFSETS
+# define ADJUST_PC(_offset) do {                                            \
+        int myoff = _offset;        /* deref only once */                   \
+        if (pc + myoff < curMethod->insns ||                                \
+            pc + myoff >= curMethod->insns + dvmGetMethodInsnsSize(curMethod)) \
+        {                                                                   \
+            char* desc;                                                     \
+            desc = dexProtoCopyMethodDescriptor(&curMethod->prototype);     \
+            LOGE("Invalid branch %d at 0x%04x in %s.%s %s\n",               \
+                myoff, (int) (pc - curMethod->insns),                       \
+                curMethod->clazz->descriptor, curMethod->name, desc);       \
+            free(desc);                                                     \
+            dvmAbort();                                                     \
+        }                                                                   \
+        pc += myoff;                                                        \
+    } while (false)
+#else
+# define ADJUST_PC(_offset) (pc += _offset)
+#endif
+
+/*
+ * If enabled, log instructions as we execute them.
+ */
+#ifdef LOG_INSTR
+# define ILOGD(...) ILOG(LOG_DEBUG, __VA_ARGS__)
+# define ILOGV(...) ILOG(LOG_VERBOSE, __VA_ARGS__)
+# define ILOG(_level, ...) do {                                             \
+        char debugStrBuf[128];                                              \
+        snprintf(debugStrBuf, sizeof(debugStrBuf), __VA_ARGS__);            \
+        if (curMethod != NULL)                                                 \
+            LOG(_level, LOG_TAG"i", "%-2d|%04x%s\n",                        \
+                self->threadId, (int)(pc - curMethod->insns), debugStrBuf); \
+        else                                                                \
+            LOG(_level, LOG_TAG"i", "%-2d|####%s\n",                        \
+                self->threadId, debugStrBuf);                               \
+    } while(false)
+void dvmDumpRegs(const Method* method, const u4* framePtr, bool inOnly);
+# define DUMP_REGS(_meth, _frame, _inOnly) dvmDumpRegs(_meth, _frame, _inOnly)
+static const char kSpacing[] = "            ";
+#else
+# define ILOGD(...) ((void)0)
+# define ILOGV(...) ((void)0)
+# define DUMP_REGS(_meth, _frame, _inOnly) ((void)0)
+#endif
+
+/* get a long from an array of u4 */
+static inline s8 getLongFromArray(const u4* ptr, int idx)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { s8 ll; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.parts[0] = ptr[0];
+    conv.parts[1] = ptr[1];
+    return conv.ll;
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    s8 val;
+    memcpy(&val, &ptr[idx], 8);
+    return val;
+#else
+    return *((s8*) &ptr[idx]);
+#endif
+}
+
+/* store a long into an array of u4 */
+static inline void putLongToArray(u4* ptr, int idx, s8 val)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { s8 ll; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.ll = val;
+    ptr[0] = conv.parts[0];
+    ptr[1] = conv.parts[1];
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    memcpy(&ptr[idx], &val, 8);
+#else
+    *((s8*) &ptr[idx]) = val;
+#endif
+}
+
+/* get a double from an array of u4 */
+static inline double getDoubleFromArray(const u4* ptr, int idx)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { double d; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.parts[0] = ptr[0];
+    conv.parts[1] = ptr[1];
+    return conv.d;
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    double dval;
+    memcpy(&dval, &ptr[idx], 8);
+    return dval;
+#else
+    return *((double*) &ptr[idx]);
+#endif
+}
+
+/* store a double into an array of u4 */
+static inline void putDoubleToArray(u4* ptr, int idx, double dval)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { double d; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.d = dval;
+    ptr[0] = conv.parts[0];
+    ptr[1] = conv.parts[1];
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    memcpy(&ptr[idx], &dval, 8);
+#else
+    *((double*) &ptr[idx]) = dval;
+#endif
+}
+
+/*
+ * If enabled, validate the register number on every access.  Otherwise,
+ * just do an array access.
+ *
+ * Assumes the existence of "u4* fp".
+ *
+ * "_idx" may be referenced more than once.
+ */
+#ifdef CHECK_REGISTER_INDICES
+# define GET_REGISTER(_idx) \
+    ( (_idx) < curMethod->registersSize ? \
+        (fp[(_idx)]) : (assert(!"bad reg"),1969) )
+# define SET_REGISTER(_idx, _val) \
+    ( (_idx) < curMethod->registersSize ? \
+        (fp[(_idx)] = (u4)(_val)) : (assert(!"bad reg"),1969) )
+# define GET_REGISTER_AS_OBJECT(_idx)       ((Object *)GET_REGISTER(_idx))
+# define SET_REGISTER_AS_OBJECT(_idx, _val) SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_INT(_idx) ((s4) GET_REGISTER(_idx))
+# define SET_REGISTER_INT(_idx, _val) SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_WIDE(_idx) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        getLongFromArray(fp, (_idx)) : (assert(!"bad reg"),1969) )
+# define SET_REGISTER_WIDE(_idx, _val) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        putLongToArray(fp, (_idx), (_val)) : (assert(!"bad reg"),1969) )
+# define GET_REGISTER_FLOAT(_idx) \
+    ( (_idx) < curMethod->registersSize ? \
+        (*((float*) &fp[(_idx)])) : (assert(!"bad reg"),1969.0f) )
+# define SET_REGISTER_FLOAT(_idx, _val) \
+    ( (_idx) < curMethod->registersSize ? \
+        (*((float*) &fp[(_idx)]) = (_val)) : (assert(!"bad reg"),1969.0f) )
+# define GET_REGISTER_DOUBLE(_idx) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        getDoubleFromArray(fp, (_idx)) : (assert(!"bad reg"),1969.0) )
+# define SET_REGISTER_DOUBLE(_idx, _val) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        putDoubleToArray(fp, (_idx), (_val)) : (assert(!"bad reg"),1969.0) )
+#else
+# define GET_REGISTER(_idx)                 (fp[(_idx)])
+# define SET_REGISTER(_idx, _val)           (fp[(_idx)] = (_val))
+# define GET_REGISTER_AS_OBJECT(_idx)       ((Object*) fp[(_idx)])
+# define SET_REGISTER_AS_OBJECT(_idx, _val) (fp[(_idx)] = (u4)(_val))
+# define GET_REGISTER_INT(_idx)             ((s4)GET_REGISTER(_idx))
+# define SET_REGISTER_INT(_idx, _val)       SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_WIDE(_idx)            getLongFromArray(fp, (_idx))
+# define SET_REGISTER_WIDE(_idx, _val)      putLongToArray(fp, (_idx), (_val))
+# define GET_REGISTER_FLOAT(_idx)           (*((float*) &fp[(_idx)]))
+# define SET_REGISTER_FLOAT(_idx, _val)     (*((float*) &fp[(_idx)]) = (_val))
+# define GET_REGISTER_DOUBLE(_idx)          getDoubleFromArray(fp, (_idx))
+# define SET_REGISTER_DOUBLE(_idx, _val)    putDoubleToArray(fp, (_idx), (_val))
+#endif
+
+/*
+ * Get 16 bits from the specified offset of the program counter.  We always
+ * want to load 16 bits at a time from the instruction stream -- it's more
+ * efficient than 8 and won't have the alignment problems that 32 might.
+ *
+ * Assumes existence of "const u2* pc".
+ */
+#define FETCH(_offset)     (pc[(_offset)])
+
+/*
+ * Extract instruction byte from 16-bit fetch (_inst is a u2).
+ */
+#define INST_INST(_inst)    ((_inst) & 0xff)
+
+/*
+ * Extract the "vA, vB" 4-bit registers from the instruction word (_inst is u2).
+ */
+#define INST_A(_inst)       (((_inst) >> 8) & 0x0f)
+#define INST_B(_inst)       ((_inst) >> 12)
+
+/*
+ * Get the 8-bit "vAA" 8-bit register index from the instruction word.
+ * (_inst is u2)
+ */
+#define INST_AA(_inst)      ((_inst) >> 8)
+
+/*
+ * The current PC must be available to Throwable constructors, e.g.
+ * those created by dvmThrowException(), so that the exception stack
+ * trace can be generated correctly.  If we don't do this, the offset
+ * within the current method won't be shown correctly.  See the notes
+ * in Exception.c.
+ *
+ * Assumes existence of "u4* fp" and "const u2* pc".
+ */
+#define EXPORT_PC()         (SAVEAREA_FROM_FP(fp)->xtra.currentPc = pc)
+
+/*
+ * Determine if we need to switch to a different interpreter.  "_current"
+ * is either INTERP_STD or INTERP_DBG.  It should be fixed for a given
+ * interpreter generation file, which should remove the outer conditional
+ * from the following.
+ *
+ * If we're building without debug and profiling support, we never switch.
+ */
+#if defined(WITH_PROFILER) || defined(WITH_DEBUGGER)
+# define NEED_INTERP_SWITCH(_current) (                                     \
+    (_current == INTERP_STD) ?                                              \
+        dvmDebuggerOrProfilerActive() : !dvmDebuggerOrProfilerActive() )
+#else
+# define NEED_INTERP_SWITCH(_current) (false)
+#endif
+
+/*
+ * Look up an interface on a class using the cache.
+ */
+INLINE Method* dvmFindInterfaceMethodInCache(ClassObject* thisClass,
+    u4 methodIdx, const Method* method, DvmDex* methodClassDex)
+{
+#define ATOMIC_CACHE_CALC \
+    dvmInterpFindInterfaceMethod(thisClass, methodIdx, method, methodClassDex)
+
+    return (Method*) ATOMIC_CACHE_LOOKUP(methodClassDex->pInterfaceCache,
+                DEX_INTERFACE_CACHE_SIZE, thisClass, methodIdx);
+
+#undef ATOMIC_CACHE_CALC
+}
+
+/*
+ * Check to see if "obj" is NULL.  If so, throw an exception.  Assumes the
+ * pc has already been exported to the stack.
+ *
+ * Perform additional checks on debug builds.
+ *
+ * Use this to check for NULL when the instruction handler calls into
+ * something that could throw an exception (so we have already called
+ * EXPORT_PC at the top).
+ */
+static inline bool checkForNull(Object* obj)
+{
+    if (obj == NULL) {
+        dvmThrowException("Ljava/lang/NullPointerException;", NULL);
+        return false;
+    }
+#ifdef WITH_EXTRA_OBJECT_VALIDATION
+    if (!dvmIsValidObject(obj)) {
+        LOGE("Invalid object %p\n", obj);
+        dvmAbort();
+    }
+#endif
+#ifndef NDEBUG
+    if (obj->clazz == NULL || ((u4) obj->clazz) <= 65536) {
+        /* probable heap corruption */
+        LOGE("Invalid object class %p (in %p)\n", obj->clazz, obj);
+        dvmAbort();
+    }
+#endif
+    return true;
+}
+
+/*
+ * Check to see if "obj" is NULL.  If so, export the PC into the stack
+ * frame and throw an exception.
+ *
+ * Perform additional checks on debug builds.
+ *
+ * Use this to check for NULL when the instruction handler doesn't do
+ * anything else that can throw an exception.
+ */
+static inline bool checkForNullExportPC(Object* obj, u4* fp, const u2* pc)
+{
+    if (obj == NULL) {
+        EXPORT_PC();
+        dvmThrowException("Ljava/lang/NullPointerException;", NULL);
+        return false;
+    }
+#ifdef WITH_EXTRA_OBJECT_VALIDATION
+    if (!dvmIsValidObject(obj)) {
+        LOGE("Invalid object %p\n", obj);
+        dvmAbort();
+    }
+#endif
+#ifndef NDEBUG
+    if (obj->clazz == NULL || ((u4) obj->clazz) <= 65536) {
+        /* probable heap corruption */
+        LOGE("Invalid object class %p (in %p)\n", obj->clazz, obj);
+        dvmAbort();
+    }
+#endif
+    return true;
+}
+
+
+/* File: portable/portstd.c */
+#define INTERP_FUNC_NAME dvmInterpretStd
+#define INTERP_TYPE INTERP_STD
+
+#define CHECK_DEBUG_AND_PROF() ((void)0)
+
+/* File: portable/stubdefs.c */
+/*
+ * In the C mterp stubs, "goto" is a function call followed immediately
+ * by a return.
+ */
+
+#define GOTO_TARGET_DECL(_target, ...)
+
+#define GOTO_TARGET(_target, ...) _target:
+
+#define GOTO_TARGET_END
+
+/* ugh */
+#define STUB_HACK(x)
+
+/*
+ * Instruction framing.  For a switch-oriented implementation this is
+ * case/break, for a threaded implementation it's a goto label and an
+ * instruction fetch/computed goto.
+ *
+ * Assumes the existence of "const u2* pc" and (for threaded operation)
+ * "u2 inst".
+ */
+#ifdef THREADED_INTERP
+# define H(_op)             &&op_##_op
+# define HANDLE_OPCODE(_op) op_##_op:
+# define FINISH(_offset) {                                                  \
+        ADJUST_PC(_offset);                                                 \
+        inst = FETCH(0);                                                    \
+        CHECK_DEBUG_AND_PROF();                                             \
+        CHECK_TRACKED_REFS();                                               \
+        goto *handlerTable[INST_INST(inst)];                                \
+    }
+#else
+# define HANDLE_OPCODE(_op) case _op:
+# define FINISH(_offset)    { ADJUST_PC(_offset); break; }
+#endif
+
+#define OP_END
+
+#if defined(WITH_TRACKREF_CHECKS)
+# define CHECK_TRACKED_REFS() \
+    dvmInterpCheckTrackedRefs(self, curMethod, debugTrackedRefStart)
+#else
+# define CHECK_TRACKED_REFS() ((void)0)
+#endif
+
+
+/*
+ * The "goto" targets just turn into goto statements.  The "arguments" are
+ * passed through local variables.
+ */
+
+#define GOTO_exceptionThrown() goto exceptionThrown;
+
+#define GOTO_returnFromMethod() goto returnFromMethod;
+
+#define GOTO_invoke(_target, _methodCallRange)                              \
+    do {                                                                    \
+        methodCallRange = _methodCallRange;                                 \
+        goto _target;                                                       \
+    } while(false)
+
+/* for this, the "args" are already in the locals */
+#define GOTO_invokeMethod(_methodCallRange, _methodToCall, _vsrc1, _vdst) goto invokeMethod;
+
+#define GOTO_bail() goto bail;
+#define GOTO_bail_switch() goto bail_switch;
+
+/*
+ * Periodically check for thread suspension.
+ *
+ * While we're at it, see if a debugger has attached or the profiler has
+ * started.  If so, switch to a different "goto" table.
+ */
+#define PERIODIC_CHECKS(_entryPoint, _pcadj) {                              \
+        dvmCheckSuspendQuick(self);                                         \
+        if (NEED_INTERP_SWITCH(INTERP_TYPE)) {                              \
+            ADJUST_PC(_pcadj);                                              \
+            interpState->entryPoint = _entryPoint;                          \
+            LOGVV("threadid=%d: switch to %s ep=%d adj=%d\n",               \
+                self->threadId,                                             \
+                (interpState->nextMode == INTERP_STD) ? "STD" : "DBG",      \
+                (_entryPoint), (_pcadj));                                   \
+            GOTO_bail_switch();                                             \
+        }                                                                   \
+    }
+
+
+/* File: c/opcommon.c */
+/* forward declarations of goto targets */
+GOTO_TARGET_DECL(filledNewArray, bool methodCallRange);
+GOTO_TARGET_DECL(invokeVirtual, bool methodCallRange);
+GOTO_TARGET_DECL(invokeSuper, bool methodCallRange);
+GOTO_TARGET_DECL(invokeInterface, bool methodCallRange);
+GOTO_TARGET_DECL(invokeDirect, bool methodCallRange);
+GOTO_TARGET_DECL(invokeStatic, bool methodCallRange);
+GOTO_TARGET_DECL(invokeVirtualQuick, bool methodCallRange);
+GOTO_TARGET_DECL(invokeSuperQuick, bool methodCallRange);
+GOTO_TARGET_DECL(invokeMethod, bool methodCallRange, const Method* methodToCall,
+    u2 count, u2 regs);
+GOTO_TARGET_DECL(returnFromMethod);
+GOTO_TARGET_DECL(exceptionThrown);
+
+/*
+ * ===========================================================================
+ *
+ * What follows are opcode definitions shared between multiple opcodes with
+ * minor substitutions handled by the C pre-processor.  These should probably
+ * use the mterp substitution mechanism instead, with the code here moved
+ * into common fragment files (like the asm "binop.S"), although it's hard
+ * to give up the C preprocessor in favor of the much simpler text subst.
+ *
+ * ===========================================================================
+ */
+
+#define HANDLE_NUMCONV(_opcode, _opname, _fromtype, _totype)                \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        SET_REGISTER##_totype(vdst,                                         \
+            GET_REGISTER##_fromtype(vsrc1));                                \
+        FINISH(1);
+
+#define HANDLE_FLOAT_TO_INT(_opcode, _opname, _fromvtype, _fromrtype,       \
+        _tovtype, _tortype)                                                 \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+    {                                                                       \
+        /* spec defines specific handling for +/- inf and NaN values */     \
+        _fromvtype val;                                                     \
+        _tovtype intMin, intMax, result;                                    \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        val = GET_REGISTER##_fromrtype(vsrc1);                              \
+        intMin = (_tovtype) 1 << (sizeof(_tovtype) * 8 -1);                 \
+        intMax = ~intMin;                                                   \
+        result = (_tovtype) val;                                            \
+        if (val >= intMax)          /* +inf */                              \
+            result = intMax;                                                \
+        else if (val <= intMin)     /* -inf */                              \
+            result = intMin;                                                \
+        else if (val != val)        /* NaN */                               \
+            result = 0;                                                     \
+        else                                                                \
+            result = (_tovtype) val;                                        \
+        SET_REGISTER##_tortype(vdst, result);                               \
+    }                                                                       \
+    FINISH(1);
+
+#define HANDLE_INT_TO_SMALL(_opcode, _opname, _type)                        \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|int-to-%s v%d,v%d", (_opname), vdst, vsrc1);                \
+        SET_REGISTER(vdst, (_type) GET_REGISTER(vsrc1));                    \
+        FINISH(1);
+
+/* NOTE: the comparison result is always a signed 4-byte integer */
+#define HANDLE_OP_CMPX(_opcode, _opname, _varType, _type, _nanVal)          \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        int result;                                                         \
+        u2 regs;                                                            \
+        _varType val1, val2;                                                \
+        vdst = INST_AA(inst);                                               \
+        regs = FETCH(1);                                                    \
+        vsrc1 = regs & 0xff;                                                \
+        vsrc2 = regs >> 8;                                                  \
+        ILOGV("|cmp%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);         \
+        val1 = GET_REGISTER##_type(vsrc1);                                  \
+        val2 = GET_REGISTER##_type(vsrc2);                                  \
+        if (val1 == val2)                                                   \
+            result = 0;                                                     \
+        else if (val1 < val2)                                               \
+            result = -1;                                                    \
+        else if (val1 > val2)                                               \
+            result = 1;                                                     \
+        else                                                                \
+            result = (_nanVal);                                             \
+        ILOGV("+ result=%d\n", result);                                     \
+        SET_REGISTER(vdst, result);                                         \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_IF_XX(_opcode, _opname, _cmp)                             \
+    HANDLE_OPCODE(_opcode /*vA, vB, +CCCC*/)                                \
+        vsrc1 = INST_A(inst);                                               \
+        vsrc2 = INST_B(inst);                                               \
+        if ((s4) GET_REGISTER(vsrc1) _cmp (s4) GET_REGISTER(vsrc2)) {       \
+            int branchOffset = (s2)FETCH(1);    /* sign-extended */         \
+            ILOGV("|if-%s v%d,v%d,+0x%04x", (_opname), vsrc1, vsrc2,        \
+                branchOffset);                                              \
+            ILOGV("> branch taken");                                        \
+            if (branchOffset < 0)                                           \
+                PERIODIC_CHECKS(kInterpEntryInstr, branchOffset);           \
+            FINISH(branchOffset);                                           \
+        } else {                                                            \
+            ILOGV("|if-%s v%d,v%d,-", (_opname), vsrc1, vsrc2);             \
+            FINISH(2);                                                      \
+        }
+
+#define HANDLE_OP_IF_XXZ(_opcode, _opname, _cmp)                            \
+    HANDLE_OPCODE(_opcode /*vAA, +BBBB*/)                                   \
+        vsrc1 = INST_AA(inst);                                              \
+        if ((s4) GET_REGISTER(vsrc1) _cmp 0) {                              \
+            int branchOffset = (s2)FETCH(1);    /* sign-extended */         \
+            ILOGV("|if-%s v%d,+0x%04x", (_opname), vsrc1, branchOffset);    \
+            ILOGV("> branch taken");                                        \
+            if (branchOffset < 0)                                           \
+                PERIODIC_CHECKS(kInterpEntryInstr, branchOffset);           \
+            FINISH(branchOffset);                                           \
+        } else {                                                            \
+            ILOGV("|if-%s v%d,-", (_opname), vsrc1);                        \
+            FINISH(2);                                                      \
+        }
+
+#define HANDLE_UNOP(_opcode, _opname, _pfx, _sfx, _type)                    \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        SET_REGISTER##_type(vdst, _pfx GET_REGISTER##_type(vsrc1) _sfx);    \
+        FINISH(1);
+
+#define HANDLE_OP_X_INT(_opcode, _opname, _op, _chkdiv)                     \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-int v%d,v%d", (_opname), vdst, vsrc1);                   \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            secondVal = GET_REGISTER(vsrc2);                                \
+            if (secondVal == 0) {                                           \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && secondVal == -1) {            \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            /* non-div/rem case */                                          \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vsrc1) _op (s4) GET_REGISTER(vsrc2));     \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_INT(_opcode, _opname, _cast, _op)                     \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-int v%d,v%d", (_opname), vdst, vsrc1);                   \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vsrc1) _op (GET_REGISTER(vsrc2) & 0x1f));    \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_INT_LIT16(_opcode, _opname, _op, _chkdiv)               \
+    HANDLE_OPCODE(_opcode /*vA, vB, #+CCCC*/)                               \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        vsrc2 = FETCH(1);                                                   \
+        ILOGV("|%s-int/lit16 v%d,v%d,#+0x%04x",                             \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, result;                                            \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            if ((s2) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                      \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && ((s2) vsrc2) == -1) {         \
+                /* won't generate /lit16 instr for this; check anyway */    \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op (s2) vsrc2;                           \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            /* non-div/rem case */                                          \
+            SET_REGISTER(vdst, GET_REGISTER(vsrc1) _op (s2) vsrc2);         \
+        }                                                                   \
+        FINISH(2);
+
+#define HANDLE_OP_X_INT_LIT8(_opcode, _opname, _op, _chkdiv)                \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, #+CC*/)                               \
+    {                                                                       \
+        u2 litInfo;                                                         \
+        vdst = INST_AA(inst);                                               \
+        litInfo = FETCH(1);                                                 \
+        vsrc1 = litInfo & 0xff;                                             \
+        vsrc2 = litInfo >> 8;       /* constant */                          \
+        ILOGV("|%s-int/lit8 v%d,v%d,#+0x%02x",                              \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, result;                                            \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            if ((s1) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && ((s1) vsrc2) == -1) {         \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op ((s1) vsrc2);                         \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vsrc1) _op (s1) vsrc2);                   \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_INT_LIT8(_opcode, _opname, _cast, _op)                \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, #+CC*/)                               \
+    {                                                                       \
+        u2 litInfo;                                                         \
+        vdst = INST_AA(inst);                                               \
+        litInfo = FETCH(1);                                                 \
+        vsrc1 = litInfo & 0xff;                                             \
+        vsrc2 = litInfo >> 8;       /* constant */                          \
+        ILOGV("|%s-int/lit8 v%d,v%d,#+0x%02x",                              \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vsrc1) _op (vsrc2 & 0x1f));                  \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_INT_2ADDR(_opcode, _opname, _op, _chkdiv)               \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-int-2addr v%d,v%d", (_opname), vdst, vsrc1);             \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER(vdst);                                  \
+            secondVal = GET_REGISTER(vsrc1);                                \
+            if (secondVal == 0) {                                           \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && secondVal == -1) {            \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vdst) _op (s4) GET_REGISTER(vsrc1));      \
+        }                                                                   \
+        FINISH(1);
+
+#define HANDLE_OP_SHX_INT_2ADDR(_opcode, _opname, _cast, _op)               \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-int-2addr v%d,v%d", (_opname), vdst, vsrc1);             \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vdst) _op (GET_REGISTER(vsrc1) & 0x1f));     \
+        FINISH(1);
+
+#define HANDLE_OP_X_LONG(_opcode, _opname, _op, _chkdiv)                    \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-long v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);       \
+        if (_chkdiv != 0) {                                                 \
+            s8 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER_WIDE(vsrc1);                            \
+            secondVal = GET_REGISTER_WIDE(vsrc2);                           \
+            if (secondVal == 0LL) {                                         \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u8)firstVal == 0x8000000000000000ULL &&                    \
+                secondVal == -1LL)                                          \
+            {                                                               \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER_WIDE(vdst, result);                                \
+        } else {                                                            \
+            SET_REGISTER_WIDE(vdst,                                         \
+                (s8) GET_REGISTER_WIDE(vsrc1) _op (s8) GET_REGISTER_WIDE(vsrc2)); \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_LONG(_opcode, _opname, _cast, _op)                    \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-long v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);       \
+        SET_REGISTER_WIDE(vdst,                                             \
+            _cast GET_REGISTER_WIDE(vsrc1) _op (GET_REGISTER(vsrc2) & 0x3f)); \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_LONG_2ADDR(_opcode, _opname, _op, _chkdiv)              \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-long-2addr v%d,v%d", (_opname), vdst, vsrc1);            \
+        if (_chkdiv != 0) {                                                 \
+            s8 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER_WIDE(vdst);                             \
+            secondVal = GET_REGISTER_WIDE(vsrc1);                           \
+            if (secondVal == 0LL) {                                         \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u8)firstVal == 0x8000000000000000ULL &&                    \
+                secondVal == -1LL)                                          \
+            {                                                               \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER_WIDE(vdst, result);                                \
+        } else {                                                            \
+            SET_REGISTER_WIDE(vdst,                                         \
+                (s8) GET_REGISTER_WIDE(vdst) _op (s8)GET_REGISTER_WIDE(vsrc1));\
+        }                                                                   \
+        FINISH(1);
+
+#define HANDLE_OP_SHX_LONG_2ADDR(_opcode, _opname, _cast, _op)              \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-long-2addr v%d,v%d", (_opname), vdst, vsrc1);            \
+        SET_REGISTER_WIDE(vdst,                                             \
+            _cast GET_REGISTER_WIDE(vdst) _op (GET_REGISTER(vsrc1) & 0x3f)); \
+        FINISH(1);
+
+#define HANDLE_OP_X_FLOAT(_opcode, _opname, _op)                            \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-float v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);      \
+        SET_REGISTER_FLOAT(vdst,                                            \
+            GET_REGISTER_FLOAT(vsrc1) _op GET_REGISTER_FLOAT(vsrc2));       \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_DOUBLE(_opcode, _opname, _op)                           \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-double v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);     \
+        SET_REGISTER_DOUBLE(vdst,                                           \
+            GET_REGISTER_DOUBLE(vsrc1) _op GET_REGISTER_DOUBLE(vsrc2));     \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_FLOAT_2ADDR(_opcode, _opname, _op)                      \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-float-2addr v%d,v%d", (_opname), vdst, vsrc1);           \
+        SET_REGISTER_FLOAT(vdst,                                            \
+            GET_REGISTER_FLOAT(vdst) _op GET_REGISTER_FLOAT(vsrc1));        \
+        FINISH(1);
+
+#define HANDLE_OP_X_DOUBLE_2ADDR(_opcode, _opname, _op)                     \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-double-2addr v%d,v%d", (_opname), vdst, vsrc1);          \
+        SET_REGISTER_DOUBLE(vdst,                                           \
+            GET_REGISTER_DOUBLE(vdst) _op GET_REGISTER_DOUBLE(vsrc1));      \
+        FINISH(1);
+
+#define HANDLE_OP_AGET(_opcode, _opname, _type, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        ArrayObject* arrayObj;                                              \
+        u2 arrayInfo;                                                       \
+        EXPORT_PC();                                                        \
+        vdst = INST_AA(inst);                                               \
+        arrayInfo = FETCH(1);                                               \
+        vsrc1 = arrayInfo & 0xff;    /* array ptr */                        \
+        vsrc2 = arrayInfo >> 8;      /* index */                            \
+        ILOGV("|aget%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);        \
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);                      \
+        if (!checkForNull((Object*) arrayObj))                              \
+            GOTO_exceptionThrown();                                         \
+        if (GET_REGISTER(vsrc2) >= arrayObj->length) {                      \
+            LOGV("Invalid array access: %p %d (len=%d)\n",                  \
+                arrayObj, vsrc2, arrayObj->length);                         \
+            dvmThrowException("Ljava/lang/ArrayIndexOutOfBoundsException;", \
+                NULL);                                                      \
+            GOTO_exceptionThrown();                                         \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst,                                        \
+            ((_type*) arrayObj->contents)[GET_REGISTER(vsrc2)]);            \
+        ILOGV("+ AGET[%d]=0x%x", GET_REGISTER(vsrc2), GET_REGISTER(vdst));  \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_APUT(_opcode, _opname, _type, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        ArrayObject* arrayObj;                                              \
+        u2 arrayInfo;                                                       \
+        EXPORT_PC();                                                        \
+        vdst = INST_AA(inst);       /* AA: source value */                  \
+        arrayInfo = FETCH(1);                                               \
+        vsrc1 = arrayInfo & 0xff;   /* BB: array ptr */                     \
+        vsrc2 = arrayInfo >> 8;     /* CC: index */                         \
+        ILOGV("|aput%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);        \
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);                      \
+        if (!checkForNull((Object*) arrayObj))                              \
+            GOTO_exceptionThrown();                                         \
+        if (GET_REGISTER(vsrc2) >= arrayObj->length) {                      \
+            dvmThrowException("Ljava/lang/ArrayIndexOutOfBoundsException;", \
+                NULL);                                                      \
+            GOTO_exceptionThrown();                                         \
+        }                                                                   \
+        ILOGV("+ APUT[%d]=0x%08x", GET_REGISTER(vsrc2), GET_REGISTER(vdst));\
+        ((_type*) arrayObj->contents)[GET_REGISTER(vsrc2)] =                \
+            GET_REGISTER##_regsize(vdst);                                   \
+    }                                                                       \
+    FINISH(2);
+
+/*
+ * It's possible to get a bad value out of a field with sub-32-bit stores
+ * because the -quick versions always operate on 32 bits.  Consider:
+ *   short foo = -1  (sets a 32-bit register to 0xffffffff)
+ *   iput-quick foo  (writes all 32 bits to the field)
+ *   short bar = 1   (sets a 32-bit register to 0x00000001)
+ *   iput-short      (writes the low 16 bits to the field)
+ *   iget-quick foo  (reads all 32 bits from the field, yielding 0xffff0001)
+ * This can only happen when optimized and non-optimized code has interleaved
+ * access to the same field.  This is unlikely but possible.
+ *
+ * The easiest way to fix this is to always read/write 32 bits at a time.  On
+ * a device with a 16-bit data bus this is sub-optimal.  (The alternative
+ * approach is to have sub-int versions of iget-quick, but now we're wasting
+ * Dalvik instruction space and making it less likely that handler code will
+ * already be in the CPU i-cache.)
+ */
+#define HANDLE_IGET_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        InstField* ifield;                                                  \
+        Object* obj;                                                        \
+        EXPORT_PC();                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|iget%s v%d,v%d,field@0x%04x", (_opname), vdst, vsrc1, ref); \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNull(obj))                                             \
+            GOTO_exceptionThrown();                                         \
+        ifield = (InstField*) dvmDexGetResolvedField(methodClassDex, ref);  \
+        if (ifield == NULL) {                                               \
+            ifield = dvmResolveInstField(curMethod->clazz, ref);            \
+            if (ifield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst,                                        \
+            dvmGetField##_ftype(obj, ifield->byteOffset));                  \
+        ILOGV("+ IGET '%s'=0x%08llx", ifield->field.name,                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+        UPDATE_FIELD_GET(&ifield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IGET_X_QUICK(_opcode, _opname, _ftype, _regsize)             \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        Object* obj;                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field offset */                          \
+        ILOGV("|iget%s-quick v%d,v%d,field@+%u",                            \
+            (_opname), vdst, vsrc1, ref);                                   \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNullExportPC(obj, fp, pc))                             \
+            GOTO_exceptionThrown();                                         \
+        SET_REGISTER##_regsize(vdst, dvmGetField##_ftype(obj, ref));        \
+        ILOGV("+ IGETQ %d=0x%08llx", ref,                                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IPUT_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        InstField* ifield;                                                  \
+        Object* obj;                                                        \
+        EXPORT_PC();                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|iput%s v%d,v%d,field@0x%04x", (_opname), vdst, vsrc1, ref); \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNull(obj))                                             \
+            GOTO_exceptionThrown();                                         \
+        ifield = (InstField*) dvmDexGetResolvedField(methodClassDex, ref);  \
+        if (ifield == NULL) {                                               \
+            ifield = dvmResolveInstField(curMethod->clazz, ref);            \
+            if (ifield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        dvmSetField##_ftype(obj, ifield->byteOffset,                        \
+            GET_REGISTER##_regsize(vdst));                                  \
+        ILOGV("+ IPUT '%s'=0x%08llx", ifield->field.name,                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+        UPDATE_FIELD_PUT(&ifield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IPUT_X_QUICK(_opcode, _opname, _ftype, _regsize)             \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        Object* obj;                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field offset */                          \
+        ILOGV("|iput%s-quick v%d,v%d,field@0x%04x",                         \
+            (_opname), vdst, vsrc1, ref);                                   \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNullExportPC(obj, fp, pc))                             \
+            GOTO_exceptionThrown();                                         \
+        dvmSetField##_ftype(obj, ref, GET_REGISTER##_regsize(vdst));        \
+        ILOGV("+ IPUTQ %d=0x%08llx", ref,                                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_SGET_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, field@BBBB*/)                              \
+    {                                                                       \
+        StaticField* sfield;                                                \
+        vdst = INST_AA(inst);                                               \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|sget%s v%d,sfield@0x%04x", (_opname), vdst, ref);           \
+        sfield = (StaticField*)dvmDexGetResolvedField(methodClassDex, ref); \
+        if (sfield == NULL) {                                               \
+            EXPORT_PC();                                                    \
+            sfield = dvmResolveStaticField(curMethod->clazz, ref);          \
+            if (sfield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst, dvmGetStaticField##_ftype(sfield));    \
+        ILOGV("+ SGET '%s'=0x%08llx",                                       \
+            sfield->field.name, (u8)GET_REGISTER##_regsize(vdst));          \
+        UPDATE_FIELD_GET(&sfield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_SPUT_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, field@BBBB*/)                              \
+    {                                                                       \
+        StaticField* sfield;                                                \
+        vdst = INST_AA(inst);                                               \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|sput%s v%d,sfield@0x%04x", (_opname), vdst, ref);           \
+        sfield = (StaticField*)dvmDexGetResolvedField(methodClassDex, ref); \
+        if (sfield == NULL) {                                               \
+            EXPORT_PC();                                                    \
+            sfield = dvmResolveStaticField(curMethod->clazz, ref);          \
+            if (sfield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        dvmSetStaticField##_ftype(sfield, GET_REGISTER##_regsize(vdst));    \
+        ILOGV("+ SPUT '%s'=0x%08llx",                                       \
+            sfield->field.name, (u8)GET_REGISTER##_regsize(vdst));          \
+        UPDATE_FIELD_PUT(&sfield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+
+/* File: portable/entry.c */
+/*
+ * Main interpreter loop.
+ *
+ * This was written with an ARM implementation in mind.
+ */
+bool INTERP_FUNC_NAME(Thread* self, InterpState* interpState)
+{
+#if defined(EASY_GDB)
+    StackSaveArea* debugSaveArea = SAVEAREA_FROM_FP(self->curFrame);
+#endif
+#if INTERP_TYPE == INTERP_DBG
+    bool debugIsMethodEntry = interpState->debugIsMethodEntry;
+#endif
+#if defined(WITH_TRACKREF_CHECKS)
+    int debugTrackedRefStart = interpState->debugTrackedRefStart;
+#endif
+    DvmDex* methodClassDex;     // curMethod->clazz->pDvmDex
+    JValue retval;
+
+    /* core state */
+    const Method* curMethod;    // method we're interpreting
+    const u2* pc;               // program counter
+    u4* fp;                     // frame pointer
+    u2 inst;                    // current instruction
+    /* instruction decoding */
+    u2 ref;                     // 16-bit quantity fetched directly
+    u2 vsrc1, vsrc2, vdst;      // usually used for register indexes
+    /* method call setup */
+    const Method* methodToCall;
+    bool methodCallRange;
+
+#if defined(THREADED_INTERP)
+    /* static computed goto table */
+    DEFINE_GOTO_TABLE(handlerTable);
+#endif
+
+    /* copy state in */
+    curMethod = interpState->method;
+    pc = interpState->pc;
+    fp = interpState->fp;
+    retval = interpState->retval;   /* only need for kInterpEntryReturn? */
+
+    methodClassDex = curMethod->clazz->pDvmDex;
+
+    LOGVV("threadid=%d: entry(%s) %s.%s pc=0x%x fp=%p ep=%d\n",
+        self->threadId, (interpState->nextMode == INTERP_STD) ? "STD" : "DBG",
+        curMethod->clazz->descriptor, curMethod->name, pc - curMethod->insns,
+        fp, interpState->entryPoint);
+
+    /*
+     * DEBUG: scramble this to ensure we're not relying on it.
+     */
+    methodToCall = (const Method*) -1;
+
+#if INTERP_TYPE == INTERP_DBG
+    if (debugIsMethodEntry) {
+        ILOGD("|-- Now interpreting %s.%s", curMethod->clazz->descriptor,
+                curMethod->name);
+        DUMP_REGS(curMethod, interpState->fp, false);
+    }
+#endif
+
+    switch (interpState->entryPoint) {
+    case kInterpEntryInstr:
+        /* just fall through to instruction loop or threaded kickstart */
+        break;
+    case kInterpEntryReturn:
+        goto returnFromMethod;
+    case kInterpEntryThrow:
+        goto exceptionThrown;
+    default:
+        dvmAbort();
+    }
+
+#ifdef THREADED_INTERP
+    FINISH(0);                  /* fetch and execute first instruction */
+#else
+    while (1) {
+        CHECK_DEBUG_AND_PROF(); /* service debugger and profiling */
+        CHECK_TRACKED_REFS();   /* check local reference tracking */
+
+        /* fetch the next 16 bits from the instruction stream */
+        inst = FETCH(0);
+
+        switch (INST_INST(inst)) {
+#endif
+
+/*--- start of opcodes ---*/
+
+/* File: c/OP_NOP.c */
+HANDLE_OPCODE(OP_NOP)
+    FINISH(1);
+OP_END
+
+/* File: c/OP_MOVE.c */
+HANDLE_OPCODE(OP_MOVE /*vA, vB*/)
+    vdst = INST_A(inst);
+    vsrc1 = INST_B(inst);
+    ILOGV("|move%s v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(1);
+OP_END
+
+/* File: c/OP_MOVE_FROM16.c */
+HANDLE_OPCODE(OP_MOVE_FROM16 /*vAA, vBBBB*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|move%s/from16 v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE_FROM16) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(2);
+OP_END
+
+/* File: c/OP_MOVE_16.c */
+HANDLE_OPCODE(OP_MOVE_16 /*vAAAA, vBBBB*/)
+    vdst = FETCH(1);
+    vsrc1 = FETCH(2);
+    ILOGV("|move%s/16 v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE_16) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(3);
+OP_END
+
+/* File: c/OP_MOVE_WIDE.c */
+HANDLE_OPCODE(OP_MOVE_WIDE /*vA, vB*/)
+    /* IMPORTANT: must correctly handle overlapping registers, e.g. both
+     * "move-wide v6, v7" and "move-wide v7, v6" */
+    vdst = INST_A(inst);
+    vsrc1 = INST_B(inst);
+    ILOGV("|move-wide v%d,v%d %s(v%d=0x%08llx)", vdst, vsrc1,
+        kSpacing+5, vdst, GET_REGISTER_WIDE(vsrc1));
+    SET_REGISTER_WIDE(vdst, GET_REGISTER_WIDE(vsrc1));
+    FINISH(1);
+OP_END
+
+/* File: c/OP_MOVE_WIDE_FROM16.c */
+HANDLE_OPCODE(OP_MOVE_WIDE_FROM16 /*vAA, vBBBB*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|move-wide/from16 v%d,v%d  (v%d=0x%08llx)", vdst, vsrc1,
+        vdst, GET_REGISTER_WIDE(vsrc1));
+    SET_REGISTER_WIDE(vdst, GET_REGISTER_WIDE(vsrc1));
+    FINISH(2);
+OP_END
+
+/* File: c/OP_MOVE_WIDE_16.c */
+HANDLE_OPCODE(OP_MOVE_WIDE_16 /*vAAAA, vBBBB*/)
+    vdst = FETCH(1);
+    vsrc1 = FETCH(2);
+    ILOGV("|move-wide/16 v%d,v%d %s(v%d=0x%08llx)", vdst, vsrc1,
+        kSpacing+8, vdst, GET_REGISTER_WIDE(vsrc1));
+    SET_REGISTER_WIDE(vdst, GET_REGISTER_WIDE(vsrc1));
+    FINISH(3);
+OP_END
+
+/* File: c/OP_MOVE_OBJECT.c */
+/* File: c/OP_MOVE.c */
+HANDLE_OPCODE(OP_MOVE_OBJECT /*vA, vB*/)
+    vdst = INST_A(inst);
+    vsrc1 = INST_B(inst);
+    ILOGV("|move%s v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(1);
+OP_END
+
+
+/* File: c/OP_MOVE_OBJECT_FROM16.c */
+/* File: c/OP_MOVE_FROM16.c */
+HANDLE_OPCODE(OP_MOVE_OBJECT_FROM16 /*vAA, vBBBB*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|move%s/from16 v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE_FROM16) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(2);
+OP_END
+
+
+/* File: c/OP_MOVE_OBJECT_16.c */
+/* File: c/OP_MOVE_16.c */
+HANDLE_OPCODE(OP_MOVE_OBJECT_16 /*vAAAA, vBBBB*/)
+    vdst = FETCH(1);
+    vsrc1 = FETCH(2);
+    ILOGV("|move%s/16 v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE_16) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(3);
+OP_END
+
+
+/* File: c/OP_MOVE_RESULT.c */
+HANDLE_OPCODE(OP_MOVE_RESULT /*vAA*/)
+    vdst = INST_AA(inst);
+    ILOGV("|move-result%s v%d %s(v%d=0x%08x)",
+         (INST_INST(inst) == OP_MOVE_RESULT) ? "" : "-object",
+         vdst, kSpacing+4, vdst,retval.i);
+    SET_REGISTER(vdst, retval.i);
+    FINISH(1);
+OP_END
+
+/* File: c/OP_MOVE_RESULT_WIDE.c */
+HANDLE_OPCODE(OP_MOVE_RESULT_WIDE /*vAA*/)
+    vdst = INST_AA(inst);
+    ILOGV("|move-result-wide v%d %s(0x%08llx)", vdst, kSpacing, retval.j);
+    SET_REGISTER_WIDE(vdst, retval.j);
+    FINISH(1);
+OP_END
+
+/* File: c/OP_MOVE_RESULT_OBJECT.c */
+/* File: c/OP_MOVE_RESULT.c */
+HANDLE_OPCODE(OP_MOVE_RESULT_OBJECT /*vAA*/)
+    vdst = INST_AA(inst);
+    ILOGV("|move-result%s v%d %s(v%d=0x%08x)",
+         (INST_INST(inst) == OP_MOVE_RESULT) ? "" : "-object",
+         vdst, kSpacing+4, vdst,retval.i);
+    SET_REGISTER(vdst, retval.i);
+    FINISH(1);
+OP_END
+
+
+/* File: c/OP_MOVE_EXCEPTION.c */
+HANDLE_OPCODE(OP_MOVE_EXCEPTION /*vAA*/)
+    vdst = INST_AA(inst);
+    ILOGV("|move-exception v%d", vdst);
+    assert(self->exception != NULL);
+    SET_REGISTER(vdst, (u4)self->exception);
+    dvmClearException(self);
+    FINISH(1);
+OP_END
+
+/* File: c/OP_RETURN_VOID.c */
+HANDLE_OPCODE(OP_RETURN_VOID /**/)
+    ILOGV("|return-void");
+#ifndef NDEBUG
+    retval.j = 0xababababULL;    // placate valgrind
+#endif
+    GOTO_returnFromMethod();
+OP_END
+
+/* File: c/OP_RETURN.c */
+HANDLE_OPCODE(OP_RETURN /*vAA*/)
+    vsrc1 = INST_AA(inst);
+    ILOGV("|return%s v%d",
+        (INST_INST(inst) == OP_RETURN) ? "" : "-object", vsrc1);
+    retval.i = GET_REGISTER(vsrc1);
+    GOTO_returnFromMethod();
+OP_END
+
+/* File: c/OP_RETURN_WIDE.c */
+HANDLE_OPCODE(OP_RETURN_WIDE /*vAA*/)
+    vsrc1 = INST_AA(inst);
+    ILOGV("|return-wide v%d", vsrc1);
+    retval.j = GET_REGISTER_WIDE(vsrc1);
+    GOTO_returnFromMethod();
+OP_END
+
+/* File: c/OP_RETURN_OBJECT.c */
+/* File: c/OP_RETURN.c */
+HANDLE_OPCODE(OP_RETURN_OBJECT /*vAA*/)
+    vsrc1 = INST_AA(inst);
+    ILOGV("|return%s v%d",
+        (INST_INST(inst) == OP_RETURN) ? "" : "-object", vsrc1);
+    retval.i = GET_REGISTER(vsrc1);
+    GOTO_returnFromMethod();
+OP_END
+
+
+/* File: c/OP_CONST_4.c */
+HANDLE_OPCODE(OP_CONST_4 /*vA, #+B*/)
+    {
+        s4 tmp;
+
+        vdst = INST_A(inst);
+        tmp = (s4) (INST_B(inst) << 28) >> 28;  // sign extend 4-bit value
+        ILOGV("|const/4 v%d,#0x%02x", vdst, (s4)tmp);
+        SET_REGISTER(vdst, tmp);
+    }
+    FINISH(1);
+OP_END
+
+/* File: c/OP_CONST_16.c */
+HANDLE_OPCODE(OP_CONST_16 /*vAA, #+BBBB*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|const/16 v%d,#0x%04x", vdst, (s2)vsrc1);
+    SET_REGISTER(vdst, (s2) vsrc1);
+    FINISH(2);
+OP_END
+
+/* File: c/OP_CONST.c */
+HANDLE_OPCODE(OP_CONST /*vAA, #+BBBBBBBB*/)
+    {
+        u4 tmp;
+
+        vdst = INST_AA(inst);
+        tmp = FETCH(1);
+        tmp |= (u4)FETCH(2) << 16;
+        ILOGV("|const v%d,#0x%08x", vdst, tmp);
+        SET_REGISTER(vdst, tmp);
+    }
+    FINISH(3);
+OP_END
+
+/* File: c/OP_CONST_HIGH16.c */
+HANDLE_OPCODE(OP_CONST_HIGH16 /*vAA, #+BBBB0000*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|const/high16 v%d,#0x%04x0000", vdst, vsrc1);
+    SET_REGISTER(vdst, vsrc1 << 16);
+    FINISH(2);
+OP_END
+
+/* File: c/OP_CONST_WIDE_16.c */
+HANDLE_OPCODE(OP_CONST_WIDE_16 /*vAA, #+BBBB*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|const-wide/16 v%d,#0x%04x", vdst, (s2)vsrc1);
+    SET_REGISTER_WIDE(vdst, (s2)vsrc1);
+    FINISH(2);
+OP_END
+
+/* File: c/OP_CONST_WIDE_32.c */
+HANDLE_OPCODE(OP_CONST_WIDE_32 /*vAA, #+BBBBBBBB*/)
+    {
+        u4 tmp;
+
+        vdst = INST_AA(inst);
+        tmp = FETCH(1);
+        tmp |= (u4)FETCH(2) << 16;
+        ILOGV("|const-wide/32 v%d,#0x%08x", vdst, tmp);
+        SET_REGISTER_WIDE(vdst, (s4) tmp);
+    }
+    FINISH(3);
+OP_END
+
+/* File: c/OP_CONST_WIDE.c */
+HANDLE_OPCODE(OP_CONST_WIDE /*vAA, #+BBBBBBBBBBBBBBBB*/)
+    {
+        u8 tmp;
+
+        vdst = INST_AA(inst);
+        tmp = FETCH(1);
+        tmp |= (u8)FETCH(2) << 16;
+        tmp |= (u8)FETCH(3) << 32;
+        tmp |= (u8)FETCH(4) << 48;
+        ILOGV("|const-wide v%d,#0x%08llx", vdst, tmp);
+        SET_REGISTER_WIDE(vdst, tmp);
+    }
+    FINISH(5);
+OP_END
+
+/* File: c/OP_CONST_WIDE_HIGH16.c */
+HANDLE_OPCODE(OP_CONST_WIDE_HIGH16 /*vAA, #+BBBB000000000000*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|const-wide/high16 v%d,#0x%04x000000000000", vdst, vsrc1);
+    SET_REGISTER_WIDE(vdst, ((u8) vsrc1) << 48);
+    FINISH(2);
+OP_END
+
+/* File: c/OP_CONST_STRING.c */
+HANDLE_OPCODE(OP_CONST_STRING /*vAA, string@BBBB*/)
+    {
+        StringObject* strObj;
+
+        vdst = INST_AA(inst);
+        ref = FETCH(1);
+        ILOGV("|const-string v%d string@0x%04x", vdst, ref);
+        strObj = dvmDexGetResolvedString(methodClassDex, ref);
+        if (strObj == NULL) {
+            EXPORT_PC();
+            strObj = dvmResolveString(curMethod->clazz, ref);
+            if (strObj == NULL)
+                GOTO_exceptionThrown();
+        }
+        SET_REGISTER(vdst, (u4) strObj);
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_CONST_STRING_JUMBO.c */
+HANDLE_OPCODE(OP_CONST_STRING_JUMBO /*vAA, string@BBBBBBBB*/)
+    {
+        StringObject* strObj;
+        u4 tmp;
+
+        vdst = INST_AA(inst);
+        tmp = FETCH(1);
+        tmp |= (u4)FETCH(2) << 16;
+        ILOGV("|const-string/jumbo v%d string@0x%08x", vdst, tmp);
+        strObj = dvmDexGetResolvedString(methodClassDex, tmp);
+        if (strObj == NULL) {
+            EXPORT_PC();
+            strObj = dvmResolveString(curMethod->clazz, tmp);
+            if (strObj == NULL)
+                GOTO_exceptionThrown();
+        }
+        SET_REGISTER(vdst, (u4) strObj);
+    }
+    FINISH(3);
+OP_END
+
+/* File: c/OP_CONST_CLASS.c */
+HANDLE_OPCODE(OP_CONST_CLASS /*vAA, class@BBBB*/)
+    {
+        ClassObject* clazz;
+
+        vdst = INST_AA(inst);
+        ref = FETCH(1);
+        ILOGV("|const-class v%d class@0x%04x", vdst, ref);
+        clazz = dvmDexGetResolvedClass(methodClassDex, ref);
+        if (clazz == NULL) {
+            EXPORT_PC();
+            clazz = dvmResolveClass(curMethod->clazz, ref, true);
+            if (clazz == NULL)
+                GOTO_exceptionThrown();
+        }
+        SET_REGISTER(vdst, (u4) clazz);
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_MONITOR_ENTER.c */
+HANDLE_OPCODE(OP_MONITOR_ENTER /*vAA*/)
+    {
+        Object* obj;
+
+        vsrc1 = INST_AA(inst);
+        ILOGV("|monitor-enter v%d %s(0x%08x)",
+            vsrc1, kSpacing+6, GET_REGISTER(vsrc1));
+        obj = (Object*)GET_REGISTER(vsrc1);
+        if (!checkForNullExportPC(obj, fp, pc))
+            GOTO_exceptionThrown();
+        ILOGV("+ locking %p %s\n", obj, obj->clazz->descriptor);
+#ifdef WITH_MONITOR_TRACKING
+        EXPORT_PC();        /* need for stack trace */
+#endif
+        dvmLockObject(self, obj);
+#ifdef WITH_DEADLOCK_PREDICTION
+        if (dvmCheckException(self))
+            GOTO_exceptionThrown();
+#endif
+    }
+    FINISH(1);
+OP_END
+
+/* File: c/OP_MONITOR_EXIT.c */
+HANDLE_OPCODE(OP_MONITOR_EXIT /*vAA*/)
+    {
+        Object* obj;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);
+        ILOGV("|monitor-exit v%d %s(0x%08x)",
+            vsrc1, kSpacing+5, GET_REGISTER(vsrc1));
+        obj = (Object*)GET_REGISTER(vsrc1);
+        if (!checkForNull(obj)) {
+            /*
+             * The exception needs to be processed at the *following*
+             * instruction, not the current instruction (see the Dalvik
+             * spec).  Because we're jumping to an exception handler,
+             * we're not actually at risk of skipping an instruction
+             * by doing so.
+             */
+            ADJUST_PC(1);           /* monitor-exit width is 1 */
+            GOTO_exceptionThrown();
+        }
+        ILOGV("+ unlocking %p %s\n", obj, obj->clazz->descriptor);
+        if (!dvmUnlockObject(self, obj)) {
+            assert(dvmCheckException(self));
+            ADJUST_PC(1);
+            GOTO_exceptionThrown();
+        }
+    }
+    FINISH(1);
+OP_END
+
+/* File: c/OP_CHECK_CAST.c */
+HANDLE_OPCODE(OP_CHECK_CAST /*vAA, class@BBBB*/)
+    {
+        ClassObject* clazz;
+        Object* obj;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);
+        ref = FETCH(1);         /* class to check against */
+        ILOGV("|check-cast v%d,class@0x%04x", vsrc1, ref);
+
+        obj = (Object*)GET_REGISTER(vsrc1);
+        if (obj != NULL) {
+#if defined(WITH_EXTRA_OBJECT_VALIDATION)
+            if (!checkForNull(obj))
+                GOTO_exceptionThrown();
+#endif
+            clazz = dvmDexGetResolvedClass(methodClassDex, ref);
+            if (clazz == NULL) {
+                clazz = dvmResolveClass(curMethod->clazz, ref, false);
+                if (clazz == NULL)
+                    GOTO_exceptionThrown();
+            }
+            if (!dvmInstanceof(obj->clazz, clazz)) {
+                dvmThrowExceptionWithClassMessage(
+                    "Ljava/lang/ClassCastException;", obj->clazz->descriptor);
+                GOTO_exceptionThrown();
+            }
+        }
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_INSTANCE_OF.c */
+HANDLE_OPCODE(OP_INSTANCE_OF /*vA, vB, class@CCCC*/)
+    {
+        ClassObject* clazz;
+        Object* obj;
+
+        vdst = INST_A(inst);
+        vsrc1 = INST_B(inst);   /* object to check */
+        ref = FETCH(1);         /* class to check against */
+        ILOGV("|instance-of v%d,v%d,class@0x%04x", vdst, vsrc1, ref);
+
+        obj = (Object*)GET_REGISTER(vsrc1);
+        if (obj == NULL) {
+            SET_REGISTER(vdst, 0);
+        } else {
+#if defined(WITH_EXTRA_OBJECT_VALIDATION)
+            if (!checkForNullExportPC(obj, fp, pc))
+                GOTO_exceptionThrown();
+#endif
+            clazz = dvmDexGetResolvedClass(methodClassDex, ref);
+            if (clazz == NULL) {
+                EXPORT_PC();
+                clazz = dvmResolveClass(curMethod->clazz, ref, true);
+                if (clazz == NULL)
+                    GOTO_exceptionThrown();
+            }
+            SET_REGISTER(vdst, dvmInstanceof(obj->clazz, clazz));
+        }
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_ARRAY_LENGTH.c */
+HANDLE_OPCODE(OP_ARRAY_LENGTH /*vA, vB*/)
+    {
+        ArrayObject* arrayObj;
+
+        vdst = INST_A(inst);
+        vsrc1 = INST_B(inst);
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);
+        ILOGV("|array-length v%d,v%d  (%p)", vdst, vsrc1, arrayObj);
+        if (!checkForNullExportPC((Object*) arrayObj, fp, pc))
+            GOTO_exceptionThrown();
+        /* verifier guarantees this is an array reference */
+        SET_REGISTER(vdst, arrayObj->length);
+    }
+    FINISH(1);
+OP_END
+
+/* File: c/OP_NEW_INSTANCE.c */
+HANDLE_OPCODE(OP_NEW_INSTANCE /*vAA, class@BBBB*/)
+    {
+        ClassObject* clazz;
+        Object* newObj;
+
+        EXPORT_PC();
+
+        vdst = INST_AA(inst);
+        ref = FETCH(1);
+        ILOGV("|new-instance v%d,class@0x%04x", vdst, ref);
+        clazz = dvmDexGetResolvedClass(methodClassDex, ref);
+        if (clazz == NULL) {
+            clazz = dvmResolveClass(curMethod->clazz, ref, false);
+            if (clazz == NULL)
+                GOTO_exceptionThrown();
+        }
+
+        if (!dvmIsClassInitialized(clazz) && !dvmInitClass(clazz))
+            GOTO_exceptionThrown();
+
+        /*
+         * Note: the verifier can ensure that this never happens, allowing us
+         * to remove the check.  However, the spec requires we throw the
+         * exception at runtime, not verify time, so the verifier would
+         * need to replace the new-instance call with a magic "throw
+         * InstantiationError" instruction.
+         *
+         * Since this relies on the verifier, which is optional, we would
+         * also need a "new-instance-quick" instruction to identify instances
+         * that don't require the check.
+         */
+        if (dvmIsInterfaceClass(clazz) || dvmIsAbstractClass(clazz)) {
+            dvmThrowExceptionWithClassMessage("Ljava/lang/InstantiationError;",
+                clazz->descriptor);
+            GOTO_exceptionThrown();
+        }
+        newObj = dvmAllocObject(clazz, ALLOC_DONT_TRACK);
+        if (newObj == NULL)
+            GOTO_exceptionThrown();
+        SET_REGISTER(vdst, (u4) newObj);
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_NEW_ARRAY.c */
+HANDLE_OPCODE(OP_NEW_ARRAY /*vA, vB, class@CCCC*/)
+    {
+        ClassObject* arrayClass;
+        ArrayObject* newArray;
+        s4 length;
+
+        EXPORT_PC();
+
+        vdst = INST_A(inst);
+        vsrc1 = INST_B(inst);       /* length reg */
+        ref = FETCH(1);
+        ILOGV("|new-array v%d,v%d,class@0x%04x  (%d elements)",
+            vdst, vsrc1, ref, (s4) GET_REGISTER(vsrc1));
+        length = (s4) GET_REGISTER(vsrc1);
+        if (length < 0) {
+            dvmThrowException("Ljava/lang/NegativeArraySizeException;", NULL);
+            GOTO_exceptionThrown();
+        }
+        arrayClass = dvmDexGetResolvedClass(methodClassDex, ref);
+        if (arrayClass == NULL) {
+            arrayClass = dvmResolveClass(curMethod->clazz, ref, false);
+            if (arrayClass == NULL)
+                GOTO_exceptionThrown();
+        }
+        /* verifier guarantees this is an array class */
+        assert(dvmIsArrayClass(arrayClass));
+        assert(dvmIsClassInitialized(arrayClass));
+
+        newArray = dvmAllocArrayByClass(arrayClass, length, ALLOC_DONT_TRACK);
+        if (newArray == NULL)
+            GOTO_exceptionThrown();
+        SET_REGISTER(vdst, (u4) newArray);
+    }
+    FINISH(2);
+OP_END
+
+
+/* File: c/OP_FILLED_NEW_ARRAY.c */
+HANDLE_OPCODE(OP_FILLED_NEW_ARRAY /*vB, {vD, vE, vF, vG, vA}, class@CCCC*/)
+    GOTO_invoke(filledNewArray, false);
+OP_END
+
+/* File: c/OP_FILLED_NEW_ARRAY_RANGE.c */
+HANDLE_OPCODE(OP_FILLED_NEW_ARRAY_RANGE /*{vCCCC..v(CCCC+AA-1)}, class@BBBB*/)
+    GOTO_invoke(filledNewArray, true);
+OP_END
+
+/* File: c/OP_FILL_ARRAY_DATA.c */
+HANDLE_OPCODE(OP_FILL_ARRAY_DATA)   /*vAA, +BBBBBBBB*/
+    {
+        const u2* arrayData;
+        s4 offset;
+        ArrayObject* arrayObj;
+
+        EXPORT_PC();
+        vsrc1 = INST_AA(inst);
+        offset = FETCH(1) | (((s4) FETCH(2)) << 16);
+        ILOGV("|fill-array-data v%d +0x%04x", vsrc1, offset);
+        arrayData = pc + offset;       // offset in 16-bit units
+#ifndef NDEBUG
+        if (arrayData < curMethod->insns ||
+            arrayData >= curMethod->insns + dvmGetMethodInsnsSize(curMethod))
+        {
+            /* should have been caught in verifier */
+            dvmThrowException("Ljava/lang/InternalError;", 
+                              "bad fill array data");
+            GOTO_exceptionThrown();
+        }
+#endif
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);
+        if (!dvmInterpHandleFillArrayData(arrayObj, arrayData)) {
+            GOTO_exceptionThrown();
+        }
+        FINISH(3);
+    }
+OP_END
+
+/* File: c/OP_THROW.c */
+HANDLE_OPCODE(OP_THROW /*vAA*/)
+    {
+        Object* obj;
+
+        vsrc1 = INST_AA(inst);
+        ILOGV("|throw v%d  (%p)", vsrc1, (void*)GET_REGISTER(vsrc1));
+        obj = (Object*) GET_REGISTER(vsrc1);
+        if (!checkForNullExportPC(obj, fp, pc)) {
+            /* will throw a null pointer exception */
+            LOGVV("Bad exception\n");
+        } else {
+            /* use the requested exception */
+            dvmSetException(self, obj);
+        }
+        GOTO_exceptionThrown();
+    }
+OP_END
+
+/* File: c/OP_GOTO.c */
+HANDLE_OPCODE(OP_GOTO /*+AA*/)
+    vdst = INST_AA(inst);
+    if ((s1)vdst < 0)
+        ILOGV("|goto -0x%02x", -((s1)vdst));
+    else
+        ILOGV("|goto +0x%02x", ((s1)vdst));
+    ILOGV("> branch taken");
+    if ((s1)vdst < 0)
+        PERIODIC_CHECKS(kInterpEntryInstr, (s1)vdst);
+    FINISH((s1)vdst);
+OP_END
+
+/* File: c/OP_GOTO_16.c */
+HANDLE_OPCODE(OP_GOTO_16 /*+AAAA*/)
+    {
+        s4 offset = (s2) FETCH(1);          /* sign-extend next code unit */
+
+        if (offset < 0)
+            ILOGV("|goto/16 -0x%04x", -offset);
+        else
+            ILOGV("|goto/16 +0x%04x", offset);
+        ILOGV("> branch taken");
+        if (offset < 0)
+            PERIODIC_CHECKS(kInterpEntryInstr, offset);
+        FINISH(offset);
+    }
+OP_END
+
+/* File: c/OP_GOTO_32.c */
+HANDLE_OPCODE(OP_GOTO_32 /*+AAAAAAAA*/)
+    {
+        s4 offset = FETCH(1);               /* low-order 16 bits */
+        offset |= ((s4) FETCH(2)) << 16;    /* high-order 16 bits */
+
+        if (offset < 0)
+            ILOGV("|goto/32 -0x%08x", -offset);
+        else
+            ILOGV("|goto/32 +0x%08x", offset);
+        ILOGV("> branch taken");
+        if (offset <= 0)    /* allowed to branch to self */
+            PERIODIC_CHECKS(kInterpEntryInstr, offset);
+        FINISH(offset);
+    }
+OP_END
+
+/* File: c/OP_PACKED_SWITCH.c */
+HANDLE_OPCODE(OP_PACKED_SWITCH /*vAA, +BBBB*/)
+    {
+        const u2* switchData;
+        u4 testVal;
+        s4 offset;
+
+        vsrc1 = INST_AA(inst);
+        offset = FETCH(1) | (((s4) FETCH(2)) << 16);
+        ILOGV("|packed-switch v%d +0x%04x", vsrc1, vsrc2);
+        switchData = pc + offset;       // offset in 16-bit units
+#ifndef NDEBUG
+        if (switchData < curMethod->insns ||
+            switchData >= curMethod->insns + dvmGetMethodInsnsSize(curMethod))
+        {
+            /* should have been caught in verifier */
+            EXPORT_PC();
+            dvmThrowException("Ljava/lang/InternalError;", "bad packed switch");
+            GOTO_exceptionThrown();
+        }
+#endif
+        testVal = GET_REGISTER(vsrc1);
+
+        offset = dvmInterpHandlePackedSwitch(switchData, testVal);
+        ILOGV("> branch taken (0x%04x)\n", offset);
+        if (offset <= 0)  /* uncommon */
+            PERIODIC_CHECKS(kInterpEntryInstr, offset);
+        FINISH(offset);
+    }
+OP_END
+
+/* File: c/OP_SPARSE_SWITCH.c */
+HANDLE_OPCODE(OP_SPARSE_SWITCH /*vAA, +BBBB*/)
+    {
+        const u2* switchData;
+        u4 testVal;
+        s4 offset;
+
+        vsrc1 = INST_AA(inst);
+        offset = FETCH(1) | (((s4) FETCH(2)) << 16);
+        ILOGV("|sparse-switch v%d +0x%04x", vsrc1, vsrc2);
+        switchData = pc + offset;       // offset in 16-bit units
+#ifndef NDEBUG
+        if (switchData < curMethod->insns ||
+            switchData >= curMethod->insns + dvmGetMethodInsnsSize(curMethod))
+        {
+            /* should have been caught in verifier */
+            EXPORT_PC();
+            dvmThrowException("Ljava/lang/InternalError;", "bad sparse switch");
+            GOTO_exceptionThrown();
+        }
+#endif
+        testVal = GET_REGISTER(vsrc1);
+
+        offset = dvmInterpHandleSparseSwitch(switchData, testVal);
+        ILOGV("> branch taken (0x%04x)\n", offset);
+        if (offset <= 0)  /* uncommon */
+            PERIODIC_CHECKS(kInterpEntryInstr, offset);
+        FINISH(offset);
+    }
+OP_END
+
+/* File: c/OP_CMPL_FLOAT.c */
+HANDLE_OP_CMPX(OP_CMPL_FLOAT, "l-float", float, _FLOAT, -1)
+OP_END
+
+/* File: c/OP_CMPG_FLOAT.c */
+HANDLE_OP_CMPX(OP_CMPG_FLOAT, "g-float", float, _FLOAT, 1)
+OP_END
+
+/* File: c/OP_CMPL_DOUBLE.c */
+HANDLE_OP_CMPX(OP_CMPL_DOUBLE, "l-double", double, _DOUBLE, -1)
+OP_END
+
+/* File: c/OP_CMPG_DOUBLE.c */
+HANDLE_OP_CMPX(OP_CMPG_DOUBLE, "g-double", double, _DOUBLE, 1)
+OP_END
+
+/* File: c/OP_CMP_LONG.c */
+HANDLE_OP_CMPX(OP_CMP_LONG, "-long", s8, _WIDE, 0)
+OP_END
+
+/* File: c/OP_IF_EQ.c */
+HANDLE_OP_IF_XX(OP_IF_EQ, "eq", ==)
+OP_END
+
+/* File: c/OP_IF_NE.c */
+HANDLE_OP_IF_XX(OP_IF_NE, "ne", !=)
+OP_END
+
+/* File: c/OP_IF_LT.c */
+HANDLE_OP_IF_XX(OP_IF_LT, "lt", <)
+OP_END
+
+/* File: c/OP_IF_GE.c */
+HANDLE_OP_IF_XX(OP_IF_GE, "ge", >=)
+OP_END
+
+/* File: c/OP_IF_GT.c */
+HANDLE_OP_IF_XX(OP_IF_GT, "gt", >)
+OP_END
+
+/* File: c/OP_IF_LE.c */
+HANDLE_OP_IF_XX(OP_IF_LE, "le", <=)
+OP_END
+
+/* File: c/OP_IF_EQZ.c */
+HANDLE_OP_IF_XXZ(OP_IF_EQZ, "eqz", ==)
+OP_END
+
+/* File: c/OP_IF_NEZ.c */
+HANDLE_OP_IF_XXZ(OP_IF_NEZ, "nez", !=)
+OP_END
+
+/* File: c/OP_IF_LTZ.c */
+HANDLE_OP_IF_XXZ(OP_IF_LTZ, "ltz", <)
+OP_END
+
+/* File: c/OP_IF_GEZ.c */
+HANDLE_OP_IF_XXZ(OP_IF_GEZ, "gez", >=)
+OP_END
+
+/* File: c/OP_IF_GTZ.c */
+HANDLE_OP_IF_XXZ(OP_IF_GTZ, "gtz", >)
+OP_END
+
+/* File: c/OP_IF_LEZ.c */
+HANDLE_OP_IF_XXZ(OP_IF_LEZ, "lez", <=)
+OP_END
+
+/* File: c/OP_UNUSED_3E.c */
+HANDLE_OPCODE(OP_UNUSED_3E)
+OP_END
+
+/* File: c/OP_UNUSED_3F.c */
+HANDLE_OPCODE(OP_UNUSED_3F)
+OP_END
+
+/* File: c/OP_UNUSED_40.c */
+HANDLE_OPCODE(OP_UNUSED_40)
+OP_END
+
+/* File: c/OP_UNUSED_41.c */
+HANDLE_OPCODE(OP_UNUSED_41)
+OP_END
+
+/* File: c/OP_UNUSED_42.c */
+HANDLE_OPCODE(OP_UNUSED_42)
+OP_END
+
+/* File: c/OP_UNUSED_43.c */
+HANDLE_OPCODE(OP_UNUSED_43)
+OP_END
+
+/* File: c/OP_AGET.c */
+HANDLE_OP_AGET(OP_AGET, "", u4, )
+OP_END
+
+/* File: c/OP_AGET_WIDE.c */
+HANDLE_OP_AGET(OP_AGET_WIDE, "-wide", s8, _WIDE)
+OP_END
+
+/* File: c/OP_AGET_OBJECT.c */
+HANDLE_OP_AGET(OP_AGET_OBJECT, "-object", u4, )
+OP_END
+
+/* File: c/OP_AGET_BOOLEAN.c */
+HANDLE_OP_AGET(OP_AGET_BOOLEAN, "-boolean", u1, )
+OP_END
+
+/* File: c/OP_AGET_BYTE.c */
+HANDLE_OP_AGET(OP_AGET_BYTE, "-byte", s1, )
+OP_END
+
+/* File: c/OP_AGET_CHAR.c */
+HANDLE_OP_AGET(OP_AGET_CHAR, "-char", u2, )
+OP_END
+
+/* File: c/OP_AGET_SHORT.c */
+HANDLE_OP_AGET(OP_AGET_SHORT, "-short", s2, )
+OP_END
+
+/* File: c/OP_APUT.c */
+HANDLE_OP_APUT(OP_APUT, "", u4, )
+OP_END
+
+/* File: c/OP_APUT_WIDE.c */
+HANDLE_OP_APUT(OP_APUT_WIDE, "-wide", s8, _WIDE)
+OP_END
+
+/* File: c/OP_APUT_OBJECT.c */
+HANDLE_OPCODE(OP_APUT_OBJECT /*vAA, vBB, vCC*/)
+    {
+        ArrayObject* arrayObj;
+        Object* obj;
+        u2 arrayInfo;
+        EXPORT_PC();
+        vdst = INST_AA(inst);       /* AA: source value */
+        arrayInfo = FETCH(1);
+        vsrc1 = arrayInfo & 0xff;   /* BB: array ptr */
+        vsrc2 = arrayInfo >> 8;     /* CC: index */
+        ILOGV("|aput%s v%d,v%d,v%d", "-object", vdst, vsrc1, vsrc2);
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);
+        if (!checkForNull((Object*) arrayObj))
+            GOTO_exceptionThrown();
+        if (GET_REGISTER(vsrc2) >= arrayObj->length) {
+            dvmThrowException("Ljava/lang/ArrayIndexOutOfBoundsException;",
+                NULL);
+            GOTO_exceptionThrown();
+        }
+        obj = (Object*) GET_REGISTER(vdst);
+        if (obj != NULL) {
+            if (!checkForNull(obj))
+                GOTO_exceptionThrown();
+            if (!dvmCanPutArrayElement(obj->clazz, arrayObj->obj.clazz)) {
+                LOGV("Can't put a '%s'(%p) into array type='%s'(%p)\n",
+                    obj->clazz->descriptor, obj,
+                    arrayObj->obj.clazz->descriptor, arrayObj);
+                //dvmDumpClass(obj->clazz);
+                //dvmDumpClass(arrayObj->obj.clazz);
+                dvmThrowException("Ljava/lang/ArrayStoreException;", NULL);
+                GOTO_exceptionThrown();
+            }
+        }
+        ILOGV("+ APUT[%d]=0x%08x", GET_REGISTER(vsrc2), GET_REGISTER(vdst));
+        ((u4*) arrayObj->contents)[GET_REGISTER(vsrc2)] =
+            GET_REGISTER(vdst);
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_APUT_BOOLEAN.c */
+HANDLE_OP_APUT(OP_APUT_BOOLEAN, "-boolean", u1, )
+OP_END
+
+/* File: c/OP_APUT_BYTE.c */
+HANDLE_OP_APUT(OP_APUT_BYTE, "-byte", s1, )
+OP_END
+
+/* File: c/OP_APUT_CHAR.c */
+HANDLE_OP_APUT(OP_APUT_CHAR, "-char", u2, )
+OP_END
+
+/* File: c/OP_APUT_SHORT.c */
+HANDLE_OP_APUT(OP_APUT_SHORT, "-short", s2, )
+OP_END
+
+/* File: c/OP_IGET.c */
+HANDLE_IGET_X(OP_IGET,                  "", Int, )
+OP_END
+
+/* File: c/OP_IGET_WIDE.c */
+HANDLE_IGET_X(OP_IGET_WIDE,             "-wide", Long, _WIDE)
+OP_END
+
+/* File: c/OP_IGET_OBJECT.c */
+HANDLE_IGET_X(OP_IGET_OBJECT,           "-object", Object, _AS_OBJECT)
+OP_END
+
+/* File: c/OP_IGET_BOOLEAN.c */
+HANDLE_IGET_X(OP_IGET_BOOLEAN,          "", Int, )
+OP_END
+
+/* File: c/OP_IGET_BYTE.c */
+HANDLE_IGET_X(OP_IGET_BYTE,             "", Int, )
+OP_END
+
+/* File: c/OP_IGET_CHAR.c */
+HANDLE_IGET_X(OP_IGET_CHAR,             "", Int, )
+OP_END
+
+/* File: c/OP_IGET_SHORT.c */
+HANDLE_IGET_X(OP_IGET_SHORT,            "", Int, )
+OP_END
+
+/* File: c/OP_IPUT.c */
+HANDLE_IPUT_X(OP_IPUT,                  "", Int, )
+OP_END
+
+/* File: c/OP_IPUT_WIDE.c */
+HANDLE_IPUT_X(OP_IPUT_WIDE,             "-wide", Long, _WIDE)
+OP_END
+
+/* File: c/OP_IPUT_OBJECT.c */
+/*
+ * The VM spec says we should verify that the reference being stored into
+ * the field is assignment compatible.  In practice, many popular VMs don't
+ * do this because it slows down a very common operation.  It's not so bad
+ * for us, since "dexopt" quickens it whenever possible, but it's still an
+ * issue.
+ *
+ * To make this spec-complaint, we'd need to add a ClassObject pointer to
+ * the Field struct, resolve the field's type descriptor at link or class
+ * init time, and then verify the type here.
+ */
+HANDLE_IPUT_X(OP_IPUT_OBJECT,           "-object", Object, _AS_OBJECT)
+OP_END
+
+/* File: c/OP_IPUT_BOOLEAN.c */
+HANDLE_IPUT_X(OP_IPUT_BOOLEAN,          "", Int, )
+OP_END
+
+/* File: c/OP_IPUT_BYTE.c */
+HANDLE_IPUT_X(OP_IPUT_BYTE,             "", Int, )
+OP_END
+
+/* File: c/OP_IPUT_CHAR.c */
+HANDLE_IPUT_X(OP_IPUT_CHAR,             "", Int, )
+OP_END
+
+/* File: c/OP_IPUT_SHORT.c */
+HANDLE_IPUT_X(OP_IPUT_SHORT,            "", Int, )
+OP_END
+
+/* File: c/OP_SGET.c */
+HANDLE_SGET_X(OP_SGET,                  "", Int, )
+OP_END
+
+/* File: c/OP_SGET_WIDE.c */
+HANDLE_SGET_X(OP_SGET_WIDE,             "-wide", Long, _WIDE)
+OP_END
+
+/* File: c/OP_SGET_OBJECT.c */
+HANDLE_SGET_X(OP_SGET_OBJECT,           "-object", Object, _AS_OBJECT)
+OP_END
+
+/* File: c/OP_SGET_BOOLEAN.c */
+HANDLE_SGET_X(OP_SGET_BOOLEAN,          "", Int, )
+OP_END
+
+/* File: c/OP_SGET_BYTE.c */
+HANDLE_SGET_X(OP_SGET_BYTE,             "", Int, )
+OP_END
+
+/* File: c/OP_SGET_CHAR.c */
+HANDLE_SGET_X(OP_SGET_CHAR,             "", Int, )
+OP_END
+
+/* File: c/OP_SGET_SHORT.c */
+HANDLE_SGET_X(OP_SGET_SHORT,            "", Int, )
+OP_END
+
+/* File: c/OP_SPUT.c */
+HANDLE_SPUT_X(OP_SPUT,                  "", Int, )
+OP_END
+
+/* File: c/OP_SPUT_WIDE.c */
+HANDLE_SPUT_X(OP_SPUT_WIDE,             "-wide", Long, _WIDE)
+OP_END
+
+/* File: c/OP_SPUT_OBJECT.c */
+HANDLE_SPUT_X(OP_SPUT_OBJECT,           "-object", Object, _AS_OBJECT)
+OP_END
+
+/* File: c/OP_SPUT_BOOLEAN.c */
+HANDLE_SPUT_X(OP_SPUT_BOOLEAN,          "", Int, )
+OP_END
+
+/* File: c/OP_SPUT_BYTE.c */
+HANDLE_SPUT_X(OP_SPUT_BYTE,             "", Int, )
+OP_END
+
+/* File: c/OP_SPUT_CHAR.c */
+HANDLE_SPUT_X(OP_SPUT_CHAR,             "", Int, )
+OP_END
+
+/* File: c/OP_SPUT_SHORT.c */
+HANDLE_SPUT_X(OP_SPUT_SHORT,            "", Int, )
+OP_END
+
+/* File: c/OP_INVOKE_VIRTUAL.c */
+HANDLE_OPCODE(OP_INVOKE_VIRTUAL /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeVirtual, false);
+OP_END
+
+/* File: c/OP_INVOKE_SUPER.c */
+HANDLE_OPCODE(OP_INVOKE_SUPER /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeSuper, false);
+OP_END
+
+/* File: c/OP_INVOKE_DIRECT.c */
+HANDLE_OPCODE(OP_INVOKE_DIRECT /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeDirect, false);
+OP_END
+
+/* File: c/OP_INVOKE_STATIC.c */
+HANDLE_OPCODE(OP_INVOKE_STATIC /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeStatic, false);
+OP_END
+
+/* File: c/OP_INVOKE_INTERFACE.c */
+HANDLE_OPCODE(OP_INVOKE_INTERFACE /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeInterface, false);
+OP_END
+
+/* File: c/OP_UNUSED_73.c */
+HANDLE_OPCODE(OP_UNUSED_73)
+OP_END
+
+/* File: c/OP_INVOKE_VIRTUAL_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_VIRTUAL_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeVirtual, true);
+OP_END
+
+/* File: c/OP_INVOKE_SUPER_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_SUPER_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeSuper, true);
+OP_END
+
+/* File: c/OP_INVOKE_DIRECT_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_DIRECT_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeDirect, true);
+OP_END
+
+/* File: c/OP_INVOKE_STATIC_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_STATIC_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeStatic, true);
+OP_END
+
+/* File: c/OP_INVOKE_INTERFACE_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_INTERFACE_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeInterface, true);
+OP_END
+
+/* File: c/OP_UNUSED_79.c */
+HANDLE_OPCODE(OP_UNUSED_79)
+OP_END
+
+/* File: c/OP_UNUSED_7A.c */
+HANDLE_OPCODE(OP_UNUSED_7A)
+OP_END
+
+/* File: c/OP_NEG_INT.c */
+HANDLE_UNOP(OP_NEG_INT, "neg-int", -, , )
+OP_END
+
+/* File: c/OP_NOT_INT.c */
+HANDLE_UNOP(OP_NOT_INT, "not-int", , ^ 0xffffffff, )
+OP_END
+
+/* File: c/OP_NEG_LONG.c */
+HANDLE_UNOP(OP_NEG_LONG, "neg-long", -, , _WIDE)
+OP_END
+
+/* File: c/OP_NOT_LONG.c */
+HANDLE_UNOP(OP_NOT_LONG, "not-long", , ^ 0xffffffffffffffffULL, _WIDE)
+OP_END
+
+/* File: c/OP_NEG_FLOAT.c */
+HANDLE_UNOP(OP_NEG_FLOAT, "neg-float", -, , _FLOAT)
+OP_END
+
+/* File: c/OP_NEG_DOUBLE.c */
+HANDLE_UNOP(OP_NEG_DOUBLE, "neg-double", -, , _DOUBLE)
+OP_END
+
+/* File: c/OP_INT_TO_LONG.c */
+HANDLE_NUMCONV(OP_INT_TO_LONG,          "int-to-long", _INT, _WIDE)
+OP_END
+
+/* File: c/OP_INT_TO_FLOAT.c */
+HANDLE_NUMCONV(OP_INT_TO_FLOAT,         "int-to-float", _INT, _FLOAT)
+OP_END
+
+/* File: c/OP_INT_TO_DOUBLE.c */
+HANDLE_NUMCONV(OP_INT_TO_DOUBLE,        "int-to-double", _INT, _DOUBLE)
+OP_END
+
+/* File: c/OP_LONG_TO_INT.c */
+HANDLE_NUMCONV(OP_LONG_TO_INT,          "long-to-int", _WIDE, _INT)
+OP_END
+
+/* File: c/OP_LONG_TO_FLOAT.c */
+HANDLE_NUMCONV(OP_LONG_TO_FLOAT,        "long-to-float", _WIDE, _FLOAT)
+OP_END
+
+/* File: c/OP_LONG_TO_DOUBLE.c */
+HANDLE_NUMCONV(OP_LONG_TO_DOUBLE,       "long-to-double", _WIDE, _DOUBLE)
+OP_END
+
+/* File: c/OP_FLOAT_TO_INT.c */
+HANDLE_FLOAT_TO_INT(OP_FLOAT_TO_INT,    "float-to-int",
+    float, _FLOAT, s4, _INT)
+OP_END
+
+/* File: c/OP_FLOAT_TO_LONG.c */
+HANDLE_FLOAT_TO_INT(OP_FLOAT_TO_LONG,   "float-to-long",
+    float, _FLOAT, s8, _WIDE)
+OP_END
+
+/* File: c/OP_FLOAT_TO_DOUBLE.c */
+HANDLE_NUMCONV(OP_FLOAT_TO_DOUBLE,      "float-to-double", _FLOAT, _DOUBLE)
+OP_END
+
+/* File: c/OP_DOUBLE_TO_INT.c */
+HANDLE_FLOAT_TO_INT(OP_DOUBLE_TO_INT,   "double-to-int",
+    double, _DOUBLE, s4, _INT)
+OP_END
+
+/* File: c/OP_DOUBLE_TO_LONG.c */
+HANDLE_FLOAT_TO_INT(OP_DOUBLE_TO_LONG,  "double-to-long",
+    double, _DOUBLE, s8, _WIDE)
+OP_END
+
+/* File: c/OP_DOUBLE_TO_FLOAT.c */
+HANDLE_NUMCONV(OP_DOUBLE_TO_FLOAT,      "double-to-float", _DOUBLE, _FLOAT)
+OP_END
+
+/* File: c/OP_INT_TO_BYTE.c */
+HANDLE_INT_TO_SMALL(OP_INT_TO_BYTE,     "byte", s1)
+OP_END
+
+/* File: c/OP_INT_TO_CHAR.c */
+HANDLE_INT_TO_SMALL(OP_INT_TO_CHAR,     "char", u2)
+OP_END
+
+/* File: c/OP_INT_TO_SHORT.c */
+HANDLE_INT_TO_SMALL(OP_INT_TO_SHORT,    "short", s2)    /* want sign bit */
+OP_END
+
+/* File: c/OP_ADD_INT.c */
+HANDLE_OP_X_INT(OP_ADD_INT, "add", +, 0)
+OP_END
+
+/* File: c/OP_SUB_INT.c */
+HANDLE_OP_X_INT(OP_SUB_INT, "sub", -, 0)
+OP_END
+
+/* File: c/OP_MUL_INT.c */
+HANDLE_OP_X_INT(OP_MUL_INT, "mul", *, 0)
+OP_END
+
+/* File: c/OP_DIV_INT.c */
+HANDLE_OP_X_INT(OP_DIV_INT, "div", /, 1)
+OP_END
+
+/* File: c/OP_REM_INT.c */
+HANDLE_OP_X_INT(OP_REM_INT, "rem", %, 2)
+OP_END
+
+/* File: c/OP_AND_INT.c */
+HANDLE_OP_X_INT(OP_AND_INT, "and", &, 0)
+OP_END
+
+/* File: c/OP_OR_INT.c */
+HANDLE_OP_X_INT(OP_OR_INT,  "or",  |, 0)
+OP_END
+
+/* File: c/OP_XOR_INT.c */
+HANDLE_OP_X_INT(OP_XOR_INT, "xor", ^, 0)
+OP_END
+
+/* File: c/OP_SHL_INT.c */
+HANDLE_OP_SHX_INT(OP_SHL_INT, "shl", (s4), <<)
+OP_END
+
+/* File: c/OP_SHR_INT.c */
+HANDLE_OP_SHX_INT(OP_SHR_INT, "shr", (s4), >>)
+OP_END
+
+/* File: c/OP_USHR_INT.c */
+HANDLE_OP_SHX_INT(OP_USHR_INT, "ushr", (u4), >>)
+OP_END
+
+/* File: c/OP_ADD_LONG.c */
+HANDLE_OP_X_LONG(OP_ADD_LONG, "add", +, 0)
+OP_END
+
+/* File: c/OP_SUB_LONG.c */
+HANDLE_OP_X_LONG(OP_SUB_LONG, "sub", -, 0)
+OP_END
+
+/* File: c/OP_MUL_LONG.c */
+HANDLE_OP_X_LONG(OP_MUL_LONG, "mul", *, 0)
+OP_END
+
+/* File: c/OP_DIV_LONG.c */
+HANDLE_OP_X_LONG(OP_DIV_LONG, "div", /, 1)
+OP_END
+
+/* File: c/OP_REM_LONG.c */
+HANDLE_OP_X_LONG(OP_REM_LONG, "rem", %, 2)
+OP_END
+
+/* File: c/OP_AND_LONG.c */
+HANDLE_OP_X_LONG(OP_AND_LONG, "and", &, 0)
+OP_END
+
+/* File: c/OP_OR_LONG.c */
+HANDLE_OP_X_LONG(OP_OR_LONG,  "or", |, 0)
+OP_END
+
+/* File: c/OP_XOR_LONG.c */
+HANDLE_OP_X_LONG(OP_XOR_LONG, "xor", ^, 0)
+OP_END
+
+/* File: c/OP_SHL_LONG.c */
+HANDLE_OP_SHX_LONG(OP_SHL_LONG, "shl", (s8), <<)
+OP_END
+
+/* File: c/OP_SHR_LONG.c */
+HANDLE_OP_SHX_LONG(OP_SHR_LONG, "shr", (s8), >>)
+OP_END
+
+/* File: c/OP_USHR_LONG.c */
+HANDLE_OP_SHX_LONG(OP_USHR_LONG, "ushr", (u8), >>)
+OP_END
+
+/* File: c/OP_ADD_FLOAT.c */
+HANDLE_OP_X_FLOAT(OP_ADD_FLOAT, "add", +)
+OP_END
+
+/* File: c/OP_SUB_FLOAT.c */
+HANDLE_OP_X_FLOAT(OP_SUB_FLOAT, "sub", -)
+OP_END
+
+/* File: c/OP_MUL_FLOAT.c */
+HANDLE_OP_X_FLOAT(OP_MUL_FLOAT, "mul", *)
+OP_END
+
+/* File: c/OP_DIV_FLOAT.c */
+HANDLE_OP_X_FLOAT(OP_DIV_FLOAT, "div", /)
+OP_END
+
+/* File: c/OP_REM_FLOAT.c */
+HANDLE_OPCODE(OP_REM_FLOAT /*vAA, vBB, vCC*/)
+    {
+        u2 srcRegs;
+        vdst = INST_AA(inst);
+        srcRegs = FETCH(1);
+        vsrc1 = srcRegs & 0xff;
+        vsrc2 = srcRegs >> 8;
+        ILOGV("|%s-float v%d,v%d,v%d", "mod", vdst, vsrc1, vsrc2);
+        SET_REGISTER_FLOAT(vdst,
+            fmodf(GET_REGISTER_FLOAT(vsrc1), GET_REGISTER_FLOAT(vsrc2)));
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_ADD_DOUBLE.c */
+HANDLE_OP_X_DOUBLE(OP_ADD_DOUBLE, "add", +)
+OP_END
+
+/* File: c/OP_SUB_DOUBLE.c */
+HANDLE_OP_X_DOUBLE(OP_SUB_DOUBLE, "sub", -)
+OP_END
+
+/* File: c/OP_MUL_DOUBLE.c */
+HANDLE_OP_X_DOUBLE(OP_MUL_DOUBLE, "mul", *)
+OP_END
+
+/* File: c/OP_DIV_DOUBLE.c */
+HANDLE_OP_X_DOUBLE(OP_DIV_DOUBLE, "div", /)
+OP_END
+
+/* File: c/OP_REM_DOUBLE.c */
+HANDLE_OPCODE(OP_REM_DOUBLE /*vAA, vBB, vCC*/)
+    {
+        u2 srcRegs;
+        vdst = INST_AA(inst);
+        srcRegs = FETCH(1);
+        vsrc1 = srcRegs & 0xff;
+        vsrc2 = srcRegs >> 8;
+        ILOGV("|%s-double v%d,v%d,v%d", "mod", vdst, vsrc1, vsrc2);
+        SET_REGISTER_DOUBLE(vdst,
+            fmod(GET_REGISTER_DOUBLE(vsrc1), GET_REGISTER_DOUBLE(vsrc2)));
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_ADD_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_ADD_INT_2ADDR, "add", +, 0)
+OP_END
+
+/* File: c/OP_SUB_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_SUB_INT_2ADDR, "sub", -, 0)
+OP_END
+
+/* File: c/OP_MUL_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_MUL_INT_2ADDR, "mul", *, 0)
+OP_END
+
+/* File: c/OP_DIV_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_DIV_INT_2ADDR, "div", /, 1)
+OP_END
+
+/* File: c/OP_REM_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_REM_INT_2ADDR, "rem", %, 2)
+OP_END
+
+/* File: c/OP_AND_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_AND_INT_2ADDR, "and", &, 0)
+OP_END
+
+/* File: c/OP_OR_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_OR_INT_2ADDR,  "or", |, 0)
+OP_END
+
+/* File: c/OP_XOR_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_XOR_INT_2ADDR, "xor", ^, 0)
+OP_END
+
+/* File: c/OP_SHL_INT_2ADDR.c */
+HANDLE_OP_SHX_INT_2ADDR(OP_SHL_INT_2ADDR, "shl", (s4), <<)
+OP_END
+
+/* File: c/OP_SHR_INT_2ADDR.c */
+HANDLE_OP_SHX_INT_2ADDR(OP_SHR_INT_2ADDR, "shr", (s4), >>)
+OP_END
+
+/* File: c/OP_USHR_INT_2ADDR.c */
+HANDLE_OP_SHX_INT_2ADDR(OP_USHR_INT_2ADDR, "ushr", (u4), >>)
+OP_END
+
+/* File: c/OP_ADD_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_ADD_LONG_2ADDR, "add", +, 0)
+OP_END
+
+/* File: c/OP_SUB_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_SUB_LONG_2ADDR, "sub", -, 0)
+OP_END
+
+/* File: c/OP_MUL_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_MUL_LONG_2ADDR, "mul", *, 0)
+OP_END
+
+/* File: c/OP_DIV_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_DIV_LONG_2ADDR, "div", /, 1)
+OP_END
+
+/* File: c/OP_REM_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_REM_LONG_2ADDR, "rem", %, 2)
+OP_END
+
+/* File: c/OP_AND_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_AND_LONG_2ADDR, "and", &, 0)
+OP_END
+
+/* File: c/OP_OR_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_OR_LONG_2ADDR,  "or", |, 0)
+OP_END
+
+/* File: c/OP_XOR_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_XOR_LONG_2ADDR, "xor", ^, 0)
+OP_END
+
+/* File: c/OP_SHL_LONG_2ADDR.c */
+HANDLE_OP_SHX_LONG_2ADDR(OP_SHL_LONG_2ADDR, "shl", (s8), <<)
+OP_END
+
+/* File: c/OP_SHR_LONG_2ADDR.c */
+HANDLE_OP_SHX_LONG_2ADDR(OP_SHR_LONG_2ADDR, "shr", (s8), >>)
+OP_END
+
+/* File: c/OP_USHR_LONG_2ADDR.c */
+HANDLE_OP_SHX_LONG_2ADDR(OP_USHR_LONG_2ADDR, "ushr", (u8), >>)
+OP_END
+
+/* File: c/OP_ADD_FLOAT_2ADDR.c */
+HANDLE_OP_X_FLOAT_2ADDR(OP_ADD_FLOAT_2ADDR, "add", +)
+OP_END
+
+/* File: c/OP_SUB_FLOAT_2ADDR.c */
+HANDLE_OP_X_FLOAT_2ADDR(OP_SUB_FLOAT_2ADDR, "sub", -)
+OP_END
+
+/* File: c/OP_MUL_FLOAT_2ADDR.c */
+HANDLE_OP_X_FLOAT_2ADDR(OP_MUL_FLOAT_2ADDR, "mul", *)
+OP_END
+
+/* File: c/OP_DIV_FLOAT_2ADDR.c */
+HANDLE_OP_X_FLOAT_2ADDR(OP_DIV_FLOAT_2ADDR, "div", /)
+OP_END
+
+/* File: c/OP_REM_FLOAT_2ADDR.c */
+HANDLE_OPCODE(OP_REM_FLOAT_2ADDR /*vA, vB*/)
+    vdst = INST_A(inst);
+    vsrc1 = INST_B(inst);
+    ILOGV("|%s-float-2addr v%d,v%d", "mod", vdst, vsrc1);
+    SET_REGISTER_FLOAT(vdst,
+        fmodf(GET_REGISTER_FLOAT(vdst), GET_REGISTER_FLOAT(vsrc1)));
+    FINISH(1);
+OP_END
+
+/* File: c/OP_ADD_DOUBLE_2ADDR.c */
+HANDLE_OP_X_DOUBLE_2ADDR(OP_ADD_DOUBLE_2ADDR, "add", +)
+OP_END
+
+/* File: c/OP_SUB_DOUBLE_2ADDR.c */
+HANDLE_OP_X_DOUBLE_2ADDR(OP_SUB_DOUBLE_2ADDR, "sub", -)
+OP_END
+
+/* File: c/OP_MUL_DOUBLE_2ADDR.c */
+HANDLE_OP_X_DOUBLE_2ADDR(OP_MUL_DOUBLE_2ADDR, "mul", *)
+OP_END
+
+/* File: c/OP_DIV_DOUBLE_2ADDR.c */
+HANDLE_OP_X_DOUBLE_2ADDR(OP_DIV_DOUBLE_2ADDR, "div", /)
+OP_END
+
+/* File: c/OP_REM_DOUBLE_2ADDR.c */
+HANDLE_OPCODE(OP_REM_DOUBLE_2ADDR /*vA, vB*/)
+    vdst = INST_A(inst);
+    vsrc1 = INST_B(inst);
+    ILOGV("|%s-double-2addr v%d,v%d", "mod", vdst, vsrc1);
+    SET_REGISTER_DOUBLE(vdst,
+        fmod(GET_REGISTER_DOUBLE(vdst), GET_REGISTER_DOUBLE(vsrc1)));
+    FINISH(1);
+OP_END
+
+/* File: c/OP_ADD_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_ADD_INT_LIT16, "add", +, 0)
+OP_END
+
+/* File: c/OP_RSUB_INT.c */
+HANDLE_OPCODE(OP_RSUB_INT /*vA, vB, #+CCCC*/)
+    {
+        vdst = INST_A(inst);
+        vsrc1 = INST_B(inst);
+        vsrc2 = FETCH(1);
+        ILOGV("|rsub-int v%d,v%d,#+0x%04x", vdst, vsrc1, vsrc2);
+        SET_REGISTER(vdst, (s2) vsrc2 - (s4) GET_REGISTER(vsrc1));
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_MUL_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_MUL_INT_LIT16, "mul", *, 0)
+OP_END
+
+/* File: c/OP_DIV_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_DIV_INT_LIT16, "div", /, 1)
+OP_END
+
+/* File: c/OP_REM_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_REM_INT_LIT16, "rem", %, 2)
+OP_END
+
+/* File: c/OP_AND_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_AND_INT_LIT16, "and", &, 0)
+OP_END
+
+/* File: c/OP_OR_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_OR_INT_LIT16,  "or",  |, 0)
+OP_END
+
+/* File: c/OP_XOR_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_XOR_INT_LIT16, "xor", ^, 0)
+OP_END
+
+/* File: c/OP_ADD_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_ADD_INT_LIT8,   "add", +, 0)
+OP_END
+
+/* File: c/OP_RSUB_INT_LIT8.c */
+HANDLE_OPCODE(OP_RSUB_INT_LIT8 /*vAA, vBB, #+CC*/)
+    {
+        u2 litInfo;
+        vdst = INST_AA(inst);
+        litInfo = FETCH(1);
+        vsrc1 = litInfo & 0xff;
+        vsrc2 = litInfo >> 8;
+        ILOGV("|%s-int/lit8 v%d,v%d,#+0x%02x", "rsub", vdst, vsrc1, vsrc2);
+        SET_REGISTER(vdst, (s1) vsrc2 - (s4) GET_REGISTER(vsrc1));
+    }
+    FINISH(2);
+OP_END
+
+/* File: c/OP_MUL_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_MUL_INT_LIT8,   "mul", *, 0)
+OP_END
+
+/* File: c/OP_DIV_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_DIV_INT_LIT8,   "div", /, 1)
+OP_END
+
+/* File: c/OP_REM_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_REM_INT_LIT8,   "rem", %, 2)
+OP_END
+
+/* File: c/OP_AND_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_AND_INT_LIT8,   "and", &, 0)
+OP_END
+
+/* File: c/OP_OR_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_OR_INT_LIT8,    "or",  |, 0)
+OP_END
+
+/* File: c/OP_XOR_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_XOR_INT_LIT8,   "xor", ^, 0)
+OP_END
+
+/* File: c/OP_SHL_INT_LIT8.c */
+HANDLE_OP_SHX_INT_LIT8(OP_SHL_INT_LIT8,   "shl", (s4), <<)
+OP_END
+
+/* File: c/OP_SHR_INT_LIT8.c */
+HANDLE_OP_SHX_INT_LIT8(OP_SHR_INT_LIT8,   "shr", (s4), >>)
+OP_END
+
+/* File: c/OP_USHR_INT_LIT8.c */
+HANDLE_OP_SHX_INT_LIT8(OP_USHR_INT_LIT8,  "ushr", (u4), >>)
+OP_END
+
+/* File: c/OP_UNUSED_E3.c */
+HANDLE_OPCODE(OP_UNUSED_E3)
+OP_END
+
+/* File: c/OP_UNUSED_E4.c */
+HANDLE_OPCODE(OP_UNUSED_E4)
+OP_END
+
+/* File: c/OP_UNUSED_E5.c */
+HANDLE_OPCODE(OP_UNUSED_E5)
+OP_END
+
+/* File: c/OP_UNUSED_E6.c */
+HANDLE_OPCODE(OP_UNUSED_E6)
+OP_END
+
+/* File: c/OP_UNUSED_E7.c */
+HANDLE_OPCODE(OP_UNUSED_E7)
+OP_END
+
+/* File: c/OP_UNUSED_E8.c */
+HANDLE_OPCODE(OP_UNUSED_E8)
+OP_END
+
+/* File: c/OP_UNUSED_E9.c */
+HANDLE_OPCODE(OP_UNUSED_E9)
+OP_END
+
+/* File: c/OP_UNUSED_EA.c */
+HANDLE_OPCODE(OP_UNUSED_EA)
+OP_END
+
+/* File: c/OP_UNUSED_EB.c */
+HANDLE_OPCODE(OP_UNUSED_EB)
+OP_END
+
+/* File: c/OP_UNUSED_EC.c */
+HANDLE_OPCODE(OP_UNUSED_EC)
+OP_END
+
+/* File: c/OP_UNUSED_ED.c */
+HANDLE_OPCODE(OP_UNUSED_ED)
+OP_END
+
+/* File: c/OP_EXECUTE_INLINE.c */
+HANDLE_OPCODE(OP_EXECUTE_INLINE /*vB, {vD, vE, vF, vG}, inline@CCCC*/)
+    {
+        /*
+         * This has the same form as other method calls, but we ignore
+         * the 5th argument (vA).  This is chiefly because the first four
+         * arguments to a function on ARM are in registers.
+         *
+         * We only set the arguments that are actually used, leaving
+         * the rest uninitialized.  We're assuming that, if the method
+         * needs them, they'll be specified in the call.
+         *
+         * This annoys gcc when optimizations are enabled, causing a
+         * "may be used uninitialized" warning.  We can quiet the warnings
+         * for a slight penalty (5%: 373ns vs. 393ns on empty method).  Note
+         * that valgrind is perfectly happy with this arrangement, because
+         * the uninitialiezd values are never actually used.
+         */
+        u4 arg0, arg1, arg2, arg3;
+        //arg0 = arg1 = arg2 = arg3 = 0;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_B(inst);       /* #of args */
+        ref = FETCH(1);             /* inline call "ref" */
+        vdst = FETCH(2);            /* 0-4 register indices */
+        ILOGV("|execute-inline args=%d @%d {regs=0x%04x}",
+            vsrc1, ref, vdst);
+
+        assert((vdst >> 16) == 0);  // 16-bit type -or- high 16 bits clear
+        assert(vsrc1 <= 4);
+
+        switch (vsrc1) {
+        case 4:
+            arg3 = GET_REGISTER(vdst >> 12);
+            /* fall through */
+        case 3:
+            arg2 = GET_REGISTER((vdst & 0x0f00) >> 8);
+            /* fall through */
+        case 2:
+            arg1 = GET_REGISTER((vdst & 0x00f0) >> 4);
+            /* fall through */
+        case 1:
+            arg0 = GET_REGISTER(vdst & 0x0f);
+            /* fall through */
+        default:        // case 0
+            ;
+        }
+
+#if INTERP_TYPE == INTERP_DBG
+        if (!dvmPerformInlineOp4Dbg(arg0, arg1, arg2, arg3, &retval, ref))
+            GOTO_exceptionThrown();
+#else
+        if (!dvmPerformInlineOp4Std(arg0, arg1, arg2, arg3, &retval, ref))
+            GOTO_exceptionThrown();
+#endif
+    }
+    FINISH(3);
+OP_END
+
+/* File: c/OP_UNUSED_EF.c */
+HANDLE_OPCODE(OP_UNUSED_EF)
+OP_END
+
+/* File: c/OP_INVOKE_DIRECT_EMPTY.c */
+HANDLE_OPCODE(OP_INVOKE_DIRECT_EMPTY /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+#if INTERP_TYPE != INTERP_DBG
+    //LOGI("Ignoring empty\n");
+    FINISH(3);
+#else
+    if (!gDvm.debuggerActive) {
+        //LOGI("Skipping empty\n");
+        FINISH(3);      // don't want it to show up in profiler output
+    } else {
+        //LOGI("Running empty\n");
+        /* fall through to OP_INVOKE_DIRECT */
+        GOTO_invoke(invokeDirect, false);
+    }
+#endif
+OP_END
+
+/* File: c/OP_UNUSED_F1.c */
+HANDLE_OPCODE(OP_UNUSED_F1)
+OP_END
+
+/* File: c/OP_IGET_QUICK.c */
+HANDLE_IGET_X_QUICK(OP_IGET_QUICK,          "", Int, )
+OP_END
+
+/* File: c/OP_IGET_WIDE_QUICK.c */
+HANDLE_IGET_X_QUICK(OP_IGET_WIDE_QUICK,     "-wide", Long, _WIDE)
+OP_END
+
+/* File: c/OP_IGET_OBJECT_QUICK.c */
+HANDLE_IGET_X_QUICK(OP_IGET_OBJECT_QUICK,   "-object", Object, _AS_OBJECT)
+OP_END
+
+/* File: c/OP_IPUT_QUICK.c */
+HANDLE_IPUT_X_QUICK(OP_IPUT_QUICK,          "", Int, )
+OP_END
+
+/* File: c/OP_IPUT_WIDE_QUICK.c */
+HANDLE_IPUT_X_QUICK(OP_IPUT_WIDE_QUICK,     "-wide", Long, _WIDE)
+OP_END
+
+/* File: c/OP_IPUT_OBJECT_QUICK.c */
+HANDLE_IPUT_X_QUICK(OP_IPUT_OBJECT_QUICK,   "-object", Object, _AS_OBJECT)
+OP_END
+
+/* File: c/OP_INVOKE_VIRTUAL_QUICK.c */
+HANDLE_OPCODE(OP_INVOKE_VIRTUAL_QUICK /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeVirtualQuick, false);
+OP_END
+
+/* File: c/OP_INVOKE_VIRTUAL_QUICK_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_VIRTUAL_QUICK_RANGE/*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeVirtualQuick, true);
+OP_END
+
+/* File: c/OP_INVOKE_SUPER_QUICK.c */
+HANDLE_OPCODE(OP_INVOKE_SUPER_QUICK /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO_invoke(invokeSuperQuick, false);
+OP_END
+
+/* File: c/OP_INVOKE_SUPER_QUICK_RANGE.c */
+HANDLE_OPCODE(OP_INVOKE_SUPER_QUICK_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO_invoke(invokeSuperQuick, true);
+OP_END
+
+/* File: c/OP_UNUSED_FC.c */
+HANDLE_OPCODE(OP_UNUSED_FC)
+OP_END
+
+/* File: c/OP_UNUSED_FD.c */
+HANDLE_OPCODE(OP_UNUSED_FD)
+OP_END
+
+/* File: c/OP_UNUSED_FE.c */
+HANDLE_OPCODE(OP_UNUSED_FE)
+OP_END
+
+/* File: c/OP_UNUSED_FF.c */
+HANDLE_OPCODE(OP_UNUSED_FF)
+    /*
+     * In portable interp, most unused opcodes will fall through to here.
+     */
+    LOGE("unknown opcode 0x%02x\n", INST_INST(inst));
+    dvmAbort();
+    FINISH(1);
+OP_END
+
+/* File: c/gotoTargets.c */
+/*
+ * C footer.  This has some common code shared by the various targets.
+ */
+
+/*
+ * Everything from here on is a "goto target".  In the basic interpreter
+ * we jump into these targets and then jump directly to the handler for
+ * next instruction.  Here, these are subroutines that return to the caller.
+ */
+
+GOTO_TARGET(filledNewArray, bool methodCallRange)
+    {
+        ClassObject* arrayClass;
+        ArrayObject* newArray;
+        u4* contents;
+        char typeCh;
+        int i;
+        u4 arg5;
+
+        EXPORT_PC();
+
+        ref = FETCH(1);             /* class ref */
+        vdst = FETCH(2);            /* first 4 regs -or- range base */
+
+        if (methodCallRange) {
+            vsrc1 = INST_AA(inst);  /* #of elements */
+            arg5 = -1;              /* silence compiler warning */
+            ILOGV("|filled-new-array-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+        } else {
+            arg5 = INST_A(inst);
+            vsrc1 = INST_B(inst);   /* #of elements */
+            ILOGV("|filled-new-array args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1, ref, vdst, arg5);
+        }
+
+        /*
+         * Resolve the array class.
+         */
+        arrayClass = dvmDexGetResolvedClass(methodClassDex, ref);
+        if (arrayClass == NULL) {
+            arrayClass = dvmResolveClass(curMethod->clazz, ref, false);
+            if (arrayClass == NULL)
+                GOTO_exceptionThrown();
+        }
+        /*
+        if (!dvmIsArrayClass(arrayClass)) {
+            dvmThrowException("Ljava/lang/RuntimeError;",
+                "filled-new-array needs array class");
+            GOTO_exceptionThrown();
+        }
+        */
+        /* verifier guarantees this is an array class */
+        assert(dvmIsArrayClass(arrayClass));
+        assert(dvmIsClassInitialized(arrayClass));
+
+        /*
+         * Create an array of the specified type.
+         */
+        LOGVV("+++ filled-new-array type is '%s'\n", arrayClass->descriptor);
+        typeCh = arrayClass->descriptor[1];
+        if (typeCh == 'D' || typeCh == 'J') {
+            /* category 2 primitives not allowed */
+            dvmThrowException("Ljava/lang/RuntimeError;",
+                "bad filled array req");
+            GOTO_exceptionThrown();
+        } else if (typeCh != 'L' && typeCh != '[' && typeCh != 'I') {
+            /* TODO: requires multiple "fill in" loops with different widths */
+            LOGE("non-int primitives not implemented\n");
+            dvmThrowException("Ljava/lang/InternalError;",
+                "filled-new-array not implemented for anything but 'int'");
+            GOTO_exceptionThrown();
+        }
+
+        newArray = dvmAllocArrayByClass(arrayClass, vsrc1, ALLOC_DONT_TRACK);
+        if (newArray == NULL)
+            GOTO_exceptionThrown();
+
+        /*
+         * Fill in the elements.  It's legal for vsrc1 to be zero.
+         */
+        contents = (u4*) newArray->contents;
+        if (methodCallRange) {
+            for (i = 0; i < vsrc1; i++)
+                contents[i] = GET_REGISTER(vdst+i);
+        } else {
+            assert(vsrc1 <= 5);
+            if (vsrc1 == 5) {
+                contents[4] = GET_REGISTER(arg5);
+                vsrc1--;
+            }
+            for (i = 0; i < vsrc1; i++) {
+                contents[i] = GET_REGISTER(vdst & 0x0f);
+                vdst >>= 4;
+            }
+        }
+
+        retval.l = newArray;
+    }
+    FINISH(3);
+GOTO_TARGET_END
+
+
+GOTO_TARGET(invokeVirtual, bool methodCallRange)
+    {
+        Method* baseMethod;
+        Object* thisPtr;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* method ref */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        /*
+         * The object against which we are executing a method is always
+         * in the first argument.
+         */
+        if (methodCallRange) {
+            assert(vsrc1 > 0);
+            ILOGV("|invoke-virtual-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisPtr = (Object*) GET_REGISTER(vdst);
+        } else {
+            assert((vsrc1>>4) > 0);
+            ILOGV("|invoke-virtual args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisPtr = (Object*) GET_REGISTER(vdst & 0x0f);
+        }
+
+        if (!checkForNull(thisPtr))
+            GOTO_exceptionThrown();
+
+        /*
+         * Resolve the method.  This is the correct method for the static
+         * type of the object.  We also verify access permissions here.
+         */
+        baseMethod = dvmDexGetResolvedMethod(methodClassDex, ref);
+        if (baseMethod == NULL) {
+            baseMethod = dvmResolveMethod(curMethod->clazz, ref,METHOD_VIRTUAL);
+            if (baseMethod == NULL) {
+                ILOGV("+ unknown method or access denied\n");
+                GOTO_exceptionThrown();
+            }
+        }
+
+        /*
+         * Combine the object we found with the vtable offset in the
+         * method.
+         */
+        assert(baseMethod->methodIndex < thisPtr->clazz->vtableCount);
+        methodToCall = thisPtr->clazz->vtable[baseMethod->methodIndex];
+
+#if 0
+        if (dvmIsAbstractMethod(methodToCall)) {
+            /*
+             * This can happen if you create two classes, Base and Sub, where
+             * Sub is a sub-class of Base.  Declare a protected abstract
+             * method foo() in Base, and invoke foo() from a method in Base.
+             * Base is an "abstract base class" and is never instantiated
+             * directly.  Now, Override foo() in Sub, and use Sub.  This
+             * Works fine unless Sub stops providing an implementation of
+             * the method.
+             */
+            dvmThrowException("Ljava/lang/AbstractMethodError;",
+                "abstract method not implemented");
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(!dvmIsAbstractMethod(methodToCall) ||
+            methodToCall->nativeFunc != NULL);
+#endif
+
+        LOGVV("+++ base=%s.%s virtual[%d]=%s.%s\n",
+            baseMethod->clazz->descriptor, baseMethod->name,
+            (u4) baseMethod->methodIndex,
+            methodToCall->clazz->descriptor, methodToCall->name);
+        assert(methodToCall != NULL);
+
+#if 0
+        if (vsrc1 != methodToCall->insSize) {
+            LOGW("WRONG METHOD: base=%s.%s virtual[%d]=%s.%s\n",
+                baseMethod->clazz->descriptor, baseMethod->name,
+                (u4) baseMethod->methodIndex,
+                methodToCall->clazz->descriptor, methodToCall->name);
+            //dvmDumpClass(baseMethod->clazz);
+            //dvmDumpClass(methodToCall->clazz);
+            dvmDumpAllClasses(0);
+        }
+#endif
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeSuper, bool methodCallRange)
+    {
+        Method* baseMethod;
+        u2 thisReg;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* method ref */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        if (methodCallRange) {
+            ILOGV("|invoke-super-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisReg = vdst;
+        } else {
+            ILOGV("|invoke-super args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisReg = vdst & 0x0f;
+        }
+        /* impossible in well-formed code, but we must check nevertheless */
+        if (!checkForNull((Object*) GET_REGISTER(thisReg)))
+            GOTO_exceptionThrown();
+
+        /*
+         * Resolve the method.  This is the correct method for the static
+         * type of the object.  We also verify access permissions here.
+         * The first arg to dvmResolveMethod() is just the referring class
+         * (used for class loaders and such), so we don't want to pass
+         * the superclass into the resolution call.
+         */
+        baseMethod = dvmDexGetResolvedMethod(methodClassDex, ref);
+        if (baseMethod == NULL) {
+            baseMethod = dvmResolveMethod(curMethod->clazz, ref,METHOD_VIRTUAL);
+            if (baseMethod == NULL) {
+                ILOGV("+ unknown method or access denied\n");
+                GOTO_exceptionThrown();
+            }
+        }
+
+        /*
+         * Combine the object we found with the vtable offset in the
+         * method's class.
+         *
+         * We're using the current method's class' superclass, not the
+         * superclass of "this".  This is because we might be executing
+         * in a method inherited from a superclass, and we want to run
+         * in that class' superclass.
+         */
+        if (baseMethod->methodIndex >= curMethod->clazz->super->vtableCount) {
+            /*
+             * Method does not exist in the superclass.  Could happen if
+             * superclass gets updated.
+             */
+            dvmThrowException("Ljava/lang/NoSuchMethodError;",
+                baseMethod->name);
+            GOTO_exceptionThrown();
+        }
+        methodToCall = curMethod->clazz->super->vtable[baseMethod->methodIndex];
+#if 0
+        if (dvmIsAbstractMethod(methodToCall)) {
+            dvmThrowException("Ljava/lang/AbstractMethodError;",
+                "abstract method not implemented");
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(!dvmIsAbstractMethod(methodToCall) ||
+            methodToCall->nativeFunc != NULL);
+#endif
+        LOGVV("+++ base=%s.%s super-virtual=%s.%s\n",
+            baseMethod->clazz->descriptor, baseMethod->name,
+            methodToCall->clazz->descriptor, methodToCall->name);
+        assert(methodToCall != NULL);
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeInterface, bool methodCallRange)
+    {
+        Object* thisPtr;
+        ClassObject* thisClass;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* method ref */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        /*
+         * The object against which we are executing a method is always
+         * in the first argument.
+         */
+        if (methodCallRange) {
+            assert(vsrc1 > 0);
+            ILOGV("|invoke-interface-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisPtr = (Object*) GET_REGISTER(vdst);
+        } else {
+            assert((vsrc1>>4) > 0);
+            ILOGV("|invoke-interface args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisPtr = (Object*) GET_REGISTER(vdst & 0x0f);
+        }
+        if (!checkForNull(thisPtr))
+            GOTO_exceptionThrown();
+
+        thisClass = thisPtr->clazz;
+
+        /*
+         * Given a class and a method index, find the Method* with the
+         * actual code we want to execute.
+         */
+        methodToCall = dvmFindInterfaceMethodInCache(thisClass, ref, curMethod,
+                        methodClassDex);
+        if (methodToCall == NULL) {
+            assert(dvmCheckException(self));
+            GOTO_exceptionThrown();
+        }
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeDirect, bool methodCallRange)
+    {
+        u2 thisReg;
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* method ref */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        EXPORT_PC();
+
+        if (methodCallRange) {
+            ILOGV("|invoke-direct-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisReg = vdst;
+        } else {
+            ILOGV("|invoke-direct args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisReg = vdst & 0x0f;
+        }
+        if (!checkForNull((Object*) GET_REGISTER(thisReg)))
+            GOTO_exceptionThrown();
+
+        methodToCall = dvmDexGetResolvedMethod(methodClassDex, ref);
+        if (methodToCall == NULL) {
+            methodToCall = dvmResolveMethod(curMethod->clazz, ref,
+                            METHOD_DIRECT);
+            if (methodToCall == NULL) {
+                ILOGV("+ unknown direct method\n");     // should be impossible
+                GOTO_exceptionThrown();
+            }
+        }
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeStatic, bool methodCallRange)
+    vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+    ref = FETCH(1);             /* method ref */
+    vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+    EXPORT_PC();
+
+    if (methodCallRange)
+        ILOGV("|invoke-static-range args=%d @0x%04x {regs=v%d-v%d}",
+            vsrc1, ref, vdst, vdst+vsrc1-1);
+    else
+        ILOGV("|invoke-static args=%d @0x%04x {regs=0x%04x %x}",
+            vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+
+    methodToCall = dvmDexGetResolvedMethod(methodClassDex, ref);
+    if (methodToCall == NULL) {
+        methodToCall = dvmResolveMethod(curMethod->clazz, ref, METHOD_STATIC);
+        if (methodToCall == NULL) {
+            ILOGV("+ unknown method\n");
+            GOTO_exceptionThrown();
+        }
+    }
+    GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeVirtualQuick, bool methodCallRange)
+    {
+        Object* thisPtr;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* vtable index */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        /*
+         * The object against which we are executing a method is always
+         * in the first argument.
+         */
+        if (methodCallRange) {
+            assert(vsrc1 > 0);
+            ILOGV("|invoke-virtual-quick-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisPtr = (Object*) GET_REGISTER(vdst);
+        } else {
+            assert((vsrc1>>4) > 0);
+            ILOGV("|invoke-virtual-quick args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisPtr = (Object*) GET_REGISTER(vdst & 0x0f);
+        }
+
+        if (!checkForNull(thisPtr))
+            GOTO_exceptionThrown();
+
+        /*
+         * Combine the object we found with the vtable offset in the
+         * method.
+         */
+        assert(ref < thisPtr->clazz->vtableCount);
+        methodToCall = thisPtr->clazz->vtable[ref];
+
+#if 0
+        if (dvmIsAbstractMethod(methodToCall)) {
+            dvmThrowException("Ljava/lang/AbstractMethodError;",
+                "abstract method not implemented");
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(!dvmIsAbstractMethod(methodToCall) ||
+            methodToCall->nativeFunc != NULL);
+#endif
+
+        LOGVV("+++ virtual[%d]=%s.%s\n",
+            ref, methodToCall->clazz->descriptor, methodToCall->name);
+        assert(methodToCall != NULL);
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeSuperQuick, bool methodCallRange)
+    {
+        u2 thisReg;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* vtable index */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        if (methodCallRange) {
+            ILOGV("|invoke-super-quick-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisReg = vdst;
+        } else {
+            ILOGV("|invoke-super-quick args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisReg = vdst & 0x0f;
+        }
+        /* impossible in well-formed code, but we must check nevertheless */
+        if (!checkForNull((Object*) GET_REGISTER(thisReg)))
+            GOTO_exceptionThrown();
+
+#if 0   /* impossible in optimized + verified code */
+        if (ref >= curMethod->clazz->super->vtableCount) {
+            dvmThrowException("Ljava/lang/NoSuchMethodError;", NULL);
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(ref < curMethod->clazz->super->vtableCount);
+#endif
+
+        /*
+         * Combine the object we found with the vtable offset in the
+         * method's class.
+         *
+         * We're using the current method's class' superclass, not the
+         * superclass of "this".  This is because we might be executing
+         * in a method inherited from a superclass, and we want to run
+         * in the method's class' superclass.
+         */
+        methodToCall = curMethod->clazz->super->vtable[ref];
+
+#if 0
+        if (dvmIsAbstractMethod(methodToCall)) {
+            dvmThrowException("Ljava/lang/AbstractMethodError;",
+                "abstract method not implemented");
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(!dvmIsAbstractMethod(methodToCall) ||
+            methodToCall->nativeFunc != NULL);
+#endif
+        LOGVV("+++ super-virtual[%d]=%s.%s\n",
+            ref, methodToCall->clazz->descriptor, methodToCall->name);
+        assert(methodToCall != NULL);
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+
+
+    /*
+     * General handling for return-void, return, and return-wide.  Put the
+     * return value in "retval" before jumping here.
+     */
+GOTO_TARGET(returnFromMethod)
+    {
+        StackSaveArea* saveArea;
+
+        /*
+         * We must do this BEFORE we pop the previous stack frame off, so
+         * that the GC can see the return value (if any) in the local vars.
+         *
+         * Since this is now an interpreter switch point, we must do it before
+         * we do anything at all.
+         */
+        PERIODIC_CHECKS(kInterpEntryReturn, 0);
+
+        ILOGV("> retval=0x%llx (leaving %s.%s %s)",
+            retval.j, curMethod->clazz->descriptor, curMethod->name,
+            curMethod->signature);
+        //DUMP_REGS(curMethod, fp);
+
+        saveArea = SAVEAREA_FROM_FP(fp);
+
+#ifdef EASY_GDB
+        debugSaveArea = saveArea;
+#endif
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_PROFILER)
+        TRACE_METHOD_EXIT(self, curMethod);
+#endif
+
+        /* back up to previous frame and see if we hit a break */
+        fp = saveArea->prevFrame;
+        assert(fp != NULL);
+        if (dvmIsBreakFrame(fp)) {
+            /* bail without popping the method frame from stack */
+            LOGVV("+++ returned into break frame\n");
+            GOTO_bail();
+        }
+
+        /* update thread FP, and reset local variables */
+        self->curFrame = fp;
+        curMethod = SAVEAREA_FROM_FP(fp)->method;
+        //methodClass = curMethod->clazz;
+        methodClassDex = curMethod->clazz->pDvmDex;
+        pc = saveArea->savedPc;
+        ILOGD("> (return to %s.%s %s)", curMethod->clazz->descriptor,
+            curMethod->name, curMethod->signature);
+
+        /* use FINISH on the caller's invoke instruction */
+        //u2 invokeInstr = INST_INST(FETCH(0));
+        if (true /*invokeInstr >= OP_INVOKE_VIRTUAL &&
+            invokeInstr <= OP_INVOKE_INTERFACE*/)
+        {
+            FINISH(3);
+        } else {
+            //LOGE("Unknown invoke instr %02x at %d\n",
+            //    invokeInstr, (int) (pc - curMethod->insns));
+            assert(false);
+        }
+    }
+GOTO_TARGET_END
+
+
+    /*
+     * Jump here when the code throws an exception.
+     *
+     * By the time we get here, the Throwable has been created and the stack
+     * trace has been saved off.
+     */
+GOTO_TARGET(exceptionThrown)
+    {
+        Object* exception;
+        int catchRelPc;
+
+        /*
+         * Since this is now an interpreter switch point, we must do it before
+         * we do anything at all.
+         */
+        PERIODIC_CHECKS(kInterpEntryThrow, 0);
+
+        /*
+         * We save off the exception and clear the exception status.  While
+         * processing the exception we might need to load some Throwable
+         * classes, and we don't want class loader exceptions to get
+         * confused with this one.
+         */
+        assert(dvmCheckException(self));
+        exception = dvmGetException(self);
+        dvmAddTrackedAlloc(exception, self);
+        dvmClearException(self);
+
+        LOGV("Handling exception %s at %s:%d\n",
+            exception->clazz->descriptor, curMethod->name,
+            dvmLineNumFromPC(curMethod, pc - curMethod->insns));
+
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_DEBUGGER)
+        /*
+         * Tell the debugger about it.
+         *
+         * TODO: if the exception was thrown by interpreted code, control
+         * fell through native, and then back to us, we will report the
+         * exception at the point of the throw and again here.  We can avoid
+         * this by not reporting exceptions when we jump here directly from
+         * the native call code above, but then we won't report exceptions
+         * that were thrown *from* the JNI code (as opposed to *through* it).
+         *
+         * The correct solution is probably to ignore from-native exceptions
+         * here, and have the JNI exception code do the reporting to the
+         * debugger.
+         */
+        if (gDvm.debuggerActive) {
+            void* catchFrame;
+            catchRelPc = dvmFindCatchBlock(self, pc - curMethod->insns,
+                        exception, true, &catchFrame);
+            dvmDbgPostException(fp, pc - curMethod->insns, catchFrame,
+                catchRelPc, exception);
+        }
+#endif
+
+        /*
+         * We need to unroll to the catch block or the nearest "break"
+         * frame.
+         *
+         * A break frame could indicate that we have reached an intermediate
+         * native call, or have gone off the top of the stack and the thread
+         * needs to exit.  Either way, we return from here, leaving the
+         * exception raised.
+         *
+         * If we do find a catch block, we want to transfer execution to
+         * that point.
+         */
+        catchRelPc = dvmFindCatchBlock(self, pc - curMethod->insns,
+                    exception, false, (void*)&fp);
+
+        /*
+         * Restore the stack bounds after an overflow.  This isn't going to
+         * be correct in all circumstances, e.g. if JNI code devours the
+         * exception this won't happen until some other exception gets
+         * thrown.  If the code keeps pushing the stack bounds we'll end
+         * up aborting the VM.
+         *
+         * Note we want to do this *after* the call to dvmFindCatchBlock,
+         * because that may need extra stack space to resolve exception
+         * classes (e.g. through a class loader).
+         */
+        if (self->stackOverflowed)
+            dvmCleanupStackOverflow(self);
+
+        if (catchRelPc < 0) {
+            /* falling through to JNI code or off the bottom of the stack */
+#if DVM_SHOW_EXCEPTION >= 2
+            LOGD("Exception %s from %s:%d not caught locally\n",
+                exception->clazz->descriptor, dvmGetMethodSourceFile(curMethod),
+                dvmLineNumFromPC(curMethod, pc - curMethod->insns));
+#endif
+            dvmSetException(self, exception);
+            dvmReleaseTrackedAlloc(exception, self);
+            GOTO_bail();
+        }
+
+#if DVM_SHOW_EXCEPTION >= 3
+        {
+            const Method* catchMethod = SAVEAREA_FROM_FP(fp)->method;
+            LOGD("Exception %s thrown from %s:%d to %s:%d\n",
+                exception->clazz->descriptor, dvmGetMethodSourceFile(curMethod),
+                dvmLineNumFromPC(curMethod, pc - curMethod->insns),
+                dvmGetMethodSourceFile(catchMethod),
+                dvmLineNumFromPC(catchMethod, catchRelPc));
+        }
+#endif
+
+        /*
+         * Adjust local variables to match self->curFrame and the
+         * updated PC.
+         */
+        //fp = (u4*) self->curFrame;
+        curMethod = SAVEAREA_FROM_FP(fp)->method;
+        //methodClass = curMethod->clazz;
+        methodClassDex = curMethod->clazz->pDvmDex;
+        pc = curMethod->insns + catchRelPc;
+        ILOGV("> pc <-- %s.%s %s", curMethod->clazz->descriptor,
+            curMethod->name, curMethod->signature);
+        DUMP_REGS(curMethod, fp, false);            // show all regs
+
+        /*
+         * Restore the exception if the handler wants it.
+         *
+         * The Dalvik spec mandates that, if an exception handler wants to
+         * do something with the exception, the first instruction executed
+         * must be "move-exception".  We can pass the exception along
+         * through the thread struct, and let the move-exception instruction
+         * clear it for us.
+         *
+         * If the handler doesn't call move-exception, we don't want to
+         * finish here with an exception still pending.
+         */
+        if (INST_INST(FETCH(0)) == OP_MOVE_EXCEPTION)
+            dvmSetException(self, exception);
+
+        dvmReleaseTrackedAlloc(exception, self);
+        FINISH(0);
+    }
+GOTO_TARGET_END
+
+
+    /*
+     * General handling for invoke-{virtual,super,direct,static,interface},
+     * including "quick" variants.
+     *
+     * Set "methodToCall" to the Method we're calling, and "methodCallRange"
+     * depending on whether this is a "/range" instruction.
+     *
+     * For a range call:
+     *  "vsrc1" holds the argument count (8 bits)
+     *  "vdst" holds the first argument in the range
+     * For a non-range call:
+     *  "vsrc1" holds the argument count (4 bits) and the 5th argument index
+     *  "vdst" holds four 4-bit register indices
+     *
+     * The caller must EXPORT_PC before jumping here, because any method
+     * call can throw a stack overflow exception.
+     */
+GOTO_TARGET(invokeMethod, bool methodCallRange, const Method* _methodToCall,
+    u2 count, u2 regs)
+    {
+        STUB_HACK(vsrc1 = count; vdst = regs; methodToCall = _methodToCall;);
+
+        //printf("range=%d call=%p count=%d regs=0x%04x\n",
+        //    methodCallRange, methodToCall, count, regs);
+        //printf(" --> %s.%s %s\n", methodToCall->clazz->descriptor,
+        //    methodToCall->name, methodToCall->signature);
+
+        u4* outs;
+        int i;
+
+        /*
+         * Copy args.  This may corrupt vsrc1/vdst.
+         */
+        if (methodCallRange) {
+            // could use memcpy or a "Duff's device"; most functions have
+            // so few args it won't matter much
+            assert(vsrc1 <= curMethod->outsSize);
+            assert(vsrc1 == methodToCall->insSize);
+            outs = OUTS_FROM_FP(fp, vsrc1);
+            for (i = 0; i < vsrc1; i++)
+                outs[i] = GET_REGISTER(vdst+i);
+        } else {
+            u4 count = vsrc1 >> 4;
+
+            assert(count <= curMethod->outsSize);
+            assert(count == methodToCall->insSize);
+            assert(count <= 5);
+
+            outs = OUTS_FROM_FP(fp, count);
+#if 0
+            if (count == 5) {
+                outs[4] = GET_REGISTER(vsrc1 & 0x0f);
+                count--;
+            }
+            for (i = 0; i < (int) count; i++) {
+                outs[i] = GET_REGISTER(vdst & 0x0f);
+                vdst >>= 4;
+            }
+#else
+            // This version executes fewer instructions but is larger
+            // overall.  Seems to be a teensy bit faster.
+            assert((vdst >> 16) == 0);  // 16 bits -or- high 16 bits clear
+            switch (count) {
+            case 5:
+                outs[4] = GET_REGISTER(vsrc1 & 0x0f);
+            case 4:
+                outs[3] = GET_REGISTER(vdst >> 12);
+            case 3:
+                outs[2] = GET_REGISTER((vdst & 0x0f00) >> 8);
+            case 2:
+                outs[1] = GET_REGISTER((vdst & 0x00f0) >> 4);
+            case 1:
+                outs[0] = GET_REGISTER(vdst & 0x0f);
+            default:
+                ;
+            }
+#endif
+        }
+    }
+
+    /*
+     * (This was originally a "goto" target; I've kept it separate from the
+     * stuff above in case we want to refactor things again.)
+     *
+     * At this point, we have the arguments stored in the "outs" area of
+     * the current method's stack frame, and the method to call in
+     * "methodToCall".  Push a new stack frame.
+     */
+    {
+        StackSaveArea* newSaveArea;
+        u4* newFp;
+
+        ILOGV("> %s%s.%s %s",
+            dvmIsNativeMethod(methodToCall) ? "(NATIVE) " : "",
+            methodToCall->clazz->descriptor, methodToCall->name,
+            methodToCall->signature);
+
+        newFp = (u4*) SAVEAREA_FROM_FP(fp) - methodToCall->registersSize;
+        newSaveArea = SAVEAREA_FROM_FP(newFp);
+
+        /* verify that we have enough space */
+        if (true) {
+            u1* bottom;
+            bottom = (u1*) newSaveArea - methodToCall->outsSize * sizeof(u4);
+            if (bottom < self->interpStackEnd) {
+                /* stack overflow */
+                LOGV("Stack overflow on method call (start=%p end=%p newBot=%p size=%d '%s')\n",
+                    self->interpStackStart, self->interpStackEnd, bottom,
+                    self->interpStackSize, methodToCall->name);
+                dvmHandleStackOverflow(self);
+                assert(dvmCheckException(self));
+                GOTO_exceptionThrown();
+            }
+            //LOGD("+++ fp=%p newFp=%p newSave=%p bottom=%p\n",
+            //    fp, newFp, newSaveArea, bottom);
+        }
+
+#ifdef LOG_INSTR
+        if (methodToCall->registersSize > methodToCall->insSize) {
+            /*
+             * This makes valgrind quiet when we print registers that
+             * haven't been initialized.  Turn it off when the debug
+             * messages are disabled -- we want valgrind to report any
+             * used-before-initialized issues.
+             */
+            memset(newFp, 0xcc,
+                (methodToCall->registersSize - methodToCall->insSize) * 4);
+        }
+#endif
+
+#ifdef EASY_GDB
+        newSaveArea->prevSave = SAVEAREA_FROM_FP(fp);
+#endif
+        newSaveArea->prevFrame = fp;
+        newSaveArea->savedPc = pc;
+        newSaveArea->method = methodToCall;
+
+        if (!dvmIsNativeMethod(methodToCall)) {
+            /*
+             * "Call" interpreted code.  Reposition the PC, update the
+             * frame pointer and other local state, and continue.
+             */
+            curMethod = methodToCall;
+            methodClassDex = curMethod->clazz->pDvmDex;
+            pc = methodToCall->insns;
+            fp = self->curFrame = newFp;
+#ifdef EASY_GDB
+            debugSaveArea = SAVEAREA_FROM_FP(newFp);
+#endif
+#if INTERP_TYPE == INTERP_DBG
+            debugIsMethodEntry = true;              // profiling, debugging
+#endif
+            ILOGD("> pc <-- %s.%s %s", curMethod->clazz->descriptor,
+                curMethod->name, curMethod->signature);
+            DUMP_REGS(curMethod, fp, true);         // show input args
+            FINISH(0);                              // jump to method start
+        } else {
+            /* set this up for JNI locals, even if not a JNI native */
+            newSaveArea->xtra.localRefTop = self->jniLocalRefTable.nextEntry;
+
+            self->curFrame = newFp;
+
+            DUMP_REGS(methodToCall, newFp, true);   // show input args
+
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_DEBUGGER)
+            if (gDvm.debuggerActive) {
+                dvmDbgPostLocationEvent(methodToCall, -1,
+                    dvmGetThisPtr(curMethod, fp), DBG_METHOD_ENTRY);
+            }
+#endif
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_PROFILER)
+            TRACE_METHOD_ENTER(self, methodToCall);
+#endif
+
+            ILOGD("> native <-- %s.%s %s", methodToCall->clazz->descriptor,
+                methodToCall->name, methodToCall->signature);
+
+            /*
+             * Jump through native call bridge.  Because we leave no
+             * space for locals on native calls, "newFp" points directly
+             * to the method arguments.
+             */
+            (*methodToCall->nativeFunc)(newFp, &retval, methodToCall, self);
+
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_DEBUGGER)
+            if (gDvm.debuggerActive) {
+                dvmDbgPostLocationEvent(methodToCall, -1,
+                    dvmGetThisPtr(curMethod, fp), DBG_METHOD_EXIT);
+            }
+#endif
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_PROFILER)
+            TRACE_METHOD_EXIT(self, methodToCall);
+#endif
+
+            /* pop frame off */
+            dvmPopJniLocals(self, newSaveArea);
+            self->curFrame = fp;
+
+            /*
+             * If the native code threw an exception, or interpreted code
+             * invoked by the native call threw one and nobody has cleared
+             * it, jump to our local exception handling.
+             */
+            if (dvmCheckException(self)) {
+                LOGV("Exception thrown by/below native code\n");
+                GOTO_exceptionThrown();
+            }
+
+            ILOGD("> retval=0x%llx (leaving native)", retval.j);
+            ILOGD("> (return from native %s.%s to %s.%s %s)",
+                methodToCall->clazz->descriptor, methodToCall->name,
+                curMethod->clazz->descriptor, curMethod->name,
+                curMethod->signature);
+
+            //u2 invokeInstr = INST_INST(FETCH(0));
+            if (true /*invokeInstr >= OP_INVOKE_VIRTUAL &&
+                invokeInstr <= OP_INVOKE_INTERFACE*/)
+            {
+                FINISH(3);
+            } else {
+                //LOGE("Unknown invoke instr %02x at %d\n",
+                //    invokeInstr, (int) (pc - curMethod->insns));
+                assert(false);
+            }
+        }
+    }
+    assert(false);      // should not get here
+GOTO_TARGET_END
+
+
+/* File: portable/enddefs.c */
+/*--- end of opcodes ---*/
+
+#ifndef THREADED_INTERP
+        } // end of "switch"
+    } // end of "while"
+#endif
+
+bail:
+    ILOGD("|-- Leaving interpreter loop");      // note "curMethod" may be NULL
+
+    interpState->retval = retval;
+    return false;
+
+bail_switch:
+    /*
+     * The standard interpreter currently doesn't set or care about the
+     * "debugIsMethodEntry" value, so setting this is only of use if we're
+     * switching between two "debug" interpreters, which we never do.
+     *
+     * TODO: figure out if preserving this makes any sense.
+     */
+#if defined(WITH_PROFILER) || defined(WITH_DEBUGGER)
+# if INTERP_TYPE == INTERP_DBG
+    interpState->debugIsMethodEntry = debugIsMethodEntry;
+# else
+    interpState->debugIsMethodEntry = false;
+# endif
+#endif
+
+    /* export state changes */
+    interpState->method = curMethod;
+    interpState->pc = pc;
+    interpState->fp = fp;
+    /* debugTrackedRefStart doesn't change */
+    interpState->retval = retval;   /* need for _entryPoint=ret */
+    interpState->nextMode =
+        (INTERP_TYPE == INTERP_STD) ? INTERP_DBG : INTERP_STD;
+    LOGVV(" meth='%s.%s' pc=0x%x fp=%p\n",
+        curMethod->clazz->descriptor, curMethod->name,
+        pc - curMethod->insns, fp);
+    return true;
+}
+
+
diff --git a/vm/mterp/out/InterpC-x86.c b/vm/mterp/out/InterpC-x86.c
new file mode 100644
index 0000000..cd5fe95
--- /dev/null
+++ b/vm/mterp/out/InterpC-x86.c
@@ -0,0 +1,2119 @@
+/*
+ * This file was generated automatically by gen-mterp.py for 'x86'.
+ *
+ * --> DO NOT EDIT <--
+ */
+
+/* File: c/header.c */
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/* common includes */
+#include "Dalvik.h"
+#include "interp/InterpDefs.h"
+#include "mterp/Mterp.h"
+#include <math.h>                   // needed for fmod, fmodf
+
+/*
+ * Configuration defines.  These affect the C implementations, i.e. the
+ * portable interpreter(s) and C stubs.
+ *
+ * Some defines are controlled by the Makefile, e.g.:
+ *   WITH_PROFILER
+ *   WITH_DEBUGGER
+ *   WITH_INSTR_CHECKS
+ *   WITH_TRACKREF_CHECKS
+ *   EASY_GDB
+ *   NDEBUG
+ *
+ * If THREADED_INTERP is not defined, we use a classic "while true / switch"
+ * interpreter.  If it is defined, then the tail end of each instruction
+ * handler fetches the next instruction and jumps directly to the handler.
+ * This increases the size of the "Std" interpreter by about 10%, but
+ * provides a speedup of about the same magnitude.
+ *
+ * There's a "hybrid" approach that uses a goto table instead of a switch
+ * statement, avoiding the "is the opcode in range" tests required for switch.
+ * The performance is close to the threaded version, and without the 10%
+ * size increase, but the benchmark results are off enough that it's not
+ * worth adding as a third option.
+ */
+#define THREADED_INTERP             /* threaded vs. while-loop interpreter */
+
+#ifdef WITH_INSTR_CHECKS            /* instruction-level paranoia */
+# define CHECK_BRANCH_OFFSETS
+# define CHECK_REGISTER_INDICES
+#endif
+
+/*
+ * ARM EABI requires 64-bit alignment for access to 64-bit data types.  We
+ * can't just use pointers to copy 64-bit values out of our interpreted
+ * register set, because gcc will generate ldrd/strd.
+ *
+ * The __UNION version copies data in and out of a union.  The __MEMCPY
+ * version uses a memcpy() call to do the transfer; gcc is smart enough to
+ * not actually call memcpy().  The __UNION version is very bad on ARM;
+ * it only uses one more instruction than __MEMCPY, but for some reason
+ * gcc thinks it needs separate storage for every instance of the union.
+ * On top of that, it feels the need to zero them out at the start of the
+ * method.  Net result is we zero out ~700 bytes of stack space at the top
+ * of the interpreter using ARM STM instructions.
+ */
+#if defined(__ARM_EABI__)
+//# define NO_UNALIGN_64__UNION
+# define NO_UNALIGN_64__MEMCPY
+#endif
+
+//#define LOG_INSTR                   /* verbose debugging */
+/* set and adjust ANDROID_LOG_TAGS='*:i jdwp:i dalvikvm:i dalvikvmi:i' */
+
+/*
+ * Keep a tally of accesses to fields.  Currently only works if full DEX
+ * optimization is disabled.
+ */
+#ifdef PROFILE_FIELD_ACCESS
+# define UPDATE_FIELD_GET(_field) { (_field)->gets++; }
+# define UPDATE_FIELD_PUT(_field) { (_field)->puts++; }
+#else
+# define UPDATE_FIELD_GET(_field) ((void)0)
+# define UPDATE_FIELD_PUT(_field) ((void)0)
+#endif
+
+/*
+ * Adjust the program counter.  "_offset" is a signed int, in 16-bit units.
+ *
+ * Assumes the existence of "const u2* pc" and "const u2* curMethod->insns".
+ *
+ * We don't advance the program counter until we finish an instruction or
+ * branch, because we do want to have to unroll the PC if there's an
+ * exception.
+ */
+#ifdef CHECK_BRANCH_OFFSETS
+# define ADJUST_PC(_offset) do {                                            \
+        int myoff = _offset;        /* deref only once */                   \
+        if (pc + myoff < curMethod->insns ||                                \
+            pc + myoff >= curMethod->insns + dvmGetMethodInsnsSize(curMethod)) \
+        {                                                                   \
+            char* desc;                                                     \
+            desc = dexProtoCopyMethodDescriptor(&curMethod->prototype);     \
+            LOGE("Invalid branch %d at 0x%04x in %s.%s %s\n",               \
+                myoff, (int) (pc - curMethod->insns),                       \
+                curMethod->clazz->descriptor, curMethod->name, desc);       \
+            free(desc);                                                     \
+            dvmAbort();                                                     \
+        }                                                                   \
+        pc += myoff;                                                        \
+    } while (false)
+#else
+# define ADJUST_PC(_offset) (pc += _offset)
+#endif
+
+/*
+ * If enabled, log instructions as we execute them.
+ */
+#ifdef LOG_INSTR
+# define ILOGD(...) ILOG(LOG_DEBUG, __VA_ARGS__)
+# define ILOGV(...) ILOG(LOG_VERBOSE, __VA_ARGS__)
+# define ILOG(_level, ...) do {                                             \
+        char debugStrBuf[128];                                              \
+        snprintf(debugStrBuf, sizeof(debugStrBuf), __VA_ARGS__);            \
+        if (curMethod != NULL)                                                 \
+            LOG(_level, LOG_TAG"i", "%-2d|%04x%s\n",                        \
+                self->threadId, (int)(pc - curMethod->insns), debugStrBuf); \
+        else                                                                \
+            LOG(_level, LOG_TAG"i", "%-2d|####%s\n",                        \
+                self->threadId, debugStrBuf);                               \
+    } while(false)
+void dvmDumpRegs(const Method* method, const u4* framePtr, bool inOnly);
+# define DUMP_REGS(_meth, _frame, _inOnly) dvmDumpRegs(_meth, _frame, _inOnly)
+static const char kSpacing[] = "            ";
+#else
+# define ILOGD(...) ((void)0)
+# define ILOGV(...) ((void)0)
+# define DUMP_REGS(_meth, _frame, _inOnly) ((void)0)
+#endif
+
+/* get a long from an array of u4 */
+static inline s8 getLongFromArray(const u4* ptr, int idx)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { s8 ll; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.parts[0] = ptr[0];
+    conv.parts[1] = ptr[1];
+    return conv.ll;
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    s8 val;
+    memcpy(&val, &ptr[idx], 8);
+    return val;
+#else
+    return *((s8*) &ptr[idx]);
+#endif
+}
+
+/* store a long into an array of u4 */
+static inline void putLongToArray(u4* ptr, int idx, s8 val)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { s8 ll; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.ll = val;
+    ptr[0] = conv.parts[0];
+    ptr[1] = conv.parts[1];
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    memcpy(&ptr[idx], &val, 8);
+#else
+    *((s8*) &ptr[idx]) = val;
+#endif
+}
+
+/* get a double from an array of u4 */
+static inline double getDoubleFromArray(const u4* ptr, int idx)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { double d; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.parts[0] = ptr[0];
+    conv.parts[1] = ptr[1];
+    return conv.d;
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    double dval;
+    memcpy(&dval, &ptr[idx], 8);
+    return dval;
+#else
+    return *((double*) &ptr[idx]);
+#endif
+}
+
+/* store a double into an array of u4 */
+static inline void putDoubleToArray(u4* ptr, int idx, double dval)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { double d; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.d = dval;
+    ptr[0] = conv.parts[0];
+    ptr[1] = conv.parts[1];
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    memcpy(&ptr[idx], &dval, 8);
+#else
+    *((double*) &ptr[idx]) = dval;
+#endif
+}
+
+/*
+ * If enabled, validate the register number on every access.  Otherwise,
+ * just do an array access.
+ *
+ * Assumes the existence of "u4* fp".
+ *
+ * "_idx" may be referenced more than once.
+ */
+#ifdef CHECK_REGISTER_INDICES
+# define GET_REGISTER(_idx) \
+    ( (_idx) < curMethod->registersSize ? \
+        (fp[(_idx)]) : (assert(!"bad reg"),1969) )
+# define SET_REGISTER(_idx, _val) \
+    ( (_idx) < curMethod->registersSize ? \
+        (fp[(_idx)] = (u4)(_val)) : (assert(!"bad reg"),1969) )
+# define GET_REGISTER_AS_OBJECT(_idx)       ((Object *)GET_REGISTER(_idx))
+# define SET_REGISTER_AS_OBJECT(_idx, _val) SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_INT(_idx) ((s4) GET_REGISTER(_idx))
+# define SET_REGISTER_INT(_idx, _val) SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_WIDE(_idx) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        getLongFromArray(fp, (_idx)) : (assert(!"bad reg"),1969) )
+# define SET_REGISTER_WIDE(_idx, _val) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        putLongToArray(fp, (_idx), (_val)) : (assert(!"bad reg"),1969) )
+# define GET_REGISTER_FLOAT(_idx) \
+    ( (_idx) < curMethod->registersSize ? \
+        (*((float*) &fp[(_idx)])) : (assert(!"bad reg"),1969.0f) )
+# define SET_REGISTER_FLOAT(_idx, _val) \
+    ( (_idx) < curMethod->registersSize ? \
+        (*((float*) &fp[(_idx)]) = (_val)) : (assert(!"bad reg"),1969.0f) )
+# define GET_REGISTER_DOUBLE(_idx) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        getDoubleFromArray(fp, (_idx)) : (assert(!"bad reg"),1969.0) )
+# define SET_REGISTER_DOUBLE(_idx, _val) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        putDoubleToArray(fp, (_idx), (_val)) : (assert(!"bad reg"),1969.0) )
+#else
+# define GET_REGISTER(_idx)                 (fp[(_idx)])
+# define SET_REGISTER(_idx, _val)           (fp[(_idx)] = (_val))
+# define GET_REGISTER_AS_OBJECT(_idx)       ((Object*) fp[(_idx)])
+# define SET_REGISTER_AS_OBJECT(_idx, _val) (fp[(_idx)] = (u4)(_val))
+# define GET_REGISTER_INT(_idx)             ((s4)GET_REGISTER(_idx))
+# define SET_REGISTER_INT(_idx, _val)       SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_WIDE(_idx)            getLongFromArray(fp, (_idx))
+# define SET_REGISTER_WIDE(_idx, _val)      putLongToArray(fp, (_idx), (_val))
+# define GET_REGISTER_FLOAT(_idx)           (*((float*) &fp[(_idx)]))
+# define SET_REGISTER_FLOAT(_idx, _val)     (*((float*) &fp[(_idx)]) = (_val))
+# define GET_REGISTER_DOUBLE(_idx)          getDoubleFromArray(fp, (_idx))
+# define SET_REGISTER_DOUBLE(_idx, _val)    putDoubleToArray(fp, (_idx), (_val))
+#endif
+
+/*
+ * Get 16 bits from the specified offset of the program counter.  We always
+ * want to load 16 bits at a time from the instruction stream -- it's more
+ * efficient than 8 and won't have the alignment problems that 32 might.
+ *
+ * Assumes existence of "const u2* pc".
+ */
+#define FETCH(_offset)     (pc[(_offset)])
+
+/*
+ * Extract instruction byte from 16-bit fetch (_inst is a u2).
+ */
+#define INST_INST(_inst)    ((_inst) & 0xff)
+
+/*
+ * Extract the "vA, vB" 4-bit registers from the instruction word (_inst is u2).
+ */
+#define INST_A(_inst)       (((_inst) >> 8) & 0x0f)
+#define INST_B(_inst)       ((_inst) >> 12)
+
+/*
+ * Get the 8-bit "vAA" 8-bit register index from the instruction word.
+ * (_inst is u2)
+ */
+#define INST_AA(_inst)      ((_inst) >> 8)
+
+/*
+ * The current PC must be available to Throwable constructors, e.g.
+ * those created by dvmThrowException(), so that the exception stack
+ * trace can be generated correctly.  If we don't do this, the offset
+ * within the current method won't be shown correctly.  See the notes
+ * in Exception.c.
+ *
+ * Assumes existence of "u4* fp" and "const u2* pc".
+ */
+#define EXPORT_PC()         (SAVEAREA_FROM_FP(fp)->xtra.currentPc = pc)
+
+/*
+ * Determine if we need to switch to a different interpreter.  "_current"
+ * is either INTERP_STD or INTERP_DBG.  It should be fixed for a given
+ * interpreter generation file, which should remove the outer conditional
+ * from the following.
+ *
+ * If we're building without debug and profiling support, we never switch.
+ */
+#if defined(WITH_PROFILER) || defined(WITH_DEBUGGER)
+# define NEED_INTERP_SWITCH(_current) (                                     \
+    (_current == INTERP_STD) ?                                              \
+        dvmDebuggerOrProfilerActive() : !dvmDebuggerOrProfilerActive() )
+#else
+# define NEED_INTERP_SWITCH(_current) (false)
+#endif
+
+/*
+ * Look up an interface on a class using the cache.
+ */
+INLINE Method* dvmFindInterfaceMethodInCache(ClassObject* thisClass,
+    u4 methodIdx, const Method* method, DvmDex* methodClassDex)
+{
+#define ATOMIC_CACHE_CALC \
+    dvmInterpFindInterfaceMethod(thisClass, methodIdx, method, methodClassDex)
+
+    return (Method*) ATOMIC_CACHE_LOOKUP(methodClassDex->pInterfaceCache,
+                DEX_INTERFACE_CACHE_SIZE, thisClass, methodIdx);
+
+#undef ATOMIC_CACHE_CALC
+}
+
+/*
+ * Check to see if "obj" is NULL.  If so, throw an exception.  Assumes the
+ * pc has already been exported to the stack.
+ *
+ * Perform additional checks on debug builds.
+ *
+ * Use this to check for NULL when the instruction handler calls into
+ * something that could throw an exception (so we have already called
+ * EXPORT_PC at the top).
+ */
+static inline bool checkForNull(Object* obj)
+{
+    if (obj == NULL) {
+        dvmThrowException("Ljava/lang/NullPointerException;", NULL);
+        return false;
+    }
+#ifdef WITH_EXTRA_OBJECT_VALIDATION
+    if (!dvmIsValidObject(obj)) {
+        LOGE("Invalid object %p\n", obj);
+        dvmAbort();
+    }
+#endif
+#ifndef NDEBUG
+    if (obj->clazz == NULL || ((u4) obj->clazz) <= 65536) {
+        /* probable heap corruption */
+        LOGE("Invalid object class %p (in %p)\n", obj->clazz, obj);
+        dvmAbort();
+    }
+#endif
+    return true;
+}
+
+/*
+ * Check to see if "obj" is NULL.  If so, export the PC into the stack
+ * frame and throw an exception.
+ *
+ * Perform additional checks on debug builds.
+ *
+ * Use this to check for NULL when the instruction handler doesn't do
+ * anything else that can throw an exception.
+ */
+static inline bool checkForNullExportPC(Object* obj, u4* fp, const u2* pc)
+{
+    if (obj == NULL) {
+        EXPORT_PC();
+        dvmThrowException("Ljava/lang/NullPointerException;", NULL);
+        return false;
+    }
+#ifdef WITH_EXTRA_OBJECT_VALIDATION
+    if (!dvmIsValidObject(obj)) {
+        LOGE("Invalid object %p\n", obj);
+        dvmAbort();
+    }
+#endif
+#ifndef NDEBUG
+    if (obj->clazz == NULL || ((u4) obj->clazz) <= 65536) {
+        /* probable heap corruption */
+        LOGE("Invalid object class %p (in %p)\n", obj->clazz, obj);
+        dvmAbort();
+    }
+#endif
+    return true;
+}
+
+
+/* File: cstubs/stubdefs.c */
+/* this is a standard (no debug support) interpreter */
+#define INTERP_TYPE INTERP_STD
+#define CHECK_DEBUG_AND_PROF() ((void)0)
+# define CHECK_TRACKED_REFS() ((void)0)
+
+/*
+ * In the C mterp stubs, "goto" is a function call followed immediately
+ * by a return.
+ */
+
+#define GOTO_TARGET_DECL(_target, ...)                                      \
+    void dvmMterp_##_target(MterpGlue* glue, ## __VA_ARGS__);
+
+#define GOTO_TARGET(_target, ...)                                           \
+    void dvmMterp_##_target(MterpGlue* glue, ## __VA_ARGS__) {              \
+        u2 ref, vsrc1, vsrc2, vdst;                                         \
+        u2 inst = FETCH(0);                                                 \
+        const Method* methodToCall;                                         \
+        StackSaveArea* debugSaveArea;
+
+#define GOTO_TARGET_END }
+
+/*
+ * Redefine what used to be local variable accesses into MterpGlue struct
+ * references.  (These are undefined down in "footer.c".)
+ */
+#define retval                  glue->retval
+#define pc                      glue->pc
+#define fp                      glue->fp
+#define curMethod               glue->method
+#define methodClassDex          glue->methodClassDex
+#define self                    glue->self
+#define debugTrackedRefStart    glue->debugTrackedRefStart
+
+/* ugh */
+#define STUB_HACK(x) x
+
+
+/*
+ * Opcode handler framing macros.  Here, each opcode is a separate function
+ * that takes a "glue" argument and returns void.  We can't declare
+ * these "static" because they may be called from an assembly stub.
+ */
+#define HANDLE_OPCODE(_op)                                                  \
+    void dvmMterp_##_op(MterpGlue* glue) {                                  \
+        u2 ref, vsrc1, vsrc2, vdst;                                         \
+        u2 inst = FETCH(0);
+
+#define OP_END }
+
+/*
+ * Like the "portable" FINISH, but don't reload "inst", and return to caller
+ * when done.
+ */
+#define FINISH(_offset) {                                                   \
+        ADJUST_PC(_offset);                                                 \
+        CHECK_DEBUG_AND_PROF();                                             \
+        CHECK_TRACKED_REFS();                                               \
+        return;                                                             \
+    }
+
+
+/*
+ * The "goto label" statements turn into function calls followed by
+ * return statements.  Some of the functions take arguments, which in the
+ * portable interpreter are handled by assigning values to globals.
+ */
+
+#define GOTO_exceptionThrown()                                              \
+    do {                                                                    \
+        dvmMterp_exceptionThrown(glue);                                     \
+        return;                                                             \
+    } while(false)
+
+#define GOTO_returnFromMethod()                                             \
+    do {                                                                    \
+        dvmMterp_returnFromMethod(glue);                                    \
+        return;                                                             \
+    } while(false)
+
+#define GOTO_invoke(_target, _methodCallRange)                              \
+    do {                                                                    \
+        dvmMterp_##_target(glue, _methodCallRange);                         \
+        return;                                                             \
+    } while(false)
+
+#define GOTO_invokeMethod(_methodCallRange, _methodToCall, _vsrc1, _vdst)   \
+    do {                                                                    \
+        dvmMterp_invokeMethod(glue, _methodCallRange, _methodToCall,        \
+            _vsrc1, _vdst);                                                 \
+        return;                                                             \
+    } while(false)
+
+/*
+ * As a special case, "goto bail" turns into a longjmp.  Use "bail_switch"
+ * if we need to switch to the other interpreter upon our return.
+ */
+#define GOTO_bail()                                                         \
+    dvmMterpStdBail(glue, false);
+#define GOTO_bail_switch()                                                  \
+    dvmMterpStdBail(glue, true);
+
+/*
+ * Periodically check for thread suspension.
+ *
+ * While we're at it, see if a debugger has attached or the profiler has
+ * started.  If so, switch to a different "goto" table.
+ */
+#define PERIODIC_CHECKS(_entryPoint, _pcadj) {                              \
+        dvmCheckSuspendQuick(self);                                         \
+        if (NEED_INTERP_SWITCH(INTERP_TYPE)) {                              \
+            ADJUST_PC(_pcadj);                                              \
+            glue->entryPoint = _entryPoint;                                 \
+            LOGVV("threadid=%d: switch to STD ep=%d adj=%d\n",              \
+                glue->self->threadId, (_entryPoint), (_pcadj));             \
+            GOTO_bail_switch();                                             \
+        }                                                                   \
+    }
+
+
+/* File: c/opcommon.c */
+/* forward declarations of goto targets */
+GOTO_TARGET_DECL(filledNewArray, bool methodCallRange);
+GOTO_TARGET_DECL(invokeVirtual, bool methodCallRange);
+GOTO_TARGET_DECL(invokeSuper, bool methodCallRange);
+GOTO_TARGET_DECL(invokeInterface, bool methodCallRange);
+GOTO_TARGET_DECL(invokeDirect, bool methodCallRange);
+GOTO_TARGET_DECL(invokeStatic, bool methodCallRange);
+GOTO_TARGET_DECL(invokeVirtualQuick, bool methodCallRange);
+GOTO_TARGET_DECL(invokeSuperQuick, bool methodCallRange);
+GOTO_TARGET_DECL(invokeMethod, bool methodCallRange, const Method* methodToCall,
+    u2 count, u2 regs);
+GOTO_TARGET_DECL(returnFromMethod);
+GOTO_TARGET_DECL(exceptionThrown);
+
+/*
+ * ===========================================================================
+ *
+ * What follows are opcode definitions shared between multiple opcodes with
+ * minor substitutions handled by the C pre-processor.  These should probably
+ * use the mterp substitution mechanism instead, with the code here moved
+ * into common fragment files (like the asm "binop.S"), although it's hard
+ * to give up the C preprocessor in favor of the much simpler text subst.
+ *
+ * ===========================================================================
+ */
+
+#define HANDLE_NUMCONV(_opcode, _opname, _fromtype, _totype)                \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        SET_REGISTER##_totype(vdst,                                         \
+            GET_REGISTER##_fromtype(vsrc1));                                \
+        FINISH(1);
+
+#define HANDLE_FLOAT_TO_INT(_opcode, _opname, _fromvtype, _fromrtype,       \
+        _tovtype, _tortype)                                                 \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+    {                                                                       \
+        /* spec defines specific handling for +/- inf and NaN values */     \
+        _fromvtype val;                                                     \
+        _tovtype intMin, intMax, result;                                    \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        val = GET_REGISTER##_fromrtype(vsrc1);                              \
+        intMin = (_tovtype) 1 << (sizeof(_tovtype) * 8 -1);                 \
+        intMax = ~intMin;                                                   \
+        result = (_tovtype) val;                                            \
+        if (val >= intMax)          /* +inf */                              \
+            result = intMax;                                                \
+        else if (val <= intMin)     /* -inf */                              \
+            result = intMin;                                                \
+        else if (val != val)        /* NaN */                               \
+            result = 0;                                                     \
+        else                                                                \
+            result = (_tovtype) val;                                        \
+        SET_REGISTER##_tortype(vdst, result);                               \
+    }                                                                       \
+    FINISH(1);
+
+#define HANDLE_INT_TO_SMALL(_opcode, _opname, _type)                        \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|int-to-%s v%d,v%d", (_opname), vdst, vsrc1);                \
+        SET_REGISTER(vdst, (_type) GET_REGISTER(vsrc1));                    \
+        FINISH(1);
+
+/* NOTE: the comparison result is always a signed 4-byte integer */
+#define HANDLE_OP_CMPX(_opcode, _opname, _varType, _type, _nanVal)          \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        int result;                                                         \
+        u2 regs;                                                            \
+        _varType val1, val2;                                                \
+        vdst = INST_AA(inst);                                               \
+        regs = FETCH(1);                                                    \
+        vsrc1 = regs & 0xff;                                                \
+        vsrc2 = regs >> 8;                                                  \
+        ILOGV("|cmp%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);         \
+        val1 = GET_REGISTER##_type(vsrc1);                                  \
+        val2 = GET_REGISTER##_type(vsrc2);                                  \
+        if (val1 == val2)                                                   \
+            result = 0;                                                     \
+        else if (val1 < val2)                                               \
+            result = -1;                                                    \
+        else if (val1 > val2)                                               \
+            result = 1;                                                     \
+        else                                                                \
+            result = (_nanVal);                                             \
+        ILOGV("+ result=%d\n", result);                                     \
+        SET_REGISTER(vdst, result);                                         \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_IF_XX(_opcode, _opname, _cmp)                             \
+    HANDLE_OPCODE(_opcode /*vA, vB, +CCCC*/)                                \
+        vsrc1 = INST_A(inst);                                               \
+        vsrc2 = INST_B(inst);                                               \
+        if ((s4) GET_REGISTER(vsrc1) _cmp (s4) GET_REGISTER(vsrc2)) {       \
+            int branchOffset = (s2)FETCH(1);    /* sign-extended */         \
+            ILOGV("|if-%s v%d,v%d,+0x%04x", (_opname), vsrc1, vsrc2,        \
+                branchOffset);                                              \
+            ILOGV("> branch taken");                                        \
+            if (branchOffset < 0)                                           \
+                PERIODIC_CHECKS(kInterpEntryInstr, branchOffset);           \
+            FINISH(branchOffset);                                           \
+        } else {                                                            \
+            ILOGV("|if-%s v%d,v%d,-", (_opname), vsrc1, vsrc2);             \
+            FINISH(2);                                                      \
+        }
+
+#define HANDLE_OP_IF_XXZ(_opcode, _opname, _cmp)                            \
+    HANDLE_OPCODE(_opcode /*vAA, +BBBB*/)                                   \
+        vsrc1 = INST_AA(inst);                                              \
+        if ((s4) GET_REGISTER(vsrc1) _cmp 0) {                              \
+            int branchOffset = (s2)FETCH(1);    /* sign-extended */         \
+            ILOGV("|if-%s v%d,+0x%04x", (_opname), vsrc1, branchOffset);    \
+            ILOGV("> branch taken");                                        \
+            if (branchOffset < 0)                                           \
+                PERIODIC_CHECKS(kInterpEntryInstr, branchOffset);           \
+            FINISH(branchOffset);                                           \
+        } else {                                                            \
+            ILOGV("|if-%s v%d,-", (_opname), vsrc1);                        \
+            FINISH(2);                                                      \
+        }
+
+#define HANDLE_UNOP(_opcode, _opname, _pfx, _sfx, _type)                    \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        SET_REGISTER##_type(vdst, _pfx GET_REGISTER##_type(vsrc1) _sfx);    \
+        FINISH(1);
+
+#define HANDLE_OP_X_INT(_opcode, _opname, _op, _chkdiv)                     \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-int v%d,v%d", (_opname), vdst, vsrc1);                   \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            secondVal = GET_REGISTER(vsrc2);                                \
+            if (secondVal == 0) {                                           \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && secondVal == -1) {            \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            /* non-div/rem case */                                          \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vsrc1) _op (s4) GET_REGISTER(vsrc2));     \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_INT(_opcode, _opname, _cast, _op)                     \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-int v%d,v%d", (_opname), vdst, vsrc1);                   \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vsrc1) _op (GET_REGISTER(vsrc2) & 0x1f));    \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_INT_LIT16(_opcode, _opname, _op, _chkdiv)               \
+    HANDLE_OPCODE(_opcode /*vA, vB, #+CCCC*/)                               \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        vsrc2 = FETCH(1);                                                   \
+        ILOGV("|%s-int/lit16 v%d,v%d,#+0x%04x",                             \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, result;                                            \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            if ((s2) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                      \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && ((s2) vsrc2) == -1) {         \
+                /* won't generate /lit16 instr for this; check anyway */    \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op (s2) vsrc2;                           \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            /* non-div/rem case */                                          \
+            SET_REGISTER(vdst, GET_REGISTER(vsrc1) _op (s2) vsrc2);         \
+        }                                                                   \
+        FINISH(2);
+
+#define HANDLE_OP_X_INT_LIT8(_opcode, _opname, _op, _chkdiv)                \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, #+CC*/)                               \
+    {                                                                       \
+        u2 litInfo;                                                         \
+        vdst = INST_AA(inst);                                               \
+        litInfo = FETCH(1);                                                 \
+        vsrc1 = litInfo & 0xff;                                             \
+        vsrc2 = litInfo >> 8;       /* constant */                          \
+        ILOGV("|%s-int/lit8 v%d,v%d,#+0x%02x",                              \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, result;                                            \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            if ((s1) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && ((s1) vsrc2) == -1) {         \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op ((s1) vsrc2);                         \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vsrc1) _op (s1) vsrc2);                   \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_INT_LIT8(_opcode, _opname, _cast, _op)                \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, #+CC*/)                               \
+    {                                                                       \
+        u2 litInfo;                                                         \
+        vdst = INST_AA(inst);                                               \
+        litInfo = FETCH(1);                                                 \
+        vsrc1 = litInfo & 0xff;                                             \
+        vsrc2 = litInfo >> 8;       /* constant */                          \
+        ILOGV("|%s-int/lit8 v%d,v%d,#+0x%02x",                              \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vsrc1) _op (vsrc2 & 0x1f));                  \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_INT_2ADDR(_opcode, _opname, _op, _chkdiv)               \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-int-2addr v%d,v%d", (_opname), vdst, vsrc1);             \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER(vdst);                                  \
+            secondVal = GET_REGISTER(vsrc1);                                \
+            if (secondVal == 0) {                                           \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && secondVal == -1) {            \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vdst) _op (s4) GET_REGISTER(vsrc1));      \
+        }                                                                   \
+        FINISH(1);
+
+#define HANDLE_OP_SHX_INT_2ADDR(_opcode, _opname, _cast, _op)               \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-int-2addr v%d,v%d", (_opname), vdst, vsrc1);             \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vdst) _op (GET_REGISTER(vsrc1) & 0x1f));     \
+        FINISH(1);
+
+#define HANDLE_OP_X_LONG(_opcode, _opname, _op, _chkdiv)                    \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-long v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);       \
+        if (_chkdiv != 0) {                                                 \
+            s8 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER_WIDE(vsrc1);                            \
+            secondVal = GET_REGISTER_WIDE(vsrc2);                           \
+            if (secondVal == 0LL) {                                         \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u8)firstVal == 0x8000000000000000ULL &&                    \
+                secondVal == -1LL)                                          \
+            {                                                               \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER_WIDE(vdst, result);                                \
+        } else {                                                            \
+            SET_REGISTER_WIDE(vdst,                                         \
+                (s8) GET_REGISTER_WIDE(vsrc1) _op (s8) GET_REGISTER_WIDE(vsrc2)); \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_LONG(_opcode, _opname, _cast, _op)                    \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-long v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);       \
+        SET_REGISTER_WIDE(vdst,                                             \
+            _cast GET_REGISTER_WIDE(vsrc1) _op (GET_REGISTER(vsrc2) & 0x3f)); \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_LONG_2ADDR(_opcode, _opname, _op, _chkdiv)              \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-long-2addr v%d,v%d", (_opname), vdst, vsrc1);            \
+        if (_chkdiv != 0) {                                                 \
+            s8 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER_WIDE(vdst);                             \
+            secondVal = GET_REGISTER_WIDE(vsrc1);                           \
+            if (secondVal == 0LL) {                                         \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u8)firstVal == 0x8000000000000000ULL &&                    \
+                secondVal == -1LL)                                          \
+            {                                                               \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER_WIDE(vdst, result);                                \
+        } else {                                                            \
+            SET_REGISTER_WIDE(vdst,                                         \
+                (s8) GET_REGISTER_WIDE(vdst) _op (s8)GET_REGISTER_WIDE(vsrc1));\
+        }                                                                   \
+        FINISH(1);
+
+#define HANDLE_OP_SHX_LONG_2ADDR(_opcode, _opname, _cast, _op)              \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-long-2addr v%d,v%d", (_opname), vdst, vsrc1);            \
+        SET_REGISTER_WIDE(vdst,                                             \
+            _cast GET_REGISTER_WIDE(vdst) _op (GET_REGISTER(vsrc1) & 0x3f)); \
+        FINISH(1);
+
+#define HANDLE_OP_X_FLOAT(_opcode, _opname, _op)                            \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-float v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);      \
+        SET_REGISTER_FLOAT(vdst,                                            \
+            GET_REGISTER_FLOAT(vsrc1) _op GET_REGISTER_FLOAT(vsrc2));       \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_DOUBLE(_opcode, _opname, _op)                           \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-double v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);     \
+        SET_REGISTER_DOUBLE(vdst,                                           \
+            GET_REGISTER_DOUBLE(vsrc1) _op GET_REGISTER_DOUBLE(vsrc2));     \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_FLOAT_2ADDR(_opcode, _opname, _op)                      \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-float-2addr v%d,v%d", (_opname), vdst, vsrc1);           \
+        SET_REGISTER_FLOAT(vdst,                                            \
+            GET_REGISTER_FLOAT(vdst) _op GET_REGISTER_FLOAT(vsrc1));        \
+        FINISH(1);
+
+#define HANDLE_OP_X_DOUBLE_2ADDR(_opcode, _opname, _op)                     \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-double-2addr v%d,v%d", (_opname), vdst, vsrc1);          \
+        SET_REGISTER_DOUBLE(vdst,                                           \
+            GET_REGISTER_DOUBLE(vdst) _op GET_REGISTER_DOUBLE(vsrc1));      \
+        FINISH(1);
+
+#define HANDLE_OP_AGET(_opcode, _opname, _type, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        ArrayObject* arrayObj;                                              \
+        u2 arrayInfo;                                                       \
+        EXPORT_PC();                                                        \
+        vdst = INST_AA(inst);                                               \
+        arrayInfo = FETCH(1);                                               \
+        vsrc1 = arrayInfo & 0xff;    /* array ptr */                        \
+        vsrc2 = arrayInfo >> 8;      /* index */                            \
+        ILOGV("|aget%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);        \
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);                      \
+        if (!checkForNull((Object*) arrayObj))                              \
+            GOTO_exceptionThrown();                                         \
+        if (GET_REGISTER(vsrc2) >= arrayObj->length) {                      \
+            LOGV("Invalid array access: %p %d (len=%d)\n",                  \
+                arrayObj, vsrc2, arrayObj->length);                         \
+            dvmThrowException("Ljava/lang/ArrayIndexOutOfBoundsException;", \
+                NULL);                                                      \
+            GOTO_exceptionThrown();                                         \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst,                                        \
+            ((_type*) arrayObj->contents)[GET_REGISTER(vsrc2)]);            \
+        ILOGV("+ AGET[%d]=0x%x", GET_REGISTER(vsrc2), GET_REGISTER(vdst));  \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_APUT(_opcode, _opname, _type, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        ArrayObject* arrayObj;                                              \
+        u2 arrayInfo;                                                       \
+        EXPORT_PC();                                                        \
+        vdst = INST_AA(inst);       /* AA: source value */                  \
+        arrayInfo = FETCH(1);                                               \
+        vsrc1 = arrayInfo & 0xff;   /* BB: array ptr */                     \
+        vsrc2 = arrayInfo >> 8;     /* CC: index */                         \
+        ILOGV("|aput%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);        \
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);                      \
+        if (!checkForNull((Object*) arrayObj))                              \
+            GOTO_exceptionThrown();                                         \
+        if (GET_REGISTER(vsrc2) >= arrayObj->length) {                      \
+            dvmThrowException("Ljava/lang/ArrayIndexOutOfBoundsException;", \
+                NULL);                                                      \
+            GOTO_exceptionThrown();                                         \
+        }                                                                   \
+        ILOGV("+ APUT[%d]=0x%08x", GET_REGISTER(vsrc2), GET_REGISTER(vdst));\
+        ((_type*) arrayObj->contents)[GET_REGISTER(vsrc2)] =                \
+            GET_REGISTER##_regsize(vdst);                                   \
+    }                                                                       \
+    FINISH(2);
+
+/*
+ * It's possible to get a bad value out of a field with sub-32-bit stores
+ * because the -quick versions always operate on 32 bits.  Consider:
+ *   short foo = -1  (sets a 32-bit register to 0xffffffff)
+ *   iput-quick foo  (writes all 32 bits to the field)
+ *   short bar = 1   (sets a 32-bit register to 0x00000001)
+ *   iput-short      (writes the low 16 bits to the field)
+ *   iget-quick foo  (reads all 32 bits from the field, yielding 0xffff0001)
+ * This can only happen when optimized and non-optimized code has interleaved
+ * access to the same field.  This is unlikely but possible.
+ *
+ * The easiest way to fix this is to always read/write 32 bits at a time.  On
+ * a device with a 16-bit data bus this is sub-optimal.  (The alternative
+ * approach is to have sub-int versions of iget-quick, but now we're wasting
+ * Dalvik instruction space and making it less likely that handler code will
+ * already be in the CPU i-cache.)
+ */
+#define HANDLE_IGET_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        InstField* ifield;                                                  \
+        Object* obj;                                                        \
+        EXPORT_PC();                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|iget%s v%d,v%d,field@0x%04x", (_opname), vdst, vsrc1, ref); \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNull(obj))                                             \
+            GOTO_exceptionThrown();                                         \
+        ifield = (InstField*) dvmDexGetResolvedField(methodClassDex, ref);  \
+        if (ifield == NULL) {                                               \
+            ifield = dvmResolveInstField(curMethod->clazz, ref);            \
+            if (ifield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst,                                        \
+            dvmGetField##_ftype(obj, ifield->byteOffset));                  \
+        ILOGV("+ IGET '%s'=0x%08llx", ifield->field.name,                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+        UPDATE_FIELD_GET(&ifield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IGET_X_QUICK(_opcode, _opname, _ftype, _regsize)             \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        Object* obj;                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field offset */                          \
+        ILOGV("|iget%s-quick v%d,v%d,field@+%u",                            \
+            (_opname), vdst, vsrc1, ref);                                   \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNullExportPC(obj, fp, pc))                             \
+            GOTO_exceptionThrown();                                         \
+        SET_REGISTER##_regsize(vdst, dvmGetField##_ftype(obj, ref));        \
+        ILOGV("+ IGETQ %d=0x%08llx", ref,                                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IPUT_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        InstField* ifield;                                                  \
+        Object* obj;                                                        \
+        EXPORT_PC();                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|iput%s v%d,v%d,field@0x%04x", (_opname), vdst, vsrc1, ref); \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNull(obj))                                             \
+            GOTO_exceptionThrown();                                         \
+        ifield = (InstField*) dvmDexGetResolvedField(methodClassDex, ref);  \
+        if (ifield == NULL) {                                               \
+            ifield = dvmResolveInstField(curMethod->clazz, ref);            \
+            if (ifield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        dvmSetField##_ftype(obj, ifield->byteOffset,                        \
+            GET_REGISTER##_regsize(vdst));                                  \
+        ILOGV("+ IPUT '%s'=0x%08llx", ifield->field.name,                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+        UPDATE_FIELD_PUT(&ifield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IPUT_X_QUICK(_opcode, _opname, _ftype, _regsize)             \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        Object* obj;                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field offset */                          \
+        ILOGV("|iput%s-quick v%d,v%d,field@0x%04x",                         \
+            (_opname), vdst, vsrc1, ref);                                   \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNullExportPC(obj, fp, pc))                             \
+            GOTO_exceptionThrown();                                         \
+        dvmSetField##_ftype(obj, ref, GET_REGISTER##_regsize(vdst));        \
+        ILOGV("+ IPUTQ %d=0x%08llx", ref,                                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_SGET_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, field@BBBB*/)                              \
+    {                                                                       \
+        StaticField* sfield;                                                \
+        vdst = INST_AA(inst);                                               \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|sget%s v%d,sfield@0x%04x", (_opname), vdst, ref);           \
+        sfield = (StaticField*)dvmDexGetResolvedField(methodClassDex, ref); \
+        if (sfield == NULL) {                                               \
+            EXPORT_PC();                                                    \
+            sfield = dvmResolveStaticField(curMethod->clazz, ref);          \
+            if (sfield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst, dvmGetStaticField##_ftype(sfield));    \
+        ILOGV("+ SGET '%s'=0x%08llx",                                       \
+            sfield->field.name, (u8)GET_REGISTER##_regsize(vdst));          \
+        UPDATE_FIELD_GET(&sfield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_SPUT_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, field@BBBB*/)                              \
+    {                                                                       \
+        StaticField* sfield;                                                \
+        vdst = INST_AA(inst);                                               \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|sput%s v%d,sfield@0x%04x", (_opname), vdst, ref);           \
+        sfield = (StaticField*)dvmDexGetResolvedField(methodClassDex, ref); \
+        if (sfield == NULL) {                                               \
+            EXPORT_PC();                                                    \
+            sfield = dvmResolveStaticField(curMethod->clazz, ref);          \
+            if (sfield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        dvmSetStaticField##_ftype(sfield, GET_REGISTER##_regsize(vdst));    \
+        ILOGV("+ SPUT '%s'=0x%08llx",                                       \
+            sfield->field.name, (u8)GET_REGISTER##_regsize(vdst));          \
+        UPDATE_FIELD_PUT(&sfield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+
+/* File: c/gotoTargets.c */
+/*
+ * C footer.  This has some common code shared by the various targets.
+ */
+
+/*
+ * Everything from here on is a "goto target".  In the basic interpreter
+ * we jump into these targets and then jump directly to the handler for
+ * next instruction.  Here, these are subroutines that return to the caller.
+ */
+
+GOTO_TARGET(filledNewArray, bool methodCallRange)
+    {
+        ClassObject* arrayClass;
+        ArrayObject* newArray;
+        u4* contents;
+        char typeCh;
+        int i;
+        u4 arg5;
+
+        EXPORT_PC();
+
+        ref = FETCH(1);             /* class ref */
+        vdst = FETCH(2);            /* first 4 regs -or- range base */
+
+        if (methodCallRange) {
+            vsrc1 = INST_AA(inst);  /* #of elements */
+            arg5 = -1;              /* silence compiler warning */
+            ILOGV("|filled-new-array-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+        } else {
+            arg5 = INST_A(inst);
+            vsrc1 = INST_B(inst);   /* #of elements */
+            ILOGV("|filled-new-array args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1, ref, vdst, arg5);
+        }
+
+        /*
+         * Resolve the array class.
+         */
+        arrayClass = dvmDexGetResolvedClass(methodClassDex, ref);
+        if (arrayClass == NULL) {
+            arrayClass = dvmResolveClass(curMethod->clazz, ref, false);
+            if (arrayClass == NULL)
+                GOTO_exceptionThrown();
+        }
+        /*
+        if (!dvmIsArrayClass(arrayClass)) {
+            dvmThrowException("Ljava/lang/RuntimeError;",
+                "filled-new-array needs array class");
+            GOTO_exceptionThrown();
+        }
+        */
+        /* verifier guarantees this is an array class */
+        assert(dvmIsArrayClass(arrayClass));
+        assert(dvmIsClassInitialized(arrayClass));
+
+        /*
+         * Create an array of the specified type.
+         */
+        LOGVV("+++ filled-new-array type is '%s'\n", arrayClass->descriptor);
+        typeCh = arrayClass->descriptor[1];
+        if (typeCh == 'D' || typeCh == 'J') {
+            /* category 2 primitives not allowed */
+            dvmThrowException("Ljava/lang/RuntimeError;",
+                "bad filled array req");
+            GOTO_exceptionThrown();
+        } else if (typeCh != 'L' && typeCh != '[' && typeCh != 'I') {
+            /* TODO: requires multiple "fill in" loops with different widths */
+            LOGE("non-int primitives not implemented\n");
+            dvmThrowException("Ljava/lang/InternalError;",
+                "filled-new-array not implemented for anything but 'int'");
+            GOTO_exceptionThrown();
+        }
+
+        newArray = dvmAllocArrayByClass(arrayClass, vsrc1, ALLOC_DONT_TRACK);
+        if (newArray == NULL)
+            GOTO_exceptionThrown();
+
+        /*
+         * Fill in the elements.  It's legal for vsrc1 to be zero.
+         */
+        contents = (u4*) newArray->contents;
+        if (methodCallRange) {
+            for (i = 0; i < vsrc1; i++)
+                contents[i] = GET_REGISTER(vdst+i);
+        } else {
+            assert(vsrc1 <= 5);
+            if (vsrc1 == 5) {
+                contents[4] = GET_REGISTER(arg5);
+                vsrc1--;
+            }
+            for (i = 0; i < vsrc1; i++) {
+                contents[i] = GET_REGISTER(vdst & 0x0f);
+                vdst >>= 4;
+            }
+        }
+
+        retval.l = newArray;
+    }
+    FINISH(3);
+GOTO_TARGET_END
+
+
+GOTO_TARGET(invokeVirtual, bool methodCallRange)
+    {
+        Method* baseMethod;
+        Object* thisPtr;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* method ref */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        /*
+         * The object against which we are executing a method is always
+         * in the first argument.
+         */
+        if (methodCallRange) {
+            assert(vsrc1 > 0);
+            ILOGV("|invoke-virtual-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisPtr = (Object*) GET_REGISTER(vdst);
+        } else {
+            assert((vsrc1>>4) > 0);
+            ILOGV("|invoke-virtual args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisPtr = (Object*) GET_REGISTER(vdst & 0x0f);
+        }
+
+        if (!checkForNull(thisPtr))
+            GOTO_exceptionThrown();
+
+        /*
+         * Resolve the method.  This is the correct method for the static
+         * type of the object.  We also verify access permissions here.
+         */
+        baseMethod = dvmDexGetResolvedMethod(methodClassDex, ref);
+        if (baseMethod == NULL) {
+            baseMethod = dvmResolveMethod(curMethod->clazz, ref,METHOD_VIRTUAL);
+            if (baseMethod == NULL) {
+                ILOGV("+ unknown method or access denied\n");
+                GOTO_exceptionThrown();
+            }
+        }
+
+        /*
+         * Combine the object we found with the vtable offset in the
+         * method.
+         */
+        assert(baseMethod->methodIndex < thisPtr->clazz->vtableCount);
+        methodToCall = thisPtr->clazz->vtable[baseMethod->methodIndex];
+
+#if 0
+        if (dvmIsAbstractMethod(methodToCall)) {
+            /*
+             * This can happen if you create two classes, Base and Sub, where
+             * Sub is a sub-class of Base.  Declare a protected abstract
+             * method foo() in Base, and invoke foo() from a method in Base.
+             * Base is an "abstract base class" and is never instantiated
+             * directly.  Now, Override foo() in Sub, and use Sub.  This
+             * Works fine unless Sub stops providing an implementation of
+             * the method.
+             */
+            dvmThrowException("Ljava/lang/AbstractMethodError;",
+                "abstract method not implemented");
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(!dvmIsAbstractMethod(methodToCall) ||
+            methodToCall->nativeFunc != NULL);
+#endif
+
+        LOGVV("+++ base=%s.%s virtual[%d]=%s.%s\n",
+            baseMethod->clazz->descriptor, baseMethod->name,
+            (u4) baseMethod->methodIndex,
+            methodToCall->clazz->descriptor, methodToCall->name);
+        assert(methodToCall != NULL);
+
+#if 0
+        if (vsrc1 != methodToCall->insSize) {
+            LOGW("WRONG METHOD: base=%s.%s virtual[%d]=%s.%s\n",
+                baseMethod->clazz->descriptor, baseMethod->name,
+                (u4) baseMethod->methodIndex,
+                methodToCall->clazz->descriptor, methodToCall->name);
+            //dvmDumpClass(baseMethod->clazz);
+            //dvmDumpClass(methodToCall->clazz);
+            dvmDumpAllClasses(0);
+        }
+#endif
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeSuper, bool methodCallRange)
+    {
+        Method* baseMethod;
+        u2 thisReg;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* method ref */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        if (methodCallRange) {
+            ILOGV("|invoke-super-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisReg = vdst;
+        } else {
+            ILOGV("|invoke-super args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisReg = vdst & 0x0f;
+        }
+        /* impossible in well-formed code, but we must check nevertheless */
+        if (!checkForNull((Object*) GET_REGISTER(thisReg)))
+            GOTO_exceptionThrown();
+
+        /*
+         * Resolve the method.  This is the correct method for the static
+         * type of the object.  We also verify access permissions here.
+         * The first arg to dvmResolveMethod() is just the referring class
+         * (used for class loaders and such), so we don't want to pass
+         * the superclass into the resolution call.
+         */
+        baseMethod = dvmDexGetResolvedMethod(methodClassDex, ref);
+        if (baseMethod == NULL) {
+            baseMethod = dvmResolveMethod(curMethod->clazz, ref,METHOD_VIRTUAL);
+            if (baseMethod == NULL) {
+                ILOGV("+ unknown method or access denied\n");
+                GOTO_exceptionThrown();
+            }
+        }
+
+        /*
+         * Combine the object we found with the vtable offset in the
+         * method's class.
+         *
+         * We're using the current method's class' superclass, not the
+         * superclass of "this".  This is because we might be executing
+         * in a method inherited from a superclass, and we want to run
+         * in that class' superclass.
+         */
+        if (baseMethod->methodIndex >= curMethod->clazz->super->vtableCount) {
+            /*
+             * Method does not exist in the superclass.  Could happen if
+             * superclass gets updated.
+             */
+            dvmThrowException("Ljava/lang/NoSuchMethodError;",
+                baseMethod->name);
+            GOTO_exceptionThrown();
+        }
+        methodToCall = curMethod->clazz->super->vtable[baseMethod->methodIndex];
+#if 0
+        if (dvmIsAbstractMethod(methodToCall)) {
+            dvmThrowException("Ljava/lang/AbstractMethodError;",
+                "abstract method not implemented");
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(!dvmIsAbstractMethod(methodToCall) ||
+            methodToCall->nativeFunc != NULL);
+#endif
+        LOGVV("+++ base=%s.%s super-virtual=%s.%s\n",
+            baseMethod->clazz->descriptor, baseMethod->name,
+            methodToCall->clazz->descriptor, methodToCall->name);
+        assert(methodToCall != NULL);
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeInterface, bool methodCallRange)
+    {
+        Object* thisPtr;
+        ClassObject* thisClass;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* method ref */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        /*
+         * The object against which we are executing a method is always
+         * in the first argument.
+         */
+        if (methodCallRange) {
+            assert(vsrc1 > 0);
+            ILOGV("|invoke-interface-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisPtr = (Object*) GET_REGISTER(vdst);
+        } else {
+            assert((vsrc1>>4) > 0);
+            ILOGV("|invoke-interface args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisPtr = (Object*) GET_REGISTER(vdst & 0x0f);
+        }
+        if (!checkForNull(thisPtr))
+            GOTO_exceptionThrown();
+
+        thisClass = thisPtr->clazz;
+
+        /*
+         * Given a class and a method index, find the Method* with the
+         * actual code we want to execute.
+         */
+        methodToCall = dvmFindInterfaceMethodInCache(thisClass, ref, curMethod,
+                        methodClassDex);
+        if (methodToCall == NULL) {
+            assert(dvmCheckException(self));
+            GOTO_exceptionThrown();
+        }
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeDirect, bool methodCallRange)
+    {
+        u2 thisReg;
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* method ref */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        EXPORT_PC();
+
+        if (methodCallRange) {
+            ILOGV("|invoke-direct-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisReg = vdst;
+        } else {
+            ILOGV("|invoke-direct args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisReg = vdst & 0x0f;
+        }
+        if (!checkForNull((Object*) GET_REGISTER(thisReg)))
+            GOTO_exceptionThrown();
+
+        methodToCall = dvmDexGetResolvedMethod(methodClassDex, ref);
+        if (methodToCall == NULL) {
+            methodToCall = dvmResolveMethod(curMethod->clazz, ref,
+                            METHOD_DIRECT);
+            if (methodToCall == NULL) {
+                ILOGV("+ unknown direct method\n");     // should be impossible
+                GOTO_exceptionThrown();
+            }
+        }
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeStatic, bool methodCallRange)
+    vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+    ref = FETCH(1);             /* method ref */
+    vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+    EXPORT_PC();
+
+    if (methodCallRange)
+        ILOGV("|invoke-static-range args=%d @0x%04x {regs=v%d-v%d}",
+            vsrc1, ref, vdst, vdst+vsrc1-1);
+    else
+        ILOGV("|invoke-static args=%d @0x%04x {regs=0x%04x %x}",
+            vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+
+    methodToCall = dvmDexGetResolvedMethod(methodClassDex, ref);
+    if (methodToCall == NULL) {
+        methodToCall = dvmResolveMethod(curMethod->clazz, ref, METHOD_STATIC);
+        if (methodToCall == NULL) {
+            ILOGV("+ unknown method\n");
+            GOTO_exceptionThrown();
+        }
+    }
+    GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeVirtualQuick, bool methodCallRange)
+    {
+        Object* thisPtr;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* vtable index */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        /*
+         * The object against which we are executing a method is always
+         * in the first argument.
+         */
+        if (methodCallRange) {
+            assert(vsrc1 > 0);
+            ILOGV("|invoke-virtual-quick-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisPtr = (Object*) GET_REGISTER(vdst);
+        } else {
+            assert((vsrc1>>4) > 0);
+            ILOGV("|invoke-virtual-quick args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisPtr = (Object*) GET_REGISTER(vdst & 0x0f);
+        }
+
+        if (!checkForNull(thisPtr))
+            GOTO_exceptionThrown();
+
+        /*
+         * Combine the object we found with the vtable offset in the
+         * method.
+         */
+        assert(ref < thisPtr->clazz->vtableCount);
+        methodToCall = thisPtr->clazz->vtable[ref];
+
+#if 0
+        if (dvmIsAbstractMethod(methodToCall)) {
+            dvmThrowException("Ljava/lang/AbstractMethodError;",
+                "abstract method not implemented");
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(!dvmIsAbstractMethod(methodToCall) ||
+            methodToCall->nativeFunc != NULL);
+#endif
+
+        LOGVV("+++ virtual[%d]=%s.%s\n",
+            ref, methodToCall->clazz->descriptor, methodToCall->name);
+        assert(methodToCall != NULL);
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+GOTO_TARGET(invokeSuperQuick, bool methodCallRange)
+    {
+        u2 thisReg;
+
+        EXPORT_PC();
+
+        vsrc1 = INST_AA(inst);      /* AA (count) or BA (count + arg 5) */
+        ref = FETCH(1);             /* vtable index */
+        vdst = FETCH(2);            /* 4 regs -or- first reg */
+
+        if (methodCallRange) {
+            ILOGV("|invoke-super-quick-range args=%d @0x%04x {regs=v%d-v%d}",
+                vsrc1, ref, vdst, vdst+vsrc1-1);
+            thisReg = vdst;
+        } else {
+            ILOGV("|invoke-super-quick args=%d @0x%04x {regs=0x%04x %x}",
+                vsrc1 >> 4, ref, vdst, vsrc1 & 0x0f);
+            thisReg = vdst & 0x0f;
+        }
+        /* impossible in well-formed code, but we must check nevertheless */
+        if (!checkForNull((Object*) GET_REGISTER(thisReg)))
+            GOTO_exceptionThrown();
+
+#if 0   /* impossible in optimized + verified code */
+        if (ref >= curMethod->clazz->super->vtableCount) {
+            dvmThrowException("Ljava/lang/NoSuchMethodError;", NULL);
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(ref < curMethod->clazz->super->vtableCount);
+#endif
+
+        /*
+         * Combine the object we found with the vtable offset in the
+         * method's class.
+         *
+         * We're using the current method's class' superclass, not the
+         * superclass of "this".  This is because we might be executing
+         * in a method inherited from a superclass, and we want to run
+         * in the method's class' superclass.
+         */
+        methodToCall = curMethod->clazz->super->vtable[ref];
+
+#if 0
+        if (dvmIsAbstractMethod(methodToCall)) {
+            dvmThrowException("Ljava/lang/AbstractMethodError;",
+                "abstract method not implemented");
+            GOTO_exceptionThrown();
+        }
+#else
+        assert(!dvmIsAbstractMethod(methodToCall) ||
+            methodToCall->nativeFunc != NULL);
+#endif
+        LOGVV("+++ super-virtual[%d]=%s.%s\n",
+            ref, methodToCall->clazz->descriptor, methodToCall->name);
+        assert(methodToCall != NULL);
+
+        GOTO_invokeMethod(methodCallRange, methodToCall, vsrc1, vdst);
+    }
+GOTO_TARGET_END
+
+
+
+    /*
+     * General handling for return-void, return, and return-wide.  Put the
+     * return value in "retval" before jumping here.
+     */
+GOTO_TARGET(returnFromMethod)
+    {
+        StackSaveArea* saveArea;
+
+        /*
+         * We must do this BEFORE we pop the previous stack frame off, so
+         * that the GC can see the return value (if any) in the local vars.
+         *
+         * Since this is now an interpreter switch point, we must do it before
+         * we do anything at all.
+         */
+        PERIODIC_CHECKS(kInterpEntryReturn, 0);
+
+        ILOGV("> retval=0x%llx (leaving %s.%s %s)",
+            retval.j, curMethod->clazz->descriptor, curMethod->name,
+            curMethod->signature);
+        //DUMP_REGS(curMethod, fp);
+
+        saveArea = SAVEAREA_FROM_FP(fp);
+
+#ifdef EASY_GDB
+        debugSaveArea = saveArea;
+#endif
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_PROFILER)
+        TRACE_METHOD_EXIT(self, curMethod);
+#endif
+
+        /* back up to previous frame and see if we hit a break */
+        fp = saveArea->prevFrame;
+        assert(fp != NULL);
+        if (dvmIsBreakFrame(fp)) {
+            /* bail without popping the method frame from stack */
+            LOGVV("+++ returned into break frame\n");
+            GOTO_bail();
+        }
+
+        /* update thread FP, and reset local variables */
+        self->curFrame = fp;
+        curMethod = SAVEAREA_FROM_FP(fp)->method;
+        //methodClass = curMethod->clazz;
+        methodClassDex = curMethod->clazz->pDvmDex;
+        pc = saveArea->savedPc;
+        ILOGD("> (return to %s.%s %s)", curMethod->clazz->descriptor,
+            curMethod->name, curMethod->signature);
+
+        /* use FINISH on the caller's invoke instruction */
+        //u2 invokeInstr = INST_INST(FETCH(0));
+        if (true /*invokeInstr >= OP_INVOKE_VIRTUAL &&
+            invokeInstr <= OP_INVOKE_INTERFACE*/)
+        {
+            FINISH(3);
+        } else {
+            //LOGE("Unknown invoke instr %02x at %d\n",
+            //    invokeInstr, (int) (pc - curMethod->insns));
+            assert(false);
+        }
+    }
+GOTO_TARGET_END
+
+
+    /*
+     * Jump here when the code throws an exception.
+     *
+     * By the time we get here, the Throwable has been created and the stack
+     * trace has been saved off.
+     */
+GOTO_TARGET(exceptionThrown)
+    {
+        Object* exception;
+        int catchRelPc;
+
+        /*
+         * Since this is now an interpreter switch point, we must do it before
+         * we do anything at all.
+         */
+        PERIODIC_CHECKS(kInterpEntryThrow, 0);
+
+        /*
+         * We save off the exception and clear the exception status.  While
+         * processing the exception we might need to load some Throwable
+         * classes, and we don't want class loader exceptions to get
+         * confused with this one.
+         */
+        assert(dvmCheckException(self));
+        exception = dvmGetException(self);
+        dvmAddTrackedAlloc(exception, self);
+        dvmClearException(self);
+
+        LOGV("Handling exception %s at %s:%d\n",
+            exception->clazz->descriptor, curMethod->name,
+            dvmLineNumFromPC(curMethod, pc - curMethod->insns));
+
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_DEBUGGER)
+        /*
+         * Tell the debugger about it.
+         *
+         * TODO: if the exception was thrown by interpreted code, control
+         * fell through native, and then back to us, we will report the
+         * exception at the point of the throw and again here.  We can avoid
+         * this by not reporting exceptions when we jump here directly from
+         * the native call code above, but then we won't report exceptions
+         * that were thrown *from* the JNI code (as opposed to *through* it).
+         *
+         * The correct solution is probably to ignore from-native exceptions
+         * here, and have the JNI exception code do the reporting to the
+         * debugger.
+         */
+        if (gDvm.debuggerActive) {
+            void* catchFrame;
+            catchRelPc = dvmFindCatchBlock(self, pc - curMethod->insns,
+                        exception, true, &catchFrame);
+            dvmDbgPostException(fp, pc - curMethod->insns, catchFrame,
+                catchRelPc, exception);
+        }
+#endif
+
+        /*
+         * We need to unroll to the catch block or the nearest "break"
+         * frame.
+         *
+         * A break frame could indicate that we have reached an intermediate
+         * native call, or have gone off the top of the stack and the thread
+         * needs to exit.  Either way, we return from here, leaving the
+         * exception raised.
+         *
+         * If we do find a catch block, we want to transfer execution to
+         * that point.
+         */
+        catchRelPc = dvmFindCatchBlock(self, pc - curMethod->insns,
+                    exception, false, (void*)&fp);
+
+        /*
+         * Restore the stack bounds after an overflow.  This isn't going to
+         * be correct in all circumstances, e.g. if JNI code devours the
+         * exception this won't happen until some other exception gets
+         * thrown.  If the code keeps pushing the stack bounds we'll end
+         * up aborting the VM.
+         *
+         * Note we want to do this *after* the call to dvmFindCatchBlock,
+         * because that may need extra stack space to resolve exception
+         * classes (e.g. through a class loader).
+         */
+        if (self->stackOverflowed)
+            dvmCleanupStackOverflow(self);
+
+        if (catchRelPc < 0) {
+            /* falling through to JNI code or off the bottom of the stack */
+#if DVM_SHOW_EXCEPTION >= 2
+            LOGD("Exception %s from %s:%d not caught locally\n",
+                exception->clazz->descriptor, dvmGetMethodSourceFile(curMethod),
+                dvmLineNumFromPC(curMethod, pc - curMethod->insns));
+#endif
+            dvmSetException(self, exception);
+            dvmReleaseTrackedAlloc(exception, self);
+            GOTO_bail();
+        }
+
+#if DVM_SHOW_EXCEPTION >= 3
+        {
+            const Method* catchMethod = SAVEAREA_FROM_FP(fp)->method;
+            LOGD("Exception %s thrown from %s:%d to %s:%d\n",
+                exception->clazz->descriptor, dvmGetMethodSourceFile(curMethod),
+                dvmLineNumFromPC(curMethod, pc - curMethod->insns),
+                dvmGetMethodSourceFile(catchMethod),
+                dvmLineNumFromPC(catchMethod, catchRelPc));
+        }
+#endif
+
+        /*
+         * Adjust local variables to match self->curFrame and the
+         * updated PC.
+         */
+        //fp = (u4*) self->curFrame;
+        curMethod = SAVEAREA_FROM_FP(fp)->method;
+        //methodClass = curMethod->clazz;
+        methodClassDex = curMethod->clazz->pDvmDex;
+        pc = curMethod->insns + catchRelPc;
+        ILOGV("> pc <-- %s.%s %s", curMethod->clazz->descriptor,
+            curMethod->name, curMethod->signature);
+        DUMP_REGS(curMethod, fp, false);            // show all regs
+
+        /*
+         * Restore the exception if the handler wants it.
+         *
+         * The Dalvik spec mandates that, if an exception handler wants to
+         * do something with the exception, the first instruction executed
+         * must be "move-exception".  We can pass the exception along
+         * through the thread struct, and let the move-exception instruction
+         * clear it for us.
+         *
+         * If the handler doesn't call move-exception, we don't want to
+         * finish here with an exception still pending.
+         */
+        if (INST_INST(FETCH(0)) == OP_MOVE_EXCEPTION)
+            dvmSetException(self, exception);
+
+        dvmReleaseTrackedAlloc(exception, self);
+        FINISH(0);
+    }
+GOTO_TARGET_END
+
+
+    /*
+     * General handling for invoke-{virtual,super,direct,static,interface},
+     * including "quick" variants.
+     *
+     * Set "methodToCall" to the Method we're calling, and "methodCallRange"
+     * depending on whether this is a "/range" instruction.
+     *
+     * For a range call:
+     *  "vsrc1" holds the argument count (8 bits)
+     *  "vdst" holds the first argument in the range
+     * For a non-range call:
+     *  "vsrc1" holds the argument count (4 bits) and the 5th argument index
+     *  "vdst" holds four 4-bit register indices
+     *
+     * The caller must EXPORT_PC before jumping here, because any method
+     * call can throw a stack overflow exception.
+     */
+GOTO_TARGET(invokeMethod, bool methodCallRange, const Method* _methodToCall,
+    u2 count, u2 regs)
+    {
+        STUB_HACK(vsrc1 = count; vdst = regs; methodToCall = _methodToCall;);
+
+        //printf("range=%d call=%p count=%d regs=0x%04x\n",
+        //    methodCallRange, methodToCall, count, regs);
+        //printf(" --> %s.%s %s\n", methodToCall->clazz->descriptor,
+        //    methodToCall->name, methodToCall->signature);
+
+        u4* outs;
+        int i;
+
+        /*
+         * Copy args.  This may corrupt vsrc1/vdst.
+         */
+        if (methodCallRange) {
+            // could use memcpy or a "Duff's device"; most functions have
+            // so few args it won't matter much
+            assert(vsrc1 <= curMethod->outsSize);
+            assert(vsrc1 == methodToCall->insSize);
+            outs = OUTS_FROM_FP(fp, vsrc1);
+            for (i = 0; i < vsrc1; i++)
+                outs[i] = GET_REGISTER(vdst+i);
+        } else {
+            u4 count = vsrc1 >> 4;
+
+            assert(count <= curMethod->outsSize);
+            assert(count == methodToCall->insSize);
+            assert(count <= 5);
+
+            outs = OUTS_FROM_FP(fp, count);
+#if 0
+            if (count == 5) {
+                outs[4] = GET_REGISTER(vsrc1 & 0x0f);
+                count--;
+            }
+            for (i = 0; i < (int) count; i++) {
+                outs[i] = GET_REGISTER(vdst & 0x0f);
+                vdst >>= 4;
+            }
+#else
+            // This version executes fewer instructions but is larger
+            // overall.  Seems to be a teensy bit faster.
+            assert((vdst >> 16) == 0);  // 16 bits -or- high 16 bits clear
+            switch (count) {
+            case 5:
+                outs[4] = GET_REGISTER(vsrc1 & 0x0f);
+            case 4:
+                outs[3] = GET_REGISTER(vdst >> 12);
+            case 3:
+                outs[2] = GET_REGISTER((vdst & 0x0f00) >> 8);
+            case 2:
+                outs[1] = GET_REGISTER((vdst & 0x00f0) >> 4);
+            case 1:
+                outs[0] = GET_REGISTER(vdst & 0x0f);
+            default:
+                ;
+            }
+#endif
+        }
+    }
+
+    /*
+     * (This was originally a "goto" target; I've kept it separate from the
+     * stuff above in case we want to refactor things again.)
+     *
+     * At this point, we have the arguments stored in the "outs" area of
+     * the current method's stack frame, and the method to call in
+     * "methodToCall".  Push a new stack frame.
+     */
+    {
+        StackSaveArea* newSaveArea;
+        u4* newFp;
+
+        ILOGV("> %s%s.%s %s",
+            dvmIsNativeMethod(methodToCall) ? "(NATIVE) " : "",
+            methodToCall->clazz->descriptor, methodToCall->name,
+            methodToCall->signature);
+
+        newFp = (u4*) SAVEAREA_FROM_FP(fp) - methodToCall->registersSize;
+        newSaveArea = SAVEAREA_FROM_FP(newFp);
+
+        /* verify that we have enough space */
+        if (true) {
+            u1* bottom;
+            bottom = (u1*) newSaveArea - methodToCall->outsSize * sizeof(u4);
+            if (bottom < self->interpStackEnd) {
+                /* stack overflow */
+                LOGV("Stack overflow on method call (start=%p end=%p newBot=%p size=%d '%s')\n",
+                    self->interpStackStart, self->interpStackEnd, bottom,
+                    self->interpStackSize, methodToCall->name);
+                dvmHandleStackOverflow(self);
+                assert(dvmCheckException(self));
+                GOTO_exceptionThrown();
+            }
+            //LOGD("+++ fp=%p newFp=%p newSave=%p bottom=%p\n",
+            //    fp, newFp, newSaveArea, bottom);
+        }
+
+#ifdef LOG_INSTR
+        if (methodToCall->registersSize > methodToCall->insSize) {
+            /*
+             * This makes valgrind quiet when we print registers that
+             * haven't been initialized.  Turn it off when the debug
+             * messages are disabled -- we want valgrind to report any
+             * used-before-initialized issues.
+             */
+            memset(newFp, 0xcc,
+                (methodToCall->registersSize - methodToCall->insSize) * 4);
+        }
+#endif
+
+#ifdef EASY_GDB
+        newSaveArea->prevSave = SAVEAREA_FROM_FP(fp);
+#endif
+        newSaveArea->prevFrame = fp;
+        newSaveArea->savedPc = pc;
+        newSaveArea->method = methodToCall;
+
+        if (!dvmIsNativeMethod(methodToCall)) {
+            /*
+             * "Call" interpreted code.  Reposition the PC, update the
+             * frame pointer and other local state, and continue.
+             */
+            curMethod = methodToCall;
+            methodClassDex = curMethod->clazz->pDvmDex;
+            pc = methodToCall->insns;
+            fp = self->curFrame = newFp;
+#ifdef EASY_GDB
+            debugSaveArea = SAVEAREA_FROM_FP(newFp);
+#endif
+#if INTERP_TYPE == INTERP_DBG
+            debugIsMethodEntry = true;              // profiling, debugging
+#endif
+            ILOGD("> pc <-- %s.%s %s", curMethod->clazz->descriptor,
+                curMethod->name, curMethod->signature);
+            DUMP_REGS(curMethod, fp, true);         // show input args
+            FINISH(0);                              // jump to method start
+        } else {
+            /* set this up for JNI locals, even if not a JNI native */
+            newSaveArea->xtra.localRefTop = self->jniLocalRefTable.nextEntry;
+
+            self->curFrame = newFp;
+
+            DUMP_REGS(methodToCall, newFp, true);   // show input args
+
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_DEBUGGER)
+            if (gDvm.debuggerActive) {
+                dvmDbgPostLocationEvent(methodToCall, -1,
+                    dvmGetThisPtr(curMethod, fp), DBG_METHOD_ENTRY);
+            }
+#endif
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_PROFILER)
+            TRACE_METHOD_ENTER(self, methodToCall);
+#endif
+
+            ILOGD("> native <-- %s.%s %s", methodToCall->clazz->descriptor,
+                methodToCall->name, methodToCall->signature);
+
+            /*
+             * Jump through native call bridge.  Because we leave no
+             * space for locals on native calls, "newFp" points directly
+             * to the method arguments.
+             */
+            (*methodToCall->nativeFunc)(newFp, &retval, methodToCall, self);
+
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_DEBUGGER)
+            if (gDvm.debuggerActive) {
+                dvmDbgPostLocationEvent(methodToCall, -1,
+                    dvmGetThisPtr(curMethod, fp), DBG_METHOD_EXIT);
+            }
+#endif
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_PROFILER)
+            TRACE_METHOD_EXIT(self, methodToCall);
+#endif
+
+            /* pop frame off */
+            dvmPopJniLocals(self, newSaveArea);
+            self->curFrame = fp;
+
+            /*
+             * If the native code threw an exception, or interpreted code
+             * invoked by the native call threw one and nobody has cleared
+             * it, jump to our local exception handling.
+             */
+            if (dvmCheckException(self)) {
+                LOGV("Exception thrown by/below native code\n");
+                GOTO_exceptionThrown();
+            }
+
+            ILOGD("> retval=0x%llx (leaving native)", retval.j);
+            ILOGD("> (return from native %s.%s to %s.%s %s)",
+                methodToCall->clazz->descriptor, methodToCall->name,
+                curMethod->clazz->descriptor, curMethod->name,
+                curMethod->signature);
+
+            //u2 invokeInstr = INST_INST(FETCH(0));
+            if (true /*invokeInstr >= OP_INVOKE_VIRTUAL &&
+                invokeInstr <= OP_INVOKE_INTERFACE*/)
+            {
+                FINISH(3);
+            } else {
+                //LOGE("Unknown invoke instr %02x at %d\n",
+                //    invokeInstr, (int) (pc - curMethod->insns));
+                assert(false);
+            }
+        }
+    }
+    assert(false);      // should not get here
+GOTO_TARGET_END
+
+
+/* File: cstubs/enddefs.c */
+
+/* undefine "magic" name remapping */
+#undef retval
+#undef pc
+#undef fp
+#undef curMethod
+#undef methodClassDex
+#undef self
+#undef debugTrackedRefStart
+