Initial Contribution
diff --git a/vm/mterp/Makefile-mterp b/vm/mterp/Makefile-mterp
new file mode 100644
index 0000000..151d5c8
--- /dev/null
+++ b/vm/mterp/Makefile-mterp
@@ -0,0 +1,50 @@
+# 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.
+
+#
+# Makefile for the Dalvik modular interpreter.  This is not currently
+# integrated into the build system.
+#
+
+SHELL := /bin/sh
+
+# Build system has TARGET_ARCH=arm, but we need the exact architecture.
+# Sooner is armv5, Dream is armv6; someday we may have x86.  The simulator
+# is "desktop".
+#
+# To generate sources for all targets:
+# for arch in desktop armv5; do TARGET_ARCH_EXT=$arch make -f Makefile-mterp; done
+#
+#TARGET_ARCH_EXT := armv5
+
+OUTPUT_DIR := out
+
+# Accumulate all possible dependencies for the generated files in a very
+# conservative fashion.  If it's not one of the generated files in "out",
+# assume it's a dependency.
+SOURCE_DEPS := \
+	$(shell find . -path ./$(OUTPUT_DIR) -prune -o -type f -print)
+
+# Source files generated by the script.  There's always one C and one
+# assembly file, though in practice one or the other could be empty.
+GEN_SOURCES := \
+	$(OUTPUT_DIR)/InterpC-$(TARGET_ARCH_EXT).c \
+	$(OUTPUT_DIR)/InterpAsm-$(TARGET_ARCH_EXT).S
+
+target: $(GEN_SOURCES)
+
+$(GEN_SOURCES): $(SOURCE_DEPS)
+	@mkdir -p out
+	./gen-mterp.py $(TARGET_ARCH_EXT) $(OUTPUT_DIR)
+
diff --git a/vm/mterp/Mterp.c b/vm/mterp/Mterp.c
new file mode 100644
index 0000000..6ef00f1
--- /dev/null
+++ b/vm/mterp/Mterp.c
@@ -0,0 +1,111 @@
+/*
+ * 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.
+ */
+/*
+ * Mterp entry point and support functions.
+ */
+#include "mterp/Mterp.h"
+
+#include <stddef.h>
+
+
+/*
+ * Verify some constants used by the mterp interpreter.
+ */
+bool dvmCheckAsmConstants(void)
+{
+    extern char dvmAsmInstructionStart[];
+    extern char dvmAsmInstructionEnd[];
+    extern char dvmAsmSisterStart[];
+    extern char dvmAsmSisterEnd[];
+
+    bool failed = false;
+
+#define ASM_DEF_VERIFY
+#include "mterp/common/asm-constants.h"
+
+    if (failed) {
+        LOGE("Please correct the values in mterp/common/asm-constants.h\n");
+        dvmAbort();
+    }
+
+    /*
+     * If an instruction overflows the 64-byte handler size limit, it will
+     * push everything up and alter the total size.  Check it here.
+     */
+    const int width = 64;
+    int interpSize = dvmAsmInstructionEnd - dvmAsmInstructionStart;
+    if (interpSize != 0 && interpSize != 256*width) {
+        LOGE("ERROR: unexpected asm interp size %d\n", interpSize);
+        LOGE("(did an instruction handler exceed %d bytes?)\n", width);
+        dvmAbort();
+    }
+    int sisterSize = dvmAsmSisterEnd - dvmAsmSisterStart;
+    LOGV("mterp: interp is %d bytes, sisters are %d bytes\n",
+        interpSize, sisterSize);
+
+    return !failed;
+}
+
+
+/*
+ * "Standard" mterp entry point.  This sets up a "glue" structure and then
+ * calls into the assembly interpreter implementation.
+ */
+bool dvmMterpStd(Thread* self, InterpState* glue)
+{
+    int changeInterp;
+
+    /* configure mterp items */
+    glue->self = self;
+    glue->methodClassDex = glue->method->clazz->pDvmDex;
+
+    glue->interpStackEnd = self->interpStackEnd;
+    glue->pSelfSuspendCount = &self->suspendCount;
+#if defined(WITH_DEBUGGER)
+    glue->pDebuggerActive = &gDvm.debuggerActive;
+#endif
+#if defined(WITH_PROFILER)
+    glue->pActiveProfilers = &gDvm.activeProfilers;
+#endif
+
+    IF_LOGVV() {
+        char* desc = dexProtoCopyMethodDescriptor(&glue->method->prototype);
+        LOGVV("mterp threadid=%d entry %d: %s.%s %s\n",
+            dvmThreadSelf()->threadId,
+            glue->entryPoint,
+            glue->method->clazz->descriptor,
+            glue->method->name,
+            desc);
+        free(desc);
+    }
+    //LOGI("glue is %p, pc=%p, fp=%p\n", glue, glue->pc, glue->fp);
+    //LOGI("first instruction is 0x%04x\n", glue->pc[0]);
+
+    changeInterp = dvmMterpStdRun(glue);
+    if (!changeInterp) {
+        /* this is a "normal" exit; we're not coming back */
+#ifdef LOG_INSTR
+        LOGD("|-- Leaving interpreter loop");
+#endif
+        return false;
+    } else {
+        /* we're "standard", so switch to "debug" */
+        LOGVV("  mterp returned, changeInterp=%d\n", changeInterp);
+        glue->nextMode = INTERP_DBG;
+        return true;
+    }
+}
+
diff --git a/vm/mterp/Mterp.h b/vm/mterp/Mterp.h
new file mode 100644
index 0000000..65e831f
--- /dev/null
+++ b/vm/mterp/Mterp.h
@@ -0,0 +1,48 @@
+/*
+ * 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.
+ */
+/*
+ * Some declarations used throughout mterp.
+ */
+#ifndef _DALVIK_MTERP_MTERP
+#define _DALVIK_MTERP_MTERP
+
+#include "Dalvik.h"
+#include "interp/InterpDefs.h"
+
+/*
+ * Interpreter state, passed into C functions from assembly stubs.  The
+ * assembly code exports all registers into the "glue" structure before
+ * calling, then extracts them when the call returns.
+ */
+typedef InterpState MterpGlue;
+
+/*
+ * Call this during initialization to verify that the values in asm-constants.h
+ * are still correct.
+ */
+bool dvmCheckAsmConstants(void);
+
+/*
+ * Local entry and exit points.  All implementations must provide these two.
+ *
+ * dvmMterpStdRun() returns the "changeInterp" argument from dvmMterpStdBail(),
+ * indicating whether we want to bail out of the interpreter or just switch
+ * between "standard" and "debug" mode.
+ */
+bool dvmMterpStdRun(MterpGlue* glue);
+void dvmMterpStdBail(MterpGlue* glue, bool changeInterp);
+
+#endif /*_DALVIK_MTERP_MTERP*/
diff --git a/vm/mterp/README.txt b/vm/mterp/README.txt
new file mode 100644
index 0000000..34d229c
--- /dev/null
+++ b/vm/mterp/README.txt
@@ -0,0 +1,188 @@
+Dalvik "mterp" README
+
+NOTE: Find rebuilding instructions at the bottom of this file.
+
+
+==== Overview ====
+
+The architecture-specific config files determine what goes into two
+generated output files (InterpC-<arch>.c, InterpAsm-<arch>.S).  The goal is
+to make it easy to swap C and assembly sources during initial development
+and testing, and to provide a way to use architecture-specific versions
+of some operations (e.g. making use of PLD instructions on ARMv6).
+
+Two basic assumptions are made about the operation of the interpreter:
+
+ - The assembly version uses fixed-size areas for each instruction
+   (e.g. 64 bytes).  "Overflow" code is tacked on to the end.
+ - When a C implementation is desired, the assembly version packs all
+   local state into a "glue" struct, and passes that into the C function.
+   Updates to the state are pulled out of the "glue" on return.
+
+
+The "arch" value should indicate an architecture family with common
+programming characteristics, so "armv5" would work for all ARMv5 CPUs,
+but might not be backward- or forward-compatible.  (We *might* want to
+specify the ABI model as well, e.g. "armv5-eabi", but currently that adds
+verbosity without value.)
+
+
+==== Config file format ====
+
+The config files are parsed from top to bottom.  Each line in the file
+may be blank, hold a comment (line starts with '#'), or be a command.
+
+The commands are:
+
+  handler-size <bytes>
+
+    Specify the size of the assembly region, in bytes.  On most platforms
+    this will need to be a power of 2.
+
+  import <filename>
+
+    The specified file is included immediately, in its entirety.  No
+    substitutions are performed.  ".c" and ".h" files are copied to the
+    C output, ".S" files are copied to the asm output.
+
+  asm-stub <filename>
+
+    The named file will be included whenever an assembly "stub" is needed.
+    Text substitution is performed on the opcode name.
+
+  op-start <directory>
+
+    Indicates the start of the opcode list.  Must precede any "op"
+    commands.  The specified directory is the default location to pull
+    instruction files from.
+
+  op <opcode> <directory>
+
+    Can only appear after "op-start" and before "op-end".  Overrides the
+    default source file location of the specified opcode.  The opcode
+    definition will come from the specified file, e.g. "op OP_NOP armv5"
+    will load from "armv5/OP_NOP.S".  A substitution dictionary will be
+    applied (see below).
+
+  op-end
+
+    Indicates the end of the opcode list.  All 256 opcodes are emitted
+    when this is seen, followed by any code that didn't fit inside the
+    fixed-size instruction handler space.
+
+
+The order of "op" directives is not significant; the generation tool will
+extract ordering info from the VM sources.
+
+Typically the form in which most opcodes currently exist is used in
+the "op-start" directive.  For a new port you would start with "c",
+and add architecture-specific "op" entries as you write instructions.
+When complete it will default to the target architecture, and you insert
+"c" ops to stub out platform-specific code.
+
+For the <directory> specified in the "op" command, the "c" directory
+is special in two ways: (1) the sources are assumed to be C code, and
+will be inserted into the generated C file; (2) when a C implementation
+is emitted, a "glue stub" is emitted in the assembly source file.
+(The generator script always emits 256 assembly instructions, unless
+"asm-stub" was left blank, in which case it only emits some labels.)
+
+
+==== Instruction file format ====
+
+The assembly instruction files are simply fragments of assembly sources.
+The starting label will be provided by the generation tool, as will
+declarations for the segment type and alignment.  The expected target
+assembler is GNU "as", but others will work (may require fiddling with
+some of the pseudo-ops emitted by the generation tool).
+
+The C files do a bunch of fancy things with macros in an attempt to keep
+the code "cut & pastable" from the portable interpreter.  (This will be
+reduced when the code bases are merged.)
+
+A substitution dictionary is applied to all opcode fragments as they are
+appended to the output.  Substitutions can look like "$value" or "${value}".
+
+The dictionary always includes:
+
+  $opcode - opcode name, e.g. "OP_NOP"
+  $opnum - opcode number, e.g. 0 for OP_NOP
+  $handler_size_bytes - max size of an instruction handler, in bytes
+  $handler_size_bits - max size of an instruction handler, log 2
+
+Both C and assembly sources will be passed through the C pre-processor,
+so you can take advantage of C-style comments and preprocessor directives
+like "#define".
+
+Some generator operations are available.
+
+  %include "filename" [subst-dict]
+
+    Includes the file, which should look like "armv5/OP_NOP.S".  You can
+    specify values for the substitution dictionary, using standard Python
+    syntax.  For example, this:
+      %include "armv5/unop.S" {"result":"r1"}
+    would insert "armv5/unop.S" at the current file position, replacing
+    occurrences of "$result" with "r1".
+
+  %default <subst-dict>
+
+    Specify default substitution dictionary values, using standard Python
+    syntax.  Useful if you want to have a "base" version and variants.
+
+  %break
+
+    Identifies the split between the main portion of the instruction
+    handler (which must fit in "handler-size" bytes) and the "sister"
+    code, which is appended to the end of the instruction handler block.
+
+  %verify "message"
+
+    Leave a note to yourself about what needs to be tested.  (This may
+    turn into something more interesting someday; for now, it just gets
+    stripped out before the output is generated.)
+
+The generation tool does *not* print a warning if your instructions
+exceed "handler-size", but the VM will abort on startup if it detects an
+oversized handler.  On architectures with fixed-width instructions this
+is easy to work with, on others this you will need to count bytes.
+
+
+==== Sharing constants ====
+
+The file "common/asm-constants.h" has some definitions for constant
+values, structure sizes, and struct member offsets.  The format is fairly
+restricted, as simple macros are used to massage it for use with both C
+(where it is verified) and assembly (where the definitions are used).
+
+If a constant in the file becomes out of sync, the VM will log an error
+message and abort during startup.
+
+
+==== Tips ====
+
+If you need to debug the initial piece of an opcode handler, and your
+debug code expands it beyond the handler size limit, you can insert a
+generic header at the top:
+
+    b       ${opcode}_start
+%break
+${opcode}_start:
+
+If you already have a %break, it's okay to leave it in place -- the second
+%break is ignored.
+
+
+==== Rebuilding ====
+
+If you change any of the source file fragments, you need to rebuild the
+combined source files in the "out" directory.  Make sure the files in
+"out" are editable, then:
+
+    $ cd mterp
+    $ ./rebuild.sh
+
+As of this writing, this requires Python 2.5. You may see inscrutible
+error messages or just general failure if you have a different version
+of Python installed.
+
diff --git a/vm/mterp/armv5/OP_ADD_DOUBLE.S b/vm/mterp/armv5/OP_ADD_DOUBLE.S
new file mode 100644
index 0000000..e1ed0d5
--- /dev/null
+++ b/vm/mterp/armv5/OP_ADD_DOUBLE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide.S" {"instr":"bl      __aeabi_dadd"}
diff --git a/vm/mterp/armv5/OP_ADD_DOUBLE_2ADDR.S b/vm/mterp/armv5/OP_ADD_DOUBLE_2ADDR.S
new file mode 100644
index 0000000..299ca37
--- /dev/null
+++ b/vm/mterp/armv5/OP_ADD_DOUBLE_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide2addr.S" {"instr":"bl      __aeabi_dadd"}
diff --git a/vm/mterp/armv5/OP_ADD_FLOAT.S b/vm/mterp/armv5/OP_ADD_FLOAT.S
new file mode 100644
index 0000000..318b733
--- /dev/null
+++ b/vm/mterp/armv5/OP_ADD_FLOAT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop.S" {"instr":"bl      __aeabi_fadd"}
diff --git a/vm/mterp/armv5/OP_ADD_FLOAT_2ADDR.S b/vm/mterp/armv5/OP_ADD_FLOAT_2ADDR.S
new file mode 100644
index 0000000..0195fd4
--- /dev/null
+++ b/vm/mterp/armv5/OP_ADD_FLOAT_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop2addr.S" {"instr":"bl      __aeabi_fadd"}
diff --git a/vm/mterp/armv5/OP_ADD_INT.S b/vm/mterp/armv5/OP_ADD_INT.S
new file mode 100644
index 0000000..b69577c
--- /dev/null
+++ b/vm/mterp/armv5/OP_ADD_INT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop.S" {"instr":"add     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_ADD_INT_2ADDR.S b/vm/mterp/armv5/OP_ADD_INT_2ADDR.S
new file mode 100644
index 0000000..0d54f91
--- /dev/null
+++ b/vm/mterp/armv5/OP_ADD_INT_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop2addr.S" {"instr":"add     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_ADD_INT_LIT16.S b/vm/mterp/armv5/OP_ADD_INT_LIT16.S
new file mode 100644
index 0000000..2710041
--- /dev/null
+++ b/vm/mterp/armv5/OP_ADD_INT_LIT16.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopLit16.S" {"instr":"add     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_ADD_INT_LIT8.S b/vm/mterp/armv5/OP_ADD_INT_LIT8.S
new file mode 100644
index 0000000..074fb90
--- /dev/null
+++ b/vm/mterp/armv5/OP_ADD_INT_LIT8.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopLit8.S" {"instr":"add     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_ADD_LONG.S b/vm/mterp/armv5/OP_ADD_LONG.S
new file mode 100644
index 0000000..b30ff53
--- /dev/null
+++ b/vm/mterp/armv5/OP_ADD_LONG.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide.S" {"preinstr":"adds    r0, r0, r2", "instr":"adc     r1, r1, r3"}
diff --git a/vm/mterp/armv5/OP_ADD_LONG_2ADDR.S b/vm/mterp/armv5/OP_ADD_LONG_2ADDR.S
new file mode 100644
index 0000000..7edc50b
--- /dev/null
+++ b/vm/mterp/armv5/OP_ADD_LONG_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide2addr.S" {"preinstr":"adds    r0, r0, r2", "instr":"adc     r1, r1, r3"}
diff --git a/vm/mterp/armv5/OP_AGET.S b/vm/mterp/armv5/OP_AGET.S
new file mode 100644
index 0000000..7a0950b
--- /dev/null
+++ b/vm/mterp/armv5/OP_AGET.S
@@ -0,0 +1,28 @@
+%default { "load":"ldr", "shift":"2" }
+%verify "executed"
+    /*
+     * 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 #$shift     @ 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
+    $load   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
+
diff --git a/vm/mterp/armv5/OP_AGET_BOOLEAN.S b/vm/mterp/armv5/OP_AGET_BOOLEAN.S
new file mode 100644
index 0000000..7745821
--- /dev/null
+++ b/vm/mterp/armv5/OP_AGET_BOOLEAN.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_AGET.S" { "load":"ldrb", "shift":"0" }
diff --git a/vm/mterp/armv5/OP_AGET_BYTE.S b/vm/mterp/armv5/OP_AGET_BYTE.S
new file mode 100644
index 0000000..cf7d01b
--- /dev/null
+++ b/vm/mterp/armv5/OP_AGET_BYTE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_AGET.S" { "load":"ldrsb", "shift":"0" }
diff --git a/vm/mterp/armv5/OP_AGET_CHAR.S b/vm/mterp/armv5/OP_AGET_CHAR.S
new file mode 100644
index 0000000..1a485b2
--- /dev/null
+++ b/vm/mterp/armv5/OP_AGET_CHAR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_AGET.S" { "load":"ldrh", "shift":"1" }
diff --git a/vm/mterp/armv5/OP_AGET_OBJECT.S b/vm/mterp/armv5/OP_AGET_OBJECT.S
new file mode 100644
index 0000000..200b7b2
--- /dev/null
+++ b/vm/mterp/armv5/OP_AGET_OBJECT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_AGET.S"
diff --git a/vm/mterp/armv5/OP_AGET_SHORT.S b/vm/mterp/armv5/OP_AGET_SHORT.S
new file mode 100644
index 0000000..07fbe94
--- /dev/null
+++ b/vm/mterp/armv5/OP_AGET_SHORT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_AGET.S" { "load":"ldrsh", "shift":"1" }
diff --git a/vm/mterp/armv5/OP_AGET_WIDE.S b/vm/mterp/armv5/OP_AGET_WIDE.S
new file mode 100644
index 0000000..ec346ca
--- /dev/null
+++ b/vm/mterp/armv5/OP_AGET_WIDE.S
@@ -0,0 +1,33 @@
+%verify "executed"
+    /*
+     * 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     .L${opcode}_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.
+%break
+
+.L${opcode}_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
+
diff --git a/vm/mterp/armv5/OP_AND_INT.S b/vm/mterp/armv5/OP_AND_INT.S
new file mode 100644
index 0000000..2994661
--- /dev/null
+++ b/vm/mterp/armv5/OP_AND_INT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop.S" {"instr":"and     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_AND_INT_2ADDR.S b/vm/mterp/armv5/OP_AND_INT_2ADDR.S
new file mode 100644
index 0000000..dab5bf9
--- /dev/null
+++ b/vm/mterp/armv5/OP_AND_INT_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop2addr.S" {"instr":"and     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_AND_INT_LIT16.S b/vm/mterp/armv5/OP_AND_INT_LIT16.S
new file mode 100644
index 0000000..9cc90dd
--- /dev/null
+++ b/vm/mterp/armv5/OP_AND_INT_LIT16.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopLit16.S" {"instr":"and     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_AND_INT_LIT8.S b/vm/mterp/armv5/OP_AND_INT_LIT8.S
new file mode 100644
index 0000000..c1db66d
--- /dev/null
+++ b/vm/mterp/armv5/OP_AND_INT_LIT8.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopLit8.S" {"instr":"and     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_AND_LONG.S b/vm/mterp/armv5/OP_AND_LONG.S
new file mode 100644
index 0000000..8fbac54
--- /dev/null
+++ b/vm/mterp/armv5/OP_AND_LONG.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide.S" {"preinstr":"and     r0, r0, r2", "instr":"and     r1, r1, r3"}
diff --git a/vm/mterp/armv5/OP_AND_LONG_2ADDR.S b/vm/mterp/armv5/OP_AND_LONG_2ADDR.S
new file mode 100644
index 0000000..61b3615
--- /dev/null
+++ b/vm/mterp/armv5/OP_AND_LONG_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide2addr.S" {"preinstr":"and     r0, r0, r2", "instr":"and     r1, r1, r3"}
diff --git a/vm/mterp/armv5/OP_APUT.S b/vm/mterp/armv5/OP_APUT.S
new file mode 100644
index 0000000..f8ee4a8
--- /dev/null
+++ b/vm/mterp/armv5/OP_APUT.S
@@ -0,0 +1,28 @@
+%default { "store":"str", "shift":"2" }
+%verify "executed"
+    /*
+     * 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 #$shift     @ 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
+    $store  r2, [r0, #offArrayObject_contents]  @ vBB[vCC]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
diff --git a/vm/mterp/armv5/OP_APUT_BOOLEAN.S b/vm/mterp/armv5/OP_APUT_BOOLEAN.S
new file mode 100644
index 0000000..3e0f32a
--- /dev/null
+++ b/vm/mterp/armv5/OP_APUT_BOOLEAN.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_APUT.S" { "store":"strb", "shift":"0" }
diff --git a/vm/mterp/armv5/OP_APUT_BYTE.S b/vm/mterp/armv5/OP_APUT_BYTE.S
new file mode 100644
index 0000000..3e0f32a
--- /dev/null
+++ b/vm/mterp/armv5/OP_APUT_BYTE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_APUT.S" { "store":"strb", "shift":"0" }
diff --git a/vm/mterp/armv5/OP_APUT_CHAR.S b/vm/mterp/armv5/OP_APUT_CHAR.S
new file mode 100644
index 0000000..038bc47
--- /dev/null
+++ b/vm/mterp/armv5/OP_APUT_CHAR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_APUT.S" { "store":"strh", "shift":"1" }
diff --git a/vm/mterp/armv5/OP_APUT_OBJECT.S b/vm/mterp/armv5/OP_APUT_OBJECT.S
new file mode 100644
index 0000000..53a188a
--- /dev/null
+++ b/vm/mterp/armv5/OP_APUT_OBJECT.S
@@ -0,0 +1,46 @@
+%verify "executed"
+    /*
+     * 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.
+     *
+     * for: aput, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* 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     .L${opcode}_finish          @ we're okay, continue on
+    b       common_errArrayIndex        @ index >= length, bail
+
+%break
+    /*
+     * On entry:
+     *  r1 = vBB (arrayObj)
+     *  r9 = vAA (obj)
+     *  r10 = offset into array (vBB + vCC * width)
+     */
+.L${opcode}_finish:
+    cmp     r9, #0                      @ storing null reference?
+    beq     .L${opcode}_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
+.L${opcode}_skip_check:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r9, [r10, #offArrayObject_contents] @ vBB[vCC]<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
diff --git a/vm/mterp/armv5/OP_APUT_SHORT.S b/vm/mterp/armv5/OP_APUT_SHORT.S
new file mode 100644
index 0000000..038bc47
--- /dev/null
+++ b/vm/mterp/armv5/OP_APUT_SHORT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_APUT.S" { "store":"strh", "shift":"1" }
diff --git a/vm/mterp/armv5/OP_APUT_WIDE.S b/vm/mterp/armv5/OP_APUT_WIDE.S
new file mode 100644
index 0000000..48738cc
--- /dev/null
+++ b/vm/mterp/armv5/OP_APUT_WIDE.S
@@ -0,0 +1,33 @@
+%verify "executed"
+    /*
+     * Array put, 64 bits.  vBB[vCC] <- vAA.
+     *
+     * Arrays of long/double are 64-bit aligned, so it's okay to use STRD.
+     */
+    /* aput-wide vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #3          @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    bcc     .L${opcode}_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.
+%break
+
+.L${opcode}_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
+
diff --git a/vm/mterp/armv5/OP_ARRAY_LENGTH.S b/vm/mterp/armv5/OP_ARRAY_LENGTH.S
new file mode 100644
index 0000000..5dc0f93
--- /dev/null
+++ b/vm/mterp/armv5/OP_ARRAY_LENGTH.S
@@ -0,0 +1,16 @@
+%verify "executed"
+    /*
+     * 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
+
diff --git a/vm/mterp/armv5/OP_CHECK_CAST.S b/vm/mterp/armv5/OP_CHECK_CAST.S
new file mode 100644
index 0000000..74f458b
--- /dev/null
+++ b/vm/mterp/armv5/OP_CHECK_CAST.S
@@ -0,0 +1,73 @@
+%verify "executed"
+%verify "null object"
+%verify "class cast exception thrown, with correct class name"
+%verify "class cast exception not thrown on same class"
+%verify "class cast exception not thrown on subclass"
+%verify "class not resolved"
+%verify "class already resolved"
+    /*
+     * 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     .L${opcode}_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     .L${opcode}_resolve         @ not resolved, do it now
+.L${opcode}_resolved:
+    cmp     r0, r1                      @ same class (trivial success)?
+    bne     .L${opcode}_fullcheck       @ no, do full check
+.L${opcode}_okay:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+%break
+
+    /*
+     * Trivial test failed, need to perform full check.  This is common.
+     *  r0 holds obj->clazz
+     *  r1 holds class resolved from BBBB
+     *  r9 holds object
+     */
+.L${opcode}_fullcheck:
+    bl      dvmInstanceofNonTrivial     @ r0<- boolean result
+    cmp     r0, #0                      @ failed?
+    bne     .L${opcode}_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
+     */
+.L${opcode}_resolve:
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    mov     r1, r2                      @ r1<- BBBB
+    mov     r2, #0                      @ r2<- false
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- resolved ClassObject ptr
+    cmp     r0, #0                      @ got null?
+    beq     common_exceptionThrown      @ yes, handle exception
+    mov     r1, r0                      @ r1<- class resolved from BBB
+    ldr     r0, [r9, #offObject_clazz]  @ r0<- obj->clazz
+    b       .L${opcode}_resolved        @ pick up where we left off
+
+.LstrClassCastExceptionPtr:
+    .word   .LstrClassCastException
+
diff --git a/vm/mterp/armv5/OP_CMPG_DOUBLE.S b/vm/mterp/armv5/OP_CMPG_DOUBLE.S
new file mode 100644
index 0000000..706dced
--- /dev/null
+++ b/vm/mterp/armv5/OP_CMPG_DOUBLE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_CMPL_DOUBLE.S" { "naninst":"mov     r1, #1" }
diff --git a/vm/mterp/armv5/OP_CMPG_FLOAT.S b/vm/mterp/armv5/OP_CMPG_FLOAT.S
new file mode 100644
index 0000000..944fff8
--- /dev/null
+++ b/vm/mterp/armv5/OP_CMPG_FLOAT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_CMPL_FLOAT.S" { "naninst":"mov     r1, #1" }
diff --git a/vm/mterp/armv5/OP_CMPL_DOUBLE.S b/vm/mterp/armv5/OP_CMPL_DOUBLE.S
new file mode 100644
index 0000000..50ff3f7
--- /dev/null
+++ b/vm/mterp/armv5/OP_CMPL_DOUBLE.S
@@ -0,0 +1,49 @@
+%default { "naninst":"mvn     r1, #0" }
+%verify "executed"
+%verify "basic lt, gt, eq */
+%verify "left arg NaN"
+%verify "right arg NaN"
+    /*
+     * 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     .L${opcode}_gt_or_nan       @ C set and Z clear, disambiguate
+    mvncc   r1, #0                      @ (less than) r1<- -1
+    moveq   r1, #0                      @ (equal) r1<- 0, trumps less than
+.L${opcode}_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
+%break
+
+    @ 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.
+.L${opcode}_gt_or_nan:
+    ldmia   r10, {r0-r1}                @ reverse order
+    ldmia   r9, {r2-r3}
+    bl      __aeabi_cdcmple             @ r0<- Z set if eq, C clear if <
+    @bleq    common_abort
+    movcc   r1, #1                      @ (greater than) r1<- 1
+    bcc     .L${opcode}_finish
+    $naninst                            @ r1<- 1 or -1 for NaN
+    b       .L${opcode}_finish
+
diff --git a/vm/mterp/armv5/OP_CMPL_FLOAT.S b/vm/mterp/armv5/OP_CMPL_FLOAT.S
new file mode 100644
index 0000000..c53d419
--- /dev/null
+++ b/vm/mterp/armv5/OP_CMPL_FLOAT.S
@@ -0,0 +1,116 @@
+%default { "naninst":"mvn     r1, #0" }
+%verify "executed"
+%verify "basic lt, gt, eq */
+%verify "left arg NaN"
+%verify "right arg NaN"
+    /*
+     * 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     .L${opcode}_gt_or_nan       @ C set and Z clear, disambiguate
+    mvncc   r1, #0                      @ (less than) r1<- -1
+    moveq   r1, #0                      @ (equal) r1<- 0, trumps less than
+.L${opcode}_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
+%break
+
+    @ 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.
+.L${opcode}_gt_or_nan:
+    mov     r1, r9                      @ reverse order
+    mov     r0, r10
+    bl      __aeabi_cfcmple             @ r0<- Z set if eq, C clear if <
+    @bleq    common_abort
+    movcc   r1, #1                      @ (greater than) r1<- 1
+    bcc     .L${opcode}_finish
+    $naninst                            @ r1<- 1 or -1 for NaN
+    b       .L${opcode}_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     ${opcode}_finish
+    mov     r0, r9                      @ r0<- vBB
+    mov     r1, r10                     @ r1<- vCC
+    bl      __aeabi_fcmplt              @ r0<- (vBB < vCC)
+    cmp     r0, #0                      @ less than?
+    b       ${opcode}_continue
+@%break
+
+${opcode}_continue:
+    mvnne   r1, #0                      @ yes, result is -1
+    bne     ${opcode}_finish
+    mov     r0, r9                      @ r0<- vBB
+    mov     r1, r10                     @ r1<- vCC
+    bl      __aeabi_fcmpgt              @ r0<- (vBB > vCC)
+    cmp     r0, #0                      @ greater than?
+    beq     ${opcode}_nan               @ no, must be NaN
+    mov     r1, #1                      @ yes, result is 1
+    @ fall through to _finish
+
+${opcode}_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).
+     */
+${opcode}_nan:
+    $naninst                            @ r1<- 1 or -1 for NaN
+    b       ${opcode}_finish
+
+#endif
+
diff --git a/vm/mterp/armv5/OP_CMP_LONG.S b/vm/mterp/armv5/OP_CMP_LONG.S
new file mode 100644
index 0000000..d456137
--- /dev/null
+++ b/vm/mterp/armv5/OP_CMP_LONG.S
@@ -0,0 +1,61 @@
+%verify "executed"
+%verify "basic lt, gt, eq"
+%verify "hi equal, lo <=>"
+%verify "lo equal, hi <=>"
+    /*
+     * Compare two 64-bit values.  Puts 0, 1, or -1 into the destination
+     * register based on the results of the comparison.
+     *
+     * We load the full values with LDM, but in practice many values could
+     * be resolved by only looking at the high word.  This could be made
+     * faster or slower by splitting the LDM into a pair of LDRs.
+     *
+     * If we just wanted to set condition flags, we could do this:
+     *  subs    ip, r0, r2
+     *  sbcs    ip, r1, r3
+     *  subeqs  ip, r0, r2
+     * Leaving { <0, 0, >0 } in ip.  However, we have to set it to a specific
+     * integer value, which we can do with 2 conditional mov/mvn instructions
+     * (set 1, set -1; if they're equal we already have 0 in ip), giving
+     * us a constant 5-cycle path plus a branch at the end to the
+     * instruction epilogue code.  The multi-compare approach below needs
+     * 2 or 3 cycles + branch if the high word doesn't match, 6 + branch
+     * in the worst case (the 64-bit values are equal).
+     */
+    /* cmp-long vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    cmp     r1, r3                      @ compare (vBB+1, vCC+1)
+    blt     .L${opcode}_less            @ signed compare on high part
+    bgt     .L${opcode}_greater
+    subs    r1, r0, r2                  @ r1<- r0 - r2
+    bhi     .L${opcode}_greater         @ unsigned compare on low part
+    bne     .L${opcode}_less
+    b       .L${opcode}_finish          @ equal; r1 already holds 0
+%break
+
+.L${opcode}_less:
+    mvn     r1, #0                      @ r1<- -1
+    @ Want to cond code the next mov so we can avoid branch, but don't see it;
+    @ instead, we just replicate the tail end.
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r9)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+.L${opcode}_greater:
+    mov     r1, #1                      @ r1<- 1
+    @ fall through to _finish
+
+.L${opcode}_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
+
diff --git a/vm/mterp/armv5/OP_CONST.S b/vm/mterp/armv5/OP_CONST.S
new file mode 100644
index 0000000..4394647
--- /dev/null
+++ b/vm/mterp/armv5/OP_CONST.S
@@ -0,0 +1,11 @@
+%verify "executed"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_CONST_16.S b/vm/mterp/armv5/OP_CONST_16.S
new file mode 100644
index 0000000..0b44c99
--- /dev/null
+++ b/vm/mterp/armv5/OP_CONST_16.S
@@ -0,0 +1,9 @@
+%verify "executed"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_CONST_4.S b/vm/mterp/armv5/OP_CONST_4.S
new file mode 100644
index 0000000..9ac53c2
--- /dev/null
+++ b/vm/mterp/armv5/OP_CONST_4.S
@@ -0,0 +1,11 @@
+%verify "executed"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_CONST_CLASS.S b/vm/mterp/armv5/OP_CONST_CLASS.S
new file mode 100644
index 0000000..b834553
--- /dev/null
+++ b/vm/mterp/armv5/OP_CONST_CLASS.S
@@ -0,0 +1,36 @@
+%verify "executed"
+%verify "Class already resolved"
+%verify "Class not yet resolved"
+%verify "Class cannot be resolved"
+    /* 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     .L${opcode}_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
+%break
+
+    /*
+     * Continuation if the Class has not yet been resolved.
+     *  r1: BBBB (Class ref)
+     *  r9: target register
+     */
+.L${opcode}_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
+
diff --git a/vm/mterp/armv5/OP_CONST_HIGH16.S b/vm/mterp/armv5/OP_CONST_HIGH16.S
new file mode 100644
index 0000000..07b5061
--- /dev/null
+++ b/vm/mterp/armv5/OP_CONST_HIGH16.S
@@ -0,0 +1,10 @@
+%verify "executed"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_CONST_STRING.S b/vm/mterp/armv5/OP_CONST_STRING.S
new file mode 100644
index 0000000..5e2e389
--- /dev/null
+++ b/vm/mterp/armv5/OP_CONST_STRING.S
@@ -0,0 +1,35 @@
+%verify "executed"
+%verify "String already resolved"
+%verify "String not yet resolved"
+%verify "String cannot be resolved"
+    /* 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     .L${opcode}_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
+%break
+
+    /*
+     * Continuation if the String has not yet been resolved.
+     *  r1: BBBB (String ref)
+     *  r9: target register
+     */
+.L${opcode}_resolve:
+    EXPORT_PC()
+    ldr     r0, [rGLUE, #offGlue_method] @ r0<- glue->method
+    ldr     r0, [r0, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveString            @ r0<- String reference
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yup, handle the exception
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
diff --git a/vm/mterp/armv5/OP_CONST_STRING_JUMBO.S b/vm/mterp/armv5/OP_CONST_STRING_JUMBO.S
new file mode 100644
index 0000000..b98e0e1
--- /dev/null
+++ b/vm/mterp/armv5/OP_CONST_STRING_JUMBO.S
@@ -0,0 +1,37 @@
+%verify "executed"
+%verify "String already resolved"
+%verify "String not yet resolved"
+%verify "String cannot be resolved"
+    /* const/string vAA, String@BBBBBBBB */
+    FETCH(r0, 1)                        @ r0<- bbbb (low)
+    FETCH(r1, 2)                        @ r1<- BBBB (high)
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- glue->methodClassDex
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    ldr     r2, [r2, #offDvmDex_pResStrings]   @ r2<- dvmDex->pResStrings
+    orr     r1, r0, r1, lsl #16         @ r1<- BBBBbbbb
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- pResStrings[BBBB]
+    cmp     r0, #0
+    beq     .L${opcode}_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
+%break
+
+    /*
+     * Continuation if the String has not yet been resolved.
+     *  r1: BBBBBBBB (String ref)
+     *  r9: target register
+     */
+.L${opcode}_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
+
diff --git a/vm/mterp/armv5/OP_CONST_WIDE.S b/vm/mterp/armv5/OP_CONST_WIDE.S
new file mode 100644
index 0000000..428d423
--- /dev/null
+++ b/vm/mterp/armv5/OP_CONST_WIDE.S
@@ -0,0 +1,15 @@
+%verify "executed"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_CONST_WIDE_16.S b/vm/mterp/armv5/OP_CONST_WIDE_16.S
new file mode 100644
index 0000000..2d37d58
--- /dev/null
+++ b/vm/mterp/armv5/OP_CONST_WIDE_16.S
@@ -0,0 +1,11 @@
+%verify "executed"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_CONST_WIDE_32.S b/vm/mterp/armv5/OP_CONST_WIDE_32.S
new file mode 100644
index 0000000..5991eb4
--- /dev/null
+++ b/vm/mterp/armv5/OP_CONST_WIDE_32.S
@@ -0,0 +1,13 @@
+%verify "executed"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_CONST_WIDE_HIGH16.S b/vm/mterp/armv5/OP_CONST_WIDE_HIGH16.S
new file mode 100644
index 0000000..2e23f79
--- /dev/null
+++ b/vm/mterp/armv5/OP_CONST_WIDE_HIGH16.S
@@ -0,0 +1,12 @@
+%verify "executed"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_DIV_DOUBLE.S b/vm/mterp/armv5/OP_DIV_DOUBLE.S
new file mode 100644
index 0000000..938875d
--- /dev/null
+++ b/vm/mterp/armv5/OP_DIV_DOUBLE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide.S" {"instr":"bl      __aeabi_ddiv"}
diff --git a/vm/mterp/armv5/OP_DIV_DOUBLE_2ADDR.S b/vm/mterp/armv5/OP_DIV_DOUBLE_2ADDR.S
new file mode 100644
index 0000000..3b6409c
--- /dev/null
+++ b/vm/mterp/armv5/OP_DIV_DOUBLE_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide2addr.S" {"instr":"bl      __aeabi_ddiv"}
diff --git a/vm/mterp/armv5/OP_DIV_FLOAT.S b/vm/mterp/armv5/OP_DIV_FLOAT.S
new file mode 100644
index 0000000..7defc9b
--- /dev/null
+++ b/vm/mterp/armv5/OP_DIV_FLOAT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop.S" {"instr":"bl      __aeabi_fdiv"}
diff --git a/vm/mterp/armv5/OP_DIV_FLOAT_2ADDR.S b/vm/mterp/armv5/OP_DIV_FLOAT_2ADDR.S
new file mode 100644
index 0000000..786514e
--- /dev/null
+++ b/vm/mterp/armv5/OP_DIV_FLOAT_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop2addr.S" {"instr":"bl      __aeabi_fdiv"}
diff --git a/vm/mterp/armv5/OP_DIV_INT.S b/vm/mterp/armv5/OP_DIV_INT.S
new file mode 100644
index 0000000..c7e9faa
--- /dev/null
+++ b/vm/mterp/armv5/OP_DIV_INT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop.S" {"instr":"bl     __aeabi_idiv","chkzero":"1"}
diff --git a/vm/mterp/armv5/OP_DIV_INT_2ADDR.S b/vm/mterp/armv5/OP_DIV_INT_2ADDR.S
new file mode 100644
index 0000000..3173c60
--- /dev/null
+++ b/vm/mterp/armv5/OP_DIV_INT_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop2addr.S" {"instr":"bl     __aeabi_idiv","chkzero":"1"}
diff --git a/vm/mterp/armv5/OP_DIV_INT_LIT16.S b/vm/mterp/armv5/OP_DIV_INT_LIT16.S
new file mode 100644
index 0000000..ca39d67
--- /dev/null
+++ b/vm/mterp/armv5/OP_DIV_INT_LIT16.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopLit16.S" {"instr":"bl     __aeabi_idiv","chkzero":"1"}
diff --git a/vm/mterp/armv5/OP_DIV_INT_LIT8.S b/vm/mterp/armv5/OP_DIV_INT_LIT8.S
new file mode 100644
index 0000000..fa518fe
--- /dev/null
+++ b/vm/mterp/armv5/OP_DIV_INT_LIT8.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopLit8.S" {"instr":"bl     __aeabi_idiv","chkzero":"1"}
diff --git a/vm/mterp/armv5/OP_DIV_LONG.S b/vm/mterp/armv5/OP_DIV_LONG.S
new file mode 100644
index 0000000..fe4009b
--- /dev/null
+++ b/vm/mterp/armv5/OP_DIV_LONG.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide.S" {"instr":"bl      __aeabi_ldivmod", "chkzero":"1"}
diff --git a/vm/mterp/armv5/OP_DIV_LONG_2ADDR.S b/vm/mterp/armv5/OP_DIV_LONG_2ADDR.S
new file mode 100644
index 0000000..41d928d
--- /dev/null
+++ b/vm/mterp/armv5/OP_DIV_LONG_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide2addr.S" {"instr":"bl      __aeabi_ldivmod", "chkzero":"1"}
diff --git a/vm/mterp/armv5/OP_DOUBLE_TO_FLOAT.S b/vm/mterp/armv5/OP_DOUBLE_TO_FLOAT.S
new file mode 100644
index 0000000..16b8e95
--- /dev/null
+++ b/vm/mterp/armv5/OP_DOUBLE_TO_FLOAT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/unopNarrower.S" {"instr":"bl      __aeabi_d2f"}
diff --git a/vm/mterp/armv5/OP_DOUBLE_TO_INT.S b/vm/mterp/armv5/OP_DOUBLE_TO_INT.S
new file mode 100644
index 0000000..df831a9
--- /dev/null
+++ b/vm/mterp/armv5/OP_DOUBLE_TO_INT.S
@@ -0,0 +1,59 @@
+%verify "executed"
+/* EABI appears to have Java-style conversions of +inf/-inf/NaN */
+%include "armv5/unopNarrower.S" {"instr":"bl      __aeabi_d2iz"}
+
+#if 0
+@include "armv5/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, .L${opcode}_maxlo       @ (double)maxint, lo
+    ldr     r3, .L${opcode}_maxhi       @ (double)maxint, hi
+    sub     sp, sp, #4                  @ align for EABI
+    mov     r4, r0                      @ save r0
+    mov     r5, r1                      @  and r1
+    bl      __aeabi_dcmpge              @ is arg >= maxint?
+    cmp     r0, #0                      @ nonzero == yes
+    mvnne   r0, #0x80000000             @ return maxint (7fffffff)
+    bne     1f
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r5
+    ldr     r3, .L${opcode}_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}
+
+.L${opcode}_maxlo:
+    .word   0xffc00000                  @ maxint, as a double (low word)
+.L${opcode}_maxhi:
+    .word   0x41dfffff                  @ maxint, as a double (high word)
+.L${opcode}_min:
+    .word   0xc1e00000                  @ minint, as a double (high word)
+#endif
+
diff --git a/vm/mterp/armv5/OP_DOUBLE_TO_LONG.S b/vm/mterp/armv5/OP_DOUBLE_TO_LONG.S
new file mode 100644
index 0000000..a64f9e7
--- /dev/null
+++ b/vm/mterp/armv5/OP_DOUBLE_TO_LONG.S
@@ -0,0 +1,57 @@
+%verify "executed"
+@include "armv5/unopWide.S" {"instr":"bl      __aeabi_d2lz"}
+%include "armv5/unopWide.S" {"instr":"bl      d2l_doconv"}
+
+%break
+/*
+ * Convert the double in r0/r1 to a long in r0/r1.
+ *
+ * We have to clip values to long min/max per the specification.  The
+ * expected common case is a "reasonable" value that converts directly
+ * to modest integer.  The EABI convert function isn't doing this for us.
+ */
+d2l_doconv:
+    stmfd   sp!, {r4, r5, lr}           @ save regs
+    ldr     r3, .L${opcode}_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, .L${opcode}_min         @ (double)minlong, hi
+    mov     r2, #0                      @ (double)minlong, lo
+    bl      __aeabi_dcmple              @ is arg <= minlong?
+    cmp     r0, #0                      @ nonzero == yes
+    movne   r0, #0                      @ return minlong (8000000000000000)
+    movne   r1, #0x80000000
+    bne     1f
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r5
+    mov     r2, r4                      @ compare against self
+    mov     r3, r5
+    bl      __aeabi_dcmpeq              @ is arg == self?
+    cmp     r0, #0                      @ zero == no
+    moveq   r1, #0                      @ return zero for NaN
+    beq     1f
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r5
+    bl      __aeabi_d2lz                @ convert double to long
+
+1:
+    add     sp, sp, #4
+    ldmfd   sp!, {r4, r5, pc}
+
+.L${opcode}_max:
+    .word   0x43e00000                  @ maxlong, as a double (high word)
+.L${opcode}_min:
+    .word   0xc3e00000                  @ minlong, as a double (high word)
+
diff --git a/vm/mterp/armv5/OP_EXECUTE_INLINE.S b/vm/mterp/armv5/OP_EXECUTE_INLINE.S
new file mode 100644
index 0000000..84836cd
--- /dev/null
+++ b/vm/mterp/armv5/OP_EXECUTE_INLINE.S
@@ -0,0 +1,61 @@
+%verify "executed"
+%verify "exception handled"
+    /*
+     * Execute a "native inline" instruction.
+     *
+     * We need to call:
+     *  dvmPerformInlineOp4Std(arg0, arg1, arg2, arg3, &retval, ref)
+     *
+     * The first four args are in r0-r3, but the last two must be pushed
+     * onto the stack.
+     */
+    /* [opt] execute-inline vAA, {vC, vD, vE, vF}, inline@BBBB */
+    FETCH(r10, 1)                       @ r10<- BBBB
+    add     r1, rGLUE, #offGlue_retval  @ r1<- &glue->retval
+    EXPORT_PC()                         @ can throw
+    sub     sp, sp, #8                  @ make room for arg(s)
+    mov     r0, rINST, lsr #12          @ r0<- B
+    str     r1, [sp]                    @ push &glue->retval
+    bl      .L${opcode}_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
+%break
+
+    /*
+     * Extract args, call function.
+     *  r0 = #of args (0-4)
+     *  r10 = call index
+     *  lr = return addr, above  [DO NOT bl in 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.
+     */
+.L${opcode}_continue:
+    rsb     r0, r0, #4                  @ r0<- 4-r0
+    FETCH(r9, 2)                        @ r9<- FEDC
+    add     pc, pc, r0, lsl #3          @ computed goto, 2 instrs each
+    bl      common_abort                @ (skipped due to ARM prefetch)
+4:  and     ip, r9, #0xf000             @ isolate F
+    ldr     r3, [rFP, ip, lsr #10]      @ r3<- vF (shift right 12, left 2)
+3:  and     ip, r9, #0x0f00             @ isolate E
+    ldr     r2, [rFP, ip, lsr #6]       @ r2<- vE
+2:  and     ip, r9, #0x00f0             @ isolate D
+    ldr     r1, [rFP, ip, lsr #2]       @ r1<- vD
+1:  and     ip, r9, #0x000f             @ isolate C
+    ldr     r0, [rFP, ip, lsl #2]       @ r0<- vC
+0:
+    @b       dvmPerformInlineOp4Std
+    ldr     r9, .L${opcode}_table       @ table of InlineOperation
+    ldr     pc, [r9, r10, lsl #4]       @ sizeof=16, "func" is first entry
+    @ (not reached)
+
+.L${opcode}_table:
+    .word   gDvmInlineOpsTable
+
diff --git a/vm/mterp/armv5/OP_FILLED_NEW_ARRAY.S b/vm/mterp/armv5/OP_FILLED_NEW_ARRAY.S
new file mode 100644
index 0000000..375c811
--- /dev/null
+++ b/vm/mterp/armv5/OP_FILLED_NEW_ARRAY.S
@@ -0,0 +1,101 @@
+%default { "isrange":"0" }
+%verify "executed"
+%verify "unimplemented array type"
+    /*
+     * Create a new array with elements filled from registers.
+     *
+     * for: filled-new-array, filled-new-array/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, type@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResClasses]    @ r3<- pDvmDex->pResClasses
+    EXPORT_PC()                         @ need for resolve and alloc
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved class
+    mov     r10, rINST, lsr #8          @ r10<- AA or BA
+    cmp     r0, #0                      @ already resolved?
+    bne     .L${opcode}_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       .L${opcode}_continue
+%break
+
+    /*
+     * On entry:
+     *  r0 holds array class
+     *  r10 holds AA or BA
+     */
+.L${opcode}_continue:
+    ldr     r3, [r0, #offClassObject_descriptor] @ r3<- arrayClass->descriptor
+    mov     r2, #ALLOC_DONT_TRACK       @ r2<- alloc flags
+    ldrb    r0, [r3, #1]                @ r0<- descriptor[1]
+    .if     $isrange
+    mov     r1, r10                     @ r1<- AA (length)
+    .else
+    mov     r1, r10, lsr #4             @ r1<- B (length)
+    .endif
+    cmp     r0, #'I'                    @ array of ints?
+    mov     r9, r1                      @ save length in r9
+    bne     .L${opcode}_notimpl         @ no, not handled yet
+    bl      dvmAllocPrimitiveArray      @ r0<- call(typeCh, 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     $isrange
+    add     r2, rFP, r1, lsl #2         @ r2<- &fp[CCCC]
+1:  ldr     r3, [r2], #4                @ r3<- *r2++
+    subs    r9, r9, #1                  @ count--
+    str     r3, [r0], #4                @ *contents++ = vX
+    bpl     1b
+    @ continue at 2
+    .else
+    cmp     r9, #4                      @ length was initially 5?
+    and     r2, r10, #15                @ r2<- A
+    bne     1f                          @ <= 4 args, branch
+    GET_VREG(r3, r2)                    @ r3<- vA
+    sub     r9, r9, #1                  @ count--
+    str     r3, [r0, #16]               @ contents[4] = vA
+1:  and     r2, r1, #15                 @ r2<- F/E/D/C
+    GET_VREG(r3, r2)                    @ r3<- vF/vE/vD/vC
+    mov     r1, r1, lsr #4              @ r1<- next reg in low 4
+    subs    r9, r9, #1                  @ count--
+    str     r3, [r0], #4                @ *contents++ = vX
+    bpl     1b
+    @ continue at 2
+    .endif
+
+2:
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    GOTO_OPCODE(ip)                     @ execute it
+
+    /*
+     * Throw an exception indicating that we have not implemented this
+     * mode of filled-new-array.
+     */
+.L${opcode}_notimpl:
+    ldr     r0, strInternalError
+    ldr     r1, strFilledNewArrayNotImpl
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+    .if     (!$isrange)                 @ define in one or the other, not both
+strFilledNewArrayNotImpl:
+    .word   .LstrFilledNewArrayNotImpl
+strInternalError:
+    .word   .LstrInternalError
+    .endif
+
diff --git a/vm/mterp/armv5/OP_FILLED_NEW_ARRAY_RANGE.S b/vm/mterp/armv5/OP_FILLED_NEW_ARRAY_RANGE.S
new file mode 100644
index 0000000..b65cc04
--- /dev/null
+++ b/vm/mterp/armv5/OP_FILLED_NEW_ARRAY_RANGE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_FILLED_NEW_ARRAY.S" { "isrange":"1" }
diff --git a/vm/mterp/armv5/OP_FILL_ARRAY_DATA.S b/vm/mterp/armv5/OP_FILL_ARRAY_DATA.S
new file mode 100644
index 0000000..a0d8399
--- /dev/null
+++ b/vm/mterp/armv5/OP_FILL_ARRAY_DATA.S
@@ -0,0 +1,15 @@
+%verify "executed"
+    /* 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
diff --git a/vm/mterp/armv5/OP_FLOAT_TO_DOUBLE.S b/vm/mterp/armv5/OP_FLOAT_TO_DOUBLE.S
new file mode 100644
index 0000000..3e00f8f
--- /dev/null
+++ b/vm/mterp/armv5/OP_FLOAT_TO_DOUBLE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/unopWider.S" {"instr":"bl      __aeabi_f2d"}
diff --git a/vm/mterp/armv5/OP_FLOAT_TO_INT.S b/vm/mterp/armv5/OP_FLOAT_TO_INT.S
new file mode 100644
index 0000000..2405551
--- /dev/null
+++ b/vm/mterp/armv5/OP_FLOAT_TO_INT.S
@@ -0,0 +1,41 @@
+%verify "executed"
+/* EABI appears to have Java-style conversions of +inf/-inf/NaN */
+%include "armv5/unop.S" {"instr":"bl      __aeabi_f2iz"}
+
+#if 0
+@include "armv5/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
+
diff --git a/vm/mterp/armv5/OP_FLOAT_TO_LONG.S b/vm/mterp/armv5/OP_FLOAT_TO_LONG.S
new file mode 100644
index 0000000..e7e27a2
--- /dev/null
+++ b/vm/mterp/armv5/OP_FLOAT_TO_LONG.S
@@ -0,0 +1,41 @@
+%verify "executed"
+@include "armv5/unopWider.S" {"instr":"bl      __aeabi_f2lz"}
+%include "armv5/unopWider.S" {"instr":"bl      f2l_doconv"}
+
+%break
+/*
+ * 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}
+
diff --git a/vm/mterp/armv5/OP_GOTO.S b/vm/mterp/armv5/OP_GOTO.S
new file mode 100644
index 0000000..3433a73
--- /dev/null
+++ b/vm/mterp/armv5/OP_GOTO.S
@@ -0,0 +1,17 @@
+%verify "executed"
+%verify "forward and backward"
+    /*
+     * 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
+
diff --git a/vm/mterp/armv5/OP_GOTO_16.S b/vm/mterp/armv5/OP_GOTO_16.S
new file mode 100644
index 0000000..479438e
--- /dev/null
+++ b/vm/mterp/armv5/OP_GOTO_16.S
@@ -0,0 +1,16 @@
+%verify "executed"
+%verify "forward and backward"
+    /*
+     * 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
+
diff --git a/vm/mterp/armv5/OP_GOTO_32.S b/vm/mterp/armv5/OP_GOTO_32.S
new file mode 100644
index 0000000..617b8ba
--- /dev/null
+++ b/vm/mterp/armv5/OP_GOTO_32.S
@@ -0,0 +1,24 @@
+%verify "executed"
+%verify "forward, backward, self"
+    /*
+     * 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
+
diff --git a/vm/mterp/armv5/OP_IF_EQ.S b/vm/mterp/armv5/OP_IF_EQ.S
new file mode 100644
index 0000000..905cb90
--- /dev/null
+++ b/vm/mterp/armv5/OP_IF_EQ.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/bincmp.S" { "revcmp":"ne" }
diff --git a/vm/mterp/armv5/OP_IF_EQZ.S b/vm/mterp/armv5/OP_IF_EQZ.S
new file mode 100644
index 0000000..bc33dec
--- /dev/null
+++ b/vm/mterp/armv5/OP_IF_EQZ.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/zcmp.S" { "revcmp":"ne" }
diff --git a/vm/mterp/armv5/OP_IF_GE.S b/vm/mterp/armv5/OP_IF_GE.S
new file mode 100644
index 0000000..aaaf770
--- /dev/null
+++ b/vm/mterp/armv5/OP_IF_GE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/bincmp.S" { "revcmp":"lt" }
diff --git a/vm/mterp/armv5/OP_IF_GEZ.S b/vm/mterp/armv5/OP_IF_GEZ.S
new file mode 100644
index 0000000..0e6bfc2
--- /dev/null
+++ b/vm/mterp/armv5/OP_IF_GEZ.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/zcmp.S" { "revcmp":"lt" }
diff --git a/vm/mterp/armv5/OP_IF_GT.S b/vm/mterp/armv5/OP_IF_GT.S
new file mode 100644
index 0000000..8ca99d9
--- /dev/null
+++ b/vm/mterp/armv5/OP_IF_GT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/bincmp.S" { "revcmp":"le" }
diff --git a/vm/mterp/armv5/OP_IF_GTZ.S b/vm/mterp/armv5/OP_IF_GTZ.S
new file mode 100644
index 0000000..264ae30
--- /dev/null
+++ b/vm/mterp/armv5/OP_IF_GTZ.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/zcmp.S" { "revcmp":"le" }
diff --git a/vm/mterp/armv5/OP_IF_LE.S b/vm/mterp/armv5/OP_IF_LE.S
new file mode 100644
index 0000000..55d4d3e
--- /dev/null
+++ b/vm/mterp/armv5/OP_IF_LE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/bincmp.S" { "revcmp":"gt" }
diff --git a/vm/mterp/armv5/OP_IF_LEZ.S b/vm/mterp/armv5/OP_IF_LEZ.S
new file mode 100644
index 0000000..1cd273d
--- /dev/null
+++ b/vm/mterp/armv5/OP_IF_LEZ.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/zcmp.S" { "revcmp":"gt" }
diff --git a/vm/mterp/armv5/OP_IF_LT.S b/vm/mterp/armv5/OP_IF_LT.S
new file mode 100644
index 0000000..8be9a8d
--- /dev/null
+++ b/vm/mterp/armv5/OP_IF_LT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/bincmp.S" { "revcmp":"ge" }
diff --git a/vm/mterp/armv5/OP_IF_LTZ.S b/vm/mterp/armv5/OP_IF_LTZ.S
new file mode 100644
index 0000000..84052b6
--- /dev/null
+++ b/vm/mterp/armv5/OP_IF_LTZ.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/zcmp.S" { "revcmp":"ge" }
diff --git a/vm/mterp/armv5/OP_IF_NE.S b/vm/mterp/armv5/OP_IF_NE.S
new file mode 100644
index 0000000..bd42a70
--- /dev/null
+++ b/vm/mterp/armv5/OP_IF_NE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/bincmp.S" { "revcmp":"eq" }
diff --git a/vm/mterp/armv5/OP_IF_NEZ.S b/vm/mterp/armv5/OP_IF_NEZ.S
new file mode 100644
index 0000000..a418241
--- /dev/null
+++ b/vm/mterp/armv5/OP_IF_NEZ.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/zcmp.S" { "revcmp":"eq" }
diff --git a/vm/mterp/armv5/OP_IGET.S b/vm/mterp/armv5/OP_IGET.S
new file mode 100644
index 0000000..a347e5c
--- /dev/null
+++ b/vm/mterp/armv5/OP_IGET.S
@@ -0,0 +1,47 @@
+%default { "load":"ldr", "sqnum":"0" }
+%verify "executed"
+%verify "null object"
+%verify "field already resolved"
+%verify "field not yet resolved"
+%verify "field cannot be resolved"
+    /*
+     * 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     .L${opcode}_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     .L${opcode}_finish
+    b       common_exceptionThrown
+%break
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.L${opcode}_finish:
+    @bl      common_squeak${sqnum}
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    $load   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
+
diff --git a/vm/mterp/armv5/OP_IGET_BOOLEAN.S b/vm/mterp/armv5/OP_IGET_BOOLEAN.S
new file mode 100644
index 0000000..acdd60d
--- /dev/null
+++ b/vm/mterp/armv5/OP_IGET_BOOLEAN.S
@@ -0,0 +1,3 @@
+%verify "executed"
+@include "armv5/OP_IGET.S" { "load":"ldrb", "sqnum":"1" }
+%include "armv5/OP_IGET.S" { "load":"ldr", "sqnum":"1" }
diff --git a/vm/mterp/armv5/OP_IGET_BYTE.S b/vm/mterp/armv5/OP_IGET_BYTE.S
new file mode 100644
index 0000000..52c0955
--- /dev/null
+++ b/vm/mterp/armv5/OP_IGET_BYTE.S
@@ -0,0 +1,4 @@
+%verify "executed"
+%verify "negative value is sign-extended"
+@include "armv5/OP_IGET.S" { "load":"ldrsb", "sqnum":"2" }
+%include "armv5/OP_IGET.S" { "load":"ldr", "sqnum":"2" }
diff --git a/vm/mterp/armv5/OP_IGET_CHAR.S b/vm/mterp/armv5/OP_IGET_CHAR.S
new file mode 100644
index 0000000..cc7501a
--- /dev/null
+++ b/vm/mterp/armv5/OP_IGET_CHAR.S
@@ -0,0 +1,4 @@
+%verify "executed"
+%verify "large values are not sign-extended"
+@include "armv5/OP_IGET.S" { "load":"ldrh", "sqnum":"3" }
+%include "armv5/OP_IGET.S" { "load":"ldr", "sqnum":"3" }
diff --git a/vm/mterp/armv5/OP_IGET_OBJECT.S b/vm/mterp/armv5/OP_IGET_OBJECT.S
new file mode 100644
index 0000000..108f9a5
--- /dev/null
+++ b/vm/mterp/armv5/OP_IGET_OBJECT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_IGET.S"
diff --git a/vm/mterp/armv5/OP_IGET_OBJECT_QUICK.S b/vm/mterp/armv5/OP_IGET_OBJECT_QUICK.S
new file mode 100644
index 0000000..546d435
--- /dev/null
+++ b/vm/mterp/armv5/OP_IGET_OBJECT_QUICK.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_IGET_QUICK.S"
diff --git a/vm/mterp/armv5/OP_IGET_QUICK.S b/vm/mterp/armv5/OP_IGET_QUICK.S
new file mode 100644
index 0000000..f248bc2
--- /dev/null
+++ b/vm/mterp/armv5/OP_IGET_QUICK.S
@@ -0,0 +1,17 @@
+%verify "executed"
+%verify "null object"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_IGET_SHORT.S b/vm/mterp/armv5/OP_IGET_SHORT.S
new file mode 100644
index 0000000..7c8b0ee
--- /dev/null
+++ b/vm/mterp/armv5/OP_IGET_SHORT.S
@@ -0,0 +1,4 @@
+%verify "executed"
+%verify "negative value is sign-extended"
+@include "armv5/OP_IGET.S" { "load":"ldrsh", "sqnum":"4" }
+%include "armv5/OP_IGET.S" { "load":"ldr", "sqnum":"4" }
diff --git a/vm/mterp/armv5/OP_IGET_WIDE.S b/vm/mterp/armv5/OP_IGET_WIDE.S
new file mode 100644
index 0000000..22377d9
--- /dev/null
+++ b/vm/mterp/armv5/OP_IGET_WIDE.S
@@ -0,0 +1,44 @@
+%verify "executed"
+%verify "null object"
+%verify "field already resolved"
+%verify "field not yet resolved"
+%verify "field cannot be resolved"
+    /*
+     * 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     .L${opcode}_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     .L${opcode}_finish
+    b       common_exceptionThrown
+%break
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.L${opcode}_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
+
diff --git a/vm/mterp/armv5/OP_IGET_WIDE_QUICK.S b/vm/mterp/armv5/OP_IGET_WIDE_QUICK.S
new file mode 100644
index 0000000..ece7e7a
--- /dev/null
+++ b/vm/mterp/armv5/OP_IGET_WIDE_QUICK.S
@@ -0,0 +1,17 @@
+%verify "executed"
+%verify "null object"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_INSTANCE_OF.S b/vm/mterp/armv5/OP_INSTANCE_OF.S
new file mode 100644
index 0000000..da9f450
--- /dev/null
+++ b/vm/mterp/armv5/OP_INSTANCE_OF.S
@@ -0,0 +1,86 @@
+%verify "executed"
+%verify "null object"
+%verify "class cast exception thrown, with correct class name"
+%verify "class cast exception not thrown on same class"
+%verify "class cast exception not thrown on subclass"
+%verify "class not resolved"
+%verify "class already resolved"
+    /*
+     * 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     .L${opcode}_store           @ null obj, not an instance, store r0
+    FETCH(r3, 1)                        @ r3<- CCCC
+    ldr     r2, [r2, #offDvmDex_pResClasses]    @ r2<- pDvmDex->pResClasses
+    ldr     r1, [r2, r3, lsl #2]        @ r1<- resolved class
+    ldr     r0, [r0, #offObject_clazz]  @ r0<- obj->clazz
+    cmp     r1, #0                      @ have we resolved this before?
+    beq     .L${opcode}_resolve         @ not resolved, do it now
+.L${opcode}_resolved: @ r0=obj->clazz, r1=resolved class
+    cmp     r0, r1                      @ same class (trivial success)?
+    beq     .L${opcode}_trivial         @ yes, trivial finish
+    b       .L${opcode}_fullcheck       @ no, do full check
+%break
+
+    /*
+     * Trivial test failed, need to perform full check.  This is common.
+     *  r0 holds obj->clazz
+     *  r1 holds class resolved from BBBB
+     *  r9 holds A
+     */
+.L${opcode}_fullcheck:
+    bl      dvmInstanceofNonTrivial     @ r0<- boolean result
+    @ fall through to ${opcode}_store
+
+    /*
+     * r0 holds boolean result
+     * r9 holds A
+     */
+.L${opcode}_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
+     */
+.L${opcode}_trivial:
+    mov     r0, #1                      @ indicate success
+    @ could b ${opcode}_store, but copying is faster and cheaper
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r0, r9)                    @ vA<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+    /*
+     * Resolution required.  This is the least-likely path.
+     *
+     *  r3 holds BBBB
+     *  r9 holds A
+     */
+.L${opcode}_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       .L${opcode}_resolved        @ pick up where we left off
+
diff --git a/vm/mterp/armv5/OP_INT_TO_BYTE.S b/vm/mterp/armv5/OP_INT_TO_BYTE.S
new file mode 100644
index 0000000..568193e
--- /dev/null
+++ b/vm/mterp/armv5/OP_INT_TO_BYTE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/unop.S" {"preinstr":"mov     r0, r0, asl #24", "instr":"mov     r0, r0, asr #24"}
diff --git a/vm/mterp/armv5/OP_INT_TO_CHAR.S b/vm/mterp/armv5/OP_INT_TO_CHAR.S
new file mode 100644
index 0000000..15da1f0
--- /dev/null
+++ b/vm/mterp/armv5/OP_INT_TO_CHAR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/unop.S" {"preinstr":"mov     r0, r0, asl #16", "instr":"mov     r0, r0, lsr #16"}
diff --git a/vm/mterp/armv5/OP_INT_TO_DOUBLE.S b/vm/mterp/armv5/OP_INT_TO_DOUBLE.S
new file mode 100644
index 0000000..342ad84
--- /dev/null
+++ b/vm/mterp/armv5/OP_INT_TO_DOUBLE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/unopWider.S" {"instr":"bl      __aeabi_i2d"}
diff --git a/vm/mterp/armv5/OP_INT_TO_FLOAT.S b/vm/mterp/armv5/OP_INT_TO_FLOAT.S
new file mode 100644
index 0000000..464df47
--- /dev/null
+++ b/vm/mterp/armv5/OP_INT_TO_FLOAT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/unop.S" {"instr":"bl      __aeabi_i2f"}
diff --git a/vm/mterp/armv5/OP_INT_TO_LONG.S b/vm/mterp/armv5/OP_INT_TO_LONG.S
new file mode 100644
index 0000000..31c0216
--- /dev/null
+++ b/vm/mterp/armv5/OP_INT_TO_LONG.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/unopWider.S" {"instr":"mov     r1, r0, asr #31"}
diff --git a/vm/mterp/armv5/OP_INT_TO_SHORT.S b/vm/mterp/armv5/OP_INT_TO_SHORT.S
new file mode 100644
index 0000000..d15de8b
--- /dev/null
+++ b/vm/mterp/armv5/OP_INT_TO_SHORT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/unop.S" {"preinstr":"mov     r0, r0, asl #16", "instr":"mov     r0, r0, asr #16"}
diff --git a/vm/mterp/armv5/OP_INVOKE_DIRECT.S b/vm/mterp/armv5/OP_INVOKE_DIRECT.S
new file mode 100644
index 0000000..c8d551c
--- /dev/null
+++ b/vm/mterp/armv5/OP_INVOKE_DIRECT.S
@@ -0,0 +1,48 @@
+%default { "isrange":"0", "routine":"NoRange" }
+%verify "executed"
+%verify "unknown method"
+    /*
+     * 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     (!$isrange)
+    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     .L${opcode}_resolve         @ not resolved, do it now
+.L${opcode}_finish:
+    cmp     r2, #0                      @ null "this" ref?
+    bne     common_invokeMethod${routine}   @ no, continue on
+    b       common_errNullObject        @ yes, throw exception
+%break
+
+    /*
+     * On entry:
+     *  r1 = reference (BBBB or CCCC)
+     *  r10 = "this" register
+     */
+.L${opcode}_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     .L${opcode}_finish          @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
diff --git a/vm/mterp/armv5/OP_INVOKE_DIRECT_EMPTY.S b/vm/mterp/armv5/OP_INVOKE_DIRECT_EMPTY.S
new file mode 100644
index 0000000..3c6b192
--- /dev/null
+++ b/vm/mterp/armv5/OP_INVOKE_DIRECT_EMPTY.S
@@ -0,0 +1,7 @@
+%verify "executed"
+    /*
+     * invoke-direct-empty is a no-op in a "standard" interpreter.
+     */
+    FETCH_ADVANCE_INST(3)               @ advance to next instr, load rINST
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    GOTO_OPCODE(ip)                     @ execute it
diff --git a/vm/mterp/armv5/OP_INVOKE_DIRECT_RANGE.S b/vm/mterp/armv5/OP_INVOKE_DIRECT_RANGE.S
new file mode 100644
index 0000000..1a6fd37
--- /dev/null
+++ b/vm/mterp/armv5/OP_INVOKE_DIRECT_RANGE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_INVOKE_DIRECT.S" { "isrange":"1", "routine":"Range" }
diff --git a/vm/mterp/armv5/OP_INVOKE_INTERFACE.S b/vm/mterp/armv5/OP_INVOKE_INTERFACE.S
new file mode 100644
index 0000000..5463d5c
--- /dev/null
+++ b/vm/mterp/armv5/OP_INVOKE_INTERFACE.S
@@ -0,0 +1,28 @@
+%default { "isrange":"0", "routine":"NoRange" }
+%verify "executed"
+%verify "unknown method"
+%verify "null object"
+    /*
+     * Handle an interface method call.
+     *
+     * for: invoke-interface, invoke-interface/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r2, 2)                        @ r2<- FEDC or CCCC
+    FETCH(r1, 1)                        @ r1<- BBBB
+    .if     (!$isrange)
+    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_invokeMethod${routine} @ jump to common handler 
+
diff --git a/vm/mterp/armv5/OP_INVOKE_INTERFACE_RANGE.S b/vm/mterp/armv5/OP_INVOKE_INTERFACE_RANGE.S
new file mode 100644
index 0000000..8ed5518
--- /dev/null
+++ b/vm/mterp/armv5/OP_INVOKE_INTERFACE_RANGE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_INVOKE_INTERFACE.S" { "isrange":"1", "routine":"Range" }
diff --git a/vm/mterp/armv5/OP_INVOKE_STATIC.S b/vm/mterp/armv5/OP_INVOKE_STATIC.S
new file mode 100644
index 0000000..57d337b
--- /dev/null
+++ b/vm/mterp/armv5/OP_INVOKE_STATIC.S
@@ -0,0 +1,25 @@
+%default { "routine":"NoRange" }
+%verify "executed"
+%verify "unknown method"
+    /*
+     * Handle a static method call.
+     *
+     * for: invoke-static, invoke-static/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved methodToCall
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    bne     common_invokeMethod${routine} @ 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_invokeMethod${routine} @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
diff --git a/vm/mterp/armv5/OP_INVOKE_STATIC_RANGE.S b/vm/mterp/armv5/OP_INVOKE_STATIC_RANGE.S
new file mode 100644
index 0000000..a972a9e
--- /dev/null
+++ b/vm/mterp/armv5/OP_INVOKE_STATIC_RANGE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_INVOKE_STATIC.S" { "routine":"Range" }
diff --git a/vm/mterp/armv5/OP_INVOKE_SUPER.S b/vm/mterp/armv5/OP_INVOKE_SUPER.S
new file mode 100644
index 0000000..f24f690
--- /dev/null
+++ b/vm/mterp/armv5/OP_INVOKE_SUPER.S
@@ -0,0 +1,61 @@
+%default { "isrange":"0", "routine":"NoRange" }
+%verify "executed"
+%verify "unknown method"
+    /*
+     * Handle a "super" method call.
+     *
+     * for: invoke-super, invoke-super/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    .if     (!$isrange)
+    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     .L${opcode}_continue        @ resolved, continue on
+    b       .L${opcode}_resolve         @ do resolve now
+%break
+
+    /*
+     * At this point:
+     *  r0 = resolved base method
+     *  r9 = method->clazz
+     */
+.L${opcode}_continue:
+    ldr     r1, [r9, #offClassObject_super]     @ r1<- method->clazz->super
+    ldrh    r2, [r0, #offMethod_methodIndex]    @ r2<- baseMethod->methodIndex
+    ldr     r3, [r1, #offClassObject_vtableCount]   @ r3<- super->vtableCount
+    EXPORT_PC()                         @ must export for invoke
+    cmp     r2, r3                      @ compare (methodIndex, vtableCount)
+    bcs     .L${opcode}_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_invokeMethod${routine} @ continue on
+
+.L${opcode}_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     .L${opcode}_continue        @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+    /*
+     * Throw a NoSuchMethodError with the method name as the message.
+     *  r0 = resolved base method
+     */
+.L${opcode}_nsm:
+    ldr     r1, [r0, #offMethod_name]   @ r1<- method name
+    b       common_errNoSuchMethod
+
diff --git a/vm/mterp/armv5/OP_INVOKE_SUPER_QUICK.S b/vm/mterp/armv5/OP_INVOKE_SUPER_QUICK.S
new file mode 100644
index 0000000..90b2a91
--- /dev/null
+++ b/vm/mterp/armv5/OP_INVOKE_SUPER_QUICK.S
@@ -0,0 +1,26 @@
+%default { "isrange":"0", "routine":"NoRange" }
+%verify "executed"
+%verify "unknown method"
+    /*
+     * 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     (!$isrange)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r2, [r2, #offMethod_clazz]  @ r2<- method->clazz
+    EXPORT_PC()                         @ must export for invoke
+    ldr     r2, [r2, #offClassObject_super]     @ r2<- method->clazz->super
+    GET_VREG(r3, r10)                   @ r3<- "this"
+    ldr     r2, [r2, #offClassObject_vtable]    @ r2<- ...clazz->super->vtable
+    cmp     r3, #0                      @ null "this" ref?
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- super->vtable[BBBB]
+    beq     common_errNullObject        @ "this" is null, throw exception
+    bl      common_invokeMethod${routine} @ continue on
+
diff --git a/vm/mterp/armv5/OP_INVOKE_SUPER_QUICK_RANGE.S b/vm/mterp/armv5/OP_INVOKE_SUPER_QUICK_RANGE.S
new file mode 100644
index 0000000..2b9ac78
--- /dev/null
+++ b/vm/mterp/armv5/OP_INVOKE_SUPER_QUICK_RANGE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_INVOKE_SUPER_QUICK.S" { "isrange":"1", "routine":"Range" }
diff --git a/vm/mterp/armv5/OP_INVOKE_SUPER_RANGE.S b/vm/mterp/armv5/OP_INVOKE_SUPER_RANGE.S
new file mode 100644
index 0000000..a0b9f75
--- /dev/null
+++ b/vm/mterp/armv5/OP_INVOKE_SUPER_RANGE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_INVOKE_SUPER.S" { "isrange":"1", "routine":"Range" }
diff --git a/vm/mterp/armv5/OP_INVOKE_VIRTUAL.S b/vm/mterp/armv5/OP_INVOKE_VIRTUAL.S
new file mode 100644
index 0000000..33dc108
--- /dev/null
+++ b/vm/mterp/armv5/OP_INVOKE_VIRTUAL.S
@@ -0,0 +1,46 @@
+%default { "isrange":"0", "routine":"NoRange" }
+%verify "executed"
+%verify "unknown method"
+%verify "null object"
+    /*
+     * 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     (!$isrange)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    bne     .L${opcode}_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     .L${opcode}_continue        @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+%break
+
+    /*
+     * At this point:
+     *  r0 = resolved base method
+     *  r10 = C or CCCC (index of first arg, which is the "this" ptr)
+     */
+.L${opcode}_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_invokeMethod${routine} @ continue on
+
diff --git a/vm/mterp/armv5/OP_INVOKE_VIRTUAL_QUICK.S b/vm/mterp/armv5/OP_INVOKE_VIRTUAL_QUICK.S
new file mode 100644
index 0000000..bc34023
--- /dev/null
+++ b/vm/mterp/armv5/OP_INVOKE_VIRTUAL_QUICK.S
@@ -0,0 +1,23 @@
+%default { "isrange":"0", "routine":"NoRange" }
+%verify "executed"
+%verify "null object"
+    /*
+     * Handle an optimized virtual method call.
+     *
+     * for: [opt] invoke-virtual-quick, invoke-virtual-quick/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r3, 2)                        @ r3<- FEDC or CCCC
+    FETCH(r1, 1)                        @ r1<- BBBB
+    .if     (!$isrange)
+    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_invokeMethod${routine} @ continue on
diff --git a/vm/mterp/armv5/OP_INVOKE_VIRTUAL_QUICK_RANGE.S b/vm/mterp/armv5/OP_INVOKE_VIRTUAL_QUICK_RANGE.S
new file mode 100644
index 0000000..c43ce3a
--- /dev/null
+++ b/vm/mterp/armv5/OP_INVOKE_VIRTUAL_QUICK_RANGE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_INVOKE_VIRTUAL_QUICK.S" { "isrange":"1", "routine":"Range" }
diff --git a/vm/mterp/armv5/OP_INVOKE_VIRTUAL_RANGE.S b/vm/mterp/armv5/OP_INVOKE_VIRTUAL_RANGE.S
new file mode 100644
index 0000000..5ca501c
--- /dev/null
+++ b/vm/mterp/armv5/OP_INVOKE_VIRTUAL_RANGE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_INVOKE_VIRTUAL.S" { "isrange":"1", "routine":"Range" }
diff --git a/vm/mterp/armv5/OP_IPUT.S b/vm/mterp/armv5/OP_IPUT.S
new file mode 100644
index 0000000..35c139f
--- /dev/null
+++ b/vm/mterp/armv5/OP_IPUT.S
@@ -0,0 +1,47 @@
+%default { "store":"str", "sqnum":"0" }
+%verify "executed"
+%verify "null object"
+%verify "field already resolved"
+%verify "field not yet resolved"
+%verify "field cannot be resolved"
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .L${opcode}_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     .L${opcode}_finish          @ yes, finish up
+    b       common_exceptionThrown
+%break
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.L${opcode}_finish:
+    @bl      common_squeak${sqnum}
+    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
+    $store  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
diff --git a/vm/mterp/armv5/OP_IPUT_BOOLEAN.S b/vm/mterp/armv5/OP_IPUT_BOOLEAN.S
new file mode 100644
index 0000000..1af9629
--- /dev/null
+++ b/vm/mterp/armv5/OP_IPUT_BOOLEAN.S
@@ -0,0 +1,3 @@
+%verify "executed"
+@include "armv5/OP_IPUT.S" { "store":"strb", "sqnum":"1" }
+%include "armv5/OP_IPUT.S" { "store":"str", "sqnum":"1" }
diff --git a/vm/mterp/armv5/OP_IPUT_BYTE.S b/vm/mterp/armv5/OP_IPUT_BYTE.S
new file mode 100644
index 0000000..8385e07
--- /dev/null
+++ b/vm/mterp/armv5/OP_IPUT_BYTE.S
@@ -0,0 +1,3 @@
+%verify "executed"
+@include "armv5/OP_IPUT.S" { "store":"strb", "sqnum":"2" }
+%include "armv5/OP_IPUT.S" { "store":"str", "sqnum":"2" }
diff --git a/vm/mterp/armv5/OP_IPUT_CHAR.S b/vm/mterp/armv5/OP_IPUT_CHAR.S
new file mode 100644
index 0000000..c0220c0
--- /dev/null
+++ b/vm/mterp/armv5/OP_IPUT_CHAR.S
@@ -0,0 +1,3 @@
+%verify "executed"
+@include "armv5/OP_IPUT.S" { "store":"strh", "sqnum":"3" }
+%include "armv5/OP_IPUT.S" { "store":"str", "sqnum":"3" }
diff --git a/vm/mterp/armv5/OP_IPUT_OBJECT.S b/vm/mterp/armv5/OP_IPUT_OBJECT.S
new file mode 100644
index 0000000..83f3665
--- /dev/null
+++ b/vm/mterp/armv5/OP_IPUT_OBJECT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_IPUT.S"
diff --git a/vm/mterp/armv5/OP_IPUT_OBJECT_QUICK.S b/vm/mterp/armv5/OP_IPUT_OBJECT_QUICK.S
new file mode 100644
index 0000000..d46b1d8
--- /dev/null
+++ b/vm/mterp/armv5/OP_IPUT_OBJECT_QUICK.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_IPUT_QUICK.S"
diff --git a/vm/mterp/armv5/OP_IPUT_QUICK.S b/vm/mterp/armv5/OP_IPUT_QUICK.S
new file mode 100644
index 0000000..c33a738
--- /dev/null
+++ b/vm/mterp/armv5/OP_IPUT_QUICK.S
@@ -0,0 +1,17 @@
+%verify "executed"
+%verify "null object"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_IPUT_SHORT.S b/vm/mterp/armv5/OP_IPUT_SHORT.S
new file mode 100644
index 0000000..f53b80b
--- /dev/null
+++ b/vm/mterp/armv5/OP_IPUT_SHORT.S
@@ -0,0 +1,3 @@
+%verify "executed"
+@include "armv5/OP_IPUT.S" { "store":"strh", "sqnum":"4" }
+%include "armv5/OP_IPUT.S" { "store":"str", "sqnum":"4" }
diff --git a/vm/mterp/armv5/OP_IPUT_WIDE.S b/vm/mterp/armv5/OP_IPUT_WIDE.S
new file mode 100644
index 0000000..75465ec
--- /dev/null
+++ b/vm/mterp/armv5/OP_IPUT_WIDE.S
@@ -0,0 +1,41 @@
+%verify "executed"
+%verify "null object"
+%verify "field already resolved"
+%verify "field not yet resolved"
+%verify "field cannot be resolved"
+    /* 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     .L${opcode}_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     .L${opcode}_finish          @ yes, finish up
+    b       common_exceptionThrown
+%break
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.L${opcode}_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
+
diff --git a/vm/mterp/armv5/OP_IPUT_WIDE_QUICK.S b/vm/mterp/armv5/OP_IPUT_WIDE_QUICK.S
new file mode 100644
index 0000000..290591c
--- /dev/null
+++ b/vm/mterp/armv5/OP_IPUT_WIDE_QUICK.S
@@ -0,0 +1,17 @@
+%verify "executed"
+%verify "null object"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_LONG_TO_DOUBLE.S b/vm/mterp/armv5/OP_LONG_TO_DOUBLE.S
new file mode 100644
index 0000000..a10258b
--- /dev/null
+++ b/vm/mterp/armv5/OP_LONG_TO_DOUBLE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/unopWide.S" {"instr":"bl      __aeabi_l2d"}
diff --git a/vm/mterp/armv5/OP_LONG_TO_FLOAT.S b/vm/mterp/armv5/OP_LONG_TO_FLOAT.S
new file mode 100644
index 0000000..9e1e998
--- /dev/null
+++ b/vm/mterp/armv5/OP_LONG_TO_FLOAT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/unopNarrower.S" {"instr":"bl      __aeabi_l2f"}
diff --git a/vm/mterp/armv5/OP_LONG_TO_INT.S b/vm/mterp/armv5/OP_LONG_TO_INT.S
new file mode 100644
index 0000000..3dfda7d
--- /dev/null
+++ b/vm/mterp/armv5/OP_LONG_TO_INT.S
@@ -0,0 +1,3 @@
+%verify "executed"
+/* we ignore the high word, making this equivalent to a 32-bit reg move */
+%include "armv5/OP_MOVE.S"
diff --git a/vm/mterp/armv5/OP_MONITOR_ENTER.S b/vm/mterp/armv5/OP_MONITOR_ENTER.S
new file mode 100644
index 0000000..6d4c2d8
--- /dev/null
+++ b/vm/mterp/armv5/OP_MONITOR_ENTER.S
@@ -0,0 +1,25 @@
+%verify "executed"
+%verify "exception for null object"
+    /*
+     * 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
+
diff --git a/vm/mterp/armv5/OP_MONITOR_EXIT.S b/vm/mterp/armv5/OP_MONITOR_EXIT.S
new file mode 100644
index 0000000..b334ae9
--- /dev/null
+++ b/vm/mterp/armv5/OP_MONITOR_EXIT.S
@@ -0,0 +1,24 @@
+%verify "executed"
+%verify "exception for null object (impossible in javac)"
+%verify "dvmUnlockObject fails"
+    /*
+     * 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
+
diff --git a/vm/mterp/armv5/OP_MOVE.S b/vm/mterp/armv5/OP_MOVE.S
new file mode 100644
index 0000000..48db45f
--- /dev/null
+++ b/vm/mterp/armv5/OP_MOVE.S
@@ -0,0 +1,12 @@
+%verify "executed"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_MOVE_16.S b/vm/mterp/armv5/OP_MOVE_16.S
new file mode 100644
index 0000000..f20f57b
--- /dev/null
+++ b/vm/mterp/armv5/OP_MOVE_16.S
@@ -0,0 +1,11 @@
+%verify "executed"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_MOVE_EXCEPTION.S b/vm/mterp/armv5/OP_MOVE_EXCEPTION.S
new file mode 100644
index 0000000..45e66fa
--- /dev/null
+++ b/vm/mterp/armv5/OP_MOVE_EXCEPTION.S
@@ -0,0 +1,12 @@
+%verify "executed"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_MOVE_FROM16.S b/vm/mterp/armv5/OP_MOVE_FROM16.S
new file mode 100644
index 0000000..fe54374
--- /dev/null
+++ b/vm/mterp/armv5/OP_MOVE_FROM16.S
@@ -0,0 +1,11 @@
+%verify "executed"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_MOVE_OBJECT.S b/vm/mterp/armv5/OP_MOVE_OBJECT.S
new file mode 100644
index 0000000..4003409
--- /dev/null
+++ b/vm/mterp/armv5/OP_MOVE_OBJECT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_MOVE.S"
diff --git a/vm/mterp/armv5/OP_MOVE_OBJECT_16.S b/vm/mterp/armv5/OP_MOVE_OBJECT_16.S
new file mode 100644
index 0000000..3ce9af7
--- /dev/null
+++ b/vm/mterp/armv5/OP_MOVE_OBJECT_16.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_MOVE_16.S"
diff --git a/vm/mterp/armv5/OP_MOVE_OBJECT_FROM16.S b/vm/mterp/armv5/OP_MOVE_OBJECT_FROM16.S
new file mode 100644
index 0000000..4abb4bf
--- /dev/null
+++ b/vm/mterp/armv5/OP_MOVE_OBJECT_FROM16.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_MOVE_FROM16.S"
diff --git a/vm/mterp/armv5/OP_MOVE_RESULT.S b/vm/mterp/armv5/OP_MOVE_RESULT.S
new file mode 100644
index 0000000..d19de23
--- /dev/null
+++ b/vm/mterp/armv5/OP_MOVE_RESULT.S
@@ -0,0 +1,10 @@
+%verify "executed"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_MOVE_RESULT_OBJECT.S b/vm/mterp/armv5/OP_MOVE_RESULT_OBJECT.S
new file mode 100644
index 0000000..0d5a4d8
--- /dev/null
+++ b/vm/mterp/armv5/OP_MOVE_RESULT_OBJECT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_MOVE_RESULT.S"
diff --git a/vm/mterp/armv5/OP_MOVE_RESULT_WIDE.S b/vm/mterp/armv5/OP_MOVE_RESULT_WIDE.S
new file mode 100644
index 0000000..7bb8254
--- /dev/null
+++ b/vm/mterp/armv5/OP_MOVE_RESULT_WIDE.S
@@ -0,0 +1,11 @@
+%verify "executed"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_MOVE_WIDE.S b/vm/mterp/armv5/OP_MOVE_WIDE.S
new file mode 100644
index 0000000..6bcc89a
--- /dev/null
+++ b/vm/mterp/armv5/OP_MOVE_WIDE.S
@@ -0,0 +1,14 @@
+%verify "executed"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_MOVE_WIDE_16.S b/vm/mterp/armv5/OP_MOVE_WIDE_16.S
new file mode 100644
index 0000000..1e0b5f2
--- /dev/null
+++ b/vm/mterp/armv5/OP_MOVE_WIDE_16.S
@@ -0,0 +1,13 @@
+%verify "executed"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_MOVE_WIDE_FROM16.S b/vm/mterp/armv5/OP_MOVE_WIDE_FROM16.S
new file mode 100644
index 0000000..0771dbc
--- /dev/null
+++ b/vm/mterp/armv5/OP_MOVE_WIDE_FROM16.S
@@ -0,0 +1,13 @@
+%verify "executed"
+    /* 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
+
diff --git a/vm/mterp/armv5/OP_MUL_DOUBLE.S b/vm/mterp/armv5/OP_MUL_DOUBLE.S
new file mode 100644
index 0000000..d3ed1b7
--- /dev/null
+++ b/vm/mterp/armv5/OP_MUL_DOUBLE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide.S" {"instr":"bl      __aeabi_dmul"}
diff --git a/vm/mterp/armv5/OP_MUL_DOUBLE_2ADDR.S b/vm/mterp/armv5/OP_MUL_DOUBLE_2ADDR.S
new file mode 100644
index 0000000..4cc9cd6
--- /dev/null
+++ b/vm/mterp/armv5/OP_MUL_DOUBLE_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide2addr.S" {"instr":"bl      __aeabi_dmul"}
diff --git a/vm/mterp/armv5/OP_MUL_FLOAT.S b/vm/mterp/armv5/OP_MUL_FLOAT.S
new file mode 100644
index 0000000..7b3d51c
--- /dev/null
+++ b/vm/mterp/armv5/OP_MUL_FLOAT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop.S" {"instr":"bl      __aeabi_fmul"}
diff --git a/vm/mterp/armv5/OP_MUL_FLOAT_2ADDR.S b/vm/mterp/armv5/OP_MUL_FLOAT_2ADDR.S
new file mode 100644
index 0000000..a440145
--- /dev/null
+++ b/vm/mterp/armv5/OP_MUL_FLOAT_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop2addr.S" {"instr":"bl      __aeabi_fmul"}
diff --git a/vm/mterp/armv5/OP_MUL_INT.S b/vm/mterp/armv5/OP_MUL_INT.S
new file mode 100644
index 0000000..178f479
--- /dev/null
+++ b/vm/mterp/armv5/OP_MUL_INT.S
@@ -0,0 +1,3 @@
+%verify "executed"
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+%include "armv5/binop.S" {"instr":"mul     r0, r1, r0"}
diff --git a/vm/mterp/armv5/OP_MUL_INT_2ADDR.S b/vm/mterp/armv5/OP_MUL_INT_2ADDR.S
new file mode 100644
index 0000000..270ccb5
--- /dev/null
+++ b/vm/mterp/armv5/OP_MUL_INT_2ADDR.S
@@ -0,0 +1,3 @@
+%verify "executed"
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+%include "armv5/binop2addr.S" {"instr":"mul     r0, r1, r0"}
diff --git a/vm/mterp/armv5/OP_MUL_INT_LIT16.S b/vm/mterp/armv5/OP_MUL_INT_LIT16.S
new file mode 100644
index 0000000..2f135d6
--- /dev/null
+++ b/vm/mterp/armv5/OP_MUL_INT_LIT16.S
@@ -0,0 +1,3 @@
+%verify "executed"
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+%include "armv5/binopLit16.S" {"instr":"mul     r0, r1, r0"}
diff --git a/vm/mterp/armv5/OP_MUL_INT_LIT8.S b/vm/mterp/armv5/OP_MUL_INT_LIT8.S
new file mode 100644
index 0000000..8a768b4
--- /dev/null
+++ b/vm/mterp/armv5/OP_MUL_INT_LIT8.S
@@ -0,0 +1,3 @@
+%verify "executed"
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+%include "armv5/binopLit8.S" {"instr":"mul     r0, r1, r0"}
diff --git a/vm/mterp/armv5/OP_MUL_LONG.S b/vm/mterp/armv5/OP_MUL_LONG.S
new file mode 100644
index 0000000..b812ca1
--- /dev/null
+++ b/vm/mterp/armv5/OP_MUL_LONG.S
@@ -0,0 +1,42 @@
+%verify "executed"
+    /*
+     * 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       .L${opcode}_finish
+%break
+
+.L${opcode}_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
+
diff --git a/vm/mterp/armv5/OP_MUL_LONG_2ADDR.S b/vm/mterp/armv5/OP_MUL_LONG_2ADDR.S
new file mode 100644
index 0000000..4c1bcb8
--- /dev/null
+++ b/vm/mterp/armv5/OP_MUL_LONG_2ADDR.S
@@ -0,0 +1,27 @@
+%verify "executed"
+    /*
+     * 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
+
diff --git a/vm/mterp/armv5/OP_NEG_DOUBLE.S b/vm/mterp/armv5/OP_NEG_DOUBLE.S
new file mode 100644
index 0000000..c8748a7
--- /dev/null
+++ b/vm/mterp/armv5/OP_NEG_DOUBLE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/unopWide.S" {"instr":"add     r1, r1, #0x80000000"}
diff --git a/vm/mterp/armv5/OP_NEG_FLOAT.S b/vm/mterp/armv5/OP_NEG_FLOAT.S
new file mode 100644
index 0000000..57e7bf3
--- /dev/null
+++ b/vm/mterp/armv5/OP_NEG_FLOAT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/unop.S" {"instr":"add     r0, r0, #0x80000000"}
diff --git a/vm/mterp/armv5/OP_NEG_INT.S b/vm/mterp/armv5/OP_NEG_INT.S
new file mode 100644
index 0000000..e3071c7
--- /dev/null
+++ b/vm/mterp/armv5/OP_NEG_INT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/unop.S" {"instr":"rsb     r0, r0, #0"}
diff --git a/vm/mterp/armv5/OP_NEG_LONG.S b/vm/mterp/armv5/OP_NEG_LONG.S
new file mode 100644
index 0000000..9c6c688
--- /dev/null
+++ b/vm/mterp/armv5/OP_NEG_LONG.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/unopWide.S" {"preinstr":"rsbs    r0, r0, #0", "instr":"rsc     r1, r1, #0"}
diff --git a/vm/mterp/armv5/OP_NEW_ARRAY.S b/vm/mterp/armv5/OP_NEW_ARRAY.S
new file mode 100644
index 0000000..55fc4f3
--- /dev/null
+++ b/vm/mterp/armv5/OP_NEW_ARRAY.S
@@ -0,0 +1,62 @@
+%verify "executed"
+%verify "negative array length"
+%verify "allocation fails"
+    /*
+     * 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     .L${opcode}_finish          @ resolved, continue
+    b       .L${opcode}_resolve         @ do resolve now
+%break
+
+
+    /*
+     * Resolve class.  (This is an uncommon case.)
+     *
+     *  r1 holds array length
+     *  r2 holds class ref CCCC
+     */
+.L${opcode}_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 ${opcode}_finish
+
+    /*
+     * Finish allocation.
+     *
+     *  r0 holds class
+     *  r1 holds array length
+     */
+.L${opcode}_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
+
diff --git a/vm/mterp/armv5/OP_NEW_INSTANCE.S b/vm/mterp/armv5/OP_NEW_INSTANCE.S
new file mode 100644
index 0000000..d1d2df6
--- /dev/null
+++ b/vm/mterp/armv5/OP_NEW_INSTANCE.S
@@ -0,0 +1,84 @@
+%verify "executed"
+%verify "class not resolved"
+%verify "class cannot be resolved"
+%verify "class not initialized"
+%verify "class fails to initialize"
+%verify "class already resolved/initialized"
+%verify "class is abstract or interface"
+%verify "allocation fails"
+    /*
+     * 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     .L${opcode}_resolve         @ no, resolve it now
+.L${opcode}_resolved:   @ r0=class
+    ldrb    r1, [r0, #offClassObject_status]    @ r1<- ClassStatus enum
+    cmp     r1, #CLASS_INITIALIZED      @ has class been initialized?
+    bne     .L${opcode}_needinit        @ no, init class now
+.L${opcode}_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     .L${opcode}_finish          @ concrete class, continue
+    b       .L${opcode}_abstract        @ fail
+%break
+
+    .balign 32                          @ minimize cache lines
+.L${opcode}_finish: @ r0=class
+    bl      dvmAllocObject              @ r0<- new object
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yes, handle the exception
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r3)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+    /*
+     * Class initialization required.
+     *
+     *  r0 holds class object
+     */
+.L${opcode}_needinit:
+    mov     r9, r0                      @ save r0
+    bl      dvmInitClass                @ initialize class
+    cmp     r0, #0                      @ check boolean result
+    mov     r0, r9                      @ restore r0
+    bne     .L${opcode}_initialized     @ success, continue
+    b       common_exceptionThrown      @ failed, deal with init exception
+
+    /*
+     * Resolution required.  This is the least-likely path.
+     *
+     *  r1 holds BBBB
+     */
+.L${opcode}_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     .L${opcode}_resolved        @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+    /*
+     * We can't instantiate an abstract class or interface, so throw an
+     * InstantiationError with the class descriptor as the message.
+     *
+     *  r0 holds class object
+     */
+.L${opcode}_abstract:
+    ldr     r1, [r0, #offClassObject_descriptor]
+    ldr     r0, .LstrInstantiationErrorPtr
+    bl      dvmThrowExceptionWithClassMessage
+    b       common_exceptionThrown
+
+.LstrInstantiationErrorPtr:
+    .word   .LstrInstantiationError
+
diff --git a/vm/mterp/armv5/OP_NOP.S b/vm/mterp/armv5/OP_NOP.S
new file mode 100644
index 0000000..1b72d3c
--- /dev/null
+++ b/vm/mterp/armv5/OP_NOP.S
@@ -0,0 +1,15 @@
+%verify "executed"
+    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
+
diff --git a/vm/mterp/armv5/OP_NOT_INT.S b/vm/mterp/armv5/OP_NOT_INT.S
new file mode 100644
index 0000000..c4fe18b
--- /dev/null
+++ b/vm/mterp/armv5/OP_NOT_INT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/unop.S" {"instr":"mvn     r0, r0"}
diff --git a/vm/mterp/armv5/OP_NOT_LONG.S b/vm/mterp/armv5/OP_NOT_LONG.S
new file mode 100644
index 0000000..46256cc
--- /dev/null
+++ b/vm/mterp/armv5/OP_NOT_LONG.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/unopWide.S" {"preinstr":"mvn     r0, r0", "instr":"mvn     r1, r1"}
diff --git a/vm/mterp/armv5/OP_OR_INT.S b/vm/mterp/armv5/OP_OR_INT.S
new file mode 100644
index 0000000..2ad6511
--- /dev/null
+++ b/vm/mterp/armv5/OP_OR_INT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop.S" {"instr":"orr     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_OR_INT_2ADDR.S b/vm/mterp/armv5/OP_OR_INT_2ADDR.S
new file mode 100644
index 0000000..6526175
--- /dev/null
+++ b/vm/mterp/armv5/OP_OR_INT_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop2addr.S" {"instr":"orr     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_OR_INT_LIT16.S b/vm/mterp/armv5/OP_OR_INT_LIT16.S
new file mode 100644
index 0000000..1c5e8e2
--- /dev/null
+++ b/vm/mterp/armv5/OP_OR_INT_LIT16.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopLit16.S" {"instr":"orr     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_OR_INT_LIT8.S b/vm/mterp/armv5/OP_OR_INT_LIT8.S
new file mode 100644
index 0000000..34be3e7
--- /dev/null
+++ b/vm/mterp/armv5/OP_OR_INT_LIT8.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopLit8.S" {"instr":"orr     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_OR_LONG.S b/vm/mterp/armv5/OP_OR_LONG.S
new file mode 100644
index 0000000..b7600c3
--- /dev/null
+++ b/vm/mterp/armv5/OP_OR_LONG.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide.S" {"preinstr":"orr     r0, r0, r2", "instr":"orr     r1, r1, r3"}
diff --git a/vm/mterp/armv5/OP_OR_LONG_2ADDR.S b/vm/mterp/armv5/OP_OR_LONG_2ADDR.S
new file mode 100644
index 0000000..7ea02a7
--- /dev/null
+++ b/vm/mterp/armv5/OP_OR_LONG_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide2addr.S" {"preinstr":"orr     r0, r0, r2", "instr":"orr     r1, r1, r3"}
diff --git a/vm/mterp/armv5/OP_PACKED_SWITCH.S b/vm/mterp/armv5/OP_PACKED_SWITCH.S
new file mode 100644
index 0000000..6fde05b
--- /dev/null
+++ b/vm/mterp/armv5/OP_PACKED_SWITCH.S
@@ -0,0 +1,26 @@
+%default { "func":"dvmInterpHandlePackedSwitch" }
+%verify executed
+    /*
+     * 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      $func                       @ 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
+
diff --git a/vm/mterp/armv5/OP_REM_DOUBLE.S b/vm/mterp/armv5/OP_REM_DOUBLE.S
new file mode 100644
index 0000000..0d5ec5f
--- /dev/null
+++ b/vm/mterp/armv5/OP_REM_DOUBLE.S
@@ -0,0 +1,3 @@
+%verify "executed"
+/* EABI doesn't define a double remainder function, but libm does */
+%include "armv5/binopWide.S" {"instr":"bl      fmod"}
diff --git a/vm/mterp/armv5/OP_REM_DOUBLE_2ADDR.S b/vm/mterp/armv5/OP_REM_DOUBLE_2ADDR.S
new file mode 100644
index 0000000..0308b4a
--- /dev/null
+++ b/vm/mterp/armv5/OP_REM_DOUBLE_2ADDR.S
@@ -0,0 +1,3 @@
+%verify "executed"
+/* EABI doesn't define a double remainder function, but libm does */
+%include "armv5/binopWide2addr.S" {"instr":"bl      fmod"}
diff --git a/vm/mterp/armv5/OP_REM_FLOAT.S b/vm/mterp/armv5/OP_REM_FLOAT.S
new file mode 100644
index 0000000..46c5fa8
--- /dev/null
+++ b/vm/mterp/armv5/OP_REM_FLOAT.S
@@ -0,0 +1,3 @@
+%verify "executed"
+/* EABI doesn't define a float remainder function, but libm does */
+%include "armv5/binop.S" {"instr":"bl      fmodf"}
diff --git a/vm/mterp/armv5/OP_REM_FLOAT_2ADDR.S b/vm/mterp/armv5/OP_REM_FLOAT_2ADDR.S
new file mode 100644
index 0000000..2cace1a
--- /dev/null
+++ b/vm/mterp/armv5/OP_REM_FLOAT_2ADDR.S
@@ -0,0 +1,3 @@
+%verify "executed"
+/* EABI doesn't define a float remainder function, but libm does */
+%include "armv5/binop2addr.S" {"instr":"bl      fmodf"}
diff --git a/vm/mterp/armv5/OP_REM_INT.S b/vm/mterp/armv5/OP_REM_INT.S
new file mode 100644
index 0000000..cd6d6fd
--- /dev/null
+++ b/vm/mterp/armv5/OP_REM_INT.S
@@ -0,0 +1,3 @@
+%verify "executed"
+/* idivmod returns quotient in r0 and remainder in r1 */
+%include "armv5/binop.S" {"instr":"bl      __aeabi_idivmod", "result":"r1", "chkzero":"1"}
diff --git a/vm/mterp/armv5/OP_REM_INT_2ADDR.S b/vm/mterp/armv5/OP_REM_INT_2ADDR.S
new file mode 100644
index 0000000..2dad3ca
--- /dev/null
+++ b/vm/mterp/armv5/OP_REM_INT_2ADDR.S
@@ -0,0 +1,3 @@
+%verify "executed"
+/* idivmod returns quotient in r0 and remainder in r1 */
+%include "armv5/binop2addr.S" {"instr":"bl      __aeabi_idivmod", "result":"r1", "chkzero":"1"}
diff --git a/vm/mterp/armv5/OP_REM_INT_LIT16.S b/vm/mterp/armv5/OP_REM_INT_LIT16.S
new file mode 100644
index 0000000..c9f5a87
--- /dev/null
+++ b/vm/mterp/armv5/OP_REM_INT_LIT16.S
@@ -0,0 +1,3 @@
+%verify "executed"
+/* idivmod returns quotient in r0 and remainder in r1 */
+%include "armv5/binopLit16.S" {"instr":"bl      __aeabi_idivmod", "result":"r1", "chkzero":"1"}
diff --git a/vm/mterp/armv5/OP_REM_INT_LIT8.S b/vm/mterp/armv5/OP_REM_INT_LIT8.S
new file mode 100644
index 0000000..aeddaba
--- /dev/null
+++ b/vm/mterp/armv5/OP_REM_INT_LIT8.S
@@ -0,0 +1,3 @@
+%verify "executed"
+/* idivmod returns quotient in r0 and remainder in r1 */
+%include "armv5/binopLit8.S" {"instr":"bl      __aeabi_idivmod", "result":"r1", "chkzero":"1"}
diff --git a/vm/mterp/armv5/OP_REM_LONG.S b/vm/mterp/armv5/OP_REM_LONG.S
new file mode 100644
index 0000000..0687b43
--- /dev/null
+++ b/vm/mterp/armv5/OP_REM_LONG.S
@@ -0,0 +1,3 @@
+%verify "executed"
+/* ldivmod returns quotient in r0/r1 and remainder in r2/r3 */
+%include "armv5/binopWide.S" {"instr":"bl      __aeabi_ldivmod", "result0":"r2", "result1":"r3", "chkzero":"1"}
diff --git a/vm/mterp/armv5/OP_REM_LONG_2ADDR.S b/vm/mterp/armv5/OP_REM_LONG_2ADDR.S
new file mode 100644
index 0000000..d4a64f8
--- /dev/null
+++ b/vm/mterp/armv5/OP_REM_LONG_2ADDR.S
@@ -0,0 +1,3 @@
+%verify "executed"
+/* ldivmod returns quotient in r0/r1 and remainder in r2/r3 */
+%include "armv5/binopWide2addr.S" {"instr":"bl      __aeabi_ldivmod", "result0":"r2", "result1":"r3", "chkzero":"1"}
diff --git a/vm/mterp/armv5/OP_RETURN.S b/vm/mterp/armv5/OP_RETURN.S
new file mode 100644
index 0000000..f76a223
--- /dev/null
+++ b/vm/mterp/armv5/OP_RETURN.S
@@ -0,0 +1,13 @@
+%verify "executed"
+    /*
+     * 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
+
diff --git a/vm/mterp/armv5/OP_RETURN_OBJECT.S b/vm/mterp/armv5/OP_RETURN_OBJECT.S
new file mode 100644
index 0000000..e4c5c49
--- /dev/null
+++ b/vm/mterp/armv5/OP_RETURN_OBJECT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_RETURN.S"
diff --git a/vm/mterp/armv5/OP_RETURN_VOID.S b/vm/mterp/armv5/OP_RETURN_VOID.S
new file mode 100644
index 0000000..647002d
--- /dev/null
+++ b/vm/mterp/armv5/OP_RETURN_VOID.S
@@ -0,0 +1,3 @@
+%verify "executed"
+    b       common_returnFromMethod
+
diff --git a/vm/mterp/armv5/OP_RETURN_WIDE.S b/vm/mterp/armv5/OP_RETURN_WIDE.S
new file mode 100644
index 0000000..2d30792
--- /dev/null
+++ b/vm/mterp/armv5/OP_RETURN_WIDE.S
@@ -0,0 +1,13 @@
+%verify "executed"
+    /*
+     * 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
+
diff --git a/vm/mterp/armv5/OP_RSUB_INT.S b/vm/mterp/armv5/OP_RSUB_INT.S
new file mode 100644
index 0000000..dbf8056
--- /dev/null
+++ b/vm/mterp/armv5/OP_RSUB_INT.S
@@ -0,0 +1,3 @@
+%verify "executed"
+/* this op is "rsub-int", but can be thought of as "rsub-int/lit16" */
+%include "armv5/binopLit16.S" {"instr":"rsb     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_RSUB_INT_LIT8.S b/vm/mterp/armv5/OP_RSUB_INT_LIT8.S
new file mode 100644
index 0000000..3bed616
--- /dev/null
+++ b/vm/mterp/armv5/OP_RSUB_INT_LIT8.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopLit8.S" {"instr":"rsb     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_SGET.S b/vm/mterp/armv5/OP_SGET.S
new file mode 100644
index 0000000..17ab26a
--- /dev/null
+++ b/vm/mterp/armv5/OP_SGET.S
@@ -0,0 +1,38 @@
+%verify "executed"
+%verify "field already resolved"
+%verify "field not yet resolved"
+%verify "field cannot be resolved"
+    /*
+     * 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     .L${opcode}_resolve         @ yes, do resolve
+.L${opcode}_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+%break
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.L${opcode}_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     .L${opcode}_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
diff --git a/vm/mterp/armv5/OP_SGET_BOOLEAN.S b/vm/mterp/armv5/OP_SGET_BOOLEAN.S
new file mode 100644
index 0000000..0cbd28f
--- /dev/null
+++ b/vm/mterp/armv5/OP_SGET_BOOLEAN.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_SGET.S"
diff --git a/vm/mterp/armv5/OP_SGET_BYTE.S b/vm/mterp/armv5/OP_SGET_BYTE.S
new file mode 100644
index 0000000..0cbd28f
--- /dev/null
+++ b/vm/mterp/armv5/OP_SGET_BYTE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_SGET.S"
diff --git a/vm/mterp/armv5/OP_SGET_CHAR.S b/vm/mterp/armv5/OP_SGET_CHAR.S
new file mode 100644
index 0000000..0cbd28f
--- /dev/null
+++ b/vm/mterp/armv5/OP_SGET_CHAR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_SGET.S"
diff --git a/vm/mterp/armv5/OP_SGET_OBJECT.S b/vm/mterp/armv5/OP_SGET_OBJECT.S
new file mode 100644
index 0000000..0cbd28f
--- /dev/null
+++ b/vm/mterp/armv5/OP_SGET_OBJECT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_SGET.S"
diff --git a/vm/mterp/armv5/OP_SGET_SHORT.S b/vm/mterp/armv5/OP_SGET_SHORT.S
new file mode 100644
index 0000000..0cbd28f
--- /dev/null
+++ b/vm/mterp/armv5/OP_SGET_SHORT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_SGET.S"
diff --git a/vm/mterp/armv5/OP_SGET_WIDE.S b/vm/mterp/armv5/OP_SGET_WIDE.S
new file mode 100644
index 0000000..1f93a2f
--- /dev/null
+++ b/vm/mterp/armv5/OP_SGET_WIDE.S
@@ -0,0 +1,37 @@
+%verify "executed"
+%verify "field already resolved"
+%verify "field not yet resolved"
+%verify "field cannot be resolved"
+    /*
+     * 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     .L${opcode}_resolve         @ yes, do resolve
+.L${opcode}_finish:
+    mov     r1, rINST, lsr #8           @ r1<- AA
+    ldrd    r2, [r0, #offStaticField_value] @ r2/r3<- field value (aligned)
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[AA]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    stmia   r1, {r2-r3}                 @ vAA/vAA+1<- r2/r3
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+%break
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.L${opcode}_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     .L${opcode}_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
diff --git a/vm/mterp/armv5/OP_SHL_INT.S b/vm/mterp/armv5/OP_SHL_INT.S
new file mode 100644
index 0000000..b8fdfcc
--- /dev/null
+++ b/vm/mterp/armv5/OP_SHL_INT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop.S" {"preinstr":"and     r1, r1, #31", "instr":"mov     r0, r0, asl r1"}
diff --git a/vm/mterp/armv5/OP_SHL_INT_2ADDR.S b/vm/mterp/armv5/OP_SHL_INT_2ADDR.S
new file mode 100644
index 0000000..753e95b
--- /dev/null
+++ b/vm/mterp/armv5/OP_SHL_INT_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop2addr.S" {"preinstr":"and     r1, r1, #31", "instr":"mov     r0, r0, asl r1"}
diff --git a/vm/mterp/armv5/OP_SHL_INT_LIT8.S b/vm/mterp/armv5/OP_SHL_INT_LIT8.S
new file mode 100644
index 0000000..8843197
--- /dev/null
+++ b/vm/mterp/armv5/OP_SHL_INT_LIT8.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopLit8.S" {"preinstr":"and     r1, r1, #31", "instr":"mov     r0, r0, asl r1"}
diff --git a/vm/mterp/armv5/OP_SHL_LONG.S b/vm/mterp/armv5/OP_SHL_LONG.S
new file mode 100644
index 0000000..3510565
--- /dev/null
+++ b/vm/mterp/armv5/OP_SHL_LONG.S
@@ -0,0 +1,33 @@
+%verify "executed"
+    /*
+     * 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       .L${opcode}_finish
+%break
+
+.L${opcode}_finish:
+    mov     r0, r0, asl r2              @  r0<- r0 << r2
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
diff --git a/vm/mterp/armv5/OP_SHL_LONG_2ADDR.S b/vm/mterp/armv5/OP_SHL_LONG_2ADDR.S
new file mode 100644
index 0000000..93f8260
--- /dev/null
+++ b/vm/mterp/armv5/OP_SHL_LONG_2ADDR.S
@@ -0,0 +1,29 @@
+%verify "executed"
+    /*
+     * Long integer shift, 2addr version.  vA is 64-bit value/result, vB is
+     * 32-bit shift distance.
+     */
+    /* shl-long/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r2, r3)                    @ r2<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    and     r2, r2, #63                 @ r2<- r2 & 0x3f
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+
+    mov     r1, r1, asl r2              @  r1<- r1 << r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r1, r1, r0, lsr r3          @  r1<- r1 | (r0 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    movpl   r1, r0, asl ip              @  if r2 >= 32, r1<- r0 << (r2-32)
+    mov     r0, r0, asl r2              @  r0<- r0 << r2
+    b       .L${opcode}_finish
+%break
+
+.L${opcode}_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
+
diff --git a/vm/mterp/armv5/OP_SHR_INT.S b/vm/mterp/armv5/OP_SHR_INT.S
new file mode 100644
index 0000000..7f35381
--- /dev/null
+++ b/vm/mterp/armv5/OP_SHR_INT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop.S" {"preinstr":"and     r1, r1, #31", "instr":"mov     r0, r0, asr r1"}
diff --git a/vm/mterp/armv5/OP_SHR_INT_2ADDR.S b/vm/mterp/armv5/OP_SHR_INT_2ADDR.S
new file mode 100644
index 0000000..0632ecc
--- /dev/null
+++ b/vm/mterp/armv5/OP_SHR_INT_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop2addr.S" {"preinstr":"and     r1, r1, #31", "instr":"mov     r0, r0, asr r1"}
diff --git a/vm/mterp/armv5/OP_SHR_INT_LIT8.S b/vm/mterp/armv5/OP_SHR_INT_LIT8.S
new file mode 100644
index 0000000..df3126c
--- /dev/null
+++ b/vm/mterp/armv5/OP_SHR_INT_LIT8.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopLit8.S" {"preinstr":"and     r1, r1, #31", "instr":"mov     r0, r0, asr r1"}
diff --git a/vm/mterp/armv5/OP_SHR_LONG.S b/vm/mterp/armv5/OP_SHR_LONG.S
new file mode 100644
index 0000000..c7f10aa
--- /dev/null
+++ b/vm/mterp/armv5/OP_SHR_LONG.S
@@ -0,0 +1,33 @@
+%verify "executed"
+    /*
+     * 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       .L${opcode}_finish
+%break
+
+.L${opcode}_finish:
+    mov     r1, r1, asr r2              @  r1<- r1 >> r2
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
diff --git a/vm/mterp/armv5/OP_SHR_LONG_2ADDR.S b/vm/mterp/armv5/OP_SHR_LONG_2ADDR.S
new file mode 100644
index 0000000..188017f
--- /dev/null
+++ b/vm/mterp/armv5/OP_SHR_LONG_2ADDR.S
@@ -0,0 +1,29 @@
+%verify "executed"
+    /*
+     * Long integer shift, 2addr version.  vA is 64-bit value/result, vB is
+     * 32-bit shift distance.
+     */
+    /* shr-long/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r2, r3)                    @ r2<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    and     r2, r2, #63                 @ r2<- r2 & 0x3f
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+
+    mov     r0, r0, lsr r2              @  r0<- r2 >> r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r0, r0, r1, asl r3          @  r0<- r0 | (r1 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    movpl   r0, r1, asr ip              @  if r2 >= 32, r0<-r1 >> (r2-32)
+    mov     r1, r1, asr r2              @  r1<- r1 >> r2
+    b       .L${opcode}_finish
+%break
+
+.L${opcode}_finish:
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
diff --git a/vm/mterp/armv5/OP_SPARSE_SWITCH.S b/vm/mterp/armv5/OP_SPARSE_SWITCH.S
new file mode 100644
index 0000000..44bc72e
--- /dev/null
+++ b/vm/mterp/armv5/OP_SPARSE_SWITCH.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_PACKED_SWITCH.S" { "func":"dvmInterpHandleSparseSwitch" }
diff --git a/vm/mterp/armv5/OP_SPUT.S b/vm/mterp/armv5/OP_SPUT.S
new file mode 100644
index 0000000..5236614
--- /dev/null
+++ b/vm/mterp/armv5/OP_SPUT.S
@@ -0,0 +1,38 @@
+%verify "executed"
+%verify "field already resolved"
+%verify "field not yet resolved"
+%verify "field cannot be resolved"
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .L${opcode}_resolve         @ yes, do resolve
+.L${opcode}_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+%break
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.L${opcode}_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     .L${opcode}_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
diff --git a/vm/mterp/armv5/OP_SPUT_BOOLEAN.S b/vm/mterp/armv5/OP_SPUT_BOOLEAN.S
new file mode 100644
index 0000000..33b968b
--- /dev/null
+++ b/vm/mterp/armv5/OP_SPUT_BOOLEAN.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_SPUT.S"
diff --git a/vm/mterp/armv5/OP_SPUT_BYTE.S b/vm/mterp/armv5/OP_SPUT_BYTE.S
new file mode 100644
index 0000000..33b968b
--- /dev/null
+++ b/vm/mterp/armv5/OP_SPUT_BYTE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_SPUT.S"
diff --git a/vm/mterp/armv5/OP_SPUT_CHAR.S b/vm/mterp/armv5/OP_SPUT_CHAR.S
new file mode 100644
index 0000000..33b968b
--- /dev/null
+++ b/vm/mterp/armv5/OP_SPUT_CHAR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_SPUT.S"
diff --git a/vm/mterp/armv5/OP_SPUT_OBJECT.S b/vm/mterp/armv5/OP_SPUT_OBJECT.S
new file mode 100644
index 0000000..33b968b
--- /dev/null
+++ b/vm/mterp/armv5/OP_SPUT_OBJECT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_SPUT.S"
diff --git a/vm/mterp/armv5/OP_SPUT_SHORT.S b/vm/mterp/armv5/OP_SPUT_SHORT.S
new file mode 100644
index 0000000..33b968b
--- /dev/null
+++ b/vm/mterp/armv5/OP_SPUT_SHORT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/OP_SPUT.S"
diff --git a/vm/mterp/armv5/OP_SPUT_WIDE.S b/vm/mterp/armv5/OP_SPUT_WIDE.S
new file mode 100644
index 0000000..a7bc5f2
--- /dev/null
+++ b/vm/mterp/armv5/OP_SPUT_WIDE.S
@@ -0,0 +1,38 @@
+%verify "executed"
+%verify "field already resolved"
+%verify "field not yet resolved"
+%verify "field cannot be resolved"
+    /*
+     * 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     .L${opcode}_resolve         @ yes, do resolve
+.L${opcode}_finish: @ field ptr in r0, AA in r9
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldmia   r9, {r2-r3}                 @ r2/r3<- vAA/vAA+1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strd    r2, [r0, #offStaticField_value] @ field<- vAA/vAA+1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+%break
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     *  r9: &fp[AA]
+     */
+.L${opcode}_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     .L${opcode}_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
diff --git a/vm/mterp/armv5/OP_SUB_DOUBLE.S b/vm/mterp/armv5/OP_SUB_DOUBLE.S
new file mode 100644
index 0000000..5aff638
--- /dev/null
+++ b/vm/mterp/armv5/OP_SUB_DOUBLE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide.S" {"instr":"bl      __aeabi_dsub"}
diff --git a/vm/mterp/armv5/OP_SUB_DOUBLE_2ADDR.S b/vm/mterp/armv5/OP_SUB_DOUBLE_2ADDR.S
new file mode 100644
index 0000000..7351d9e
--- /dev/null
+++ b/vm/mterp/armv5/OP_SUB_DOUBLE_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide2addr.S" {"instr":"bl      __aeabi_dsub"}
diff --git a/vm/mterp/armv5/OP_SUB_FLOAT.S b/vm/mterp/armv5/OP_SUB_FLOAT.S
new file mode 100644
index 0000000..c902fa2
--- /dev/null
+++ b/vm/mterp/armv5/OP_SUB_FLOAT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop.S" {"instr":"bl      __aeabi_fsub"}
diff --git a/vm/mterp/armv5/OP_SUB_FLOAT_2ADDR.S b/vm/mterp/armv5/OP_SUB_FLOAT_2ADDR.S
new file mode 100644
index 0000000..994c820
--- /dev/null
+++ b/vm/mterp/armv5/OP_SUB_FLOAT_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop2addr.S" {"instr":"bl      __aeabi_fsub"}
diff --git a/vm/mterp/armv5/OP_SUB_INT.S b/vm/mterp/armv5/OP_SUB_INT.S
new file mode 100644
index 0000000..cbd865e
--- /dev/null
+++ b/vm/mterp/armv5/OP_SUB_INT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop.S" {"instr":"sub     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_SUB_INT_2ADDR.S b/vm/mterp/armv5/OP_SUB_INT_2ADDR.S
new file mode 100644
index 0000000..93a32ec
--- /dev/null
+++ b/vm/mterp/armv5/OP_SUB_INT_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop2addr.S" {"instr":"sub     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_SUB_LONG.S b/vm/mterp/armv5/OP_SUB_LONG.S
new file mode 100644
index 0000000..34b0a4d
--- /dev/null
+++ b/vm/mterp/armv5/OP_SUB_LONG.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide.S" {"preinstr":"subs    r0, r0, r2", "instr":"sbc     r1, r1, r3"}
diff --git a/vm/mterp/armv5/OP_SUB_LONG_2ADDR.S b/vm/mterp/armv5/OP_SUB_LONG_2ADDR.S
new file mode 100644
index 0000000..ee5bf9e
--- /dev/null
+++ b/vm/mterp/armv5/OP_SUB_LONG_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide2addr.S" {"preinstr":"subs    r0, r0, r2", "instr":"sbc     r1, r1, r3"}
diff --git a/vm/mterp/armv5/OP_THROW.S b/vm/mterp/armv5/OP_THROW.S
new file mode 100644
index 0000000..c8c5c36
--- /dev/null
+++ b/vm/mterp/armv5/OP_THROW.S
@@ -0,0 +1,15 @@
+%verify "executed"
+%verify "exception for null object"
+    /*
+     * 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
+
diff --git a/vm/mterp/armv5/OP_UNUSED_3E.S b/vm/mterp/armv5/OP_UNUSED_3E.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_3E.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_3F.S b/vm/mterp/armv5/OP_UNUSED_3F.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_3F.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_40.S b/vm/mterp/armv5/OP_UNUSED_40.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_40.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_41.S b/vm/mterp/armv5/OP_UNUSED_41.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_41.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_42.S b/vm/mterp/armv5/OP_UNUSED_42.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_42.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_43.S b/vm/mterp/armv5/OP_UNUSED_43.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_43.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_73.S b/vm/mterp/armv5/OP_UNUSED_73.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_73.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_79.S b/vm/mterp/armv5/OP_UNUSED_79.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_79.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_7A.S b/vm/mterp/armv5/OP_UNUSED_7A.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_7A.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_E3.S b/vm/mterp/armv5/OP_UNUSED_E3.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_E3.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_E4.S b/vm/mterp/armv5/OP_UNUSED_E4.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_E4.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_E5.S b/vm/mterp/armv5/OP_UNUSED_E5.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_E5.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_E6.S b/vm/mterp/armv5/OP_UNUSED_E6.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_E6.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_E7.S b/vm/mterp/armv5/OP_UNUSED_E7.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_E7.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_E8.S b/vm/mterp/armv5/OP_UNUSED_E8.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_E8.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_E9.S b/vm/mterp/armv5/OP_UNUSED_E9.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_E9.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_EA.S b/vm/mterp/armv5/OP_UNUSED_EA.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_EA.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_EB.S b/vm/mterp/armv5/OP_UNUSED_EB.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_EB.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_EC.S b/vm/mterp/armv5/OP_UNUSED_EC.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_EC.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_ED.S b/vm/mterp/armv5/OP_UNUSED_ED.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_ED.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_EF.S b/vm/mterp/armv5/OP_UNUSED_EF.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_EF.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_F1.S b/vm/mterp/armv5/OP_UNUSED_F1.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_F1.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_FC.S b/vm/mterp/armv5/OP_UNUSED_FC.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_FC.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_FD.S b/vm/mterp/armv5/OP_UNUSED_FD.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_FD.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_FE.S b/vm/mterp/armv5/OP_UNUSED_FE.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_FE.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_UNUSED_FF.S b/vm/mterp/armv5/OP_UNUSED_FF.S
new file mode 100644
index 0000000..9fdef1d
--- /dev/null
+++ b/vm/mterp/armv5/OP_UNUSED_FF.S
@@ -0,0 +1 @@
+%include "armv5/unused.S"
diff --git a/vm/mterp/armv5/OP_USHR_INT.S b/vm/mterp/armv5/OP_USHR_INT.S
new file mode 100644
index 0000000..dc06b4c
--- /dev/null
+++ b/vm/mterp/armv5/OP_USHR_INT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop.S" {"preinstr":"and     r1, r1, #31", "instr":"mov     r0, r0, lsr r1"}
diff --git a/vm/mterp/armv5/OP_USHR_INT_2ADDR.S b/vm/mterp/armv5/OP_USHR_INT_2ADDR.S
new file mode 100644
index 0000000..f4cba40
--- /dev/null
+++ b/vm/mterp/armv5/OP_USHR_INT_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop2addr.S" {"preinstr":"and     r1, r1, #31", "instr":"mov     r0, r0, lsr r1"}
diff --git a/vm/mterp/armv5/OP_USHR_INT_LIT8.S b/vm/mterp/armv5/OP_USHR_INT_LIT8.S
new file mode 100644
index 0000000..bc08d54
--- /dev/null
+++ b/vm/mterp/armv5/OP_USHR_INT_LIT8.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopLit8.S" {"preinstr":"and     r1, r1, #31", "instr":"mov     r0, r0, lsr r1"}
diff --git a/vm/mterp/armv5/OP_USHR_LONG.S b/vm/mterp/armv5/OP_USHR_LONG.S
new file mode 100644
index 0000000..fc14b57
--- /dev/null
+++ b/vm/mterp/armv5/OP_USHR_LONG.S
@@ -0,0 +1,33 @@
+%verify "executed"
+    /*
+     * 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       .L${opcode}_finish
+%break
+
+.L${opcode}_finish:
+    mov     r1, r1, lsr r2              @  r1<- r1 >>> r2
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
diff --git a/vm/mterp/armv5/OP_USHR_LONG_2ADDR.S b/vm/mterp/armv5/OP_USHR_LONG_2ADDR.S
new file mode 100644
index 0000000..18167e3
--- /dev/null
+++ b/vm/mterp/armv5/OP_USHR_LONG_2ADDR.S
@@ -0,0 +1,29 @@
+%verify "executed"
+    /*
+     * Long integer shift, 2addr version.  vA is 64-bit value/result, vB is
+     * 32-bit shift distance.
+     */
+    /* ushr-long/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r2, r3)                    @ r2<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    and     r2, r2, #63                 @ r2<- r2 & 0x3f
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+
+    mov     r0, r0, lsr r2              @  r0<- r2 >> r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r0, r0, r1, asl r3          @  r0<- r0 | (r1 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    movpl   r0, r1, lsr ip              @  if r2 >= 32, r0<-r1 >>> (r2-32)
+    mov     r1, r1, lsr r2              @  r1<- r1 >>> r2
+    b       .L${opcode}_finish
+%break
+
+.L${opcode}_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
+
diff --git a/vm/mterp/armv5/OP_XOR_INT.S b/vm/mterp/armv5/OP_XOR_INT.S
new file mode 100644
index 0000000..101c3b2
--- /dev/null
+++ b/vm/mterp/armv5/OP_XOR_INT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop.S" {"instr":"eor     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_XOR_INT_2ADDR.S b/vm/mterp/armv5/OP_XOR_INT_2ADDR.S
new file mode 100644
index 0000000..70c3e36
--- /dev/null
+++ b/vm/mterp/armv5/OP_XOR_INT_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binop2addr.S" {"instr":"eor     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_XOR_INT_LIT16.S b/vm/mterp/armv5/OP_XOR_INT_LIT16.S
new file mode 100644
index 0000000..3e12b63
--- /dev/null
+++ b/vm/mterp/armv5/OP_XOR_INT_LIT16.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopLit16.S" {"instr":"eor     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_XOR_INT_LIT8.S b/vm/mterp/armv5/OP_XOR_INT_LIT8.S
new file mode 100644
index 0000000..063b2cb
--- /dev/null
+++ b/vm/mterp/armv5/OP_XOR_INT_LIT8.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopLit8.S" {"instr":"eor     r0, r0, r1"}
diff --git a/vm/mterp/armv5/OP_XOR_LONG.S b/vm/mterp/armv5/OP_XOR_LONG.S
new file mode 100644
index 0000000..1ecdaad
--- /dev/null
+++ b/vm/mterp/armv5/OP_XOR_LONG.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide.S" {"preinstr":"eor     r0, r0, r2", "instr":"eor     r1, r1, r3"}
diff --git a/vm/mterp/armv5/OP_XOR_LONG_2ADDR.S b/vm/mterp/armv5/OP_XOR_LONG_2ADDR.S
new file mode 100644
index 0000000..5423e6a
--- /dev/null
+++ b/vm/mterp/armv5/OP_XOR_LONG_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "armv5/binopWide2addr.S" {"preinstr":"eor     r0, r0, r2", "instr":"eor     r1, r1, r3"}
diff --git a/vm/mterp/armv5/bincmp.S b/vm/mterp/armv5/bincmp.S
new file mode 100644
index 0000000..9b574a3
--- /dev/null
+++ b/vm/mterp/armv5/bincmp.S
@@ -0,0 +1,25 @@
+%verify "branch taken"
+%verify "branch not taken"
+    /*
+     * 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)
+    b${revcmp}  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
+
diff --git a/vm/mterp/armv5/binop.S b/vm/mterp/armv5/binop.S
new file mode 100644
index 0000000..d33ce01
--- /dev/null
+++ b/vm/mterp/armv5/binop.S
@@ -0,0 +1,34 @@
+%default {"preinstr":"", "result":"r0", "chkzero":"0"}
+    /*
+     * 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.
+     *
+     * 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 $chkzero
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    $preinstr                           @ optional op; may set condition codes
+    $instr                              @ $result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG($result, r9)               @ vAA<- $result
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
diff --git a/vm/mterp/armv5/binop2addr.S b/vm/mterp/armv5/binop2addr.S
new file mode 100644
index 0000000..fc170a0
--- /dev/null
+++ b/vm/mterp/armv5/binop2addr.S
@@ -0,0 +1,34 @@
+%default {"preinstr":"", "result":"r0", "chkzero":"0"}
+    /*
+     * 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 $chkzero
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    $preinstr                           @ optional op; may set condition codes
+    $instr                              @ $result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG($result, r9)               @ vAA<- $result
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
diff --git a/vm/mterp/armv5/binopLit16.S b/vm/mterp/armv5/binopLit16.S
new file mode 100644
index 0000000..3cb1875
--- /dev/null
+++ b/vm/mterp/armv5/binopLit16.S
@@ -0,0 +1,31 @@
+%default {"result":"r0", "chkzero":"0"}
+    /*
+     * 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 $chkzero
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    $instr                              @ $result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG($result, r9)               @ vAA<- $result
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
diff --git a/vm/mterp/armv5/binopLit8.S b/vm/mterp/armv5/binopLit8.S
new file mode 100644
index 0000000..4406259
--- /dev/null
+++ b/vm/mterp/armv5/binopLit8.S
@@ -0,0 +1,33 @@
+%default {"preinstr":"", "result":"r0", "chkzero":"0"}
+    /*
+     * 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 $chkzero
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    $preinstr                           @ optional op; may set condition codes
+    $instr                              @ $result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG($result, r9)               @ vAA<- $result
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
diff --git a/vm/mterp/armv5/binopWide.S b/vm/mterp/armv5/binopWide.S
new file mode 100644
index 0000000..87c5f07
--- /dev/null
+++ b/vm/mterp/armv5/binopWide.S
@@ -0,0 +1,39 @@
+%default {"preinstr":"", "result0":"r0", "result1":"r1", "chkzero":"0"}
+    /*
+     * 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 $chkzero
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    $preinstr                           @ optional op; may set condition codes
+    $instr                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {$result0,$result1}     @ vAA/vAA+1<- $result0/$result1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
diff --git a/vm/mterp/armv5/binopWide2addr.S b/vm/mterp/armv5/binopWide2addr.S
new file mode 100644
index 0000000..98c6c98
--- /dev/null
+++ b/vm/mterp/armv5/binopWide2addr.S
@@ -0,0 +1,36 @@
+%default {"preinstr":"", "result0":"r0", "result1":"r1", "chkzero":"0"}
+    /*
+     * 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 $chkzero
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    $preinstr                           @ optional op; may set condition codes
+    $instr                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {$result0,$result1}     @ vAA/vAA+1<- $result0/$result1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
diff --git a/vm/mterp/armv5/debug.c b/vm/mterp/armv5/debug.c
new file mode 100644
index 0000000..301e27a
--- /dev/null
+++ b/vm/mterp/armv5/debug.c
@@ -0,0 +1,79 @@
+#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/armv5/entry.S b/vm/mterp/armv5/entry.S
new file mode 100644
index 0000000..a1e3b4e
--- /dev/null
+++ b/vm/mterp/armv5/entry.S
@@ -0,0 +1,118 @@
+/*
+ * 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   sp!, {r4-r10,fp,pc}             @ restore 9 regs
+
+
+/*
+ * String references.
+ */
+strBadEntryPoint:
+    .word   .LstrBadEntryPoint
+
diff --git a/vm/mterp/armv5/footer.S b/vm/mterp/armv5/footer.S
new file mode 100644
index 0000000..5efb24a
--- /dev/null
+++ b/vm/mterp/armv5/footer.S
@@ -0,0 +1,714 @@
+/*
+ * ===========================================================================
+ *  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
+
+    mov     pc, 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
+
+    @ 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)
+    ldr     rFP, [sp, #4]               @ retrieve the updated rFP
+    cmp     r0, #0                      @ is catchRelPc < 0?
+    add     sp, sp, #8                  @ restore stack
+    bmi     .LnotCaughtLocally
+
+    /* fix stack overflow if necessary; must preserve r0 */
+    ldrb    r1, [r10, #offThread_stackOverflowed]
+    cmp     r1, #0                      @ did we overflow earlier?
+    beq     1f                          @ no, skip ahead
+    mov     r9, r0                      @ r9<- r0 (save it)
+    mov     r0, r10                     @ r0<- self
+    bl      dvmCleanupStackOverflow     @ call(self)
+    mov     r0, r9                      @ r0<- r9 (restore it)
+    ldr     r9, [r10, #offThread_exception] @ r9<- self->exception
+1:
+
+    /* 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, pc}
+    .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, pc}
+
+/*
+ * 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, pc}
+
+    /*
+     * 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, pc}
+
+/*
+ * 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, pc}
+
+/*
+ * 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, pc}
+
+/*
+ * 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, pc}
+    .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 '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/armv5/header.S b/vm/mterp/armv5/header.S
new file mode 100644
index 0000000..460930d
--- /dev/null
+++ b/vm/mterp/armv5/header.S
@@ -0,0 +1,168 @@
+/*
+ * 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.
+ */
+#define GOTO_OPCODE(_reg)       add     pc, rIBASE, _reg, lsl #${handler_size_bits}
+
+/*
+ * Get/set the 32-bit value from a Dalvik register.
+ */
+#define GET_VREG(_reg, _vreg)   ldr     _reg, [rFP, _vreg, lsl #2]
+#define SET_VREG(_reg, _vreg)   str     _reg, [rFP, _vreg, lsl #2]
+
+/*
+ * This is a #include, not a %include, because we want the C pre-processor
+ * to expand the macros into assembler assignment statements.
+ */
+#include "../common/asm-constants.h"
+
diff --git a/vm/mterp/armv5/stub.S b/vm/mterp/armv5/stub.S
new file mode 100644
index 0000000..7607f09
--- /dev/null
+++ b/vm/mterp/armv5/stub.S
@@ -0,0 +1,9 @@
+    /* (stub) */
+    SAVE_PC_FP_TO_GLUE()            @ only need to export these two
+    mov     r0, rGLUE               @ glue is first arg to function
+    bl      dvmMterp_${opcode}      @ call
+    LOAD_PC_FP_FROM_GLUE()          @ retrieve updated values
+    FETCH_INST()                    @ load next instruction from rPC
+    GET_INST_OPCODE(ip)             @ ...trim down to just the opcode
+    GOTO_OPCODE(ip)                 @ ...and jump to the handler
+
diff --git a/vm/mterp/armv5/unop.S b/vm/mterp/armv5/unop.S
new file mode 100644
index 0000000..12d8206
--- /dev/null
+++ b/vm/mterp/armv5/unop.S
@@ -0,0 +1,21 @@
+%default {"preinstr":""}
+    /*
+     * 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
+    $preinstr                           @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    $instr                              @ 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 */
diff --git a/vm/mterp/armv5/unopNarrower.S b/vm/mterp/armv5/unopNarrower.S
new file mode 100644
index 0000000..f1ad902
--- /dev/null
+++ b/vm/mterp/armv5/unopNarrower.S
@@ -0,0 +1,24 @@
+%default {"preinstr":""}
+    /*
+     * 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
+    $preinstr                           @ optional op; may set condition codes
+    $instr                              @ 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 */
diff --git a/vm/mterp/armv5/unopWide.S b/vm/mterp/armv5/unopWide.S
new file mode 100644
index 0000000..e097317
--- /dev/null
+++ b/vm/mterp/armv5/unopWide.S
@@ -0,0 +1,23 @@
+%default {"preinstr":""}
+    /*
+     * 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
+    $preinstr                           @ optional op; may set condition codes
+    $instr                              @ 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 */
+
diff --git a/vm/mterp/armv5/unopWider.S b/vm/mterp/armv5/unopWider.S
new file mode 100644
index 0000000..df1baea
--- /dev/null
+++ b/vm/mterp/armv5/unopWider.S
@@ -0,0 +1,21 @@
+%default {"preinstr":""}
+    /*
+     * 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]
+    $preinstr                           @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    $instr                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vA/vA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-11 instructions */
diff --git a/vm/mterp/armv5/unused.S b/vm/mterp/armv5/unused.S
new file mode 100644
index 0000000..1c82919
--- /dev/null
+++ b/vm/mterp/armv5/unused.S
@@ -0,0 +1,2 @@
+    bl      common_abort
+
diff --git a/vm/mterp/armv5/zcmp.S b/vm/mterp/armv5/zcmp.S
new file mode 100644
index 0000000..7942632
--- /dev/null
+++ b/vm/mterp/armv5/zcmp.S
@@ -0,0 +1,22 @@
+%verify "branch taken"
+%verify "branch not taken"
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    b${revcmp}  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
+
diff --git a/vm/mterp/c/OP_ADD_DOUBLE.c b/vm/mterp/c/OP_ADD_DOUBLE.c
new file mode 100644
index 0000000..571aeb8
--- /dev/null
+++ b/vm/mterp/c/OP_ADD_DOUBLE.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_DOUBLE(OP_ADD_DOUBLE, "add", +)
+OP_END
diff --git a/vm/mterp/c/OP_ADD_DOUBLE_2ADDR.c b/vm/mterp/c/OP_ADD_DOUBLE_2ADDR.c
new file mode 100644
index 0000000..af952cb
--- /dev/null
+++ b/vm/mterp/c/OP_ADD_DOUBLE_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_DOUBLE_2ADDR(OP_ADD_DOUBLE_2ADDR, "add", +)
+OP_END
diff --git a/vm/mterp/c/OP_ADD_FLOAT.c b/vm/mterp/c/OP_ADD_FLOAT.c
new file mode 100644
index 0000000..dab7d33
--- /dev/null
+++ b/vm/mterp/c/OP_ADD_FLOAT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_FLOAT(OP_ADD_FLOAT, "add", +)
+OP_END
diff --git a/vm/mterp/c/OP_ADD_FLOAT_2ADDR.c b/vm/mterp/c/OP_ADD_FLOAT_2ADDR.c
new file mode 100644
index 0000000..a068fd0
--- /dev/null
+++ b/vm/mterp/c/OP_ADD_FLOAT_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_FLOAT_2ADDR(OP_ADD_FLOAT_2ADDR, "add", +)
+OP_END
diff --git a/vm/mterp/c/OP_ADD_INT.c b/vm/mterp/c/OP_ADD_INT.c
new file mode 100644
index 0000000..dda2cb8
--- /dev/null
+++ b/vm/mterp/c/OP_ADD_INT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT(OP_ADD_INT, "add", +, false)
+OP_END
diff --git a/vm/mterp/c/OP_ADD_INT_2ADDR.c b/vm/mterp/c/OP_ADD_INT_2ADDR.c
new file mode 100644
index 0000000..29f32dd
--- /dev/null
+++ b/vm/mterp/c/OP_ADD_INT_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_2ADDR(OP_ADD_INT_2ADDR, "add", +, false)
+OP_END
diff --git a/vm/mterp/c/OP_ADD_INT_LIT16.c b/vm/mterp/c/OP_ADD_INT_LIT16.c
new file mode 100644
index 0000000..578bd50
--- /dev/null
+++ b/vm/mterp/c/OP_ADD_INT_LIT16.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_LIT16(OP_ADD_INT_LIT16, "add", (s4), +, false)
+OP_END
diff --git a/vm/mterp/c/OP_ADD_INT_LIT8.c b/vm/mterp/c/OP_ADD_INT_LIT8.c
new file mode 100644
index 0000000..a25b5f6
--- /dev/null
+++ b/vm/mterp/c/OP_ADD_INT_LIT8.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_LIT8(OP_ADD_INT_LIT8,   "add", +, false)
+OP_END
diff --git a/vm/mterp/c/OP_ADD_LONG.c b/vm/mterp/c/OP_ADD_LONG.c
new file mode 100644
index 0000000..24a12bd
--- /dev/null
+++ b/vm/mterp/c/OP_ADD_LONG.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_LONG(OP_ADD_LONG, "add", +, false)
+OP_END
diff --git a/vm/mterp/c/OP_ADD_LONG_2ADDR.c b/vm/mterp/c/OP_ADD_LONG_2ADDR.c
new file mode 100644
index 0000000..0b7f740
--- /dev/null
+++ b/vm/mterp/c/OP_ADD_LONG_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_LONG_2ADDR(OP_ADD_LONG_2ADDR, "add", +, false)
+OP_END
diff --git a/vm/mterp/c/OP_AGET.c b/vm/mterp/c/OP_AGET.c
new file mode 100644
index 0000000..766beaf
--- /dev/null
+++ b/vm/mterp/c/OP_AGET.c
@@ -0,0 +1,2 @@
+HANDLE_OP_AGET(OP_AGET, "", u4, )
+OP_END
diff --git a/vm/mterp/c/OP_AGET_BOOLEAN.c b/vm/mterp/c/OP_AGET_BOOLEAN.c
new file mode 100644
index 0000000..d63bc10
--- /dev/null
+++ b/vm/mterp/c/OP_AGET_BOOLEAN.c
@@ -0,0 +1,2 @@
+HANDLE_OP_AGET(OP_AGET_BOOLEAN, "-boolean", u1, )
+OP_END
diff --git a/vm/mterp/c/OP_AGET_BYTE.c b/vm/mterp/c/OP_AGET_BYTE.c
new file mode 100644
index 0000000..61ecc05
--- /dev/null
+++ b/vm/mterp/c/OP_AGET_BYTE.c
@@ -0,0 +1,2 @@
+HANDLE_OP_AGET(OP_AGET_BYTE, "-byte", s1, )
+OP_END
diff --git a/vm/mterp/c/OP_AGET_CHAR.c b/vm/mterp/c/OP_AGET_CHAR.c
new file mode 100644
index 0000000..55e16ef
--- /dev/null
+++ b/vm/mterp/c/OP_AGET_CHAR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_AGET(OP_AGET_CHAR, "-char", u2, )
+OP_END
diff --git a/vm/mterp/c/OP_AGET_OBJECT.c b/vm/mterp/c/OP_AGET_OBJECT.c
new file mode 100644
index 0000000..903637c
--- /dev/null
+++ b/vm/mterp/c/OP_AGET_OBJECT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_AGET(OP_AGET_OBJECT, "-object", u4, )
+OP_END
diff --git a/vm/mterp/c/OP_AGET_SHORT.c b/vm/mterp/c/OP_AGET_SHORT.c
new file mode 100644
index 0000000..176b4a6
--- /dev/null
+++ b/vm/mterp/c/OP_AGET_SHORT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_AGET(OP_AGET_SHORT, "-short", s2, )
+OP_END
diff --git a/vm/mterp/c/OP_AGET_WIDE.c b/vm/mterp/c/OP_AGET_WIDE.c
new file mode 100644
index 0000000..e7974cb
--- /dev/null
+++ b/vm/mterp/c/OP_AGET_WIDE.c
@@ -0,0 +1,2 @@
+HANDLE_OP_AGET(OP_AGET_WIDE, "-wide", s8, _WIDE)
+OP_END
diff --git a/vm/mterp/c/OP_AND_INT.c b/vm/mterp/c/OP_AND_INT.c
new file mode 100644
index 0000000..779b2d3
--- /dev/null
+++ b/vm/mterp/c/OP_AND_INT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT(OP_AND_INT, "and", &, false)
+OP_END
diff --git a/vm/mterp/c/OP_AND_INT_2ADDR.c b/vm/mterp/c/OP_AND_INT_2ADDR.c
new file mode 100644
index 0000000..4330c35
--- /dev/null
+++ b/vm/mterp/c/OP_AND_INT_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_2ADDR(OP_AND_INT_2ADDR, "and", &, false)
+OP_END
diff --git a/vm/mterp/c/OP_AND_INT_LIT16.c b/vm/mterp/c/OP_AND_INT_LIT16.c
new file mode 100644
index 0000000..01f48ba
--- /dev/null
+++ b/vm/mterp/c/OP_AND_INT_LIT16.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_LIT16(OP_AND_INT_LIT16, "and", (s4), &, false)
+OP_END
diff --git a/vm/mterp/c/OP_AND_INT_LIT8.c b/vm/mterp/c/OP_AND_INT_LIT8.c
new file mode 100644
index 0000000..7bcfa0a
--- /dev/null
+++ b/vm/mterp/c/OP_AND_INT_LIT8.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_LIT8(OP_AND_INT_LIT8,   "and", &, false)
+OP_END
diff --git a/vm/mterp/c/OP_AND_LONG.c b/vm/mterp/c/OP_AND_LONG.c
new file mode 100644
index 0000000..a92a5d4
--- /dev/null
+++ b/vm/mterp/c/OP_AND_LONG.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_LONG(OP_AND_LONG, "and", &, false)
+OP_END
diff --git a/vm/mterp/c/OP_AND_LONG_2ADDR.c b/vm/mterp/c/OP_AND_LONG_2ADDR.c
new file mode 100644
index 0000000..8161b8c
--- /dev/null
+++ b/vm/mterp/c/OP_AND_LONG_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_LONG_2ADDR(OP_AND_LONG_2ADDR, "and", &, false)
+OP_END
diff --git a/vm/mterp/c/OP_APUT.c b/vm/mterp/c/OP_APUT.c
new file mode 100644
index 0000000..07d3e04
--- /dev/null
+++ b/vm/mterp/c/OP_APUT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_APUT(OP_APUT, "", u4, )
+OP_END
diff --git a/vm/mterp/c/OP_APUT_BOOLEAN.c b/vm/mterp/c/OP_APUT_BOOLEAN.c
new file mode 100644
index 0000000..fc69147
--- /dev/null
+++ b/vm/mterp/c/OP_APUT_BOOLEAN.c
@@ -0,0 +1,2 @@
+HANDLE_OP_APUT(OP_APUT_BOOLEAN, "-boolean", u1, )
+OP_END
diff --git a/vm/mterp/c/OP_APUT_BYTE.c b/vm/mterp/c/OP_APUT_BYTE.c
new file mode 100644
index 0000000..45aeb0b
--- /dev/null
+++ b/vm/mterp/c/OP_APUT_BYTE.c
@@ -0,0 +1,2 @@
+HANDLE_OP_APUT(OP_APUT_BYTE, "-byte", s1, )
+OP_END
diff --git a/vm/mterp/c/OP_APUT_CHAR.c b/vm/mterp/c/OP_APUT_CHAR.c
new file mode 100644
index 0000000..1553c27
--- /dev/null
+++ b/vm/mterp/c/OP_APUT_CHAR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_APUT(OP_APUT_CHAR, "-char", u2, )
+OP_END
diff --git a/vm/mterp/c/OP_APUT_OBJECT.c b/vm/mterp/c/OP_APUT_OBJECT.c
new file mode 100644
index 0000000..6533b7d
--- /dev/null
+++ b/vm/mterp/c/OP_APUT_OBJECT.c
@@ -0,0 +1,39 @@
+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
diff --git a/vm/mterp/c/OP_APUT_SHORT.c b/vm/mterp/c/OP_APUT_SHORT.c
new file mode 100644
index 0000000..a72b5ea
--- /dev/null
+++ b/vm/mterp/c/OP_APUT_SHORT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_APUT(OP_APUT_SHORT, "-short", s2, )
+OP_END
diff --git a/vm/mterp/c/OP_APUT_WIDE.c b/vm/mterp/c/OP_APUT_WIDE.c
new file mode 100644
index 0000000..39c8cfa
--- /dev/null
+++ b/vm/mterp/c/OP_APUT_WIDE.c
@@ -0,0 +1,2 @@
+HANDLE_OP_APUT(OP_APUT_WIDE, "-wide", s8, _WIDE)
+OP_END
diff --git a/vm/mterp/c/OP_ARRAY_LENGTH.c b/vm/mterp/c/OP_ARRAY_LENGTH.c
new file mode 100644
index 0000000..f38b057
--- /dev/null
+++ b/vm/mterp/c/OP_ARRAY_LENGTH.c
@@ -0,0 +1,15 @@
+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
diff --git a/vm/mterp/c/OP_CHECK_CAST.c b/vm/mterp/c/OP_CHECK_CAST.c
new file mode 100644
index 0000000..9bc1ecf
--- /dev/null
+++ b/vm/mterp/c/OP_CHECK_CAST.c
@@ -0,0 +1,32 @@
+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(method->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
diff --git a/vm/mterp/c/OP_CMPG_DOUBLE.c b/vm/mterp/c/OP_CMPG_DOUBLE.c
new file mode 100644
index 0000000..3f4082c
--- /dev/null
+++ b/vm/mterp/c/OP_CMPG_DOUBLE.c
@@ -0,0 +1,2 @@
+HANDLE_OP_CMPX(OP_CMPG_DOUBLE, "g-double", double, _DOUBLE, 1)
+OP_END
diff --git a/vm/mterp/c/OP_CMPG_FLOAT.c b/vm/mterp/c/OP_CMPG_FLOAT.c
new file mode 100644
index 0000000..0bba49e
--- /dev/null
+++ b/vm/mterp/c/OP_CMPG_FLOAT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_CMPX(OP_CMPG_FLOAT, "g-float", float, _FLOAT, 1)
+OP_END
diff --git a/vm/mterp/c/OP_CMPL_DOUBLE.c b/vm/mterp/c/OP_CMPL_DOUBLE.c
new file mode 100644
index 0000000..4da18b4
--- /dev/null
+++ b/vm/mterp/c/OP_CMPL_DOUBLE.c
@@ -0,0 +1,2 @@
+HANDLE_OP_CMPX(OP_CMPL_DOUBLE, "l-double", double, _DOUBLE, -1)
+OP_END
diff --git a/vm/mterp/c/OP_CMPL_FLOAT.c b/vm/mterp/c/OP_CMPL_FLOAT.c
new file mode 100644
index 0000000..7916193
--- /dev/null
+++ b/vm/mterp/c/OP_CMPL_FLOAT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_CMPX(OP_CMPL_FLOAT, "l-float", float, _FLOAT, -1)
+OP_END
diff --git a/vm/mterp/c/OP_CMP_LONG.c b/vm/mterp/c/OP_CMP_LONG.c
new file mode 100644
index 0000000..a0e412c
--- /dev/null
+++ b/vm/mterp/c/OP_CMP_LONG.c
@@ -0,0 +1,2 @@
+HANDLE_OP_CMPX(OP_CMP_LONG, "-long", s8, _WIDE, 0)
+OP_END
diff --git a/vm/mterp/c/OP_CONST.c b/vm/mterp/c/OP_CONST.c
new file mode 100644
index 0000000..e281a51
--- /dev/null
+++ b/vm/mterp/c/OP_CONST.c
@@ -0,0 +1,12 @@
+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
diff --git a/vm/mterp/c/OP_CONST_16.c b/vm/mterp/c/OP_CONST_16.c
new file mode 100644
index 0000000..f58f50c
--- /dev/null
+++ b/vm/mterp/c/OP_CONST_16.c
@@ -0,0 +1,7 @@
+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
diff --git a/vm/mterp/c/OP_CONST_4.c b/vm/mterp/c/OP_CONST_4.c
new file mode 100644
index 0000000..800ef9a
--- /dev/null
+++ b/vm/mterp/c/OP_CONST_4.c
@@ -0,0 +1,11 @@
+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
diff --git a/vm/mterp/c/OP_CONST_CLASS.c b/vm/mterp/c/OP_CONST_CLASS.c
new file mode 100644
index 0000000..d0aefa3
--- /dev/null
+++ b/vm/mterp/c/OP_CONST_CLASS.c
@@ -0,0 +1,18 @@
+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(method->clazz, ref, true);
+            if (clazz == NULL)
+                GOTO(exceptionThrown);
+        }
+        SET_REGISTER(vdst, (u4) clazz);
+    }
+    FINISH(2);
+OP_END
diff --git a/vm/mterp/c/OP_CONST_HIGH16.c b/vm/mterp/c/OP_CONST_HIGH16.c
new file mode 100644
index 0000000..26b22f4
--- /dev/null
+++ b/vm/mterp/c/OP_CONST_HIGH16.c
@@ -0,0 +1,7 @@
+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
diff --git a/vm/mterp/c/OP_CONST_STRING.c b/vm/mterp/c/OP_CONST_STRING.c
new file mode 100644
index 0000000..c74635f
--- /dev/null
+++ b/vm/mterp/c/OP_CONST_STRING.c
@@ -0,0 +1,18 @@
+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(method->clazz, ref);
+            if (strObj == NULL)
+                GOTO(exceptionThrown);
+        }
+        SET_REGISTER(vdst, (u4) strObj);
+    }
+    FINISH(2);
+OP_END
diff --git a/vm/mterp/c/OP_CONST_STRING_JUMBO.c b/vm/mterp/c/OP_CONST_STRING_JUMBO.c
new file mode 100644
index 0000000..3afe47f
--- /dev/null
+++ b/vm/mterp/c/OP_CONST_STRING_JUMBO.c
@@ -0,0 +1,20 @@
+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(method->clazz, tmp);
+            if (strObj == NULL)
+                GOTO(exceptionThrown);
+        }
+        SET_REGISTER(vdst, (u4) strObj);
+    }
+    FINISH(3);
+OP_END
diff --git a/vm/mterp/c/OP_CONST_WIDE.c b/vm/mterp/c/OP_CONST_WIDE.c
new file mode 100644
index 0000000..ccb3955
--- /dev/null
+++ b/vm/mterp/c/OP_CONST_WIDE.c
@@ -0,0 +1,14 @@
+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
diff --git a/vm/mterp/c/OP_CONST_WIDE_16.c b/vm/mterp/c/OP_CONST_WIDE_16.c
new file mode 100644
index 0000000..da69f37
--- /dev/null
+++ b/vm/mterp/c/OP_CONST_WIDE_16.c
@@ -0,0 +1,7 @@
+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
diff --git a/vm/mterp/c/OP_CONST_WIDE_32.c b/vm/mterp/c/OP_CONST_WIDE_32.c
new file mode 100644
index 0000000..ad4acbb
--- /dev/null
+++ b/vm/mterp/c/OP_CONST_WIDE_32.c
@@ -0,0 +1,12 @@
+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
diff --git a/vm/mterp/c/OP_CONST_WIDE_HIGH16.c b/vm/mterp/c/OP_CONST_WIDE_HIGH16.c
new file mode 100644
index 0000000..bcc0664
--- /dev/null
+++ b/vm/mterp/c/OP_CONST_WIDE_HIGH16.c
@@ -0,0 +1,7 @@
+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
diff --git a/vm/mterp/c/OP_DIV_DOUBLE.c b/vm/mterp/c/OP_DIV_DOUBLE.c
new file mode 100644
index 0000000..d6e4b55
--- /dev/null
+++ b/vm/mterp/c/OP_DIV_DOUBLE.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_DOUBLE(OP_DIV_DOUBLE, "div", /)
+OP_END
diff --git a/vm/mterp/c/OP_DIV_DOUBLE_2ADDR.c b/vm/mterp/c/OP_DIV_DOUBLE_2ADDR.c
new file mode 100644
index 0000000..85a1523
--- /dev/null
+++ b/vm/mterp/c/OP_DIV_DOUBLE_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_DOUBLE_2ADDR(OP_DIV_DOUBLE_2ADDR, "div", /)
+OP_END
diff --git a/vm/mterp/c/OP_DIV_FLOAT.c b/vm/mterp/c/OP_DIV_FLOAT.c
new file mode 100644
index 0000000..2c5049b
--- /dev/null
+++ b/vm/mterp/c/OP_DIV_FLOAT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_FLOAT(OP_DIV_FLOAT, "div", /)
+OP_END
diff --git a/vm/mterp/c/OP_DIV_FLOAT_2ADDR.c b/vm/mterp/c/OP_DIV_FLOAT_2ADDR.c
new file mode 100644
index 0000000..cd7b4d9
--- /dev/null
+++ b/vm/mterp/c/OP_DIV_FLOAT_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_FLOAT_2ADDR(OP_DIV_FLOAT_2ADDR, "div", /)
+OP_END
diff --git a/vm/mterp/c/OP_DIV_INT.c b/vm/mterp/c/OP_DIV_INT.c
new file mode 100644
index 0000000..70b1cb9
--- /dev/null
+++ b/vm/mterp/c/OP_DIV_INT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT(OP_DIV_INT, "div", /, true)
+OP_END
diff --git a/vm/mterp/c/OP_DIV_INT_2ADDR.c b/vm/mterp/c/OP_DIV_INT_2ADDR.c
new file mode 100644
index 0000000..0491373
--- /dev/null
+++ b/vm/mterp/c/OP_DIV_INT_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_2ADDR(OP_DIV_INT_2ADDR, "div", /, true)
+OP_END
diff --git a/vm/mterp/c/OP_DIV_INT_LIT16.c b/vm/mterp/c/OP_DIV_INT_LIT16.c
new file mode 100644
index 0000000..0975d06
--- /dev/null
+++ b/vm/mterp/c/OP_DIV_INT_LIT16.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_LIT16(OP_DIV_INT_LIT16, "div", (s4), /, true)
+OP_END
diff --git a/vm/mterp/c/OP_DIV_INT_LIT8.c b/vm/mterp/c/OP_DIV_INT_LIT8.c
new file mode 100644
index 0000000..dfd84c0
--- /dev/null
+++ b/vm/mterp/c/OP_DIV_INT_LIT8.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_LIT8(OP_DIV_INT_LIT8,   "div", /, true)
+OP_END
diff --git a/vm/mterp/c/OP_DIV_LONG.c b/vm/mterp/c/OP_DIV_LONG.c
new file mode 100644
index 0000000..da07f1c
--- /dev/null
+++ b/vm/mterp/c/OP_DIV_LONG.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_LONG(OP_DIV_LONG, "div", /, true)
+OP_END
diff --git a/vm/mterp/c/OP_DIV_LONG_2ADDR.c b/vm/mterp/c/OP_DIV_LONG_2ADDR.c
new file mode 100644
index 0000000..4fb234d
--- /dev/null
+++ b/vm/mterp/c/OP_DIV_LONG_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_LONG_2ADDR(OP_DIV_LONG_2ADDR, "div", /, true)
+OP_END
diff --git a/vm/mterp/c/OP_DOUBLE_TO_FLOAT.c b/vm/mterp/c/OP_DOUBLE_TO_FLOAT.c
new file mode 100644
index 0000000..152e5fd
--- /dev/null
+++ b/vm/mterp/c/OP_DOUBLE_TO_FLOAT.c
@@ -0,0 +1,2 @@
+HANDLE_NUMCONV(OP_DOUBLE_TO_FLOAT,      "double-to-float", _DOUBLE, _FLOAT)
+OP_END
diff --git a/vm/mterp/c/OP_DOUBLE_TO_INT.c b/vm/mterp/c/OP_DOUBLE_TO_INT.c
new file mode 100644
index 0000000..e210b92
--- /dev/null
+++ b/vm/mterp/c/OP_DOUBLE_TO_INT.c
@@ -0,0 +1,3 @@
+HANDLE_FLOAT_TO_INT(OP_DOUBLE_TO_INT,   "double-to-int",
+    double, _DOUBLE, s4, _INT)
+OP_END
diff --git a/vm/mterp/c/OP_DOUBLE_TO_LONG.c b/vm/mterp/c/OP_DOUBLE_TO_LONG.c
new file mode 100644
index 0000000..44d548c
--- /dev/null
+++ b/vm/mterp/c/OP_DOUBLE_TO_LONG.c
@@ -0,0 +1,3 @@
+HANDLE_FLOAT_TO_INT(OP_DOUBLE_TO_LONG,  "double-to-long",
+    double, _DOUBLE, s8, _WIDE)
+OP_END
diff --git a/vm/mterp/c/OP_EXECUTE_INLINE.c b/vm/mterp/c/OP_EXECUTE_INLINE.c
new file mode 100644
index 0000000..0081dfe
--- /dev/null
+++ b/vm/mterp/c/OP_EXECUTE_INLINE.c
@@ -0,0 +1,58 @@
+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
diff --git a/vm/mterp/c/OP_FILLED_NEW_ARRAY.c b/vm/mterp/c/OP_FILLED_NEW_ARRAY.c
new file mode 100644
index 0000000..b925847
--- /dev/null
+++ b/vm/mterp/c/OP_FILLED_NEW_ARRAY.c
@@ -0,0 +1,3 @@
+HANDLE_OPCODE(OP_FILLED_NEW_ARRAY /*vB, {vD, vE, vF, vG, vA}, class@CCCC*/)
+    GOTO(filledNewArray, false);
+OP_END
diff --git a/vm/mterp/c/OP_FILLED_NEW_ARRAY_RANGE.c b/vm/mterp/c/OP_FILLED_NEW_ARRAY_RANGE.c
new file mode 100644
index 0000000..714e248
--- /dev/null
+++ b/vm/mterp/c/OP_FILLED_NEW_ARRAY_RANGE.c
@@ -0,0 +1,3 @@
+HANDLE_OPCODE(OP_FILLED_NEW_ARRAY_RANGE /*{vCCCC..v(CCCC+AA-1)}, class@BBBB*/)
+    GOTO(filledNewArray, true);
+OP_END
diff --git a/vm/mterp/c/OP_FILL_ARRAY_DATA.c b/vm/mterp/c/OP_FILL_ARRAY_DATA.c
new file mode 100644
index 0000000..22d9d66
--- /dev/null
+++ b/vm/mterp/c/OP_FILL_ARRAY_DATA.c
@@ -0,0 +1,28 @@
+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 < method->insns ||
+            arrayData >= method->insns + dvmGetMethodInsnsSize(method))
+        {
+            /* 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
diff --git a/vm/mterp/c/OP_FLOAT_TO_DOUBLE.c b/vm/mterp/c/OP_FLOAT_TO_DOUBLE.c
new file mode 100644
index 0000000..ea5e7a6
--- /dev/null
+++ b/vm/mterp/c/OP_FLOAT_TO_DOUBLE.c
@@ -0,0 +1,2 @@
+HANDLE_NUMCONV(OP_FLOAT_TO_DOUBLE,      "float-to-double", _FLOAT, _DOUBLE)
+OP_END
diff --git a/vm/mterp/c/OP_FLOAT_TO_INT.c b/vm/mterp/c/OP_FLOAT_TO_INT.c
new file mode 100644
index 0000000..15522f8
--- /dev/null
+++ b/vm/mterp/c/OP_FLOAT_TO_INT.c
@@ -0,0 +1,3 @@
+HANDLE_FLOAT_TO_INT(OP_FLOAT_TO_INT,    "float-to-int",
+    float, _FLOAT, s4, _INT)
+OP_END
diff --git a/vm/mterp/c/OP_FLOAT_TO_LONG.c b/vm/mterp/c/OP_FLOAT_TO_LONG.c
new file mode 100644
index 0000000..03bd30d
--- /dev/null
+++ b/vm/mterp/c/OP_FLOAT_TO_LONG.c
@@ -0,0 +1,3 @@
+HANDLE_FLOAT_TO_INT(OP_FLOAT_TO_LONG,   "float-to-long",
+    float, _FLOAT, s8, _WIDE)
+OP_END
diff --git a/vm/mterp/c/OP_GOTO.c b/vm/mterp/c/OP_GOTO.c
new file mode 100644
index 0000000..eed7b9f
--- /dev/null
+++ b/vm/mterp/c/OP_GOTO.c
@@ -0,0 +1,11 @@
+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
diff --git a/vm/mterp/c/OP_GOTO_16.c b/vm/mterp/c/OP_GOTO_16.c
new file mode 100644
index 0000000..afdccb3
--- /dev/null
+++ b/vm/mterp/c/OP_GOTO_16.c
@@ -0,0 +1,14 @@
+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
diff --git a/vm/mterp/c/OP_GOTO_32.c b/vm/mterp/c/OP_GOTO_32.c
new file mode 100644
index 0000000..d1cee32
--- /dev/null
+++ b/vm/mterp/c/OP_GOTO_32.c
@@ -0,0 +1,15 @@
+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
diff --git a/vm/mterp/c/OP_IF_EQ.c b/vm/mterp/c/OP_IF_EQ.c
new file mode 100644
index 0000000..2c3b9b5
--- /dev/null
+++ b/vm/mterp/c/OP_IF_EQ.c
@@ -0,0 +1,2 @@
+HANDLE_OP_IF_XX(OP_IF_EQ, "eq", ==)
+OP_END
diff --git a/vm/mterp/c/OP_IF_EQZ.c b/vm/mterp/c/OP_IF_EQZ.c
new file mode 100644
index 0000000..d2dd1aa
--- /dev/null
+++ b/vm/mterp/c/OP_IF_EQZ.c
@@ -0,0 +1,2 @@
+HANDLE_OP_IF_XXZ(OP_IF_EQZ, "eqz", ==)
+OP_END
diff --git a/vm/mterp/c/OP_IF_GE.c b/vm/mterp/c/OP_IF_GE.c
new file mode 100644
index 0000000..8aa85c4
--- /dev/null
+++ b/vm/mterp/c/OP_IF_GE.c
@@ -0,0 +1,2 @@
+HANDLE_OP_IF_XX(OP_IF_GE, "ge", >=)
+OP_END
diff --git a/vm/mterp/c/OP_IF_GEZ.c b/vm/mterp/c/OP_IF_GEZ.c
new file mode 100644
index 0000000..8c4b78a
--- /dev/null
+++ b/vm/mterp/c/OP_IF_GEZ.c
@@ -0,0 +1,2 @@
+HANDLE_OP_IF_XXZ(OP_IF_GEZ, "gez", >=)
+OP_END
diff --git a/vm/mterp/c/OP_IF_GT.c b/vm/mterp/c/OP_IF_GT.c
new file mode 100644
index 0000000..d35eb29
--- /dev/null
+++ b/vm/mterp/c/OP_IF_GT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_IF_XX(OP_IF_GT, "gt", >)
+OP_END
diff --git a/vm/mterp/c/OP_IF_GTZ.c b/vm/mterp/c/OP_IF_GTZ.c
new file mode 100644
index 0000000..63a0073
--- /dev/null
+++ b/vm/mterp/c/OP_IF_GTZ.c
@@ -0,0 +1,2 @@
+HANDLE_OP_IF_XXZ(OP_IF_GTZ, "gtz", >)
+OP_END
diff --git a/vm/mterp/c/OP_IF_LE.c b/vm/mterp/c/OP_IF_LE.c
new file mode 100644
index 0000000..f4b213a
--- /dev/null
+++ b/vm/mterp/c/OP_IF_LE.c
@@ -0,0 +1,2 @@
+HANDLE_OP_IF_XX(OP_IF_LE, "le", <=)
+OP_END
diff --git a/vm/mterp/c/OP_IF_LEZ.c b/vm/mterp/c/OP_IF_LEZ.c
new file mode 100644
index 0000000..1d57a50
--- /dev/null
+++ b/vm/mterp/c/OP_IF_LEZ.c
@@ -0,0 +1,2 @@
+HANDLE_OP_IF_XXZ(OP_IF_LEZ, "lez", <=)
+OP_END
diff --git a/vm/mterp/c/OP_IF_LT.c b/vm/mterp/c/OP_IF_LT.c
new file mode 100644
index 0000000..0233892
--- /dev/null
+++ b/vm/mterp/c/OP_IF_LT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_IF_XX(OP_IF_LT, "lt", <)
+OP_END
diff --git a/vm/mterp/c/OP_IF_LTZ.c b/vm/mterp/c/OP_IF_LTZ.c
new file mode 100644
index 0000000..b4b9be2
--- /dev/null
+++ b/vm/mterp/c/OP_IF_LTZ.c
@@ -0,0 +1,2 @@
+HANDLE_OP_IF_XXZ(OP_IF_LTZ, "ltz", <)
+OP_END
diff --git a/vm/mterp/c/OP_IF_NE.c b/vm/mterp/c/OP_IF_NE.c
new file mode 100644
index 0000000..8da70a5
--- /dev/null
+++ b/vm/mterp/c/OP_IF_NE.c
@@ -0,0 +1,2 @@
+HANDLE_OP_IF_XX(OP_IF_NE, "ne", !=)
+OP_END
diff --git a/vm/mterp/c/OP_IF_NEZ.c b/vm/mterp/c/OP_IF_NEZ.c
new file mode 100644
index 0000000..209e836
--- /dev/null
+++ b/vm/mterp/c/OP_IF_NEZ.c
@@ -0,0 +1,2 @@
+HANDLE_OP_IF_XXZ(OP_IF_NEZ, "nez", !=)
+OP_END
diff --git a/vm/mterp/c/OP_IGET.c b/vm/mterp/c/OP_IGET.c
new file mode 100644
index 0000000..c6333e5
--- /dev/null
+++ b/vm/mterp/c/OP_IGET.c
@@ -0,0 +1,2 @@
+HANDLE_IGET_X(OP_IGET,                  "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_IGET_BOOLEAN.c b/vm/mterp/c/OP_IGET_BOOLEAN.c
new file mode 100644
index 0000000..a5a47be
--- /dev/null
+++ b/vm/mterp/c/OP_IGET_BOOLEAN.c
@@ -0,0 +1,2 @@
+HANDLE_IGET_X(OP_IGET_BOOLEAN,          "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_IGET_BYTE.c b/vm/mterp/c/OP_IGET_BYTE.c
new file mode 100644
index 0000000..647f311
--- /dev/null
+++ b/vm/mterp/c/OP_IGET_BYTE.c
@@ -0,0 +1,2 @@
+HANDLE_IGET_X(OP_IGET_BYTE,             "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_IGET_CHAR.c b/vm/mterp/c/OP_IGET_CHAR.c
new file mode 100644
index 0000000..9a8adb0
--- /dev/null
+++ b/vm/mterp/c/OP_IGET_CHAR.c
@@ -0,0 +1,2 @@
+HANDLE_IGET_X(OP_IGET_CHAR,             "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_IGET_OBJECT.c b/vm/mterp/c/OP_IGET_OBJECT.c
new file mode 100644
index 0000000..03c9e50
--- /dev/null
+++ b/vm/mterp/c/OP_IGET_OBJECT.c
@@ -0,0 +1,2 @@
+HANDLE_IGET_X(OP_IGET_OBJECT,           "-object", Object, _AS_OBJECT)
+OP_END
diff --git a/vm/mterp/c/OP_IGET_OBJECT_QUICK.c b/vm/mterp/c/OP_IGET_OBJECT_QUICK.c
new file mode 100644
index 0000000..2ac3a54
--- /dev/null
+++ b/vm/mterp/c/OP_IGET_OBJECT_QUICK.c
@@ -0,0 +1,2 @@
+HANDLE_IGET_X_QUICK(OP_IGET_OBJECT_QUICK,   "-object", Object, _AS_OBJECT)
+OP_END
diff --git a/vm/mterp/c/OP_IGET_QUICK.c b/vm/mterp/c/OP_IGET_QUICK.c
new file mode 100644
index 0000000..b5724cc
--- /dev/null
+++ b/vm/mterp/c/OP_IGET_QUICK.c
@@ -0,0 +1,2 @@
+HANDLE_IGET_X_QUICK(OP_IGET_QUICK,          "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_IGET_SHORT.c b/vm/mterp/c/OP_IGET_SHORT.c
new file mode 100644
index 0000000..3e77789
--- /dev/null
+++ b/vm/mterp/c/OP_IGET_SHORT.c
@@ -0,0 +1,2 @@
+HANDLE_IGET_X(OP_IGET_SHORT,            "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_IGET_WIDE.c b/vm/mterp/c/OP_IGET_WIDE.c
new file mode 100644
index 0000000..cb1fcca
--- /dev/null
+++ b/vm/mterp/c/OP_IGET_WIDE.c
@@ -0,0 +1,2 @@
+HANDLE_IGET_X(OP_IGET_WIDE,             "-wide", Long, _WIDE)
+OP_END
diff --git a/vm/mterp/c/OP_IGET_WIDE_QUICK.c b/vm/mterp/c/OP_IGET_WIDE_QUICK.c
new file mode 100644
index 0000000..adb4fc1
--- /dev/null
+++ b/vm/mterp/c/OP_IGET_WIDE_QUICK.c
@@ -0,0 +1,2 @@
+HANDLE_IGET_X_QUICK(OP_IGET_WIDE_QUICK,     "-wide", Long, _WIDE)
+OP_END
diff --git a/vm/mterp/c/OP_INSTANCE_OF.c b/vm/mterp/c/OP_INSTANCE_OF.c
new file mode 100644
index 0000000..5855b5a
--- /dev/null
+++ b/vm/mterp/c/OP_INSTANCE_OF.c
@@ -0,0 +1,30 @@
+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(method->clazz, ref, true);
+                if (clazz == NULL)
+                    GOTO(exceptionThrown);
+            }
+            SET_REGISTER(vdst, dvmInstanceof(obj->clazz, clazz));
+        }
+    }
+    FINISH(2);
+OP_END
diff --git a/vm/mterp/c/OP_INT_TO_BYTE.c b/vm/mterp/c/OP_INT_TO_BYTE.c
new file mode 100644
index 0000000..ea75747
--- /dev/null
+++ b/vm/mterp/c/OP_INT_TO_BYTE.c
@@ -0,0 +1,2 @@
+HANDLE_INT_TO_SMALL(OP_INT_TO_BYTE,     "byte", s1)
+OP_END
diff --git a/vm/mterp/c/OP_INT_TO_CHAR.c b/vm/mterp/c/OP_INT_TO_CHAR.c
new file mode 100644
index 0000000..45ae0df
--- /dev/null
+++ b/vm/mterp/c/OP_INT_TO_CHAR.c
@@ -0,0 +1,2 @@
+HANDLE_INT_TO_SMALL(OP_INT_TO_CHAR,     "char", u2)
+OP_END
diff --git a/vm/mterp/c/OP_INT_TO_DOUBLE.c b/vm/mterp/c/OP_INT_TO_DOUBLE.c
new file mode 100644
index 0000000..624c702
--- /dev/null
+++ b/vm/mterp/c/OP_INT_TO_DOUBLE.c
@@ -0,0 +1,2 @@
+HANDLE_NUMCONV(OP_INT_TO_DOUBLE,        "int-to-double", _INT, _DOUBLE)
+OP_END
diff --git a/vm/mterp/c/OP_INT_TO_FLOAT.c b/vm/mterp/c/OP_INT_TO_FLOAT.c
new file mode 100644
index 0000000..fd15199
--- /dev/null
+++ b/vm/mterp/c/OP_INT_TO_FLOAT.c
@@ -0,0 +1,2 @@
+HANDLE_NUMCONV(OP_INT_TO_FLOAT,         "int-to-float", _INT, _FLOAT)
+OP_END
diff --git a/vm/mterp/c/OP_INT_TO_LONG.c b/vm/mterp/c/OP_INT_TO_LONG.c
new file mode 100644
index 0000000..8bc4223
--- /dev/null
+++ b/vm/mterp/c/OP_INT_TO_LONG.c
@@ -0,0 +1,2 @@
+HANDLE_NUMCONV(OP_INT_TO_LONG,          "int-to-long", _INT, _WIDE)
+OP_END
diff --git a/vm/mterp/c/OP_INT_TO_SHORT.c b/vm/mterp/c/OP_INT_TO_SHORT.c
new file mode 100644
index 0000000..0f06739
--- /dev/null
+++ b/vm/mterp/c/OP_INT_TO_SHORT.c
@@ -0,0 +1,2 @@
+HANDLE_INT_TO_SMALL(OP_INT_TO_SHORT,    "short", s2)    /* want sign bit */
+OP_END
diff --git a/vm/mterp/c/OP_INVOKE_DIRECT.c b/vm/mterp/c/OP_INVOKE_DIRECT.c
new file mode 100644
index 0000000..2f8858f
--- /dev/null
+++ b/vm/mterp/c/OP_INVOKE_DIRECT.c
@@ -0,0 +1,3 @@
+HANDLE_OPCODE(OP_INVOKE_DIRECT /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO(invokeDirect, false);
+OP_END
diff --git a/vm/mterp/c/OP_INVOKE_DIRECT_EMPTY.c b/vm/mterp/c/OP_INVOKE_DIRECT_EMPTY.c
new file mode 100644
index 0000000..f816490
--- /dev/null
+++ b/vm/mterp/c/OP_INVOKE_DIRECT_EMPTY.c
@@ -0,0 +1,4 @@
+HANDLE_OPCODE(OP_INVOKE_DIRECT_EMPTY /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    //LOGI("Ignoring empty\n");
+    FINISH(3);
+OP_END
diff --git a/vm/mterp/c/OP_INVOKE_DIRECT_RANGE.c b/vm/mterp/c/OP_INVOKE_DIRECT_RANGE.c
new file mode 100644
index 0000000..7f13c5f
--- /dev/null
+++ b/vm/mterp/c/OP_INVOKE_DIRECT_RANGE.c
@@ -0,0 +1,3 @@
+HANDLE_OPCODE(OP_INVOKE_DIRECT_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO(invokeDirect, true);
+OP_END
diff --git a/vm/mterp/c/OP_INVOKE_INTERFACE.c b/vm/mterp/c/OP_INVOKE_INTERFACE.c
new file mode 100644
index 0000000..f78dc7e
--- /dev/null
+++ b/vm/mterp/c/OP_INVOKE_INTERFACE.c
@@ -0,0 +1,3 @@
+HANDLE_OPCODE(OP_INVOKE_INTERFACE /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO(invokeInterface, false);
+OP_END
diff --git a/vm/mterp/c/OP_INVOKE_INTERFACE_RANGE.c b/vm/mterp/c/OP_INVOKE_INTERFACE_RANGE.c
new file mode 100644
index 0000000..6ca2d7d
--- /dev/null
+++ b/vm/mterp/c/OP_INVOKE_INTERFACE_RANGE.c
@@ -0,0 +1,3 @@
+HANDLE_OPCODE(OP_INVOKE_INTERFACE_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO(invokeInterface, true);
+OP_END
diff --git a/vm/mterp/c/OP_INVOKE_STATIC.c b/vm/mterp/c/OP_INVOKE_STATIC.c
new file mode 100644
index 0000000..b04ba95
--- /dev/null
+++ b/vm/mterp/c/OP_INVOKE_STATIC.c
@@ -0,0 +1,3 @@
+HANDLE_OPCODE(OP_INVOKE_STATIC /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO(invokeStatic, false);
+OP_END
diff --git a/vm/mterp/c/OP_INVOKE_STATIC_RANGE.c b/vm/mterp/c/OP_INVOKE_STATIC_RANGE.c
new file mode 100644
index 0000000..2a74039
--- /dev/null
+++ b/vm/mterp/c/OP_INVOKE_STATIC_RANGE.c
@@ -0,0 +1,3 @@
+HANDLE_OPCODE(OP_INVOKE_STATIC_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO(invokeStatic, true);
+OP_END
diff --git a/vm/mterp/c/OP_INVOKE_SUPER.c b/vm/mterp/c/OP_INVOKE_SUPER.c
new file mode 100644
index 0000000..22cab39
--- /dev/null
+++ b/vm/mterp/c/OP_INVOKE_SUPER.c
@@ -0,0 +1,3 @@
+HANDLE_OPCODE(OP_INVOKE_SUPER /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO(invokeSuper, false);
+OP_END
diff --git a/vm/mterp/c/OP_INVOKE_SUPER_QUICK.c b/vm/mterp/c/OP_INVOKE_SUPER_QUICK.c
new file mode 100644
index 0000000..fcd074e
--- /dev/null
+++ b/vm/mterp/c/OP_INVOKE_SUPER_QUICK.c
@@ -0,0 +1,3 @@
+HANDLE_OPCODE(OP_INVOKE_SUPER_QUICK /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO(invokeSuperQuick, false);
+OP_END
diff --git a/vm/mterp/c/OP_INVOKE_SUPER_QUICK_RANGE.c b/vm/mterp/c/OP_INVOKE_SUPER_QUICK_RANGE.c
new file mode 100644
index 0000000..cb468fd
--- /dev/null
+++ b/vm/mterp/c/OP_INVOKE_SUPER_QUICK_RANGE.c
@@ -0,0 +1,3 @@
+HANDLE_OPCODE(OP_INVOKE_SUPER_QUICK_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO(invokeSuperQuick, true);
+OP_END
diff --git a/vm/mterp/c/OP_INVOKE_SUPER_RANGE.c b/vm/mterp/c/OP_INVOKE_SUPER_RANGE.c
new file mode 100644
index 0000000..fce1ff3
--- /dev/null
+++ b/vm/mterp/c/OP_INVOKE_SUPER_RANGE.c
@@ -0,0 +1,3 @@
+HANDLE_OPCODE(OP_INVOKE_SUPER_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO(invokeSuper, true);
+OP_END
diff --git a/vm/mterp/c/OP_INVOKE_VIRTUAL.c b/vm/mterp/c/OP_INVOKE_VIRTUAL.c
new file mode 100644
index 0000000..0cef190
--- /dev/null
+++ b/vm/mterp/c/OP_INVOKE_VIRTUAL.c
@@ -0,0 +1,3 @@
+HANDLE_OPCODE(OP_INVOKE_VIRTUAL /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO(invokeVirtual, false);
+OP_END
diff --git a/vm/mterp/c/OP_INVOKE_VIRTUAL_QUICK.c b/vm/mterp/c/OP_INVOKE_VIRTUAL_QUICK.c
new file mode 100644
index 0000000..50862b1
--- /dev/null
+++ b/vm/mterp/c/OP_INVOKE_VIRTUAL_QUICK.c
@@ -0,0 +1,3 @@
+HANDLE_OPCODE(OP_INVOKE_VIRTUAL_QUICK /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO(invokeVirtualQuick, false);
+OP_END
diff --git a/vm/mterp/c/OP_INVOKE_VIRTUAL_QUICK_RANGE.c b/vm/mterp/c/OP_INVOKE_VIRTUAL_QUICK_RANGE.c
new file mode 100644
index 0000000..388bc73
--- /dev/null
+++ b/vm/mterp/c/OP_INVOKE_VIRTUAL_QUICK_RANGE.c
@@ -0,0 +1,3 @@
+HANDLE_OPCODE(OP_INVOKE_VIRTUAL_QUICK_RANGE/*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO(invokeVirtualQuick, true);
+OP_END
diff --git a/vm/mterp/c/OP_INVOKE_VIRTUAL_RANGE.c b/vm/mterp/c/OP_INVOKE_VIRTUAL_RANGE.c
new file mode 100644
index 0000000..0cac69f
--- /dev/null
+++ b/vm/mterp/c/OP_INVOKE_VIRTUAL_RANGE.c
@@ -0,0 +1,3 @@
+HANDLE_OPCODE(OP_INVOKE_VIRTUAL_RANGE /*{vCCCC..v(CCCC+AA-1)}, meth@BBBB*/)
+    GOTO(invokeVirtual, true);
+OP_END
diff --git a/vm/mterp/c/OP_IPUT.c b/vm/mterp/c/OP_IPUT.c
new file mode 100644
index 0000000..9d503ef
--- /dev/null
+++ b/vm/mterp/c/OP_IPUT.c
@@ -0,0 +1,2 @@
+HANDLE_IPUT_X(OP_IPUT,                  "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_IPUT_BOOLEAN.c b/vm/mterp/c/OP_IPUT_BOOLEAN.c
new file mode 100644
index 0000000..7fe4929
--- /dev/null
+++ b/vm/mterp/c/OP_IPUT_BOOLEAN.c
@@ -0,0 +1,2 @@
+HANDLE_IPUT_X(OP_IPUT_BOOLEAN,          "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_IPUT_BYTE.c b/vm/mterp/c/OP_IPUT_BYTE.c
new file mode 100644
index 0000000..8a49fb7
--- /dev/null
+++ b/vm/mterp/c/OP_IPUT_BYTE.c
@@ -0,0 +1,2 @@
+HANDLE_IPUT_X(OP_IPUT_BYTE,             "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_IPUT_CHAR.c b/vm/mterp/c/OP_IPUT_CHAR.c
new file mode 100644
index 0000000..b2812c2
--- /dev/null
+++ b/vm/mterp/c/OP_IPUT_CHAR.c
@@ -0,0 +1,2 @@
+HANDLE_IPUT_X(OP_IPUT_CHAR,             "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_IPUT_OBJECT.c b/vm/mterp/c/OP_IPUT_OBJECT.c
new file mode 100644
index 0000000..799e846
--- /dev/null
+++ b/vm/mterp/c/OP_IPUT_OBJECT.c
@@ -0,0 +1,2 @@
+HANDLE_IPUT_X(OP_IPUT_OBJECT,           "-object", Object, _AS_OBJECT)
+OP_END
diff --git a/vm/mterp/c/OP_IPUT_OBJECT_QUICK.c b/vm/mterp/c/OP_IPUT_OBJECT_QUICK.c
new file mode 100644
index 0000000..8670188
--- /dev/null
+++ b/vm/mterp/c/OP_IPUT_OBJECT_QUICK.c
@@ -0,0 +1,2 @@
+HANDLE_IPUT_X_QUICK(OP_IPUT_OBJECT_QUICK,   "-object", Object, _AS_OBJECT)
+OP_END
diff --git a/vm/mterp/c/OP_IPUT_QUICK.c b/vm/mterp/c/OP_IPUT_QUICK.c
new file mode 100644
index 0000000..483b9b1
--- /dev/null
+++ b/vm/mterp/c/OP_IPUT_QUICK.c
@@ -0,0 +1,2 @@
+HANDLE_IPUT_X_QUICK(OP_IPUT_QUICK,          "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_IPUT_SHORT.c b/vm/mterp/c/OP_IPUT_SHORT.c
new file mode 100644
index 0000000..0a63ebc
--- /dev/null
+++ b/vm/mterp/c/OP_IPUT_SHORT.c
@@ -0,0 +1,2 @@
+HANDLE_IPUT_X(OP_IPUT_SHORT,            "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_IPUT_WIDE.c b/vm/mterp/c/OP_IPUT_WIDE.c
new file mode 100644
index 0000000..bb4926c
--- /dev/null
+++ b/vm/mterp/c/OP_IPUT_WIDE.c
@@ -0,0 +1,2 @@
+HANDLE_IPUT_X(OP_IPUT_WIDE,             "-wide", Long, _WIDE)
+OP_END
diff --git a/vm/mterp/c/OP_IPUT_WIDE_QUICK.c b/vm/mterp/c/OP_IPUT_WIDE_QUICK.c
new file mode 100644
index 0000000..691630b
--- /dev/null
+++ b/vm/mterp/c/OP_IPUT_WIDE_QUICK.c
@@ -0,0 +1,2 @@
+HANDLE_IPUT_X_QUICK(OP_IPUT_WIDE_QUICK,     "-wide", Long, _WIDE)
+OP_END
diff --git a/vm/mterp/c/OP_LONG_TO_DOUBLE.c b/vm/mterp/c/OP_LONG_TO_DOUBLE.c
new file mode 100644
index 0000000..91b5eb2
--- /dev/null
+++ b/vm/mterp/c/OP_LONG_TO_DOUBLE.c
@@ -0,0 +1,2 @@
+HANDLE_NUMCONV(OP_LONG_TO_DOUBLE,       "long-to-double", _WIDE, _DOUBLE)
+OP_END
diff --git a/vm/mterp/c/OP_LONG_TO_FLOAT.c b/vm/mterp/c/OP_LONG_TO_FLOAT.c
new file mode 100644
index 0000000..ff1f5fb
--- /dev/null
+++ b/vm/mterp/c/OP_LONG_TO_FLOAT.c
@@ -0,0 +1,2 @@
+HANDLE_NUMCONV(OP_LONG_TO_FLOAT,        "long-to-float", _WIDE, _FLOAT)
+OP_END
diff --git a/vm/mterp/c/OP_LONG_TO_INT.c b/vm/mterp/c/OP_LONG_TO_INT.c
new file mode 100644
index 0000000..87c9a2e
--- /dev/null
+++ b/vm/mterp/c/OP_LONG_TO_INT.c
@@ -0,0 +1,2 @@
+HANDLE_NUMCONV(OP_LONG_TO_INT,          "long-to-int", _WIDE, _INT)
+OP_END
diff --git a/vm/mterp/c/OP_MONITOR_ENTER.c b/vm/mterp/c/OP_MONITOR_ENTER.c
new file mode 100644
index 0000000..57329d2
--- /dev/null
+++ b/vm/mterp/c/OP_MONITOR_ENTER.c
@@ -0,0 +1,22 @@
+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
diff --git a/vm/mterp/c/OP_MONITOR_EXIT.c b/vm/mterp/c/OP_MONITOR_EXIT.c
new file mode 100644
index 0000000..d13a95f
--- /dev/null
+++ b/vm/mterp/c/OP_MONITOR_EXIT.c
@@ -0,0 +1,30 @@
+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
diff --git a/vm/mterp/c/OP_MOVE.c b/vm/mterp/c/OP_MOVE.c
new file mode 100644
index 0000000..6666199
--- /dev/null
+++ b/vm/mterp/c/OP_MOVE.c
@@ -0,0 +1,9 @@
+HANDLE_OPCODE($opcode /*vA, vB*/)
+    vdst = INST_A(inst);
+    vsrc1 = INST_B(inst);
+    ILOGV("|move%s v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(1);
+OP_END
diff --git a/vm/mterp/c/OP_MOVE_16.c b/vm/mterp/c/OP_MOVE_16.c
new file mode 100644
index 0000000..53af5d5
--- /dev/null
+++ b/vm/mterp/c/OP_MOVE_16.c
@@ -0,0 +1,9 @@
+HANDLE_OPCODE($opcode /*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
diff --git a/vm/mterp/c/OP_MOVE_EXCEPTION.c b/vm/mterp/c/OP_MOVE_EXCEPTION.c
new file mode 100644
index 0000000..86587ca
--- /dev/null
+++ b/vm/mterp/c/OP_MOVE_EXCEPTION.c
@@ -0,0 +1,8 @@
+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
diff --git a/vm/mterp/c/OP_MOVE_FROM16.c b/vm/mterp/c/OP_MOVE_FROM16.c
new file mode 100644
index 0000000..59fc285
--- /dev/null
+++ b/vm/mterp/c/OP_MOVE_FROM16.c
@@ -0,0 +1,9 @@
+HANDLE_OPCODE($opcode /*vAA, vBBBB*/)
+    vdst = INST_AA(inst);
+    vsrc1 = FETCH(1);
+    ILOGV("|move%s/from16 v%d,v%d %s(v%d=0x%08x)",
+        (INST_INST(inst) == OP_MOVE_FROM16) ? "" : "-object", vdst, vsrc1,
+        kSpacing, vdst, GET_REGISTER(vsrc1));
+    SET_REGISTER(vdst, GET_REGISTER(vsrc1));
+    FINISH(2);
+OP_END
diff --git a/vm/mterp/c/OP_MOVE_OBJECT.c b/vm/mterp/c/OP_MOVE_OBJECT.c
new file mode 100644
index 0000000..bc03c80
--- /dev/null
+++ b/vm/mterp/c/OP_MOVE_OBJECT.c
@@ -0,0 +1,2 @@
+%include "c/OP_MOVE.c"
+//OP_END
diff --git a/vm/mterp/c/OP_MOVE_OBJECT_16.c b/vm/mterp/c/OP_MOVE_OBJECT_16.c
new file mode 100644
index 0000000..336a17e
--- /dev/null
+++ b/vm/mterp/c/OP_MOVE_OBJECT_16.c
@@ -0,0 +1,2 @@
+%include "c/OP_MOVE_16.c"
+//OP_END
diff --git a/vm/mterp/c/OP_MOVE_OBJECT_FROM16.c b/vm/mterp/c/OP_MOVE_OBJECT_FROM16.c
new file mode 100644
index 0000000..c65ecf9
--- /dev/null
+++ b/vm/mterp/c/OP_MOVE_OBJECT_FROM16.c
@@ -0,0 +1,2 @@
+%include "c/OP_MOVE_FROM16.c"
+//OP_END
diff --git a/vm/mterp/c/OP_MOVE_RESULT.c b/vm/mterp/c/OP_MOVE_RESULT.c
new file mode 100644
index 0000000..ddf535b
--- /dev/null
+++ b/vm/mterp/c/OP_MOVE_RESULT.c
@@ -0,0 +1,8 @@
+HANDLE_OPCODE($opcode /*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
diff --git a/vm/mterp/c/OP_MOVE_RESULT_OBJECT.c b/vm/mterp/c/OP_MOVE_RESULT_OBJECT.c
new file mode 100644
index 0000000..a52f7c2
--- /dev/null
+++ b/vm/mterp/c/OP_MOVE_RESULT_OBJECT.c
@@ -0,0 +1,2 @@
+%include "c/OP_MOVE_RESULT.c"
+//OP_END
diff --git a/vm/mterp/c/OP_MOVE_RESULT_WIDE.c b/vm/mterp/c/OP_MOVE_RESULT_WIDE.c
new file mode 100644
index 0000000..f6ec8d9
--- /dev/null
+++ b/vm/mterp/c/OP_MOVE_RESULT_WIDE.c
@@ -0,0 +1,6 @@
+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
diff --git a/vm/mterp/c/OP_MOVE_WIDE.c b/vm/mterp/c/OP_MOVE_WIDE.c
new file mode 100644
index 0000000..9ee323d
--- /dev/null
+++ b/vm/mterp/c/OP_MOVE_WIDE.c
@@ -0,0 +1,10 @@
+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
diff --git a/vm/mterp/c/OP_MOVE_WIDE_16.c b/vm/mterp/c/OP_MOVE_WIDE_16.c
new file mode 100644
index 0000000..e3d0e16
--- /dev/null
+++ b/vm/mterp/c/OP_MOVE_WIDE_16.c
@@ -0,0 +1,8 @@
+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
diff --git a/vm/mterp/c/OP_MOVE_WIDE_FROM16.c b/vm/mterp/c/OP_MOVE_WIDE_FROM16.c
new file mode 100644
index 0000000..cdbaa2e
--- /dev/null
+++ b/vm/mterp/c/OP_MOVE_WIDE_FROM16.c
@@ -0,0 +1,8 @@
+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
diff --git a/vm/mterp/c/OP_MUL_DOUBLE.c b/vm/mterp/c/OP_MUL_DOUBLE.c
new file mode 100644
index 0000000..3e65efa
--- /dev/null
+++ b/vm/mterp/c/OP_MUL_DOUBLE.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_DOUBLE(OP_MUL_DOUBLE, "mul", *)
+OP_END
diff --git a/vm/mterp/c/OP_MUL_DOUBLE_2ADDR.c b/vm/mterp/c/OP_MUL_DOUBLE_2ADDR.c
new file mode 100644
index 0000000..905b6a7
--- /dev/null
+++ b/vm/mterp/c/OP_MUL_DOUBLE_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_DOUBLE_2ADDR(OP_MUL_DOUBLE_2ADDR, "mul", *)
+OP_END
diff --git a/vm/mterp/c/OP_MUL_FLOAT.c b/vm/mterp/c/OP_MUL_FLOAT.c
new file mode 100644
index 0000000..310495c
--- /dev/null
+++ b/vm/mterp/c/OP_MUL_FLOAT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_FLOAT(OP_MUL_FLOAT, "mul", *)
+OP_END
diff --git a/vm/mterp/c/OP_MUL_FLOAT_2ADDR.c b/vm/mterp/c/OP_MUL_FLOAT_2ADDR.c
new file mode 100644
index 0000000..03623ca
--- /dev/null
+++ b/vm/mterp/c/OP_MUL_FLOAT_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_FLOAT_2ADDR(OP_MUL_FLOAT_2ADDR, "mul", *)
+OP_END
diff --git a/vm/mterp/c/OP_MUL_INT.c b/vm/mterp/c/OP_MUL_INT.c
new file mode 100644
index 0000000..19c2c48
--- /dev/null
+++ b/vm/mterp/c/OP_MUL_INT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT(OP_MUL_INT, "mul", *, false)
+OP_END
diff --git a/vm/mterp/c/OP_MUL_INT_2ADDR.c b/vm/mterp/c/OP_MUL_INT_2ADDR.c
new file mode 100644
index 0000000..d9619a6
--- /dev/null
+++ b/vm/mterp/c/OP_MUL_INT_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_2ADDR(OP_MUL_INT_2ADDR, "mul", *, false)
+OP_END
diff --git a/vm/mterp/c/OP_MUL_INT_LIT16.c b/vm/mterp/c/OP_MUL_INT_LIT16.c
new file mode 100644
index 0000000..b8eb4ba
--- /dev/null
+++ b/vm/mterp/c/OP_MUL_INT_LIT16.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_LIT16(OP_MUL_INT_LIT16, "mul", (s4), *, false)
+OP_END
diff --git a/vm/mterp/c/OP_MUL_INT_LIT8.c b/vm/mterp/c/OP_MUL_INT_LIT8.c
new file mode 100644
index 0000000..0d3a9a5
--- /dev/null
+++ b/vm/mterp/c/OP_MUL_INT_LIT8.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_LIT8(OP_MUL_INT_LIT8,   "mul", *, false)
+OP_END
diff --git a/vm/mterp/c/OP_MUL_LONG.c b/vm/mterp/c/OP_MUL_LONG.c
new file mode 100644
index 0000000..145acbc
--- /dev/null
+++ b/vm/mterp/c/OP_MUL_LONG.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_LONG(OP_MUL_LONG, "mul", *, false)
+OP_END
diff --git a/vm/mterp/c/OP_MUL_LONG_2ADDR.c b/vm/mterp/c/OP_MUL_LONG_2ADDR.c
new file mode 100644
index 0000000..9cebcaa
--- /dev/null
+++ b/vm/mterp/c/OP_MUL_LONG_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_LONG_2ADDR(OP_MUL_LONG_2ADDR, "mul", *, false)
+OP_END
diff --git a/vm/mterp/c/OP_NEG_DOUBLE.c b/vm/mterp/c/OP_NEG_DOUBLE.c
new file mode 100644
index 0000000..805082c
--- /dev/null
+++ b/vm/mterp/c/OP_NEG_DOUBLE.c
@@ -0,0 +1,2 @@
+HANDLE_UNOP(OP_NEG_DOUBLE, "neg-double", -, , _DOUBLE)
+OP_END
diff --git a/vm/mterp/c/OP_NEG_FLOAT.c b/vm/mterp/c/OP_NEG_FLOAT.c
new file mode 100644
index 0000000..00e14f5
--- /dev/null
+++ b/vm/mterp/c/OP_NEG_FLOAT.c
@@ -0,0 +1,2 @@
+HANDLE_UNOP(OP_NEG_FLOAT, "neg-float", -, , _FLOAT)
+OP_END
diff --git a/vm/mterp/c/OP_NEG_INT.c b/vm/mterp/c/OP_NEG_INT.c
new file mode 100644
index 0000000..9b97bef
--- /dev/null
+++ b/vm/mterp/c/OP_NEG_INT.c
@@ -0,0 +1,2 @@
+HANDLE_UNOP(OP_NEG_INT, "neg-int", -, , )
+OP_END
diff --git a/vm/mterp/c/OP_NEG_LONG.c b/vm/mterp/c/OP_NEG_LONG.c
new file mode 100644
index 0000000..52d553a
--- /dev/null
+++ b/vm/mterp/c/OP_NEG_LONG.c
@@ -0,0 +1,2 @@
+HANDLE_UNOP(OP_NEG_LONG, "neg-long", -, , _WIDE)
+OP_END
diff --git a/vm/mterp/c/OP_NEW_ARRAY.c b/vm/mterp/c/OP_NEW_ARRAY.c
new file mode 100644
index 0000000..dd20505
--- /dev/null
+++ b/vm/mterp/c/OP_NEW_ARRAY.c
@@ -0,0 +1,36 @@
+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(method->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
+
diff --git a/vm/mterp/c/OP_NEW_INSTANCE.c b/vm/mterp/c/OP_NEW_INSTANCE.c
new file mode 100644
index 0000000..86da6e3
--- /dev/null
+++ b/vm/mterp/c/OP_NEW_INSTANCE.c
@@ -0,0 +1,43 @@
+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(method->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
diff --git a/vm/mterp/c/OP_NOP.c b/vm/mterp/c/OP_NOP.c
new file mode 100644
index 0000000..d9fd744
--- /dev/null
+++ b/vm/mterp/c/OP_NOP.c
@@ -0,0 +1,3 @@
+HANDLE_OPCODE(OP_NOP)
+    FINISH(1);
+OP_END
diff --git a/vm/mterp/c/OP_NOT_INT.c b/vm/mterp/c/OP_NOT_INT.c
new file mode 100644
index 0000000..e585f62
--- /dev/null
+++ b/vm/mterp/c/OP_NOT_INT.c
@@ -0,0 +1,2 @@
+HANDLE_UNOP(OP_NOT_INT, "not-int", , ^ 0xffffffff, )
+OP_END
diff --git a/vm/mterp/c/OP_NOT_LONG.c b/vm/mterp/c/OP_NOT_LONG.c
new file mode 100644
index 0000000..4aafe8c
--- /dev/null
+++ b/vm/mterp/c/OP_NOT_LONG.c
@@ -0,0 +1,2 @@
+HANDLE_UNOP(OP_NOT_LONG, "not-long", , & 0xffffffffffffffffULL, _WIDE)
+OP_END
diff --git a/vm/mterp/c/OP_OR_INT.c b/vm/mterp/c/OP_OR_INT.c
new file mode 100644
index 0000000..f1a74b1
--- /dev/null
+++ b/vm/mterp/c/OP_OR_INT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT(OP_OR_INT,  "or",  |, false)
+OP_END
diff --git a/vm/mterp/c/OP_OR_INT_2ADDR.c b/vm/mterp/c/OP_OR_INT_2ADDR.c
new file mode 100644
index 0000000..f53f06d
--- /dev/null
+++ b/vm/mterp/c/OP_OR_INT_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_2ADDR(OP_OR_INT_2ADDR,  "or", |, false)
+OP_END
diff --git a/vm/mterp/c/OP_OR_INT_LIT16.c b/vm/mterp/c/OP_OR_INT_LIT16.c
new file mode 100644
index 0000000..8db4686
--- /dev/null
+++ b/vm/mterp/c/OP_OR_INT_LIT16.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_LIT16(OP_OR_INT_LIT16,  "or",  (s4), |, false)
+OP_END
diff --git a/vm/mterp/c/OP_OR_INT_LIT8.c b/vm/mterp/c/OP_OR_INT_LIT8.c
new file mode 100644
index 0000000..0a31a25
--- /dev/null
+++ b/vm/mterp/c/OP_OR_INT_LIT8.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_LIT8(OP_OR_INT_LIT8,    "or",  |, false)
+OP_END
diff --git a/vm/mterp/c/OP_OR_LONG.c b/vm/mterp/c/OP_OR_LONG.c
new file mode 100644
index 0000000..8d977a0
--- /dev/null
+++ b/vm/mterp/c/OP_OR_LONG.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_LONG(OP_OR_LONG,  "or", |, false)
+OP_END
diff --git a/vm/mterp/c/OP_OR_LONG_2ADDR.c b/vm/mterp/c/OP_OR_LONG_2ADDR.c
new file mode 100644
index 0000000..5977d49
--- /dev/null
+++ b/vm/mterp/c/OP_OR_LONG_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_LONG_2ADDR(OP_OR_LONG_2ADDR,  "or", |, false)
+OP_END
diff --git a/vm/mterp/c/OP_PACKED_SWITCH.c b/vm/mterp/c/OP_PACKED_SWITCH.c
new file mode 100644
index 0000000..5e490cf
--- /dev/null
+++ b/vm/mterp/c/OP_PACKED_SWITCH.c
@@ -0,0 +1,29 @@
+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 < method->insns ||
+            switchData >= method->insns + dvmGetMethodInsnsSize(method))
+        {
+            /* 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
diff --git a/vm/mterp/c/OP_REM_DOUBLE.c b/vm/mterp/c/OP_REM_DOUBLE.c
new file mode 100644
index 0000000..343e25e
--- /dev/null
+++ b/vm/mterp/c/OP_REM_DOUBLE.c
@@ -0,0 +1,13 @@
+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
diff --git a/vm/mterp/c/OP_REM_DOUBLE_2ADDR.c b/vm/mterp/c/OP_REM_DOUBLE_2ADDR.c
new file mode 100644
index 0000000..392eacf
--- /dev/null
+++ b/vm/mterp/c/OP_REM_DOUBLE_2ADDR.c
@@ -0,0 +1,8 @@
+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
diff --git a/vm/mterp/c/OP_REM_FLOAT.c b/vm/mterp/c/OP_REM_FLOAT.c
new file mode 100644
index 0000000..9604b30
--- /dev/null
+++ b/vm/mterp/c/OP_REM_FLOAT.c
@@ -0,0 +1,13 @@
+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
diff --git a/vm/mterp/c/OP_REM_FLOAT_2ADDR.c b/vm/mterp/c/OP_REM_FLOAT_2ADDR.c
new file mode 100644
index 0000000..87bb31e
--- /dev/null
+++ b/vm/mterp/c/OP_REM_FLOAT_2ADDR.c
@@ -0,0 +1,8 @@
+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
diff --git a/vm/mterp/c/OP_REM_INT.c b/vm/mterp/c/OP_REM_INT.c
new file mode 100644
index 0000000..fb277bd
--- /dev/null
+++ b/vm/mterp/c/OP_REM_INT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT(OP_REM_INT, "rem", %, true)
+OP_END
diff --git a/vm/mterp/c/OP_REM_INT_2ADDR.c b/vm/mterp/c/OP_REM_INT_2ADDR.c
new file mode 100644
index 0000000..dca7647
--- /dev/null
+++ b/vm/mterp/c/OP_REM_INT_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_2ADDR(OP_REM_INT_2ADDR, "rem", %, true)
+OP_END
diff --git a/vm/mterp/c/OP_REM_INT_LIT16.c b/vm/mterp/c/OP_REM_INT_LIT16.c
new file mode 100644
index 0000000..69e9b4e
--- /dev/null
+++ b/vm/mterp/c/OP_REM_INT_LIT16.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_LIT16(OP_REM_INT_LIT16, "rem", (s4), %, true)
+OP_END
diff --git a/vm/mterp/c/OP_REM_INT_LIT8.c b/vm/mterp/c/OP_REM_INT_LIT8.c
new file mode 100644
index 0000000..49cc2f6
--- /dev/null
+++ b/vm/mterp/c/OP_REM_INT_LIT8.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_LIT8(OP_REM_INT_LIT8,   "rem", %, true)
+OP_END
diff --git a/vm/mterp/c/OP_REM_LONG.c b/vm/mterp/c/OP_REM_LONG.c
new file mode 100644
index 0000000..04ccc1b
--- /dev/null
+++ b/vm/mterp/c/OP_REM_LONG.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_LONG(OP_REM_LONG, "rem", %, true)
+OP_END
diff --git a/vm/mterp/c/OP_REM_LONG_2ADDR.c b/vm/mterp/c/OP_REM_LONG_2ADDR.c
new file mode 100644
index 0000000..5879dd6
--- /dev/null
+++ b/vm/mterp/c/OP_REM_LONG_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_LONG_2ADDR(OP_REM_LONG_2ADDR, "rem", %, true)
+OP_END
diff --git a/vm/mterp/c/OP_RETURN.c b/vm/mterp/c/OP_RETURN.c
new file mode 100644
index 0000000..6d5ffa3
--- /dev/null
+++ b/vm/mterp/c/OP_RETURN.c
@@ -0,0 +1,7 @@
+HANDLE_OPCODE($opcode /*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
diff --git a/vm/mterp/c/OP_RETURN_OBJECT.c b/vm/mterp/c/OP_RETURN_OBJECT.c
new file mode 100644
index 0000000..49bf01e
--- /dev/null
+++ b/vm/mterp/c/OP_RETURN_OBJECT.c
@@ -0,0 +1,2 @@
+%include "c/OP_RETURN.c"
+//OP_END
diff --git a/vm/mterp/c/OP_RETURN_VOID.c b/vm/mterp/c/OP_RETURN_VOID.c
new file mode 100644
index 0000000..38649e9
--- /dev/null
+++ b/vm/mterp/c/OP_RETURN_VOID.c
@@ -0,0 +1,7 @@
+HANDLE_OPCODE(OP_RETURN_VOID /**/)
+    ILOGV("|return-void");
+#ifndef NDEBUG
+    retval.j = 0xababababULL;    // placate valgrind
+#endif
+    GOTO(returnFromMethod);
+OP_END
diff --git a/vm/mterp/c/OP_RETURN_WIDE.c b/vm/mterp/c/OP_RETURN_WIDE.c
new file mode 100644
index 0000000..5bd9849
--- /dev/null
+++ b/vm/mterp/c/OP_RETURN_WIDE.c
@@ -0,0 +1,6 @@
+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
diff --git a/vm/mterp/c/OP_RSUB_INT.c b/vm/mterp/c/OP_RSUB_INT.c
new file mode 100644
index 0000000..336ca55
--- /dev/null
+++ b/vm/mterp/c/OP_RSUB_INT.c
@@ -0,0 +1,10 @@
+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
diff --git a/vm/mterp/c/OP_RSUB_INT_LIT8.c b/vm/mterp/c/OP_RSUB_INT_LIT8.c
new file mode 100644
index 0000000..742854b
--- /dev/null
+++ b/vm/mterp/c/OP_RSUB_INT_LIT8.c
@@ -0,0 +1,12 @@
+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
diff --git a/vm/mterp/c/OP_SGET.c b/vm/mterp/c/OP_SGET.c
new file mode 100644
index 0000000..5297cd7
--- /dev/null
+++ b/vm/mterp/c/OP_SGET.c
@@ -0,0 +1,2 @@
+HANDLE_SGET_X(OP_SGET,                  "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_SGET_BOOLEAN.c b/vm/mterp/c/OP_SGET_BOOLEAN.c
new file mode 100644
index 0000000..7c5d45e
--- /dev/null
+++ b/vm/mterp/c/OP_SGET_BOOLEAN.c
@@ -0,0 +1,2 @@
+HANDLE_SGET_X(OP_SGET_BOOLEAN,          "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_SGET_BYTE.c b/vm/mterp/c/OP_SGET_BYTE.c
new file mode 100644
index 0000000..b37cab4
--- /dev/null
+++ b/vm/mterp/c/OP_SGET_BYTE.c
@@ -0,0 +1,2 @@
+HANDLE_SGET_X(OP_SGET_BYTE,             "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_SGET_CHAR.c b/vm/mterp/c/OP_SGET_CHAR.c
new file mode 100644
index 0000000..7ede5ec
--- /dev/null
+++ b/vm/mterp/c/OP_SGET_CHAR.c
@@ -0,0 +1,2 @@
+HANDLE_SGET_X(OP_SGET_CHAR,             "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_SGET_OBJECT.c b/vm/mterp/c/OP_SGET_OBJECT.c
new file mode 100644
index 0000000..9f3b63d
--- /dev/null
+++ b/vm/mterp/c/OP_SGET_OBJECT.c
@@ -0,0 +1,2 @@
+HANDLE_SGET_X(OP_SGET_OBJECT,           "-object", Object, _AS_OBJECT)
+OP_END
diff --git a/vm/mterp/c/OP_SGET_SHORT.c b/vm/mterp/c/OP_SGET_SHORT.c
new file mode 100644
index 0000000..cd1fe4c
--- /dev/null
+++ b/vm/mterp/c/OP_SGET_SHORT.c
@@ -0,0 +1,2 @@
+HANDLE_SGET_X(OP_SGET_SHORT,            "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_SGET_WIDE.c b/vm/mterp/c/OP_SGET_WIDE.c
new file mode 100644
index 0000000..817c6e7
--- /dev/null
+++ b/vm/mterp/c/OP_SGET_WIDE.c
@@ -0,0 +1,2 @@
+HANDLE_SGET_X(OP_SGET_WIDE,             "-wide", Long, _WIDE)
+OP_END
diff --git a/vm/mterp/c/OP_SHL_INT.c b/vm/mterp/c/OP_SHL_INT.c
new file mode 100644
index 0000000..e32af49
--- /dev/null
+++ b/vm/mterp/c/OP_SHL_INT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_SHX_INT(OP_SHL_INT, "shl", (s4), <<)
+OP_END
diff --git a/vm/mterp/c/OP_SHL_INT_2ADDR.c b/vm/mterp/c/OP_SHL_INT_2ADDR.c
new file mode 100644
index 0000000..c5f5399
--- /dev/null
+++ b/vm/mterp/c/OP_SHL_INT_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_SHX_INT_2ADDR(OP_SHL_INT_2ADDR, "shl", (s4), <<)
+OP_END
diff --git a/vm/mterp/c/OP_SHL_INT_LIT8.c b/vm/mterp/c/OP_SHL_INT_LIT8.c
new file mode 100644
index 0000000..009d14e
--- /dev/null
+++ b/vm/mterp/c/OP_SHL_INT_LIT8.c
@@ -0,0 +1,2 @@
+HANDLE_OP_SHX_INT_LIT8(OP_SHL_INT_LIT8,   "shl", (s4), <<)
+OP_END
diff --git a/vm/mterp/c/OP_SHL_LONG.c b/vm/mterp/c/OP_SHL_LONG.c
new file mode 100644
index 0000000..f6b502a
--- /dev/null
+++ b/vm/mterp/c/OP_SHL_LONG.c
@@ -0,0 +1,2 @@
+HANDLE_OP_SHX_LONG(OP_SHL_LONG, "shl", (s8), <<)
+OP_END
diff --git a/vm/mterp/c/OP_SHL_LONG_2ADDR.c b/vm/mterp/c/OP_SHL_LONG_2ADDR.c
new file mode 100644
index 0000000..b8a9954
--- /dev/null
+++ b/vm/mterp/c/OP_SHL_LONG_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_SHX_LONG_2ADDR(OP_SHL_LONG_2ADDR, "shl", (s8), <<)
+OP_END
diff --git a/vm/mterp/c/OP_SHR_INT.c b/vm/mterp/c/OP_SHR_INT.c
new file mode 100644
index 0000000..3834824
--- /dev/null
+++ b/vm/mterp/c/OP_SHR_INT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_SHX_INT(OP_SHR_INT, "shr", (s4), >>)
+OP_END
diff --git a/vm/mterp/c/OP_SHR_INT_2ADDR.c b/vm/mterp/c/OP_SHR_INT_2ADDR.c
new file mode 100644
index 0000000..c76c178
--- /dev/null
+++ b/vm/mterp/c/OP_SHR_INT_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_SHX_INT_2ADDR(OP_SHR_INT_2ADDR, "shr", (s4), >>)
+OP_END
diff --git a/vm/mterp/c/OP_SHR_INT_LIT8.c b/vm/mterp/c/OP_SHR_INT_LIT8.c
new file mode 100644
index 0000000..e2657d7
--- /dev/null
+++ b/vm/mterp/c/OP_SHR_INT_LIT8.c
@@ -0,0 +1,2 @@
+HANDLE_OP_SHX_INT_LIT8(OP_SHR_INT_LIT8,   "shr", (s4), >>)
+OP_END
diff --git a/vm/mterp/c/OP_SHR_LONG.c b/vm/mterp/c/OP_SHR_LONG.c
new file mode 100644
index 0000000..357a666
--- /dev/null
+++ b/vm/mterp/c/OP_SHR_LONG.c
@@ -0,0 +1,2 @@
+HANDLE_OP_SHX_LONG(OP_SHR_LONG, "shr", (s8), >>)
+OP_END
diff --git a/vm/mterp/c/OP_SHR_LONG_2ADDR.c b/vm/mterp/c/OP_SHR_LONG_2ADDR.c
new file mode 100644
index 0000000..43e27ea
--- /dev/null
+++ b/vm/mterp/c/OP_SHR_LONG_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_SHX_LONG_2ADDR(OP_SHR_LONG_2ADDR, "shr", (s8), >>)
+OP_END
diff --git a/vm/mterp/c/OP_SPARSE_SWITCH.c b/vm/mterp/c/OP_SPARSE_SWITCH.c
new file mode 100644
index 0000000..4886830
--- /dev/null
+++ b/vm/mterp/c/OP_SPARSE_SWITCH.c
@@ -0,0 +1,29 @@
+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 < method->insns ||
+            switchData >= method->insns + dvmGetMethodInsnsSize(method))
+        {
+            /* 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
diff --git a/vm/mterp/c/OP_SPUT.c b/vm/mterp/c/OP_SPUT.c
new file mode 100644
index 0000000..286e64c
--- /dev/null
+++ b/vm/mterp/c/OP_SPUT.c
@@ -0,0 +1,2 @@
+HANDLE_SPUT_X(OP_SPUT,                  "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_SPUT_BOOLEAN.c b/vm/mterp/c/OP_SPUT_BOOLEAN.c
new file mode 100644
index 0000000..55ceb11
--- /dev/null
+++ b/vm/mterp/c/OP_SPUT_BOOLEAN.c
@@ -0,0 +1,2 @@
+HANDLE_SPUT_X(OP_SPUT_BOOLEAN,          "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_SPUT_BYTE.c b/vm/mterp/c/OP_SPUT_BYTE.c
new file mode 100644
index 0000000..d242fe1
--- /dev/null
+++ b/vm/mterp/c/OP_SPUT_BYTE.c
@@ -0,0 +1,2 @@
+HANDLE_SPUT_X(OP_SPUT_BYTE,             "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_SPUT_CHAR.c b/vm/mterp/c/OP_SPUT_CHAR.c
new file mode 100644
index 0000000..18a2f06
--- /dev/null
+++ b/vm/mterp/c/OP_SPUT_CHAR.c
@@ -0,0 +1,2 @@
+HANDLE_SPUT_X(OP_SPUT_CHAR,             "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_SPUT_OBJECT.c b/vm/mterp/c/OP_SPUT_OBJECT.c
new file mode 100644
index 0000000..fb223d6
--- /dev/null
+++ b/vm/mterp/c/OP_SPUT_OBJECT.c
@@ -0,0 +1,2 @@
+HANDLE_SPUT_X(OP_SPUT_OBJECT,           "-object", Object, _AS_OBJECT)
+OP_END
diff --git a/vm/mterp/c/OP_SPUT_SHORT.c b/vm/mterp/c/OP_SPUT_SHORT.c
new file mode 100644
index 0000000..c6cd8d6
--- /dev/null
+++ b/vm/mterp/c/OP_SPUT_SHORT.c
@@ -0,0 +1,2 @@
+HANDLE_SPUT_X(OP_SPUT_SHORT,            "", Int, )
+OP_END
diff --git a/vm/mterp/c/OP_SPUT_WIDE.c b/vm/mterp/c/OP_SPUT_WIDE.c
new file mode 100644
index 0000000..0c74651
--- /dev/null
+++ b/vm/mterp/c/OP_SPUT_WIDE.c
@@ -0,0 +1,2 @@
+HANDLE_SPUT_X(OP_SPUT_WIDE,             "-wide", Long, _WIDE)
+OP_END
diff --git a/vm/mterp/c/OP_SUB_DOUBLE.c b/vm/mterp/c/OP_SUB_DOUBLE.c
new file mode 100644
index 0000000..64a112d
--- /dev/null
+++ b/vm/mterp/c/OP_SUB_DOUBLE.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_DOUBLE(OP_SUB_DOUBLE, "sub", -)
+OP_END
diff --git a/vm/mterp/c/OP_SUB_DOUBLE_2ADDR.c b/vm/mterp/c/OP_SUB_DOUBLE_2ADDR.c
new file mode 100644
index 0000000..5870400
--- /dev/null
+++ b/vm/mterp/c/OP_SUB_DOUBLE_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_DOUBLE_2ADDR(OP_SUB_DOUBLE_2ADDR, "sub", -)
+OP_END
diff --git a/vm/mterp/c/OP_SUB_FLOAT.c b/vm/mterp/c/OP_SUB_FLOAT.c
new file mode 100644
index 0000000..96c5fbd
--- /dev/null
+++ b/vm/mterp/c/OP_SUB_FLOAT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_FLOAT(OP_SUB_FLOAT, "sub", -)
+OP_END
diff --git a/vm/mterp/c/OP_SUB_FLOAT_2ADDR.c b/vm/mterp/c/OP_SUB_FLOAT_2ADDR.c
new file mode 100644
index 0000000..802935c
--- /dev/null
+++ b/vm/mterp/c/OP_SUB_FLOAT_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_FLOAT_2ADDR(OP_SUB_FLOAT_2ADDR, "sub", -)
+OP_END
diff --git a/vm/mterp/c/OP_SUB_INT.c b/vm/mterp/c/OP_SUB_INT.c
new file mode 100644
index 0000000..f16524b
--- /dev/null
+++ b/vm/mterp/c/OP_SUB_INT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT(OP_SUB_INT, "sub", -, false)
+OP_END
diff --git a/vm/mterp/c/OP_SUB_INT_2ADDR.c b/vm/mterp/c/OP_SUB_INT_2ADDR.c
new file mode 100644
index 0000000..744eb58
--- /dev/null
+++ b/vm/mterp/c/OP_SUB_INT_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_2ADDR(OP_SUB_INT_2ADDR, "sub", -, false)
+OP_END
diff --git a/vm/mterp/c/OP_SUB_LONG.c b/vm/mterp/c/OP_SUB_LONG.c
new file mode 100644
index 0000000..07f782c
--- /dev/null
+++ b/vm/mterp/c/OP_SUB_LONG.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_LONG(OP_SUB_LONG, "sub", -, false)
+OP_END
diff --git a/vm/mterp/c/OP_SUB_LONG_2ADDR.c b/vm/mterp/c/OP_SUB_LONG_2ADDR.c
new file mode 100644
index 0000000..09c4653
--- /dev/null
+++ b/vm/mterp/c/OP_SUB_LONG_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_LONG_2ADDR(OP_SUB_LONG_2ADDR, "sub", -, false)
+OP_END
diff --git a/vm/mterp/c/OP_THROW.c b/vm/mterp/c/OP_THROW.c
new file mode 100644
index 0000000..12e002a
--- /dev/null
+++ b/vm/mterp/c/OP_THROW.c
@@ -0,0 +1,17 @@
+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
diff --git a/vm/mterp/c/OP_UNUSED_3E.c b/vm/mterp/c/OP_UNUSED_3E.c
new file mode 100644
index 0000000..9ecf8e3
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_3E.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_3E)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_3F.c b/vm/mterp/c/OP_UNUSED_3F.c
new file mode 100644
index 0000000..9d1d68d
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_3F.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_3F)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_40.c b/vm/mterp/c/OP_UNUSED_40.c
new file mode 100644
index 0000000..f73a59c
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_40.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_40)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_41.c b/vm/mterp/c/OP_UNUSED_41.c
new file mode 100644
index 0000000..38747e6
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_41.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_41)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_42.c b/vm/mterp/c/OP_UNUSED_42.c
new file mode 100644
index 0000000..154d293
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_42.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_42)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_43.c b/vm/mterp/c/OP_UNUSED_43.c
new file mode 100644
index 0000000..c7e702c
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_43.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_43)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_73.c b/vm/mterp/c/OP_UNUSED_73.c
new file mode 100644
index 0000000..85aa95f
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_73.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_73)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_79.c b/vm/mterp/c/OP_UNUSED_79.c
new file mode 100644
index 0000000..1fa86e9
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_79.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_79)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_7A.c b/vm/mterp/c/OP_UNUSED_7A.c
new file mode 100644
index 0000000..beab006
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_7A.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_7A)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_E3.c b/vm/mterp/c/OP_UNUSED_E3.c
new file mode 100644
index 0000000..d52836b
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_E3.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_E3)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_E4.c b/vm/mterp/c/OP_UNUSED_E4.c
new file mode 100644
index 0000000..30a714f
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_E4.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_E4)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_E5.c b/vm/mterp/c/OP_UNUSED_E5.c
new file mode 100644
index 0000000..0cc6ff9
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_E5.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_E5)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_E6.c b/vm/mterp/c/OP_UNUSED_E6.c
new file mode 100644
index 0000000..255dcd1
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_E6.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_E6)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_E7.c b/vm/mterp/c/OP_UNUSED_E7.c
new file mode 100644
index 0000000..b910f71
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_E7.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_E7)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_E8.c b/vm/mterp/c/OP_UNUSED_E8.c
new file mode 100644
index 0000000..ba11d34
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_E8.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_E8)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_E9.c b/vm/mterp/c/OP_UNUSED_E9.c
new file mode 100644
index 0000000..6f4d8fb
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_E9.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_E9)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_EA.c b/vm/mterp/c/OP_UNUSED_EA.c
new file mode 100644
index 0000000..8456083
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_EA.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_EA)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_EB.c b/vm/mterp/c/OP_UNUSED_EB.c
new file mode 100644
index 0000000..3525262
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_EB.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_EB)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_EC.c b/vm/mterp/c/OP_UNUSED_EC.c
new file mode 100644
index 0000000..fcb8c2e
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_EC.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_EC)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_ED.c b/vm/mterp/c/OP_UNUSED_ED.c
new file mode 100644
index 0000000..c11348f
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_ED.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_ED)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_EF.c b/vm/mterp/c/OP_UNUSED_EF.c
new file mode 100644
index 0000000..c5e1863
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_EF.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_EF)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_F1.c b/vm/mterp/c/OP_UNUSED_F1.c
new file mode 100644
index 0000000..af26195
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_F1.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_F1)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_FC.c b/vm/mterp/c/OP_UNUSED_FC.c
new file mode 100644
index 0000000..4b49684
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_FC.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_FC)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_FD.c b/vm/mterp/c/OP_UNUSED_FD.c
new file mode 100644
index 0000000..c10efaf
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_FD.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_FD)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_FE.c b/vm/mterp/c/OP_UNUSED_FE.c
new file mode 100644
index 0000000..7b097d9
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_FE.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_FE)
+OP_END
diff --git a/vm/mterp/c/OP_UNUSED_FF.c b/vm/mterp/c/OP_UNUSED_FF.c
new file mode 100644
index 0000000..ed9e11c
--- /dev/null
+++ b/vm/mterp/c/OP_UNUSED_FF.c
@@ -0,0 +1,2 @@
+HANDLE_OPCODE(OP_UNUSED_FF)
+OP_END
diff --git a/vm/mterp/c/OP_USHR_INT.c b/vm/mterp/c/OP_USHR_INT.c
new file mode 100644
index 0000000..7596c94
--- /dev/null
+++ b/vm/mterp/c/OP_USHR_INT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_SHX_INT(OP_USHR_INT, "ushr", (u4), >>)
+OP_END
diff --git a/vm/mterp/c/OP_USHR_INT_2ADDR.c b/vm/mterp/c/OP_USHR_INT_2ADDR.c
new file mode 100644
index 0000000..5fa2b94
--- /dev/null
+++ b/vm/mterp/c/OP_USHR_INT_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_SHX_INT_2ADDR(OP_USHR_INT_2ADDR, "ushr", (u4), >>)
+OP_END
diff --git a/vm/mterp/c/OP_USHR_INT_LIT8.c b/vm/mterp/c/OP_USHR_INT_LIT8.c
new file mode 100644
index 0000000..0d325d7
--- /dev/null
+++ b/vm/mterp/c/OP_USHR_INT_LIT8.c
@@ -0,0 +1,2 @@
+HANDLE_OP_SHX_INT_LIT8(OP_USHR_INT_LIT8,  "ushr", (u4), >>)
+OP_END
diff --git a/vm/mterp/c/OP_USHR_LONG.c b/vm/mterp/c/OP_USHR_LONG.c
new file mode 100644
index 0000000..9b7e757
--- /dev/null
+++ b/vm/mterp/c/OP_USHR_LONG.c
@@ -0,0 +1,2 @@
+HANDLE_OP_SHX_LONG(OP_USHR_LONG, "ushr", (u8), >>)
+OP_END
diff --git a/vm/mterp/c/OP_USHR_LONG_2ADDR.c b/vm/mterp/c/OP_USHR_LONG_2ADDR.c
new file mode 100644
index 0000000..4ac0598
--- /dev/null
+++ b/vm/mterp/c/OP_USHR_LONG_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_SHX_LONG_2ADDR(OP_USHR_LONG_2ADDR, "ushr", (u8), >>)
+OP_END
diff --git a/vm/mterp/c/OP_XOR_INT.c b/vm/mterp/c/OP_XOR_INT.c
new file mode 100644
index 0000000..76b7791
--- /dev/null
+++ b/vm/mterp/c/OP_XOR_INT.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT(OP_XOR_INT, "xor", ^, false)
+OP_END
diff --git a/vm/mterp/c/OP_XOR_INT_2ADDR.c b/vm/mterp/c/OP_XOR_INT_2ADDR.c
new file mode 100644
index 0000000..9b8ff1d
--- /dev/null
+++ b/vm/mterp/c/OP_XOR_INT_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_2ADDR(OP_XOR_INT_2ADDR, "xor", ^, false)
+OP_END
diff --git a/vm/mterp/c/OP_XOR_INT_LIT16.c b/vm/mterp/c/OP_XOR_INT_LIT16.c
new file mode 100644
index 0000000..b9d3127
--- /dev/null
+++ b/vm/mterp/c/OP_XOR_INT_LIT16.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_LIT16(OP_XOR_INT_LIT16, "xor", (s4), ^, false)
+OP_END
diff --git a/vm/mterp/c/OP_XOR_INT_LIT8.c b/vm/mterp/c/OP_XOR_INT_LIT8.c
new file mode 100644
index 0000000..11c641a
--- /dev/null
+++ b/vm/mterp/c/OP_XOR_INT_LIT8.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_INT_LIT8(OP_XOR_INT_LIT8,   "xor", ^, false)
+OP_END
diff --git a/vm/mterp/c/OP_XOR_LONG.c b/vm/mterp/c/OP_XOR_LONG.c
new file mode 100644
index 0000000..289af20
--- /dev/null
+++ b/vm/mterp/c/OP_XOR_LONG.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_LONG(OP_XOR_LONG, "xor", ^, false)
+OP_END
diff --git a/vm/mterp/c/OP_XOR_LONG_2ADDR.c b/vm/mterp/c/OP_XOR_LONG_2ADDR.c
new file mode 100644
index 0000000..8e4976e
--- /dev/null
+++ b/vm/mterp/c/OP_XOR_LONG_2ADDR.c
@@ -0,0 +1,2 @@
+HANDLE_OP_X_LONG_2ADDR(OP_XOR_LONG_2ADDR, "xor", ^, false)
+OP_END
diff --git a/vm/mterp/c/footer.c b/vm/mterp/c/footer.c
new file mode 100644
index 0000000..89e7347
--- /dev/null
+++ b/vm/mterp/c/footer.c
@@ -0,0 +1,977 @@
+/*
+ * C footer.  This has some common code shared by the various targets.
+ */
+
+#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 }
+
+
+/*
+ * 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;
+        int* 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(method->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 == '[') {
+            /* create array of objects or array of arrays */
+            /* TODO: need some work in the verifier before we allow this */
+            LOGE("fnao not implemented\n");
+            dvmThrowException("Ljava/lang/InternalError;",
+                "filled-new-array not implemented for reference types");
+            GOTO(exceptionThrown);
+        } else if (typeCh != 'I') {
+            /* TODO: requires multiple "fill in" loops with different widths */
+            LOGE("non-int not implemented\n");
+            dvmThrowException("Ljava/lang/InternalError;",
+                "filled-new-array not implemented for anything but 'int'");
+            GOTO(exceptionThrown);
+        }
+
+        assert(strchr("BCIFZ", typeCh) != NULL);
+        newArray = dvmAllocPrimitiveArray(arrayClass->descriptor[1], vsrc1,
+                    ALLOC_DONT_TRACK);
+        if (newArray == NULL)
+            GOTO(exceptionThrown);
+
+        /*
+         * Fill in the elements.  It's legal for vsrc1 to be zero.
+         */
+        contents = (int*) 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 */
+
+        if (methodCallRange) {
+            /*
+             * The object against which we are executing a method is always
+             * in the first argument.
+             */
+            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 {
+            /*
+             * The object against which we are executing a method is always
+             * in the first argument.
+             */
+            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(method->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(method->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 >= method->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 = method->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 */
+
+        if (methodCallRange) {
+            /*
+             * The object against which we are executing a method is always
+             * in the first argument.
+             */
+            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 {
+            /*
+             * The object against which we are executing a method is always
+             * in the first argument.
+             */
+            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, method,
+                        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(method->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(method->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 */
+
+        if (methodCallRange) {
+            /*
+             * The object against which we are executing a method is always
+             * in the first argument.
+             */
+            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 {
+            /*
+             * The object against which we are executing a method is always
+             * in the first argument.
+             */
+            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 >= method->clazz->super->vtableCount) {
+            dvmThrowException("Ljava/lang/NoSuchMethodError;", NULL);
+            GOTO(exceptionThrown);
+        }
+#else
+        assert(ref < method->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 = method->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, method->clazz->descriptor, method->name,
+            method->signature);
+        //DUMP_REGS(method, fp);
+
+        saveArea = SAVEAREA_FROM_FP(fp);
+
+#ifdef EASY_GDB
+        debugSaveArea = saveArea;
+#endif
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_PROFILER)
+        TRACE_METHOD_EXIT(self, method);
+#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(false);
+        }
+
+        /* update thread FP, and reset local variables */
+        self->curFrame = fp;
+        method =
+#undef method       // ARRGH!
+            SAVEAREA_FROM_FP(fp)->method;
+#define method glue->method
+        //methodClass = method->clazz;
+        methodClassDex = method->clazz->pDvmDex;
+        pc = saveArea->savedPc;
+        ILOGD("> (return to %s.%s %s)", method->clazz->descriptor,
+            method->name, method->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 - method->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, method->name,
+            dvmLineNumFromPC(method, pc - method->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 - method->insns,
+                        exception, true, &catchFrame);
+            dvmDbgPostException(fp, pc - method->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 - method->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.
+         */
+        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(method),
+                dvmLineNumFromPC(method, pc - method->insns));
+#endif
+            dvmSetException(self, exception);
+            dvmReleaseTrackedAlloc(exception, self);
+            GOTO_BAIL(false);
+        }
+
+#if DVM_SHOW_EXCEPTION >= 3
+        {
+            const Method* catchMethod =
+#undef method
+                SAVEAREA_FROM_FP(fp)->method;
+#define method glue->method
+            LOGD("Exception %s thrown from %s:%d to %s:%d\n",
+                exception->clazz->descriptor, dvmGetMethodSourceFile(method),
+                dvmLineNumFromPC(method, pc - method->insns),
+                dvmGetMethodSourceFile(catchMethod),
+                dvmLineNumFromPC(catchMethod, catchRelPc));
+        }
+#endif
+
+        /*
+         * Adjust local variables to match self->curFrame and the
+         * updated PC.
+         */
+        //fp = (u4*) self->curFrame;
+        method =
+#undef method
+            SAVEAREA_FROM_FP(fp)->method;
+#define method glue->method
+        //methodClass = method->clazz;
+        methodClassDex = method->clazz->pDvmDex;
+        pc = method->insns + catchRelPc;
+        ILOGV("> pc <-- %s.%s %s", method->clazz->descriptor, method->name,
+            method->signature);
+        DUMP_REGS(method, 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)
+    {
+        vsrc1 = count; vdst = regs; methodToCall = _methodToCall;  /* ADDED */
+
+        //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 <= method->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 <= method->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;
+#undef method
+        newSaveArea->method = methodToCall;
+#define method glue->method
+
+        if (!dvmIsNativeMethod(methodToCall)) {
+            /*
+             * "Call" interpreted code.  Reposition the PC, update the
+             * frame pointer and other local state, and continue.
+             */
+            method = methodToCall;
+            methodClassDex = method->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", method->clazz->descriptor, method->name,
+                method->signature);
+            DUMP_REGS(method, 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(method, 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(method, 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,
+                method->clazz->descriptor, method->name,
+                method->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 - method->insns));
+                assert(false);
+            }
+        }
+    }
+    assert(false);      // should not get here
+GOTO_TARGET_END
+
+
+/* undefine "magic" name remapping */
+#undef retval
+#undef pc
+#undef fp
+#undef method
+#undef methodClassDex
+#undef self
+#undef debugTrackedRefStart
diff --git a/vm/mterp/c/header.c b/vm/mterp/c/header.c
new file mode 100644
index 0000000..1c3b20e
--- /dev/null
+++ b/vm/mterp/c/header.c
@@ -0,0 +1,39 @@
+/*
+ * 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
+
+
+#define GOTO_TARGET_DECL(_target, ...)                                      \
+    void dvmMterp_##_target(MterpGlue* glue, ## __VA_ARGS__);
+
+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);
+
diff --git a/vm/mterp/c/opcommon.c b/vm/mterp/c/opcommon.c
new file mode 100644
index 0000000..c3f824d
--- /dev/null
+++ b/vm/mterp/c/opcommon.c
@@ -0,0 +1,627 @@
+/*
+ * 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 method                  glue->method
+#define methodClassDex          glue->methodClassDex
+#define self                    glue->self
+//#define entryPoint              glue->entryPoint
+#define debugTrackedRefStart    glue->debugTrackedRefStart
+
+
+/*
+ * Replace the opcode definition 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.
+ */
+#undef HANDLE_OPCODE
+#undef OP_END
+#undef FINISH
+
+#define HANDLE_OPCODE(_op)                                                  \
+    void dvmMterp_##_op(MterpGlue* glue) {                                  \
+        u2 ref, vsrc1, vsrc2, vdst;                                         \
+        u2 inst = FETCH(0);
+
+#define OP_END }
+
+/*
+ * Like standard 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.
+ */
+#define GOTO(_target, ...)                                                  \
+    do {                                                                    \
+        dvmMterp_##_target(glue, ## __VA_ARGS__);                           \
+        return;                                                             \
+    } while(false)
+
+/*
+ * As a special case, "goto bail" turns into a longjmp.  "_switch" should be
+ * "true" if we need to switch to the other interpreter upon our return.
+ */
+#define GOTO_BAIL(_switch)                                                  \
+    dvmMterpStdBail(glue, _switch);
+
+/* for now, mterp is always a "standard" interpreter */
+#define INTERP_TYPE INTERP_STD
+
+/*
+ * Periodic checks macro, slightly modified.
+ */
+#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(true);                                                \
+        }                                                                   \
+    }
+
+
+/*
+ * ===========================================================================
+ *
+ * What follows are the "common" opcode definitions copied & pasted from the
+ * basic interpreter.  The only changes that need to be made to the original
+ * sources are:
+ *  - replace "goto exceptionThrown" with "GOTO(exceptionThrown)"
+ *
+ * ===========================================================================
+ */
+
+
+#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) {                                                      \
+            if (GET_REGISTER(vsrc2) == 0) {                                 \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO(exceptionThrown);                                      \
+            }                                                               \
+        }                                                                   \
+        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, _cast, _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) {                                                      \
+            if ((s2) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO(exceptionThrown);                                      \
+            }                                                               \
+        }                                                                   \
+        SET_REGISTER(vdst,                                                  \
+            _cast 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) {                                                      \
+            if ((s1) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO(exceptionThrown);                                      \
+            }                                                               \
+        }                                                                   \
+        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) {                                                      \
+            if (GET_REGISTER(vsrc1) == 0) {                                 \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO(exceptionThrown);                                      \
+            }                                                               \
+        }                                                                   \
+        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) {                                                      \
+            if (GET_REGISTER_WIDE(vsrc2) == 0) {                            \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO(exceptionThrown);                                      \
+            }                                                               \
+        }                                                                   \
+        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) {                                                      \
+            if (GET_REGISTER_WIDE(vsrc1) == 0) {                            \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO(exceptionThrown);                                      \
+            }                                                               \
+        }                                                                   \
+        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(method->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(method->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(method->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(method->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);
+
diff --git a/vm/mterp/common/asm-constants.h b/vm/mterp/common/asm-constants.h
new file mode 100644
index 0000000..d7a975f
--- /dev/null
+++ b/vm/mterp/common/asm-constants.h
@@ -0,0 +1,227 @@
+/*
+ * Copyright 2008 The Android Open Source Project
+ *
+ * Constants used by the assembler and verified by the C compiler.
+ */
+
+#if defined(ASM_DEF_VERIFY)
+  /*
+   * Generate C fragments that verify values; assumes "bool failed" exists.
+   * These are all constant expressions, so on success these will compile
+   * down to nothing.
+   */
+# define MTERP_OFFSET(_name, _type, _field, _offset)                        \
+    if (offsetof(_type, _field) != _offset) {                               \
+        LOGE("Bad asm offset %s (%d), should be %d\n",                      \
+            #_name, _offset, offsetof(_type, _field));                      \
+        failed = true;                                                      \
+    }
+# define MTERP_SIZEOF(_name, _type, _size)                                  \
+    if (sizeof(_type) != (_size)) {                                         \
+        LOGE("Bad asm sizeof %s (%d), should be %d\n",                      \
+            #_name, (_size), sizeof(_type));                                \
+        failed = true;                                                      \
+    }
+# define MTERP_CONSTANT(_name, _value)                                      \
+    if ((_name) != (_value)) {                                              \
+        LOGE("Bad asm constant %s (%d), should be %d\n",                    \
+            #_name, (_value), (_name));                                     \
+        failed = true;                                                      \
+    }
+#else
+  /* generate constant labels for the assembly output */
+# define MTERP_OFFSET(name, type, field, offset)    name = offset
+# define MTERP_SIZEOF(name, type, size)             name = size
+# define MTERP_CONSTANT(name, value)                name = value
+#endif
+
+/*
+ * Platform dependencies.  Some platforms require 64-bit alignment of 64-bit
+ * data structures.  Some versions of gcc will hold small enumerated types
+ * in a char instead of an int.
+ */
+#if defined(__ARM_EABI__)
+# define MTERP_NO_UNALIGN_64
+#endif
+#if defined(__ARM_EABI__)           // TODO: should be gcc version?
+# define MTERP_SMALL_ENUM   1
+#else
+# define MTERP_SMALL_ENUM   4
+#endif
+
+/*
+ * This file must only contain the following kinds of statements:
+ *
+ *  MTERP_OFFSET(name, StructType, fieldname, offset)
+ *
+ *   Declares that the expected offset of StructType.fieldname is "offset".
+ *   This will break whenever the contents of StructType are rearranged.
+ *
+ *  MTERP_SIZEOF(name, Type, size)
+ *
+ *   Declares that the expected size of Type is "size".
+ *
+ *  MTERP_CONSTANT(name, value)
+ *
+ *   Declares that the expected value of "name" is "value".  Useful for
+ *   enumerations and defined constants that are inaccessible to the
+ *   assembly source.  (Note this assumes you will use the same name in
+ *   both C and assembly, which is good practice.)
+ *
+ * In all cases the "name" field is the label you will use in the assembler.
+ *
+ * The "value" field must always be an actual number, not a symbol, unless
+ * you are sure that the symbol's value will be visible to both C and
+ * assembly sources.  There may be restrictions on the possible range of
+ * values (which are usually provided as immediate operands), so it's best
+ * to restrict numbers assuming a signed 8-bit field.
+ *
+ * On the assembly side, these just become "name=value" constants.  On the
+ * C side, these turn into assertions that cause the VM to abort if the
+ * values are incorrect.
+ */
+
+/* globals (sanity check for LDR vs LDRB) */
+MTERP_SIZEOF(sizeofGlobal_debuggerActive, gDvm.debuggerActive, MTERP_SMALL_ENUM)
+#if defined(WITH_PROFILER)
+MTERP_SIZEOF(sizeofGlobal_activeProfilers, gDvm.activeProfilers, 4)
+#endif
+
+/* MterpGlue fields */
+MTERP_OFFSET(offGlue_pc,                MterpGlue, pc, 0)
+MTERP_OFFSET(offGlue_fp,                MterpGlue, fp, 4)
+MTERP_OFFSET(offGlue_retval,            MterpGlue, retval, 8)
+MTERP_OFFSET(offGlue_method,            MterpGlue, method, 16)
+MTERP_OFFSET(offGlue_methodClassDex,    MterpGlue, methodClassDex, 20)
+MTERP_OFFSET(offGlue_self,              MterpGlue, self, 24)
+MTERP_OFFSET(offGlue_bailPtr,           MterpGlue, bailPtr, 28)
+MTERP_OFFSET(offGlue_interpStackEnd,    MterpGlue, interpStackEnd, 32)
+MTERP_OFFSET(offGlue_pSelfSuspendCount, MterpGlue, pSelfSuspendCount, 36)
+#if defined(WITH_DEBUGGER) && defined(WITH_PROFILER)
+MTERP_OFFSET(offGlue_pDebuggerActive,   MterpGlue, pDebuggerActive, 40)
+MTERP_OFFSET(offGlue_pActiveProfilers,  MterpGlue, pActiveProfilers, 44)
+MTERP_OFFSET(offGlue_entryPoint,        MterpGlue, entryPoint, 48)
+#elif defined(WITH_DEBUGGER)
+MTERP_OFFSET(offGlue_pDebuggerActive,   MterpGlue, pDebuggerActive, 40)
+MTERP_OFFSET(offGlue_entryPoint,        MterpGlue, entryPoint, 44)
+#elif defined(WITH_PROFILER)
+MTERP_OFFSET(offGlue_pActiveProfilers,  MterpGlue, pActiveProfilers, 40)
+MTERP_OFFSET(offGlue_entryPoint,        MterpGlue, entryPoint, 44)
+#else
+MTERP_OFFSET(offGlue_entryPoint,        MterpGlue, entryPoint, 40)
+#endif
+/* make sure all JValue union members are stored at the same offset */
+MTERP_OFFSET(offGlue_retval_z,          MterpGlue, retval.z, 8)
+MTERP_OFFSET(offGlue_retval_i,          MterpGlue, retval.i, 8)
+MTERP_OFFSET(offGlue_retval_j,          MterpGlue, retval.j, 8)
+MTERP_OFFSET(offGlue_retval_l,          MterpGlue, retval.l, 8)
+
+/* DvmDex fields */
+MTERP_OFFSET(offDvmDex_pResStrings,     DvmDex, pResStrings, 8)
+MTERP_OFFSET(offDvmDex_pResClasses,     DvmDex, pResClasses, 12)
+MTERP_OFFSET(offDvmDex_pResMethods,     DvmDex, pResMethods, 16)
+MTERP_OFFSET(offDvmDex_pResFields,      DvmDex, pResFields, 20)
+MTERP_OFFSET(offDvmDex_pInterfaceCache, DvmDex, pInterfaceCache, 24)
+
+/* StackSaveArea fields */
+#ifdef EASY_GDB
+MTERP_OFFSET(offStackSaveArea_prevSave, StackSaveArea, prevSave, 0)
+MTERP_OFFSET(offStackSaveArea_prevFrame, StackSaveArea, prevFrame, 4)
+MTERP_OFFSET(offStackSaveArea_savedPc,  StackSaveArea, savedPc, 8)
+MTERP_OFFSET(offStackSaveArea_method,   StackSaveArea, method, 12)
+MTERP_OFFSET(offStackSaveArea_currentPc, StackSaveArea, xtra.currentPc, 16)
+MTERP_OFFSET(offStackSaveArea_localRefTop, StackSaveArea, xtra.localRefTop, 16)
+MTERP_SIZEOF(sizeofStackSaveArea,       StackSaveArea, 20)
+#else
+MTERP_OFFSET(offStackSaveArea_prevFrame, StackSaveArea, prevFrame, 0)
+MTERP_OFFSET(offStackSaveArea_savedPc,  StackSaveArea, savedPc, 4)
+MTERP_OFFSET(offStackSaveArea_method,   StackSaveArea, method, 8)
+MTERP_OFFSET(offStackSaveArea_currentPc, StackSaveArea, xtra.currentPc, 12)
+MTERP_OFFSET(offStackSaveArea_localRefTop, StackSaveArea, xtra.localRefTop, 12)
+MTERP_SIZEOF(sizeofStackSaveArea,       StackSaveArea, 16)
+#endif
+
+/* InstField fields */
+#ifdef PROFILE_FIELD_ACCESS
+MTERP_OFFSET(offInstField_byteOffset,   InstField, byteOffset, 24)
+#else
+MTERP_OFFSET(offInstField_byteOffset,   InstField, byteOffset, 16)
+#endif
+
+/* StaticField fields */
+#ifdef PROFILE_FIELD_ACCESS
+MTERP_OFFSET(offStaticField_value,      StaticField, value, 24)
+#else
+MTERP_OFFSET(offStaticField_value,      StaticField, value, 16)
+#endif
+
+/* Method fields */
+MTERP_OFFSET(offMethod_clazz,           Method, clazz, 0)
+MTERP_OFFSET(offMethod_accessFlags,     Method, accessFlags, 4)
+MTERP_OFFSET(offMethod_methodIndex,     Method, methodIndex, 8)
+MTERP_OFFSET(offMethod_registersSize,   Method, registersSize, 10)
+MTERP_OFFSET(offMethod_outsSize,        Method, outsSize, 12)
+MTERP_OFFSET(offMethod_name,            Method, name, 16)
+MTERP_OFFSET(offMethod_insns,           Method, insns, 32)
+MTERP_OFFSET(offMethod_nativeFunc,      Method, nativeFunc, 40)
+
+/* InlineOperation fields -- code assumes "func" offset is zero, do not alter */
+MTERP_OFFSET(offInlineOperation_func,   InlineOperation, func, 0)
+
+/* Thread fields */
+MTERP_OFFSET(offThread_stackOverflowed, Thread, stackOverflowed, 40)
+MTERP_OFFSET(offThread_curFrame,        Thread, curFrame, 44)
+MTERP_OFFSET(offThread_exception,       Thread, exception, 48)
+MTERP_OFFSET(offThread_jniLocal_nextEntry, \
+                                        Thread, jniLocalRefTable.nextEntry, 80)
+
+/* Object fields */
+MTERP_OFFSET(offObject_clazz,           Object, clazz, 0)
+
+/* ArrayObject fields */
+MTERP_OFFSET(offArrayObject_length,     ArrayObject, length, 8)
+#ifdef MTERP_NO_UNALIGN_64
+MTERP_OFFSET(offArrayObject_contents,   ArrayObject, contents, 16)
+#else
+MTERP_OFFSET(offArrayObject_contents,   ArrayObject, contents, 12)
+#endif
+
+/* ClassObject fields */
+MTERP_OFFSET(offClassObject_descriptor, ClassObject, descriptor, 24)
+MTERP_OFFSET(offClassObject_accessFlags, ClassObject, accessFlags, 32)
+MTERP_OFFSET(offClassObject_pDvmDex,    ClassObject, pDvmDex, 36)
+MTERP_OFFSET(offClassObject_status,     ClassObject, status, 40)
+MTERP_OFFSET(offClassObject_super,      ClassObject, super, 72)
+MTERP_OFFSET(offClassObject_vtableCount, ClassObject, vtableCount, 112)
+MTERP_OFFSET(offClassObject_vtable,     ClassObject, vtable, 116)
+
+/* InterpEntry enumeration */
+MTERP_SIZEOF(sizeofClassStatus,         InterpEntry, MTERP_SMALL_ENUM)
+MTERP_CONSTANT(kInterpEntryInstr,   0)
+MTERP_CONSTANT(kInterpEntryReturn,  1)
+MTERP_CONSTANT(kInterpEntryThrow,   2)
+
+/* ClassStatus enumeration */
+MTERP_SIZEOF(sizeofClassStatus,         ClassStatus, MTERP_SMALL_ENUM)
+MTERP_CONSTANT(CLASS_INITIALIZED,   7)
+
+/* MethodType enumeration */
+MTERP_SIZEOF(sizeofMethodType,          MethodType, MTERP_SMALL_ENUM)
+MTERP_CONSTANT(METHOD_DIRECT,       1)
+MTERP_CONSTANT(METHOD_STATIC,       2)
+MTERP_CONSTANT(METHOD_VIRTUAL,      3)
+MTERP_CONSTANT(METHOD_INTERFACE,    4)
+
+/* ClassObject constants */
+MTERP_CONSTANT(ACC_PRIVATE,         0x0002)
+MTERP_CONSTANT(ACC_STATIC,          0x0008)
+MTERP_CONSTANT(ACC_NATIVE,          0x0100)
+MTERP_CONSTANT(ACC_INTERFACE,       0x0200)
+MTERP_CONSTANT(ACC_ABSTRACT,        0x0400)
+
+/* flags for dvmMalloc */
+MTERP_CONSTANT(ALLOC_DONT_TRACK,    0x02)
+
+/* opcode number */
+MTERP_CONSTANT(OP_MOVE_EXCEPTION,   0x0d)
+
diff --git a/vm/mterp/config-armv5 b/vm/mterp/config-armv5
new file mode 100644
index 0000000..4bf4562
--- /dev/null
+++ b/vm/mterp/config-armv5
@@ -0,0 +1,45 @@
+# 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.
+
+#
+# Configuration for ARMv5 architecture targets.
+#
+
+handler-size 64
+
+# source for the instruction table stub
+asm-stub armv5/stub.S
+
+# file header and basic definitions
+import c/header.c
+import armv5/header.S
+
+# common defs for the C helpers; include this before the instruction handlers
+import c/opcommon.c
+
+# arch-specific entry point to interpreter
+import armv5/entry.S
+
+# opcode list; argument to op-start is default directory
+op-start armv5
+	#op OP_FILL_ARRAY_DATA c
+op-end
+
+# "helper" code for C; include this after the instruction handlers
+import c/footer.c
+
+# common subroutines for asm
+import armv5/footer.S
+import armv5/debug.c
+
diff --git a/vm/mterp/config-desktop b/vm/mterp/config-desktop
new file mode 100644
index 0000000..06a977a
--- /dev/null
+++ b/vm/mterp/config-desktop
@@ -0,0 +1,37 @@
+# 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.
+
+#
+# Configuration for "desktop" targets.
+#
+
+handler-size 64
+
+# C file header and basic definitions
+import c/header.c
+
+# common defs for the C opcodes
+import c/opcommon.c
+
+# opcode list; argument to op-start is default directory
+op-start c
+    # use 100% C implementations
+op-end
+
+# arch-specific entry point to interpreter
+import desktop/entry.c
+
+# "helper" code
+import c/footer.c
+
diff --git a/vm/mterp/desktop/entry.c b/vm/mterp/desktop/entry.c
new file mode 100644
index 0000000..ba1eb09
--- /dev/null
+++ b/vm/mterp/desktop/entry.c
@@ -0,0 +1,79 @@
+/*
+ * 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.
+ */
+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);
+}
+
diff --git a/vm/mterp/gen-mterp.py b/vm/mterp/gen-mterp.py
new file mode 100755
index 0000000..e342266
--- /dev/null
+++ b/vm/mterp/gen-mterp.py
@@ -0,0 +1,474 @@
+#!/usr/bin/env python
+#
+# Copyright (C) 2007 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.
+
+#
+# Using instructions from an architecture-specific config file, generate C
+# and assembly source files for the Dalvik interpreter.
+#
+
+import sys, string, re, time
+from string import Template
+
+interp_defs_file = "../../libdex/OpCode.h" # need opcode list
+
+handler_size_bits = -1000
+handler_size_bytes = -1000
+in_op_start = 0             # 0=not started, 1=started, 2=ended
+default_op_dir = None
+opcode_locations = {}
+asm_stub_text = []
+label_prefix = ".L"         # use ".L" to hide labels from gdb
+
+
+# Exception class.
+class DataParseError(SyntaxError):
+    "Failure when parsing data file"
+
+#
+# Set any omnipresent substitution values.
+#
+def getGlobalSubDict():
+    return { "handler_size_bits":handler_size_bits,
+             "handler_size_bytes":handler_size_bytes }
+
+#
+# Parse arch config file --
+# Set handler_size_bytes to the value of tokens[1], and handler_size_bits to
+# log2(handler_size_bytes).  Throws an exception if "bytes" is not a power
+# of two.
+#
+def setHandlerSize(tokens):
+    global handler_size_bits, handler_size_bytes
+    if len(tokens) != 2:
+        raise DataParseError("handler-size requires one argument")
+    if handler_size_bits != -1000:
+        raise DataParseError("handler-size may only be set once")
+
+    # compute log2(n), and make sure n is a power of 2
+    handler_size_bytes = bytes = int(tokens[1])
+    bits = -1
+    while bytes > 0:
+        bytes //= 2     # halve with truncating division
+        bits += 1
+
+    if handler_size_bytes == 0 or handler_size_bytes != (1 << bits):
+        raise DataParseError("handler-size (%d) must be power of 2 and > 0" \
+                % orig_bytes)
+    handler_size_bits = bits
+
+#
+# Parse arch config file --
+# Copy a file in to the C or asm output file.
+#
+def importFile(tokens):
+    if len(tokens) != 2:
+        raise DataParseError("import requires one argument")
+    source = tokens[1]
+    if source.endswith(".c"):
+        appendSourceFile(tokens[1], getGlobalSubDict(), c_fp, None)
+    elif source.endswith(".S"):
+        appendSourceFile(tokens[1], getGlobalSubDict(), asm_fp, None)
+    else:
+        raise DataParseError("don't know how to import %s (expecting .c/.S)"
+                % source)
+
+#
+# Parse arch config file --
+# Copy a file in to the C or asm output file.
+#
+def setAsmStub(tokens):
+    global asm_stub_text
+    if len(tokens) != 2:
+        raise DataParseError("import requires one argument")
+    try:
+        stub_fp = open(tokens[1])
+        asm_stub_text = stub_fp.readlines()
+    except IOError, err:
+        stub_fp.close()
+        raise DataParseError("unable to load asm-stub: %s" % str(err))
+    stub_fp.close()
+
+#
+# Parse arch config file --
+# Start of opcode list.
+#
+def opStart(tokens):
+    global in_op_start
+    global default_op_dir
+    if len(tokens) != 2:
+        raise DataParseError("opStart takes a directory name argument")
+    if in_op_start != 0:
+        raise DataParseError("opStart can only be specified once")
+    default_op_dir = tokens[1]
+    in_op_start = 1
+
+#
+# Parse arch config file --
+# Set location of a single opcode's source file.
+#
+def opEntry(tokens):
+    #global opcode_locations
+    if len(tokens) != 3:
+        raise DataParseError("op requires exactly two arguments")
+    if in_op_start != 1:
+        raise DataParseError("op statements must be between opStart/opEnd")
+    try:
+        index = opcodes.index(tokens[1])
+    except ValueError:
+        raise DataParseError("unknown opcode %s" % tokens[1])
+    opcode_locations[tokens[1]] = tokens[2]
+
+#
+# Parse arch config file --
+# End of opcode list; emit instruction blocks.
+#
+def opEnd(tokens):
+    global in_op_start
+    if len(tokens) != 1:
+        raise DataParseError("opEnd takes no arguments")
+    if in_op_start != 1:
+        raise DataParseError("opEnd must follow opStart, and only appear once")
+    in_op_start = 2
+
+    loadAndEmitOpcodes()
+
+
+#
+# Extract an ordered list of instructions from the VM sources.  We use the
+# "goto table" definition macro, which has exactly 256 entries.
+#
+def getOpcodeList():
+    opcodes = []
+    opcode_fp = open(interp_defs_file)
+    opcode_re = re.compile(r"^\s*H\(OP_(\w+)\),.*", re.DOTALL)
+    for line in opcode_fp:
+        match = opcode_re.match(line)
+        if not match:
+            continue
+        opcodes.append("OP_" + match.group(1))
+    opcode_fp.close()
+
+    if len(opcodes) != 256:
+        print "ERROR: found %d opcodes in Interp.h (expected 256)" \
+                % len(opcodes)
+        raise SyntaxError, "bad opcode count"
+    return opcodes
+
+
+#
+# Load and emit opcodes for all 256 instructions.
+#
+def loadAndEmitOpcodes():
+    sister_list = []
+    assert len(opcodes) == 256
+    need_dummy_start = False
+
+    # point dvmAsmInstructionStart at the first handler or stub
+    asm_fp.write("\n    .global dvmAsmInstructionStart\n")
+    asm_fp.write("    .type   dvmAsmInstructionStart, %function\n")
+    asm_fp.write("dvmAsmInstructionStart = " + label_prefix + "_OP_NOP\n")
+    asm_fp.write("    .text\n\n")
+
+    for i in xrange(256):
+        op = opcodes[i]
+
+        if opcode_locations.has_key(op):
+            location = opcode_locations[op]
+        else:
+            location = default_op_dir
+
+        if location == "c":
+            loadAndEmitC(location, i)
+            if len(asm_stub_text) == 0:
+                need_dummy_start = True
+        else:
+            loadAndEmitAsm(location, i, sister_list)
+
+    # For a 100% C implementation, there are no asm handlers or stubs.  We
+    # need to have the dvmAsmInstructionStart label point at OP_NOP, and it's
+    # too annoying to try to slide it in after the alignment psuedo-op, so
+    # we take the low road and just emit a dummy OP_NOP here.
+    if need_dummy_start:
+        asm_fp.write("    .balign %d\n" % handler_size_bytes)
+        asm_fp.write(label_prefix + "_OP_NOP:   /* dummy */\n");
+
+    asm_fp.write("\n    .balign %d\n" % handler_size_bytes)
+    asm_fp.write("    .size   dvmAsmInstructionStart, .-dvmAsmInstructionStart\n")
+    asm_fp.write("    .global dvmAsmInstructionEnd\n")
+    asm_fp.write("dvmAsmInstructionEnd:\n")
+
+    emitSectionComment("Sister implementations", asm_fp)
+    asm_fp.write("    .global dvmAsmSisterStart\n")
+    asm_fp.write("    .type   dvmAsmSisterStart, %function\n")
+    asm_fp.write("    .text\n")
+    asm_fp.write("    .balign 4\n")
+    asm_fp.write("dvmAsmSisterStart:\n")
+    asm_fp.writelines(sister_list)
+
+    asm_fp.write("\n    .size   dvmAsmSisterStart, .-dvmAsmSisterStart\n")
+    asm_fp.write("    .global dvmAsmSisterEnd\n")
+    asm_fp.write("dvmAsmSisterEnd:\n\n")
+
+#
+# Load a C fragment and emit it, then output an assembly stub.
+#
+def loadAndEmitC(location, opindex):
+    op = opcodes[opindex]
+    source = "%s/%s.c" % (location, op)
+    print " emit %s --> C" % source
+    dict = getGlobalSubDict()
+    dict.update({ "opcode":op, "opnum":opindex })
+
+    appendSourceFile(source, dict, c_fp, None)
+
+    if len(asm_stub_text) != 0:
+        emitAsmStub(asm_fp, dict)
+
+#
+# Load an assembly fragment and emit it.
+#
+def loadAndEmitAsm(location, opindex, sister_list):
+    op = opcodes[opindex]
+    source = "%s/%s.S" % (location, op)
+    dict = getGlobalSubDict()
+    dict.update({ "opcode":op, "opnum":opindex })
+    print " emit %s --> asm" % source
+
+    emitAsmHeader(asm_fp, dict)
+    appendSourceFile(source, dict, asm_fp, sister_list)
+
+#
+# Output the alignment directive and label for an assembly piece.
+#
+def emitAsmHeader(outfp, dict):
+    outfp.write("/* ------------------------------ */\n")
+    # The alignment directive ensures that the handler occupies
+    # at least the correct amount of space.  We don't try to deal
+    # with overflow here.
+    outfp.write("    .balign %d\n" % handler_size_bytes)
+    # Emit a label so that gdb will say the right thing.  We prepend an
+    # underscore so the symbol name doesn't clash with the OpCode enum.
+    outfp.write(label_prefix + "_%(opcode)s: /* 0x%(opnum)02x */\n" % dict)
+
+#
+# Output a generic instruction stub that updates the "glue" struct and
+# calls the C implementation.
+#
+def emitAsmStub(outfp, dict):
+    emitAsmHeader(outfp, dict)
+    for line in asm_stub_text:
+        templ = Template(line)
+        outfp.write(templ.substitute(dict))
+
+#
+# Append the file specified by "source" to the open "outfp".  Each line will
+# be template-replaced using the substitution dictionary "dict".
+#
+# If the first line of the file starts with "%" it is taken as a directive.
+# A "%include" line contains a filename and, optionally, a Python-style
+# dictionary declaration with substitution strings.  (This is implemented
+# with recursion.)
+#
+# If "sister_list" is provided, and we find a line that contains only "&",
+# all subsequent lines from the file will be appended to sister_list instead
+# of copied to the output.
+#
+# This may modify "dict".
+#
+def appendSourceFile(source, dict, outfp, sister_list):
+    outfp.write("/* File: %s */\n" % source)
+    infp = open(source, "r")
+    in_sister = False
+    for line in infp:
+        if line.startswith("%include"):
+            # Parse the "include" line
+            tokens = line.strip().split(' ', 2)
+            if len(tokens) < 2:
+                raise DataParseError("malformed %%include in %s" % source)
+
+            alt_source = tokens[1].strip("\"")
+            if alt_source == source:
+                raise DataParseError("self-referential %%include in %s"
+                        % source)
+
+            new_dict = dict.copy()
+            if len(tokens) == 3:
+                new_dict.update(eval(tokens[2]))
+            #print " including src=%s dict=%s" % (alt_source, new_dict)
+            appendSourceFile(alt_source, new_dict, outfp, sister_list)
+            continue
+
+        elif line.startswith("%default"):
+            # copy keywords into dictionary
+            tokens = line.strip().split(' ', 1)
+            if len(tokens) < 2:
+                raise DataParseError("malformed %%default in %s" % source)
+            defaultValues = eval(tokens[1])
+            for entry in defaultValues:
+                dict.setdefault(entry, defaultValues[entry])
+            continue
+
+        elif line.startswith("%verify"):
+            # more to come, someday
+            continue
+
+        elif line.startswith("%break") and sister_list != None:
+            # allow more than one %break, ignoring all following the first
+            if not in_sister:
+                in_sister = True
+                sister_list.append("\n/* continuation for %(opcode)s */\n"%dict)
+            continue
+
+        # perform keyword substitution if a dictionary was provided
+        if dict != None:
+            templ = Template(line)
+            try:
+                subline = templ.substitute(dict)
+            except KeyError, err:
+                raise DataParseError("keyword substitution failed in %s: %s"
+                        % (source, str(err)))
+            except:
+                print "ERROR: substitution failed: " + line
+                raise
+        else:
+            subline = line
+
+        # write output to appropriate file
+        if in_sister:
+            sister_list.append(subline)
+        else:
+            outfp.write(subline)
+    outfp.write("\n")
+    infp.close()
+
+#
+# Emit a C-style section header comment.
+#
+def emitSectionComment(str, fp):
+    equals = "========================================" \
+             "==================================="
+
+    fp.write("\n/*\n * %s\n *  %s\n * %s\n */\n" %
+        (equals, str, equals))
+
+
+#
+# ===========================================================================
+# "main" code
+#
+
+#
+# Check args.
+#
+if len(sys.argv) != 3:
+    print "Usage: %s target-arch output-dir" % sys.argv[0]
+    sys.exit(2)
+
+target_arch = sys.argv[1]
+output_dir = sys.argv[2]
+
+#
+# Extract opcode list.
+#
+opcodes = getOpcodeList()
+#for op in opcodes:
+#    print "  %s" % op
+
+#
+# Open config file.
+#
+try:
+    config_fp = open("config-%s" % target_arch)
+except:
+    print "Unable to open config file 'config-%s'" % target_arch
+    sys.exit(1)
+
+#
+# Open and prepare output files.
+#
+try:
+    c_fp = open("%s/InterpC-%s.c" % (output_dir, target_arch), "w")
+    asm_fp = open("%s/InterpAsm-%s.S" % (output_dir, target_arch), "w")
+except:
+    print "Unable to open output files"
+    print "Make sure directory '%s' exists and existing files are writable" \
+            % output_dir
+    # Ideally we'd remove the files to avoid confusing "make", but if they
+    # failed to open we probably won't be able to remove them either.
+    sys.exit(1)
+
+print "Generating %s, %s" % (c_fp.name, asm_fp.name)
+
+file_header = """/*
+ * This file was generated automatically by gen-mterp.py for '%s'.
+ *
+ * --> DO NOT EDIT <--
+ */
+
+""" % (target_arch)
+
+c_fp.write(file_header)
+asm_fp.write(file_header)
+
+#
+# Process the config file.
+#
+failed = False
+try:
+    for line in config_fp:
+        line = line.strip()         # remove CRLF, leading spaces
+        tokens = line.split(' ')    # tokenize
+        #print "%d: %s" % (len(tokens), tokens)
+        if len(tokens[0]) == 0:
+            #print "  blank"
+            pass
+        elif tokens[0][0] == '#':
+            #print "  comment"
+            pass
+        else:
+            if tokens[0] == "handler-size":
+                setHandlerSize(tokens)
+            elif tokens[0] == "import":
+                importFile(tokens)
+            elif tokens[0] == "asm-stub":
+                setAsmStub(tokens)
+            elif tokens[0] == "op-start":
+                opStart(tokens)
+            elif tokens[0] == "op-end":
+                opEnd(tokens)
+            elif tokens[0] == "op":
+                opEntry(tokens)
+            else:
+                raise DataParseError, "unrecognized command '%s'" % tokens[0]
+except DataParseError, err:
+    print "Failed: " + str(err)
+    # TODO: remove output files so "make" doesn't get confused
+    failed = True
+    c_fp.close()
+    asm_fp.close()
+    c_fp = asm_fp = None
+
+config_fp.close()
+
+#
+# Done!
+#
+if c_fp:
+    c_fp.close()
+if asm_fp:
+    asm_fp.close()
+
+sys.exit(failed)
diff --git a/vm/mterp/out/InterpAsm-armv5.S b/vm/mterp/out/InterpAsm-armv5.S
new file mode 100644
index 0000000..1e776ca
--- /dev/null
+++ b/vm/mterp/out/InterpAsm-armv5.S
@@ -0,0 +1,9925 @@
+/*
+ * This file was generated automatically by gen-mterp.py for 'armv5'.
+ *
+ * --> DO NOT EDIT <--
+ */
+
+/* File: armv5/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.
+ */
+#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: armv5/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   sp!, {r4-r10,fp,pc}             @ restore 9 regs
+
+
+/*
+ * String references.
+ */
+strBadEntryPoint:
+    .word   .LstrBadEntryPoint
+
+
+
+    .global dvmAsmInstructionStart
+    .type   dvmAsmInstructionStart, %function
+dvmAsmInstructionStart = .L_OP_NOP
+    .text
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NOP: /* 0x00 */
+/* File: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/OP_MOVE_OBJECT.S */
+/* File: armv5/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: armv5/OP_MOVE_OBJECT_FROM16.S */
+/* File: armv5/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: armv5/OP_MOVE_OBJECT_16.S */
+/* File: armv5/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: armv5/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: armv5/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: armv5/OP_MOVE_RESULT_OBJECT.S */
+/* File: armv5/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: armv5/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: armv5/OP_RETURN_VOID.S */
+    b       common_returnFromMethod
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RETURN: /* 0x0f */
+/* File: armv5/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: armv5/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: armv5/OP_RETURN_OBJECT.S */
+/* File: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/OP_FILLED_NEW_ARRAY_RANGE.S */
+/* File: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/OP_SPARSE_SWITCH.S */
+/* File: armv5/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: armv5/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: armv5/OP_CMPG_FLOAT.S */
+/* File: armv5/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: armv5/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: armv5/OP_CMPG_DOUBLE.S */
+/* File: armv5/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: armv5/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: armv5/OP_IF_EQ.S */
+/* File: armv5/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: armv5/OP_IF_NE.S */
+/* File: armv5/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: armv5/OP_IF_LT.S */
+/* File: armv5/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: armv5/OP_IF_GE.S */
+/* File: armv5/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: armv5/OP_IF_GT.S */
+/* File: armv5/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: armv5/OP_IF_LE.S */
+/* File: armv5/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: armv5/OP_IF_EQZ.S */
+/* File: armv5/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: armv5/OP_IF_NEZ.S */
+/* File: armv5/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: armv5/OP_IF_LTZ.S */
+/* File: armv5/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: armv5/OP_IF_GEZ.S */
+/* File: armv5/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: armv5/OP_IF_GTZ.S */
+/* File: armv5/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: armv5/OP_IF_LEZ.S */
+/* File: armv5/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: armv5/OP_UNUSED_3E.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_3F: /* 0x3f */
+/* File: armv5/OP_UNUSED_3F.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_40: /* 0x40 */
+/* File: armv5/OP_UNUSED_40.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_41: /* 0x41 */
+/* File: armv5/OP_UNUSED_41.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_42: /* 0x42 */
+/* File: armv5/OP_UNUSED_42.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_43: /* 0x43 */
+/* File: armv5/OP_UNUSED_43.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET: /* 0x44 */
+/* File: armv5/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: armv5/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: armv5/OP_AGET_OBJECT.S */
+/* File: armv5/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: armv5/OP_AGET_BOOLEAN.S */
+/* File: armv5/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: armv5/OP_AGET_BYTE.S */
+/* File: armv5/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: armv5/OP_AGET_CHAR.S */
+/* File: armv5/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: armv5/OP_AGET_SHORT.S */
+/* File: armv5/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: armv5/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: armv5/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: armv5/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.
+     *
+     * for: aput, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* 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: armv5/OP_APUT_BOOLEAN.S */
+/* File: armv5/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: armv5/OP_APUT_BYTE.S */
+/* File: armv5/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: armv5/OP_APUT_CHAR.S */
+/* File: armv5/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: armv5/OP_APUT_SHORT.S */
+/* File: armv5/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: armv5/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: armv5/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: armv5/OP_IGET_OBJECT.S */
+/* File: armv5/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: armv5/OP_IGET_BOOLEAN.S */
+@include "armv5/OP_IGET.S" { "load":"ldrb", "sqnum":"1" }
+/* File: armv5/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: armv5/OP_IGET_BYTE.S */
+@include "armv5/OP_IGET.S" { "load":"ldrsb", "sqnum":"2" }
+/* File: armv5/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: armv5/OP_IGET_CHAR.S */
+@include "armv5/OP_IGET.S" { "load":"ldrh", "sqnum":"3" }
+/* File: armv5/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: armv5/OP_IGET_SHORT.S */
+@include "armv5/OP_IGET.S" { "load":"ldrsh", "sqnum":"4" }
+/* File: armv5/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: armv5/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: armv5/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: armv5/OP_IPUT_OBJECT.S */
+/* File: armv5/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: armv5/OP_IPUT_BOOLEAN.S */
+@include "armv5/OP_IPUT.S" { "store":"strb", "sqnum":"1" }
+/* File: armv5/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: armv5/OP_IPUT_BYTE.S */
+@include "armv5/OP_IPUT.S" { "store":"strb", "sqnum":"2" }
+/* File: armv5/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: armv5/OP_IPUT_CHAR.S */
+@include "armv5/OP_IPUT.S" { "store":"strh", "sqnum":"3" }
+/* File: armv5/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: armv5/OP_IPUT_SHORT.S */
+@include "armv5/OP_IPUT.S" { "store":"strh", "sqnum":"4" }
+/* File: armv5/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: armv5/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: armv5/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: armv5/OP_SGET_OBJECT.S */
+/* File: armv5/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: armv5/OP_SGET_BOOLEAN.S */
+/* File: armv5/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: armv5/OP_SGET_BYTE.S */
+/* File: armv5/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: armv5/OP_SGET_CHAR.S */
+/* File: armv5/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: armv5/OP_SGET_SHORT.S */
+/* File: armv5/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: armv5/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: armv5/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: armv5/OP_SPUT_OBJECT.S */
+/* File: armv5/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: armv5/OP_SPUT_BOOLEAN.S */
+/* File: armv5/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: armv5/OP_SPUT_BYTE.S */
+/* File: armv5/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: armv5/OP_SPUT_CHAR.S */
+/* File: armv5/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: armv5/OP_SPUT_SHORT.S */
+/* File: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/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: armv5/OP_UNUSED_73.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_VIRTUAL_RANGE: /* 0x74 */
+/* File: armv5/OP_INVOKE_VIRTUAL_RANGE.S */
+/* File: armv5/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: armv5/OP_INVOKE_SUPER_RANGE.S */
+/* File: armv5/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: armv5/OP_INVOKE_DIRECT_RANGE.S */
+/* File: armv5/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: armv5/OP_INVOKE_STATIC_RANGE.S */
+/* File: armv5/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: armv5/OP_INVOKE_INTERFACE_RANGE.S */
+/* File: armv5/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: armv5/OP_UNUSED_79.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_7A: /* 0x7a */
+/* File: armv5/OP_UNUSED_7A.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEG_INT: /* 0x7b */
+/* File: armv5/OP_NEG_INT.S */
+/* File: armv5/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: armv5/OP_NOT_INT.S */
+/* File: armv5/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: armv5/OP_NEG_LONG.S */
+/* File: armv5/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: armv5/OP_NOT_LONG.S */
+/* File: armv5/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: armv5/OP_NEG_FLOAT.S */
+/* File: armv5/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: armv5/OP_NEG_DOUBLE.S */
+/* File: armv5/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: armv5/OP_INT_TO_LONG.S */
+/* File: armv5/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: armv5/OP_INT_TO_FLOAT.S */
+/* File: armv5/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: armv5/OP_INT_TO_DOUBLE.S */
+/* File: armv5/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: armv5/OP_LONG_TO_INT.S */
+/* we ignore the high word, making this equivalent to a 32-bit reg move */
+/* File: armv5/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: armv5/OP_LONG_TO_FLOAT.S */
+/* File: armv5/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: armv5/OP_LONG_TO_DOUBLE.S */
+/* File: armv5/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: armv5/OP_FLOAT_TO_INT.S */
+/* EABI appears to have Java-style conversions of +inf/-inf/NaN */
+/* File: armv5/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 "armv5/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: armv5/OP_FLOAT_TO_LONG.S */
+@include "armv5/unopWider.S" {"instr":"bl      __aeabi_f2lz"}
+/* File: armv5/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: armv5/OP_FLOAT_TO_DOUBLE.S */
+/* File: armv5/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: armv5/OP_DOUBLE_TO_INT.S */
+/* EABI appears to have Java-style conversions of +inf/-inf/NaN */
+/* File: armv5/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 "armv5/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: armv5/OP_DOUBLE_TO_LONG.S */
+@include "armv5/unopWide.S" {"instr":"bl      __aeabi_d2lz"}
+/* File: armv5/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: armv5/OP_DOUBLE_TO_FLOAT.S */
+/* File: armv5/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: armv5/OP_INT_TO_BYTE.S */
+/* File: armv5/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: armv5/OP_INT_TO_CHAR.S */
+/* File: armv5/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: armv5/OP_INT_TO_SHORT.S */
+/* File: armv5/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: armv5/OP_ADD_INT.S */
+/* File: armv5/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.
+     *
+     * 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: armv5/OP_SUB_INT.S */
+/* File: armv5/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.
+     *
+     * 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: armv5/OP_MUL_INT.S */
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+/* File: armv5/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.
+     *
+     * 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: armv5/OP_DIV_INT.S */
+/* File: armv5/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.
+     *
+     * 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: armv5/OP_REM_INT.S */
+/* idivmod returns quotient in r0 and remainder in r1 */
+/* File: armv5/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.
+     *
+     * 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: armv5/OP_AND_INT.S */
+/* File: armv5/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.
+     *
+     * 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: armv5/OP_OR_INT.S */
+/* File: armv5/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.
+     *
+     * 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: armv5/OP_XOR_INT.S */
+/* File: armv5/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.
+     *
+     * 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: armv5/OP_SHL_INT.S */
+/* File: armv5/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.
+     *
+     * 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: armv5/OP_SHR_INT.S */
+/* File: armv5/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.
+     *
+     * 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: armv5/OP_USHR_INT.S */
+/* File: armv5/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.
+     *
+     * 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: armv5/OP_ADD_LONG.S */
+/* File: armv5/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: armv5/OP_SUB_LONG.S */
+/* File: armv5/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: armv5/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: armv5/OP_DIV_LONG.S */
+/* File: armv5/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: armv5/OP_REM_LONG.S */
+/* ldivmod returns quotient in r0/r1 and remainder in r2/r3 */
+/* File: armv5/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: armv5/OP_AND_LONG.S */
+/* File: armv5/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: armv5/OP_OR_LONG.S */
+/* File: armv5/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: armv5/OP_XOR_LONG.S */
+/* File: armv5/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: armv5/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: armv5/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: armv5/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: armv5/OP_ADD_FLOAT.S */
+/* File: armv5/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.
+     *
+     * 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: armv5/OP_SUB_FLOAT.S */
+/* File: armv5/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.
+     *
+     * 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: armv5/OP_MUL_FLOAT.S */
+/* File: armv5/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.
+     *
+     * 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: armv5/OP_DIV_FLOAT.S */
+/* File: armv5/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.
+     *
+     * 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: armv5/OP_REM_FLOAT.S */
+/* EABI doesn't define a float remainder function, but libm does */
+/* File: armv5/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.
+     *
+     * 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: armv5/OP_ADD_DOUBLE.S */
+/* File: armv5/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: armv5/OP_SUB_DOUBLE.S */
+/* File: armv5/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: armv5/OP_MUL_DOUBLE.S */
+/* File: armv5/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: armv5/OP_DIV_DOUBLE.S */
+/* File: armv5/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: armv5/OP_REM_DOUBLE.S */
+/* EABI doesn't define a double remainder function, but libm does */
+/* File: armv5/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: armv5/OP_ADD_INT_2ADDR.S */
+/* File: armv5/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: armv5/OP_SUB_INT_2ADDR.S */
+/* File: armv5/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: armv5/OP_MUL_INT_2ADDR.S */
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+/* File: armv5/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: armv5/OP_DIV_INT_2ADDR.S */
+/* File: armv5/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: armv5/OP_REM_INT_2ADDR.S */
+/* idivmod returns quotient in r0 and remainder in r1 */
+/* File: armv5/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: armv5/OP_AND_INT_2ADDR.S */
+/* File: armv5/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: armv5/OP_OR_INT_2ADDR.S */
+/* File: armv5/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: armv5/OP_XOR_INT_2ADDR.S */
+/* File: armv5/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: armv5/OP_SHL_INT_2ADDR.S */
+/* File: armv5/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: armv5/OP_SHR_INT_2ADDR.S */
+/* File: armv5/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: armv5/OP_USHR_INT_2ADDR.S */
+/* File: armv5/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: armv5/OP_ADD_LONG_2ADDR.S */
+/* File: armv5/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: armv5/OP_SUB_LONG_2ADDR.S */
+/* File: armv5/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: armv5/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: armv5/OP_DIV_LONG_2ADDR.S */
+/* File: armv5/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: armv5/OP_REM_LONG_2ADDR.S */
+/* ldivmod returns quotient in r0/r1 and remainder in r2/r3 */
+/* File: armv5/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: armv5/OP_AND_LONG_2ADDR.S */
+/* File: armv5/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: armv5/OP_OR_LONG_2ADDR.S */
+/* File: armv5/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: armv5/OP_XOR_LONG_2ADDR.S */
+/* File: armv5/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: armv5/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: armv5/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: armv5/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: armv5/OP_ADD_FLOAT_2ADDR.S */
+/* File: armv5/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: armv5/OP_SUB_FLOAT_2ADDR.S */
+/* File: armv5/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: armv5/OP_MUL_FLOAT_2ADDR.S */
+/* File: armv5/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: armv5/OP_DIV_FLOAT_2ADDR.S */
+/* File: armv5/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: armv5/OP_REM_FLOAT_2ADDR.S */
+/* EABI doesn't define a float remainder function, but libm does */
+/* File: armv5/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: armv5/OP_ADD_DOUBLE_2ADDR.S */
+/* File: armv5/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: armv5/OP_SUB_DOUBLE_2ADDR.S */
+/* File: armv5/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: armv5/OP_MUL_DOUBLE_2ADDR.S */
+/* File: armv5/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: armv5/OP_DIV_DOUBLE_2ADDR.S */
+/* File: armv5/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: armv5/OP_REM_DOUBLE_2ADDR.S */
+/* EABI doesn't define a double remainder function, but libm does */
+/* File: armv5/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: armv5/OP_ADD_INT_LIT16.S */
+/* File: armv5/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: armv5/OP_RSUB_INT.S */
+/* this op is "rsub-int", but can be thought of as "rsub-int/lit16" */
+/* File: armv5/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: armv5/OP_MUL_INT_LIT16.S */
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+/* File: armv5/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: armv5/OP_DIV_INT_LIT16.S */
+/* File: armv5/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: armv5/OP_REM_INT_LIT16.S */
+/* idivmod returns quotient in r0 and remainder in r1 */
+/* File: armv5/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: armv5/OP_AND_INT_LIT16.S */
+/* File: armv5/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: armv5/OP_OR_INT_LIT16.S */
+/* File: armv5/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: armv5/OP_XOR_INT_LIT16.S */
+/* File: armv5/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: armv5/OP_ADD_INT_LIT8.S */
+/* File: armv5/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: armv5/OP_RSUB_INT_LIT8.S */
+/* File: armv5/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: armv5/OP_MUL_INT_LIT8.S */
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+/* File: armv5/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: armv5/OP_DIV_INT_LIT8.S */
+/* File: armv5/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: armv5/OP_REM_INT_LIT8.S */
+/* idivmod returns quotient in r0 and remainder in r1 */
+/* File: armv5/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: armv5/OP_AND_INT_LIT8.S */
+/* File: armv5/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: armv5/OP_OR_INT_LIT8.S */
+/* File: armv5/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: armv5/OP_XOR_INT_LIT8.S */
+/* File: armv5/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: armv5/OP_SHL_INT_LIT8.S */
+/* File: armv5/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: armv5/OP_SHR_INT_LIT8.S */
+/* File: armv5/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: armv5/OP_USHR_INT_LIT8.S */
+/* File: armv5/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: armv5/OP_UNUSED_E3.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E4: /* 0xe4 */
+/* File: armv5/OP_UNUSED_E4.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E5: /* 0xe5 */
+/* File: armv5/OP_UNUSED_E5.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E6: /* 0xe6 */
+/* File: armv5/OP_UNUSED_E6.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E7: /* 0xe7 */
+/* File: armv5/OP_UNUSED_E7.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E8: /* 0xe8 */
+/* File: armv5/OP_UNUSED_E8.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E9: /* 0xe9 */
+/* File: armv5/OP_UNUSED_E9.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EA: /* 0xea */
+/* File: armv5/OP_UNUSED_EA.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EB: /* 0xeb */
+/* File: armv5/OP_UNUSED_EB.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EC: /* 0xec */
+/* File: armv5/OP_UNUSED_EC.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_ED: /* 0xed */
+/* File: armv5/OP_UNUSED_ED.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_EXECUTE_INLINE: /* 0xee */
+/* File: armv5/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: armv5/OP_UNUSED_EF.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_DIRECT_EMPTY: /* 0xf0 */
+/* File: armv5/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: armv5/OP_UNUSED_F1.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_QUICK: /* 0xf2 */
+/* File: armv5/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: armv5/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: armv5/OP_IGET_OBJECT_QUICK.S */
+/* File: armv5/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: armv5/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: armv5/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: armv5/OP_IPUT_OBJECT_QUICK.S */
+/* File: armv5/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: armv5/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: armv5/OP_INVOKE_VIRTUAL_QUICK_RANGE.S */
+/* File: armv5/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: armv5/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: armv5/OP_INVOKE_SUPER_QUICK_RANGE.S */
+/* File: armv5/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: armv5/OP_UNUSED_FC.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FD: /* 0xfd */
+/* File: armv5/OP_UNUSED_FD.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FE: /* 0xfe */
+/* File: armv5/OP_UNUSED_FE.S */
+/* File: armv5/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FF: /* 0xff */
+/* File: armv5/OP_UNUSED_FF.S */
+/* File: armv5/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    r0, [r3, #1]                @ r0<- descriptor[1]
+    .if     0
+    mov     r1, r10                     @ r1<- AA (length)
+    .else
+    mov     r1, r10, lsr #4             @ r1<- B (length)
+    .endif
+    cmp     r0, #'I'                    @ array of ints?
+    mov     r9, r1                      @ save length in r9
+    bne     .LOP_FILLED_NEW_ARRAY_notimpl         @ no, not handled yet
+    bl      dvmAllocPrimitiveArray      @ r0<- call(typeCh, 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, strInternalError
+    ldr     r1, strFilledNewArrayNotImpl
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+    .if     (!0)                 @ define in one or the other, not both
+strFilledNewArrayNotImpl:
+    .word   .LstrFilledNewArrayNotImpl
+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    r0, [r3, #1]                @ r0<- descriptor[1]
+    .if     1
+    mov     r1, r10                     @ r1<- AA (length)
+    .else
+    mov     r1, r10, lsr #4             @ r1<- B (length)
+    .endif
+    cmp     r0, #'I'                    @ array of ints?
+    mov     r9, r1                      @ save length in r9
+    bne     .LOP_FILLED_NEW_ARRAY_RANGE_notimpl         @ no, not handled yet
+    bl      dvmAllocPrimitiveArray      @ r0<- call(typeCh, 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, strInternalError
+    ldr     r1, strFilledNewArrayNotImpl
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+    .if     (!1)                 @ define in one or the other, not both
+strFilledNewArrayNotImpl:
+    .word   .LstrFilledNewArrayNotImpl
+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 in 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:
+    @b       dvmPerformInlineOp4Std
+    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: armv5/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
+
+    mov     pc, 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
+
+    @ 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)
+    ldr     rFP, [sp, #4]               @ retrieve the updated rFP
+    cmp     r0, #0                      @ is catchRelPc < 0?
+    add     sp, sp, #8                  @ restore stack
+    bmi     .LnotCaughtLocally
+
+    /* fix stack overflow if necessary; must preserve r0 */
+    ldrb    r1, [r10, #offThread_stackOverflowed]
+    cmp     r1, #0                      @ did we overflow earlier?
+    beq     1f                          @ no, skip ahead
+    mov     r9, r0                      @ r9<- r0 (save it)
+    mov     r0, r10                     @ r0<- self
+    bl      dvmCleanupStackOverflow     @ call(self)
+    mov     r0, r9                      @ r0<- r9 (restore it)
+    ldr     r9, [r10, #offThread_exception] @ r9<- self->exception
+1:
+
+    /* 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, pc}
+    .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, pc}
+
+/*
+ * 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, pc}
+
+    /*
+     * 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, pc}
+
+/*
+ * 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, pc}
+
+/*
+ * 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, pc}
+
+/*
+ * 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, pc}
+    .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 '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-desktop.S b/vm/mterp/out/InterpAsm-desktop.S
new file mode 100644
index 0000000..6d7689c
--- /dev/null
+++ b/vm/mterp/out/InterpAsm-desktop.S
@@ -0,0 +1,35 @@
+/*
+ * This file was generated automatically by gen-mterp.py for 'desktop'.
+ *
+ * --> 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/InterpC-armv5.c b/vm/mterp/out/InterpC-armv5.c
new file mode 100644
index 0000000..1e112ad
--- /dev/null
+++ b/vm/mterp/out/InterpC-armv5.c
@@ -0,0 +1,1736 @@
+/*
+ * This file was generated automatically by gen-mterp.py for 'armv5'.
+ *
+ * --> 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
+
+
+#define GOTO_TARGET_DECL(_target, ...)                                      \
+    void dvmMterp_##_target(MterpGlue* glue, ## __VA_ARGS__);
+
+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);
+
+
+/* File: c/opcommon.c */
+/*
+ * 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 method                  glue->method
+#define methodClassDex          glue->methodClassDex
+#define self                    glue->self
+//#define entryPoint              glue->entryPoint
+#define debugTrackedRefStart    glue->debugTrackedRefStart
+
+
+/*
+ * Replace the opcode definition 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.
+ */
+#undef HANDLE_OPCODE
+#undef OP_END
+#undef FINISH
+
+#define HANDLE_OPCODE(_op)                                                  \
+    void dvmMterp_##_op(MterpGlue* glue) {                                  \
+        u2 ref, vsrc1, vsrc2, vdst;                                         \
+        u2 inst = FETCH(0);
+
+#define OP_END }
+
+/*
+ * Like standard 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.
+ */
+#define GOTO(_target, ...)                                                  \
+    do {                                                                    \
+        dvmMterp_##_target(glue, ## __VA_ARGS__);                           \
+        return;                                                             \
+    } while(false)
+
+/*
+ * As a special case, "goto bail" turns into a longjmp.  "_switch" should be
+ * "true" if we need to switch to the other interpreter upon our return.
+ */
+#define GOTO_BAIL(_switch)                                                  \
+    dvmMterpStdBail(glue, _switch);
+
+/* for now, mterp is always a "standard" interpreter */
+#define INTERP_TYPE INTERP_STD
+
+/*
+ * Periodic checks macro, slightly modified.
+ */
+#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(true);                                                \
+        }                                                                   \
+    }
+
+
+/*
+ * ===========================================================================
+ *
+ * What follows are the "common" opcode definitions copied & pasted from the
+ * basic interpreter.  The only changes that need to be made to the original
+ * sources are:
+ *  - replace "goto exceptionThrown" with "GOTO(exceptionThrown)"
+ *
+ * ===========================================================================
+ */
+
+
+#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) {                                                      \
+            if (GET_REGISTER(vsrc2) == 0) {                                 \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO(exceptionThrown);                                      \
+            }                                                               \
+        }                                                                   \
+        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, _cast, _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) {                                                      \
+            if ((s2) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO(exceptionThrown);                                      \
+            }                                                               \
+        }                                                                   \
+        SET_REGISTER(vdst,                                                  \
+            _cast 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) {                                                      \
+            if ((s1) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO(exceptionThrown);                                      \
+            }                                                               \
+        }                                                                   \
+        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) {                                                      \
+            if (GET_REGISTER(vsrc1) == 0) {                                 \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO(exceptionThrown);                                      \
+            }                                                               \
+        }                                                                   \
+        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) {                                                      \
+            if (GET_REGISTER_WIDE(vsrc2) == 0) {                            \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO(exceptionThrown);                                      \
+            }                                                               \
+        }                                                                   \
+        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) {                                                      \
+            if (GET_REGISTER_WIDE(vsrc1) == 0) {                            \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO(exceptionThrown);                                      \
+            }                                                               \
+        }                                                                   \
+        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(method->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(method->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(method->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(method->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/footer.c */
+/*
+ * C footer.  This has some common code shared by the various targets.
+ */
+
+#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 }
+
+
+/*
+ * 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;
+        int* 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(method->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 == '[') {
+            /* create array of objects or array of arrays */
+            /* TODO: need some work in the verifier before we allow this */
+            LOGE("fnao not implemented\n");
+            dvmThrowException("Ljava/lang/InternalError;",
+                "filled-new-array not implemented for reference types");
+            GOTO(exceptionThrown);
+        } else if (typeCh != 'I') {
+            /* TODO: requires multiple "fill in" loops with different widths */
+            LOGE("non-int not implemented\n");
+            dvmThrowException("Ljava/lang/InternalError;",
+                "filled-new-array not implemented for anything but 'int'");
+            GOTO(exceptionThrown);
+        }
+
+        assert(strchr("BCIFZ", typeCh) != NULL);
+        newArray = dvmAllocPrimitiveArray(arrayClass->descriptor[1], vsrc1,
+                    ALLOC_DONT_TRACK);
+        if (newArray == NULL)
+            GOTO(exceptionThrown);
+
+        /*
+         * Fill in the elements.  It's legal for vsrc1 to be zero.
+         */
+        contents = (int*) 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 */
+
+        if (methodCallRange) {
+            /*
+             * The object against which we are executing a method is always
+             * in the first argument.
+             */
+            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 {
+            /*
+             * The object against which we are executing a method is always
+             * in the first argument.
+             */
+            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(method->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(method->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 >= method->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 = method->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 */
+
+        if (methodCallRange) {
+            /*
+             * The object against which we are executing a method is always
+             * in the first argument.
+             */
+            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 {
+            /*
+             * The object against which we are executing a method is always
+             * in the first argument.
+             */
+            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, method,
+                        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(method->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(method->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 */
+
+        if (methodCallRange) {
+            /*
+             * The object against which we are executing a method is always
+             * in the first argument.
+             */
+            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 {
+            /*
+             * The object against which we are executing a method is always
+             * in the first argument.
+             */
+            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 >= method->clazz->super->vtableCount) {
+            dvmThrowException("Ljava/lang/NoSuchMethodError;", NULL);
+            GOTO(exceptionThrown);
+        }
+#else
+        assert(ref < method->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 = method->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, method->clazz->descriptor, method->name,
+            method->signature);
+        //DUMP_REGS(method, fp);
+
+        saveArea = SAVEAREA_FROM_FP(fp);
+
+#ifdef EASY_GDB
+        debugSaveArea = saveArea;
+#endif
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_PROFILER)
+        TRACE_METHOD_EXIT(self, method);
+#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(false);
+        }
+
+        /* update thread FP, and reset local variables */
+        self->curFrame = fp;
+        method =
+#undef method       // ARRGH!
+            SAVEAREA_FROM_FP(fp)->method;
+#define method glue->method
+        //methodClass = method->clazz;
+        methodClassDex = method->clazz->pDvmDex;
+        pc = saveArea->savedPc;
+        ILOGD("> (return to %s.%s %s)", method->clazz->descriptor,
+            method->name, method->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 - method->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, method->name,
+            dvmLineNumFromPC(method, pc - method->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 - method->insns,
+                        exception, true, &catchFrame);
+            dvmDbgPostException(fp, pc - method->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 - method->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.
+         */
+        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(method),
+                dvmLineNumFromPC(method, pc - method->insns));
+#endif
+            dvmSetException(self, exception);
+            dvmReleaseTrackedAlloc(exception, self);
+            GOTO_BAIL(false);
+        }
+
+#if DVM_SHOW_EXCEPTION >= 3
+        {
+            const Method* catchMethod =
+#undef method
+                SAVEAREA_FROM_FP(fp)->method;
+#define method glue->method
+            LOGD("Exception %s thrown from %s:%d to %s:%d\n",
+                exception->clazz->descriptor, dvmGetMethodSourceFile(method),
+                dvmLineNumFromPC(method, pc - method->insns),
+                dvmGetMethodSourceFile(catchMethod),
+                dvmLineNumFromPC(catchMethod, catchRelPc));
+        }
+#endif
+
+        /*
+         * Adjust local variables to match self->curFrame and the
+         * updated PC.
+         */
+        //fp = (u4*) self->curFrame;
+        method =
+#undef method
+            SAVEAREA_FROM_FP(fp)->method;
+#define method glue->method
+        //methodClass = method->clazz;
+        methodClassDex = method->clazz->pDvmDex;
+        pc = method->insns + catchRelPc;
+        ILOGV("> pc <-- %s.%s %s", method->clazz->descriptor, method->name,
+            method->signature);
+        DUMP_REGS(method, 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)
+    {
+        vsrc1 = count; vdst = regs; methodToCall = _methodToCall;  /* ADDED */
+
+        //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 <= method->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 <= method->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;
+#undef method
+        newSaveArea->method = methodToCall;
+#define method glue->method
+
+        if (!dvmIsNativeMethod(methodToCall)) {
+            /*
+             * "Call" interpreted code.  Reposition the PC, update the
+             * frame pointer and other local state, and continue.
+             */
+            method = methodToCall;
+            methodClassDex = method->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", method->clazz->descriptor, method->name,
+                method->signature);
+            DUMP_REGS(method, 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(method, 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(method, 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,
+                method->clazz->descriptor, method->name,
+                method->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 - method->insns));
+                assert(false);
+            }
+        }
+    }
+    assert(false);      // should not get here
+GOTO_TARGET_END
+
+
+/* undefine "magic" name remapping */
+#undef retval
+#undef pc
+#undef fp
+#undef method
+#undef methodClassDex
+#undef self
+#undef debugTrackedRefStart
+
+/* File: armv5/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-desktop.c b/vm/mterp/out/InterpC-desktop.c
new file mode 100644
index 0000000..061e095
--- /dev/null
+++ b/vm/mterp/out/InterpC-desktop.c
@@ -0,0 +1,3480 @@
+/*
+ * This file was generated automatically by gen-mterp.py for 'desktop'.
+ *
+ * --> 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
+
+
+#define GOTO_TARGET_DECL(_target, ...)                                      \
+    void dvmMterp_##_target(MterpGlue* glue, ## __VA_ARGS__);
+
+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);
+
+
+/* File: c/opcommon.c */
+/*
+ * 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 method                  glue->method
+#define methodClassDex          glue->methodClassDex
+#define self                    glue->self
+//#define entryPoint              glue->entryPoint
+#define debugTrackedRefStart    glue->debugTrackedRefStart
+
+
+/*
+ * Replace the opcode definition 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.
+ */
+#undef HANDLE_OPCODE
+#undef OP_END
+#undef FINISH
+
+#define HANDLE_OPCODE(_op)                                                  \
+    void dvmMterp_##_op(MterpGlue* glue) {                                  \
+        u2 ref, vsrc1, vsrc2, vdst;                                         \
+        u2 inst = FETCH(0);
+
+#define OP_END }
+
+/*
+ * Like standard 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.
+ */
+#define GOTO(_target, ...)                                                  \
+    do {                                                                    \
+        dvmMterp_##_target(glue, ## __VA_ARGS__);                           \
+        return;                                                             \
+    } while(false)
+
+/*
+ * As a special case, "goto bail" turns into a longjmp.  "_switch" should be
+ * "true" if we need to switch to the other interpreter upon our return.
+ */
+#define GOTO_BAIL(_switch)                                                  \
+    dvmMterpStdBail(glue, _switch);
+
+/* for now, mterp is always a "standard" interpreter */
+#define INTERP_TYPE INTERP_STD
+
+/*
+ * Periodic checks macro, slightly modified.
+ */
+#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(true);                                                \
+        }                                                                   \
+    }
+
+
+/*
+ * ===========================================================================
+ *
+ * What follows are the "common" opcode definitions copied & pasted from the
+ * basic interpreter.  The only changes that need to be made to the original
+ * sources are:
+ *  - replace "goto exceptionThrown" with "GOTO(exceptionThrown)"
+ *
+ * ===========================================================================
+ */
+
+
+#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) {                                                      \
+            if (GET_REGISTER(vsrc2) == 0) {                                 \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO(exceptionThrown);                                      \
+            }                                                               \
+        }                                                                   \
+        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, _cast, _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) {                                                      \
+            if ((s2) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO(exceptionThrown);                                      \
+            }                                                               \
+        }                                                                   \
+        SET_REGISTER(vdst,                                                  \
+            _cast 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) {                                                      \
+            if ((s1) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO(exceptionThrown);                                      \
+            }                                                               \
+        }                                                                   \
+        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) {                                                      \
+            if (GET_REGISTER(vsrc1) == 0) {                                 \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO(exceptionThrown);                                      \
+            }                                                               \
+        }                                                                   \
+        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) {                                                      \
+            if (GET_REGISTER_WIDE(vsrc2) == 0) {                            \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO(exceptionThrown);                                      \
+            }                                                               \
+        }                                                                   \
+        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) {                                                      \
+            if (GET_REGISTER_WIDE(vsrc1) == 0) {                            \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO(exceptionThrown);                                      \
+            }                                                               \
+        }                                                                   \
+        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(method->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(method->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(method->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(method->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
+
+//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
+
+//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
+
+//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
+
+//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
+
+//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(method->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(method->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(method->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(method->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(method->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(method->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(method->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(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(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 < method->insns ||
+            arrayData >= method->insns + dvmGetMethodInsnsSize(method))
+        {
+            /* 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 < method->insns ||
+            switchData >= method->insns + dvmGetMethodInsnsSize(method))
+        {
+            /* 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 < method->insns ||
+            switchData >= method->insns + dvmGetMethodInsnsSize(method))
+        {
+            /* 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 */
+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(invokeVirtual, false);
+OP_END
+
+/* File: c/OP_INVOKE_SUPER.c */
+HANDLE_OPCODE(OP_INVOKE_SUPER /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO(invokeSuper, false);
+OP_END
+
+/* File: c/OP_INVOKE_DIRECT.c */
+HANDLE_OPCODE(OP_INVOKE_DIRECT /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO(invokeDirect, false);
+OP_END
+
+/* File: c/OP_INVOKE_STATIC.c */
+HANDLE_OPCODE(OP_INVOKE_STATIC /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO(invokeStatic, false);
+OP_END
+
+/* File: c/OP_INVOKE_INTERFACE.c */
+HANDLE_OPCODE(OP_INVOKE_INTERFACE /*vB, {vD, vE, vF, vG, vA}, meth@CCCC*/)
+    GOTO(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(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(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(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(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(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", +, false)
+OP_END
+
+/* File: c/OP_SUB_INT.c */
+HANDLE_OP_X_INT(OP_SUB_INT, "sub", -, false)
+OP_END
+
+/* File: c/OP_MUL_INT.c */
+HANDLE_OP_X_INT(OP_MUL_INT, "mul", *, false)
+OP_END
+
+/* File: c/OP_DIV_INT.c */
+HANDLE_OP_X_INT(OP_DIV_INT, "div", /, true)
+OP_END
+
+/* File: c/OP_REM_INT.c */
+HANDLE_OP_X_INT(OP_REM_INT, "rem", %, true)
+OP_END
+
+/* File: c/OP_AND_INT.c */
+HANDLE_OP_X_INT(OP_AND_INT, "and", &, false)
+OP_END
+
+/* File: c/OP_OR_INT.c */
+HANDLE_OP_X_INT(OP_OR_INT,  "or",  |, false)
+OP_END
+
+/* File: c/OP_XOR_INT.c */
+HANDLE_OP_X_INT(OP_XOR_INT, "xor", ^, false)
+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", +, false)
+OP_END
+
+/* File: c/OP_SUB_LONG.c */
+HANDLE_OP_X_LONG(OP_SUB_LONG, "sub", -, false)
+OP_END
+
+/* File: c/OP_MUL_LONG.c */
+HANDLE_OP_X_LONG(OP_MUL_LONG, "mul", *, false)
+OP_END
+
+/* File: c/OP_DIV_LONG.c */
+HANDLE_OP_X_LONG(OP_DIV_LONG, "div", /, true)
+OP_END
+
+/* File: c/OP_REM_LONG.c */
+HANDLE_OP_X_LONG(OP_REM_LONG, "rem", %, true)
+OP_END
+
+/* File: c/OP_AND_LONG.c */
+HANDLE_OP_X_LONG(OP_AND_LONG, "and", &, false)
+OP_END
+
+/* File: c/OP_OR_LONG.c */
+HANDLE_OP_X_LONG(OP_OR_LONG,  "or", |, false)
+OP_END
+
+/* File: c/OP_XOR_LONG.c */
+HANDLE_OP_X_LONG(OP_XOR_LONG, "xor", ^, false)
+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", +, false)
+OP_END
+
+/* File: c/OP_SUB_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_SUB_INT_2ADDR, "sub", -, false)
+OP_END
+
+/* File: c/OP_MUL_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_MUL_INT_2ADDR, "mul", *, false)
+OP_END
+
+/* File: c/OP_DIV_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_DIV_INT_2ADDR, "div", /, true)
+OP_END
+
+/* File: c/OP_REM_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_REM_INT_2ADDR, "rem", %, true)
+OP_END
+
+/* File: c/OP_AND_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_AND_INT_2ADDR, "and", &, false)
+OP_END
+
+/* File: c/OP_OR_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_OR_INT_2ADDR,  "or", |, false)
+OP_END
+
+/* File: c/OP_XOR_INT_2ADDR.c */
+HANDLE_OP_X_INT_2ADDR(OP_XOR_INT_2ADDR, "xor", ^, false)
+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", +, false)
+OP_END
+
+/* File: c/OP_SUB_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_SUB_LONG_2ADDR, "sub", -, false)
+OP_END
+
+/* File: c/OP_MUL_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_MUL_LONG_2ADDR, "mul", *, false)
+OP_END
+
+/* File: c/OP_DIV_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_DIV_LONG_2ADDR, "div", /, true)
+OP_END
+
+/* File: c/OP_REM_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_REM_LONG_2ADDR, "rem", %, true)
+OP_END
+
+/* File: c/OP_AND_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_AND_LONG_2ADDR, "and", &, false)
+OP_END
+
+/* File: c/OP_OR_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_OR_LONG_2ADDR,  "or", |, false)
+OP_END
+
+/* File: c/OP_XOR_LONG_2ADDR.c */
+HANDLE_OP_X_LONG_2ADDR(OP_XOR_LONG_2ADDR, "xor", ^, false)
+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", (s4), +, false)
+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", (s4), *, false)
+OP_END
+
+/* File: c/OP_DIV_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_DIV_INT_LIT16, "div", (s4), /, true)
+OP_END
+
+/* File: c/OP_REM_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_REM_INT_LIT16, "rem", (s4), %, true)
+OP_END
+
+/* File: c/OP_AND_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_AND_INT_LIT16, "and", (s4), &, false)
+OP_END
+
+/* File: c/OP_OR_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_OR_INT_LIT16,  "or",  (s4), |, false)
+OP_END
+
+/* File: c/OP_XOR_INT_LIT16.c */
+HANDLE_OP_X_INT_LIT16(OP_XOR_INT_LIT16, "xor", (s4), ^, false)
+OP_END
+
+/* File: c/OP_ADD_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_ADD_INT_LIT8,   "add", +, false)
+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", *, false)
+OP_END
+
+/* File: c/OP_DIV_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_DIV_INT_LIT8,   "div", /, true)
+OP_END
+
+/* File: c/OP_REM_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_REM_INT_LIT8,   "rem", %, true)
+OP_END
+
+/* File: c/OP_AND_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_AND_INT_LIT8,   "and", &, false)
+OP_END
+
+/* File: c/OP_OR_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_OR_INT_LIT8,    "or",  |, false)
+OP_END
+
+/* File: c/OP_XOR_INT_LIT8.c */
+HANDLE_OP_X_INT_LIT8(OP_XOR_INT_LIT8,   "xor", ^, false)
+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*/)
+    //LOGI("Ignoring empty\n");
+    FINISH(3);
+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(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(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(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(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)
+OP_END
+
+/* File: desktop/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.
+ */
+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/footer.c */
+/*
+ * C footer.  This has some common code shared by the various targets.
+ */
+
+#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 }
+
+
+/*
+ * 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;
+        int* 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(method->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 == '[') {
+            /* create array of objects or array of arrays */
+            /* TODO: need some work in the verifier before we allow this */
+            LOGE("fnao not implemented\n");
+            dvmThrowException("Ljava/lang/InternalError;",
+                "filled-new-array not implemented for reference types");
+            GOTO(exceptionThrown);
+        } else if (typeCh != 'I') {
+            /* TODO: requires multiple "fill in" loops with different widths */
+            LOGE("non-int not implemented\n");
+            dvmThrowException("Ljava/lang/InternalError;",
+                "filled-new-array not implemented for anything but 'int'");
+            GOTO(exceptionThrown);
+        }
+
+        assert(strchr("BCIFZ", typeCh) != NULL);
+        newArray = dvmAllocPrimitiveArray(arrayClass->descriptor[1], vsrc1,
+                    ALLOC_DONT_TRACK);
+        if (newArray == NULL)
+            GOTO(exceptionThrown);
+
+        /*
+         * Fill in the elements.  It's legal for vsrc1 to be zero.
+         */
+        contents = (int*) 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 */
+
+        if (methodCallRange) {
+            /*
+             * The object against which we are executing a method is always
+             * in the first argument.
+             */
+            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 {
+            /*
+             * The object against which we are executing a method is always
+             * in the first argument.
+             */
+            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(method->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(method->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 >= method->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 = method->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 */
+
+        if (methodCallRange) {
+            /*
+             * The object against which we are executing a method is always
+             * in the first argument.
+             */
+            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 {
+            /*
+             * The object against which we are executing a method is always
+             * in the first argument.
+             */
+            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, method,
+                        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(method->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(method->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 */
+
+        if (methodCallRange) {
+            /*
+             * The object against which we are executing a method is always
+             * in the first argument.
+             */
+            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 {
+            /*
+             * The object against which we are executing a method is always
+             * in the first argument.
+             */
+            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 >= method->clazz->super->vtableCount) {
+            dvmThrowException("Ljava/lang/NoSuchMethodError;", NULL);
+            GOTO(exceptionThrown);
+        }
+#else
+        assert(ref < method->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 = method->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, method->clazz->descriptor, method->name,
+            method->signature);
+        //DUMP_REGS(method, fp);
+
+        saveArea = SAVEAREA_FROM_FP(fp);
+
+#ifdef EASY_GDB
+        debugSaveArea = saveArea;
+#endif
+#if (INTERP_TYPE == INTERP_DBG) && defined(WITH_PROFILER)
+        TRACE_METHOD_EXIT(self, method);
+#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(false);
+        }
+
+        /* update thread FP, and reset local variables */
+        self->curFrame = fp;
+        method =
+#undef method       // ARRGH!
+            SAVEAREA_FROM_FP(fp)->method;
+#define method glue->method
+        //methodClass = method->clazz;
+        methodClassDex = method->clazz->pDvmDex;
+        pc = saveArea->savedPc;
+        ILOGD("> (return to %s.%s %s)", method->clazz->descriptor,
+            method->name, method->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 - method->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, method->name,
+            dvmLineNumFromPC(method, pc - method->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 - method->insns,
+                        exception, true, &catchFrame);
+            dvmDbgPostException(fp, pc - method->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 - method->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.
+         */
+        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(method),
+                dvmLineNumFromPC(method, pc - method->insns));
+#endif
+            dvmSetException(self, exception);
+            dvmReleaseTrackedAlloc(exception, self);
+            GOTO_BAIL(false);
+        }
+
+#if DVM_SHOW_EXCEPTION >= 3
+        {
+            const Method* catchMethod =
+#undef method
+                SAVEAREA_FROM_FP(fp)->method;
+#define method glue->method
+            LOGD("Exception %s thrown from %s:%d to %s:%d\n",
+                exception->clazz->descriptor, dvmGetMethodSourceFile(method),
+                dvmLineNumFromPC(method, pc - method->insns),
+                dvmGetMethodSourceFile(catchMethod),
+                dvmLineNumFromPC(catchMethod, catchRelPc));
+        }
+#endif
+
+        /*
+         * Adjust local variables to match self->curFrame and the
+         * updated PC.
+         */
+        //fp = (u4*) self->curFrame;
+        method =
+#undef method
+            SAVEAREA_FROM_FP(fp)->method;
+#define method glue->method
+        //methodClass = method->clazz;
+        methodClassDex = method->clazz->pDvmDex;
+        pc = method->insns + catchRelPc;
+        ILOGV("> pc <-- %s.%s %s", method->clazz->descriptor, method->name,
+            method->signature);
+        DUMP_REGS(method, 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)
+    {
+        vsrc1 = count; vdst = regs; methodToCall = _methodToCall;  /* ADDED */
+
+        //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 <= method->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 <= method->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;
+#undef method
+        newSaveArea->method = methodToCall;
+#define method glue->method
+
+        if (!dvmIsNativeMethod(methodToCall)) {
+            /*
+             * "Call" interpreted code.  Reposition the PC, update the
+             * frame pointer and other local state, and continue.
+             */
+            method = methodToCall;
+            methodClassDex = method->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", method->clazz->descriptor, method->name,
+                method->signature);
+            DUMP_REGS(method, 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(method, 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(method, 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,
+                method->clazz->descriptor, method->name,
+                method->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 - method->insns));
+                assert(false);
+            }
+        }
+    }
+    assert(false);      // should not get here
+GOTO_TARGET_END
+
+
+/* undefine "magic" name remapping */
+#undef retval
+#undef pc
+#undef fp
+#undef method
+#undef methodClassDex
+#undef self
+#undef debugTrackedRefStart
+
diff --git a/vm/mterp/rebuild.sh b/vm/mterp/rebuild.sh
new file mode 100755
index 0000000..79e37ac
--- /dev/null
+++ b/vm/mterp/rebuild.sh
@@ -0,0 +1,22 @@
+#!/bin/sh
+#
+# 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.
+
+#
+# Rebuild for all known targets.  Necessary until the stuff in "out" gets
+# generated as part of the build.
+#
+set -e
+for arch in desktop armv5; do TARGET_ARCH_EXT=$arch make -f Makefile-mterp; done