vm/compiler/codegen/arm/Codegen.c

/*
 * Copyright (C) 2009 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.
 */

/*
 * This file contains codegen and support common to all supported
 * ARM variants.  It is included by:
 *
 *        Codegen-$(TARGET_ARCH_VARIANT).c
 *
 * which combines this common code with specific support found in the
 * applicable directory below this one.
 */

#include "compiler/Loop.h"

/* Array holding the entry offset of each template relative to the first one */
static intptr_t templateEntryOffsets[TEMPLATE_LAST_MARK];

/* Track exercised opcodes */
static int opcodeCoverage[256];

#if defined(WITH_SELF_VERIFICATION)
/* Prevent certain opcodes from being jitted */
static inline bool selfVerificationPuntOps(OpCode op)
{
  return (op == OP_MONITOR_ENTER || op == OP_MONITOR_EXIT ||
          op == OP_NEW_INSTANCE  || op == OP_NEW_ARRAY);
}

/*
 * The following are used to keep compiled loads and stores from modifying
 * memory during self verification mode.
 *
 * Stores do not modify memory. Instead, the address and value pair are stored
 * into heapSpace. Addresses within heapSpace are unique. For accesses smaller
 * than a word, the word containing the address is loaded first before being
 * updated.
 *
 * Loads check heapSpace first and return data from there if an entry exists.
 * Otherwise, data is loaded from memory as usual.
 */

/* Decode contents of heapArgSpace to determine addr to load from */
static void selfVerificationLoadDecode(HeapArgSpace* heapArgSpace, int* addr)
{
    int reg = heapArgSpace->regMap & 0xFF;
    if (!FPREG(reg)) {
        assert(reg < 16);
        *addr = heapArgSpace->coreRegs[reg];
    } else {
        assert(!DOUBLEREG(reg));
        *addr = heapArgSpace->fpRegs[(reg & FP_REG_MASK)];
    }
}

/* Decode contents of heapArgSpace to determine reg to load into */
static void selfVerificationLoadDecodeData(HeapArgSpace* heapArgSpace,
                                           int data, int reg)
{
    if (!FPREG(reg)) {
        assert(reg < 16);
        heapArgSpace->coreRegs[reg] = data;
    } else {
        assert(!DOUBLEREG(reg));
        heapArgSpace->fpRegs[(reg & FP_REG_MASK)] = data;
    }
}

static void selfVerificationLoad(InterpState* interpState)
{
    Thread *self = dvmThreadSelf();
    ShadowHeap *heapSpacePtr;
    ShadowSpace *shadowSpace = self->shadowSpace;
    HeapArgSpace *heapArgSpace = &(interpState->heapArgSpace);

    int addr, data;
    selfVerificationLoadDecode(heapArgSpace, &addr);

    for (heapSpacePtr = shadowSpace->heapSpace;
         heapSpacePtr != shadowSpace->heapSpaceTail; heapSpacePtr++) {
        if (heapSpacePtr->addr == addr) {
            data = heapSpacePtr->data;
            break;
        }
    }

    if (heapSpacePtr == shadowSpace->heapSpaceTail)
        data = *((unsigned int*) addr);

    int reg = (heapArgSpace->regMap >> 8) & 0xFF;

    // LOGD("*** HEAP LOAD: Reg:%d Addr: 0x%x Data: 0x%x", reg, addr, data);

    selfVerificationLoadDecodeData(heapArgSpace, data, reg);
}

static void selfVerificationLoadByte(InterpState* interpState)
{
    Thread *self = dvmThreadSelf();
    ShadowHeap *heapSpacePtr;
    ShadowSpace *shadowSpace = self->shadowSpace;
    HeapArgSpace *heapArgSpace = &(interpState->heapArgSpace);

    int addr, data;
    selfVerificationLoadDecode(heapArgSpace, &addr);

    int maskedAddr = addr & 0xFFFFFFFC;
    int alignment = addr & 0x3;

    for (heapSpacePtr = shadowSpace->heapSpace;
         heapSpacePtr != shadowSpace->heapSpaceTail; heapSpacePtr++) {
        if (heapSpacePtr->addr == maskedAddr) {
            addr = ((unsigned int) &(heapSpacePtr->data)) | alignment;
            data = *((unsigned char*) addr);
            break;
        }
    }

    if (heapSpacePtr == shadowSpace->heapSpaceTail)
        data = *((unsigned char*) addr);

    //LOGD("*** HEAP LOAD BYTE: Addr: 0x%x Data: 0x%x", addr, data);

    int reg = (heapArgSpace->regMap >> 8) & 0xFF;
    selfVerificationLoadDecodeData(heapArgSpace, data, reg);
}

static void selfVerificationLoadHalfword(InterpState* interpState)
{
    Thread *self = dvmThreadSelf();
    ShadowHeap *heapSpacePtr;
    ShadowSpace *shadowSpace = self->shadowSpace;
    HeapArgSpace *heapArgSpace = &(interpState->heapArgSpace);

    int addr, data;
    selfVerificationLoadDecode(heapArgSpace, &addr);

    int maskedAddr = addr & 0xFFFFFFFC;
    int alignment = addr & 0x2;

    for (heapSpacePtr = shadowSpace->heapSpace;
         heapSpacePtr != shadowSpace->heapSpaceTail; heapSpacePtr++) {
        if (heapSpacePtr->addr == maskedAddr) {
            addr = ((unsigned int) &(heapSpacePtr->data)) | alignment;
            data = *((unsigned short*) addr);
            break;
        }
    }

    if (heapSpacePtr == shadowSpace->heapSpaceTail)
        data = *((unsigned short*) addr);

    //LOGD("*** HEAP LOAD kHalfWord: Addr: 0x%x Data: 0x%x", addr, data);

    int reg = (heapArgSpace->regMap >> 8) & 0xFF;
    selfVerificationLoadDecodeData(heapArgSpace, data, reg);
}

static void selfVerificationLoadSignedByte(InterpState* interpState)
{
    Thread *self = dvmThreadSelf();
    ShadowHeap* heapSpacePtr;
    ShadowSpace* shadowSpace = self->shadowSpace;
    HeapArgSpace *heapArgSpace = &(interpState->heapArgSpace);

    int addr, data;
    selfVerificationLoadDecode(heapArgSpace, &addr);

    int maskedAddr = addr & 0xFFFFFFFC;
    int alignment = addr & 0x3;

    for (heapSpacePtr = shadowSpace->heapSpace;
         heapSpacePtr != shadowSpace->heapSpaceTail; heapSpacePtr++) {
        if (heapSpacePtr->addr == maskedAddr) {
            addr = ((unsigned int) &(heapSpacePtr->data)) | alignment;
            data = *((signed char*) addr);
            break;
        }
    }

    if (heapSpacePtr == shadowSpace->heapSpaceTail)
        data = *((signed char*) addr);

    //LOGD("*** HEAP LOAD SIGNED BYTE: Addr: 0x%x Data: 0x%x", addr, data);

    int reg = (heapArgSpace->regMap >> 8) & 0xFF;
    selfVerificationLoadDecodeData(heapArgSpace, data, reg);
}

static void selfVerificationLoadSignedHalfword(InterpState* interpState)
{
    Thread *self = dvmThreadSelf();
    ShadowHeap* heapSpacePtr;
    ShadowSpace* shadowSpace = self->shadowSpace;
    HeapArgSpace *heapArgSpace = &(interpState->heapArgSpace);

    int addr, data;
    selfVerificationLoadDecode(heapArgSpace, &addr);

    int maskedAddr = addr & 0xFFFFFFFC;
    int alignment = addr & 0x2;

    for (heapSpacePtr = shadowSpace->heapSpace;
         heapSpacePtr != shadowSpace->heapSpaceTail; heapSpacePtr++) {
        if (heapSpacePtr->addr == maskedAddr) {
            addr = ((unsigned int) &(heapSpacePtr->data)) | alignment;
            data = *((signed short*) addr);
            break;
        }
    }

    if (heapSpacePtr == shadowSpace->heapSpaceTail)
        data = *((signed short*) addr);

    //LOGD("*** HEAP LOAD SIGNED kHalfWord: Addr: 0x%x Data: 0x%x", addr, data);

    int reg = (heapArgSpace->regMap >> 8) & 0xFF;
    selfVerificationLoadDecodeData(heapArgSpace, data, reg);
}

static void selfVerificationLoadDoubleword(InterpState* interpState)
{
    Thread *self = dvmThreadSelf();
    ShadowHeap* heapSpacePtr;
    ShadowSpace* shadowSpace = self->shadowSpace;
    HeapArgSpace *heapArgSpace = &(interpState->heapArgSpace);

    int addr;
    selfVerificationLoadDecode(heapArgSpace, &addr);

    int addr2 = addr+4;
    unsigned int data = *((unsigned int*) addr);
    unsigned int data2 = *((unsigned int*) addr2);

    for (heapSpacePtr = shadowSpace->heapSpace;
         heapSpacePtr != shadowSpace->heapSpaceTail; heapSpacePtr++) {
        if (heapSpacePtr->addr == addr) {
            data = heapSpacePtr->data;
        } else if (heapSpacePtr->addr == addr2) {
            data2 = heapSpacePtr->data;
        }
    }

    // LOGD("*** HEAP LOAD DOUBLEWORD: Addr: 0x%x Data: 0x%x Data2: 0x%x",
    //    addr, data, data2);

    int reg = (heapArgSpace->regMap >> 8) & 0xFF;
    int reg2 = (heapArgSpace->regMap >> 16) & 0xFF;
    selfVerificationLoadDecodeData(heapArgSpace, data, reg);
    selfVerificationLoadDecodeData(heapArgSpace, data2, reg2);
}

/* Decode contents of heapArgSpace to determine arguments to store. */
static void selfVerificationStoreDecode(HeapArgSpace* heapArgSpace,
                                        int* value, int reg)
{
    if (!FPREG(reg)) {
        assert(reg < 16);
        *value = heapArgSpace->coreRegs[reg];
    } else {
        assert(!DOUBLEREG(reg));
        *value = heapArgSpace->fpRegs[(reg & FP_REG_MASK)];
    }
}

static void selfVerificationStore(InterpState* interpState)
{
    Thread *self = dvmThreadSelf();
    ShadowHeap *heapSpacePtr;
    ShadowSpace *shadowSpace = self->shadowSpace;
    HeapArgSpace *heapArgSpace = &(interpState->heapArgSpace);

    int addr, data;
    int reg0 = heapArgSpace->regMap & 0xFF;
    int reg1 = (heapArgSpace->regMap >> 8) & 0xFF;
    selfVerificationStoreDecode(heapArgSpace, &addr, reg0);
    selfVerificationStoreDecode(heapArgSpace, &data, reg1);

    //LOGD("*** HEAP STORE: Addr: 0x%x Data: 0x%x", addr, data);

    for (heapSpacePtr = shadowSpace->heapSpace;
         heapSpacePtr != shadowSpace->heapSpaceTail; heapSpacePtr++) {
        if (heapSpacePtr->addr == addr) break;
    }

    if (heapSpacePtr == shadowSpace->heapSpaceTail) {
        heapSpacePtr->addr = addr;
        shadowSpace->heapSpaceTail++;
    }

    heapSpacePtr->data = data;
}

static void selfVerificationStoreByte(InterpState* interpState)
{
    Thread *self = dvmThreadSelf();
    ShadowHeap *heapSpacePtr;
    ShadowSpace *shadowSpace = self->shadowSpace;
    HeapArgSpace *heapArgSpace = &(interpState->heapArgSpace);

    int addr, data;
    int reg0 = heapArgSpace->regMap & 0xFF;
    int reg1 = (heapArgSpace->regMap >> 8) & 0xFF;
    selfVerificationStoreDecode(heapArgSpace, &addr, reg0);
    selfVerificationStoreDecode(heapArgSpace, &data, reg1);

    int maskedAddr = addr & 0xFFFFFFFC;
    int alignment = addr & 0x3;

    //LOGD("*** HEAP STORE BYTE: Addr: 0x%x Data: 0x%x", addr, data);

    for (heapSpacePtr = shadowSpace->heapSpace;
         heapSpacePtr != shadowSpace->heapSpaceTail; heapSpacePtr++) {
        if (heapSpacePtr->addr == maskedAddr) break;
    }

    if (heapSpacePtr == shadowSpace->heapSpaceTail)  {
        heapSpacePtr->addr = maskedAddr;
        heapSpacePtr->data = *((unsigned int*) maskedAddr);
        shadowSpace->heapSpaceTail++;
    }

    addr = ((unsigned int) &(heapSpacePtr->data)) | alignment;
    *((unsigned char*) addr) = (char) data;

    //LOGD("*** HEAP STORE BYTE: Addr: 0x%x Final Data: 0x%x",
    //    addr, heapSpacePtr->data);
}

static void selfVerificationStoreHalfword(InterpState* interpState)
{
    Thread *self = dvmThreadSelf();
    ShadowHeap *heapSpacePtr;
    ShadowSpace *shadowSpace = self->shadowSpace;
    HeapArgSpace *heapArgSpace = &(interpState->heapArgSpace);

    int addr, data;
    int reg0 = heapArgSpace->regMap & 0xFF;
    int reg1 = (heapArgSpace->regMap >> 8) & 0xFF;
    selfVerificationStoreDecode(heapArgSpace, &addr, reg0);
    selfVerificationStoreDecode(heapArgSpace, &data, reg1);

    int maskedAddr = addr & 0xFFFFFFFC;
    int alignment = addr & 0x2;

    //LOGD("*** HEAP STORE kHalfWord: Addr: 0x%x Data: 0x%x", addr, data);

    for (heapSpacePtr = shadowSpace->heapSpace;
         heapSpacePtr != shadowSpace->heapSpaceTail; heapSpacePtr++) {
        if (heapSpacePtr->addr == maskedAddr) break;
    }

    if (heapSpacePtr == shadowSpace->heapSpaceTail)  {
        heapSpacePtr->addr = maskedAddr;
        heapSpacePtr->data = *((unsigned int*) maskedAddr);
        shadowSpace->heapSpaceTail++;
    }

    addr = ((unsigned int) &(heapSpacePtr->data)) | alignment;
    *((unsigned short*) addr) = (short) data;

    //LOGD("*** HEAP STORE kHalfWord: Addr: 0x%x Final Data: 0x%x",
    //    addr, heapSpacePtr->data);
}

static void selfVerificationStoreDoubleword(InterpState* interpState)
{
    Thread *self = dvmThreadSelf();
    ShadowHeap *heapSpacePtr;
    ShadowSpace *shadowSpace = self->shadowSpace;
    HeapArgSpace *heapArgSpace = &(interpState->heapArgSpace);

    int addr, data, data2;
    int reg0 = heapArgSpace->regMap & 0xFF;
    int reg1 = (heapArgSpace->regMap >> 8) & 0xFF;
    int reg2 = (heapArgSpace->regMap >> 16) & 0xFF;
    selfVerificationStoreDecode(heapArgSpace, &addr, reg0);
    selfVerificationStoreDecode(heapArgSpace, &data, reg1);
    selfVerificationStoreDecode(heapArgSpace, &data2, reg2);

    int addr2 = addr+4;
    bool store1 = false, store2 = false;

    //LOGD("*** HEAP STORE DOUBLEWORD: Addr: 0x%x Data: 0x%x, Data2: 0x%x",
    //    addr, data, data2);

    for (heapSpacePtr = shadowSpace->heapSpace;
         heapSpacePtr != shadowSpace->heapSpaceTail; heapSpacePtr++) {
        if (heapSpacePtr->addr == addr) {
            heapSpacePtr->data = data;
            store1 = true;
        } else if (heapSpacePtr->addr == addr2) {
            heapSpacePtr->data = data2;
            store2 = true;
        }
    }

    if (!store1) {
        shadowSpace->heapSpaceTail->addr = addr;
        shadowSpace->heapSpaceTail->data = data;
        shadowSpace->heapSpaceTail++;
    }
    if (!store2) {
        shadowSpace->heapSpaceTail->addr = addr2;
        shadowSpace->heapSpaceTail->data = data2;
        shadowSpace->heapSpaceTail++;
    }
}

/* Common wrapper function for all memory operations */
static void selfVerificationMemOpWrapper(CompilationUnit *cUnit, int regMap,
                                         void* funct)
{
    /* push r0 and r7 to give us a foothold */
    newLIR1(cUnit, kThumbPush, (1 << r0) | (1 << r7));

    /* Let the save handler know where the save record is */
    loadConstant(cUnit, r0, offsetof(InterpState, heapArgSpace));

    /* Load the regMap and call the save handler [note: handler pops r0/r7] */
    loadConstant(cUnit, r7, regMap);
    genDispatchToHandler(cUnit, TEMPLATE_SAVE_STATE);

    /* Set function pointer, pass rGLUE and branch */
    loadConstant(cUnit, r1, (int) funct);
    newLIR2(cUnit, kThumbMovRR, r0, rGLUE);
    newLIR1(cUnit, kThumbBlxR, r1);

    /* Let the recover handler know where coreRegs[0] and restore regs */
    loadConstant(cUnit, r0, offsetof(InterpState, heapArgSpace) +
                            offsetof(HeapArgSpace, coreRegs));
    genDispatchToHandler(cUnit, TEMPLATE_RESTORE_STATE);
}
#endif

/*
 * Load a Dalvik register into a physical register.  Take care when
 * using this routine, as it doesn't perform any bookkeeping regarding
 * register liveness.  That is the responsibility of the caller.
 */
static void loadValueDirect(CompilationUnit *cUnit, RegLocation rlSrc,
                                int reg1)
{
    rlSrc = updateLoc(cUnit, rlSrc);  /* Is our value hiding in a live temp? */
    if (rlSrc.location == kLocPhysReg) {
        genRegCopy(cUnit, reg1, rlSrc.lowReg);
    } else  if (rlSrc.location == kLocRetval) {
        loadWordDisp(cUnit, rGLUE, offsetof(InterpState, retval), reg1);
    } else {
        assert(rlSrc.location == kLocDalvikFrame);
        loadWordDisp(cUnit, rFP, sReg2vReg(cUnit, rlSrc.sRegLow) << 2,
                     reg1);
    }
}

/*
 * Similar to loadValueDirect, but clobbers and allocates the target
 * register.  Should be used when loading to a fixed register (for example,
 * loading arguments to an out of line call.
 */
static void loadValueDirectFixed(CompilationUnit *cUnit, RegLocation rlSrc,
                                 int reg1)
{
    clobberReg(cUnit, reg1);
    markRegInUse(cUnit, reg1);
    loadValueDirect(cUnit, rlSrc, reg1);
}

/*
 * Load a Dalvik register pair into a physical register[s].  Take care when
 * using this routine, as it doesn't perform any bookkeeping regarding
 * register liveness.  That is the responsibility of the caller.
 */
static void loadValueDirectWide(CompilationUnit *cUnit, RegLocation rlSrc,
                                int regLo, int regHi)
{
    rlSrc = updateLocWide(cUnit, rlSrc);
    if (rlSrc.location == kLocPhysReg) {
        genRegCopyWide(cUnit, regLo, regHi, rlSrc.lowReg, rlSrc.highReg);
    } else if (rlSrc.location == kLocRetval) {
        loadBaseDispWide(cUnit, NULL, rGLUE, offsetof(InterpState, retval),
                         regLo, regHi, false, INVALID_SREG);
    } else {
        assert(rlSrc.location == kLocDalvikFrame);
            loadBaseDispWide(cUnit, NULL, rFP,
                             sReg2vReg(cUnit, rlSrc.sRegLow) << 2,
                             regLo, regHi, false, INVALID_SREG);
    }
}

/*
 * Similar to loadValueDirect, but clobbers and allocates the target
 * registers.  Should be used when loading to a fixed registers (for example,
 * loading arguments to an out of line call.
 */
static void loadValueDirectWideFixed(CompilationUnit *cUnit, RegLocation rlSrc,
                                     int regLo, int regHi)
{
    clobberReg(cUnit, regLo);
    clobberReg(cUnit, regHi);
    markRegInUse(cUnit, regLo);
    markRegInUse(cUnit, regHi);
    loadValueDirectWide(cUnit, rlSrc, regLo, regHi);
}

static RegLocation loadValue(CompilationUnit *cUnit, RegLocation rlSrc,
                             RegisterClass opKind)
{
    RegisterInfo *pReg;
    rlSrc = evalLoc(cUnit, rlSrc, opKind, false);
    if (rlSrc.location == kLocDalvikFrame) {
        loadValueDirect(cUnit, rlSrc, rlSrc.lowReg);
        rlSrc.location = kLocPhysReg;
        markRegLive(cUnit, rlSrc.lowReg, rlSrc.sRegLow);
    } else if (rlSrc.location == kLocRetval) {
        loadWordDisp(cUnit, rGLUE, offsetof(InterpState, retval), rlSrc.lowReg);
        rlSrc.location = kLocPhysReg;
        clobberReg(cUnit, rlSrc.lowReg);
    }
    return rlSrc;
}

static RegLocation loadValueWide(CompilationUnit *cUnit, RegLocation rlSrc,
                                 RegisterClass opKind)
{
    RegisterInfo *pRegLo;
    RegisterInfo *pRegHi;
    assert(rlSrc.wide);
    rlSrc = evalLoc(cUnit, rlSrc, opKind, false);
    if (rlSrc.location == kLocDalvikFrame) {
        loadValueDirectWide(cUnit, rlSrc, rlSrc.lowReg, rlSrc.highReg);
        rlSrc.location = kLocPhysReg;
        markRegLive(cUnit, rlSrc.lowReg, rlSrc.sRegLow);
        markRegLive(cUnit, rlSrc.highReg, hiSReg(rlSrc.sRegLow));
    } else if (rlSrc.location == kLocRetval) {
        loadBaseDispWide(cUnit, NULL, rGLUE, offsetof(InterpState, retval),
                         rlSrc.lowReg, rlSrc.highReg, false, INVALID_SREG);
        rlSrc.location = kLocPhysReg;
        clobberReg(cUnit, rlSrc.lowReg);
        clobberReg(cUnit, rlSrc.highReg);
    }
    return rlSrc;
}

static void storeValue(CompilationUnit *cUnit, RegLocation rlDest,
                       RegLocation rlSrc)
{
    RegisterInfo *pRegLo;
    LIR *defStart;
    LIR *defEnd;
    assert(!rlDest.wide);
    assert(!rlSrc.wide);
    killNullCheckedLocation(cUnit, rlDest);
    rlSrc = updateLoc(cUnit, rlSrc);
    rlDest = updateLoc(cUnit, rlDest);
    if (rlSrc.location == kLocPhysReg) {
        if (isLive(cUnit, rlSrc.lowReg) || (rlDest.location == kLocPhysReg)) {
            // Src is live or Dest has assigned reg.
            rlDest = evalLoc(cUnit, rlDest, kAnyReg, false);
            genRegCopy(cUnit, rlDest.lowReg, rlSrc.lowReg);
        } else {
            // Just re-assign the registers.  Dest gets Src's regs
            rlDest.lowReg = rlSrc.lowReg;
            clobberReg(cUnit, rlSrc.lowReg);
        }
    } else {
        // Load Src either into promoted Dest or temps allocated for Dest
        rlDest = evalLoc(cUnit, rlDest, kAnyReg, false);
        loadValueDirect(cUnit, rlSrc, rlDest.lowReg);
    }

    // Dest is now live and dirty (until/if we flush it to home location)
    markRegLive(cUnit, rlDest.lowReg, rlDest.sRegLow);
    markRegDirty(cUnit, rlDest.lowReg);


    if (rlDest.location == kLocRetval) {
        storeBaseDisp(cUnit, rGLUE, offsetof(InterpState, retval),
                      rlDest.lowReg, kWord);
        clobberReg(cUnit, rlDest.lowReg);
    } else {
        resetDefLoc(cUnit, rlDest);
        if (liveOut(cUnit, rlDest.sRegLow)) {
            defStart = (LIR *)cUnit->lastLIRInsn;
            int vReg = sReg2vReg(cUnit, rlDest.sRegLow);
            storeBaseDisp(cUnit, rFP, vReg << 2, rlDest.lowReg, kWord);
            markRegClean(cUnit, rlDest.lowReg);
            defEnd = (LIR *)cUnit->lastLIRInsn;
            markDef(cUnit, rlDest, defStart, defEnd);
        }
    }
}

static void storeValueWide(CompilationUnit *cUnit, RegLocation rlDest,
                       RegLocation rlSrc)
{
    RegisterInfo *pRegLo;
    RegisterInfo *pRegHi;
    LIR *defStart;
    LIR *defEnd;
    bool srcFP = FPREG(rlSrc.lowReg) && FPREG(rlSrc.highReg);
    assert(FPREG(rlSrc.lowReg)==FPREG(rlSrc.highReg));
    assert(rlDest.wide);
    assert(rlSrc.wide);
    killNullCheckedLocation(cUnit, rlDest);
    if (rlSrc.location == kLocPhysReg) {
        if (isLive(cUnit, rlSrc.lowReg) || isLive(cUnit, rlSrc.highReg) ||
            (rlDest.location == kLocPhysReg)) {
            // Src is live or Dest has assigned reg.
            rlDest = evalLoc(cUnit, rlDest, kAnyReg, false);
            genRegCopyWide(cUnit, rlDest.lowReg, rlDest.highReg,
                           rlSrc.lowReg, rlSrc.highReg);
        } else {
            // Just re-assign the registers.  Dest gets Src's regs
            rlDest.lowReg = rlSrc.lowReg;
            rlDest.highReg = rlSrc.highReg;
            clobberReg(cUnit, rlSrc.lowReg);
            clobberReg(cUnit, rlSrc.highReg);
        }
    } else {
        // Load Src either into promoted Dest or temps allocated for Dest
        rlDest = evalLoc(cUnit, rlDest, kAnyReg, false);
        loadValueDirectWide(cUnit, rlSrc, rlDest.lowReg,
                            rlDest.highReg);
    }

    // Dest is now live and dirty (until/if we flush it to home location)
    markRegLive(cUnit, rlDest.lowReg, rlDest.sRegLow);
    markRegLive(cUnit, rlDest.highReg, hiSReg(rlDest.sRegLow));
    markRegDirty(cUnit, rlDest.lowReg);
    markRegDirty(cUnit, rlDest.highReg);
    markRegPair(cUnit, rlDest.lowReg, rlDest.highReg);


    if (rlDest.location == kLocRetval) {
        storeBaseDispWide(cUnit, rGLUE, offsetof(InterpState, retval),
                          rlDest.lowReg, rlDest.highReg);
        clobberReg(cUnit, rlDest.lowReg);
        clobberReg(cUnit, rlDest.highReg);
    } else {
        resetDefLocWide(cUnit, rlDest);
        if (liveOut(cUnit, rlDest.sRegLow) ||
            liveOut(cUnit, hiSReg(rlDest.sRegLow))) {
            defStart = (LIR *)cUnit->lastLIRInsn;
            int vReg = sReg2vReg(cUnit, rlDest.sRegLow);
            assert((vReg+1) == sReg2vReg(cUnit, hiSReg(rlDest.sRegLow)));
            storeBaseDispWide(cUnit, rFP, vReg << 2, rlDest.lowReg,
                              rlDest.highReg);
            markRegClean(cUnit, rlDest.lowReg);
            markRegClean(cUnit, rlDest.highReg);
            defEnd = (LIR *)cUnit->lastLIRInsn;
            markDefWide(cUnit, rlDest, defStart, defEnd);
        }
    }
}

/*
 * Load an immediate value into a fixed or temp register.  Target
 * register is clobbered, and marked inUse.
 */
static ArmLIR *loadConstant(CompilationUnit *cUnit, int rDest, int value)
{
    if (isTemp(cUnit, rDest)) {
        clobberReg(cUnit, rDest);
        markRegInUse(cUnit, rDest);
    }
    return loadConstantValue(cUnit, rDest, value);
}

/*
 * Mark load/store instructions that access Dalvik registers through rFP +
 * offset.
 */
static void annotateDalvikRegAccess(ArmLIR *lir, int regId, bool isLoad)
{
    if (isLoad) {
        lir->useMask |= ENCODE_DALVIK_REG;
    } else {
        lir->defMask |= ENCODE_DALVIK_REG;
    }

    /*
     * Store the Dalvik register id in aliasInfo. Mark he MSB if it is a 64-bit
     * access.
     */
    lir->aliasInfo = regId;
    if (DOUBLEREG(lir->operands[0])) {
        lir->aliasInfo |= 0x80000000;
    }
}

/*
 * Decode the register id and mark the corresponding bit(s).
 */
static inline void setupRegMask(u8 *mask, int reg)
{
    u8 seed;
    int shift;
    int regId = reg & 0x1f;

    /*
     * Each double register is equal to a pair of single-precision FP registers
     */
    seed = DOUBLEREG(reg) ? 3 : 1;
    /* FP register starts at bit position 16 */
    shift = FPREG(reg) ? kFPReg0 : 0;
    /* Expand the double register id into single offset */
    shift += regId;
    *mask |= seed << shift;
}

/*
 * Set up the proper fields in the resource mask
 */
static void setupResourceMasks(ArmLIR *lir)
{
    int opCode = lir->opCode;
    int flags;

    if (opCode <= 0) {
        lir->useMask = lir->defMask = 0;
        return;
    }

    flags = EncodingMap[lir->opCode].flags;

    /* Set up the mask for resources that are updated */
    if (flags & IS_BRANCH) {
        lir->defMask |= ENCODE_REG_PC;
        lir->useMask |= ENCODE_REG_PC;
    }

    if (flags & REG_DEF0) {
        setupRegMask(&lir->defMask, lir->operands[0]);
    }

    if (flags & REG_DEF1) {
        setupRegMask(&lir->defMask, lir->operands[1]);
    }

    if (flags & REG_DEF_SP) {
        lir->defMask |= ENCODE_REG_SP;
    }

    if (flags & REG_DEF_SP) {
        lir->defMask |= ENCODE_REG_LR;
    }

    if (flags & REG_DEF_LIST0) {
        lir->defMask |= ENCODE_REG_LIST(lir->operands[0]);
    }

    if (flags & REG_DEF_LIST1) {
        lir->defMask |= ENCODE_REG_LIST(lir->operands[1]);
    }

    if (flags & SETS_CCODES) {
        lir->defMask |= ENCODE_CCODE;
    }

    /* Conservatively treat the IT block */
    if (flags & IS_IT) {
        lir->defMask = ENCODE_ALL;
    }

    /* Set up the mask for resources that are used */
    if (flags & IS_BRANCH) {
        lir->useMask |= ENCODE_REG_PC;
    }

    if (flags & (REG_USE0 | REG_USE1 | REG_USE2 | REG_USE3)) {
        int i;

        for (i = 0; i < 4; i++) {
            if (flags & (1 << (kRegUse0 + i))) {
                setupRegMask(&lir->useMask, lir->operands[i]);
            }
        }
    }

    if (flags & REG_USE_PC) {
        lir->useMask |= ENCODE_REG_PC;
    }

    if (flags & REG_USE_SP) {
        lir->useMask |= ENCODE_REG_SP;
    }

    if (flags & REG_USE_LIST0) {
        lir->useMask |= ENCODE_REG_LIST(lir->operands[0]);
    }

    if (flags & REG_USE_LIST1) {
        lir->useMask |= ENCODE_REG_LIST(lir->operands[1]);
    }

    if (flags & USES_CCODES) {
        lir->useMask |= ENCODE_CCODE;
    }
}

/*
 * The following are building blocks to construct low-level IRs with 0 - 4
 * operands.
 */
static ArmLIR *newLIR0(CompilationUnit *cUnit, ArmOpCode opCode)
{
    ArmLIR *insn = dvmCompilerNew(sizeof(ArmLIR), true);
    assert(isPseudoOpCode(opCode) || (EncodingMap[opCode].flags & NO_OPERAND));
    insn->opCode = opCode;
    setupResourceMasks(insn);
    dvmCompilerAppendLIR(cUnit, (LIR *) insn);
    return insn;
}

static ArmLIR *newLIR1(CompilationUnit *cUnit, ArmOpCode opCode,
                           int dest)
{
    ArmLIR *insn = dvmCompilerNew(sizeof(ArmLIR), true);
    assert(isPseudoOpCode(opCode) || (EncodingMap[opCode].flags & IS_UNARY_OP));
    insn->opCode = opCode;
    insn->operands[0] = dest;
    setupResourceMasks(insn);
    dvmCompilerAppendLIR(cUnit, (LIR *) insn);
    return insn;
}

static ArmLIR *newLIR2(CompilationUnit *cUnit, ArmOpCode opCode,
                           int dest, int src1)
{
    ArmLIR *insn = dvmCompilerNew(sizeof(ArmLIR), true);
    assert(isPseudoOpCode(opCode) ||
           (EncodingMap[opCode].flags & IS_BINARY_OP));
    insn->opCode = opCode;
    insn->operands[0] = dest;
    insn->operands[1] = src1;
    setupResourceMasks(insn);
    dvmCompilerAppendLIR(cUnit, (LIR *) insn);
    return insn;
}

static ArmLIR *newLIR3(CompilationUnit *cUnit, ArmOpCode opCode,
                           int dest, int src1, int src2)
{
    ArmLIR *insn = dvmCompilerNew(sizeof(ArmLIR), true);
    if (!(EncodingMap[opCode].flags & IS_TERTIARY_OP)) {
        LOGE("Bad LIR3: %s[%d]",EncodingMap[opCode].name,opCode);
    }
    assert(isPseudoOpCode(opCode) ||
           (EncodingMap[opCode].flags & IS_TERTIARY_OP));
    insn->opCode = opCode;
    insn->operands[0] = dest;
    insn->operands[1] = src1;
    insn->operands[2] = src2;
    setupResourceMasks(insn);
    dvmCompilerAppendLIR(cUnit, (LIR *) insn);
    return insn;
}

static ArmLIR *newLIR4(CompilationUnit *cUnit, ArmOpCode opCode,
                           int dest, int src1, int src2, int info)
{
    ArmLIR *insn = dvmCompilerNew(sizeof(ArmLIR), true);
    assert(isPseudoOpCode(opCode) ||
           (EncodingMap[opCode].flags & IS_QUAD_OP));
    insn->opCode = opCode;
    insn->operands[0] = dest;
    insn->operands[1] = src1;
    insn->operands[2] = src2;
    insn->operands[3] = info;
    setupResourceMasks(insn);
    dvmCompilerAppendLIR(cUnit, (LIR *) insn);
    return insn;
}

/*
 * If the next instruction is a move-result or move-result-long,
 * return the target Dalvik sReg[s] and convert the next to a
 * nop.  Otherwise, return INVALID_SREG.  Used to optimize method inlining.
 */
static RegLocation inlinedTarget(CompilationUnit *cUnit, MIR *mir,
                                  bool fpHint)
{
    if (mir->next &&
        ((mir->next->dalvikInsn.opCode == OP_MOVE_RESULT) ||
         (mir->next->dalvikInsn.opCode == OP_MOVE_RESULT_OBJECT))) {
        mir->next->dalvikInsn.opCode = OP_NOP;
        return getDestLoc(cUnit, mir->next, 0);
    } else {
        RegLocation res = LOC_DALVIK_RETURN_VAL;
        res.fp = fpHint;
        return res;
    }
}

static RegLocation inlinedTargetWide(CompilationUnit *cUnit, MIR *mir,
                                      bool fpHint)
{
    if (mir->next &&
        (mir->next->dalvikInsn.opCode == OP_MOVE_RESULT_WIDE)) {
        mir->next->dalvikInsn.opCode = OP_NOP;
        return getDestLocWide(cUnit, mir->next, 0, 1);
    } else {
        RegLocation res = LOC_DALVIK_RETURN_VAL_WIDE;
        res.fp = fpHint;
        return res;
    }
}

/*
 * The following are building blocks to insert constants into the pool or
 * instruction streams.
 */

/* Add a 32-bit constant either in the constant pool or mixed with code */
static ArmLIR *addWordData(CompilationUnit *cUnit, int value, bool inPlace)
{
    /* Add the constant to the literal pool */
    if (!inPlace) {
        ArmLIR *newValue = dvmCompilerNew(sizeof(ArmLIR), true);
        newValue->operands[0] = value;
        newValue->generic.next = cUnit->wordList;
        cUnit->wordList = (LIR *) newValue;
        return newValue;
    } else {
        /* Add the constant in the middle of code stream */
        newLIR1(cUnit, kArm16BitData, (value & 0xffff));
        newLIR1(cUnit, kArm16BitData, (value >> 16));
    }
    return NULL;
}

/*
 * Search the existing constants in the literal pool for an exact or close match
 * within specified delta (greater or equal to 0).
 */
static ArmLIR *scanLiteralPool(CompilationUnit *cUnit, int value,
                                   unsigned int delta)
{
    LIR *dataTarget = cUnit->wordList;
    while (dataTarget) {
        if (((unsigned) (value - ((ArmLIR *) dataTarget)->operands[0])) <=
            delta)
            return (ArmLIR *) dataTarget;
        dataTarget = dataTarget->next;
    }
    return NULL;
}

/* Create the PC reconstruction slot if not already done */
static inline ArmLIR *genCheckCommon(CompilationUnit *cUnit, int dOffset,
                                         ArmLIR *branch,
                                         ArmLIR *pcrLabel)
{
    /* Set up the place holder to reconstruct this Dalvik PC */
    if (pcrLabel == NULL) {
        int dPC = (int) (cUnit->method->insns + dOffset);
        pcrLabel = dvmCompilerNew(sizeof(ArmLIR), true);
        pcrLabel->opCode = ARM_PSEUDO_kPCReconstruction_CELL;
        pcrLabel->operands[0] = dPC;
        pcrLabel->operands[1] = dOffset;
        /* Insert the place holder to the growable list */
        dvmInsertGrowableList(&cUnit->pcReconstructionList, pcrLabel);
    }
    /* Branch to the PC reconstruction code */
    branch->generic.target = (LIR *) pcrLabel;
    return pcrLabel;
}


/*
 * Perform a "reg cmp reg" operation and jump to the PCR region if condition
 * satisfies.
 */
static inline ArmLIR *genRegRegCheck(CompilationUnit *cUnit,
                                     ArmConditionCode cond,
                                     int reg1, int reg2, int dOffset,
                                     ArmLIR *pcrLabel)
{
    ArmLIR *res;
    res = opRegReg(cUnit, kOpCmp, reg1, reg2);
    ArmLIR *branch = opCondBranch(cUnit, cond);
    genCheckCommon(cUnit, dOffset, branch, pcrLabel);
    return res;
}

/*
 * Perform null-check on a register. sReg is the ssa register being checked,
 * and mReg is the machine register holding the actual value. If internal state
 * indicates that sReg has been checked before the check request is ignored.
 */
static ArmLIR *genNullCheck(CompilationUnit *cUnit, int sReg, int mReg,
                                int dOffset, ArmLIR *pcrLabel)
{
    /* This particular Dalvik register has been null-checked */
    if (dvmIsBitSet(cUnit->regPool->nullCheckedRegs, sReg)) {
        return pcrLabel;
    }
    dvmSetBit(cUnit->regPool->nullCheckedRegs, sReg);
    return genRegImmCheck(cUnit, kArmCondEq, mReg, 0, dOffset, pcrLabel);
}

/*
 * Perform zero-check on a register. Similar to genNullCheck but the value being
 * checked does not have a corresponding Dalvik register.
 */
static ArmLIR *genZeroCheck(CompilationUnit *cUnit, int mReg,
                                int dOffset, ArmLIR *pcrLabel)
{
    return genRegImmCheck(cUnit, kArmCondEq, mReg, 0, dOffset, pcrLabel);
}

/* Perform bound check on two registers */
static ArmLIR *genBoundsCheck(CompilationUnit *cUnit, int rIndex,
                                  int rBound, int dOffset, ArmLIR *pcrLabel)
{
    return genRegRegCheck(cUnit, kArmCondCs, rIndex, rBound, dOffset,
                            pcrLabel);
}

/* Generate a unconditional branch to go to the interpreter */
static inline ArmLIR *genTrap(CompilationUnit *cUnit, int dOffset,
                                  ArmLIR *pcrLabel)
{
    ArmLIR *branch = opNone(cUnit, kOpUncondBr);
    return genCheckCommon(cUnit, dOffset, branch, pcrLabel);
}

/* Load a wide field from an object instance */
static void genIGetWide(CompilationUnit *cUnit, MIR *mir, int fieldOffset)
{
    DecodedInstruction *dInsn = &mir->dalvikInsn;
    RegLocation rlObj = getSrcLoc(cUnit, mir, 0);
    RegLocation rlDest = getDestLocWide(cUnit, mir, 0, 1);
    RegLocation rlResult;
    rlObj = loadValue(cUnit, rlObj, kCoreReg);
    int regPtr = allocTemp(cUnit);

    assert(rlDest.wide);

    genNullCheck(cUnit, rlObj.sRegLow, rlObj.lowReg, mir->offset,
                 NULL);/* null object? */
    opRegRegImm(cUnit, kOpAdd, regPtr, rlObj.lowReg, fieldOffset);
    rlResult = evalLoc(cUnit, rlDest, kAnyReg, true);
#if !defined(WITH_SELF_VERIFICATION)
    loadPair(cUnit, regPtr, rlResult.lowReg, rlResult.highReg);
#else
    int regMap = rlResult.highReg << 16 | rlResult.lowReg << 8 | regPtr;
    selfVerificationMemOpWrapper(cUnit, regMap,
        &selfVerificationLoadDoubleword);
#endif
    freeTemp(cUnit, regPtr);
    storeValueWide(cUnit, rlDest, rlResult);
}

/* Store a wide field to an object instance */
static void genIPutWide(CompilationUnit *cUnit, MIR *mir, int fieldOffset)
{
    DecodedInstruction *dInsn = &mir->dalvikInsn;
    RegLocation rlSrc = getSrcLocWide(cUnit, mir, 0, 1);
    RegLocation rlObj = getSrcLoc(cUnit, mir, 2);
    rlObj = loadValue(cUnit, rlObj, kCoreReg);
    int regPtr;
    rlSrc = loadValueWide(cUnit, rlSrc, kAnyReg);
    genNullCheck(cUnit, rlObj.sRegLow, rlObj.lowReg, mir->offset,
                 NULL);/* null object? */
    regPtr = allocTemp(cUnit);
    opRegRegImm(cUnit, kOpAdd, regPtr, rlObj.lowReg, fieldOffset);
#if !defined(WITH_SELF_VERIFICATION)
    storePair(cUnit, regPtr, rlSrc.lowReg, rlSrc.highReg);
#else
    int regMap = rlSrc.highReg << 16 | rlSrc.lowReg << 8 | regPtr;
    selfVerificationMemOpWrapper(cUnit, regMap,
        &selfVerificationStoreDoubleword);
#endif
    freeTemp(cUnit, regPtr);
}

/*
 * Load a field from an object instance
 *
 */
static void genIGet(CompilationUnit *cUnit, MIR *mir, OpSize size,
                    int fieldOffset)
{
    int regPtr;
    RegLocation rlResult;
    DecodedInstruction *dInsn = &mir->dalvikInsn;
    RegLocation rlObj = getSrcLoc(cUnit, mir, 0);
    RegLocation rlDest = getDestLoc(cUnit, mir, 0);
    rlObj = loadValue(cUnit, rlObj, kCoreReg);
    rlResult = evalLoc(cUnit, rlDest, kAnyReg, true);
#if !defined(WITH_SELF_VERIFICATION)
    loadBaseDisp(cUnit, mir, rlObj.lowReg, fieldOffset, rlResult.lowReg,
                 size, true, rlObj.sRegLow);
#else
    genNullCheck(cUnit, rlObj.sRegLow, rlObj.lowReg, mir->offset,
                 NULL);/* null object? */
    /* Combine address and offset */
    regPtr = allocTemp(cUnit);
    opRegRegImm(cUnit, kOpAdd, regPtr, rlObj.lowReg, fieldOffset);

    int regMap = rlResult.lowReg << 8 | regPtr;
    selfVerificationMemOpWrapper(cUnit, regMap, &selfVerificationLoad);
    freeTemp(cUnit, regPtr);
#endif
    storeValue(cUnit, rlDest, rlResult);
}

/*
 * Store a field to an object instance
 *
 */
static void genIPut(CompilationUnit *cUnit, MIR *mir, OpSize size,
                    int fieldOffset)
{
    DecodedInstruction *dInsn = &mir->dalvikInsn;
    RegLocation rlSrc = getSrcLoc(cUnit, mir, 0);
    RegLocation rlObj = getSrcLoc(cUnit, mir, 1);
    rlObj = loadValue(cUnit, rlObj, kCoreReg);
    rlSrc = loadValue(cUnit, rlSrc, kAnyReg);
    int regPtr;
    genNullCheck(cUnit, rlObj.sRegLow, rlObj.lowReg, mir->offset,
                 NULL);/* null object? */
#if !defined(WITH_SELF_VERIFICATION)
    storeBaseDisp(cUnit, rlObj.lowReg, fieldOffset, rlSrc.lowReg, size);
#else
    /* Combine address and offset */
    regPtr = allocTemp(cUnit);
    opRegRegImm(cUnit, kOpAdd, regPtr, rlObj.lowReg, fieldOffset);

    int regMap = rlSrc.lowReg << 8 | regPtr;
    selfVerificationMemOpWrapper(cUnit, regMap, &selfVerificationStore);
#endif
}


/*
 * Generate array load
 */
static void genArrayGet(CompilationUnit *cUnit, MIR *mir, OpSize size,
                        RegLocation rlArray, RegLocation rlIndex,
                        RegLocation rlDest, int scale)
{
    int lenOffset = offsetof(ArrayObject, length);
    int dataOffset = offsetof(ArrayObject, contents);
    RegLocation rlResult;
    rlArray = loadValue(cUnit, rlArray, kCoreReg);
    rlIndex = loadValue(cUnit, rlIndex, kCoreReg);
    int regPtr;

    /* null object? */
    ArmLIR * pcrLabel = NULL;

    if (!(mir->OptimizationFlags & MIR_IGNORE_NULL_CHECK)) {
        pcrLabel = genNullCheck(cUnit, rlArray.sRegLow,
                                rlArray.lowReg, mir->offset, NULL);
    }

    regPtr = allocTemp(cUnit);

    if (!(mir->OptimizationFlags & MIR_IGNORE_RANGE_CHECK)) {
        int regLen = allocTemp(cUnit);
        /* Get len */
        loadWordDisp(cUnit, rlArray.lowReg, lenOffset, regLen);
        /* regPtr -> array data */
        opRegRegImm(cUnit, kOpAdd, regPtr, rlArray.lowReg, dataOffset);
        genBoundsCheck(cUnit, rlIndex.lowReg, regLen, mir->offset,
                       pcrLabel);
        freeTemp(cUnit, regLen);
    } else {
        /* regPtr -> array data */
        opRegRegImm(cUnit, kOpAdd, regPtr, rlArray.lowReg, dataOffset);
    }
#if !defined(WITH_SELF_VERIFICATION)
    if ((size == kLong) || (size == kDouble)) {
        if (scale) {
            int rNewIndex = allocTemp(cUnit);
            opRegRegImm(cUnit, kOpLsl, rNewIndex, rlIndex.lowReg, scale);
            opRegReg(cUnit, kOpAdd, regPtr, rNewIndex);
            freeTemp(cUnit, rNewIndex);
        } else {
            opRegReg(cUnit, kOpAdd, regPtr, rlIndex.lowReg);
        }
        rlResult = evalLoc(cUnit, rlDest, kAnyReg, true);
        loadPair(cUnit, regPtr, rlResult.lowReg, rlResult.highReg);
        freeTemp(cUnit, regPtr);
        storeValueWide(cUnit, rlDest, rlResult);
    } else {
        rlResult = evalLoc(cUnit, rlDest, kAnyReg, true);
        loadBaseIndexed(cUnit, regPtr, rlIndex.lowReg, rlResult.lowReg,
                        scale, size);
        freeTemp(cUnit, regPtr);
        storeValue(cUnit, rlDest, rlResult);
    }
#else
    //TODO: probably want to move this into loadBaseIndexed
    void *funct = NULL;
    switch(size) {
        case kLong:
        case kDouble:
            funct = (void*) &selfVerificationLoadDoubleword;
            break;
        case kWord:
            funct = (void*) &selfVerificationLoad;
            break;
        case kUnsignedHalf:
            funct = (void*) &selfVerificationLoadHalfword;
            break;
        case kSignedHalf:
            funct = (void*) &selfVerificationLoadSignedHalfword;
            break;
        case kUnsignedByte:
            funct = (void*) &selfVerificationLoadByte;
            break;
        case kSignedByte:
            funct = (void*) &selfVerificationLoadSignedByte;
            break;
        default:
            assert(0);
            dvmAbort();
    }
    /* Combine address and index */
    if (scale) {
        int regTmp = allocTemp(cUnit);
        opRegRegImm(cUnit, kOpLsl, regTmp, rlIndex.lowReg, scale);
        opRegReg(cUnit, kOpAdd, regPtr, regTmp);
        freeTemp(cUnit, regTmp);
    } else {
        opRegReg(cUnit, kOpAdd, regPtr, rlIndex.lowReg);
    }

    rlResult = evalLoc(cUnit, rlDest, kAnyReg, true);
    int regMap = rlResult.highReg << 16 | rlResult.lowReg << 8 | regPtr;
    selfVerificationMemOpWrapper(cUnit, regMap, funct);

    freeTemp(cUnit, regPtr);
    if ((size == kLong) || (size == kDouble))
        storeValueWide(cUnit, rlDest, rlResult);
    else
        storeValue(cUnit, rlDest, rlResult);
#endif
}

/*
 * Generate array store
 *
 */
static void genArrayPut(CompilationUnit *cUnit, MIR *mir, OpSize size,
                        RegLocation rlArray, RegLocation rlIndex,
                        RegLocation rlSrc, int scale)
{
    int lenOffset = offsetof(ArrayObject, length);
    int dataOffset = offsetof(ArrayObject, contents);

    int regPtr;
    rlArray = loadValue(cUnit, rlArray, kCoreReg);
    rlIndex = loadValue(cUnit, rlIndex, kCoreReg);

    if (isTemp(cUnit, rlArray.lowReg)) {
        clobberReg(cUnit, rlArray.lowReg);
        regPtr = rlArray.lowReg;
    } else {
        regPtr = allocTemp(cUnit);
        genRegCopy(cUnit, regPtr, rlArray.lowReg);
    }

    /* null object? */
    ArmLIR * pcrLabel = NULL;

    if (!(mir->OptimizationFlags & MIR_IGNORE_NULL_CHECK)) {
        pcrLabel = genNullCheck(cUnit, rlArray.sRegLow, rlArray.lowReg,
                                mir->offset, NULL);
    }

    if (!(mir->OptimizationFlags & MIR_IGNORE_RANGE_CHECK)) {
        int regLen = allocTemp(cUnit);
        //NOTE: max live temps(4) here.
        /* Get len */
        loadWordDisp(cUnit, rlArray.lowReg, lenOffset, regLen);
        /* regPtr -> array data */
        opRegImm(cUnit, kOpAdd, regPtr, dataOffset);
        genBoundsCheck(cUnit, rlIndex.lowReg, regLen, mir->offset,
                       pcrLabel);
        freeTemp(cUnit, regLen);
    } else {
        /* regPtr -> array data */
        opRegImm(cUnit, kOpAdd, regPtr, dataOffset);
    }
    /* at this point, regPtr points to array, 2 live temps */
#if !defined(WITH_SELF_VERIFICATION)
    if ((size == kLong) || (size == kDouble)) {
        //TODO: need specific wide routine that can handle fp regs
        if (scale) {
            int rNewIndex = allocTemp(cUnit);
            opRegRegImm(cUnit, kOpLsl, rNewIndex, rlIndex.lowReg, scale);
            opRegReg(cUnit, kOpAdd, regPtr, rNewIndex);
            freeTemp(cUnit, rNewIndex);
        } else {
            opRegReg(cUnit, kOpAdd, regPtr, rlIndex.lowReg);
        }
        rlSrc = loadValueWide(cUnit, rlSrc, kAnyReg);
        storePair(cUnit, regPtr, rlSrc.lowReg, rlSrc.highReg);
        freeTemp(cUnit, regPtr);
    } else {
        rlSrc = loadValue(cUnit, rlSrc, kAnyReg);
        storeBaseIndexed(cUnit, regPtr, rlIndex.lowReg, rlSrc.lowReg,
                         scale, size);
    }
#else
    //TODO: probably want to move this into storeBaseIndexed
    void *funct = NULL;
    switch(size) {
        case kLong:
        case kDouble:
            funct = (void*) &selfVerificationStoreDoubleword;
            break;
        case kWord:
            funct = (void*) &selfVerificationStore;
            break;
        case kSignedHalf:
        case kUnsignedHalf:
            funct = (void*) &selfVerificationStoreHalfword;
            break;
        case kSignedByte:
        case kUnsignedByte:
            funct = (void*) &selfVerificationStoreByte;
            break;
        default:
            assert(0);
            dvmAbort();
    }

    if (scale) {
        int regTmpIndex = allocTemp(cUnit);
        // 3 live temps
        opRegRegImm(cUnit, kOpLsl, regTmpIndex, rlIndex.lowReg, scale);
        opRegReg(cUnit, kOpAdd, regPtr, regTmpIndex);
        freeTemp(cUnit, regTmpIndex);
    } else {
        opRegReg(cUnit, kOpAdd, regPtr, rlIndex.lowReg);
    }
    /* Combine address and index */
    if ((size == kLong) || (size == kDouble)) {
        rlSrc = loadValueWide(cUnit, rlSrc, kAnyReg);
    } else {
        rlSrc = loadValue(cUnit, rlSrc, kAnyReg);
    }

    int regMap = rlSrc.highReg << 16 | rlSrc.lowReg << 8 | regPtr;
    selfVerificationMemOpWrapper(cUnit, regMap, funct);

#endif
}

static bool handleShiftOpLong(CompilationUnit *cUnit, MIR *mir,
                              RegLocation rlDest, RegLocation rlSrc1,
                              RegLocation rlShift)
{
    /*
     * Don't mess with the regsiters here as there is a particular calling
     * convention to the out-of-line handler.
     */
    RegLocation rlResult;

    loadValueDirectWideFixed(cUnit, rlSrc1, r0, r1);
    loadValueDirect(cUnit, rlShift, r2);
    switch( mir->dalvikInsn.opCode) {
        case OP_SHL_LONG:
        case OP_SHL_LONG_2ADDR:
            genDispatchToHandler(cUnit, TEMPLATE_SHL_LONG);
            break;
        case OP_SHR_LONG:
        case OP_SHR_LONG_2ADDR:
            genDispatchToHandler(cUnit, TEMPLATE_SHR_LONG);
            break;
        case OP_USHR_LONG:
        case OP_USHR_LONG_2ADDR:
            genDispatchToHandler(cUnit, TEMPLATE_USHR_LONG);
            break;
        default:
            return true;
    }
    rlResult = getReturnLocWide(cUnit);
    storeValueWide(cUnit, rlDest, rlResult);
    return false;
}
bool handleArithOpFloatPortable(CompilationUnit *cUnit, MIR *mir,
                                RegLocation rlDest, RegLocation rlSrc1,
                                RegLocation rlSrc2)
{
    RegLocation rlResult;
    void* funct;

    /* TODO: use a proper include file to define these */
    float __aeabi_fadd(float a, float b);
    float __aeabi_fsub(float a, float b);
    float __aeabi_fdiv(float a, float b);
    float __aeabi_fmul(float a, float b);
    float fmodf(float a, float b);

    switch (mir->dalvikInsn.opCode) {
        case OP_ADD_FLOAT_2ADDR:
        case OP_ADD_FLOAT:
            funct = (void*) __aeabi_fadd;
            break;
        case OP_SUB_FLOAT_2ADDR:
        case OP_SUB_FLOAT:
            funct = (void*) __aeabi_fsub;
            break;
        case OP_DIV_FLOAT_2ADDR:
        case OP_DIV_FLOAT:
            funct = (void*) __aeabi_fdiv;
            break;
        case OP_MUL_FLOAT_2ADDR:
        case OP_MUL_FLOAT:
            funct = (void*) __aeabi_fmul;
            break;
        case OP_REM_FLOAT_2ADDR:
        case OP_REM_FLOAT:
            funct = (void*) fmodf;
            break;
        case OP_NEG_FLOAT: {
            genNegFloat(cUnit, rlDest, rlSrc1);
            return false;
        }
        default:
            return true;
    }
    loadValueDirectFixed(cUnit, rlSrc1, r0);
    loadValueDirectFixed(cUnit, rlSrc2, r1);
    loadConstant(cUnit, r2, (int)funct);
    opReg(cUnit, kOpBlx, r2);
    clobberCallRegs(cUnit);
    rlResult = getReturnLoc(cUnit);
    storeValue(cUnit, rlDest, rlResult);
    return false;
}

bool handleArithOpDoublePortable(CompilationUnit *cUnit, MIR *mir,
                                 RegLocation rlDest, RegLocation rlSrc1,
                                 RegLocation rlSrc2)
{
    RegLocation rlResult;
    void* funct;

    /* TODO: use a proper include file to define these */
    double __aeabi_dadd(double a, double b);
    double __aeabi_dsub(double a, double b);
    double __aeabi_ddiv(double a, double b);
    double __aeabi_dmul(double a, double b);
    double fmod(double a, double b);

    switch (mir->dalvikInsn.opCode) {
        case OP_ADD_DOUBLE_2ADDR:
        case OP_ADD_DOUBLE:
            funct = (void*) __aeabi_dadd;
            break;
        case OP_SUB_DOUBLE_2ADDR:
        case OP_SUB_DOUBLE:
            funct = (void*) __aeabi_dsub;
            break;
        case OP_DIV_DOUBLE_2ADDR:
        case OP_DIV_DOUBLE:
            funct = (void*) __aeabi_ddiv;
            break;
        case OP_MUL_DOUBLE_2ADDR:
        case OP_MUL_DOUBLE:
            funct = (void*) __aeabi_dmul;
            break;
        case OP_REM_DOUBLE_2ADDR:
        case OP_REM_DOUBLE:
            funct = (void*) fmod;
            break;
        case OP_NEG_DOUBLE: {
            genNegDouble(cUnit, rlDest, rlSrc1);
            return false;
        }
        default:
            return true;
    }
    loadConstant(cUnit, rlr, (int)funct);
    loadValueDirectWideFixed(cUnit, rlSrc1, r0, r1);
    loadValueDirectWideFixed(cUnit, rlSrc2, r2, r3);
    opReg(cUnit, kOpBlx, rlr);
    clobberCallRegs(cUnit);
    rlResult = getReturnLocWide(cUnit);
    storeValueWide(cUnit, rlDest, rlResult);
    return false;
}

static bool handleArithOpLong(CompilationUnit *cUnit, MIR *mir,
                              RegLocation rlDest, RegLocation rlSrc1,
                              RegLocation rlSrc2)
{
    RegLocation rlResult;
    OpKind firstOp = kOpBkpt;
    OpKind secondOp = kOpBkpt;
    bool callOut = false;
    void *callTgt;
    int retReg = r0;
    /* TODO - find proper .h file to declare these */
    long long __aeabi_ldivmod(long long op1, long long op2);

    switch (mir->dalvikInsn.opCode) {
        case OP_NOT_LONG:
            rlSrc2 = loadValueWide(cUnit, rlSrc2, kCoreReg);
            rlResult = evalLoc(cUnit, rlDest, kCoreReg, true);
            opRegReg(cUnit, kOpMvn, rlResult.lowReg, rlSrc2.lowReg);
            opRegReg(cUnit, kOpMvn, rlResult.highReg, rlSrc2.highReg);
            storeValueWide(cUnit, rlDest, rlResult);
            return false;
            break;
        case OP_ADD_LONG:
        case OP_ADD_LONG_2ADDR:
            firstOp = kOpAdd;
            secondOp = kOpAdc;
            break;
        case OP_SUB_LONG:
        case OP_SUB_LONG_2ADDR:
            firstOp = kOpSub;
            secondOp = kOpSbc;
            break;
        case OP_MUL_LONG:
        case OP_MUL_LONG_2ADDR:
            genMulLong(cUnit, rlDest, rlSrc1, rlSrc2);
            return false;
        case OP_DIV_LONG:
        case OP_DIV_LONG_2ADDR:
            callOut = true;
            retReg = r0;
            callTgt = (void*)__aeabi_ldivmod;
            break;
        /* NOTE - result is in r2/r3 instead of r0/r1 */
        case OP_REM_LONG:
        case OP_REM_LONG_2ADDR:
            callOut = true;
            callTgt = (void*)__aeabi_ldivmod;
            retReg = r2;
            break;
        case OP_AND_LONG_2ADDR:
        case OP_AND_LONG:
            firstOp = kOpAnd;
            secondOp = kOpAnd;
            break;
        case OP_OR_LONG:
        case OP_OR_LONG_2ADDR:
            firstOp = kOpOr;
            secondOp = kOpOr;
            break;
        case OP_XOR_LONG:
        case OP_XOR_LONG_2ADDR:
            firstOp = kOpXor;
            secondOp = kOpXor;
            break;
        case OP_NEG_LONG: {
            rlSrc2 = loadValueWide(cUnit, rlSrc2, kCoreReg);
            rlResult = evalLoc(cUnit, rlDest, kCoreReg, true);
            loadConstantValue(cUnit, rlResult.highReg, 0);
            opRegRegReg(cUnit, kOpSub, rlResult.lowReg,
                        rlResult.highReg, rlSrc2.lowReg);
            opRegReg(cUnit, kOpSbc, rlResult.highReg, rlSrc2.highReg);
            storeValueWide(cUnit, rlDest, rlResult);
            return false;
        }
        default:
            LOGE("Invalid long arith op");
            dvmAbort();
    }
    if (!callOut) {
        genLong3Addr(cUnit, firstOp, secondOp, rlDest, rlSrc1, rlSrc2);
    } else {
        // Adjust return regs in to handle case of rem returning r2/r3
        loadValueDirectWideFixed(cUnit, rlSrc1, r0, r1);
        loadConstant(cUnit, rlr, (int) callTgt);
        loadValueDirectWideFixed(cUnit, rlSrc2, r2, r3);
        opReg(cUnit, kOpBlx, rlr);
        clobberCallRegs(cUnit);
        if (retReg == r0)
            rlResult = getReturnLocWide(cUnit);
        else
            rlResult = getReturnLocWideAlt(cUnit);
        storeValueWide(cUnit, rlDest, rlResult);
    }
    return false;
}

static bool handleArithOpInt(CompilationUnit *cUnit, MIR *mir,
                             RegLocation rlDest, RegLocation rlSrc1,
                             RegLocation rlSrc2)
{
    OpKind op = kOpBkpt;
    bool callOut = false;
    bool checkZero = false;
    bool unary = false;
    int retReg = r0;
    void *callTgt;
    RegLocation rlResult;

    /* TODO - find proper .h file to declare these */
    int __aeabi_idivmod(int op1, int op2);
    int __aeabi_idiv(int op1, int op2);

    switch (mir->dalvikInsn.opCode) {
        case OP_NEG_INT:
            op = kOpNeg;
            unary = true;
            break;
        case OP_NOT_INT:
            op = kOpMvn;
            unary = true;
            break;
        case OP_ADD_INT:
        case OP_ADD_INT_2ADDR:
            op = kOpAdd;
            break;
        case OP_SUB_INT:
        case OP_SUB_INT_2ADDR:
            op = kOpSub;
            break;
        case OP_MUL_INT:
        case OP_MUL_INT_2ADDR:
            op = kOpMul;
            break;
        case OP_DIV_INT:
        case OP_DIV_INT_2ADDR:
            callOut = true;
            checkZero = true;
            callTgt = __aeabi_idiv;
            retReg = r0;
            break;
        /* NOTE: returns in r1 */
        case OP_REM_INT:
        case OP_REM_INT_2ADDR:
            callOut = true;
            checkZero = true;
            callTgt = __aeabi_idivmod;
            retReg = r1;
            break;
        case OP_AND_INT:
        case OP_AND_INT_2ADDR:
            op = kOpAnd;
            break;
        case OP_OR_INT:
        case OP_OR_INT_2ADDR:
            op = kOpOr;
            break;
        case OP_XOR_INT:
        case OP_XOR_INT_2ADDR:
            op = kOpXor;
            break;
        case OP_SHL_INT:
        case OP_SHL_INT_2ADDR:
            op = kOpLsl;
            break;
        case OP_SHR_INT:
        case OP_SHR_INT_2ADDR:
            op = kOpAsr;
            break;
        case OP_USHR_INT:
        case OP_USHR_INT_2ADDR:
            op = kOpLsr;
            break;
        default:
            LOGE("Invalid word arith op: 0x%x(%d)",
                 mir->dalvikInsn.opCode, mir->dalvikInsn.opCode);
            dvmAbort();
    }
    if (!callOut) {
        rlSrc1 = loadValue(cUnit, rlSrc1, kCoreReg);
        if (unary) {
            rlResult = evalLoc(cUnit, rlDest, kCoreReg, true);
            opRegReg(cUnit, op, rlResult.lowReg,
                     rlSrc1.lowReg);
        } else {
            rlSrc2 = loadValue(cUnit, rlSrc2, kCoreReg);
            rlResult = evalLoc(cUnit, rlDest, kCoreReg, true);
            opRegRegReg(cUnit, op, rlResult.lowReg,
                        rlSrc1.lowReg, rlSrc2.lowReg);
        }
        storeValue(cUnit, rlDest, rlResult);
    } else {
        RegLocation rlResult;
        loadValueDirectFixed(cUnit, rlSrc2, r1);
        loadConstant(cUnit, r2, (int) callTgt);
        loadValueDirectFixed(cUnit, rlSrc1, r0);
        if (checkZero) {
            genNullCheck(cUnit, rlSrc2.sRegLow, r1, mir->offset, NULL);
        }
        opReg(cUnit, kOpBlx, r2);
        clobberCallRegs(cUnit);
        if (retReg == r0)
            rlResult = getReturnLoc(cUnit);
        else
            rlResult = getReturnLocAlt(cUnit);
        storeValue(cUnit, rlDest, rlResult);
    }
    return false;
}

static bool handleArithOp(CompilationUnit *cUnit, MIR *mir)
{
    OpCode opCode = mir->dalvikInsn.opCode;
    RegLocation rlDest;
    RegLocation rlSrc1;
    RegLocation rlSrc2;
    /* Deduce sizes of operands */
    if (mir->ssaRep->numUses == 2) {
        rlSrc1 = getSrcLoc(cUnit, mir, 0);
        rlSrc2 = getSrcLoc(cUnit, mir, 1);
    } else if (mir->ssaRep->numUses == 3) {
        rlSrc1 = getSrcLocWide(cUnit, mir, 0, 1);
        rlSrc2 = getSrcLoc(cUnit, mir, 2);
    } else {
        rlSrc1 = getSrcLocWide(cUnit, mir, 0, 1);
        rlSrc2 = getSrcLocWide(cUnit, mir, 2, 3);
        assert(mir->ssaRep->numUses == 4);
    }
    if (mir->ssaRep->numDefs == 1) {
        rlDest = getDestLoc(cUnit, mir, 0);
    } else {
        assert(mir->ssaRep->numDefs == 2);
        rlDest = getDestLocWide(cUnit, mir, 0, 1);
    }

    if ((opCode >= OP_ADD_LONG_2ADDR) && (opCode <= OP_XOR_LONG_2ADDR)) {
        return handleArithOpLong(cUnit,mir, rlDest, rlSrc1, rlSrc2);
    }
    if ((opCode >= OP_ADD_LONG) && (opCode <= OP_XOR_LONG)) {
        return handleArithOpLong(cUnit,mir, rlDest, rlSrc1, rlSrc2);
    }
    if ((opCode >= OP_SHL_LONG_2ADDR) && (opCode <= OP_USHR_LONG_2ADDR)) {
        return handleShiftOpLong(cUnit,mir, rlDest, rlSrc1, rlSrc2);
    }
    if ((opCode >= OP_SHL_LONG) && (opCode <= OP_USHR_LONG)) {
        return handleShiftOpLong(cUnit,mir, rlDest, rlSrc1, rlSrc2);
    }
    if ((opCode >= OP_ADD_INT_2ADDR) && (opCode <= OP_USHR_INT_2ADDR)) {
        return handleArithOpInt(cUnit,mir, rlDest, rlSrc1, rlSrc2);
    }
    if ((opCode >= OP_ADD_INT) && (opCode <= OP_USHR_INT)) {
        return handleArithOpInt(cUnit,mir, rlDest, rlSrc1, rlSrc2);
    }
    if ((opCode >= OP_ADD_FLOAT_2ADDR) && (opCode <= OP_REM_FLOAT_2ADDR)) {
        return handleArithOpFloat(cUnit,mir, rlDest, rlSrc1, rlSrc2);
    }
    if ((opCode >= OP_ADD_FLOAT) && (opCode <= OP_REM_FLOAT)) {
        return handleArithOpFloat(cUnit, mir, rlDest, rlSrc1, rlSrc2);
    }
    if ((opCode >= OP_ADD_DOUBLE_2ADDR) && (opCode <= OP_REM_DOUBLE_2ADDR)) {
        return handleArithOpDouble(cUnit,mir, rlDest, rlSrc1, rlSrc2);
    }
    if ((opCode >= OP_ADD_DOUBLE) && (opCode <= OP_REM_DOUBLE)) {
        return handleArithOpDouble(cUnit,mir, rlDest, rlSrc1, rlSrc2);
    }
    return true;
}

/* Generate conditional branch instructions */
static ArmLIR *genConditionalBranch(CompilationUnit *cUnit,
                                    ArmConditionCode cond,
                                    ArmLIR *target)
{
    ArmLIR *branch = opCondBranch(cUnit, cond);
    branch->generic.target = (LIR *) target;
    return branch;
}

/* Generate unconditional branch instructions */
static ArmLIR *genUnconditionalBranch(CompilationUnit *cUnit, ArmLIR *target)
{
    ArmLIR *branch = opNone(cUnit, kOpUncondBr);
    branch->generic.target = (LIR *) target;
    return branch;
}

/*
 * Generate an kArmPseudoBarrier marker to indicate the boundary of special
 * blocks.
 */
static void genBarrier(CompilationUnit *cUnit)
{
    ArmLIR *barrier = newLIR0(cUnit, kArmPseudoBarrier);
    /* Mark all resources as being clobbered */
    barrier->defMask = -1;
}

/* Perform the actual operation for OP_RETURN_* */
static void genReturnCommon(CompilationUnit *cUnit, MIR *mir)
{
    genDispatchToHandler(cUnit, TEMPLATE_RETURN);
#if defined(INVOKE_STATS)
    gDvmJit.returnOp++;
#endif
    int dPC = (int) (cUnit->method->insns + mir->offset);
    /* Insert branch, but defer setting of target */
    ArmLIR *branch = genUnconditionalBranch(cUnit, NULL);
    /* Set up the place holder to reconstruct this Dalvik PC */
    ArmLIR *pcrLabel = dvmCompilerNew(sizeof(ArmLIR), true);
    pcrLabel->opCode = ARM_PSEUDO_kPCReconstruction_CELL;
    pcrLabel->operands[0] = dPC;
    pcrLabel->operands[1] = mir->offset;
    /* Insert the place holder to the growable list */
    dvmInsertGrowableList(&cUnit->pcReconstructionList, pcrLabel);
    /* Branch to the PC reconstruction code */
    branch->generic.target = (LIR *) pcrLabel;
}

static bool genConversionCall(CompilationUnit *cUnit, MIR *mir, void *funct,
                                     int srcSize, int tgtSize)
{
    /*
     * Don't optimize the register usage since it calls out to template
     * functions
     */
    RegLocation rlSrc;
    RegLocation rlDest;
    if (srcSize == 1) {
        rlSrc = getSrcLoc(cUnit, mir, 0);
        loadValueDirectFixed(cUnit, rlSrc, r0);
    } else {
        rlSrc = getSrcLocWide(cUnit, mir, 0, 1);
        loadValueDirectWideFixed(cUnit, rlSrc, r0, r1);
    }
    loadConstant(cUnit, r2, (int)funct);
    opReg(cUnit, kOpBlx, r2);
    clobberCallRegs(cUnit);
    if (tgtSize == 1) {
        RegLocation rlResult;
        rlDest = getDestLoc(cUnit, mir, 0);
        rlResult = getReturnLoc(cUnit);
        storeValue(cUnit, rlDest, rlResult);
    } else {
        RegLocation rlResult;
        rlDest = getDestLocWide(cUnit, mir, 0, 1);
        rlResult = getReturnLocWide(cUnit);
        storeValueWide(cUnit, rlDest, rlResult);
    }
    return false;
}

static void genProcessArgsNoRange(CompilationUnit *cUnit, MIR *mir,
                                  DecodedInstruction *dInsn,
                                  ArmLIR **pcrLabel)
{
    unsigned int i;
    unsigned int regMask = 0;
    RegLocation rlArg;
    int numDone = 0;

    /*
     * Load arguments to r0..r4.  Note that these registers may contain
     * live values, so we clobber them immediately after loading to prevent
     * them from being used as sources for subsequent loads.
     */
    lockAllTemps(cUnit);
    for (i = 0; i < dInsn->vA; i++) {
        regMask |= 1 << i;
        rlArg = getSrcLoc(cUnit, mir, numDone++);
        loadValueDirectFixed(cUnit, rlArg, i);
    }
    if (regMask) {
        /* Up to 5 args are pushed on top of FP - sizeofStackSaveArea */
        opRegRegImm(cUnit, kOpSub, r7, rFP,
                    sizeof(StackSaveArea) + (dInsn->vA << 2));
        /* generate null check */
        if (pcrLabel) {
            *pcrLabel = genNullCheck(cUnit, getSrcSSAName(mir, 0), r0,
                                     mir->offset, NULL);
        }
        storeMultiple(cUnit, r7, regMask);
    }
}

static void genProcessArgsRange(CompilationUnit *cUnit, MIR *mir,
                                DecodedInstruction *dInsn,
                                ArmLIR **pcrLabel)
{
    int srcOffset = dInsn->vC << 2;
    int numArgs = dInsn->vA;
    int regMask;

    /*
     * Note: here, all promoted registers will have been flushed
     * back to the Dalvik base locations, so register usage restrictins
     * are lifted.  All parms loaded from original Dalvik register
     * region - even though some might conceivably have valid copies
     * cached in a preserved register.
     */
    lockAllTemps(cUnit);

    /*
     * r4PC     : &rFP[vC]
     * r7: &newFP[0]
     */
    opRegRegImm(cUnit, kOpAdd, r4PC, rFP, srcOffset);
    /* load [r0 .. min(numArgs,4)] */
    regMask = (1 << ((numArgs < 4) ? numArgs : 4)) - 1;
    /*
     * Protect the loadMultiple instruction from being reordered with other
     * Dalvik stack accesses.
     */
    loadMultiple(cUnit, r4PC, regMask);

    opRegRegImm(cUnit, kOpSub, r7, rFP,
                sizeof(StackSaveArea) + (numArgs << 2));
    /* generate null check */
    if (pcrLabel) {
        *pcrLabel = genNullCheck(cUnit, getSrcSSAName(mir, 0), r0,
                                 mir->offset, NULL);
    }

    /*
     * Handle remaining 4n arguments:
     * store previously loaded 4 values and load the next 4 values
     */
    if (numArgs >= 8) {
        ArmLIR *loopLabel = NULL;
        /*
         * r0 contains "this" and it will be used later, so push it to the stack
         * first. Pushing r5 (rFP) is just for stack alignment purposes.
         */
        opImm(cUnit, kOpPush, (1 << r0 | 1 << rFP));
        /* No need to generate the loop structure if numArgs <= 11 */
        if (numArgs > 11) {
            loadConstant(cUnit, 5, ((numArgs - 4) >> 2) << 2);
            loopLabel = newLIR0(cUnit, kArmPseudoTargetLabel);
            loopLabel->defMask = ENCODE_ALL;
        }
        storeMultiple(cUnit, r7, regMask);
        /*
         * Protect the loadMultiple instruction from being reordered with other
         * Dalvik stack accesses.
         */
        loadMultiple(cUnit, r4PC, regMask);
        /* No need to generate the loop structure if numArgs <= 11 */
        if (numArgs > 11) {
            opRegImm(cUnit, kOpSub, rFP, 4);
            genConditionalBranch(cUnit, kArmCondNe, loopLabel);
        }
    }

    /* Save the last batch of loaded values */
    storeMultiple(cUnit, r7, regMask);

    /* Generate the loop epilogue - don't use r0 */
    if ((numArgs > 4) && (numArgs % 4)) {
        regMask = ((1 << (numArgs & 0x3)) - 1) << 1;
        /*
         * Protect the loadMultiple instruction from being reordered with other
         * Dalvik stack accesses.
         */
        loadMultiple(cUnit, r4PC, regMask);
    }
    if (numArgs >= 8)
        opImm(cUnit, kOpPop, (1 << r0 | 1 << rFP));

    /* Save the modulo 4 arguments */
    if ((numArgs > 4) && (numArgs % 4)) {
        storeMultiple(cUnit, r7, regMask);
    }
}

/*
 * Generate code to setup the call stack then jump to the chaining cell if it
 * is not a native method.
 */
static void genInvokeSingletonCommon(CompilationUnit *cUnit, MIR *mir,
                                     BasicBlock *bb, ArmLIR *labelList,
                                     ArmLIR *pcrLabel,
                                     const Method *calleeMethod)
{
    /*
     * Note: all Dalvik register state should be flushed to
     * memory by the point, so register usage restrictions no
     * longer apply.  All temp & preserved registers may be used.
     */
    lockAllTemps(cUnit);
    ArmLIR *retChainingCell = &labelList[bb->fallThrough->id];

    /* r1 = &retChainingCell */
    lockTemp(cUnit, r1);
    ArmLIR *addrRetChain = opRegRegImm(cUnit, kOpAdd, r1, rpc, 0);
    /* r4PC = dalvikCallsite */
    loadConstant(cUnit, r4PC,
                 (int) (cUnit->method->insns + mir->offset));
    addrRetChain->generic.target = (LIR *) retChainingCell;
    /*
     * r0 = calleeMethod (loaded upon calling genInvokeSingletonCommon)
     * r1 = &ChainingCell
     * r4PC = callsiteDPC
     */
    if (dvmIsNativeMethod(calleeMethod)) {
        genDispatchToHandler(cUnit, TEMPLATE_INVOKE_METHOD_NATIVE);
#if defined(INVOKE_STATS)
        gDvmJit.invokeNative++;
#endif
    } else {
        genDispatchToHandler(cUnit, TEMPLATE_INVOKE_METHOD_CHAIN);
#if defined(INVOKE_STATS)
        gDvmJit.invokeChain++;
#endif
        /* Branch to the chaining cell */
        genUnconditionalBranch(cUnit, &labelList[bb->taken->id]);
    }
    /* Handle exceptions using the interpreter */
    genTrap(cUnit, mir->offset, pcrLabel);
}

/*
 * Generate code to check the validity of a predicted chain and take actions
 * based on the result.
 *
 * 0x426a99aa : ldr     r4, [pc, #72] --> r4 <- dalvikPC of this invoke
 * 0x426a99ac : add     r1, pc, #32   --> r1 <- &retChainingCell
 * 0x426a99ae : add     r2, pc, #40   --> r2 <- &predictedChainingCell
 * 0x426a99b0 : blx_1   0x426a918c    --+ TEMPLATE_INVOKE_METHOD_PREDICTED_CHAIN
 * 0x426a99b2 : blx_2   see above     --+
 * 0x426a99b4 : b       0x426a99d8    --> off to the predicted chain
 * 0x426a99b6 : b       0x426a99c8    --> punt to the interpreter
 * 0x426a99b8 : ldr     r0, [r7, #44] --> r0 <- this->class->vtable[methodIdx]
 * 0x426a99ba : cmp     r1, #0        --> compare r1 (rechain count) against 0
 * 0x426a99bc : bgt     0x426a99c2    --> >=0? don't rechain
 * 0x426a99be : ldr     r7, [r6, #96] --+ dvmJitToPatchPredictedChain
 * 0x426a99c0 : blx     r7            --+
 * 0x426a99c2 : add     r1, pc, #12   --> r1 <- &retChainingCell
 * 0x426a99c4 : blx_1   0x426a9098    --+ TEMPLATE_INVOKE_METHOD_NO_OPT
 * 0x426a99c6 : blx_2   see above     --+
 */
static void genInvokeVirtualCommon(CompilationUnit *cUnit, MIR *mir,
                                   int methodIndex,
                                   ArmLIR *retChainingCell,
                                   ArmLIR *predChainingCell,
                                   ArmLIR *pcrLabel)
{
    /*
     * Note: all Dalvik register state should be flushed to
     * memory by the point, so register usage restrictions no
     * longer apply.  Lock temps to prevent them from being
     * allocated by utility routines.
     */
    lockAllTemps(cUnit);

    /* "this" is already left in r0 by genProcessArgs* */

    /* r4PC = dalvikCallsite */
    loadConstant(cUnit, r4PC,
                 (int) (cUnit->method->insns + mir->offset));

    /* r1 = &retChainingCell */
    ArmLIR *addrRetChain = opRegRegImm(cUnit, kOpAdd, r1, rpc, 0);
    addrRetChain->generic.target = (LIR *) retChainingCell;

    /* r2 = &predictedChainingCell */
    ArmLIR *predictedChainingCell = opRegRegImm(cUnit, kOpAdd, r2, rpc, 0);
    predictedChainingCell->generic.target = (LIR *) predChainingCell;

    genDispatchToHandler(cUnit, TEMPLATE_INVOKE_METHOD_PREDICTED_CHAIN);

    /* return through lr - jump to the chaining cell */
    genUnconditionalBranch(cUnit, predChainingCell);

    /*
     * null-check on "this" may have been eliminated, but we still need a PC-
     * reconstruction label for stack overflow bailout.
     */
    if (pcrLabel == NULL) {
        int dPC = (int) (cUnit->method->insns + mir->offset);
        pcrLabel = dvmCompilerNew(sizeof(ArmLIR), true);
        pcrLabel->opCode = ARM_PSEUDO_kPCReconstruction_CELL;
        pcrLabel->operands[0] = dPC;
        pcrLabel->operands[1] = mir->offset;
        /* Insert the place holder to the growable list */
        dvmInsertGrowableList(&cUnit->pcReconstructionList, pcrLabel);
    }

    /* return through lr+2 - punt to the interpreter */
    genUnconditionalBranch(cUnit, pcrLabel);

    /*
     * return through lr+4 - fully resolve the callee method.
     * r1 <- count
     * r2 <- &predictedChainCell
     * r3 <- this->class
     * r4 <- dPC
     * r7 <- this->class->vtable
     */

    /* r0 <- calleeMethod */
    loadWordDisp(cUnit, r7, methodIndex * 4, r0);

    /* Check if rechain limit is reached */
    opRegImm(cUnit, kOpCmp, r1, 0);

    ArmLIR *bypassRechaining = opCondBranch(cUnit, kArmCondGt);

    loadWordDisp(cUnit, rGLUE, offsetof(InterpState,
                 jitToInterpEntries.dvmJitToPatchPredictedChain), r7);

    /*
     * r0 = calleeMethod
     * r2 = &predictedChainingCell
     * r3 = class
     *
     * &returnChainingCell has been loaded into r1 but is not needed
     * when patching the chaining cell and will be clobbered upon
     * returning so it will be reconstructed again.
     */
    opReg(cUnit, kOpBlx, r7);

    /* r1 = &retChainingCell */
    addrRetChain = opRegRegImm(cUnit, kOpAdd, r1, rpc, 0);
    addrRetChain->generic.target = (LIR *) retChainingCell;

    bypassRechaining->generic.target = (LIR *) addrRetChain;
    /*
     * r0 = calleeMethod,
     * r1 = &ChainingCell,
     * r4PC = callsiteDPC,
     */
    genDispatchToHandler(cUnit, TEMPLATE_INVOKE_METHOD_NO_OPT);
#if defined(INVOKE_STATS)
    gDvmJit.invokePredictedChain++;
#endif
    /* Handle exceptions using the interpreter */
    genTrap(cUnit, mir->offset, pcrLabel);
}

/*
 * Up calling this function, "this" is stored in r0. The actual class will be
 * chased down off r0 and the predicted one will be retrieved through
 * predictedChainingCell then a comparison is performed to see whether the
 * previously established chaining is still valid.
 *
 * The return LIR is a branch based on the comparison result. The actual branch
 * target will be setup in the caller.
 */
static ArmLIR *genCheckPredictedChain(CompilationUnit *cUnit,
                                          ArmLIR *predChainingCell,
                                          ArmLIR *retChainingCell,
                                          MIR *mir)
{
    /*
     * Note: all Dalvik register state should be flushed to
     * memory by the point, so register usage restrictions no
     * longer apply.  All temp & preserved registers may be used.
     */
    lockAllTemps(cUnit);

    /* r3 now contains this->clazz */
    loadWordDisp(cUnit, r0, offsetof(Object, clazz), r3);

    /*
     * r2 now contains predicted class. The starting offset of the
     * cached value is 4 bytes into the chaining cell.
     */
    ArmLIR *getPredictedClass =
         loadWordDisp(cUnit, rpc, offsetof(PredictedChainingCell, clazz), r2);
    getPredictedClass->generic.target = (LIR *) predChainingCell;

    /*
     * r0 now contains predicted method. The starting offset of the
     * cached value is 8 bytes into the chaining cell.
     */
    ArmLIR *getPredictedMethod =
        loadWordDisp(cUnit, rpc, offsetof(PredictedChainingCell, method), r0);
    getPredictedMethod->generic.target = (LIR *) predChainingCell;

    /* Load the stats counter to see if it is time to unchain and refresh */
    ArmLIR *getRechainingRequestCount =
        loadWordDisp(cUnit, rpc, offsetof(PredictedChainingCell, counter), r7);
    getRechainingRequestCount->generic.target =
        (LIR *) predChainingCell;

    /* r4PC = dalvikCallsite */
    loadConstant(cUnit, r4PC,
                 (int) (cUnit->method->insns + mir->offset));

    /* r1 = &retChainingCell */
    ArmLIR *addrRetChain = opRegRegImm(cUnit, kOpAdd, r1, rpc, 0);
    addrRetChain->generic.target = (LIR *) retChainingCell;

    /* Check if r2 (predicted class) == r3 (actual class) */
    opRegReg(cUnit, kOpCmp, r2, r3);

    return opCondBranch(cUnit, kArmCondEq);
}

/* Geneate a branch to go back to the interpreter */
static void genPuntToInterp(CompilationUnit *cUnit, unsigned int offset)
{
    /* r0 = dalvik pc */
    flushAllRegs(cUnit);
    loadConstant(cUnit, r0, (int) (cUnit->method->insns + offset));
    loadWordDisp(cUnit, r0, offsetof(Object, clazz), r3);
    loadWordDisp(cUnit, rGLUE, offsetof(InterpState,
                 jitToInterpEntries.dvmJitToInterpPunt), r1);
    opReg(cUnit, kOpBlx, r1);
}

/*
 * Attempt to single step one instruction using the interpreter and return
 * to the compiled code for the next Dalvik instruction
 */
static void genInterpSingleStep(CompilationUnit *cUnit, MIR *mir)
{
    int flags = dexGetInstrFlags(gDvm.instrFlags, mir->dalvikInsn.opCode);
    int flagsToCheck = kInstrCanBranch | kInstrCanSwitch | kInstrCanReturn |
                       kInstrCanThrow;

    //Ugly, but necessary.  Flush all Dalvik regs so Interp can find them
    flushAllRegs(cUnit);

    if ((mir->next == NULL) || (flags & flagsToCheck)) {
       genPuntToInterp(cUnit, mir->offset);
       return;
    }
    int entryAddr = offsetof(InterpState,
                             jitToInterpEntries.dvmJitToInterpSingleStep);
    loadWordDisp(cUnit, rGLUE, entryAddr, r2);
    /* r0 = dalvik pc */
    loadConstant(cUnit, r0, (int) (cUnit->method->insns + mir->offset));
    /* r1 = dalvik pc of following instruction */
    loadConstant(cUnit, r1, (int) (cUnit->method->insns + mir->next->offset));
    opReg(cUnit, kOpBlx, r2);
}

static void handleMonitorPortable(CompilationUnit *cUnit, MIR *mir)
{
    genExportPC(cUnit, mir);
    RegLocation rlSrc = getSrcLoc(cUnit, mir, 0);
    loadValueDirectFixed(cUnit, rlSrc, r1);
    loadWordDisp(cUnit, rGLUE, offsetof(InterpState, self), r0);
    if (mir->dalvikInsn.opCode == OP_MONITOR_ENTER) {
        loadConstant(cUnit, r2, (int)dvmLockObject);
    } else {
        loadConstant(cUnit, r2, (int)dvmUnlockObject);
    }
    genNullCheck(cUnit, rlSrc.sRegLow, r1, mir->offset, NULL);
    /* Do the call */
    opReg(cUnit, kOpBlx, r2);
    clobberCallRegs(cUnit);
}

/* Load a word at base + displacement.  Displacement must be word multiple */
static ArmLIR *loadWordDisp(CompilationUnit *cUnit, int rBase, int displacement,
                            int rDest)
{
    return loadBaseDisp(cUnit, NULL, rBase, displacement, rDest, kWord, false,
                        INVALID_SREG);
}

static ArmLIR *storeWordDisp(CompilationUnit *cUnit, int rBase,
                             int displacement, int rSrc)
{
    return storeBaseDisp(cUnit, rBase, displacement, rSrc, kWord);
}

static ArmLIR *genRegCopy(CompilationUnit *cUnit, int rDest, int rSrc)
{
    ArmLIR *res = dvmCompilerRegCopy(cUnit, rDest, rSrc);
    dvmCompilerAppendLIR(cUnit, (LIR*)res);
    return res;
}

/*
 * The following are the first-level codegen routines that analyze the format
 * of each bytecode then either dispatch special purpose codegen routines
 * or produce corresponding Thumb instructions directly.
 */

static bool handleFmt10t_Fmt20t_Fmt30t(CompilationUnit *cUnit, MIR *mir,
                                       BasicBlock *bb, ArmLIR *labelList)
{
    /* For OP_GOTO, OP_GOTO_16, and OP_GOTO_32 */
    genUnconditionalBranch(cUnit, &labelList[bb->taken->id]);
    return false;
}

static bool handleFmt10x(CompilationUnit *cUnit, MIR *mir)
{
    OpCode dalvikOpCode = mir->dalvikInsn.opCode;
    if (((dalvikOpCode >= OP_UNUSED_3E) && (dalvikOpCode <= OP_UNUSED_43)) ||
        ((dalvikOpCode >= OP_UNUSED_E3) && (dalvikOpCode <= OP_UNUSED_EC))) {
        LOGE("Codegen: got unused opcode 0x%x\n",dalvikOpCode);
        return true;
    }
    switch (dalvikOpCode) {
        case OP_RETURN_VOID:
            genReturnCommon(cUnit,mir);
            break;
        case OP_UNUSED_73:
        case OP_UNUSED_79:
        case OP_UNUSED_7A:
            LOGE("Codegen: got unused opcode 0x%x\n",dalvikOpCode);
            return true;
        case OP_NOP:
            break;
        default:
            return true;
    }
    return false;
}

static bool handleFmt11n_Fmt31i(CompilationUnit *cUnit, MIR *mir)
{
    RegLocation rlDest;
    RegLocation rlResult;
    if (mir->ssaRep->numDefs == 2) {
        rlDest = getDestLocWide(cUnit, mir, 0, 1);
    } else {
        rlDest = getDestLoc(cUnit, mir, 0);
    }

    switch (mir->dalvikInsn.opCode) {
        case OP_CONST:
        case OP_CONST_4: {
            rlResult = evalLoc(cUnit, rlDest, kAnyReg, true);
            loadConstantValue(cUnit, rlResult.lowReg, mir->dalvikInsn.vB);
            storeValue(cUnit, rlDest, rlResult);
            break;
        }
        case OP_CONST_WIDE_32: {
            //TUNING: single routine to load constant pair for support doubles
            rlResult = evalLoc(cUnit, rlDest, kCoreReg, true);
            loadConstantValue(cUnit, rlResult.lowReg, mir->dalvikInsn.vB);
            opRegRegImm(cUnit, kOpAsr, rlResult.highReg,
                        rlResult.lowReg, 31);
            storeValueWide(cUnit, rlDest, rlResult);
            break;
        }
        default:
            return true;
    }
    return false;
}

static bool handleFmt21h(CompilationUnit *cUnit, MIR *mir)
{
    RegLocation rlDest;
    RegLocation rlResult;
    if (mir->ssaRep->numDefs == 2) {
        rlDest = getDestLocWide(cUnit, mir, 0, 1);
    } else {
        rlDest = getDestLoc(cUnit, mir, 0);
    }
    rlResult = evalLoc(cUnit, rlDest, kAnyReg, true);

    switch (mir->dalvikInsn.opCode) {
        case OP_CONST_HIGH16: {
            loadConstantValue(cUnit, rlResult.lowReg, mir->dalvikInsn.vB << 16);
            storeValue(cUnit, rlDest, rlResult);
            break;
        }
        case OP_CONST_WIDE_HIGH16: {
            loadConstantValueWide(cUnit, rlResult.lowReg, rlResult.highReg,
                                  0, mir->dalvikInsn.vB << 16);
            storeValueWide(cUnit, rlDest, rlResult);
            break;
        }
        default:
            return true;
    }
    return false;
}

static bool handleFmt20bc(CompilationUnit *cUnit, MIR *mir)
{
    /* For OP_THROW_VERIFICATION_ERROR */
    genInterpSingleStep(cUnit, mir);
    return false;
}

static bool handleFmt21c_Fmt31c(CompilationUnit *cUnit, MIR *mir)
{
    RegLocation rlResult;
    RegLocation rlDest;
    RegLocation rlSrc;

    switch (mir->dalvikInsn.opCode) {
        case OP_CONST_STRING_JUMBO:
        case OP_CONST_STRING: {
            void *strPtr = (void*)
              (cUnit->method->clazz->pDvmDex->pResStrings[mir->dalvikInsn.vB]);
            assert(strPtr != NULL);
            rlDest = getDestLoc(cUnit, mir, 0);
            rlResult = evalLoc(cUnit, rlDest, kCoreReg, true);
            loadConstantValue(cUnit, rlResult.lowReg, (int) strPtr );
            storeValue(cUnit, rlDest, rlResult);
            break;
        }
        case OP_CONST_CLASS: {
            void *classPtr = (void*)
              (cUnit->method->clazz->pDvmDex->pResClasses[mir->dalvikInsn.vB]);
            assert(classPtr != NULL);
            rlDest = getDestLoc(cUnit, mir, 0);
            rlResult = evalLoc(cUnit, rlDest, kCoreReg, true);
            loadConstantValue(cUnit, rlResult.lowReg, (int) classPtr );
            storeValue(cUnit, rlDest, rlResult);
            break;
        }
        case OP_SGET_OBJECT:
        case OP_SGET_BOOLEAN:
        case OP_SGET_CHAR:
        case OP_SGET_BYTE:
        case OP_SGET_SHORT:
        case OP_SGET: {
            int valOffset = offsetof(StaticField, value);
            int tReg = allocTemp(cUnit);
            void *fieldPtr = (void*)
              (cUnit->method->clazz->pDvmDex->pResFields[mir->dalvikInsn.vB]);
            assert(fieldPtr != NULL);
            rlDest = getDestLoc(cUnit, mir, 0);
            rlResult = evalLoc(cUnit, rlDest, kAnyReg, true);
            loadConstant(cUnit, tReg,  (int) fieldPtr + valOffset);
#if !defined(WITH_SELF_VERIFICATION)
            loadWordDisp(cUnit, tReg, 0, rlResult.lowReg);
#else
            int regMap = rlResult.lowReg << 8 | tReg;
            selfVerificationMemOpWrapper(cUnit, regMap, &selfVerificationLoad);

#endif
            storeValue(cUnit, rlDest, rlResult);
            break;
        }
        case OP_SGET_WIDE: {
            int valOffset = offsetof(StaticField, value);
            void *fieldPtr = (void*)
              (cUnit->method->clazz->pDvmDex->pResFields[mir->dalvikInsn.vB]);
            int tReg = allocTemp(cUnit);
            assert(fieldPtr != NULL);
            rlDest = getDestLocWide(cUnit, mir, 0, 1);
            rlResult = evalLoc(cUnit, rlDest, kAnyReg, true);
            loadConstant(cUnit, tReg,  (int) fieldPtr + valOffset);
#if !defined(WITH_SELF_VERIFICATION)
            loadPair(cUnit, tReg, rlResult.lowReg, rlResult.highReg);
#else
            int regMap = rlResult.highReg << 16 |
                         rlResult.lowReg << 8 | tReg;
            selfVerificationMemOpWrapper(cUnit, regMap,
                &selfVerificationLoadDoubleword);

#endif
            storeValueWide(cUnit, rlDest, rlResult);
            break;
        }
        case OP_SPUT_OBJECT:
        case OP_SPUT_BOOLEAN:
        case OP_SPUT_CHAR:
        case OP_SPUT_BYTE:
        case OP_SPUT_SHORT:
        case OP_SPUT: {
            int valOffset = offsetof(StaticField, value);
            int tReg = allocTemp(cUnit);
            void *fieldPtr = (void*)
              (cUnit->method->clazz->pDvmDex->pResFields[mir->dalvikInsn.vB]);

            assert(fieldPtr != NULL);
            rlSrc = getSrcLoc(cUnit, mir, 0);
            rlSrc = loadValue(cUnit, rlSrc, kAnyReg);
            loadConstant(cUnit, tReg,  (int) fieldPtr + valOffset);
#if !defined(WITH_SELF_VERIFICATION)
            storeWordDisp(cUnit, tReg, 0 ,rlSrc.lowReg);
#else
            int regMap = rlSrc.lowReg << 8 | tReg;
            selfVerificationMemOpWrapper(cUnit, regMap, &selfVerificationStore);
#endif
            break;
        }
        case OP_SPUT_WIDE: {
            int tReg = allocTemp(cUnit);
            int valOffset = offsetof(StaticField, value);
            void *fieldPtr = (void*)
              (cUnit->method->clazz->pDvmDex->pResFields[mir->dalvikInsn.vB]);

            assert(fieldPtr != NULL);
            rlSrc = getSrcLocWide(cUnit, mir, 0, 1);
            rlSrc = loadValueWide(cUnit, rlSrc, kAnyReg);
            loadConstant(cUnit, tReg,  (int) fieldPtr + valOffset);
#if !defined(WITH_SELF_VERIFICATION)
            storePair(cUnit, tReg, rlSrc.lowReg, rlSrc.highReg);
#else
            int regMap = rlSrc.highReg << 16 | rlSrc.lowReg << 8 | tReg;
            selfVerificationMemOpWrapper(cUnit, regMap,
                &selfVerificationStoreDoubleword);
#endif
            break;
        }
        case OP_NEW_INSTANCE: {
            /*
             * Obey the calling convention and don't mess with the register
             * usage.
             */
            ClassObject *classPtr = (void*)
              (cUnit->method->clazz->pDvmDex->pResClasses[mir->dalvikInsn.vB]);
            assert(classPtr != NULL);
            assert(classPtr->status & CLASS_INITIALIZED);
            /*
             * If it is going to throw, it should not make to the trace to begin
             * with.  However, Alloc might throw, so we need to genExportPC()
             */
            assert((classPtr->accessFlags & (ACC_INTERFACE|ACC_ABSTRACT)) == 0);
            genExportPC(cUnit, mir);
            loadConstant(cUnit, r2, (int)dvmAllocObject);
            loadConstant(cUnit, r0, (int) classPtr);
            loadConstant(cUnit, r1, ALLOC_DONT_TRACK);
            opReg(cUnit, kOpBlx, r2);
            clobberCallRegs(cUnit);
            /* generate a branch over if allocation is successful */
            opRegImm(cUnit, kOpCmp, r0, 0); /* NULL? */
            ArmLIR *branchOver = opCondBranch(cUnit, kArmCondNe);
            /*
             * OOM exception needs to be thrown here and cannot re-execute
             */
            loadConstant(cUnit, r0,
                         (int) (cUnit->method->insns + mir->offset));
            genDispatchToHandler(cUnit, TEMPLATE_THROW_EXCEPTION_COMMON);
            /* noreturn */

            ArmLIR *target = newLIR0(cUnit, kArmPseudoTargetLabel);
            target->defMask = ENCODE_ALL;
            branchOver->generic.target = (LIR *) target;
            rlDest = getDestLoc(cUnit, mir, 0);
            rlResult = getReturnLoc(cUnit);
            storeValue(cUnit, rlDest, rlResult);
            break;
        }
        case OP_CHECK_CAST: {
            /*
             * Obey the calling convention and don't mess with the register
             * usage.
             */
            ClassObject *classPtr =
              (cUnit->method->clazz->pDvmDex->pResClasses[mir->dalvikInsn.vB]);
            loadConstant(cUnit, r1, (int) classPtr );
            rlSrc = getSrcLoc(cUnit, mir, 0);
            rlSrc = loadValue(cUnit, rlSrc, kCoreReg);
            opRegImm(cUnit, kOpCmp, rlSrc.lowReg, 0);   /* Null? */
            ArmLIR *branch1 = opCondBranch(cUnit, kArmCondEq);
            /*
             *  rlSrc.lowReg now contains object->clazz.  Note that
             *  it could have been allocated r0, but we're okay so long
             *  as we don't do anything desctructive until r0 is loaded
             *  with clazz.
             */
            /* r0 now contains object->clazz */
            loadWordDisp(cUnit, rlSrc.lowReg, offsetof(Object, clazz), r0);
            loadConstant(cUnit, r2, (int)dvmInstanceofNonTrivial);
            opRegReg(cUnit, kOpCmp, r0, r1);
            ArmLIR *branch2 = opCondBranch(cUnit, kArmCondEq);
            opReg(cUnit, kOpBlx, r2);
            clobberCallRegs(cUnit);
            /*
             * If null, check cast failed - punt to the interpreter.  Because
             * interpreter will be the one throwing, we don't need to
             * genExportPC() here.
             */
            genZeroCheck(cUnit, r0, mir->offset, NULL);
            /* check cast passed - branch target here */
            ArmLIR *target = newLIR0(cUnit, kArmPseudoTargetLabel);
            target->defMask = ENCODE_ALL;
            branch1->generic.target = (LIR *)target;
            branch2->generic.target = (LIR *)target;
            break;
        }
        default:
            return true;
    }
    return false;
}

static bool handleFmt11x(CompilationUnit *cUnit, MIR *mir)
{
    OpCode dalvikOpCode = mir->dalvikInsn.opCode;
    RegLocation rlResult;
    switch (dalvikOpCode) {
        case OP_MOVE_EXCEPTION: {
            int offset = offsetof(InterpState, self);
            int exOffset = offsetof(Thread, exception);
            int selfReg = allocTemp(cUnit);
            RegLocation rlDest = getDestLoc(cUnit, mir, 0);
            rlResult = evalLoc(cUnit, rlDest, kCoreReg, true);
            loadWordDisp(cUnit, rGLUE, offset, selfReg);
            loadWordDisp(cUnit, selfReg, exOffset, rlResult.lowReg);
            storeValue(cUnit, rlDest, rlResult);
           break;
        }
        case OP_MOVE_RESULT:
        case OP_MOVE_RESULT_OBJECT: {
            RegLocation rlDest = getDestLoc(cUnit, mir, 0);
            RegLocation rlSrc = LOC_DALVIK_RETURN_VAL;
            rlSrc.fp = rlDest.fp;
            storeValue(cUnit, rlDest, rlSrc);
            break;
        }
        case OP_MOVE_RESULT_WIDE: {
            RegLocation rlDest = getDestLocWide(cUnit, mir, 0, 1);
            RegLocation rlSrc = LOC_DALVIK_RETURN_VAL_WIDE;
            rlSrc.fp = rlDest.fp;
            storeValueWide(cUnit, rlDest, rlSrc);
            break;
        }
        case OP_RETURN_WIDE: {
            RegLocation rlSrc = getSrcLocWide(cUnit, mir, 0, 1);
            RegLocation rlDest = LOC_DALVIK_RETURN_VAL_WIDE;
            rlDest.fp = rlSrc.fp;
            storeValueWide(cUnit, rlDest, rlSrc);
            genReturnCommon(cUnit,mir);
            break;
        }
        case OP_RETURN:
        case OP_RETURN_OBJECT: {
            RegLocation rlSrc = getSrcLoc(cUnit, mir, 0);
            RegLocation rlDest = LOC_DALVIK_RETURN_VAL;
            rlDest.fp = rlSrc.fp;
            storeValue(cUnit, rlDest, rlSrc);
            genReturnCommon(cUnit,mir);
            break;
        }
        case OP_MONITOR_EXIT:
        case OP_MONITOR_ENTER:
#if defined(WITH_DEADLOCK_PREDICTION) || defined(WITH_MONITOR_TRACKING)
            handleMonitorPortable(cUnit, mir);
#else
            handleMonitor(cUnit, mir);
#endif
            break;
        case OP_THROW: {
            genInterpSingleStep(cUnit, mir);
            break;
        }
        default:
            return true;
    }
    return false;
}

static bool handleConversionPortable(CompilationUnit *cUnit, MIR *mir)
{
    OpCode opCode = mir->dalvikInsn.opCode;

    float  __aeabi_i2f(  int op1 );
    int    __aeabi_f2iz( float op1 );
    float  __aeabi_d2f(  double op1 );
    double __aeabi_f2d(  float op1 );
    double __aeabi_i2d(  int op1 );
    int    __aeabi_d2iz( double op1 );
    float  __aeabi_l2f(  long op1 );
    double __aeabi_l2d(  long op1 );
    s8 dvmJitf2l( float op1 );
    s8 dvmJitd2l( double op1 );

    switch (opCode) {
        case OP_INT_TO_FLOAT:
            return genConversionCall(cUnit, mir, (void*)__aeabi_i2f, 1, 1);
        case OP_FLOAT_TO_INT:
            return genConversionCall(cUnit, mir, (void*)__aeabi_f2iz, 1, 1);
        case OP_DOUBLE_TO_FLOAT:
            return genConversionCall(cUnit, mir, (void*)__aeabi_d2f, 2, 1);
        case OP_FLOAT_TO_DOUBLE:
            return genConversionCall(cUnit, mir, (void*)__aeabi_f2d, 1, 2);
        case OP_INT_TO_DOUBLE:
            return genConversionCall(cUnit, mir, (void*)__aeabi_i2d, 1, 2);
        case OP_DOUBLE_TO_INT:
            return genConversionCall(cUnit, mir, (void*)__aeabi_d2iz, 2, 1);
        case OP_FLOAT_TO_LONG:
            return genConversionCall(cUnit, mir, (void*)dvmJitf2l, 1, 2);
        case OP_LONG_TO_FLOAT:
            return genConversionCall(cUnit, mir, (void*)__aeabi_l2f, 2, 1);
        case OP_DOUBLE_TO_LONG:
            return genConversionCall(cUnit, mir, (void*)dvmJitd2l, 2, 2);
        case OP_LONG_TO_DOUBLE:
            return genConversionCall(cUnit, mir, (void*)__aeabi_l2d, 2, 2);
        default:
            return true;
    }
    return false;
}

static bool handleFmt12x(CompilationUnit *cUnit, MIR *mir)
{
    OpCode opCode = mir->dalvikInsn.opCode;
    RegLocation rlDest;
    RegLocation rlSrc;
    RegLocation rlResult;

    if ( (opCode >= OP_ADD_INT_2ADDR) && (opCode <= OP_REM_DOUBLE_2ADDR)) {
        return handleArithOp( cUnit, mir );
    }

    if (mir->ssaRep->numUses == 2)
        rlSrc = getSrcLocWide(cUnit, mir, 0, 1);
    else
        rlSrc = getSrcLoc(cUnit, mir, 0);
    if (mir->ssaRep->numDefs == 2)
        rlDest = getDestLocWide(cUnit, mir, 0, 1);
    else
        rlDest = getDestLoc(cUnit, mir, 0);

    switch (opCode) {
        case OP_DOUBLE_TO_INT:
        case OP_INT_TO_FLOAT:
        case OP_FLOAT_TO_INT:
        case OP_DOUBLE_TO_FLOAT:
        case OP_FLOAT_TO_DOUBLE:
        case OP_INT_TO_DOUBLE:
        case OP_FLOAT_TO_LONG:
        case OP_LONG_TO_FLOAT:
        case OP_DOUBLE_TO_LONG:
        case OP_LONG_TO_DOUBLE:
            return handleConversion(cUnit, mir);
        case OP_NEG_INT:
        case OP_NOT_INT:
            return handleArithOpInt(cUnit, mir, rlDest, rlSrc, rlSrc);
        case OP_NEG_LONG:
        case OP_NOT_LONG:
            return handleArithOpLong(cUnit, mir, rlDest, rlSrc, rlSrc);
        case OP_NEG_FLOAT:
            return handleArithOpFloat(cUnit, mir, rlDest, rlSrc, rlSrc);
        case OP_NEG_DOUBLE:
            return handleArithOpDouble(cUnit, mir, rlDest, rlSrc, rlSrc);
        case OP_MOVE_WIDE:
            storeValueWide(cUnit, rlDest, rlSrc);
            break;
        case OP_INT_TO_LONG:
            rlSrc = updateLoc(cUnit, rlSrc);
            rlResult = evalLoc(cUnit, rlDest, kCoreReg, true);
            if (rlSrc.location == kLocPhysReg) {
                genRegCopy(cUnit, rlResult.lowReg, rlSrc.lowReg);
            } else {
                loadValueDirect(cUnit, rlSrc, rlResult.lowReg);
            }
            opRegRegImm(cUnit, kOpAsr, rlResult.highReg,
                        rlResult.lowReg, 31);
            storeValueWide(cUnit, rlDest, rlResult);
            break;
        case OP_LONG_TO_INT:
            rlSrc = updateLocWide(cUnit, rlSrc);
            rlSrc = wideToNarrowLoc(cUnit, rlSrc);
            // Intentional fallthrough
        case OP_MOVE:
        case OP_MOVE_OBJECT:
            storeValue(cUnit, rlDest, rlSrc);
            break;
        case OP_INT_TO_BYTE:
            rlSrc = loadValue(cUnit, rlSrc, kCoreReg);
            rlResult = evalLoc(cUnit, rlDest, kCoreReg, true);
            opRegReg(cUnit, kOp2Byte, rlResult.lowReg, rlSrc.lowReg);
            storeValue(cUnit, rlDest, rlResult);
            break;
        case OP_INT_TO_SHORT:
            rlSrc = loadValue(cUnit, rlSrc, kCoreReg);
            rlResult = evalLoc(cUnit, rlDest, kCoreReg, true);
            opRegReg(cUnit, kOp2Short, rlResult.lowReg, rlSrc.lowReg);
            storeValue(cUnit, rlDest, rlResult);
            break;
        case OP_INT_TO_CHAR:
            rlSrc = loadValue(cUnit, rlSrc, kCoreReg);
            rlResult = evalLoc(cUnit, rlDest, kCoreReg, true);
            opRegReg(cUnit, kOp2Char, rlResult.lowReg, rlSrc.lowReg);
            storeValue(cUnit, rlDest, rlResult);
            break;
        case OP_ARRAY_LENGTH: {
            int lenOffset = offsetof(ArrayObject, length);
            rlSrc = loadValue(cUnit, rlSrc, kCoreReg);
            genNullCheck(cUnit, rlSrc.sRegLow, rlSrc.lowReg,
                         mir->offset, NULL);
            rlResult = evalLoc(cUnit, rlDest, kCoreReg, true);
            loadWordDisp(cUnit, rlSrc.lowReg, lenOffset,
                         rlResult.lowReg);
            storeValue(cUnit, rlDest, rlResult);
            break;
        }
        default:
            return true;
    }
    return false;
}

static bool handleFmt21s(CompilationUnit *cUnit, MIR *mir)
{
    OpCode dalvikOpCode = mir->dalvikInsn.opCode;
    RegLocation rlDest;
    RegLocation rlResult;
    int BBBB = mir->dalvikInsn.vB;
    if (dalvikOpCode == OP_CONST_WIDE_16) {
        rlDest = getDestLocWide(cUnit, mir, 0, 1);
        rlResult = evalLoc(cUnit, rlDest, kCoreReg, true);
        loadConstantValue(cUnit, rlResult.lowReg, BBBB);
        opRegRegImm(cUnit, kOpAsr, rlResult.highReg, rlResult.lowReg, 31);
        storeValueWide(cUnit, rlDest, rlResult);
    } else if (dalvikOpCode == OP_CONST_16) {
        rlDest = getDestLoc(cUnit, mir, 0);
        rlResult = evalLoc(cUnit, rlDest, kAnyReg, true);
        loadConstantValue(cUnit, rlResult.lowReg, BBBB);
        storeValue(cUnit, rlDest, rlResult);
    } else
        return true;
    return false;
}

/* Compare agaist zero */
static bool handleFmt21t(CompilationUnit *cUnit, MIR *mir, BasicBlock *bb,
                         ArmLIR *labelList)
{
    OpCode dalvikOpCode = mir->dalvikInsn.opCode;
    ArmConditionCode cond;
    RegLocation rlSrc = getSrcLoc(cUnit, mir, 0);
    rlSrc = loadValue(cUnit, rlSrc, kCoreReg);
    opRegImm(cUnit, kOpCmp, rlSrc.lowReg, 0);

//TUNING: break this out to allow use of Thumb2 CB[N]Z
    switch (dalvikOpCode) {
        case OP_IF_EQZ:
            cond = kArmCondEq;
            break;
        case OP_IF_NEZ:
            cond = kArmCondNe;
            break;
        case OP_IF_LTZ:
            cond = kArmCondLt;
            break;
        case OP_IF_GEZ:
            cond = kArmCondGe;
            break;
        case OP_IF_GTZ:
            cond = kArmCondGt;
            break;
        case OP_IF_LEZ:
            cond = kArmCondLe;
            break;
        default:
            cond = 0;
            LOGE("Unexpected opcode (%d) for Fmt21t\n", dalvikOpCode);
            dvmAbort();
    }
    genConditionalBranch(cUnit, cond, &labelList[bb->taken->id]);
    /* This mostly likely will be optimized away in a later phase */
    genUnconditionalBranch(cUnit, &labelList[bb->fallThrough->id]);
    return false;
}

static bool handleFmt22b_Fmt22s(CompilationUnit *cUnit, MIR *mir)
{
    OpCode dalvikOpCode = mir->dalvikInsn.opCode;
    RegLocation rlSrc = getSrcLoc(cUnit, mir, 0);
    RegLocation rlDest = getDestLoc(cUnit, mir, 0);
    RegLocation rlResult;
    int lit = mir->dalvikInsn.vC;
    OpKind op = 0;      /* Make gcc happy */
    int shiftOp = false;
    bool isDiv = false;

    int __aeabi_idivmod(int op1, int op2);
    int __aeabi_idiv(int op1, int op2);

    switch (dalvikOpCode) {
        case OP_RSUB_INT_LIT8:
        case OP_RSUB_INT: {
            int tReg;
            //TUNING: add support for use of Arm rsub op
            rlSrc = loadValue(cUnit, rlSrc, kCoreReg);
            tReg = allocTemp(cUnit);
            loadConstant(cUnit, tReg, lit);
            rlResult = evalLoc(cUnit, rlDest, kCoreReg, true);
            opRegRegReg(cUnit, kOpSub, rlResult.lowReg,
                        tReg, rlSrc.lowReg);
            storeValue(cUnit, rlDest, rlResult);
            return false;
            break;
        }

        case OP_ADD_INT_LIT8:
        case OP_ADD_INT_LIT16:
            op = kOpAdd;
            break;
        case OP_MUL_INT_LIT8:
        case OP_MUL_INT_LIT16:
            op = kOpMul;
            break;
        case OP_AND_INT_LIT8:
        case OP_AND_INT_LIT16:
            op = kOpAnd;
            break;
        case OP_OR_INT_LIT8:
        case OP_OR_INT_LIT16:
            op = kOpOr;
            break;
        case OP_XOR_INT_LIT8:
        case OP_XOR_INT_LIT16:
            op = kOpXor;
            break;
        case OP_SHL_INT_LIT8:
            shiftOp = true;
            op = kOpLsl;
            break;
        case OP_SHR_INT_LIT8:
            shiftOp = true;
            op = kOpAsr;
            break;
        case OP_USHR_INT_LIT8:
            shiftOp = true;
            op = kOpLsr;
            break;

        case OP_DIV_INT_LIT8:
        case OP_DIV_INT_LIT16:
        case OP_REM_INT_LIT8:
        case OP_REM_INT_LIT16:
            if (lit == 0) {
                /* Let the interpreter deal with div by 0 */
                genInterpSingleStep(cUnit, mir);
                return false;
            }
            loadValueDirectFixed(cUnit, rlSrc, r0);
            clobberReg(cUnit, r0);
            if ((dalvikOpCode == OP_DIV_INT_LIT8) ||
                (dalvikOpCode == OP_DIV_INT_LIT16)) {
                loadConstant(cUnit, r2, (int)__aeabi_idiv);
                isDiv = true;
            } else {
                loadConstant(cUnit, r2, (int)__aeabi_idivmod);
                isDiv = false;
            }
            loadConstant(cUnit, r1, lit);
            opReg(cUnit, kOpBlx, r2);
            clobberCallRegs(cUnit);
            if (isDiv)
                rlResult = getReturnLoc(cUnit);
            else
                rlResult = getReturnLocAlt(cUnit);
            storeValue(cUnit, rlDest, rlResult);
            return false;
            break;
        default:
            return true;
    }
    rlSrc = loadValue(cUnit, rlSrc, kCoreReg);
    rlResult = evalLoc(cUnit, rlDest, kCoreReg, true);
    // Avoid shifts by literal 0 - no support in Thumb.  Change to copy
    if (shiftOp && (lit == 0)) {
        genRegCopy(cUnit, rlResult.lowReg, rlSrc.lowReg);
    } else {
        opRegRegImm(cUnit, op, rlResult.lowReg, rlSrc.lowReg, lit);
    }
    storeValue(cUnit, rlDest, rlResult);
    return false;
}

static bool handleFmt22c(CompilationUnit *cUnit, MIR *mir)
{
    OpCode dalvikOpCode = mir->dalvikInsn.opCode;
    int fieldOffset;

    if (dalvikOpCode >= OP_IGET && dalvikOpCode <= OP_IPUT_SHORT) {
        InstField *pInstField = (InstField *)
            cUnit->method->clazz->pDvmDex->pResFields[mir->dalvikInsn.vC];

        assert(pInstField != NULL);
        fieldOffset = pInstField->byteOffset;
    } else {
        /* Deliberately break the code while make the compiler happy */
        fieldOffset = -1;
    }
    switch (dalvikOpCode) {
        case OP_NEW_ARRAY: {
            // Generates a call - use explicit registers
            RegLocation rlSrc = getSrcLoc(cUnit, mir, 0);
            RegLocation rlDest = getDestLoc(cUnit, mir, 0);
            RegLocation rlResult;
            void *classPtr = (void*)
              (cUnit->method->clazz->pDvmDex->pResClasses[mir->dalvikInsn.vC]);
            assert(classPtr != NULL);
            genExportPC(cUnit, mir);
            loadValueDirectFixed(cUnit, rlSrc, r1);   /* Len */
            loadConstant(cUnit, r0, (int) classPtr );
            loadConstant(cUnit, r3, (int)dvmAllocArrayByClass);
            /*
             * "len < 0": bail to the interpreter to re-execute the
             * instruction
             */
            ArmLIR *pcrLabel =
                genRegImmCheck(cUnit, kArmCondMi, r1, 0, mir->offset, NULL);
            loadConstant(cUnit, r2, ALLOC_DONT_TRACK);
            opReg(cUnit, kOpBlx, r3);
            clobberCallRegs(cUnit);
            /* generate a branch over if allocation is successful */
            opRegImm(cUnit, kOpCmp, r0, 0); /* NULL? */
            ArmLIR *branchOver = opCondBranch(cUnit, kArmCondNe);
            /*
             * OOM exception needs to be thrown here and cannot re-execute
             */
            loadConstant(cUnit, r0,
                         (int) (cUnit->method->insns + mir->offset));
            genDispatchToHandler(cUnit, TEMPLATE_THROW_EXCEPTION_COMMON);
            /* noreturn */

            ArmLIR *target = newLIR0(cUnit, kArmPseudoTargetLabel);
            target->defMask = ENCODE_ALL;
            branchOver->generic.target = (LIR *) target;
            rlResult = getReturnLoc(cUnit);
            storeValue(cUnit, rlDest, rlResult);
            break;
        }
        case OP_INSTANCE_OF: {
            // May generate a call - use explicit registers
            RegLocation rlSrc = getSrcLoc(cUnit, mir, 0);
            RegLocation rlDest = getDestLoc(cUnit, mir, 0);
            RegLocation rlResult;
            ClassObject *classPtr =
              (cUnit->method->clazz->pDvmDex->pResClasses[mir->dalvikInsn.vC]);
            assert(classPtr != NULL);
            loadValueDirectFixed(cUnit, rlSrc, r0);  /* Ref */
            loadConstant(cUnit, r2, (int) classPtr );
//TUNING: compare to 0 primative to allow use of CB[N]Z
            opRegImm(cUnit, kOpCmp, r0, 0); /* NULL? */
            /* When taken r0 has NULL which can be used for store directly */
            ArmLIR *branch1 = opCondBranch(cUnit, kArmCondEq);
            /* r1 now contains object->clazz */
            loadWordDisp(cUnit, r0, offsetof(Object, clazz), r1);
            /* r1 now contains object->clazz */
            loadConstant(cUnit, r3, (int)dvmInstanceofNonTrivial);
            loadConstant(cUnit, r0, 1);                /* Assume true */
            opRegReg(cUnit, kOpCmp, r1, r2);
            ArmLIR *branch2 = opCondBranch(cUnit, kArmCondEq);
            genRegCopy(cUnit, r0, r1);
            genRegCopy(cUnit, r1, r2);
            opReg(cUnit, kOpBlx, r3);
            clobberCallRegs(cUnit);
            /* branch target here */
            ArmLIR *target = newLIR0(cUnit, kArmPseudoTargetLabel);
            target->defMask = ENCODE_ALL;
            rlResult = getReturnLoc(cUnit);
            storeValue(cUnit, rlDest, rlResult);
            branch1->generic.target = (LIR *)target;
            branch2->generic.target = (LIR *)target;
            break;
        }
        case OP_IGET_WIDE:
            genIGetWide(cUnit, mir, fieldOffset);
            break;
        case OP_IGET:
        case OP_IGET_OBJECT:
            genIGet(cUnit, mir, kWord, fieldOffset);
            break;
        case OP_IGET_BOOLEAN:
            genIGet(cUnit, mir, kUnsignedByte, fieldOffset);
            break;
        case OP_IGET_BYTE:
            genIGet(cUnit, mir, kSignedByte, fieldOffset);
            break;
        case OP_IGET_CHAR:
            genIGet(cUnit, mir, kUnsignedHalf, fieldOffset);
            break;
        case OP_IGET_SHORT:
            genIGet(cUnit, mir, kSignedHalf, fieldOffset);
            break;
        case OP_IPUT_WIDE:
            genIPutWide(cUnit, mir, fieldOffset);
            break;
        case OP_IPUT:
        case OP_IPUT_OBJECT:
            genIPut(cUnit, mir, kWord, fieldOffset);
            break;
        case OP_IPUT_SHORT:
        case OP_IPUT_CHAR:
            genIPut(cUnit, mir, kUnsignedHalf, fieldOffset);
            break;
        case OP_IPUT_BYTE:
        case OP_IPUT_BOOLEAN:
            genIPut(cUnit, mir, kUnsignedByte, fieldOffset);
            break;
        default:
            return true;
    }
    return false;
}

static bool handleFmt22cs(CompilationUnit *cUnit, MIR *mir)
{
    OpCode dalvikOpCode = mir->dalvikInsn.opCode;
    int fieldOffset =  mir->dalvikInsn.vC;
    switch (dalvikOpCode) {
        case OP_IGET_QUICK:
        case OP_IGET_OBJECT_QUICK:
            genIGet(cUnit, mir, kWord, fieldOffset);
            break;
        case OP_IPUT_QUICK:
        case OP_IPUT_OBJECT_QUICK:
            genIPut(cUnit, mir, kWord, fieldOffset);
            break;
        case OP_IGET_WIDE_QUICK:
            genIGetWide(cUnit, mir, fieldOffset);
            break;
        case OP_IPUT_WIDE_QUICK:
            genIPutWide(cUnit, mir, fieldOffset);
            break;
        default:
            return true;
    }
    return false;

}

/* Compare agaist zero */
static bool handleFmt22t(CompilationUnit *cUnit, MIR *mir, BasicBlock *bb,
                         ArmLIR *labelList)
{
    OpCode dalvikOpCode = mir->dalvikInsn.opCode;
    ArmConditionCode cond;
    RegLocation rlSrc1 = getSrcLoc(cUnit, mir, 0);
    RegLocation rlSrc2 = getSrcLoc(cUnit, mir, 1);

    rlSrc1 = loadValue(cUnit, rlSrc1, kCoreReg);
    rlSrc2 = loadValue(cUnit, rlSrc2, kCoreReg);
    opRegReg(cUnit, kOpCmp, rlSrc1.lowReg, rlSrc2.lowReg);

    switch (dalvikOpCode) {
        case OP_IF_EQ:
            cond = kArmCondEq;
            break;
        case OP_IF_NE:
            cond = kArmCondNe;
            break;
        case OP_IF_LT:
            cond = kArmCondLt;
            break;
        case OP_IF_GE:
            cond = kArmCondGe;
            break;
        case OP_IF_GT:
            cond = kArmCondGt;
            break;
        case OP_IF_LE:
            cond = kArmCondLe;
            break;
        default:
            cond = 0;
            LOGE("Unexpected opcode (%d) for Fmt22t\n", dalvikOpCode);
            dvmAbort();
    }
    genConditionalBranch(cUnit, cond, &labelList[bb->taken->id]);
    /* This mostly likely will be optimized away in a later phase */
    genUnconditionalBranch(cUnit, &labelList[bb->fallThrough->id]);
    return false;
}

static bool handleFmt22x_Fmt32x(CompilationUnit *cUnit, MIR *mir)
{
    OpCode opCode = mir->dalvikInsn.opCode;

    switch (opCode) {
        case OP_MOVE_16:
        case OP_MOVE_OBJECT_16:
        case OP_MOVE_FROM16:
        case OP_MOVE_OBJECT_FROM16: {
            storeValue(cUnit, getDestLoc(cUnit, mir, 0),
                       getSrcLoc(cUnit, mir, 0));
            break;
        }
        case OP_MOVE_WIDE_16:
        case OP_MOVE_WIDE_FROM16: {
            storeValueWide(cUnit, getDestLocWide(cUnit, mir, 0, 1),
                           getSrcLocWide(cUnit, mir, 0, 1));
            break;
        }
        default:
            return true;
    }
    return false;
}

static bool handleFmt23x(CompilationUnit *cUnit, MIR *mir)
{
    OpCode opCode = mir->dalvikInsn.opCode;
    RegLocation rlSrc1;
    RegLocation rlSrc2;
    RegLocation rlDest;

    if ( (opCode >= OP_ADD_INT) && (opCode <= OP_REM_DOUBLE)) {
        return handleArithOp( cUnit, mir );
    }

    /* APUTs have 3 sources and no targets */
    if (mir->ssaRep->numDefs == 0) {
        if (mir->ssaRep->numUses == 3) {
            rlDest = getSrcLoc(cUnit, mir, 0);
            rlSrc1 = getSrcLoc(cUnit, mir, 1);
            rlSrc2 = getSrcLoc(cUnit, mir, 2);
        } else {
            assert(mir->ssaRep->numUses == 4);
            rlDest = getSrcLocWide(cUnit, mir, 0, 1);
            rlSrc1 = getSrcLoc(cUnit, mir, 2);
            rlSrc2 = getSrcLoc(cUnit, mir, 3);
        }
    } else {
        /* Two sources and 1 dest.  Deduce the operand sizes */
        if (mir->ssaRep->numUses == 4) {
            rlSrc1 = getSrcLocWide(cUnit, mir, 0, 1);
            rlSrc2 = getSrcLocWide(cUnit, mir, 2, 3);
        } else {
            assert(mir->ssaRep->numUses == 2);
            rlSrc1 = getSrcLoc(cUnit, mir, 0);
            rlSrc2 = getSrcLoc(cUnit, mir, 1);
        }
        if (mir->ssaRep->numDefs == 2) {
            rlDest = getDestLocWide(cUnit, mir, 0, 1);
        } else {
            assert(mir->ssaRep->numDefs == 1);
            rlDest = getDestLoc(cUnit, mir, 0);
        }
    }


    switch (opCode) {
        case OP_CMPL_FLOAT:
        case OP_CMPG_FLOAT:
        case OP_CMPL_DOUBLE:
        case OP_CMPG_DOUBLE:
            return handleCmpFP(cUnit, mir, rlDest, rlSrc1, rlSrc2);
        case OP_CMP_LONG:
            genCmpLong(cUnit, mir, rlDest, rlSrc1, rlSrc2);
            break;
        case OP_AGET_WIDE:
            genArrayGet(cUnit, mir, kLong, rlSrc1, rlSrc2, rlDest, 3);
            break;
        case OP_AGET:
        case OP_AGET_OBJECT:
            genArrayGet(cUnit, mir, kWord, rlSrc1, rlSrc2, rlDest, 2);
            break;
        case OP_AGET_BOOLEAN:
            genArrayGet(cUnit, mir, kUnsignedByte, rlSrc1, rlSrc2, rlDest, 0);
            break;
        case OP_AGET_BYTE:
            genArrayGet(cUnit, mir, kSignedByte, rlSrc1, rlSrc2, rlDest, 0);
            break;
        case OP_AGET_CHAR:
            genArrayGet(cUnit, mir, kUnsignedHalf, rlSrc1, rlSrc2, rlDest, 1);
            break;
        case OP_AGET_SHORT:
            genArrayGet(cUnit, mir, kSignedHalf, rlSrc1, rlSrc2, rlDest, 1);
            break;
        case OP_APUT_WIDE:
            genArrayPut(cUnit, mir, kLong, rlSrc1, rlSrc2, rlDest, 3);
            break;
        case OP_APUT:
        case OP_APUT_OBJECT:
            genArrayPut(cUnit, mir, kWord, rlSrc1, rlSrc2, rlDest, 2);
            break;
        case OP_APUT_SHORT:
        case OP_APUT_CHAR:
            genArrayPut(cUnit, mir, kUnsignedHalf, rlSrc1, rlSrc2, rlDest, 1);
            break;
        case OP_APUT_BYTE:
        case OP_APUT_BOOLEAN:
            genArrayPut(cUnit, mir, kUnsignedByte, rlSrc1, rlSrc2, rlDest, 0);
            break;
        default:
            return true;
    }
    return false;
}

static bool handleFmt31t(CompilationUnit *cUnit, MIR *mir)
{
    OpCode dalvikOpCode = mir->dalvikInsn.opCode;
    switch (dalvikOpCode) {
        case OP_FILL_ARRAY_DATA: {
            RegLocation rlSrc = getSrcLoc(cUnit, mir, 0);
            // Making a call - use explicit registers
            genExportPC(cUnit, mir);
            loadValueDirectFixed(cUnit, rlSrc, r0);
            loadConstant(cUnit, r3, (int)dvmInterpHandleFillArrayData);
            loadConstant(cUnit, r1, (mir->dalvikInsn.vB << 1) +
                 (int) (cUnit->method->insns + mir->offset));
            opReg(cUnit, kOpBlx, r2);
            clobberCallRegs(cUnit);
            genZeroCheck(cUnit, r0, mir->offset, NULL);
            break;
        }
        /*
         * TODO
         * - Add a 1 to 3-entry per-location cache here to completely
         *   bypass the dvmInterpHandle[Packed/Sparse]Switch call w/ chaining
         * - Use out-of-line handlers for both of these.  These ops
         *   handle their own register allocation.
         */
        case OP_PACKED_SWITCH:
        case OP_SPARSE_SWITCH: {
            RegLocation rlSrc = getSrcLoc(cUnit, mir, 0);
            loadValueDirectFixed(cUnit, rlSrc, r1);
            lockAllTemps(cUnit);
            // Exit to the interpreter, setting up r4PC
            if (dalvikOpCode == OP_PACKED_SWITCH) {
                loadConstant(cUnit, r4PC, (int)dvmInterpHandlePackedSwitch);
            } else {
                loadConstant(cUnit, r4PC, (int)dvmInterpHandleSparseSwitch);
            }
            loadConstant(cUnit, r0, (mir->dalvikInsn.vB << 1) +
                 (int) (cUnit->method->insns + mir->offset));
            opReg(cUnit, kOpBlx, r4PC);
            clobberCallRegs(cUnit);
            loadConstant(cUnit, r1, (int)(cUnit->method->insns + mir->offset));
            loadWordDisp(cUnit, rGLUE, offsetof(InterpState,
                         jitToInterpEntries.dvmJitToInterpNoChain), r2);
            opRegReg(cUnit, kOpAdd, r0, r0);
            opRegRegReg(cUnit, kOpAdd, r4PC, r0, r1);
            opReg(cUnit, kOpBlx, r2);
            clobberCallRegs(cUnit);
            break;
        }
        default:
            return true;
    }
    return false;
}

static bool handleFmt35c_3rc(CompilationUnit *cUnit, MIR *mir, BasicBlock *bb,
                             ArmLIR *labelList)
{
    ArmLIR *retChainingCell = NULL;
    ArmLIR *pcrLabel = NULL;

    if (bb->fallThrough != NULL)
        retChainingCell = &labelList[bb->fallThrough->id];

    DecodedInstruction *dInsn = &mir->dalvikInsn;
    switch (mir->dalvikInsn.opCode) {
        /*
         * calleeMethod = this->clazz->vtable[
         *     method->clazz->pDvmDex->pResMethods[BBBB]->methodIndex
         * ]
         */
        case OP_INVOKE_VIRTUAL:
        case OP_INVOKE_VIRTUAL_RANGE: {
            ArmLIR *predChainingCell = &labelList[bb->taken->id];
            int methodIndex =
                cUnit->method->clazz->pDvmDex->pResMethods[dInsn->vB]->
                methodIndex;

            if (mir->dalvikInsn.opCode == OP_INVOKE_VIRTUAL)
                genProcessArgsNoRange(cUnit, mir, dInsn, &pcrLabel);
            else
                genProcessArgsRange(cUnit, mir, dInsn, &pcrLabel);

            genInvokeVirtualCommon(cUnit, mir, methodIndex,
                                   retChainingCell,
                                   predChainingCell,
                                   pcrLabel);
            break;
        }
        /*
         * calleeMethod = method->clazz->super->vtable[method->clazz->pDvmDex
         *                ->pResMethods[BBBB]->methodIndex]
         */
        /* TODO - not excersized in RunPerf.jar */
        case OP_INVOKE_SUPER:
        case OP_INVOKE_SUPER_RANGE: {
            int mIndex = cUnit->method->clazz->pDvmDex->
                pResMethods[dInsn->vB]->methodIndex;
            const Method *calleeMethod =
                cUnit->method->clazz->super->vtable[mIndex];

            if (mir->dalvikInsn.opCode == OP_INVOKE_SUPER)
                genProcessArgsNoRange(cUnit, mir, dInsn, &pcrLabel);
            else
                genProcessArgsRange(cUnit, mir, dInsn, &pcrLabel);

            /* r0 = calleeMethod */
            loadConstant(cUnit, r0, (int) calleeMethod);

            genInvokeSingletonCommon(cUnit, mir, bb, labelList, pcrLabel,
                                     calleeMethod);
            break;
        }
        /* calleeMethod = method->clazz->pDvmDex->pResMethods[BBBB] */
        case OP_INVOKE_DIRECT:
        case OP_INVOKE_DIRECT_RANGE: {
            const Method *calleeMethod =
                cUnit->method->clazz->pDvmDex->pResMethods[dInsn->vB];

            if (mir->dalvikInsn.opCode == OP_INVOKE_DIRECT)
                genProcessArgsNoRange(cUnit, mir, dInsn, &pcrLabel);
            else
                genProcessArgsRange(cUnit, mir, dInsn, &pcrLabel);

            /* r0 = calleeMethod */
            loadConstant(cUnit, r0, (int) calleeMethod);

            genInvokeSingletonCommon(cUnit, mir, bb, labelList, pcrLabel,
                                     calleeMethod);
            break;
        }
        /* calleeMethod = method->clazz->pDvmDex->pResMethods[BBBB] */
        case OP_INVOKE_STATIC:
        case OP_INVOKE_STATIC_RANGE: {
            const Method *calleeMethod =
                cUnit->method->clazz->pDvmDex->pResMethods[dInsn->vB];

            if (mir->dalvikInsn.opCode == OP_INVOKE_STATIC)
                genProcessArgsNoRange(cUnit, mir, dInsn,
                                      NULL /* no null check */);
            else
                genProcessArgsRange(cUnit, mir, dInsn,
                                    NULL /* no null check */);

            /* r0 = calleeMethod */
            loadConstant(cUnit, r0, (int) calleeMethod);

            genInvokeSingletonCommon(cUnit, mir, bb, labelList, pcrLabel,
                                     calleeMethod);
            break;
        }
    /*
         * calleeMethod = dvmFindInterfaceMethodInCache(this->clazz,
         *                    BBBB, method, method->clazz->pDvmDex)
         *
         *  Given "invoke-interface {v0}", the following is the generated code:
         *
         * 0x426a9abe : ldr     r0, [r5, #0]   --+
         * 0x426a9ac0 : mov     r7, r5           |
         * 0x426a9ac2 : sub     r7, #24          |
         * 0x426a9ac4 : cmp     r0, #0           | genProcessArgsNoRange
         * 0x426a9ac6 : beq     0x426a9afe       |
         * 0x426a9ac8 : stmia   r7, <r0>       --+
         * 0x426a9aca : ldr     r4, [pc, #104] --> r4 <- dalvikPC of this invoke
         * 0x426a9acc : add     r1, pc, #52    --> r1 <- &retChainingCell
         * 0x426a9ace : add     r2, pc, #60    --> r2 <- &predictedChainingCell
         * 0x426a9ad0 : blx_1   0x426a918c     --+ TEMPLATE_INVOKE_METHOD_
         * 0x426a9ad2 : blx_2   see above      --+     PREDICTED_CHAIN
         * 0x426a9ad4 : b       0x426a9b0c     --> off to the predicted chain
         * 0x426a9ad6 : b       0x426a9afe     --> punt to the interpreter
         * 0x426a9ad8 : mov     r8, r1         --+
         * 0x426a9ada : mov     r9, r2           |
         * 0x426a9adc : mov     r10, r3          |
         * 0x426a9ade : mov     r0, r3           |
         * 0x426a9ae0 : mov     r1, #74          | dvmFindInterfaceMethodInCache
         * 0x426a9ae2 : ldr     r2, [pc, #76]    |
         * 0x426a9ae4 : ldr     r3, [pc, #68]    |
         * 0x426a9ae6 : ldr     r7, [pc, #64]    |
         * 0x426a9ae8 : blx     r7             --+
         * 0x426a9aea : mov     r1, r8         --> r1 <- rechain count
         * 0x426a9aec : cmp     r1, #0         --> compare against 0
         * 0x426a9aee : bgt     0x426a9af8     --> >=0? don't rechain
         * 0x426a9af0 : ldr     r7, [r6, #96]  --+
         * 0x426a9af2 : mov     r2, r9           | dvmJitToPatchPredictedChain
         * 0x426a9af4 : mov     r3, r10          |
         * 0x426a9af6 : blx     r7             --+
         * 0x426a9af8 : add     r1, pc, #8     --> r1 <- &retChainingCell
         * 0x426a9afa : blx_1   0x426a9098     --+ TEMPLATE_INVOKE_METHOD_NO_OPT
         * 0x426a9afc : blx_2   see above      --+
         * -------- reconstruct dalvik PC : 0x428b786c @ +0x001e
         * 0x426a9afe (0042): ldr     r0, [pc, #52]
         * Exception_Handling:
         * 0x426a9b00 (0044): ldr     r1, [r6, #84]
         * 0x426a9b02 (0046): blx     r1
         * 0x426a9b04 (0048): .align4
         * -------- chaining cell (hot): 0x0021
         * 0x426a9b04 (0048): ldr     r0, [r6, #92]
         * 0x426a9b06 (004a): blx     r0
         * 0x426a9b08 (004c): data    0x7872(30834)
         * 0x426a9b0a (004e): data    0x428b(17035)
         * 0x426a9b0c (0050): .align4
         * -------- chaining cell (predicted)
         * 0x426a9b0c (0050): data    0x0000(0) --> will be patched into bx
         * 0x426a9b0e (0052): data    0x0000(0)
         * 0x426a9b10 (0054): data    0x0000(0) --> class
         * 0x426a9b12 (0056): data    0x0000(0)
         * 0x426a9b14 (0058): data    0x0000(0) --> method
         * 0x426a9b16 (005a): data    0x0000(0)
         * 0x426a9b18 (005c): data    0x0000(0) --> reset count
         * 0x426a9b1a (005e): data    0x0000(0)
         * 0x426a9b28 (006c): .word (0xad0392a5)
         * 0x426a9b2c (0070): .word (0x6e750)
         * 0x426a9b30 (0074): .word (0x4109a618)
         * 0x426a9b34 (0078): .word (0x428b786c)
         */
        case OP_INVOKE_INTERFACE:
        case OP_INVOKE_INTERFACE_RANGE: {
            ArmLIR *predChainingCell = &labelList[bb->taken->id];
            int methodIndex = dInsn->vB;

            /* Ensure that nothing is both live and dirty */
            flushAllRegs(cUnit);

            if (mir->dalvikInsn.opCode == OP_INVOKE_INTERFACE)
                genProcessArgsNoRange(cUnit, mir, dInsn, &pcrLabel);
            else
                genProcessArgsRange(cUnit, mir, dInsn, &pcrLabel);

            /* "this" is already left in r0 by genProcessArgs* */

            /* r4PC = dalvikCallsite */
            loadConstant(cUnit, r4PC,
                         (int) (cUnit->method->insns + mir->offset));

            /* r1 = &retChainingCell */
            ArmLIR *addrRetChain =
                opRegRegImm(cUnit, kOpAdd, r1, rpc, 0);
            addrRetChain->generic.target = (LIR *) retChainingCell;

            /* r2 = &predictedChainingCell */
            ArmLIR *predictedChainingCell =
                opRegRegImm(cUnit, kOpAdd, r2, rpc, 0);
            predictedChainingCell->generic.target = (LIR *) predChainingCell;

            genDispatchToHandler(cUnit, TEMPLATE_INVOKE_METHOD_PREDICTED_CHAIN);

            /* return through lr - jump to the chaining cell */
            genUnconditionalBranch(cUnit, predChainingCell);

            /*
             * null-check on "this" may have been eliminated, but we still need
             * a PC-reconstruction label for stack overflow bailout.
             */
            if (pcrLabel == NULL) {
                int dPC = (int) (cUnit->method->insns + mir->offset);
                pcrLabel = dvmCompilerNew(sizeof(ArmLIR), true);
                pcrLabel->opCode = ARM_PSEUDO_kPCReconstruction_CELL;
                pcrLabel->operands[0] = dPC;
                pcrLabel->operands[1] = mir->offset;
                /* Insert the place holder to the growable list */
                dvmInsertGrowableList(&cUnit->pcReconstructionList, pcrLabel);
            }

            /* return through lr+2 - punt to the interpreter */
            genUnconditionalBranch(cUnit, pcrLabel);

            /*
             * return through lr+4 - fully resolve the callee method.
             * r1 <- count
             * r2 <- &predictedChainCell
             * r3 <- this->class
             * r4 <- dPC
             * r7 <- this->class->vtable
             */

            /* Save count, &predictedChainCell, and class to high regs first */
            genRegCopy(cUnit, r8, r1);
            genRegCopy(cUnit, r9, r2);
            genRegCopy(cUnit, r10, r3);

            /* r0 now contains this->clazz */
            genRegCopy(cUnit, r0, r3);

            /* r1 = BBBB */
            loadConstant(cUnit, r1, dInsn->vB);

            /* r2 = method (caller) */
            loadConstant(cUnit, r2, (int) cUnit->method);

            /* r3 = pDvmDex */
            loadConstant(cUnit, r3, (int) cUnit->method->clazz->pDvmDex);

            loadConstant(cUnit, r7,
                         (intptr_t) dvmFindInterfaceMethodInCache);
            opReg(cUnit, kOpBlx, r7);

            /* r0 = calleeMethod (returned from dvmFindInterfaceMethodInCache */

            genRegCopy(cUnit, r1, r8);

            /* Check if rechain limit is reached */
            opRegImm(cUnit, kOpCmp, r1, 0);

            ArmLIR *bypassRechaining = opCondBranch(cUnit, kArmCondGt);

            loadWordDisp(cUnit, rGLUE, offsetof(InterpState,
                         jitToInterpEntries.dvmJitToPatchPredictedChain), r7);

            genRegCopy(cUnit, r2, r9);
            genRegCopy(cUnit, r3, r10);

            /*
             * r0 = calleeMethod
             * r2 = &predictedChainingCell
             * r3 = class
             *
             * &returnChainingCell has been loaded into r1 but is not needed
             * when patching the chaining cell and will be clobbered upon
             * returning so it will be reconstructed again.
             */
            opReg(cUnit, kOpBlx, r7);

            /* r1 = &retChainingCell */
            addrRetChain = opRegRegImm(cUnit, kOpAdd, r1, rpc, 0);
            addrRetChain->generic.target = (LIR *) retChainingCell;

            bypassRechaining->generic.target = (LIR *) addrRetChain;

            /*
             * r0 = this, r1 = calleeMethod,
             * r1 = &ChainingCell,
             * r4PC = callsiteDPC,
             */
            genDispatchToHandler(cUnit, TEMPLATE_INVOKE_METHOD_NO_OPT);
#if defined(INVOKE_STATS)
            gDvmJit.invokePredictedChain++;
#endif
            /* Handle exceptions using the interpreter */
            genTrap(cUnit, mir->offset, pcrLabel);
            break;
        }
        /* NOP */
        case OP_INVOKE_DIRECT_EMPTY: {
            return false;
        }
        case OP_FILLED_NEW_ARRAY:
        case OP_FILLED_NEW_ARRAY_RANGE: {
            /* Just let the interpreter deal with these */
            genInterpSingleStep(cUnit, mir);
            break;
        }
        default:
            return true;
    }
    return false;
}

static bool handleFmt35ms_3rms(CompilationUnit *cUnit, MIR *mir,
                               BasicBlock *bb, ArmLIR *labelList)
{
    ArmLIR *retChainingCell = &labelList[bb->fallThrough->id];
    ArmLIR *predChainingCell = &labelList[bb->taken->id];
    ArmLIR *pcrLabel = NULL;

    DecodedInstruction *dInsn = &mir->dalvikInsn;
    switch (mir->dalvikInsn.opCode) {
        /* calleeMethod = this->clazz->vtable[BBBB] */
        case OP_INVOKE_VIRTUAL_QUICK_RANGE:
        case OP_INVOKE_VIRTUAL_QUICK: {
            int methodIndex = dInsn->vB;
            if (mir->dalvikInsn.opCode == OP_INVOKE_VIRTUAL_QUICK)
                genProcessArgsNoRange(cUnit, mir, dInsn, &pcrLabel);
            else
                genProcessArgsRange(cUnit, mir, dInsn, &pcrLabel);

            genInvokeVirtualCommon(cUnit, mir, methodIndex,
                                   retChainingCell,
                                   predChainingCell,
                                   pcrLabel);
            break;
        }
        /* calleeMethod = method->clazz->super->vtable[BBBB] */
        case OP_INVOKE_SUPER_QUICK:
        case OP_INVOKE_SUPER_QUICK_RANGE: {
            const Method *calleeMethod =
                cUnit->method->clazz->super->vtable[dInsn->vB];

            if (mir->dalvikInsn.opCode == OP_INVOKE_SUPER_QUICK)
                genProcessArgsNoRange(cUnit, mir, dInsn, &pcrLabel);
            else
                genProcessArgsRange(cUnit, mir, dInsn, &pcrLabel);

            /* r0 = calleeMethod */
            loadConstant(cUnit, r0, (int) calleeMethod);

            genInvokeSingletonCommon(cUnit, mir, bb, labelList, pcrLabel,
                                     calleeMethod);
            /* Handle exceptions using the interpreter */
            genTrap(cUnit, mir->offset, pcrLabel);
            break;
        }
        default:
            return true;
    }
    return false;
}

/*
 * NOTE: We assume here that the special native inline routines
 * are side-effect free.  By making this assumption, we can safely
 * re-execute the routine from the interpreter if it decides it
 * wants to throw an exception. We still need to EXPORT_PC(), though.
 */
static bool handleFmt3inline(CompilationUnit *cUnit, MIR *mir)
{
    DecodedInstruction *dInsn = &mir->dalvikInsn;
    switch( mir->dalvikInsn.opCode) {
        case OP_EXECUTE_INLINE: {
            unsigned int i;
            const InlineOperation* inLineTable = dvmGetInlineOpsTable();
            int offset = offsetof(InterpState, retval);
            int operation = dInsn->vB;
            int tReg1;
            int tReg2;
            switch (operation) {
                case INLINE_EMPTYINLINEMETHOD:
                    return false;  /* Nop */
                case INLINE_STRING_LENGTH:
                    return genInlinedStringLength(cUnit, mir);
                case INLINE_MATH_ABS_INT:
                    return genInlinedAbsInt(cUnit, mir);
                case INLINE_MATH_ABS_LONG:
                    return genInlinedAbsLong(cUnit, mir);
                case INLINE_MATH_MIN_INT:
                    return genInlinedMinMaxInt(cUnit, mir, true);
                case INLINE_MATH_MAX_INT:
                    return genInlinedMinMaxInt(cUnit, mir, false);
                case INLINE_STRING_CHARAT:
                    return genInlinedStringCharAt(cUnit, mir);
                case INLINE_MATH_SQRT:
                    if (genInlineSqrt(cUnit, mir))
                        return false;
                    else
                        break;   /* Handle with C routine */
                case INLINE_MATH_COS:
                case INLINE_MATH_SIN:
                        break;   /* Handle with C routine */
                case INLINE_MATH_ABS_FLOAT:
                    if (genInlinedAbsFloat(cUnit, mir))
                        return false;
                    else
                        break;
                case INLINE_MATH_ABS_DOUBLE:
                    if (genInlinedAbsDouble(cUnit, mir))
                        return false;
                    else
                        break;
                case INLINE_STRING_COMPARETO:
                case INLINE_STRING_EQUALS:
                case INLINE_STRING_INDEXOF_I:
                case INLINE_STRING_INDEXOF_II:
                    break;
                default:
                    dvmAbort();
            }
            clobberCallRegs(cUnit);
            clobberReg(cUnit, r4PC);
            clobberReg(cUnit, r7);
            opRegRegImm(cUnit, kOpAdd, r4PC, rGLUE, offset);
            opImm(cUnit, kOpPush, (1<<r4PC) | (1<<r7));
            loadConstant(cUnit, r4PC, (int)inLineTable[operation].func);
            genExportPC(cUnit, mir);
            for (i=0; i < dInsn->vA; i++) {
                loadValueDirect(cUnit, getSrcLoc(cUnit, mir, i), i);
            }
            opReg(cUnit, kOpBlx, r4PC);
            opRegImm(cUnit, kOpAdd, r13, 8);
            genZeroCheck(cUnit, r0, mir->offset, NULL);
            break;
        }
        default:
            return true;
    }
    return false;
}

static bool handleFmt51l(CompilationUnit *cUnit, MIR *mir)
{
    //TUNING: We're using core regs here - not optimal when target is a double
    RegLocation rlDest = getDestLocWide(cUnit, mir, 0, 1);
    RegLocation rlResult = evalLoc(cUnit, rlDest, kCoreReg, true);
    loadConstantValue(cUnit, rlResult.lowReg,
                      mir->dalvikInsn.vB_wide & 0xFFFFFFFFUL);
    loadConstantValue(cUnit, rlResult.highReg,
                      (mir->dalvikInsn.vB_wide>>32) & 0xFFFFFFFFUL);
    storeValueWide(cUnit, rlDest, rlResult);
    return false;
}

/*
 * The following are special processing routines that handle transfer of
 * controls between compiled code and the interpreter. Certain VM states like
 * Dalvik PC and special-purpose registers are reconstructed here.
 */

/* Chaining cell for code that may need warmup. */
static void handleNormalChainingCell(CompilationUnit *cUnit,
                                     unsigned int offset)
{
    loadWordDisp(cUnit, rGLUE, offsetof(InterpState,
                 jitToInterpEntries.dvmJitToInterpNormal), r0);
    opReg(cUnit, kOpBlx, r0);
    addWordData(cUnit, (int) (cUnit->method->insns + offset), true);
}

/*
 * Chaining cell for instructions that immediately following already translated
 * code.
 */
static void handleHotChainingCell(CompilationUnit *cUnit,
                                  unsigned int offset)
{
    loadWordDisp(cUnit, rGLUE, offsetof(InterpState,
                 jitToInterpEntries.dvmJitToTraceSelect), r0);
    opReg(cUnit, kOpBlx, r0);
    addWordData(cUnit, (int) (cUnit->method->insns + offset), true);
}

#if defined(WITH_SELF_VERIFICATION) || defined(WITH_JIT_TUNING)
/* Chaining cell for branches that branch back into the same basic block */
static void handleBackwardBranchChainingCell(CompilationUnit *cUnit,
                                             unsigned int offset)
{
#if defined(WITH_SELF_VERIFICATION)
    newLIR3(cUnit, kThumbLdrRRI5, r0, rGLUE,
        offsetof(InterpState, jitToInterpEntries.dvmJitToBackwardBranch) >> 2);
#else
    newLIR3(cUnit, kThumbLdrRRI5, r0, rGLUE,
        offsetof(InterpState, jitToInterpEntries.dvmJitToInterpNormal) >> 2);
#endif
    newLIR1(cUnit, kThumbBlxR, r0);
    addWordData(cUnit, (int) (cUnit->method->insns + offset), true);
}

#endif
/* Chaining cell for monomorphic method invocations. */
static void handleInvokeSingletonChainingCell(CompilationUnit *cUnit,
                                              const Method *callee)
{
    loadWordDisp(cUnit, rGLUE, offsetof(InterpState,
                 jitToInterpEntries.dvmJitToTraceSelect), r0);
    opReg(cUnit, kOpBlx, r0);
    addWordData(cUnit, (int) (callee->insns), true);
}

/* Chaining cell for monomorphic method invocations. */
static void handleInvokePredictedChainingCell(CompilationUnit *cUnit)
{

    /* Should not be executed in the initial state */
    addWordData(cUnit, PREDICTED_CHAIN_BX_PAIR_INIT, true);
    /* To be filled: class */
    addWordData(cUnit, PREDICTED_CHAIN_CLAZZ_INIT, true);
    /* To be filled: method */
    addWordData(cUnit, PREDICTED_CHAIN_METHOD_INIT, true);
    /*
     * Rechain count. The initial value of 0 here will trigger chaining upon
     * the first invocation of this callsite.
     */
    addWordData(cUnit, PREDICTED_CHAIN_COUNTER_INIT, true);
}

/* Load the Dalvik PC into r0 and jump to the specified target */
static void handlePCReconstruction(CompilationUnit *cUnit,
                                   ArmLIR *targetLabel)
{
    ArmLIR **pcrLabel =
        (ArmLIR **) cUnit->pcReconstructionList.elemList;
    int numElems = cUnit->pcReconstructionList.numUsed;
    int i;
    for (i = 0; i < numElems; i++) {
        dvmCompilerAppendLIR(cUnit, (LIR *) pcrLabel[i]);
        /* r0 = dalvik PC */
        loadConstant(cUnit, r0, pcrLabel[i]->operands[0]);
        genUnconditionalBranch(cUnit, targetLabel);
    }
}

static char *extendedMIROpNames[kMirOpLast - kMirOpFirst] = {
    "kMirOpPhi",
    "kMirOpNullNRangeUpCheck",
    "kMirOpNullNRangeDownCheck",
    "kMirOpLowerBound",
    "kMirOpPunt",
};

/*
 * vA = arrayReg;
 * vB = idxReg;
 * vC = endConditionReg;
 * arg[0] = maxC
 * arg[1] = minC
 * arg[2] = loopBranchConditionCode
 */
static void genHoistedChecksForCountUpLoop(CompilationUnit *cUnit, MIR *mir)
{
    /*
     * NOTE: these synthesized blocks don't have ssa names assigned
     * for Dalvik registers.  However, because they dominate the following
     * blocks we can simply use the Dalvik name w/ subscript 0 as the
     * ssa name.
     */
    DecodedInstruction *dInsn = &mir->dalvikInsn;
    const int lenOffset = offsetof(ArrayObject, length);
    const int maxC = dInsn->arg[0];
    const int minC = dInsn->arg[1];
    int regLength;
    RegLocation rlArray = cUnit->regLocation[mir->dalvikInsn.vA];
    RegLocation rlIdxEnd = cUnit->regLocation[mir->dalvikInsn.vC];

    /* regArray <- arrayRef */
    rlArray = loadValue(cUnit, rlArray, kCoreReg);
    rlIdxEnd = loadValue(cUnit, rlIdxEnd, kCoreReg);
    genRegImmCheck(cUnit, kArmCondEq, rlArray.lowReg, 0, 0,
                   (ArmLIR *) cUnit->loopAnalysis->branchToPCR);

    /* regLength <- len(arrayRef) */
    regLength = allocTemp(cUnit);
    loadWordDisp(cUnit, rlArray.lowReg, lenOffset, regLength);

    int delta = maxC;
    /*
     * If the loop end condition is ">=" instead of ">", then the largest value
     * of the index is "endCondition - 1".
     */
    if (dInsn->arg[2] == OP_IF_GE) {
        delta--;
    }

    if (delta) {
        int tReg = allocTemp(cUnit);
        opRegRegImm(cUnit, kOpAdd, tReg, rlIdxEnd.lowReg, delta);
        rlIdxEnd.lowReg = tReg;
        freeTemp(cUnit, tReg);
    }
    /* Punt if "regIdxEnd < len(Array)" is false */
    genRegRegCheck(cUnit, kArmCondGe, rlIdxEnd.lowReg, regLength, 0,
                   (ArmLIR *) cUnit->loopAnalysis->branchToPCR);
}

/*
 * vA = arrayReg;
 * vB = idxReg;
 * vC = endConditionReg;
 * arg[0] = maxC
 * arg[1] = minC
 * arg[2] = loopBranchConditionCode
 */
static void genHoistedChecksForCountDownLoop(CompilationUnit *cUnit, MIR *mir)
{
    DecodedInstruction *dInsn = &mir->dalvikInsn;
    const int lenOffset = offsetof(ArrayObject, length);
    const int regLength = allocTemp(cUnit);
    const int maxC = dInsn->arg[0];
    const int minC = dInsn->arg[1];
    RegLocation rlArray = cUnit->regLocation[mir->dalvikInsn.vA];
    RegLocation rlIdxInit = cUnit->regLocation[mir->dalvikInsn.vB];

    /* regArray <- arrayRef */
    rlArray = loadValue(cUnit, rlArray, kCoreReg);
    rlIdxInit = loadValue(cUnit, rlIdxInit, kCoreReg);
    genRegImmCheck(cUnit, kArmCondEq, rlArray.lowReg, 0, 0,
                   (ArmLIR *) cUnit->loopAnalysis->branchToPCR);

    /* regLength <- len(arrayRef) */
    loadWordDisp(cUnit, rlArray.lowReg, lenOffset, regLength);

    if (maxC) {
        int tReg = allocTemp(cUnit);
        opRegRegImm(cUnit, kOpAdd, tReg, rlIdxInit.lowReg, maxC);
        rlIdxInit.lowReg = tReg;
        freeTemp(cUnit, tReg);
    }

    /* Punt if "regIdxInit < len(Array)" is false */
    genRegRegCheck(cUnit, kArmCondGe, rlIdxInit.lowReg, regLength, 0,
                   (ArmLIR *) cUnit->loopAnalysis->branchToPCR);
}

/*
 * vA = idxReg;
 * vB = minC;
 */
static void genHoistedLowerBoundCheck(CompilationUnit *cUnit, MIR *mir)
{
    DecodedInstruction *dInsn = &mir->dalvikInsn;
    const int minC = dInsn->vB;
    RegLocation rlIdx = cUnit->regLocation[mir->dalvikInsn.vA];

    /* regIdx <- initial index value */
    rlIdx = loadValue(cUnit, rlIdx, kCoreReg);

    /* Punt if "regIdxInit + minC >= 0" is false */
    genRegImmCheck(cUnit, kArmCondLt, rlIdx.lowReg, -minC, 0,
                   (ArmLIR *) cUnit->loopAnalysis->branchToPCR);
}

/* Extended MIR instructions like PHI */
static void handleExtendedMIR(CompilationUnit *cUnit, MIR *mir)
{
    int opOffset = mir->dalvikInsn.opCode - kMirOpFirst;
    char *msg = dvmCompilerNew(strlen(extendedMIROpNames[opOffset]) + 1,
                               false);
    strcpy(msg, extendedMIROpNames[opOffset]);
    newLIR1(cUnit, kArmPseudoExtended, (int) msg);

    switch (mir->dalvikInsn.opCode) {
        case kMirOpPhi: {
            char *ssaString = dvmCompilerGetSSAString(cUnit, mir->ssaRep);
            newLIR1(cUnit, kArmPseudoSSARep, (int) ssaString);
            break;
        }
        case kMirOpNullNRangeUpCheck: {
            genHoistedChecksForCountUpLoop(cUnit, mir);
            break;
        }
        case kMirOpNullNRangeDownCheck: {
            genHoistedChecksForCountDownLoop(cUnit, mir);
            break;
        }
        case kMirOpLowerBound: {
            genHoistedLowerBoundCheck(cUnit, mir);
            break;
        }
        case kMirOpPunt: {
            genUnconditionalBranch(cUnit,
                                   (ArmLIR *) cUnit->loopAnalysis->branchToPCR);
            break;
        }
        default:
            break;
    }
}

/*
 * Create a PC-reconstruction cell for the starting offset of this trace.
 * Since the PCR cell is placed near the end of the compiled code which is
 * usually out of range for a conditional branch, we put two branches (one
 * branch over to the loop body and one layover branch to the actual PCR) at the
 * end of the entry block.
 */
static void setupLoopEntryBlock(CompilationUnit *cUnit, BasicBlock *entry,
                                ArmLIR *bodyLabel)
{
    /* Set up the place holder to reconstruct this Dalvik PC */
    ArmLIR *pcrLabel = dvmCompilerNew(sizeof(ArmLIR), true);
    pcrLabel->opCode = ARM_PSEUDO_kPCReconstruction_CELL;
    pcrLabel->operands[0] =
        (int) (cUnit->method->insns + entry->startOffset);
    pcrLabel->operands[1] = entry->startOffset;
    /* Insert the place holder to the growable list */
    dvmInsertGrowableList(&cUnit->pcReconstructionList, pcrLabel);

    /*
     * Next, create two branches - one branch over to the loop body and the
     * other branch to the PCR cell to punt.
     */
    ArmLIR *branchToBody = dvmCompilerNew(sizeof(ArmLIR), true);
    branchToBody->opCode = kThumbBUncond;
    branchToBody->generic.target = (LIR *) bodyLabel;
    setupResourceMasks(branchToBody);
    cUnit->loopAnalysis->branchToBody = (LIR *) branchToBody;

    ArmLIR *branchToPCR = dvmCompilerNew(sizeof(ArmLIR), true);
    branchToPCR->opCode = kThumbBUncond;
    branchToPCR->generic.target = (LIR *) pcrLabel;
    setupResourceMasks(branchToPCR);
    cUnit->loopAnalysis->branchToPCR = (LIR *) branchToPCR;
}

void dvmCompilerMIR2LIR(CompilationUnit *cUnit)
{
    /* Used to hold the labels of each block */
    ArmLIR *labelList =
        dvmCompilerNew(sizeof(ArmLIR) * cUnit->numBlocks, true);
    GrowableList chainingListByType[kChainingCellLast];
    int i;

    /*
     * Initialize various types chaining lists.
     */
    for (i = 0; i < kChainingCellLast; i++) {
        dvmInitGrowableList(&chainingListByType[i], 2);
    }

    BasicBlock **blockList = cUnit->blockList;

    if (cUnit->executionCount) {
        /*
         * Reserve 6 bytes at the beginning of the trace
         *        +----------------------------+
         *        | execution count (4 bytes)  |
         *        +----------------------------+
         *        | chain cell offset (2 bytes)|
         *        +----------------------------+
         * ...and then code to increment the execution
         * count:
         *       mov   r0, pc       @ move adr of "mov r0,pc" + 4 to r0
         *       sub   r0, #10      @ back up to addr of executionCount
         *       ldr   r1, [r0]
         *       add   r1, #1
         *       str   r1, [r0]
         */
        newLIR1(cUnit, kArm16BitData, 0);
        newLIR1(cUnit, kArm16BitData, 0);
        cUnit->chainCellOffsetLIR =
            (LIR *) newLIR1(cUnit, kArm16BitData, CHAIN_CELL_OFFSET_TAG);
        cUnit->headerSize = 6;
        /* Thumb instruction used directly here to ensure correct size */
        newLIR2(cUnit, kThumbMovRR_H2L, r0, rpc);
        newLIR2(cUnit, kThumbSubRI8, r0, 10);
        newLIR3(cUnit, kThumbLdrRRI5, r1, r0, 0);
        newLIR2(cUnit, kThumbAddRI8, r1, 1);
        newLIR3(cUnit, kThumbStrRRI5, r1, r0, 0);
    } else {
         /* Just reserve 2 bytes for the chain cell offset */
        cUnit->chainCellOffsetLIR =
            (LIR *) newLIR1(cUnit, kArm16BitData, CHAIN_CELL_OFFSET_TAG);
        cUnit->headerSize = 2;
    }

    /* Handle the content in each basic block */
    for (i = 0; i < cUnit->numBlocks; i++) {
        blockList[i]->visited = true;
        MIR *mir;

        labelList[i].operands[0] = blockList[i]->startOffset;

        if (blockList[i]->blockType >= kChainingCellLast) {
            /*
             * Append the label pseudo LIR first. Chaining cells will be handled
             * separately afterwards.
             */
            dvmCompilerAppendLIR(cUnit, (LIR *) &labelList[i]);
        }

        if (blockList[i]->blockType == kEntryBlock) {
            labelList[i].opCode = ARM_PSEUDO_kEntryBlock;
            if (blockList[i]->firstMIRInsn == NULL) {
                continue;
            } else {
              setupLoopEntryBlock(cUnit, blockList[i],
                                  &labelList[blockList[i]->fallThrough->id]);
            }
        } else if (blockList[i]->blockType == kExitBlock) {
            labelList[i].opCode = ARM_PSEUDO_kExitBlock;
            goto gen_fallthrough;
        } else if (blockList[i]->blockType == kDalvikByteCode) {
            labelList[i].opCode = kArmPseudoNormalBlockLabel;
            /* Reset the register state */
            resetRegPool(cUnit);
            clobberAllRegs(cUnit);
            resetNullCheckTracker(cUnit);
        } else {
            switch (blockList[i]->blockType) {
                case kChainingCellNormal:
                    labelList[i].opCode = ARM_PSEUDO_kChainingCellNormal;
                    /* handle the codegen later */
                    dvmInsertGrowableList(
                        &chainingListByType[kChainingCellNormal], (void *) i);
                    break;
                case kChainingCellInvokeSingleton:
                    labelList[i].opCode =
                        ARM_PSEUDO_kChainingCellInvokeSingleton;
                    labelList[i].operands[0] =
                        (int) blockList[i]->containingMethod;
                    /* handle the codegen later */
                    dvmInsertGrowableList(
                        &chainingListByType[kChainingCellInvokeSingleton],
                        (void *) i);
                    break;
                case kChainingCellInvokePredicted:
                    labelList[i].opCode =
                        ARM_PSEUDO_kChainingCellInvokePredicted;
                    /* handle the codegen later */
                    dvmInsertGrowableList(
                        &chainingListByType[kChainingCellInvokePredicted],
                        (void *) i);
                    break;
                case kChainingCellHot:
                    labelList[i].opCode =
                        ARM_PSEUDO_kChainingCellHot;
                    /* handle the codegen later */
                    dvmInsertGrowableList(
                        &chainingListByType[kChainingCellHot],
                        (void *) i);
                    break;
                case kPCReconstruction:
                    /* Make sure exception handling block is next */
                    labelList[i].opCode =
                        ARM_PSEUDO_kPCReconstruction_BLOCK_LABEL;
                    assert (i == cUnit->numBlocks - 2);
                    handlePCReconstruction(cUnit, &labelList[i+1]);
                    break;
                case kExceptionHandling:
                    labelList[i].opCode = kArmPseudoEHBlockLabel;
                    if (cUnit->pcReconstructionList.numUsed) {
                        loadWordDisp(cUnit, rGLUE, offsetof(InterpState,
                                     jitToInterpEntries.dvmJitToInterpPunt),
                                     r1);
                        opReg(cUnit, kOpBlx, r1);
                    }
                    break;
#if defined(WITH_SELF_VERIFICATION) || defined(WITH_JIT_TUNING)
                case kChainingCellBackwardBranch:
                    labelList[i].opCode =
                        ARM_PSEUDO_kChainingCellBackwardBranch;
                    /* handle the codegen later */
                    dvmInsertGrowableList(
                        &chainingListByType[kChainingCellBackwardBranch],
                        (void *) i);
                    break;
#endif
                default:
                    break;
            }
            continue;
        }

        ArmLIR *headLIR = NULL;

        for (mir = blockList[i]->firstMIRInsn; mir; mir = mir->next) {

            resetRegPool(cUnit);
            if (gDvmJit.disableOpt & (1 << kTrackLiveTemps)) {
                clobberAllRegs(cUnit);
            }

            if (gDvmJit.disableOpt & (1 << kSuppressLoads)) {
                resetDefTracking(cUnit);
            }

            if (mir->dalvikInsn.opCode >= kMirOpFirst) {
                handleExtendedMIR(cUnit, mir);
                continue;
            }


            OpCode dalvikOpCode = mir->dalvikInsn.opCode;
            InstructionFormat dalvikFormat =
                dexGetInstrFormat(gDvm.instrFormat, dalvikOpCode);
            ArmLIR *boundaryLIR =
                newLIR2(cUnit, ARM_PSEUDO_kDalvikByteCode_BOUNDARY,
                        mir->offset,
                        (int) dvmCompilerGetDalvikDisassembly(&mir->dalvikInsn)
                       );
            if (mir->ssaRep) {
                char *ssaString = dvmCompilerGetSSAString(cUnit, mir->ssaRep);
                newLIR1(cUnit, kArmPseudoSSARep, (int) ssaString);
            }

            /* Remember the first LIR for this block */
            if (headLIR == NULL) {
                headLIR = boundaryLIR;
                /* Set the first boundaryLIR as a scheduling barrier */
                headLIR->defMask = ENCODE_ALL;
            }

            bool notHandled;
            /*
             * Debugging: screen the opcode first to see if it is in the
             * do[-not]-compile list
             */
            bool singleStepMe =
                gDvmJit.includeSelectedOp !=
                ((gDvmJit.opList[dalvikOpCode >> 3] &
                  (1 << (dalvikOpCode & 0x7))) !=
                 0);
#if defined(WITH_SELF_VERIFICATION)
            /* Punt on opcodes we can't replay */
            if (selfVerificationPuntOps(dalvikOpCode))
                singleStepMe = true;
#endif
            if (singleStepMe || cUnit->allSingleStep) {
                notHandled = false;
                genInterpSingleStep(cUnit, mir);
            } else {
                opcodeCoverage[dalvikOpCode]++;
                switch (dalvikFormat) {
                    case kFmt10t:
                    case kFmt20t:
                    case kFmt30t:
                        notHandled = handleFmt10t_Fmt20t_Fmt30t(cUnit,
                                  mir, blockList[i], labelList);
                        break;
                    case kFmt10x:
                        notHandled = handleFmt10x(cUnit, mir);
                        break;
                    case kFmt11n:
                    case kFmt31i:
                        notHandled = handleFmt11n_Fmt31i(cUnit, mir);
                        break;
                    case kFmt11x:
                        notHandled = handleFmt11x(cUnit, mir);
                        break;
                    case kFmt12x:
                        notHandled = handleFmt12x(cUnit, mir);
                        break;
                    case kFmt20bc:
                        notHandled = handleFmt20bc(cUnit, mir);
                        break;
                    case kFmt21c:
                    case kFmt31c:
                        notHandled = handleFmt21c_Fmt31c(cUnit, mir);
                        break;
                    case kFmt21h:
                        notHandled = handleFmt21h(cUnit, mir);
                        break;
                    case kFmt21s:
                        notHandled = handleFmt21s(cUnit, mir);
                        break;
                    case kFmt21t:
                        notHandled = handleFmt21t(cUnit, mir, blockList[i],
                                                  labelList);
                        break;
                    case kFmt22b:
                    case kFmt22s:
                        notHandled = handleFmt22b_Fmt22s(cUnit, mir);
                        break;
                    case kFmt22c:
                        notHandled = handleFmt22c(cUnit, mir);
                        break;
                    case kFmt22cs:
                        notHandled = handleFmt22cs(cUnit, mir);
                        break;
                    case kFmt22t:
                        notHandled = handleFmt22t(cUnit, mir, blockList[i],
                                                  labelList);
                        break;
                    case kFmt22x:
                    case kFmt32x:
                        notHandled = handleFmt22x_Fmt32x(cUnit, mir);
                        break;
                    case kFmt23x:
                        notHandled = handleFmt23x(cUnit, mir);
                        break;
                    case kFmt31t:
                        notHandled = handleFmt31t(cUnit, mir);
                        break;
                    case kFmt3rc:
                    case kFmt35c:
                        notHandled = handleFmt35c_3rc(cUnit, mir, blockList[i],
                                                      labelList);
                        break;
                    case kFmt3rms:
                    case kFmt35ms:
                        notHandled = handleFmt35ms_3rms(cUnit, mir,blockList[i],
                                                        labelList);
                        break;
                    case kFmt3inline:
                        notHandled = handleFmt3inline(cUnit, mir);
                        break;
                    case kFmt51l:
                        notHandled = handleFmt51l(cUnit, mir);
                        break;
                    default:
                        notHandled = true;
                        break;
                }
            }
            if (notHandled) {
                LOGE("%#06x: Opcode 0x%x (%s) / Fmt %d not handled\n",
                     mir->offset,
                     dalvikOpCode, getOpcodeName(dalvikOpCode),
                     dalvikFormat);
                dvmAbort();
                break;
            }
        }

        if (blockList[i]->blockType == kEntryBlock) {
            dvmCompilerAppendLIR(cUnit,
                                 (LIR *) cUnit->loopAnalysis->branchToBody);
            dvmCompilerAppendLIR(cUnit,
                                 (LIR *) cUnit->loopAnalysis->branchToPCR);
        }

        if (headLIR) {
            /*
             * Eliminate redundant loads/stores and delay stores into later
             * slots
             */
            dvmCompilerApplyLocalOptimizations(cUnit, (LIR *) headLIR,
                                               cUnit->lastLIRInsn);
        }

gen_fallthrough:
        /*
         * Check if the block is terminated due to trace length constraint -
         * insert an unconditional branch to the chaining cell.
         */
        if (blockList[i]->needFallThroughBranch) {
            genUnconditionalBranch(cUnit,
                                   &labelList[blockList[i]->fallThrough->id]);
        }

    }

    /* Handle the chaining cells in predefined order */
    for (i = 0; i < kChainingCellLast; i++) {
        size_t j;
        int *blockIdList = (int *) chainingListByType[i].elemList;

        cUnit->numChainingCells[i] = chainingListByType[i].numUsed;

        /* No chaining cells of this type */
        if (cUnit->numChainingCells[i] == 0)
            continue;

        /* Record the first LIR for a new type of chaining cell */
        cUnit->firstChainingLIR[i] = (LIR *) &labelList[blockIdList[0]];

        for (j = 0; j < chainingListByType[i].numUsed; j++) {
            int blockId = blockIdList[j];

            /* Align this chaining cell first */
            newLIR0(cUnit, kArmPseudoPseudoAlign4);

            /* Insert the pseudo chaining instruction */
            dvmCompilerAppendLIR(cUnit, (LIR *) &labelList[blockId]);


            switch (blockList[blockId]->blockType) {
                case kChainingCellNormal:
                    handleNormalChainingCell(cUnit,
                      blockList[blockId]->startOffset);
                    break;
                case kChainingCellInvokeSingleton:
                    handleInvokeSingletonChainingCell(cUnit,
                        blockList[blockId]->containingMethod);
                    break;
                case kChainingCellInvokePredicted:
                    handleInvokePredictedChainingCell(cUnit);
                    break;
                case kChainingCellHot:
                    handleHotChainingCell(cUnit,
                        blockList[blockId]->startOffset);
                    break;
#if defined(WITH_SELF_VERIFICATION) || defined(WITH_JIT_TUNING)
                case kChainingCellBackwardBranch:
                    handleBackwardBranchChainingCell(cUnit,
                        blockList[blockId]->startOffset);
                    break;
#endif
                default:
                    LOGE("Bad blocktype %d", blockList[blockId]->blockType);
                    dvmAbort();
                    break;
            }
        }
    }

    dvmCompilerApplyGlobalOptimizations(cUnit);
}

/* Accept the work and start compiling */
bool dvmCompilerDoWork(CompilerWorkOrder *work)
{
    bool res;

    if (gDvmJit.codeCacheFull) {
        return false;
    }

    switch (work->kind) {
        case kWorkOrderMethod:
            res = dvmCompileMethod(work->info, &work->result);
            break;
        case kWorkOrderTrace:
            /* Start compilation with maximally allowed trace length */
            res = dvmCompileTrace(work->info, JIT_MAX_TRACE_LEN, &work->result);
            break;
        case kWorkOrderTraceDebug: {
            bool oldPrintMe = gDvmJit.printMe;
            gDvmJit.printMe = true;
            /* Start compilation with maximally allowed trace length */
            res = dvmCompileTrace(work->info, JIT_MAX_TRACE_LEN, &work->result);
            gDvmJit.printMe = oldPrintMe;;
            break;
        }
        default:
            res = false;
            dvmAbort();
    }
    return res;
}

/* Architectural-specific debugging helpers go here */
void dvmCompilerArchDump(void)
{
    /* Print compiled opcode in this VM instance */
    int i, start, streak;
    char buf[1024];

    streak = i = 0;
    buf[0] = 0;
    while (opcodeCoverage[i] == 0 && i < 256) {
        i++;
    }
    if (i == 256) {
        return;
    }
    for (start = i++, streak = 1; i < 256; i++) {
        if (opcodeCoverage[i]) {
            streak++;
        } else {
            if (streak == 1) {
                sprintf(buf+strlen(buf), "%x,", start);
            } else {
                sprintf(buf+strlen(buf), "%x-%x,", start, start + streak - 1);
            }
            streak = 0;
            while (opcodeCoverage[i] == 0 && i < 256) {
                i++;
            }
            if (i < 256) {
                streak = 1;
                start = i;
            }
        }
    }
    if (streak) {
        if (streak == 1) {
            sprintf(buf+strlen(buf), "%x", start);
        } else {
            sprintf(buf+strlen(buf), "%x-%x", start, start + streak - 1);
        }
    }
    if (strlen(buf)) {
        LOGD("dalvik.vm.jit.op = %s", buf);
    }
}

/* Common initialization routine for an architecture family */
bool dvmCompilerArchInit()
{
    int i;

    for (i = 0; i < kArmLast; i++) {
        if (EncodingMap[i].opCode != i) {
            LOGE("Encoding order for %s is wrong: expecting %d, seeing %d",
                 EncodingMap[i].name, i, EncodingMap[i].opCode);
            dvmAbort();
        }
    }

    return compilerArchVariantInit();
}