Added basic VFP support to Dalvik interpreter.

This adds opcode handlers for all instructions that have a VFP equivalent
or (for the compare operations) near-equivalent.

The inclusion of VFP is keyed off of TARGET_ARCH_VERSION, so enabling
this requires setting "TARGET_ARCH_VERSION=armv5te-vfp" in a buildspec.
diff --git a/vm/Android.mk b/vm/Android.mk
index 2ef1ec9..cb6a971 100644
--- a/vm/Android.mk
+++ b/vm/Android.mk
@@ -232,27 +232,15 @@
 endif
 
 ifeq ($(TARGET_ARCH),arm)
+  #TARGET_ARCH_VERSION := armv5te-vfp
   LOCAL_SRC_FILES += \
 		arch/arm/CallOldABI.S \
 		arch/arm/CallEABI.S \
 		arch/arm/HintsEABI.c
-  #
-  # The armv4 configation in mterp is actually armv4t, it's just
-  # wrongly named
-  #
-  # TODO: Rename mterp/config-armv4 and mterp/armv4 (to armv4t)
-  #       then remove this ifeq.
-  #
-  ifeq ($(TARGET_ARCH_VERSION),armv4t)
-    LOCAL_SRC_FILES += \
-		mterp/out/InterpC-armv4.c.arm \
-		mterp/out/InterpAsm-armv4.S
-  else
-    # Select architecture specific sources (armv4,armv5te etc)
-    LOCAL_SRC_FILES += \
+  # Select architecture specific sources (armv4t,armv5te etc)
+  LOCAL_SRC_FILES += \
 		mterp/out/InterpC-$(TARGET_ARCH_VERSION).c.arm \
 		mterp/out/InterpAsm-$(TARGET_ARCH_VERSION).S
-  endif
   LOCAL_SHARED_LIBRARIES += libdl
 else
   ifeq ($(TARGET_ARCH),x86)
diff --git a/vm/mterp/armv5te/debug.c b/vm/mterp/armv5te/debug.c
index 301e27a..22fbd6d 100644
--- a/vm/mterp/armv5te/debug.c
+++ b/vm/mterp/armv5te/debug.c
@@ -63,12 +63,12 @@
      * It is a direct (non-virtual) method if it is static, private,
      * or a constructor.
      */
-    bool isDirect = 
+    bool isDirect =
         ((method->accessFlags & (ACC_STATIC|ACC_PRIVATE)) != 0) ||
         (method->name[0] == '<');
 
     char* desc = dexProtoCopyMethodDescriptor(&method->prototype);
-        
+
     printf("<%c:%s.%s %s> ",
             isDirect ? 'D' : 'V',
             method->clazz->descriptor,
diff --git a/vm/mterp/armv5te/header.S b/vm/mterp/armv5te/header.S
index 2e5c6ed..586418f 100644
--- a/vm/mterp/armv5te/header.S
+++ b/vm/mterp/armv5te/header.S
@@ -33,7 +33,9 @@
 r0 holds returns of <= 4 bytes
 r0-r1 hold returns of 8 bytes, low word in r0
 
-Callee must save/restore r4+ (except r12) if it modifies them.
+Callee must save/restore r4+ (except r12) if it modifies them.  If VFP
+is present, registers s16-s31 (a/k/a d8-d15, a/k/a q4-q7) must be preserved,
+s0-s15 (d0-d7, q0-a3) do not need to be.
 
 Stack is "full descending".  Only the arguments that don't fit in the first 4
 registers are placed on the stack.  "sp" points at the first stacked argument
@@ -174,6 +176,12 @@
 #define SET_VREG(_reg, _vreg)   str     _reg, [rFP, _vreg, lsl #2]
 
 /*
+ * Convert a virtual register index into an address.
+ */
+#define VREG_INDEX_TO_ADDR(_reg, _vreg) \
+        add     _reg, rFP, _vreg, lsl #2
+
+/*
  * This is a #include, not a %include, because we want the C pre-processor
  * to expand the macros into assembler assignment statements.
  */
diff --git a/vm/mterp/config-armv5te-vfp b/vm/mterp/config-armv5te-vfp
new file mode 100644
index 0000000..863f09f
--- /dev/null
+++ b/vm/mterp/config-armv5te-vfp
@@ -0,0 +1,100 @@
+# 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.
+
+#
+# Configuration for ARMv5TE targets with VFP support.
+#
+# This is just ARMv5TE with replacements for the handlers that can benefit
+# from floating-point instructions.  Essentially all float/double
+# operations except for "remainder" and conversions to/from 64-bit ints.
+#
+
+handler-size 64
+
+# source for the instruction table stub
+asm-stub armv5te/stub.S
+
+# file header and basic definitions
+import c/header.c
+import armv5te/header.S
+
+# C pre-processor defines for stub C instructions
+import cstubs/stubdefs.c
+
+# highly-platform-specific defs
+import armv5te/platform.S
+
+# common defs for the C helpers; include this before the instruction handlers
+import c/opcommon.c
+
+# arch-specific entry point to interpreter
+import armv5te/entry.S
+
+# opcode list; argument to op-start is default directory
+op-start armv5te
+    op OP_ADD_DOUBLE vfp
+    op OP_ADD_DOUBLE_2ADDR vfp
+    op OP_ADD_FLOAT vfp
+    op OP_ADD_FLOAT_2ADDR vfp
+    op OP_CMPG_DOUBLE vfp
+    op OP_CMPG_FLOAT vfp
+    op OP_CMPL_DOUBLE vfp
+    op OP_CMPL_FLOAT vfp
+    op OP_DIV_DOUBLE vfp
+    op OP_DIV_DOUBLE_2ADDR vfp
+    op OP_DIV_FLOAT vfp
+    op OP_DIV_FLOAT_2ADDR vfp
+    op OP_DOUBLE_TO_FLOAT vfp
+    op OP_DOUBLE_TO_INT vfp
+    op OP_FLOAT_TO_DOUBLE vfp
+    op OP_FLOAT_TO_INT vfp
+    op OP_INT_TO_DOUBLE vfp
+    op OP_INT_TO_FLOAT vfp
+    op OP_MUL_DOUBLE vfp
+    op OP_MUL_DOUBLE_2ADDR vfp
+    op OP_MUL_FLOAT vfp
+    op OP_MUL_FLOAT_2ADDR vfp
+    op OP_SUB_DOUBLE vfp
+    op OP_SUB_DOUBLE_2ADDR vfp
+    op OP_SUB_FLOAT vfp
+    op OP_SUB_FLOAT_2ADDR vfp
+
+    # use trivial integer operation
+    #op OP_NEG_DOUBLE armv5te
+    #op OP_NEG_FLOAT armv5te
+
+    # use __aeabi_* functions
+    #op OP_DOUBLE_TO_LONG armv5te
+    #op OP_FLOAT_TO_LONG armv5te
+    #op OP_LONG_TO_DOUBLE armv5te
+    #op OP_LONG_TO_FLOAT armv5te
+
+    # no "remainder" op in vfp or libgcc.a; use libc function
+    #op OP_REM_DOUBLE armv5te
+    #op OP_REM_DOUBLE_2ADDR armv5te
+    #op OP_REM_FLOAT armv5te
+    #op OP_REM_FLOAT_2ADDR armv5te
+op-end
+
+# "helper" code for C; include if you use any of the C stubs (this generates
+# object code, so it's normally excluded)
+##import c/gotoTargets.c
+
+# end of defs; include this when cstubs/stubdefs.c is included
+import cstubs/enddefs.c
+
+# common subroutines for asm
+import armv5te/footer.S
+import armv5te/debug.c
+
diff --git a/vm/mterp/out/InterpAsm-armv4t.S b/vm/mterp/out/InterpAsm-armv4t.S
index b231a30..a348e31 100644
--- a/vm/mterp/out/InterpAsm-armv4t.S
+++ b/vm/mterp/out/InterpAsm-armv4t.S
@@ -40,7 +40,9 @@
 r0 holds returns of <= 4 bytes
 r0-r1 hold returns of 8 bytes, low word in r0
 
-Callee must save/restore r4+ (except r12) if it modifies them.
+Callee must save/restore r4+ (except r12) if it modifies them.  If VFP
+is present, registers s16-s31 (a/k/a d8-d15, a/k/a q4-q7) must be preserved,
+s0-s15 (d0-d7, q0-a3) do not need to be.
 
 Stack is "full descending".  Only the arguments that don't fit in the first 4
 registers are placed on the stack.  "sp" points at the first stacked argument
@@ -181,6 +183,12 @@
 #define SET_VREG(_reg, _vreg)   str     _reg, [rFP, _vreg, lsl #2]
 
 /*
+ * Convert a virtual register index into an address.
+ */
+#define VREG_INDEX_TO_ADDR(_reg, _vreg) \
+        add     _reg, rFP, _vreg, lsl #2
+
+/*
  * This is a #include, not a %include, because we want the C pre-processor
  * to expand the macros into assembler assignment statements.
  */
diff --git a/vm/mterp/out/InterpAsm-armv5te-vfp.S b/vm/mterp/out/InterpAsm-armv5te-vfp.S
new file mode 100644
index 0000000..aecd11d
--- /dev/null
+++ b/vm/mterp/out/InterpAsm-armv5te-vfp.S
@@ -0,0 +1,9544 @@
+/*
+ * This file was generated automatically by gen-mterp.py for 'armv5te-vfp'.
+ *
+ * --> DO NOT EDIT <--
+ */
+
+/* File: armv5te/header.S */
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ * ARMv5 definitions and declarations.
+ */
+
+/*
+ARM EABI general notes:
+
+r0-r3 hold first 4 args to a method; they are not preserved across method calls
+r4-r8 are available for general use
+r9 is given special treatment in some situations, but not for us
+r10 (sl) seems to be generally available
+r11 (fp) is used by gcc (unless -fomit-frame-pointer is set)
+r12 (ip) is scratch -- not preserved across method calls
+r13 (sp) should be managed carefully in case a signal arrives
+r14 (lr) must be preserved
+r15 (pc) can be tinkered with directly
+
+r0 holds returns of <= 4 bytes
+r0-r1 hold returns of 8 bytes, low word in r0
+
+Callee must save/restore r4+ (except r12) if it modifies them.  If VFP
+is present, registers s16-s31 (a/k/a d8-d15, a/k/a q4-q7) must be preserved,
+s0-s15 (d0-d7, q0-a3) do not need to be.
+
+Stack is "full descending".  Only the arguments that don't fit in the first 4
+registers are placed on the stack.  "sp" points at the first stacked argument
+(i.e. the 5th arg).
+
+VFP: single-precision results in s0, double-precision results in d0.
+
+In the EABI, "sp" must be 64-bit aligned on entry to a function, and any
+64-bit quantities (long long, double) must be 64-bit aligned.
+*/
+
+/*
+Mterp and ARM notes:
+
+The following registers have fixed assignments:
+
+  reg nick      purpose
+  r4  rPC       interpreted program counter, used for fetching instructions
+  r5  rFP       interpreted frame pointer, used for accessing locals and args
+  r6  rGLUE     MterpGlue pointer
+  r7  rIBASE    interpreted instruction base pointer, used for computed goto
+  r8  rINST     first 16-bit code unit of current instruction
+
+Macros are provided for common operations.  Each macro MUST emit only
+one instruction to make instruction-counting easier.  They MUST NOT alter
+unspecified registers or condition codes.
+*/
+
+/* single-purpose registers, given names for clarity */
+#define rPC     r4
+#define rFP     r5
+#define rGLUE   r6
+#define rIBASE  r7
+#define rINST   r8
+
+/* save/restore the PC and/or FP from the glue struct */
+#define LOAD_PC_FROM_GLUE()     ldr     rPC, [rGLUE, #offGlue_pc]
+#define SAVE_PC_TO_GLUE()       str     rPC, [rGLUE, #offGlue_pc]
+#define LOAD_FP_FROM_GLUE()     ldr     rFP, [rGLUE, #offGlue_fp]
+#define SAVE_FP_TO_GLUE()       str     rFP, [rGLUE, #offGlue_fp]
+#define LOAD_PC_FP_FROM_GLUE()  ldmia   rGLUE, {rPC, rFP}
+#define SAVE_PC_FP_TO_GLUE()    stmia   rGLUE, {rPC, rFP}
+
+/*
+ * "export" the PC to the stack frame, f/b/o future exception objects.  Must
+ * be done *before* something calls dvmThrowException.
+ *
+ * In C this is "SAVEAREA_FROM_FP(fp)->xtra.currentPc = pc", i.e.
+ * fp - sizeof(StackSaveArea) + offsetof(SaveArea, xtra.currentPc)
+ *
+ * It's okay to do this more than once.
+ */
+#define EXPORT_PC() \
+    str     rPC, [rFP, #(-sizeofStackSaveArea + offStackSaveArea_currentPc)]
+
+/*
+ * Given a frame pointer, find the stack save area.
+ *
+ * In C this is "((StackSaveArea*)(_fp) -1)".
+ */
+#define SAVEAREA_FROM_FP(_reg, _fpreg) \
+    sub     _reg, _fpreg, #sizeofStackSaveArea
+
+/*
+ * Fetch the next instruction from rPC into rINST.  Does not advance rPC.
+ */
+#define FETCH_INST()            ldrh    rINST, [rPC]
+
+/*
+ * Fetch the next instruction from the specified offset.  Advances rPC
+ * to point to the next instruction.  "_count" is in 16-bit code units.
+ *
+ * Because of the limited size of immediate constants on ARM, this is only
+ * suitable for small forward movements (i.e. don't try to implement "goto"
+ * with this).
+ *
+ * This must come AFTER anything that can throw an exception, or the
+ * exception catch may miss.  (This also implies that it must come after
+ * EXPORT_PC().)
+ */
+#define FETCH_ADVANCE_INST(_count) ldrh    rINST, [rPC, #(_count*2)]!
+
+/*
+ * The operation performed here is similar to FETCH_ADVANCE_INST, except the
+ * src and dest registers are parameterized (not hard-wired to rPC and rINST).
+ */
+#define PREFETCH_ADVANCE_INST(_dreg, _sreg, _count) \
+        ldrh    _dreg, [_sreg, #(_count*2)]!
+
+/*
+ * Fetch the next instruction from an offset specified by _reg.  Updates
+ * rPC to point to the next instruction.  "_reg" must specify the distance
+ * in bytes, *not* 16-bit code units, and may be a signed value.
+ *
+ * We want to write "ldrh rINST, [rPC, _reg, lsl #2]!", but some of the
+ * bits that hold the shift distance are used for the half/byte/sign flags.
+ * In some cases we can pre-double _reg for free, so we require a byte offset
+ * here.
+ */
+#define FETCH_ADVANCE_INST_RB(_reg) ldrh    rINST, [rPC, _reg]!
+
+/*
+ * Fetch a half-word code unit from an offset past the current PC.  The
+ * "_count" value is in 16-bit code units.  Does not advance rPC.
+ *
+ * The "_S" variant works the same but treats the value as signed.
+ */
+#define FETCH(_reg, _count)     ldrh    _reg, [rPC, #(_count*2)]
+#define FETCH_S(_reg, _count)   ldrsh   _reg, [rPC, #(_count*2)]
+
+/*
+ * Fetch one byte from an offset past the current PC.  Pass in the same
+ * "_count" as you would for FETCH, and an additional 0/1 indicating which
+ * byte of the halfword you want (lo/hi).
+ */
+#define FETCH_B(_reg, _count, _byte) ldrb     _reg, [rPC, #(_count*2+_byte)]
+
+/*
+ * Put the instruction's opcode field into the specified register.
+ */
+#define GET_INST_OPCODE(_reg)   and     _reg, rINST, #255
+
+/*
+ * Put the prefetched instruction's opcode field into the specified register.
+ */
+#define GET_PREFETCHED_OPCODE(_oreg, _ireg)   and     _oreg, _ireg, #255
+
+/*
+ * Begin executing the opcode in _reg.  Because this only jumps within the
+ * interpreter, we don't have to worry about pre-ARMv5 THUMB interwork.
+ */
+#define GOTO_OPCODE(_reg)       add     pc, rIBASE, _reg, lsl #6
+
+/*
+ * Get/set the 32-bit value from a Dalvik register.
+ */
+#define GET_VREG(_reg, _vreg)   ldr     _reg, [rFP, _vreg, lsl #2]
+#define SET_VREG(_reg, _vreg)   str     _reg, [rFP, _vreg, lsl #2]
+
+/*
+ * Convert a virtual register index into an address.
+ */
+#define VREG_INDEX_TO_ADDR(_reg, _vreg) \
+        add     _reg, rFP, _vreg, lsl #2
+
+/*
+ * This is a #include, not a %include, because we want the C pre-processor
+ * to expand the macros into assembler assignment statements.
+ */
+#include "../common/asm-constants.h"
+
+
+/* File: armv5te/platform.S */
+/*
+ * ===========================================================================
+ *  CPU-version-specific defines
+ * ===========================================================================
+ */
+
+/*
+ * Macro for "LDR PC,xxx", which is not allowed pre-ARMv5.  Essentially a
+ * one-way branch.
+ *
+ * May modify IP.  Does not modify LR.
+ */
+.macro  LDR_PC source
+    ldr     pc, \source
+.endm
+
+/*
+ * Macro for "MOV LR,PC / LDR PC,xxx", which is not allowed pre-ARMv5.
+ * Jump to subroutine.
+ *
+ * May modify IP and LR.
+ */
+.macro  LDR_PC_LR source
+    mov     lr, pc
+    ldr     pc, \source
+.endm
+
+/*
+ * Macro for "LDMFD SP!, {...regs...,PC}".
+ *
+ * May modify IP and LR.
+ */
+.macro  LDMFD_PC regs
+    ldmfd   sp!, {\regs,pc}
+.endm
+
+
+/* File: armv5te/entry.S */
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ * Interpreter entry point.
+ */
+
+/*
+ * We don't have formal stack frames, so gdb scans upward in the code
+ * to find the start of the function (a label with the %function type),
+ * and then looks at the next few instructions to figure out what
+ * got pushed onto the stack.  From this it figures out how to restore
+ * the registers, including PC, for the previous stack frame.  If gdb
+ * sees a non-function label, it stops scanning, so either we need to
+ * have nothing but assembler-local labels between the entry point and
+ * the break, or we need to fake it out.
+ *
+ * When this is defined, we add some stuff to make gdb less confused.
+ */
+#define ASSIST_DEBUGGER 1
+
+    .text
+    .align  2
+    .global dvmMterpStdRun
+    .type   dvmMterpStdRun, %function
+
+/*
+ * On entry:
+ *  r0  MterpGlue* glue
+ *
+ * This function returns a boolean "changeInterp" value.  The return comes
+ * via a call to dvmMterpStdBail().
+ */
+dvmMterpStdRun:
+#define MTERP_ENTRY1 \
+    .save {r4-r10,fp,lr}; \
+    stmfd   sp!, {r4-r10,fp,lr}         @ save 9 regs
+#define MTERP_ENTRY2 \
+    .pad    #4; \
+    sub     sp, sp, #4                  @ align 64
+
+    .fnstart
+    MTERP_ENTRY1
+    MTERP_ENTRY2
+
+    /* save stack pointer, add magic word for debuggerd */
+    str     sp, [r0, #offGlue_bailPtr]  @ save SP for eventual return
+
+    /* set up "named" registers, figure out entry point */
+    mov     rGLUE, r0                   @ set rGLUE
+    ldrb    r1, [r0, #offGlue_entryPoint]   @ InterpEntry enum is char
+    LOAD_PC_FP_FROM_GLUE()              @ load rPC and rFP from "glue"
+    adr     rIBASE, dvmAsmInstructionStart  @ set rIBASE
+    cmp     r1, #kInterpEntryInstr      @ usual case?
+    bne     .Lnot_instr                 @ no, handle it
+
+    /* start executing the instruction at rPC */
+    FETCH_INST()                        @ load rINST from rPC
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+.Lnot_instr:
+    cmp     r1, #kInterpEntryReturn     @ were we returning from a method?
+    beq     common_returnFromMethod
+
+.Lnot_return:
+    cmp     r1, #kInterpEntryThrow      @ were we throwing an exception?
+    beq     common_exceptionThrown
+
+.Lbad_arg:
+    ldr     r0, strBadEntryPoint
+    @ r1 holds value of entryPoint
+    bl      printf
+    bl      dvmAbort
+    .fnend
+
+
+    .global dvmMterpStdBail
+    .type   dvmMterpStdBail, %function
+
+/*
+ * Restore the stack pointer and PC from the save point established on entry.
+ * This is essentially the same as a longjmp, but should be cheaper.  The
+ * last instruction causes us to return to whoever called dvmMterpStdRun.
+ *
+ * We pushed some registers on the stack in dvmMterpStdRun, then saved
+ * SP and LR.  Here we restore SP, restore the registers, and then restore
+ * LR to PC.
+ *
+ * On entry:
+ *  r0  MterpGlue* glue
+ *  r1  bool changeInterp
+ */
+dvmMterpStdBail:
+    ldr     sp, [r0, #offGlue_bailPtr]      @ sp<- saved SP
+    mov     r0, r1                          @ return the changeInterp value
+    add     sp, sp, #4                      @ un-align 64
+    LDMFD_PC "r4-r10,fp"                    @ restore 9 regs and return
+
+
+/*
+ * String references.
+ */
+strBadEntryPoint:
+    .word   .LstrBadEntryPoint
+
+
+
+    .global dvmAsmInstructionStart
+    .type   dvmAsmInstructionStart, %function
+dvmAsmInstructionStart = .L_OP_NOP
+    .text
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NOP: /* 0x00 */
+/* File: armv5te/OP_NOP.S */
+    FETCH_ADVANCE_INST(1)               @ advance to next instr, load rINST
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    GOTO_OPCODE(ip)                     @ execute it
+
+#ifdef ASSIST_DEBUGGER
+    /* insert fake function header to help gdb find the stack frame */
+    .type   dalvik_inst, %function
+dalvik_inst:
+    .fnstart
+    MTERP_ENTRY1
+    MTERP_ENTRY2
+    .fnend
+#endif
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE: /* 0x01 */
+/* File: armv5te/OP_MOVE.S */
+    /* for move, move-object, long-to-int */
+    /* op vA, vB */
+    mov     r1, rINST, lsr #12          @ r1<- B from 15:12
+    mov     r0, rINST, lsr #8           @ r0<- A from 11:8
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[B]
+    and     r0, r0, #15
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[A]<- r2
+    GOTO_OPCODE(ip)                     @ execute next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_FROM16: /* 0x02 */
+/* File: armv5te/OP_MOVE_FROM16.S */
+    /* for: move/from16, move-object/from16 */
+    /* op vAA, vBBBB */
+    FETCH(r1, 1)                        @ r1<- BBBB
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[BBBB]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[AA]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_16: /* 0x03 */
+/* File: armv5te/OP_MOVE_16.S */
+    /* for: move/16, move-object/16 */
+    /* op vAAAA, vBBBB */
+    FETCH(r1, 2)                        @ r1<- BBBB
+    FETCH(r0, 1)                        @ r0<- AAAA
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[BBBB]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[AAAA]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_WIDE: /* 0x04 */
+/* File: armv5te/OP_MOVE_WIDE.S */
+    /* move-wide vA, vB */
+    /* NOTE: regs can overlap, e.g. "move v6,v7" or "move v7,v6" */
+    mov     r2, rINST, lsr #8           @ r2<- A(+)
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r2, r2, #15
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[A]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- fp[B]
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r2, {r0-r1}                 @ fp[A]<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_WIDE_FROM16: /* 0x05 */
+/* File: armv5te/OP_MOVE_WIDE_FROM16.S */
+    /* move-wide/from16 vAA, vBBBB */
+    /* NOTE: regs can overlap, e.g. "move v6,v7" or "move v7,v6" */
+    FETCH(r3, 1)                        @ r3<- BBBB
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[BBBB]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[AA]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- fp[BBBB]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r2, {r0-r1}                 @ fp[AA]<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_WIDE_16: /* 0x06 */
+/* File: armv5te/OP_MOVE_WIDE_16.S */
+    /* move-wide/16 vAAAA, vBBBB */
+    /* NOTE: regs can overlap, e.g. "move v6,v7" or "move v7,v6" */
+    FETCH(r3, 2)                        @ r3<- BBBB
+    FETCH(r2, 1)                        @ r2<- AAAA
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[BBBB]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[AAAA]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- fp[BBBB]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r2, {r0-r1}                 @ fp[AAAA]<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_OBJECT: /* 0x07 */
+/* File: armv5te/OP_MOVE_OBJECT.S */
+/* File: armv5te/OP_MOVE.S */
+    /* for move, move-object, long-to-int */
+    /* op vA, vB */
+    mov     r1, rINST, lsr #12          @ r1<- B from 15:12
+    mov     r0, rINST, lsr #8           @ r0<- A from 11:8
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[B]
+    and     r0, r0, #15
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[A]<- r2
+    GOTO_OPCODE(ip)                     @ execute next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_OBJECT_FROM16: /* 0x08 */
+/* File: armv5te/OP_MOVE_OBJECT_FROM16.S */
+/* File: armv5te/OP_MOVE_FROM16.S */
+    /* for: move/from16, move-object/from16 */
+    /* op vAA, vBBBB */
+    FETCH(r1, 1)                        @ r1<- BBBB
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[BBBB]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[AA]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_OBJECT_16: /* 0x09 */
+/* File: armv5te/OP_MOVE_OBJECT_16.S */
+/* File: armv5te/OP_MOVE_16.S */
+    /* for: move/16, move-object/16 */
+    /* op vAAAA, vBBBB */
+    FETCH(r1, 2)                        @ r1<- BBBB
+    FETCH(r0, 1)                        @ r0<- AAAA
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[BBBB]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[AAAA]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_RESULT: /* 0x0a */
+/* File: armv5te/OP_MOVE_RESULT.S */
+    /* for: move-result, move-result-object */
+    /* op vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    ldr     r0, [rGLUE, #offGlue_retval]    @ r0<- glue->retval.i
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[AA]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_RESULT_WIDE: /* 0x0b */
+/* File: armv5te/OP_MOVE_RESULT_WIDE.S */
+    /* move-result-wide vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    add     r3, rGLUE, #offGlue_retval  @ r3<- &glue->retval
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[AA]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- retval.j
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r2, {r0-r1}                 @ fp[AA]<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_RESULT_OBJECT: /* 0x0c */
+/* File: armv5te/OP_MOVE_RESULT_OBJECT.S */
+/* File: armv5te/OP_MOVE_RESULT.S */
+    /* for: move-result, move-result-object */
+    /* op vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    ldr     r0, [rGLUE, #offGlue_retval]    @ r0<- glue->retval.i
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[AA]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MOVE_EXCEPTION: /* 0x0d */
+/* File: armv5te/OP_MOVE_EXCEPTION.S */
+    /* move-exception vAA */
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- glue->self
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    ldr     r3, [r0, #offThread_exception]  @ r3<- dvmGetException bypass
+    mov     r1, #0                      @ r1<- 0
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    SET_VREG(r3, r2)                    @ fp[AA]<- exception obj
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offThread_exception]  @ dvmClearException bypass
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RETURN_VOID: /* 0x0e */
+/* File: armv5te/OP_RETURN_VOID.S */
+    b       common_returnFromMethod
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RETURN: /* 0x0f */
+/* File: armv5te/OP_RETURN.S */
+    /*
+     * Return a 32-bit value.  Copies the return value into the "glue"
+     * structure, then jumps to the return handler.
+     *
+     * for: return, return-object
+     */
+    /* op vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    GET_VREG(r0, r2)                    @ r0<- vAA
+    str     r0, [rGLUE, #offGlue_retval] @ retval.i <- vAA
+    b       common_returnFromMethod
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RETURN_WIDE: /* 0x10 */
+/* File: armv5te/OP_RETURN_WIDE.S */
+    /*
+     * Return a 64-bit value.  Copies the return value into the "glue"
+     * structure, then jumps to the return handler.
+     */
+    /* return-wide vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[AA]
+    add     r3, rGLUE, #offGlue_retval  @ r3<- &glue->retval
+    ldmia   r2, {r0-r1}                 @ r0/r1 <- vAA/vAA+1
+    stmia   r3, {r0-r1}                 @ retval<- r0/r1
+    b       common_returnFromMethod
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RETURN_OBJECT: /* 0x11 */
+/* File: armv5te/OP_RETURN_OBJECT.S */
+/* File: armv5te/OP_RETURN.S */
+    /*
+     * Return a 32-bit value.  Copies the return value into the "glue"
+     * structure, then jumps to the return handler.
+     *
+     * for: return, return-object
+     */
+    /* op vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    GET_VREG(r0, r2)                    @ r0<- vAA
+    str     r0, [rGLUE, #offGlue_retval] @ retval.i <- vAA
+    b       common_returnFromMethod
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_4: /* 0x12 */
+/* File: armv5te/OP_CONST_4.S */
+    /* const/4 vA, #+B */
+    mov     r1, rINST, lsl #16          @ r1<- Bxxx0000
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mov     r1, r1, asr #28             @ r1<- sssssssB (sign-extended)
+    and     r0, r0, #15
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    SET_VREG(r1, r0)                    @ fp[A]<- r1
+    GOTO_OPCODE(ip)                     @ execute next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_16: /* 0x13 */
+/* File: armv5te/OP_CONST_16.S */
+    /* const/16 vAA, #+BBBB */
+    FETCH_S(r0, 1)                      @ r0<- ssssBBBB (sign-extended)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r0, r3)                    @ vAA<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST: /* 0x14 */
+/* File: armv5te/OP_CONST.S */
+    /* const vAA, #+BBBBbbbb */
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH(r0, 1)                        @ r0<- bbbb (low)
+    FETCH(r1, 2)                        @ r1<- BBBB (high)
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    orr     r0, r0, r1, lsl #16         @ r0<- BBBBbbbb
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r3)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_HIGH16: /* 0x15 */
+/* File: armv5te/OP_CONST_HIGH16.S */
+    /* const/high16 vAA, #+BBBB0000 */
+    FETCH(r0, 1)                        @ r0<- 0000BBBB (zero-extended)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    mov     r0, r0, lsl #16             @ r0<- BBBB0000
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r0, r3)                    @ vAA<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_WIDE_16: /* 0x16 */
+/* File: armv5te/OP_CONST_WIDE_16.S */
+    /* const-wide/16 vAA, #+BBBB */
+    FETCH_S(r0, 1)                      @ r0<- ssssBBBB (sign-extended)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    mov     r1, r0, asr #31             @ r1<- ssssssss
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r3, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_WIDE_32: /* 0x17 */
+/* File: armv5te/OP_CONST_WIDE_32.S */
+    /* const-wide/32 vAA, #+BBBBbbbb */
+    FETCH(r0, 1)                        @ r0<- 0000bbbb (low)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH_S(r2, 2)                      @ r2<- ssssBBBB (high)
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    orr     r0, r0, r2, lsl #16         @ r0<- BBBBbbbb
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[AA]
+    mov     r1, r0, asr #31             @ r1<- ssssssss
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r3, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_WIDE: /* 0x18 */
+/* File: armv5te/OP_CONST_WIDE.S */
+    /* const-wide vAA, #+HHHHhhhhBBBBbbbb */
+    FETCH(r0, 1)                        @ r0<- bbbb (low)
+    FETCH(r1, 2)                        @ r1<- BBBB (low middle)
+    FETCH(r2, 3)                        @ r2<- hhhh (high middle)
+    orr     r0, r0, r1, lsl #16         @ r0<- BBBBbbbb (low word)
+    FETCH(r3, 4)                        @ r3<- HHHH (high)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    orr     r1, r2, r3, lsl #16         @ r1<- HHHHhhhh (high word)
+    FETCH_ADVANCE_INST(5)               @ advance rPC, load rINST
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_WIDE_HIGH16: /* 0x19 */
+/* File: armv5te/OP_CONST_WIDE_HIGH16.S */
+    /* const-wide/high16 vAA, #+BBBB000000000000 */
+    FETCH(r1, 1)                        @ r1<- 0000BBBB (zero-extended)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    mov     r0, #0                      @ r0<- 00000000
+    mov     r1, r1, lsl #16             @ r1<- BBBB0000
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r3, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_STRING: /* 0x1a */
+/* File: armv5te/OP_CONST_STRING.S */
+    /* const/string vAA, String@BBBB */
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- glue->methodClassDex
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    ldr     r2, [r2, #offDvmDex_pResStrings]   @ r2<- dvmDex->pResStrings
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- pResStrings[BBBB]
+    cmp     r0, #0                      @ not yet resolved?
+    beq     .LOP_CONST_STRING_resolve
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_STRING_JUMBO: /* 0x1b */
+/* File: armv5te/OP_CONST_STRING_JUMBO.S */
+    /* const/string vAA, String@BBBBBBBB */
+    FETCH(r0, 1)                        @ r0<- bbbb (low)
+    FETCH(r1, 2)                        @ r1<- BBBB (high)
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- glue->methodClassDex
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    ldr     r2, [r2, #offDvmDex_pResStrings]   @ r2<- dvmDex->pResStrings
+    orr     r1, r0, r1, lsl #16         @ r1<- BBBBbbbb
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- pResStrings[BBBB]
+    cmp     r0, #0
+    beq     .LOP_CONST_STRING_JUMBO_resolve
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CONST_CLASS: /* 0x1c */
+/* File: armv5te/OP_CONST_CLASS.S */
+    /* const/class vAA, Class@BBBB */
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- glue->methodClassDex
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    ldr     r2, [r2, #offDvmDex_pResClasses]   @ r2<- dvmDex->pResClasses
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- pResClasses[BBBB]
+    cmp     r0, #0                      @ not yet resolved?
+    beq     .LOP_CONST_CLASS_resolve
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MONITOR_ENTER: /* 0x1d */
+/* File: armv5te/OP_MONITOR_ENTER.S */
+    /*
+     * Synchronize on an object.
+     */
+    /* monitor-enter vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    GET_VREG(r1, r2)                    @ r1<- vAA (object)
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- glue->self
+    cmp     r1, #0                      @ null object?
+    EXPORT_PC()                         @ need for precise GC, MONITOR_TRACKING
+    beq     common_errNullObject        @ null object, throw an exception
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    bl      dvmLockObject               @ call(self, obj)
+#ifdef WITH_DEADLOCK_PREDICTION /* implies WITH_MONITOR_TRACKING */
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- glue->self
+    ldr     r1, [r0, #offThread_exception] @ check for exception
+    cmp     r1, #0
+    bne     common_exceptionThrown      @ exception raised, bail out
+#endif
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MONITOR_EXIT: /* 0x1e */
+/* File: armv5te/OP_MONITOR_EXIT.S */
+    /*
+     * Unlock an object.
+     *
+     * Exceptions that occur when unlocking a monitor need to appear as
+     * if they happened at the following instruction.  See the Dalvik
+     * instruction spec.
+     */
+    /* monitor-exit vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    EXPORT_PC()                         @ before fetch: export the PC
+    GET_VREG(r1, r2)                    @ r1<- vAA (object)
+    cmp     r1, #0                      @ null object?
+    beq     common_errNullObject        @ yes
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- glue->self
+    bl      dvmUnlockObject             @ r0<- success for unlock(self, obj)
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yes, exception is pending
+    FETCH_ADVANCE_INST(1)               @ before throw: advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CHECK_CAST: /* 0x1f */
+/* File: armv5te/OP_CHECK_CAST.S */
+    /*
+     * Check to see if a cast from one class to another is allowed.
+     */
+    /* check-cast vAA, class@BBBB */
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    FETCH(r2, 1)                        @ r2<- BBBB
+    GET_VREG(r9, r3)                    @ r9<- object
+    ldr     r0, [rGLUE, #offGlue_methodClassDex]    @ r0<- pDvmDex
+    cmp     r9, #0                      @ is object null?
+    ldr     r0, [r0, #offDvmDex_pResClasses]    @ r0<- pDvmDex->pResClasses
+    beq     .LOP_CHECK_CAST_okay            @ null obj, cast always succeeds
+    ldr     r1, [r0, r2, lsl #2]        @ r1<- resolved class
+    ldr     r0, [r9, #offObject_clazz]  @ r0<- obj->clazz
+    cmp     r1, #0                      @ have we resolved this before?
+    beq     .LOP_CHECK_CAST_resolve         @ not resolved, do it now
+.LOP_CHECK_CAST_resolved:
+    cmp     r0, r1                      @ same class (trivial success)?
+    bne     .LOP_CHECK_CAST_fullcheck       @ no, do full check
+.LOP_CHECK_CAST_okay:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INSTANCE_OF: /* 0x20 */
+/* File: armv5te/OP_INSTANCE_OF.S */
+    /*
+     * Check to see if an object reference is an instance of a class.
+     *
+     * Most common situation is a non-null object, being compared against
+     * an already-resolved class.
+     */
+    /* instance-of vA, vB, class@CCCC */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB (object)
+    and     r9, r9, #15                 @ r9<- A
+    cmp     r0, #0                      @ is object null?
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- pDvmDex
+    beq     .LOP_INSTANCE_OF_store           @ null obj, not an instance, store r0
+    FETCH(r3, 1)                        @ r3<- CCCC
+    ldr     r2, [r2, #offDvmDex_pResClasses]    @ r2<- pDvmDex->pResClasses
+    ldr     r1, [r2, r3, lsl #2]        @ r1<- resolved class
+    ldr     r0, [r0, #offObject_clazz]  @ r0<- obj->clazz
+    cmp     r1, #0                      @ have we resolved this before?
+    beq     .LOP_INSTANCE_OF_resolve         @ not resolved, do it now
+.LOP_INSTANCE_OF_resolved: @ r0=obj->clazz, r1=resolved class
+    cmp     r0, r1                      @ same class (trivial success)?
+    beq     .LOP_INSTANCE_OF_trivial         @ yes, trivial finish
+    b       .LOP_INSTANCE_OF_fullcheck       @ no, do full check
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ARRAY_LENGTH: /* 0x21 */
+/* File: armv5te/OP_ARRAY_LENGTH.S */
+    /*
+     * Return the length of an array.
+     */
+    mov     r1, rINST, lsr #12          @ r1<- B
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    GET_VREG(r0, r1)                    @ r0<- vB (object ref)
+    and     r2, r2, #15                 @ r2<- A
+    cmp     r0, #0                      @ is object null?
+    beq     common_errNullObject        @ yup, fail
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- array length
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r3, r2)                    @ vB<- length
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEW_INSTANCE: /* 0x22 */
+/* File: armv5te/OP_NEW_INSTANCE.S */
+    /*
+     * Create a new instance of a class.
+     */
+    /* new-instance vAA, class@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResClasses]    @ r3<- pDvmDex->pResClasses
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved class
+    EXPORT_PC()                         @ req'd for init, resolve, alloc
+    cmp     r0, #0                      @ already resolved?
+    beq     .LOP_NEW_INSTANCE_resolve         @ no, resolve it now
+.LOP_NEW_INSTANCE_resolved:   @ r0=class
+    ldrb    r1, [r0, #offClassObject_status]    @ r1<- ClassStatus enum
+    cmp     r1, #CLASS_INITIALIZED      @ has class been initialized?
+    bne     .LOP_NEW_INSTANCE_needinit        @ no, init class now
+.LOP_NEW_INSTANCE_initialized: @ r0=class
+    mov     r1, #ALLOC_DONT_TRACK       @ flags for alloc call
+    bl      dvmAllocObject              @ r0<- new object
+    b       .LOP_NEW_INSTANCE_finish          @ continue
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEW_ARRAY: /* 0x23 */
+/* File: armv5te/OP_NEW_ARRAY.S */
+    /*
+     * Allocate an array of objects, specified with the array class
+     * and a count.
+     *
+     * The verifier guarantees that this is an array class, so we don't
+     * check for it here.
+     */
+    /* new-array vA, vB, class@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    FETCH(r2, 1)                        @ r2<- CCCC
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    GET_VREG(r1, r0)                    @ r1<- vB (array length)
+    ldr     r3, [r3, #offDvmDex_pResClasses]    @ r3<- pDvmDex->pResClasses
+    cmp     r1, #0                      @ check length
+    ldr     r0, [r3, r2, lsl #2]        @ r0<- resolved class
+    bmi     common_errNegativeArraySize @ negative length, bail
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ req'd for resolve, alloc
+    bne     .LOP_NEW_ARRAY_finish          @ resolved, continue
+    b       .LOP_NEW_ARRAY_resolve         @ do resolve now
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FILLED_NEW_ARRAY: /* 0x24 */
+/* File: armv5te/OP_FILLED_NEW_ARRAY.S */
+    /*
+     * Create a new array with elements filled from registers.
+     *
+     * for: filled-new-array, filled-new-array/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, type@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResClasses]    @ r3<- pDvmDex->pResClasses
+    EXPORT_PC()                         @ need for resolve and alloc
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved class
+    mov     r10, rINST, lsr #8          @ r10<- AA or BA
+    cmp     r0, #0                      @ already resolved?
+    bne     .LOP_FILLED_NEW_ARRAY_continue        @ yes, continue on
+8:  ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    mov     r2, #0                      @ r2<- false
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- call(clazz, ref)
+    cmp     r0, #0                      @ got null?
+    beq     common_exceptionThrown      @ yes, handle exception
+    b       .LOP_FILLED_NEW_ARRAY_continue
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FILLED_NEW_ARRAY_RANGE: /* 0x25 */
+/* File: armv5te/OP_FILLED_NEW_ARRAY_RANGE.S */
+/* File: armv5te/OP_FILLED_NEW_ARRAY.S */
+    /*
+     * Create a new array with elements filled from registers.
+     *
+     * for: filled-new-array, filled-new-array/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, type@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResClasses]    @ r3<- pDvmDex->pResClasses
+    EXPORT_PC()                         @ need for resolve and alloc
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved class
+    mov     r10, rINST, lsr #8          @ r10<- AA or BA
+    cmp     r0, #0                      @ already resolved?
+    bne     .LOP_FILLED_NEW_ARRAY_RANGE_continue        @ yes, continue on
+8:  ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    mov     r2, #0                      @ r2<- false
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- call(clazz, ref)
+    cmp     r0, #0                      @ got null?
+    beq     common_exceptionThrown      @ yes, handle exception
+    b       .LOP_FILLED_NEW_ARRAY_RANGE_continue
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FILL_ARRAY_DATA: /* 0x26 */
+/* File: armv5te/OP_FILL_ARRAY_DATA.S */
+    /* fill-array-data vAA, +BBBBBBBB */
+    FETCH(r0, 1)                        @ r0<- bbbb (lo)
+    FETCH(r1, 2)                        @ r1<- BBBB (hi)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    orr     r1, r0, r1, lsl #16         @ r1<- BBBBbbbb
+    GET_VREG(r0, r3)                    @ r0<- vAA (array object)
+    add     r1, rPC, r1, lsl #1         @ r1<- PC + BBBBbbbb*2 (array data off.)
+    EXPORT_PC();
+    bl      dvmInterpHandleFillArrayData@ fill the array with predefined data
+    cmp     r0, #0                      @ 0 means an exception is thrown
+    beq     common_exceptionThrown      @ has exception
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_THROW: /* 0x27 */
+/* File: armv5te/OP_THROW.S */
+    /*
+     * Throw an exception object in the current thread.
+     */
+    /* throw vAA */
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    GET_VREG(r1, r2)                    @ r1<- vAA (exception object)
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- glue->self
+    cmp     r1, #0                      @ null object?
+    beq     common_errNullObject        @ yes, throw an NPE instead
+    @ bypass dvmSetException, just store it
+    str     r1, [r0, #offThread_exception]  @ thread->exception<- obj
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_GOTO: /* 0x28 */
+/* File: armv5te/OP_GOTO.S */
+    /*
+     * Unconditional branch, 8-bit offset.
+     *
+     * The branch distance is a signed code-unit offset, which we need to
+     * double to get a byte offset.
+     */
+    /* goto +AA */
+    mov     r0, rINST, lsl #16          @ r0<- AAxx0000
+    movs    r9, r0, asr #24             @ r9<- ssssssAA (sign-extended)
+    mov     r9, r9, lsl #1              @ r9<- byte offset
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+    FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_GOTO_16: /* 0x29 */
+/* File: armv5te/OP_GOTO_16.S */
+    /*
+     * Unconditional branch, 16-bit offset.
+     *
+     * The branch distance is a signed code-unit offset, which we need to
+     * double to get a byte offset.
+     */
+    /* goto/16 +AAAA */
+    FETCH_S(r0, 1)                      @ r0<- ssssAAAA (sign-extended)
+    movs    r9, r0, asl #1              @ r9<- byte offset, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+    FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_GOTO_32: /* 0x2a */
+/* File: armv5te/OP_GOTO_32.S */
+    /*
+     * Unconditional branch, 32-bit offset.
+     *
+     * The branch distance is a signed code-unit offset, which we need to
+     * double to get a byte offset.
+     *
+     * Unlike most opcodes, this one is allowed to branch to itself, so
+     * our "backward branch" test must be "<=0" instead of "<0".  The ORRS
+     * instruction doesn't affect the V flag, so we need to clear it
+     * explicitly.
+     */
+    /* goto/32 +AAAAAAAA */
+    FETCH(r0, 1)                        @ r0<- aaaa (lo)
+    FETCH(r1, 2)                        @ r1<- AAAA (hi)
+    cmp     ip, ip                      @ (clear V flag during stall)
+    orrs    r0, r0, r1, lsl #16         @ r0<- AAAAaaaa, check sign
+    mov     r9, r0, asl #1              @ r9<- byte offset
+    ble     common_backwardBranch       @ backward branch, do periodic checks
+    FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_PACKED_SWITCH: /* 0x2b */
+/* File: armv5te/OP_PACKED_SWITCH.S */
+    /*
+     * Handle a packed-switch or sparse-switch instruction.  In both cases
+     * we decode it and hand it off to a helper function.
+     *
+     * We don't really expect backward branches in a switch statement, but
+     * they're perfectly legal, so we check for them here.
+     *
+     * for: packed-switch, sparse-switch
+     */
+    /* op vAA, +BBBB */
+    FETCH(r0, 1)                        @ r0<- bbbb (lo)
+    FETCH(r1, 2)                        @ r1<- BBBB (hi)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    orr     r0, r0, r1, lsl #16         @ r0<- BBBBbbbb
+    GET_VREG(r1, r3)                    @ r1<- vAA
+    add     r0, rPC, r0, lsl #1         @ r0<- PC + BBBBbbbb*2
+    bl      dvmInterpHandlePackedSwitch                       @ r0<- code-unit branch offset
+    movs    r9, r0, asl #1              @ r9<- branch byte offset, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+    beq     common_backwardBranch       @ (want to use BLE but V is unknown)
+    FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPARSE_SWITCH: /* 0x2c */
+/* File: armv5te/OP_SPARSE_SWITCH.S */
+/* File: armv5te/OP_PACKED_SWITCH.S */
+    /*
+     * Handle a packed-switch or sparse-switch instruction.  In both cases
+     * we decode it and hand it off to a helper function.
+     *
+     * We don't really expect backward branches in a switch statement, but
+     * they're perfectly legal, so we check for them here.
+     *
+     * for: packed-switch, sparse-switch
+     */
+    /* op vAA, +BBBB */
+    FETCH(r0, 1)                        @ r0<- bbbb (lo)
+    FETCH(r1, 2)                        @ r1<- BBBB (hi)
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    orr     r0, r0, r1, lsl #16         @ r0<- BBBBbbbb
+    GET_VREG(r1, r3)                    @ r1<- vAA
+    add     r0, rPC, r0, lsl #1         @ r0<- PC + BBBBbbbb*2
+    bl      dvmInterpHandleSparseSwitch                       @ r0<- code-unit branch offset
+    movs    r9, r0, asl #1              @ r9<- branch byte offset, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+    beq     common_backwardBranch       @ (want to use BLE but V is unknown)
+    FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMPL_FLOAT: /* 0x2d */
+/* File: vfp/OP_CMPL_FLOAT.S */
+    /*
+     * Compare two floating-point values.  Puts 0, 1, or -1 into the
+     * destination register based on the results of the comparison.
+     *
+     * int compare(x, y) {
+     *     if (x == y) {
+     *         return 0;
+     *     } else if (x > y) {
+     *         return 1;
+     *     } else if (x < y) {
+     *         return -1;
+     *     } else {
+     *         return -1;
+     *     }
+     * }
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    VREG_INDEX_TO_ADDR(r2, r2)          @ r2<- &vBB
+    flds    s0, [r2]                    @ s0<- vBB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vCC
+    flds    s1, [r3]                    @ s1<- vCC
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    fcmpes  s0, s1                      @ compare (vBB, vCC)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    mvn     r0, #0                      @ r0<- -1 (default)
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    fmstat                              @ export status flags
+    movgt   r0, #1                      @ (greater than) r1<- 1
+    moveq   r0, #0                      @ (equal) r1<- 0
+    bl      .LOP_CMPL_FLOAT_finish          @ argh
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMPG_FLOAT: /* 0x2e */
+/* File: vfp/OP_CMPG_FLOAT.S */
+    /*
+     * Compare two floating-point values.  Puts 0, 1, or -1 into the
+     * destination register based on the results of the comparison.
+     *
+     * int compare(x, y) {
+     *     if (x == y) {
+     *         return 0;
+     *     } else if (x < y) {
+     *         return -1;
+     *     } else if (x > y) {
+     *         return 1;
+     *     } else {
+     *         return 1;
+     *     }
+     * }
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    VREG_INDEX_TO_ADDR(r2, r2)          @ r2<- &vBB
+    flds    s0, [r2]                    @ s0<- vBB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vCC
+    flds    s1, [r3]                    @ s1<- vCC
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    fcmpes  s0, s1                      @ compare (vBB, vCC)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    mov     r0, #1                      @ r0<- 1 (default)
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    fmstat                              @ export status flags
+    mvnmi   r0, #0                      @ (less than) r1<- -1
+    moveq   r0, #0                      @ (equal) r1<- 0
+    bl      .LOP_CMPG_FLOAT_finish          @ argh
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMPL_DOUBLE: /* 0x2f */
+/* File: vfp/OP_CMPL_DOUBLE.S */
+    /*
+     * Compare two floating-point values.  Puts 0, 1, or -1 into the
+     * destination register based on the results of the comparison.
+     *
+     * int compare(x, y) {
+     *     if (x == y) {
+     *         return 0;
+     *     } else if (x > y) {
+     *         return 1;
+     *     } else if (x < y) {
+     *         return -1;
+     *     } else {
+     *         return -1;
+     *     }
+     * }
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    VREG_INDEX_TO_ADDR(r2, r2)          @ r2<- &vBB
+    fldd    d0, [r2]                    @ d0<- vBB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vCC
+    fldd    d1, [r3]                    @ d1<- vCC
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    fcmped  d0, d1                      @ compare (vBB, vCC)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    mvn     r0, #0                      @ r0<- -1 (default)
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    fmstat                              @ export status flags
+    movgt   r0, #1                      @ (greater than) r1<- 1
+    moveq   r0, #0                      @ (equal) r1<- 0
+    bl      .LOP_CMPL_DOUBLE_finish          @ argh
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMPG_DOUBLE: /* 0x30 */
+/* File: vfp/OP_CMPG_DOUBLE.S */
+    /*
+     * Compare two floating-point values.  Puts 0, 1, or -1 into the
+     * destination register based on the results of the comparison.
+     *
+     * int compare(x, y) {
+     *     if (x == y) {
+     *         return 0;
+     *     } else if (x < y) {
+     *         return -1;
+     *     } else if (x > y) {
+     *         return 1;
+     *     } else {
+     *         return 1;
+     *     }
+     * }
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    VREG_INDEX_TO_ADDR(r2, r2)          @ r2<- &vBB
+    fldd    d0, [r2]                    @ d0<- vBB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vCC
+    fldd    d1, [r3]                    @ d1<- vCC
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    fcmped  d0, d1                      @ compare (vBB, vCC)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    mov     r0, #1                      @ r0<- 1 (default)
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    fmstat                              @ export status flags
+    mvnmi   r0, #0                      @ (less than) r1<- -1
+    moveq   r0, #0                      @ (equal) r1<- 0
+    bl      .LOP_CMPG_DOUBLE_finish          @ argh
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_CMP_LONG: /* 0x31 */
+/* File: armv5te/OP_CMP_LONG.S */
+    /*
+     * Compare two 64-bit values.  Puts 0, 1, or -1 into the destination
+     * register based on the results of the comparison.
+     *
+     * We load the full values with LDM, but in practice many values could
+     * be resolved by only looking at the high word.  This could be made
+     * faster or slower by splitting the LDM into a pair of LDRs.
+     *
+     * If we just wanted to set condition flags, we could do this:
+     *  subs    ip, r0, r2
+     *  sbcs    ip, r1, r3
+     *  subeqs  ip, r0, r2
+     * Leaving { <0, 0, >0 } in ip.  However, we have to set it to a specific
+     * integer value, which we can do with 2 conditional mov/mvn instructions
+     * (set 1, set -1; if they're equal we already have 0 in ip), giving
+     * us a constant 5-cycle path plus a branch at the end to the
+     * instruction epilogue code.  The multi-compare approach below needs
+     * 2 or 3 cycles + branch if the high word doesn't match, 6 + branch
+     * in the worst case (the 64-bit values are equal).
+     */
+    /* cmp-long vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    cmp     r1, r3                      @ compare (vBB+1, vCC+1)
+    blt     .LOP_CMP_LONG_less            @ signed compare on high part
+    bgt     .LOP_CMP_LONG_greater
+    subs    r1, r0, r2                  @ r1<- r0 - r2
+    bhi     .LOP_CMP_LONG_greater         @ unsigned compare on low part
+    bne     .LOP_CMP_LONG_less
+    b       .LOP_CMP_LONG_finish          @ equal; r1 already holds 0
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_EQ: /* 0x32 */
+/* File: armv5te/OP_IF_EQ.S */
+/* File: armv5te/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r3, r1)                    @ r3<- vB
+    GET_VREG(r2, r0)                    @ r2<- vA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, r3                      @ compare (vA, vB)
+    bne  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ yes, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_NE: /* 0x33 */
+/* File: armv5te/OP_IF_NE.S */
+/* File: armv5te/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r3, r1)                    @ r3<- vB
+    GET_VREG(r2, r0)                    @ r2<- vA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, r3                      @ compare (vA, vB)
+    beq  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ yes, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_LT: /* 0x34 */
+/* File: armv5te/OP_IF_LT.S */
+/* File: armv5te/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r3, r1)                    @ r3<- vB
+    GET_VREG(r2, r0)                    @ r2<- vA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, r3                      @ compare (vA, vB)
+    bge  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ yes, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_GE: /* 0x35 */
+/* File: armv5te/OP_IF_GE.S */
+/* File: armv5te/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r3, r1)                    @ r3<- vB
+    GET_VREG(r2, r0)                    @ r2<- vA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, r3                      @ compare (vA, vB)
+    blt  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ yes, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_GT: /* 0x36 */
+/* File: armv5te/OP_IF_GT.S */
+/* File: armv5te/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r3, r1)                    @ r3<- vB
+    GET_VREG(r2, r0)                    @ r2<- vA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, r3                      @ compare (vA, vB)
+    ble  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ yes, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_LE: /* 0x37 */
+/* File: armv5te/OP_IF_LE.S */
+/* File: armv5te/bincmp.S */
+    /*
+     * Generic two-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * For: if-eq, if-ne, if-lt, if-ge, if-gt, if-le
+     */
+    /* if-cmp vA, vB, +CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r3, r1)                    @ r3<- vB
+    GET_VREG(r2, r0)                    @ r2<- vA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, r3                      @ compare (vA, vB)
+    bgt  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ yes, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_EQZ: /* 0x38 */
+/* File: armv5te/OP_IF_EQZ.S */
+/* File: armv5te/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    bne  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_NEZ: /* 0x39 */
+/* File: armv5te/OP_IF_NEZ.S */
+/* File: armv5te/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    beq  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_LTZ: /* 0x3a */
+/* File: armv5te/OP_IF_LTZ.S */
+/* File: armv5te/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    bge  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_GEZ: /* 0x3b */
+/* File: armv5te/OP_IF_GEZ.S */
+/* File: armv5te/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    blt  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_GTZ: /* 0x3c */
+/* File: armv5te/OP_IF_GTZ.S */
+/* File: armv5te/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    ble  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IF_LEZ: /* 0x3d */
+/* File: armv5te/OP_IF_LEZ.S */
+/* File: armv5te/zcmp.S */
+    /*
+     * Generic one-operand compare-and-branch operation.  Provide a "revcmp"
+     * fragment that specifies the *reverse* comparison to perform, e.g.
+     * for "if-le" you would use "gt".
+     *
+     * for: if-eqz, if-nez, if-ltz, if-gez, if-gtz, if-lez
+     */
+    /* if-cmp vAA, +BBBB */
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    GET_VREG(r2, r0)                    @ r2<- vAA
+    mov     r9, #4                      @ r0<- BYTE branch dist for not-taken
+    cmp     r2, #0                      @ compare (vA, 0)
+    bgt  1f                      @ branch to 1 if comparison failed
+    FETCH_S(r9, 1)                      @ r9<- branch offset, in code units
+    movs    r9, r9, asl #1              @ convert to bytes, check sign
+    bmi     common_backwardBranch       @ backward branch, do periodic checks
+1:  FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_3E: /* 0x3e */
+/* File: armv5te/OP_UNUSED_3E.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_3F: /* 0x3f */
+/* File: armv5te/OP_UNUSED_3F.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_40: /* 0x40 */
+/* File: armv5te/OP_UNUSED_40.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_41: /* 0x41 */
+/* File: armv5te/OP_UNUSED_41.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_42: /* 0x42 */
+/* File: armv5te/OP_UNUSED_42.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_43: /* 0x43 */
+/* File: armv5te/OP_UNUSED_43.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET: /* 0x44 */
+/* File: armv5te/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #2     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldr   r2, [r0, #offArrayObject_contents]  @ r2<- vBB[vCC]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r9)                    @ vAA<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_WIDE: /* 0x45 */
+/* File: armv5te/OP_AGET_WIDE.S */
+    /*
+     * Array get, 64 bits.  vAA <- vBB[vCC].
+     *
+     * Arrays of long/double are 64-bit aligned, so it's okay to use LDRD.
+     */
+    /* aget-wide vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #3          @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcc     .LOP_AGET_WIDE_finish          @ okay, continue below
+    b       common_errArrayIndex        @ index >= length, bail
+    @ May want to swap the order of these two branches depending on how the
+    @ branch prediction (if any) handles conditional forward branches vs.
+    @ unconditional forward branches.
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_OBJECT: /* 0x46 */
+/* File: armv5te/OP_AGET_OBJECT.S */
+/* File: armv5te/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #2     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldr   r2, [r0, #offArrayObject_contents]  @ r2<- vBB[vCC]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r9)                    @ vAA<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_BOOLEAN: /* 0x47 */
+/* File: armv5te/OP_AGET_BOOLEAN.S */
+/* File: armv5te/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #0     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldrb   r2, [r0, #offArrayObject_contents]  @ r2<- vBB[vCC]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r9)                    @ vAA<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_BYTE: /* 0x48 */
+/* File: armv5te/OP_AGET_BYTE.S */
+/* File: armv5te/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #0     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldrsb   r2, [r0, #offArrayObject_contents]  @ r2<- vBB[vCC]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r9)                    @ vAA<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_CHAR: /* 0x49 */
+/* File: armv5te/OP_AGET_CHAR.S */
+/* File: armv5te/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #1     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldrh   r2, [r0, #offArrayObject_contents]  @ r2<- vBB[vCC]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r9)                    @ vAA<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AGET_SHORT: /* 0x4a */
+/* File: armv5te/OP_AGET_SHORT.S */
+/* File: armv5te/OP_AGET.S */
+    /*
+     * Array get, 32 bits or less.  vAA <- vBB[vCC].
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aget, aget-object, aget-boolean, aget-byte, aget-char, aget-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #1     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldrsh   r2, [r0, #offArrayObject_contents]  @ r2<- vBB[vCC]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r2, r9)                    @ vAA<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT: /* 0x4b */
+/* File: armv5te/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA.
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aput, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #2     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r9)                    @ r2<- vAA
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r2, [r0, #offArrayObject_contents]  @ vBB[vCC]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_WIDE: /* 0x4c */
+/* File: armv5te/OP_APUT_WIDE.S */
+    /*
+     * Array put, 64 bits.  vBB[vCC] <- vAA.
+     *
+     * Arrays of long/double are 64-bit aligned, so it's okay to use STRD.
+     */
+    /* aput-wide vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #3          @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    bcc     .LOP_APUT_WIDE_finish          @ okay, continue below
+    b       common_errArrayIndex        @ index >= length, bail
+    @ May want to swap the order of these two branches depending on how the
+    @ branch prediction (if any) handles conditional forward branches vs.
+    @ unconditional forward branches.
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_OBJECT: /* 0x4d */
+/* File: armv5te/OP_APUT_OBJECT.S */
+    /*
+     * Store an object into an array.  vBB[vCC] <- vAA.
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    GET_VREG(r1, r2)                    @ r1<- vBB (array object)
+    GET_VREG(r0, r3)                    @ r0<- vCC (requested index)
+    cmp     r1, #0                      @ null array object?
+    GET_VREG(r9, r9)                    @ r9<- vAA
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r1, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r10, r1, r0, lsl #2         @ r10<- arrayObj + index*width
+    cmp     r0, r3                      @ compare unsigned index, length
+    bcc     .LOP_APUT_OBJECT_finish          @ we're okay, continue on
+    b       common_errArrayIndex        @ index >= length, bail
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_BOOLEAN: /* 0x4e */
+/* File: armv5te/OP_APUT_BOOLEAN.S */
+/* File: armv5te/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA.
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aput, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #0     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r9)                    @ r2<- vAA
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strb  r2, [r0, #offArrayObject_contents]  @ vBB[vCC]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_BYTE: /* 0x4f */
+/* File: armv5te/OP_APUT_BYTE.S */
+/* File: armv5te/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA.
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aput, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #0     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r9)                    @ r2<- vAA
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strb  r2, [r0, #offArrayObject_contents]  @ vBB[vCC]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_CHAR: /* 0x50 */
+/* File: armv5te/OP_APUT_CHAR.S */
+/* File: armv5te/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA.
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aput, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #1     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r9)                    @ r2<- vAA
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strh  r2, [r0, #offArrayObject_contents]  @ vBB[vCC]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_APUT_SHORT: /* 0x51 */
+/* File: armv5te/OP_APUT_SHORT.S */
+/* File: armv5te/OP_APUT.S */
+    /*
+     * Array put, 32 bits or less.  vBB[vCC] <- vAA.
+     *
+     * Note: using the usual FETCH/and/shift stuff, this fits in exactly 17
+     * instructions.  We use a pair of FETCH_Bs instead.
+     *
+     * for: aput, aput-boolean, aput-byte, aput-char, aput-short
+     */
+    /* op vAA, vBB, vCC */
+    FETCH_B(r2, 1, 0)                   @ r2<- BB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    FETCH_B(r3, 1, 1)                   @ r3<- CC
+    GET_VREG(r0, r2)                    @ r0<- vBB (array object)
+    GET_VREG(r1, r3)                    @ r1<- vCC (requested index)
+    cmp     r0, #0                      @ null array object?
+    beq     common_errNullObject        @ yes, bail
+    ldr     r3, [r0, #offArrayObject_length]    @ r3<- arrayObj->length
+    add     r0, r0, r1, lsl #1     @ r0<- arrayObj + index*width
+    cmp     r1, r3                      @ compare unsigned index, length
+    bcs     common_errArrayIndex        @ index >= length, bail
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r2, r9)                    @ r2<- vAA
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strh  r2, [r0, #offArrayObject_contents]  @ vBB[vCC]<- r2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET: /* 0x52 */
+/* File: armv5te/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_finish
+    b       common_exceptionThrown
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_WIDE: /* 0x53 */
+/* File: armv5te/OP_IGET_WIDE.S */
+    /*
+     * Wide 32-bit instance field get.
+     */
+    /* iget-wide vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_WIDE_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method] @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_WIDE_finish
+    b       common_exceptionThrown
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_OBJECT: /* 0x54 */
+/* File: armv5te/OP_IGET_OBJECT.S */
+/* File: armv5te/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_OBJECT_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_OBJECT_finish
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_BOOLEAN: /* 0x55 */
+/* File: armv5te/OP_IGET_BOOLEAN.S */
+@include "armv5te/OP_IGET.S" { "load":"ldrb", "sqnum":"1" }
+/* File: armv5te/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_BOOLEAN_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_BOOLEAN_finish
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_BYTE: /* 0x56 */
+/* File: armv5te/OP_IGET_BYTE.S */
+@include "armv5te/OP_IGET.S" { "load":"ldrsb", "sqnum":"2" }
+/* File: armv5te/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_BYTE_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_BYTE_finish
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_CHAR: /* 0x57 */
+/* File: armv5te/OP_IGET_CHAR.S */
+@include "armv5te/OP_IGET.S" { "load":"ldrh", "sqnum":"3" }
+/* File: armv5te/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_CHAR_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_CHAR_finish
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_SHORT: /* 0x58 */
+/* File: armv5te/OP_IGET_SHORT.S */
+@include "armv5te/OP_IGET.S" { "load":"ldrsh", "sqnum":"4" }
+/* File: armv5te/OP_IGET.S */
+    /*
+     * General 32-bit instance field get.
+     *
+     * for: iget, iget-object, iget-boolean, iget-byte, iget-char, iget-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IGET_SHORT_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0
+    bne     .LOP_IGET_SHORT_finish
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT: /* 0x59 */
+/* File: armv5te/OP_IPUT.S */
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_WIDE: /* 0x5a */
+/* File: armv5te/OP_IPUT_WIDE.S */
+    /* iput-wide vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_WIDE_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method] @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_WIDE_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_OBJECT: /* 0x5b */
+/* File: armv5te/OP_IPUT_OBJECT.S */
+/* File: armv5te/OP_IPUT.S */
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_OBJECT_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_OBJECT_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_BOOLEAN: /* 0x5c */
+/* File: armv5te/OP_IPUT_BOOLEAN.S */
+@include "armv5te/OP_IPUT.S" { "store":"strb", "sqnum":"1" }
+/* File: armv5te/OP_IPUT.S */
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_BOOLEAN_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_BOOLEAN_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_BYTE: /* 0x5d */
+/* File: armv5te/OP_IPUT_BYTE.S */
+@include "armv5te/OP_IPUT.S" { "store":"strb", "sqnum":"2" }
+/* File: armv5te/OP_IPUT.S */
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_BYTE_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_BYTE_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_CHAR: /* 0x5e */
+/* File: armv5te/OP_IPUT_CHAR.S */
+@include "armv5te/OP_IPUT.S" { "store":"strh", "sqnum":"3" }
+/* File: armv5te/OP_IPUT.S */
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_CHAR_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_CHAR_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_SHORT: /* 0x5f */
+/* File: armv5te/OP_IPUT_SHORT.S */
+@include "armv5te/OP_IPUT.S" { "store":"strh", "sqnum":"4" }
+/* File: armv5te/OP_IPUT.S */
+    /*
+     * General 32-bit instance field put.
+     *
+     * for: iput, iput-object, iput-boolean, iput-byte, iput-char, iput-short
+     */
+    /* op vA, vB, field@CCCC */
+    mov     r0, rINST, lsr #12          @ r0<- B
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref CCCC
+    ldr     r2, [r3, #offDvmDex_pResFields] @ r2<- pDvmDex->pResFields
+    GET_VREG(r9, r0)                    @ r9<- fp[B], the object pointer
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    bne     .LOP_IPUT_SHORT_finish          @ no, already resolved
+8:  ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveInstField         @ r0<- resolved InstField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_IPUT_SHORT_finish          @ yes, finish up
+    b       common_exceptionThrown
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET: /* 0x60 */
+/* File: armv5te/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_resolve         @ yes, do resolve
+.LOP_SGET_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_WIDE: /* 0x61 */
+/* File: armv5te/OP_SGET_WIDE.S */
+    /*
+     * 64-bit SGET handler.
+     */
+    /* sget-wide vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_WIDE_resolve         @ yes, do resolve
+.LOP_SGET_WIDE_finish:
+    mov     r1, rINST, lsr #8           @ r1<- AA
+    ldrd    r2, [r0, #offStaticField_value] @ r2/r3<- field value (aligned)
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[AA]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    stmia   r1, {r2-r3}                 @ vAA/vAA+1<- r2/r3
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_OBJECT: /* 0x62 */
+/* File: armv5te/OP_SGET_OBJECT.S */
+/* File: armv5te/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_OBJECT_resolve         @ yes, do resolve
+.LOP_SGET_OBJECT_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_BOOLEAN: /* 0x63 */
+/* File: armv5te/OP_SGET_BOOLEAN.S */
+/* File: armv5te/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_BOOLEAN_resolve         @ yes, do resolve
+.LOP_SGET_BOOLEAN_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_BYTE: /* 0x64 */
+/* File: armv5te/OP_SGET_BYTE.S */
+/* File: armv5te/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_BYTE_resolve         @ yes, do resolve
+.LOP_SGET_BYTE_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_CHAR: /* 0x65 */
+/* File: armv5te/OP_SGET_CHAR.S */
+/* File: armv5te/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_CHAR_resolve         @ yes, do resolve
+.LOP_SGET_CHAR_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SGET_SHORT: /* 0x66 */
+/* File: armv5te/OP_SGET_SHORT.S */
+/* File: armv5te/OP_SGET.S */
+    /*
+     * General 32-bit SGET handler.
+     *
+     * for: sget, sget-object, sget-boolean, sget-byte, sget-char, sget-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SGET_SHORT_resolve         @ yes, do resolve
+.LOP_SGET_SHORT_finish: @ field ptr in r0
+    ldr     r1, [r0, #offStaticField_value] @ r1<- field value
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r2)                    @ fp[AA]<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT: /* 0x67 */
+/* File: armv5te/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_resolve         @ yes, do resolve
+.LOP_SPUT_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_WIDE: /* 0x68 */
+/* File: armv5te/OP_SPUT_WIDE.S */
+    /*
+     * 64-bit SPUT handler.
+     */
+    /* sput-wide vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_WIDE_resolve         @ yes, do resolve
+.LOP_SPUT_WIDE_finish: @ field ptr in r0, AA in r9
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldmia   r9, {r2-r3}                 @ r2/r3<- vAA/vAA+1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strd    r2, [r0, #offStaticField_value] @ field<- vAA/vAA+1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_OBJECT: /* 0x69 */
+/* File: armv5te/OP_SPUT_OBJECT.S */
+/* File: armv5te/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_OBJECT_resolve         @ yes, do resolve
+.LOP_SPUT_OBJECT_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_BOOLEAN: /* 0x6a */
+/* File: armv5te/OP_SPUT_BOOLEAN.S */
+/* File: armv5te/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_BOOLEAN_resolve         @ yes, do resolve
+.LOP_SPUT_BOOLEAN_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_BYTE: /* 0x6b */
+/* File: armv5te/OP_SPUT_BYTE.S */
+/* File: armv5te/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_BYTE_resolve         @ yes, do resolve
+.LOP_SPUT_BYTE_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_CHAR: /* 0x6c */
+/* File: armv5te/OP_SPUT_CHAR.S */
+/* File: armv5te/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_CHAR_resolve         @ yes, do resolve
+.LOP_SPUT_CHAR_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SPUT_SHORT: /* 0x6d */
+/* File: armv5te/OP_SPUT_SHORT.S */
+/* File: armv5te/OP_SPUT.S */
+    /*
+     * General 32-bit SPUT handler.
+     *
+     * for: sput, sput-object, sput-boolean, sput-byte, sput-char, sput-short
+     */
+    /* op vAA, field@BBBB */
+    ldr     r2, [rGLUE, #offGlue_methodClassDex]    @ r2<- DvmDex
+    FETCH(r1, 1)                        @ r1<- field ref BBBB
+    ldr     r2, [r2, #offDvmDex_pResFields] @ r2<- dvmDex->pResFields
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ is resolved entry null?
+    beq     .LOP_SPUT_SHORT_resolve         @ yes, do resolve
+.LOP_SPUT_SHORT_finish:   @ field ptr in r0
+    mov     r2, rINST, lsr #8           @ r2<- AA
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_VREG(r1, r2)                    @ r1<- fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r1, [r0, #offStaticField_value] @ field<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_VIRTUAL: /* 0x6e */
+/* File: armv5te/OP_INVOKE_VIRTUAL.S */
+    /*
+     * Handle a virtual method call.
+     *
+     * for: invoke-virtual, invoke-virtual/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved baseMethod
+    .if     (!0)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    bne     .LOP_INVOKE_VIRTUAL_continue        @ yes, continue on
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    mov     r2, #METHOD_VIRTUAL         @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    bne     .LOP_INVOKE_VIRTUAL_continue        @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_SUPER: /* 0x6f */
+/* File: armv5te/OP_INVOKE_SUPER.S */
+    /*
+     * Handle a "super" method call.
+     *
+     * for: invoke-super, invoke-super/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    .if     (!0)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    GET_VREG(r2, r10)                   @ r2<- "this" ptr
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved baseMethod
+    cmp     r2, #0                      @ null "this"?
+    ldr     r9, [rGLUE, #offGlue_method] @ r9<- current method
+    beq     common_errNullObject        @ null "this", throw exception
+    cmp     r0, #0                      @ already resolved?
+    ldr     r9, [r9, #offMethod_clazz]  @ r9<- method->clazz
+    EXPORT_PC()                         @ must export for invoke
+    bne     .LOP_INVOKE_SUPER_continue        @ resolved, continue on
+    b       .LOP_INVOKE_SUPER_resolve         @ do resolve now
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_DIRECT: /* 0x70 */
+/* File: armv5te/OP_INVOKE_DIRECT.S */
+    /*
+     * Handle a direct method call.
+     *
+     * (We could defer the "is 'this' pointer null" test to the common
+     * method invocation code, and use a flag to indicate that static
+     * calls don't count.  If we do this as part of copying the arguments
+     * out we could avoiding loading the first arg twice.)
+     *
+     * for: invoke-direct, invoke-direct/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved methodToCall
+    .if     (!0)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    GET_VREG(r2, r10)                   @ r2<- "this" ptr
+    beq     .LOP_INVOKE_DIRECT_resolve         @ not resolved, do it now
+.LOP_INVOKE_DIRECT_finish:
+    cmp     r2, #0                      @ null "this" ref?
+    bne     common_invokeMethodNoRange   @ no, continue on
+    b       common_errNullObject        @ yes, throw exception
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_STATIC: /* 0x71 */
+/* File: armv5te/OP_INVOKE_STATIC.S */
+    /*
+     * Handle a static method call.
+     *
+     * for: invoke-static, invoke-static/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved methodToCall
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    bne     common_invokeMethodNoRange @ yes, continue on
+0:  ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    mov     r2, #METHOD_STATIC          @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    bne     common_invokeMethodNoRange @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_INTERFACE: /* 0x72 */
+/* File: armv5te/OP_INVOKE_INTERFACE.S */
+    /*
+     * Handle an interface method call.
+     *
+     * for: invoke-interface, invoke-interface/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r2, 2)                        @ r2<- FEDC or CCCC
+    FETCH(r1, 1)                        @ r1<- BBBB
+    .if     (!0)
+    and     r2, r2, #15                 @ r2<- C (or stays CCCC)
+    .endif
+    EXPORT_PC()                         @ must export for invoke
+    GET_VREG(r0, r2)                    @ r0<- first arg ("this")
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- methodClassDex
+    cmp     r0, #0                      @ null obj?
+    ldr     r2, [rGLUE, #offGlue_method]  @ r2<- method
+    beq     common_errNullObject        @ yes, fail
+    ldr     r0, [r0, #offObject_clazz]  @ r0<- thisPtr->clazz
+    bl      dvmFindInterfaceMethodInCache @ r0<- call(class, ref, method, dex)
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yes, handle exception
+    b       common_invokeMethodNoRange @ jump to common handler 
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_73: /* 0x73 */
+/* File: armv5te/OP_UNUSED_73.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_VIRTUAL_RANGE: /* 0x74 */
+/* File: armv5te/OP_INVOKE_VIRTUAL_RANGE.S */
+/* File: armv5te/OP_INVOKE_VIRTUAL.S */
+    /*
+     * Handle a virtual method call.
+     *
+     * for: invoke-virtual, invoke-virtual/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved baseMethod
+    .if     (!1)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    bne     .LOP_INVOKE_VIRTUAL_RANGE_continue        @ yes, continue on
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    mov     r2, #METHOD_VIRTUAL         @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    bne     .LOP_INVOKE_VIRTUAL_RANGE_continue        @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_SUPER_RANGE: /* 0x75 */
+/* File: armv5te/OP_INVOKE_SUPER_RANGE.S */
+/* File: armv5te/OP_INVOKE_SUPER.S */
+    /*
+     * Handle a "super" method call.
+     *
+     * for: invoke-super, invoke-super/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    .if     (!1)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    GET_VREG(r2, r10)                   @ r2<- "this" ptr
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved baseMethod
+    cmp     r2, #0                      @ null "this"?
+    ldr     r9, [rGLUE, #offGlue_method] @ r9<- current method
+    beq     common_errNullObject        @ null "this", throw exception
+    cmp     r0, #0                      @ already resolved?
+    ldr     r9, [r9, #offMethod_clazz]  @ r9<- method->clazz
+    EXPORT_PC()                         @ must export for invoke
+    bne     .LOP_INVOKE_SUPER_RANGE_continue        @ resolved, continue on
+    b       .LOP_INVOKE_SUPER_RANGE_resolve         @ do resolve now
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_DIRECT_RANGE: /* 0x76 */
+/* File: armv5te/OP_INVOKE_DIRECT_RANGE.S */
+/* File: armv5te/OP_INVOKE_DIRECT.S */
+    /*
+     * Handle a direct method call.
+     *
+     * (We could defer the "is 'this' pointer null" test to the common
+     * method invocation code, and use a flag to indicate that static
+     * calls don't count.  If we do this as part of copying the arguments
+     * out we could avoiding loading the first arg twice.)
+     *
+     * for: invoke-direct, invoke-direct/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved methodToCall
+    .if     (!1)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    GET_VREG(r2, r10)                   @ r2<- "this" ptr
+    beq     .LOP_INVOKE_DIRECT_RANGE_resolve         @ not resolved, do it now
+.LOP_INVOKE_DIRECT_RANGE_finish:
+    cmp     r2, #0                      @ null "this" ref?
+    bne     common_invokeMethodRange   @ no, continue on
+    b       common_errNullObject        @ yes, throw exception
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_STATIC_RANGE: /* 0x77 */
+/* File: armv5te/OP_INVOKE_STATIC_RANGE.S */
+/* File: armv5te/OP_INVOKE_STATIC.S */
+    /*
+     * Handle a static method call.
+     *
+     * for: invoke-static, invoke-static/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- pDvmDex
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r3, [r3, #offDvmDex_pResMethods]    @ r3<- pDvmDex->pResMethods
+    ldr     r0, [r3, r1, lsl #2]        @ r0<- resolved methodToCall
+    cmp     r0, #0                      @ already resolved?
+    EXPORT_PC()                         @ must export for invoke
+    bne     common_invokeMethodRange @ yes, continue on
+0:  ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    mov     r2, #METHOD_STATIC          @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    bne     common_invokeMethodRange @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_INTERFACE_RANGE: /* 0x78 */
+/* File: armv5te/OP_INVOKE_INTERFACE_RANGE.S */
+/* File: armv5te/OP_INVOKE_INTERFACE.S */
+    /*
+     * Handle an interface method call.
+     *
+     * for: invoke-interface, invoke-interface/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r2, 2)                        @ r2<- FEDC or CCCC
+    FETCH(r1, 1)                        @ r1<- BBBB
+    .if     (!1)
+    and     r2, r2, #15                 @ r2<- C (or stays CCCC)
+    .endif
+    EXPORT_PC()                         @ must export for invoke
+    GET_VREG(r0, r2)                    @ r0<- first arg ("this")
+    ldr     r3, [rGLUE, #offGlue_methodClassDex]    @ r3<- methodClassDex
+    cmp     r0, #0                      @ null obj?
+    ldr     r2, [rGLUE, #offGlue_method]  @ r2<- method
+    beq     common_errNullObject        @ yes, fail
+    ldr     r0, [r0, #offObject_clazz]  @ r0<- thisPtr->clazz
+    bl      dvmFindInterfaceMethodInCache @ r0<- call(class, ref, method, dex)
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yes, handle exception
+    b       common_invokeMethodRange @ jump to common handler 
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_79: /* 0x79 */
+/* File: armv5te/OP_UNUSED_79.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_7A: /* 0x7a */
+/* File: armv5te/OP_UNUSED_7A.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEG_INT: /* 0x7b */
+/* File: armv5te/OP_NEG_INT.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    rsb     r0, r0, #0                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NOT_INT: /* 0x7c */
+/* File: armv5te/OP_NOT_INT.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mvn     r0, r0                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEG_LONG: /* 0x7d */
+/* File: armv5te/OP_NEG_LONG.S */
+/* File: armv5te/unopWide.S */
+    /*
+     * Generic 64-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0/r1".
+     * This could be an ARM instruction or a function call.
+     *
+     * For: neg-long, not-long, neg-double, long-to-double, double-to-long
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vAA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    rsbs    r0, r0, #0                           @ optional op; may set condition codes
+    rsc     r1, r1, #0                              @ r0/r1<- op, r2-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NOT_LONG: /* 0x7e */
+/* File: armv5te/OP_NOT_LONG.S */
+/* File: armv5te/unopWide.S */
+    /*
+     * Generic 64-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0/r1".
+     * This could be an ARM instruction or a function call.
+     *
+     * For: neg-long, not-long, neg-double, long-to-double, double-to-long
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vAA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mvn     r0, r0                           @ optional op; may set condition codes
+    mvn     r1, r1                              @ r0/r1<- op, r2-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEG_FLOAT: /* 0x7f */
+/* File: armv5te/OP_NEG_FLOAT.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    add     r0, r0, #0x80000000                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_NEG_DOUBLE: /* 0x80 */
+/* File: armv5te/OP_NEG_DOUBLE.S */
+/* File: armv5te/unopWide.S */
+    /*
+     * Generic 64-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0/r1".
+     * This could be an ARM instruction or a function call.
+     *
+     * For: neg-long, not-long, neg-double, long-to-double, double-to-long
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vAA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    add     r1, r1, #0x80000000                              @ r0/r1<- op, r2-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_LONG: /* 0x81 */
+/* File: armv5te/OP_INT_TO_LONG.S */
+/* File: armv5te/unopWider.S */
+    /*
+     * Generic 32bit-to-64bit unary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = op r0", where
+     * "result" is a 64-bit quantity in r0/r1.
+     *
+     * For: int-to-long, int-to-double, float-to-long, float-to-double
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r3)                    @ r0<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mov     r1, r0, asr #31                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vA/vA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-11 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_FLOAT: /* 0x82 */
+/* File: vfp/OP_INT_TO_FLOAT.S */
+/* File: vfp/funop.S */
+    /*
+     * Generic 32-bit unary floating-point operation.  Provide an "instr"
+     * line that specifies an instruction that performs "s1 = op s0".
+     *
+     * for: int-to-float, float-to-int
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    flds    s0, [r3]                    @ s0<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    and     r9, r9, #15                 @ r9<- A
+    fsitos  s1, s0                              @ s1<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    fsts    s1, [r9]                    @ vA<- s1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_DOUBLE: /* 0x83 */
+/* File: vfp/OP_INT_TO_DOUBLE.S */
+/* File: vfp/funopWider.S */
+    /*
+     * Generic 32bit-to-64bit floating point unary operation.  Provide an
+     * "instr" line that specifies an instruction that performs "d0 = op s0".
+     *
+     * For: int-to-double, float-to-double
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    flds    s0, [r3]                    @ s0<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    and     r9, r9, #15                 @ r9<- A
+    fsitod  d0, s0                              @ d0<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    fstd    d0, [r9]                    @ vA<- d0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_LONG_TO_INT: /* 0x84 */
+/* File: armv5te/OP_LONG_TO_INT.S */
+/* we ignore the high word, making this equivalent to a 32-bit reg move */
+/* File: armv5te/OP_MOVE.S */
+    /* for move, move-object, long-to-int */
+    /* op vA, vB */
+    mov     r1, rINST, lsr #12          @ r1<- B from 15:12
+    mov     r0, rINST, lsr #8           @ r0<- A from 11:8
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    GET_VREG(r2, r1)                    @ r2<- fp[B]
+    and     r0, r0, #15
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    SET_VREG(r2, r0)                    @ fp[A]<- r2
+    GOTO_OPCODE(ip)                     @ execute next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_LONG_TO_FLOAT: /* 0x85 */
+/* File: armv5te/OP_LONG_TO_FLOAT.S */
+/* File: armv5te/unopNarrower.S */
+    /*
+     * Generic 64bit-to-32bit unary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = op r0/r1", where
+     * "result" is a 32-bit quantity in r0.
+     *
+     * For: long-to-float, double-to-int, double-to-float
+     *
+     * (This would work for long-to-int, but that instruction is actually
+     * an exact match for OP_MOVE.)
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    and     r9, r9, #15
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vB/vB+1
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_l2f                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-11 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_LONG_TO_DOUBLE: /* 0x86 */
+/* File: armv5te/OP_LONG_TO_DOUBLE.S */
+/* File: armv5te/unopWide.S */
+    /*
+     * Generic 64-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0/r1".
+     * This could be an ARM instruction or a function call.
+     *
+     * For: neg-long, not-long, neg-double, long-to-double, double-to-long
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vAA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_l2d                              @ r0/r1<- op, r2-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FLOAT_TO_INT: /* 0x87 */
+/* File: vfp/OP_FLOAT_TO_INT.S */
+/* File: vfp/funop.S */
+    /*
+     * Generic 32-bit unary floating-point operation.  Provide an "instr"
+     * line that specifies an instruction that performs "s1 = op s0".
+     *
+     * for: int-to-float, float-to-int
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    flds    s0, [r3]                    @ s0<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    and     r9, r9, #15                 @ r9<- A
+    ftosizs s1, s0                              @ s1<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    fsts    s1, [r9]                    @ vA<- s1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FLOAT_TO_LONG: /* 0x88 */
+/* File: armv5te/OP_FLOAT_TO_LONG.S */
+@include "armv5te/unopWider.S" {"instr":"bl      __aeabi_f2lz"}
+/* File: armv5te/unopWider.S */
+    /*
+     * Generic 32bit-to-64bit unary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = op r0", where
+     * "result" is a 64-bit quantity in r0/r1.
+     *
+     * For: int-to-long, int-to-double, float-to-long, float-to-double
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r3)                    @ r0<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+                               @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    bl      f2l_doconv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vA/vA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-11 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_FLOAT_TO_DOUBLE: /* 0x89 */
+/* File: vfp/OP_FLOAT_TO_DOUBLE.S */
+/* File: vfp/funopWider.S */
+    /*
+     * Generic 32bit-to-64bit floating point unary operation.  Provide an
+     * "instr" line that specifies an instruction that performs "d0 = op s0".
+     *
+     * For: int-to-double, float-to-double
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    flds    s0, [r3]                    @ s0<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    and     r9, r9, #15                 @ r9<- A
+    fcvtds  d0, s0                              @ d0<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    fstd    d0, [r9]                    @ vA<- d0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DOUBLE_TO_INT: /* 0x8a */
+/* File: vfp/OP_DOUBLE_TO_INT.S */
+/* File: vfp/funopNarrower.S */
+    /*
+     * Generic 64bit-to-32bit unary floating point operation.  Provide an
+     * "instr" line that specifies an instruction that performs "s0 = op d0".
+     *
+     * For: double-to-int, double-to-float
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    fldd    d0, [r3]                    @ d0<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    and     r9, r9, #15                 @ r9<- A
+    ftosizd  s0, d0                              @ s0<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    fsts    s0, [r9]                    @ vA<- s0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DOUBLE_TO_LONG: /* 0x8b */
+/* File: armv5te/OP_DOUBLE_TO_LONG.S */
+@include "armv5te/unopWide.S" {"instr":"bl      __aeabi_d2lz"}
+/* File: armv5te/unopWide.S */
+    /*
+     * Generic 64-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0/r1".
+     * This could be an ARM instruction or a function call.
+     *
+     * For: neg-long, not-long, neg-double, long-to-double, double-to-long
+     */
+    /* unop vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vAA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      d2l_doconv                              @ r0/r1<- op, r2-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-13 instructions */
+
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DOUBLE_TO_FLOAT: /* 0x8c */
+/* File: vfp/OP_DOUBLE_TO_FLOAT.S */
+/* File: vfp/funopNarrower.S */
+    /*
+     * Generic 64bit-to-32bit unary floating point operation.  Provide an
+     * "instr" line that specifies an instruction that performs "s0 = op d0".
+     *
+     * For: double-to-int, double-to-float
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    fldd    d0, [r3]                    @ d0<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    and     r9, r9, #15                 @ r9<- A
+    fcvtsd  s0, d0                              @ s0<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    fsts    s0, [r9]                    @ vA<- s0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_BYTE: /* 0x8d */
+/* File: armv5te/OP_INT_TO_BYTE.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+    mov     r0, r0, asl #24                           @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mov     r0, r0, asr #24                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_CHAR: /* 0x8e */
+/* File: armv5te/OP_INT_TO_CHAR.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+    mov     r0, r0, asl #16                           @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mov     r0, r0, lsr #16                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INT_TO_SHORT: /* 0x8f */
+/* File: armv5te/OP_INT_TO_SHORT.S */
+/* File: armv5te/unop.S */
+    /*
+     * Generic 32-bit unary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = op r0".
+     * This could be an ARM instruction or a function call.
+     *
+     * for: neg-int, not-int, neg-float, int-to-float, float-to-int,
+     *      int-to-byte, int-to-char, int-to-short
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r3)                    @ r0<- vB
+    and     r9, r9, #15
+    mov     r0, r0, asl #16                           @ optional op; may set condition codes
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    mov     r0, r0, asr #16                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 9-10 instructions */
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_INT: /* 0x90 */
+/* File: armv5te/OP_ADD_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    add     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_INT: /* 0x91 */
+/* File: armv5te/OP_SUB_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    sub     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_INT: /* 0x92 */
+/* File: armv5te/OP_MUL_INT.S */
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    mul     r0, r1, r0                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_INT: /* 0x93 */
+/* File: armv5te/OP_DIV_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 1
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl     __aeabi_idiv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_INT: /* 0x94 */
+/* File: armv5te/OP_REM_INT.S */
+/* idivmod returns quotient in r0 and remainder in r1 */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 1
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      __aeabi_idivmod                              @ r1<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r1, r9)               @ vAA<- r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_INT: /* 0x95 */
+/* File: armv5te/OP_AND_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    and     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_INT: /* 0x96 */
+/* File: armv5te/OP_OR_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    orr     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_INT: /* 0x97 */
+/* File: armv5te/OP_XOR_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    eor     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_INT: /* 0x98 */
+/* File: armv5te/OP_SHL_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, asl r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_INT: /* 0x99 */
+/* File: armv5te/OP_SHR_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, asr r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_INT: /* 0x9a */
+/* File: armv5te/OP_USHR_INT.S */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, lsr r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_LONG: /* 0x9b */
+/* File: armv5te/OP_ADD_LONG.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    adds    r0, r0, r2                           @ optional op; may set condition codes
+    adc     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_LONG: /* 0x9c */
+/* File: armv5te/OP_SUB_LONG.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    subs    r0, r0, r2                           @ optional op; may set condition codes
+    sbc     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_LONG: /* 0x9d */
+/* File: armv5te/OP_MUL_LONG.S */
+    /*
+     * Signed 64-bit integer multiply.
+     *
+     * Consider WXxYZ (r1r0 x r3r2) with a long multiply:
+     *        WX
+     *      x YZ
+     *  --------
+     *     ZW ZX
+     *  YW YX
+     *
+     * The low word of the result holds ZX, the high word holds
+     * (ZW+YX) + (the high overflow from ZX).  YW doesn't matter because
+     * it doesn't fit in the low 64 bits.
+     *
+     * Unlike most ARM math operations, multiply instructions have
+     * restrictions on using the same register more than once (Rd and Rm
+     * cannot be the same).
+     */
+    /* mul-long vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    mul     ip, r2, r1                  @  ip<- ZxW
+    umull   r9, r10, r2, r0             @  r9/r10 <- ZxX
+    mla     r2, r0, r3, ip              @  r2<- YxX + (ZxW)
+    mov     r0, rINST, lsr #8           @ r0<- AA
+    add     r10, r2, r10                @  r10<- r10 + low(ZxW + (YxX))
+    add     r0, rFP, r0, lsl #2         @ r0<- &fp[AA]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    b       .LOP_MUL_LONG_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_LONG: /* 0x9e */
+/* File: armv5te/OP_DIV_LONG.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 1
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_ldivmod                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_LONG: /* 0x9f */
+/* File: armv5te/OP_REM_LONG.S */
+/* ldivmod returns quotient in r0/r1 and remainder in r2/r3 */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 1
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_ldivmod                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r2,r3}     @ vAA/vAA+1<- r2/r3
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_LONG: /* 0xa0 */
+/* File: armv5te/OP_AND_LONG.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    and     r0, r0, r2                           @ optional op; may set condition codes
+    and     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_LONG: /* 0xa1 */
+/* File: armv5te/OP_OR_LONG.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    orr     r0, r0, r2                           @ optional op; may set condition codes
+    orr     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_LONG: /* 0xa2 */
+/* File: armv5te/OP_XOR_LONG.S */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    eor     r0, r0, r2                           @ optional op; may set condition codes
+    eor     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_LONG: /* 0xa3 */
+/* File: armv5te/OP_SHL_LONG.S */
+    /*
+     * Long integer shift.  This is different from the generic 32/64-bit
+     * binary operations because vAA/vBB are 64-bit but vCC (the shift
+     * distance) is 32-bit.  Also, Dalvik requires us to mask off the low
+     * 6 bits of the shift distance.
+     */
+    /* shl-long vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r3, r0, #255                @ r3<- BB
+    mov     r0, r0, lsr #8              @ r0<- CC
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[BB]
+    GET_VREG(r2, r0)                    @ r2<- vCC
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    and     r2, r2, #63                 @ r2<- r2 & 0x3f
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+
+    mov     r1, r1, asl r2              @  r1<- r1 << r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r1, r1, r0, lsr r3          @  r1<- r1 | (r0 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    movpl   r1, r0, asl ip              @  if r2 >= 32, r1<- r0 << (r2-32)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    b       .LOP_SHL_LONG_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_LONG: /* 0xa4 */
+/* File: armv5te/OP_SHR_LONG.S */
+    /*
+     * Long integer shift.  This is different from the generic 32/64-bit
+     * binary operations because vAA/vBB are 64-bit but vCC (the shift
+     * distance) is 32-bit.  Also, Dalvik requires us to mask off the low
+     * 6 bits of the shift distance.
+     */
+    /* shr-long vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r3, r0, #255                @ r3<- BB
+    mov     r0, r0, lsr #8              @ r0<- CC
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[BB]
+    GET_VREG(r2, r0)                    @ r2<- vCC
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    and     r2, r2, #63                 @ r0<- r0 & 0x3f
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+
+    mov     r0, r0, lsr r2              @  r0<- r2 >> r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r0, r0, r1, asl r3          @  r0<- r0 | (r1 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    movpl   r0, r1, asr ip              @  if r2 >= 32, r0<-r1 >> (r2-32)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    b       .LOP_SHR_LONG_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_LONG: /* 0xa5 */
+/* File: armv5te/OP_USHR_LONG.S */
+    /*
+     * Long integer shift.  This is different from the generic 32/64-bit
+     * binary operations because vAA/vBB are 64-bit but vCC (the shift
+     * distance) is 32-bit.  Also, Dalvik requires us to mask off the low
+     * 6 bits of the shift distance.
+     */
+    /* ushr-long vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r3, r0, #255                @ r3<- BB
+    mov     r0, r0, lsr #8              @ r0<- CC
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[BB]
+    GET_VREG(r2, r0)                    @ r2<- vCC
+    ldmia   r3, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    and     r2, r2, #63                 @ r0<- r0 & 0x3f
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+
+    mov     r0, r0, lsr r2              @  r0<- r2 >> r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r0, r0, r1, asl r3          @  r0<- r0 | (r1 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    movpl   r0, r1, lsr ip              @  if r2 >= 32, r0<-r1 >>> (r2-32)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    b       .LOP_USHR_LONG_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_FLOAT: /* 0xa6 */
+/* File: vfp/OP_ADD_FLOAT.S */
+/* File: vfp/fbinop.S */
+    /*
+     * Generic 32-bit floating-point operation.  Provide an "instr" line that
+     * specifies an instruction that performs "s2 = s0 op s1".  Because we
+     * use the "softfp" ABI, this must be an instruction, not a function call.
+     *
+     * For: add-float, sub-float, mul-float, div-float
+     */
+    /* floatop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vCC
+    flds    s1, [r3]                    @ s1<- vCC
+    and     r2, r0, #255                @ r2<- BB
+    VREG_INDEX_TO_ADDR(r2, r2)          @ r2<- &vBB
+    flds    s0, [r2]                    @ s0<- vBB
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    fadds   s2, s0, s1                              @ s2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vAA
+    fsts    s2, [r9]                    @ vAA<- s2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_FLOAT: /* 0xa7 */
+/* File: vfp/OP_SUB_FLOAT.S */
+/* File: vfp/fbinop.S */
+    /*
+     * Generic 32-bit floating-point operation.  Provide an "instr" line that
+     * specifies an instruction that performs "s2 = s0 op s1".  Because we
+     * use the "softfp" ABI, this must be an instruction, not a function call.
+     *
+     * For: add-float, sub-float, mul-float, div-float
+     */
+    /* floatop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vCC
+    flds    s1, [r3]                    @ s1<- vCC
+    and     r2, r0, #255                @ r2<- BB
+    VREG_INDEX_TO_ADDR(r2, r2)          @ r2<- &vBB
+    flds    s0, [r2]                    @ s0<- vBB
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    fsubs   s2, s0, s1                              @ s2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vAA
+    fsts    s2, [r9]                    @ vAA<- s2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_FLOAT: /* 0xa8 */
+/* File: vfp/OP_MUL_FLOAT.S */
+/* File: vfp/fbinop.S */
+    /*
+     * Generic 32-bit floating-point operation.  Provide an "instr" line that
+     * specifies an instruction that performs "s2 = s0 op s1".  Because we
+     * use the "softfp" ABI, this must be an instruction, not a function call.
+     *
+     * For: add-float, sub-float, mul-float, div-float
+     */
+    /* floatop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vCC
+    flds    s1, [r3]                    @ s1<- vCC
+    and     r2, r0, #255                @ r2<- BB
+    VREG_INDEX_TO_ADDR(r2, r2)          @ r2<- &vBB
+    flds    s0, [r2]                    @ s0<- vBB
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    fmuls   s2, s0, s1                              @ s2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vAA
+    fsts    s2, [r9]                    @ vAA<- s2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_FLOAT: /* 0xa9 */
+/* File: vfp/OP_DIV_FLOAT.S */
+/* File: vfp/fbinop.S */
+    /*
+     * Generic 32-bit floating-point operation.  Provide an "instr" line that
+     * specifies an instruction that performs "s2 = s0 op s1".  Because we
+     * use the "softfp" ABI, this must be an instruction, not a function call.
+     *
+     * For: add-float, sub-float, mul-float, div-float
+     */
+    /* floatop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vCC
+    flds    s1, [r3]                    @ s1<- vCC
+    and     r2, r0, #255                @ r2<- BB
+    VREG_INDEX_TO_ADDR(r2, r2)          @ r2<- &vBB
+    flds    s0, [r2]                    @ s0<- vBB
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    fdivs   s2, s0, s1                              @ s2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vAA
+    fsts    s2, [r9]                    @ vAA<- s2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_FLOAT: /* 0xaa */
+/* File: armv5te/OP_REM_FLOAT.S */
+/* EABI doesn't define a float remainder function, but libm does */
+/* File: armv5te/binop.S */
+    /*
+     * Generic 32-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.  Note that we
+     * *don't* check for (INT_MIN / -1) here, because the ARM math lib
+     * handles it correctly.
+     *
+     * For: add-int, sub-int, mul-int, div-int, rem-int, and-int, or-int,
+     *      xor-int, shl-int, shr-int, ushr-int, add-float, sub-float,
+     *      mul-float, div-float, rem-float
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    and     r2, r0, #255                @ r2<- BB
+    GET_VREG(r1, r3)                    @ r1<- vCC
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+                               @ optional op; may set condition codes
+    bl      fmodf                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 11-14 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_DOUBLE: /* 0xab */
+/* File: vfp/OP_ADD_DOUBLE.S */
+/* File: vfp/fbinopWide.S */
+    /*
+     * Generic 64-bit double-precision floating point binary operation.
+     * Provide an "instr" line that specifies an instruction that performs
+     * "d2 = d0 op d1".
+     *
+     * for: add-double, sub-double, mul-double, div-double
+     */
+    /* doubleop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vCC
+    fldd    d1, [r3]                    @ d1<- vCC
+    and     r2, r0, #255                @ r2<- BB
+    VREG_INDEX_TO_ADDR(r2, r2)          @ r2<- &vBB
+    fldd    d0, [r2]                    @ d0<- vBB
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    faddd   d2, d0, d1                              @ s2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vAA
+    fstd    d2, [r9]                    @ vAA<- d2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_DOUBLE: /* 0xac */
+/* File: vfp/OP_SUB_DOUBLE.S */
+/* File: vfp/fbinopWide.S */
+    /*
+     * Generic 64-bit double-precision floating point binary operation.
+     * Provide an "instr" line that specifies an instruction that performs
+     * "d2 = d0 op d1".
+     *
+     * for: add-double, sub-double, mul-double, div-double
+     */
+    /* doubleop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vCC
+    fldd    d1, [r3]                    @ d1<- vCC
+    and     r2, r0, #255                @ r2<- BB
+    VREG_INDEX_TO_ADDR(r2, r2)          @ r2<- &vBB
+    fldd    d0, [r2]                    @ d0<- vBB
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    fsubd   d2, d0, d1                              @ s2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vAA
+    fstd    d2, [r9]                    @ vAA<- d2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_DOUBLE: /* 0xad */
+/* File: vfp/OP_MUL_DOUBLE.S */
+/* File: vfp/fbinopWide.S */
+    /*
+     * Generic 64-bit double-precision floating point binary operation.
+     * Provide an "instr" line that specifies an instruction that performs
+     * "d2 = d0 op d1".
+     *
+     * for: add-double, sub-double, mul-double, div-double
+     */
+    /* doubleop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vCC
+    fldd    d1, [r3]                    @ d1<- vCC
+    and     r2, r0, #255                @ r2<- BB
+    VREG_INDEX_TO_ADDR(r2, r2)          @ r2<- &vBB
+    fldd    d0, [r2]                    @ d0<- vBB
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    fmuld   d2, d0, d1                              @ s2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vAA
+    fstd    d2, [r9]                    @ vAA<- d2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_DOUBLE: /* 0xae */
+/* File: vfp/OP_DIV_DOUBLE.S */
+/* File: vfp/fbinopWide.S */
+    /*
+     * Generic 64-bit double-precision floating point binary operation.
+     * Provide an "instr" line that specifies an instruction that performs
+     * "d2 = d0 op d1".
+     *
+     * for: add-double, sub-double, mul-double, div-double
+     */
+    /* doubleop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vCC
+    fldd    d1, [r3]                    @ d1<- vCC
+    and     r2, r0, #255                @ r2<- BB
+    VREG_INDEX_TO_ADDR(r2, r2)          @ r2<- &vBB
+    fldd    d0, [r2]                    @ d0<- vBB
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    fdivd   d2, d0, d1                              @ s2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vAA
+    fstd    d2, [r9]                    @ vAA<- d2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_DOUBLE: /* 0xaf */
+/* File: armv5te/OP_REM_DOUBLE.S */
+/* EABI doesn't define a double remainder function, but libm does */
+/* File: armv5te/binopWide.S */
+    /*
+     * Generic 64-bit binary operation.  Provide an "instr" line that
+     * specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * for: add-long, sub-long, div-long, rem-long, and-long, or-long,
+     *      xor-long, add-double, sub-double, mul-double, div-double,
+     *      rem-double
+     *
+     * IMPORTANT: you may specify "chkzero" or "preinstr" but not both.
+     */
+    /* binop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r0, #255                @ r2<- BB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    add     r2, rFP, r2, lsl #2         @ r2<- &fp[BB]
+    add     r3, rFP, r3, lsl #2         @ r3<- &fp[CC]
+    ldmia   r2, {r0-r1}                 @ r0/r1<- vBB/vBB+1
+    ldmia   r3, {r2-r3}                 @ r2/r3<- vCC/vCC+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      fmod                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 14-17 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_INT_2ADDR: /* 0xb0 */
+/* File: armv5te/OP_ADD_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    add     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_INT_2ADDR: /* 0xb1 */
+/* File: armv5te/OP_SUB_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    sub     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_INT_2ADDR: /* 0xb2 */
+/* File: armv5te/OP_MUL_INT_2ADDR.S */
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    mul     r0, r1, r0                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_INT_2ADDR: /* 0xb3 */
+/* File: armv5te/OP_DIV_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 1
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl     __aeabi_idiv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_INT_2ADDR: /* 0xb4 */
+/* File: armv5te/OP_REM_INT_2ADDR.S */
+/* idivmod returns quotient in r0 and remainder in r1 */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 1
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_idivmod                              @ r1<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r1, r9)               @ vAA<- r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_INT_2ADDR: /* 0xb5 */
+/* File: armv5te/OP_AND_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    and     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_INT_2ADDR: /* 0xb6 */
+/* File: armv5te/OP_OR_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    orr     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_INT_2ADDR: /* 0xb7 */
+/* File: armv5te/OP_XOR_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    eor     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_INT_2ADDR: /* 0xb8 */
+/* File: armv5te/OP_SHL_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, asl r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_INT_2ADDR: /* 0xb9 */
+/* File: armv5te/OP_SHR_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, asr r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_INT_2ADDR: /* 0xba */
+/* File: armv5te/OP_USHR_INT_2ADDR.S */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, lsr r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_LONG_2ADDR: /* 0xbb */
+/* File: armv5te/OP_ADD_LONG_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    adds    r0, r0, r2                           @ optional op; may set condition codes
+    adc     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_LONG_2ADDR: /* 0xbc */
+/* File: armv5te/OP_SUB_LONG_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    subs    r0, r0, r2                           @ optional op; may set condition codes
+    sbc     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_LONG_2ADDR: /* 0xbd */
+/* File: armv5te/OP_MUL_LONG_2ADDR.S */
+    /*
+     * Signed 64-bit integer multiply, "/2addr" version.
+     *
+     * See OP_MUL_LONG for an explanation.
+     *
+     * We get a little tight on registers, so to avoid looking up &fp[A]
+     * again we stuff it into rINST.
+     */
+    /* mul-long/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     rINST, rFP, r9, lsl #2      @ rINST<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   rINST, {r0-r1}              @ r0/r1<- vAA/vAA+1
+    mul     ip, r2, r1                  @  ip<- ZxW
+    umull   r9, r10, r2, r0             @  r9/r10 <- ZxX
+    mla     r2, r0, r3, ip              @  r2<- YxX + (ZxW)
+    mov     r0, rINST                   @ r0<- &fp[A] (free up rINST)
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    add     r10, r2, r10                @  r10<- r10 + low(ZxW + (YxX))
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r0, {r9-r10}                @ vAA/vAA+1<- r9/r10
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_LONG_2ADDR: /* 0xbe */
+/* File: armv5te/OP_DIV_LONG_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 1
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_ldivmod                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_LONG_2ADDR: /* 0xbf */
+/* File: armv5te/OP_REM_LONG_2ADDR.S */
+/* ldivmod returns quotient in r0/r1 and remainder in r2/r3 */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 1
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_ldivmod                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r2,r3}     @ vAA/vAA+1<- r2/r3
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_LONG_2ADDR: /* 0xc0 */
+/* File: armv5te/OP_AND_LONG_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    and     r0, r0, r2                           @ optional op; may set condition codes
+    and     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_LONG_2ADDR: /* 0xc1 */
+/* File: armv5te/OP_OR_LONG_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    orr     r0, r0, r2                           @ optional op; may set condition codes
+    orr     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_LONG_2ADDR: /* 0xc2 */
+/* File: armv5te/OP_XOR_LONG_2ADDR.S */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+    eor     r0, r0, r2                           @ optional op; may set condition codes
+    eor     r1, r1, r3                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_LONG_2ADDR: /* 0xc3 */
+/* File: armv5te/OP_SHL_LONG_2ADDR.S */
+    /*
+     * Long integer shift, 2addr version.  vA is 64-bit value/result, vB is
+     * 32-bit shift distance.
+     */
+    /* shl-long/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r2, r3)                    @ r2<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    and     r2, r2, #63                 @ r2<- r2 & 0x3f
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+
+    mov     r1, r1, asl r2              @  r1<- r1 << r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r1, r1, r0, lsr r3          @  r1<- r1 | (r0 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    movpl   r1, r0, asl ip              @  if r2 >= 32, r1<- r0 << (r2-32)
+    mov     r0, r0, asl r2              @  r0<- r0 << r2
+    b       .LOP_SHL_LONG_2ADDR_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_LONG_2ADDR: /* 0xc4 */
+/* File: armv5te/OP_SHR_LONG_2ADDR.S */
+    /*
+     * Long integer shift, 2addr version.  vA is 64-bit value/result, vB is
+     * 32-bit shift distance.
+     */
+    /* shr-long/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r2, r3)                    @ r2<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    and     r2, r2, #63                 @ r2<- r2 & 0x3f
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+
+    mov     r0, r0, lsr r2              @  r0<- r2 >> r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r0, r0, r1, asl r3          @  r0<- r0 | (r1 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    movpl   r0, r1, asr ip              @  if r2 >= 32, r0<-r1 >> (r2-32)
+    mov     r1, r1, asr r2              @  r1<- r1 >> r2
+    b       .LOP_SHR_LONG_2ADDR_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_LONG_2ADDR: /* 0xc5 */
+/* File: armv5te/OP_USHR_LONG_2ADDR.S */
+    /*
+     * Long integer shift, 2addr version.  vA is 64-bit value/result, vB is
+     * 32-bit shift distance.
+     */
+    /* ushr-long/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r2, r3)                    @ r2<- vB
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    and     r2, r2, #63                 @ r2<- r2 & 0x3f
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+
+    mov     r0, r0, lsr r2              @  r0<- r2 >> r2
+    rsb     r3, r2, #32                 @  r3<- 32 - r2
+    orr     r0, r0, r1, asl r3          @  r0<- r0 | (r1 << (32-r2))
+    subs    ip, r2, #32                 @  ip<- r2 - 32
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    movpl   r0, r1, lsr ip              @  if r2 >= 32, r0<-r1 >>> (r2-32)
+    mov     r1, r1, lsr r2              @  r1<- r1 >>> r2
+    b       .LOP_USHR_LONG_2ADDR_finish
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_FLOAT_2ADDR: /* 0xc6 */
+/* File: vfp/OP_ADD_FLOAT_2ADDR.S */
+/* File: vfp/fbinop2addr.S */
+    /*
+     * Generic 32-bit floating point "/2addr" binary operation.  Provide
+     * an "instr" line that specifies an instruction that performs
+     * "s2 = s0 op s1".
+     *
+     * For: add-float/2addr, sub-float/2addr, mul-float/2addr, div-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    flds    s1, [r3]                    @ s1<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    and     r9, r9, #15                 @ r9<- A
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    flds    s0, [r9]                    @ s0<- vA
+
+    fadds   s2, s0, s1                              @ s2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    fsts    s2, [r9]                    @ vAA<- s2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_FLOAT_2ADDR: /* 0xc7 */
+/* File: vfp/OP_SUB_FLOAT_2ADDR.S */
+/* File: vfp/fbinop2addr.S */
+    /*
+     * Generic 32-bit floating point "/2addr" binary operation.  Provide
+     * an "instr" line that specifies an instruction that performs
+     * "s2 = s0 op s1".
+     *
+     * For: add-float/2addr, sub-float/2addr, mul-float/2addr, div-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    flds    s1, [r3]                    @ s1<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    and     r9, r9, #15                 @ r9<- A
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    flds    s0, [r9]                    @ s0<- vA
+
+    fsubs   s2, s0, s1                              @ s2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    fsts    s2, [r9]                    @ vAA<- s2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_FLOAT_2ADDR: /* 0xc8 */
+/* File: vfp/OP_MUL_FLOAT_2ADDR.S */
+/* File: vfp/fbinop2addr.S */
+    /*
+     * Generic 32-bit floating point "/2addr" binary operation.  Provide
+     * an "instr" line that specifies an instruction that performs
+     * "s2 = s0 op s1".
+     *
+     * For: add-float/2addr, sub-float/2addr, mul-float/2addr, div-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    flds    s1, [r3]                    @ s1<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    and     r9, r9, #15                 @ r9<- A
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    flds    s0, [r9]                    @ s0<- vA
+
+    fmuls   s2, s0, s1                              @ s2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    fsts    s2, [r9]                    @ vAA<- s2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_FLOAT_2ADDR: /* 0xc9 */
+/* File: vfp/OP_DIV_FLOAT_2ADDR.S */
+/* File: vfp/fbinop2addr.S */
+    /*
+     * Generic 32-bit floating point "/2addr" binary operation.  Provide
+     * an "instr" line that specifies an instruction that performs
+     * "s2 = s0 op s1".
+     *
+     * For: add-float/2addr, sub-float/2addr, mul-float/2addr, div-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    flds    s1, [r3]                    @ s1<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    and     r9, r9, #15                 @ r9<- A
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    flds    s0, [r9]                    @ s0<- vA
+
+    fdivs   s2, s0, s1                              @ s2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    fsts    s2, [r9]                    @ vAA<- s2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_FLOAT_2ADDR: /* 0xca */
+/* File: armv5te/OP_REM_FLOAT_2ADDR.S */
+/* EABI doesn't define a float remainder function, but libm does */
+/* File: armv5te/binop2addr.S */
+    /*
+     * Generic 32-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/2addr, sub-int/2addr, mul-int/2addr, div-int/2addr,
+     *      rem-int/2addr, and-int/2addr, or-int/2addr, xor-int/2addr,
+     *      shl-int/2addr, shr-int/2addr, ushr-int/2addr, add-float/2addr,
+     *      sub-float/2addr, mul-float/2addr, div-float/2addr, rem-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r3, rINST, lsr #12          @ r3<- B
+    and     r9, r9, #15
+    GET_VREG(r0, r9)                    @ r0<- vA
+    GET_VREG(r1, r3)                    @ r1<- vB
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      fmodf                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_DOUBLE_2ADDR: /* 0xcb */
+/* File: vfp/OP_ADD_DOUBLE_2ADDR.S */
+/* File: vfp/fbinopWide2addr.S */
+    /*
+     * Generic 64-bit floating point "/2addr" binary operation.  Provide
+     * an "instr" line that specifies an instruction that performs
+     * "d2 = d0 op d1".
+     *
+     * For: add-double/2addr, sub-double/2addr, mul-double/2addr,
+     *      div-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    fldd    d1, [r3]                    @ d1<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    and     r9, r9, #15                 @ r9<- A
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    fldd    d0, [r9]                    @ d0<- vA
+
+    faddd   d2, d0, d1                              @ d2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    fstd    d2, [r9]                    @ vAA<- d2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SUB_DOUBLE_2ADDR: /* 0xcc */
+/* File: vfp/OP_SUB_DOUBLE_2ADDR.S */
+/* File: vfp/fbinopWide2addr.S */
+    /*
+     * Generic 64-bit floating point "/2addr" binary operation.  Provide
+     * an "instr" line that specifies an instruction that performs
+     * "d2 = d0 op d1".
+     *
+     * For: add-double/2addr, sub-double/2addr, mul-double/2addr,
+     *      div-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    fldd    d1, [r3]                    @ d1<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    and     r9, r9, #15                 @ r9<- A
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    fldd    d0, [r9]                    @ d0<- vA
+
+    fsubd   d2, d0, d1                              @ d2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    fstd    d2, [r9]                    @ vAA<- d2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_DOUBLE_2ADDR: /* 0xcd */
+/* File: vfp/OP_MUL_DOUBLE_2ADDR.S */
+/* File: vfp/fbinopWide2addr.S */
+    /*
+     * Generic 64-bit floating point "/2addr" binary operation.  Provide
+     * an "instr" line that specifies an instruction that performs
+     * "d2 = d0 op d1".
+     *
+     * For: add-double/2addr, sub-double/2addr, mul-double/2addr,
+     *      div-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    fldd    d1, [r3]                    @ d1<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    and     r9, r9, #15                 @ r9<- A
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    fldd    d0, [r9]                    @ d0<- vA
+
+    fmuld   d2, d0, d1                              @ d2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    fstd    d2, [r9]                    @ vAA<- d2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_DOUBLE_2ADDR: /* 0xce */
+/* File: vfp/OP_DIV_DOUBLE_2ADDR.S */
+/* File: vfp/fbinopWide2addr.S */
+    /*
+     * Generic 64-bit floating point "/2addr" binary operation.  Provide
+     * an "instr" line that specifies an instruction that performs
+     * "d2 = d0 op d1".
+     *
+     * For: add-double/2addr, sub-double/2addr, mul-double/2addr,
+     *      div-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    fldd    d1, [r3]                    @ d1<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    and     r9, r9, #15                 @ r9<- A
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    fldd    d0, [r9]                    @ d0<- vA
+
+    fdivd   d2, d0, d1                              @ d2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    fstd    d2, [r9]                    @ vAA<- d2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_DOUBLE_2ADDR: /* 0xcf */
+/* File: armv5te/OP_REM_DOUBLE_2ADDR.S */
+/* EABI doesn't define a double remainder function, but libm does */
+/* File: armv5te/binopWide2addr.S */
+    /*
+     * Generic 64-bit "/2addr" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0-r1 op r2-r3".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-long/2addr, sub-long/2addr, div-long/2addr, rem-long/2addr,
+     *      and-long/2addr, or-long/2addr, xor-long/2addr, add-double/2addr,
+     *      sub-double/2addr, mul-double/2addr, div-double/2addr,
+     *      rem-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r9, r9, #15
+    add     r1, rFP, r1, lsl #2         @ r1<- &fp[B]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[A]
+    ldmia   r1, {r2-r3}                 @ r2/r3<- vBB/vBB+1
+    ldmia   r9, {r0-r1}                 @ r0/r1<- vAA/vAA+1
+    .if 0
+    orrs    ip, r2, r3                  @ second arg (r2-r3) is zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      fmod                              @ result<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0,r1}     @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 12-15 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_INT_LIT16: /* 0xd0 */
+/* File: armv5te/OP_ADD_INT_LIT16.S */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    add     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RSUB_INT: /* 0xd1 */
+/* File: armv5te/OP_RSUB_INT.S */
+/* this op is "rsub-int", but can be thought of as "rsub-int/lit16" */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    rsb     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_INT_LIT16: /* 0xd2 */
+/* File: armv5te/OP_MUL_INT_LIT16.S */
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    mul     r0, r1, r0                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_INT_LIT16: /* 0xd3 */
+/* File: armv5te/OP_DIV_INT_LIT16.S */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 1
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    bl     __aeabi_idiv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_INT_LIT16: /* 0xd4 */
+/* File: armv5te/OP_REM_INT_LIT16.S */
+/* idivmod returns quotient in r0 and remainder in r1 */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 1
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    bl      __aeabi_idivmod                              @ r1<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r1, r9)               @ vAA<- r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_INT_LIT16: /* 0xd5 */
+/* File: armv5te/OP_AND_INT_LIT16.S */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    and     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_INT_LIT16: /* 0xd6 */
+/* File: armv5te/OP_OR_INT_LIT16.S */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    orr     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_INT_LIT16: /* 0xd7 */
+/* File: armv5te/OP_XOR_INT_LIT16.S */
+/* File: armv5te/binopLit16.S */
+    /*
+     * Generic 32-bit "lit16" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit16, rsub-int, mul-int/lit16, div-int/lit16,
+     *      rem-int/lit16, and-int/lit16, or-int/lit16, xor-int/lit16
+     */
+    /* binop/lit16 vA, vB, #+CCCC */
+    FETCH_S(r1, 1)                      @ r1<- ssssCCCC (sign-extended)
+    mov     r2, rINST, lsr #12          @ r2<- B
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    GET_VREG(r0, r2)                    @ r0<- vB
+    and     r9, r9, #15
+    .if 0
+    cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    eor     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-13 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_ADD_INT_LIT8: /* 0xd8 */
+/* File: armv5te/OP_ADD_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    add     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_RSUB_INT_LIT8: /* 0xd9 */
+/* File: armv5te/OP_RSUB_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    rsb     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_MUL_INT_LIT8: /* 0xda */
+/* File: armv5te/OP_MUL_INT_LIT8.S */
+/* must be "mul r0, r1, r0" -- "r0, r0, r1" is illegal */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    mul     r0, r1, r0                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_DIV_INT_LIT8: /* 0xdb */
+/* File: armv5te/OP_DIV_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 1
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl     __aeabi_idiv                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_REM_INT_LIT8: /* 0xdc */
+/* File: armv5te/OP_REM_INT_LIT8.S */
+/* idivmod returns quotient in r0 and remainder in r1 */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 1
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    bl      __aeabi_idivmod                              @ r1<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r1, r9)               @ vAA<- r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_AND_INT_LIT8: /* 0xdd */
+/* File: armv5te/OP_AND_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    and     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_OR_INT_LIT8: /* 0xde */
+/* File: armv5te/OP_OR_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    orr     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_XOR_INT_LIT8: /* 0xdf */
+/* File: armv5te/OP_XOR_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+                               @ optional op; may set condition codes
+    eor     r0, r0, r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHL_INT_LIT8: /* 0xe0 */
+/* File: armv5te/OP_SHL_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, asl r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_SHR_INT_LIT8: /* 0xe1 */
+/* File: armv5te/OP_SHR_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, asr r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_USHR_INT_LIT8: /* 0xe2 */
+/* File: armv5te/OP_USHR_INT_LIT8.S */
+/* File: armv5te/binopLit8.S */
+    /*
+     * Generic 32-bit "lit8" binary operation.  Provide an "instr" line
+     * that specifies an instruction that performs "result = r0 op r1".
+     * This could be an ARM instruction or a function call.  (If the result
+     * comes back in a register other than r0, you can override "result".)
+     *
+     * If "chkzero" is set to 1, we perform a divide-by-zero check on
+     * vCC (r1).  Useful for integer division and modulus.
+     *
+     * For: add-int/lit8, rsub-int/lit8, mul-int/lit8, div-int/lit8,
+     *      rem-int/lit8, and-int/lit8, or-int/lit8, xor-int/lit8,
+     *      shl-int/lit8, shr-int/lit8, ushr-int/lit8
+     */
+    /* binop/lit8 vAA, vBB, #+CC */
+    FETCH_S(r3, 1)                      @ r3<- ssssCCBB (sign-extended for CC)
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    and     r2, r3, #255                @ r2<- BB
+    GET_VREG(r0, r2)                    @ r0<- vBB
+    movs    r1, r3, asr #8              @ r1<- ssssssCC (sign extended)
+    .if 0
+    @cmp     r1, #0                      @ is second operand zero?
+    beq     common_errDivideByZero
+    .endif
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+
+    and     r1, r1, #31                           @ optional op; may set condition codes
+    mov     r0, r0, lsr r1                              @ r0<- op, r0-r3 changed
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)               @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+    /* 10-12 instructions */
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E3: /* 0xe3 */
+/* File: armv5te/OP_UNUSED_E3.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E4: /* 0xe4 */
+/* File: armv5te/OP_UNUSED_E4.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E5: /* 0xe5 */
+/* File: armv5te/OP_UNUSED_E5.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E6: /* 0xe6 */
+/* File: armv5te/OP_UNUSED_E6.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E7: /* 0xe7 */
+/* File: armv5te/OP_UNUSED_E7.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E8: /* 0xe8 */
+/* File: armv5te/OP_UNUSED_E8.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_E9: /* 0xe9 */
+/* File: armv5te/OP_UNUSED_E9.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EA: /* 0xea */
+/* File: armv5te/OP_UNUSED_EA.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EB: /* 0xeb */
+/* File: armv5te/OP_UNUSED_EB.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EC: /* 0xec */
+/* File: armv5te/OP_UNUSED_EC.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_THROW_VERIFICATION_ERROR: /* 0xed */
+/* File: armv5te/OP_THROW_VERIFICATION_ERROR.S */
+    /*
+     * Handle a throw-verification-error instruction.  This throws an
+     * exception for an error discovered during verification.  The
+     * exception is indicated by AA, with some detail provided by BBBB.
+     */
+    /* op AA, ref@BBBB */
+    ldr     r0, [rGLUE, #offGlue_method]    @ r0<- glue->method
+    FETCH(r2, 1)                        @ r2<- BBBB
+    EXPORT_PC()                         @ export the PC
+    mov     r1, rINST, lsr #8           @ r1<- AA
+    bl      dvmThrowVerificationError   @ always throws
+    b       common_exceptionThrown      @ handle exception
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_EXECUTE_INLINE: /* 0xee */
+/* File: armv5te/OP_EXECUTE_INLINE.S */
+    /*
+     * Execute a "native inline" instruction.
+     *
+     * We need to call:
+     *  dvmPerformInlineOp4Std(arg0, arg1, arg2, arg3, &retval, ref)
+     *
+     * The first four args are in r0-r3, but the last two must be pushed
+     * onto the stack.
+     */
+    /* [opt] execute-inline vAA, {vC, vD, vE, vF}, inline@BBBB */
+    FETCH(r10, 1)                       @ r10<- BBBB
+    add     r1, rGLUE, #offGlue_retval  @ r1<- &glue->retval
+    EXPORT_PC()                         @ can throw
+    sub     sp, sp, #8                  @ make room for arg(s)
+    mov     r0, rINST, lsr #12          @ r0<- B
+    str     r1, [sp]                    @ push &glue->retval
+    bl      .LOP_EXECUTE_INLINE_continue        @ make call; will return after
+    add     sp, sp, #8                  @ pop stack
+    cmp     r0, #0                      @ test boolean result of inline
+    beq     common_exceptionThrown      @ returned false, handle exception
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_EF: /* 0xef */
+/* File: armv5te/OP_UNUSED_EF.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_DIRECT_EMPTY: /* 0xf0 */
+/* File: armv5te/OP_INVOKE_DIRECT_EMPTY.S */
+    /*
+     * invoke-direct-empty is a no-op in a "standard" interpreter.
+     */
+    FETCH_ADVANCE_INST(3)               @ advance to next instr, load rINST
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    GOTO_OPCODE(ip)                     @ execute it
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_F1: /* 0xf1 */
+/* File: armv5te/OP_UNUSED_F1.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_QUICK: /* 0xf2 */
+/* File: armv5te/OP_IGET_QUICK.S */
+    /* For: iget-quick, iget-object-quick */
+    /* op vA, vB, offset@CCCC */
+    mov     r2, rINST, lsr #12          @ r2<- B
+    GET_VREG(r3, r2)                    @ r3<- object we're operating on
+    FETCH(r1, 1)                        @ r1<- field byte offset
+    cmp     r3, #0                      @ check object for null
+    mov     r2, rINST, lsr #8           @ r2<- A(+)
+    beq     common_errNullObject        @ object was null
+    ldr     r0, [r3, r1]                @ r0<- obj.field (always 32 bits)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_WIDE_QUICK: /* 0xf3 */
+/* File: armv5te/OP_IGET_WIDE_QUICK.S */
+    /* iget-wide-quick vA, vB, offset@CCCC */
+    mov     r2, rINST, lsr #12          @ r2<- B
+    GET_VREG(r3, r2)                    @ r3<- object we're operating on
+    FETCH(r1, 1)                        @ r1<- field byte offset
+    cmp     r3, #0                      @ check object for null
+    mov     r2, rINST, lsr #8           @ r2<- A(+)
+    beq     common_errNullObject        @ object was null
+    ldrd    r0, [r3, r1]                @ r0<- obj.field (64 bits, aligned)
+    and     r2, r2, #15
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    add     r3, rFP, r2, lsl #2         @ r3<- &fp[A]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r3, {r0-r1}                 @ fp[A]<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IGET_OBJECT_QUICK: /* 0xf4 */
+/* File: armv5te/OP_IGET_OBJECT_QUICK.S */
+/* File: armv5te/OP_IGET_QUICK.S */
+    /* For: iget-quick, iget-object-quick */
+    /* op vA, vB, offset@CCCC */
+    mov     r2, rINST, lsr #12          @ r2<- B
+    GET_VREG(r3, r2)                    @ r3<- object we're operating on
+    FETCH(r1, 1)                        @ r1<- field byte offset
+    cmp     r3, #0                      @ check object for null
+    mov     r2, rINST, lsr #8           @ r2<- A(+)
+    beq     common_errNullObject        @ object was null
+    ldr     r0, [r3, r1]                @ r0<- obj.field (always 32 bits)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_QUICK: /* 0xf5 */
+/* File: armv5te/OP_IPUT_QUICK.S */
+    /* For: iput-quick, iput-object-quick */
+    /* op vA, vB, offset@CCCC */
+    mov     r2, rINST, lsr #12          @ r2<- B
+    GET_VREG(r3, r2)                    @ r3<- fp[B], the object pointer
+    FETCH(r1, 1)                        @ r1<- field byte offset
+    cmp     r3, #0                      @ check object for null
+    mov     r2, rINST, lsr #8           @ r2<- A(+)
+    beq     common_errNullObject        @ object was null
+    and     r2, r2, #15
+    GET_VREG(r0, r2)                    @ r0<- fp[A]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    str     r0, [r3, r1]                @ obj.field (always 32 bits)<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_WIDE_QUICK: /* 0xf6 */
+/* File: armv5te/OP_IPUT_WIDE_QUICK.S */
+    /* iput-wide-quick vA, vB, offset@CCCC */
+    mov     r0, rINST, lsr #8           @ r0<- A(+)
+    mov     r1, rINST, lsr #12          @ r1<- B
+    and     r0, r0, #15
+    GET_VREG(r2, r1)                    @ r2<- fp[B], the object pointer
+    add     r3, rFP, r0, lsl #2         @ r3<- &fp[A]
+    cmp     r2, #0                      @ check object for null
+    ldmia   r3, {r0-r1}                 @ r0/r1<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH(r3, 1)                        @ r3<- field byte offset
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    strd    r0, [r2, r3]                @ obj.field (64 bits, aligned)<- r0/r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_IPUT_OBJECT_QUICK: /* 0xf7 */
+/* File: armv5te/OP_IPUT_OBJECT_QUICK.S */
+/* File: armv5te/OP_IPUT_QUICK.S */
+    /* For: iput-quick, iput-object-quick */
+    /* op vA, vB, offset@CCCC */
+    mov     r2, rINST, lsr #12          @ r2<- B
+    GET_VREG(r3, r2)                    @ r3<- fp[B], the object pointer
+    FETCH(r1, 1)                        @ r1<- field byte offset
+    cmp     r3, #0                      @ check object for null
+    mov     r2, rINST, lsr #8           @ r2<- A(+)
+    beq     common_errNullObject        @ object was null
+    and     r2, r2, #15
+    GET_VREG(r0, r2)                    @ r0<- fp[A]
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    str     r0, [r3, r1]                @ obj.field (always 32 bits)<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_VIRTUAL_QUICK: /* 0xf8 */
+/* File: armv5te/OP_INVOKE_VIRTUAL_QUICK.S */
+    /*
+     * Handle an optimized virtual method call.
+     *
+     * for: [opt] invoke-virtual-quick, invoke-virtual-quick/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r3, 2)                        @ r3<- FEDC or CCCC
+    FETCH(r1, 1)                        @ r1<- BBBB
+    .if     (!0)
+    and     r3, r3, #15                 @ r3<- C (or stays CCCC)
+    .endif
+    GET_VREG(r2, r3)                    @ r2<- vC ("this" ptr)
+    cmp     r2, #0                      @ is "this" null?
+    beq     common_errNullObject        @ null "this", throw exception
+    ldr     r2, [r2, #offObject_clazz]  @ r2<- thisPtr->clazz
+    ldr     r2, [r2, #offClassObject_vtable]    @ r2<- thisPtr->clazz->vtable
+    EXPORT_PC()                         @ invoke must export
+    ldr     r0, [r2, r1, lsl #2]        @ r3<- vtable[BBBB]
+    bl      common_invokeMethodNoRange @ continue on
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_VIRTUAL_QUICK_RANGE: /* 0xf9 */
+/* File: armv5te/OP_INVOKE_VIRTUAL_QUICK_RANGE.S */
+/* File: armv5te/OP_INVOKE_VIRTUAL_QUICK.S */
+    /*
+     * Handle an optimized virtual method call.
+     *
+     * for: [opt] invoke-virtual-quick, invoke-virtual-quick/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r3, 2)                        @ r3<- FEDC or CCCC
+    FETCH(r1, 1)                        @ r1<- BBBB
+    .if     (!1)
+    and     r3, r3, #15                 @ r3<- C (or stays CCCC)
+    .endif
+    GET_VREG(r2, r3)                    @ r2<- vC ("this" ptr)
+    cmp     r2, #0                      @ is "this" null?
+    beq     common_errNullObject        @ null "this", throw exception
+    ldr     r2, [r2, #offObject_clazz]  @ r2<- thisPtr->clazz
+    ldr     r2, [r2, #offClassObject_vtable]    @ r2<- thisPtr->clazz->vtable
+    EXPORT_PC()                         @ invoke must export
+    ldr     r0, [r2, r1, lsl #2]        @ r3<- vtable[BBBB]
+    bl      common_invokeMethodRange @ continue on
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_SUPER_QUICK: /* 0xfa */
+/* File: armv5te/OP_INVOKE_SUPER_QUICK.S */
+    /*
+     * Handle an optimized "super" method call.
+     *
+     * for: [opt] invoke-super-quick, invoke-super-quick/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    .if     (!0)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r2, [r2, #offMethod_clazz]  @ r2<- method->clazz
+    EXPORT_PC()                         @ must export for invoke
+    ldr     r2, [r2, #offClassObject_super]     @ r2<- method->clazz->super
+    GET_VREG(r3, r10)                   @ r3<- "this"
+    ldr     r2, [r2, #offClassObject_vtable]    @ r2<- ...clazz->super->vtable
+    cmp     r3, #0                      @ null "this" ref?
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- super->vtable[BBBB]
+    beq     common_errNullObject        @ "this" is null, throw exception
+    bl      common_invokeMethodNoRange @ continue on
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_INVOKE_SUPER_QUICK_RANGE: /* 0xfb */
+/* File: armv5te/OP_INVOKE_SUPER_QUICK_RANGE.S */
+/* File: armv5te/OP_INVOKE_SUPER_QUICK.S */
+    /*
+     * Handle an optimized "super" method call.
+     *
+     * for: [opt] invoke-super-quick, invoke-super-quick/range
+     */
+    /* op vB, {vD, vE, vF, vG, vA}, class@CCCC */
+    /* op vAA, {vCCCC..v(CCCC+AA-1)}, meth@BBBB */
+    FETCH(r10, 2)                       @ r10<- GFED or CCCC
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    .if     (!1)
+    and     r10, r10, #15               @ r10<- D (or stays CCCC)
+    .endif
+    FETCH(r1, 1)                        @ r1<- BBBB
+    ldr     r2, [r2, #offMethod_clazz]  @ r2<- method->clazz
+    EXPORT_PC()                         @ must export for invoke
+    ldr     r2, [r2, #offClassObject_super]     @ r2<- method->clazz->super
+    GET_VREG(r3, r10)                   @ r3<- "this"
+    ldr     r2, [r2, #offClassObject_vtable]    @ r2<- ...clazz->super->vtable
+    cmp     r3, #0                      @ null "this" ref?
+    ldr     r0, [r2, r1, lsl #2]        @ r0<- super->vtable[BBBB]
+    beq     common_errNullObject        @ "this" is null, throw exception
+    bl      common_invokeMethodRange @ continue on
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FC: /* 0xfc */
+/* File: armv5te/OP_UNUSED_FC.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FD: /* 0xfd */
+/* File: armv5te/OP_UNUSED_FD.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FE: /* 0xfe */
+/* File: armv5te/OP_UNUSED_FE.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+/* ------------------------------ */
+    .balign 64
+.L_OP_UNUSED_FF: /* 0xff */
+/* File: armv5te/OP_UNUSED_FF.S */
+/* File: armv5te/unused.S */
+    bl      common_abort
+
+
+
+
+    .balign 64
+    .size   dvmAsmInstructionStart, .-dvmAsmInstructionStart
+    .global dvmAsmInstructionEnd
+dvmAsmInstructionEnd:
+
+/*
+ * ===========================================================================
+ *  Sister implementations
+ * ===========================================================================
+ */
+    .global dvmAsmSisterStart
+    .type   dvmAsmSisterStart, %function
+    .text
+    .balign 4
+dvmAsmSisterStart:
+
+/* continuation for OP_CONST_STRING */
+
+    /*
+     * Continuation if the String has not yet been resolved.
+     *  r1: BBBB (String ref)
+     *  r9: target register
+     */
+.LOP_CONST_STRING_resolve:
+    EXPORT_PC()
+    ldr     r0, [rGLUE, #offGlue_method] @ r0<- glue->method
+    ldr     r0, [r0, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveString            @ r0<- String reference
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yup, handle the exception
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_CONST_STRING_JUMBO */
+
+    /*
+     * Continuation if the String has not yet been resolved.
+     *  r1: BBBBBBBB (String ref)
+     *  r9: target register
+     */
+.LOP_CONST_STRING_JUMBO_resolve:
+    EXPORT_PC()
+    ldr     r0, [rGLUE, #offGlue_method] @ r0<- glue->method
+    ldr     r0, [r0, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveString            @ r0<- String reference
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yup, handle the exception
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_CONST_CLASS */
+
+    /*
+     * Continuation if the Class has not yet been resolved.
+     *  r1: BBBB (Class ref)
+     *  r9: target register
+     */
+.LOP_CONST_CLASS_resolve:
+    EXPORT_PC()
+    ldr     r0, [rGLUE, #offGlue_method] @ r0<- glue->method
+    mov     r2, #1                      @ r2<- true
+    ldr     r0, [r0, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- Class reference
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yup, handle the exception
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_CHECK_CAST */
+
+    /*
+     * Trivial test failed, need to perform full check.  This is common.
+     *  r0 holds obj->clazz
+     *  r1 holds class resolved from BBBB
+     *  r9 holds object
+     */
+.LOP_CHECK_CAST_fullcheck:
+    bl      dvmInstanceofNonTrivial     @ r0<- boolean result
+    cmp     r0, #0                      @ failed?
+    bne     .LOP_CHECK_CAST_okay            @ no, success
+
+    @ A cast has failed.  We need to throw a ClassCastException with the
+    @ class of the object that failed to be cast.
+    EXPORT_PC()                         @ about to throw
+    ldr     r3, [r9, #offObject_clazz]  @ r3<- obj->clazz
+    ldr     r0, .LstrClassCastExceptionPtr
+    ldr     r1, [r3, #offClassObject_descriptor] @ r1<- obj->clazz->descriptor
+    bl      dvmThrowExceptionWithClassMessage
+    b       common_exceptionThrown
+
+    /*
+     * Resolution required.  This is the least-likely path.
+     *
+     *  r2 holds BBBB
+     *  r9 holds object
+     */
+.LOP_CHECK_CAST_resolve:
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    mov     r1, r2                      @ r1<- BBBB
+    mov     r2, #0                      @ r2<- false
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- resolved ClassObject ptr
+    cmp     r0, #0                      @ got null?
+    beq     common_exceptionThrown      @ yes, handle exception
+    mov     r1, r0                      @ r1<- class resolved from BBB
+    ldr     r0, [r9, #offObject_clazz]  @ r0<- obj->clazz
+    b       .LOP_CHECK_CAST_resolved        @ pick up where we left off
+
+.LstrClassCastExceptionPtr:
+    .word   .LstrClassCastException
+
+
+/* continuation for OP_INSTANCE_OF */
+
+    /*
+     * Trivial test failed, need to perform full check.  This is common.
+     *  r0 holds obj->clazz
+     *  r1 holds class resolved from BBBB
+     *  r9 holds A
+     */
+.LOP_INSTANCE_OF_fullcheck:
+    bl      dvmInstanceofNonTrivial     @ r0<- boolean result
+    @ fall through to OP_INSTANCE_OF_store
+
+    /*
+     * r0 holds boolean result
+     * r9 holds A
+     */
+.LOP_INSTANCE_OF_store:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r0, r9)                    @ vA<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+    /*
+     * Trivial test succeeded, save and bail.
+     *  r9 holds A
+     */
+.LOP_INSTANCE_OF_trivial:
+    mov     r0, #1                      @ indicate success
+    @ could b OP_INSTANCE_OF_store, but copying is faster and cheaper
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r0, r9)                    @ vA<- r0
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+    /*
+     * Resolution required.  This is the least-likely path.
+     *
+     *  r3 holds BBBB
+     *  r9 holds A
+     */
+.LOP_INSTANCE_OF_resolve:
+    EXPORT_PC()                         @ resolve() could throw
+    ldr     r0, [rGLUE, #offGlue_method]    @ r0<- glue->method
+    mov     r1, r3                      @ r1<- BBBB
+    mov     r2, #1                      @ r2<- true
+    ldr     r0, [r0, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- resolved ClassObject ptr
+    cmp     r0, #0                      @ got null?
+    beq     common_exceptionThrown      @ yes, handle exception
+    mov     r1, r0                      @ r1<- class resolved from BBB
+    mov     r3, rINST, lsr #12          @ r3<- B
+    GET_VREG(r0, r3)                    @ r0<- vB (object)
+    ldr     r0, [r0, #offObject_clazz]  @ r0<- obj->clazz
+    b       .LOP_INSTANCE_OF_resolved        @ pick up where we left off
+
+
+/* continuation for OP_NEW_INSTANCE */
+
+    .balign 32                          @ minimize cache lines
+.LOP_NEW_INSTANCE_finish: @ r0=new object
+    mov     r3, rINST, lsr #8           @ r3<- AA
+    cmp     r0, #0                      @ failed?
+    beq     common_exceptionThrown      @ yes, handle the exception
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r3)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+    /*
+     * Class initialization required.
+     *
+     *  r0 holds class object
+     */
+.LOP_NEW_INSTANCE_needinit:
+    mov     r9, r0                      @ save r0
+    bl      dvmInitClass                @ initialize class
+    cmp     r0, #0                      @ check boolean result
+    mov     r0, r9                      @ restore r0
+    bne     .LOP_NEW_INSTANCE_initialized     @ success, continue
+    b       common_exceptionThrown      @ failed, deal with init exception
+
+    /*
+     * Resolution required.  This is the least-likely path.
+     *
+     *  r1 holds BBBB
+     */
+.LOP_NEW_INSTANCE_resolve:
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    mov     r2, #0                      @ r2<- false
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- resolved ClassObject ptr
+    cmp     r0, #0                      @ got null?
+    bne     .LOP_NEW_INSTANCE_resolved        @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+.LstrInstantiationErrorPtr:
+    .word   .LstrInstantiationError
+
+
+/* continuation for OP_NEW_ARRAY */
+
+
+    /*
+     * Resolve class.  (This is an uncommon case.)
+     *
+     *  r1 holds array length
+     *  r2 holds class ref CCCC
+     */
+.LOP_NEW_ARRAY_resolve:
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    mov     r9, r1                      @ r9<- length (save)
+    mov     r1, r2                      @ r1<- CCCC
+    mov     r2, #0                      @ r2<- false
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveClass             @ r0<- call(clazz, ref)
+    cmp     r0, #0                      @ got null?
+    mov     r1, r9                      @ r1<- length (restore)
+    beq     common_exceptionThrown      @ yes, handle exception
+    @ fall through to OP_NEW_ARRAY_finish
+
+    /*
+     * Finish allocation.
+     *
+     *  r0 holds class
+     *  r1 holds array length
+     */
+.LOP_NEW_ARRAY_finish:
+    mov     r2, #ALLOC_DONT_TRACK       @ don't track in local refs table
+    bl      dvmAllocArrayByClass        @ r0<- call(clazz, length, flags)
+    cmp     r0, #0                      @ failed?
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    beq     common_exceptionThrown      @ yes, handle the exception
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ vA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_FILLED_NEW_ARRAY */
+
+    /*
+     * On entry:
+     *  r0 holds array class
+     *  r10 holds AA or BA
+     */
+.LOP_FILLED_NEW_ARRAY_continue:
+    ldr     r3, [r0, #offClassObject_descriptor] @ r3<- arrayClass->descriptor
+    mov     r2, #ALLOC_DONT_TRACK       @ r2<- alloc flags
+    ldrb    r3, [r3, #1]                @ r3<- descriptor[1]
+    .if     0
+    mov     r1, r10                     @ r1<- AA (length)
+    .else
+    mov     r1, r10, lsr #4             @ r1<- B (length)
+    .endif
+    cmp     r3, #'I'                    @ array of ints?
+    cmpne   r3, #'L'                    @ array of objects?
+    cmpne   r3, #'['                    @ array of arrays?
+    mov     r9, r1                      @ save length in r9
+    bne     .LOP_FILLED_NEW_ARRAY_notimpl         @ no, not handled yet
+    bl      dvmAllocArrayByClass        @ r0<- call(arClass, length, flags)
+    cmp     r0, #0                      @ null return?
+    beq     common_exceptionThrown      @ alloc failed, handle exception
+
+    FETCH(r1, 2)                        @ r1<- FEDC or CCCC
+    str     r0, [rGLUE, #offGlue_retval]    @ retval.l <- new array
+    add     r0, r0, #offArrayObject_contents @ r0<- newArray->contents
+    subs    r9, r9, #1                  @ length--, check for neg
+    FETCH_ADVANCE_INST(3)               @ advance to next instr, load rINST
+    bmi     2f                          @ was zero, bail
+
+    @ copy values from registers into the array
+    @ r0=array, r1=CCCC/FEDC, r9=length (from AA or B), r10=AA/BA
+    .if     0
+    add     r2, rFP, r1, lsl #2         @ r2<- &fp[CCCC]
+1:  ldr     r3, [r2], #4                @ r3<- *r2++
+    subs    r9, r9, #1                  @ count--
+    str     r3, [r0], #4                @ *contents++ = vX
+    bpl     1b
+    @ continue at 2
+    .else
+    cmp     r9, #4                      @ length was initially 5?
+    and     r2, r10, #15                @ r2<- A
+    bne     1f                          @ <= 4 args, branch
+    GET_VREG(r3, r2)                    @ r3<- vA
+    sub     r9, r9, #1                  @ count--
+    str     r3, [r0, #16]               @ contents[4] = vA
+1:  and     r2, r1, #15                 @ r2<- F/E/D/C
+    GET_VREG(r3, r2)                    @ r3<- vF/vE/vD/vC
+    mov     r1, r1, lsr #4              @ r1<- next reg in low 4
+    subs    r9, r9, #1                  @ count--
+    str     r3, [r0], #4                @ *contents++ = vX
+    bpl     1b
+    @ continue at 2
+    .endif
+
+2:
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    GOTO_OPCODE(ip)                     @ execute it
+
+    /*
+     * Throw an exception indicating that we have not implemented this
+     * mode of filled-new-array.
+     */
+.LOP_FILLED_NEW_ARRAY_notimpl:
+    ldr     r0, .L_strInternalError
+    ldr     r1, .L_strFilledNewArrayNotImpl
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+    .if     (!0)                 @ define in one or the other, not both
+.L_strFilledNewArrayNotImpl:
+    .word   .LstrFilledNewArrayNotImpl
+.L_strInternalError:
+    .word   .LstrInternalError
+    .endif
+
+
+/* continuation for OP_FILLED_NEW_ARRAY_RANGE */
+
+    /*
+     * On entry:
+     *  r0 holds array class
+     *  r10 holds AA or BA
+     */
+.LOP_FILLED_NEW_ARRAY_RANGE_continue:
+    ldr     r3, [r0, #offClassObject_descriptor] @ r3<- arrayClass->descriptor
+    mov     r2, #ALLOC_DONT_TRACK       @ r2<- alloc flags
+    ldrb    r3, [r3, #1]                @ r3<- descriptor[1]
+    .if     1
+    mov     r1, r10                     @ r1<- AA (length)
+    .else
+    mov     r1, r10, lsr #4             @ r1<- B (length)
+    .endif
+    cmp     r3, #'I'                    @ array of ints?
+    cmpne   r3, #'L'                    @ array of objects?
+    cmpne   r3, #'['                    @ array of arrays?
+    mov     r9, r1                      @ save length in r9
+    bne     .LOP_FILLED_NEW_ARRAY_RANGE_notimpl         @ no, not handled yet
+    bl      dvmAllocArrayByClass        @ r0<- call(arClass, length, flags)
+    cmp     r0, #0                      @ null return?
+    beq     common_exceptionThrown      @ alloc failed, handle exception
+
+    FETCH(r1, 2)                        @ r1<- FEDC or CCCC
+    str     r0, [rGLUE, #offGlue_retval]    @ retval.l <- new array
+    add     r0, r0, #offArrayObject_contents @ r0<- newArray->contents
+    subs    r9, r9, #1                  @ length--, check for neg
+    FETCH_ADVANCE_INST(3)               @ advance to next instr, load rINST
+    bmi     2f                          @ was zero, bail
+
+    @ copy values from registers into the array
+    @ r0=array, r1=CCCC/FEDC, r9=length (from AA or B), r10=AA/BA
+    .if     1
+    add     r2, rFP, r1, lsl #2         @ r2<- &fp[CCCC]
+1:  ldr     r3, [r2], #4                @ r3<- *r2++
+    subs    r9, r9, #1                  @ count--
+    str     r3, [r0], #4                @ *contents++ = vX
+    bpl     1b
+    @ continue at 2
+    .else
+    cmp     r9, #4                      @ length was initially 5?
+    and     r2, r10, #15                @ r2<- A
+    bne     1f                          @ <= 4 args, branch
+    GET_VREG(r3, r2)                    @ r3<- vA
+    sub     r9, r9, #1                  @ count--
+    str     r3, [r0, #16]               @ contents[4] = vA
+1:  and     r2, r1, #15                 @ r2<- F/E/D/C
+    GET_VREG(r3, r2)                    @ r3<- vF/vE/vD/vC
+    mov     r1, r1, lsr #4              @ r1<- next reg in low 4
+    subs    r9, r9, #1                  @ count--
+    str     r3, [r0], #4                @ *contents++ = vX
+    bpl     1b
+    @ continue at 2
+    .endif
+
+2:
+    GET_INST_OPCODE(ip)                 @ ip<- opcode from rINST
+    GOTO_OPCODE(ip)                     @ execute it
+
+    /*
+     * Throw an exception indicating that we have not implemented this
+     * mode of filled-new-array.
+     */
+.LOP_FILLED_NEW_ARRAY_RANGE_notimpl:
+    ldr     r0, .L_strInternalError
+    ldr     r1, .L_strFilledNewArrayNotImpl
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+    .if     (!1)                 @ define in one or the other, not both
+.L_strFilledNewArrayNotImpl:
+    .word   .LstrFilledNewArrayNotImpl
+.L_strInternalError:
+    .word   .LstrInternalError
+    .endif
+
+
+/* continuation for OP_CMPL_FLOAT */
+.LOP_CMPL_FLOAT_finish:
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_CMPG_FLOAT */
+.LOP_CMPG_FLOAT_finish:
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_CMPL_DOUBLE */
+.LOP_CMPL_DOUBLE_finish:
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_CMPG_DOUBLE */
+.LOP_CMPG_DOUBLE_finish:
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_CMP_LONG */
+
+.LOP_CMP_LONG_less:
+    mvn     r1, #0                      @ r1<- -1
+    @ Want to cond code the next mov so we can avoid branch, but don't see it;
+    @ instead, we just replicate the tail end.
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r9)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+.LOP_CMP_LONG_greater:
+    mov     r1, #1                      @ r1<- 1
+    @ fall through to _finish
+
+.LOP_CMP_LONG_finish:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    SET_VREG(r1, r9)                    @ vAA<- r1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_AGET_WIDE */
+
+.LOP_AGET_WIDE_finish:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldrd    r2, [r0, #offArrayObject_contents]  @ r2/r3<- vBB[vCC]
+    add     r9, rFP, r9, lsl #2         @ r9<- &fp[AA]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r2-r3}                 @ vAA/vAA+1<- r2/r3
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_APUT_WIDE */
+
+.LOP_APUT_WIDE_finish:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldmia   r9, {r2-r3}                 @ r2/r3<- vAA/vAA+1
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strd    r2, [r0, #offArrayObject_contents]  @ r2/r3<- vBB[vCC]
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_APUT_OBJECT */
+    /*
+     * On entry:
+     *  r1 = vBB (arrayObj)
+     *  r9 = vAA (obj)
+     *  r10 = offset into array (vBB + vCC * width)
+     */
+.LOP_APUT_OBJECT_finish:
+    cmp     r9, #0                      @ storing null reference?
+    beq     .LOP_APUT_OBJECT_skip_check      @ yes, skip type checks
+    ldr     r0, [r9, #offObject_clazz]  @ r0<- obj->clazz
+    ldr     r1, [r1, #offObject_clazz]  @ r1<- arrayObj->clazz
+    bl      dvmCanPutArrayElement       @ test object type vs. array type
+    cmp     r0, #0                      @ okay?
+    beq     common_errArrayStore        @ no
+.LOP_APUT_OBJECT_skip_check:
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str     r9, [r10, #offArrayObject_contents] @ vBB[vCC]<- vAA
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_finish:
+    @bl      common_squeak0
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    ldr   r0, [r9, r3]                @ r0<- obj.field (8/16/32 bits)
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET_WIDE */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_WIDE_finish:
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    ldrd    r0, [r9, r3]                @ r0/r1<- obj.field (64-bit align ok)
+    and     r2, r2, #15                 @ r2<- A
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    add     r3, rFP, r2, lsl #2         @ r3<- &fp[A]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r3, {r0-r1}                 @ fp[A]<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET_OBJECT */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_OBJECT_finish:
+    @bl      common_squeak0
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    ldr   r0, [r9, r3]                @ r0<- obj.field (8/16/32 bits)
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET_BOOLEAN */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_BOOLEAN_finish:
+    @bl      common_squeak1
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    ldr   r0, [r9, r3]                @ r0<- obj.field (8/16/32 bits)
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET_BYTE */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_BYTE_finish:
+    @bl      common_squeak2
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    ldr   r0, [r9, r3]                @ r0<- obj.field (8/16/32 bits)
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET_CHAR */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_CHAR_finish:
+    @bl      common_squeak3
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    ldr   r0, [r9, r3]                @ r0<- obj.field (8/16/32 bits)
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IGET_SHORT */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IGET_SHORT_finish:
+    @bl      common_squeak4
+    cmp     r9, #0                      @ check object for null
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    beq     common_errNullObject        @ object was null
+    ldr   r0, [r9, r3]                @ r0<- obj.field (8/16/32 bits)
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    and     r2, r2, #15                 @ r2<- A
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    SET_VREG(r0, r2)                    @ fp[A]<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_finish:
+    @bl      common_squeak0
+    mov     r1, rINST, lsr #8           @ r1<- A+
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    and     r1, r1, #15                 @ r1<- A
+    cmp     r9, #0                      @ check object for null
+    GET_VREG(r0, r1)                    @ r0<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT_WIDE */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_WIDE_finish:
+    mov     r2, rINST, lsr #8           @ r2<- A+
+    cmp     r9, #0                      @ check object for null
+    and     r2, r2, #15                 @ r2<- A
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    add     r2, rFP, r2, lsl #2         @ r3<- &fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    ldmia   r2, {r0-r1}                 @ r0/r1<- fp[A]
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    strd    r0, [r9, r3]                @ obj.field (64 bits, aligned)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT_OBJECT */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_OBJECT_finish:
+    @bl      common_squeak0
+    mov     r1, rINST, lsr #8           @ r1<- A+
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    and     r1, r1, #15                 @ r1<- A
+    cmp     r9, #0                      @ check object for null
+    GET_VREG(r0, r1)                    @ r0<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT_BOOLEAN */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_BOOLEAN_finish:
+    @bl      common_squeak1
+    mov     r1, rINST, lsr #8           @ r1<- A+
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    and     r1, r1, #15                 @ r1<- A
+    cmp     r9, #0                      @ check object for null
+    GET_VREG(r0, r1)                    @ r0<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT_BYTE */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_BYTE_finish:
+    @bl      common_squeak2
+    mov     r1, rINST, lsr #8           @ r1<- A+
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    and     r1, r1, #15                 @ r1<- A
+    cmp     r9, #0                      @ check object for null
+    GET_VREG(r0, r1)                    @ r0<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT_CHAR */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_CHAR_finish:
+    @bl      common_squeak3
+    mov     r1, rINST, lsr #8           @ r1<- A+
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    and     r1, r1, #15                 @ r1<- A
+    cmp     r9, #0                      @ check object for null
+    GET_VREG(r0, r1)                    @ r0<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_IPUT_SHORT */
+
+    /*
+     * Currently:
+     *  r0 holds resolved field
+     *  r9 holds object
+     */
+.LOP_IPUT_SHORT_finish:
+    @bl      common_squeak4
+    mov     r1, rINST, lsr #8           @ r1<- A+
+    ldr     r3, [r0, #offInstField_byteOffset]  @ r3<- byte offset of field
+    and     r1, r1, #15                 @ r1<- A
+    cmp     r9, #0                      @ check object for null
+    GET_VREG(r0, r1)                    @ r0<- fp[A]
+    beq     common_errNullObject        @ object was null
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    str  r0, [r9, r3]                @ obj.field (8/16/32 bits)<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_SGET */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SGET_WIDE */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_WIDE_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_WIDE_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SGET_OBJECT */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_OBJECT_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_OBJECT_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SGET_BOOLEAN */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_BOOLEAN_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_BOOLEAN_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SGET_BYTE */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_BYTE_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_BYTE_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SGET_CHAR */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_CHAR_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_CHAR_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SGET_SHORT */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SGET_SHORT_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SGET_SHORT_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SPUT_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT_WIDE */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     *  r9: &fp[AA]
+     */
+.LOP_SPUT_WIDE_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_WIDE_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT_OBJECT */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SPUT_OBJECT_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_OBJECT_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT_BOOLEAN */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SPUT_BOOLEAN_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_BOOLEAN_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT_BYTE */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SPUT_BYTE_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_BYTE_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT_CHAR */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SPUT_CHAR_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_CHAR_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_SPUT_SHORT */
+
+    /*
+     * Continuation if the field has not yet been resolved.
+     *  r1: BBBB field ref
+     */
+.LOP_SPUT_SHORT_resolve:
+    ldr     r2, [rGLUE, #offGlue_method]    @ r2<- current method
+    EXPORT_PC()                         @ resolve() could throw, so export now
+    ldr     r0, [r2, #offMethod_clazz]  @ r0<- method->clazz
+    bl      dvmResolveStaticField       @ r0<- resolved StaticField ptr
+    cmp     r0, #0                      @ success?
+    bne     .LOP_SPUT_SHORT_finish          @ yes, finish
+    b       common_exceptionThrown      @ no, handle exception
+
+
+/* continuation for OP_INVOKE_VIRTUAL */
+
+    /*
+     * At this point:
+     *  r0 = resolved base method
+     *  r10 = C or CCCC (index of first arg, which is the "this" ptr)
+     */
+.LOP_INVOKE_VIRTUAL_continue:
+    GET_VREG(r1, r10)                   @ r1<- "this" ptr
+    ldrh    r2, [r0, #offMethod_methodIndex]    @ r2<- baseMethod->methodIndex
+    cmp     r1, #0                      @ is "this" null?
+    beq     common_errNullObject        @ null "this", throw exception
+    ldr     r3, [r1, #offObject_clazz]  @ r1<- thisPtr->clazz
+    ldr     r3, [r3, #offClassObject_vtable]    @ r3<- thisPtr->clazz->vtable
+    ldr     r0, [r3, r2, lsl #2]        @ r3<- vtable[methodIndex]
+    bl      common_invokeMethodNoRange @ continue on
+
+
+/* continuation for OP_INVOKE_SUPER */
+
+    /*
+     * At this point:
+     *  r0 = resolved base method
+     *  r9 = method->clazz
+     */
+.LOP_INVOKE_SUPER_continue:
+    ldr     r1, [r9, #offClassObject_super]     @ r1<- method->clazz->super
+    ldrh    r2, [r0, #offMethod_methodIndex]    @ r2<- baseMethod->methodIndex
+    ldr     r3, [r1, #offClassObject_vtableCount]   @ r3<- super->vtableCount
+    EXPORT_PC()                         @ must export for invoke
+    cmp     r2, r3                      @ compare (methodIndex, vtableCount)
+    bcs     .LOP_INVOKE_SUPER_nsm             @ method not present in superclass
+    ldr     r1, [r1, #offClassObject_vtable]    @ r1<- ...clazz->super->vtable
+    ldr     r0, [r1, r2, lsl #2]        @ r3<- vtable[methodIndex]
+    bl      common_invokeMethodNoRange @ continue on
+
+.LOP_INVOKE_SUPER_resolve:
+    mov     r0, r9                      @ r0<- method->clazz
+    mov     r2, #METHOD_VIRTUAL         @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    bne     .LOP_INVOKE_SUPER_continue        @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+    /*
+     * Throw a NoSuchMethodError with the method name as the message.
+     *  r0 = resolved base method
+     */
+.LOP_INVOKE_SUPER_nsm:
+    ldr     r1, [r0, #offMethod_name]   @ r1<- method name
+    b       common_errNoSuchMethod
+
+
+/* continuation for OP_INVOKE_DIRECT */
+
+    /*
+     * On entry:
+     *  r1 = reference (BBBB or CCCC)
+     *  r10 = "this" register
+     */
+.LOP_INVOKE_DIRECT_resolve:
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    mov     r2, #METHOD_DIRECT          @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    GET_VREG(r2, r10)                   @ r2<- "this" ptr (reload)
+    bne     .LOP_INVOKE_DIRECT_finish          @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+
+/* continuation for OP_INVOKE_VIRTUAL_RANGE */
+
+    /*
+     * At this point:
+     *  r0 = resolved base method
+     *  r10 = C or CCCC (index of first arg, which is the "this" ptr)
+     */
+.LOP_INVOKE_VIRTUAL_RANGE_continue:
+    GET_VREG(r1, r10)                   @ r1<- "this" ptr
+    ldrh    r2, [r0, #offMethod_methodIndex]    @ r2<- baseMethod->methodIndex
+    cmp     r1, #0                      @ is "this" null?
+    beq     common_errNullObject        @ null "this", throw exception
+    ldr     r3, [r1, #offObject_clazz]  @ r1<- thisPtr->clazz
+    ldr     r3, [r3, #offClassObject_vtable]    @ r3<- thisPtr->clazz->vtable
+    ldr     r0, [r3, r2, lsl #2]        @ r3<- vtable[methodIndex]
+    bl      common_invokeMethodRange @ continue on
+
+
+/* continuation for OP_INVOKE_SUPER_RANGE */
+
+    /*
+     * At this point:
+     *  r0 = resolved base method
+     *  r9 = method->clazz
+     */
+.LOP_INVOKE_SUPER_RANGE_continue:
+    ldr     r1, [r9, #offClassObject_super]     @ r1<- method->clazz->super
+    ldrh    r2, [r0, #offMethod_methodIndex]    @ r2<- baseMethod->methodIndex
+    ldr     r3, [r1, #offClassObject_vtableCount]   @ r3<- super->vtableCount
+    EXPORT_PC()                         @ must export for invoke
+    cmp     r2, r3                      @ compare (methodIndex, vtableCount)
+    bcs     .LOP_INVOKE_SUPER_RANGE_nsm             @ method not present in superclass
+    ldr     r1, [r1, #offClassObject_vtable]    @ r1<- ...clazz->super->vtable
+    ldr     r0, [r1, r2, lsl #2]        @ r3<- vtable[methodIndex]
+    bl      common_invokeMethodRange @ continue on
+
+.LOP_INVOKE_SUPER_RANGE_resolve:
+    mov     r0, r9                      @ r0<- method->clazz
+    mov     r2, #METHOD_VIRTUAL         @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    bne     .LOP_INVOKE_SUPER_RANGE_continue        @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+    /*
+     * Throw a NoSuchMethodError with the method name as the message.
+     *  r0 = resolved base method
+     */
+.LOP_INVOKE_SUPER_RANGE_nsm:
+    ldr     r1, [r0, #offMethod_name]   @ r1<- method name
+    b       common_errNoSuchMethod
+
+
+/* continuation for OP_INVOKE_DIRECT_RANGE */
+
+    /*
+     * On entry:
+     *  r1 = reference (BBBB or CCCC)
+     *  r10 = "this" register
+     */
+.LOP_INVOKE_DIRECT_RANGE_resolve:
+    ldr     r3, [rGLUE, #offGlue_method] @ r3<- glue->method
+    ldr     r0, [r3, #offMethod_clazz]  @ r0<- method->clazz
+    mov     r2, #METHOD_DIRECT          @ resolver method type
+    bl      dvmResolveMethod            @ r0<- call(clazz, ref, flags)
+    cmp     r0, #0                      @ got null?
+    GET_VREG(r2, r10)                   @ r2<- "this" ptr (reload)
+    bne     .LOP_INVOKE_DIRECT_RANGE_finish          @ no, continue
+    b       common_exceptionThrown      @ yes, handle exception
+
+
+/* continuation for OP_FLOAT_TO_LONG */
+/*
+ * Convert the float in r0 to a long in r0/r1.
+ *
+ * We have to clip values to long min/max per the specification.  The
+ * expected common case is a "reasonable" value that converts directly
+ * to modest integer.  The EABI convert function isn't doing this for us.
+ */
+f2l_doconv:
+    stmfd   sp!, {r4, lr}
+    mov     r1, #0x5f000000             @ (float)maxlong
+    mov     r4, r0
+    bl      __aeabi_fcmpge              @ is arg >= maxlong?
+    cmp     r0, #0                      @ nonzero == yes
+    mvnne   r0, #0                      @ return maxlong (7fffffff)
+    mvnne   r1, #0x80000000
+    ldmnefd sp!, {r4, pc}
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, #0xdf000000             @ (float)minlong
+    bl      __aeabi_fcmple              @ is arg <= minlong?
+    cmp     r0, #0                      @ nonzero == yes
+    movne   r0, #0                      @ return minlong (80000000)
+    movne   r1, #0x80000000
+    ldmnefd sp!, {r4, pc}
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r4
+    bl      __aeabi_fcmpeq              @ is arg == self?
+    cmp     r0, #0                      @ zero == no
+    moveq   r1, #0                      @ return zero for NaN
+    ldmeqfd sp!, {r4, pc}
+
+    mov     r0, r4                      @ recover arg
+    bl      __aeabi_f2lz                @ convert float to long
+    ldmfd   sp!, {r4, pc}
+
+
+/* continuation for OP_DOUBLE_TO_LONG */
+/*
+ * Convert the double in r0/r1 to a long in r0/r1.
+ *
+ * We have to clip values to long min/max per the specification.  The
+ * expected common case is a "reasonable" value that converts directly
+ * to modest integer.  The EABI convert function isn't doing this for us.
+ */
+d2l_doconv:
+    stmfd   sp!, {r4, r5, lr}           @ save regs
+    ldr     r3, .LOP_DOUBLE_TO_LONG_max         @ (double)maxlong, hi
+    sub     sp, sp, #4                  @ align for EABI
+    mov     r2, #0                      @ (double)maxlong, lo
+    mov     r4, r0                      @ save r0
+    mov     r5, r1                      @  and r1
+    bl      __aeabi_dcmpge              @ is arg >= maxlong?
+    cmp     r0, #0                      @ nonzero == yes
+    mvnne   r0, #0                      @ return maxlong (7fffffffffffffff)
+    mvnne   r1, #0x80000000
+    bne     1f
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r5
+    ldr     r3, .LOP_DOUBLE_TO_LONG_min         @ (double)minlong, hi
+    mov     r2, #0                      @ (double)minlong, lo
+    bl      __aeabi_dcmple              @ is arg <= minlong?
+    cmp     r0, #0                      @ nonzero == yes
+    movne   r0, #0                      @ return minlong (8000000000000000)
+    movne   r1, #0x80000000
+    bne     1f
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r5
+    mov     r2, r4                      @ compare against self
+    mov     r3, r5
+    bl      __aeabi_dcmpeq              @ is arg == self?
+    cmp     r0, #0                      @ zero == no
+    moveq   r1, #0                      @ return zero for NaN
+    beq     1f
+
+    mov     r0, r4                      @ recover arg
+    mov     r1, r5
+    bl      __aeabi_d2lz                @ convert double to long
+
+1:
+    add     sp, sp, #4
+    ldmfd   sp!, {r4, r5, pc}
+
+.LOP_DOUBLE_TO_LONG_max:
+    .word   0x43e00000                  @ maxlong, as a double (high word)
+.LOP_DOUBLE_TO_LONG_min:
+    .word   0xc3e00000                  @ minlong, as a double (high word)
+
+
+/* continuation for OP_MUL_LONG */
+
+.LOP_MUL_LONG_finish:
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r0, {r9-r10}                @ vAA/vAA+1<- r9/r10
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_SHL_LONG */
+
+.LOP_SHL_LONG_finish:
+    mov     r0, r0, asl r2              @  r0<- r0 << r2
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_SHR_LONG */
+
+.LOP_SHR_LONG_finish:
+    mov     r1, r1, asr r2              @  r1<- r1 >> r2
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_USHR_LONG */
+
+.LOP_USHR_LONG_finish:
+    mov     r1, r1, lsr r2              @  r1<- r1 >>> r2
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_SHL_LONG_2ADDR */
+
+.LOP_SHL_LONG_2ADDR_finish:
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_SHR_LONG_2ADDR */
+
+.LOP_SHR_LONG_2ADDR_finish:
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_USHR_LONG_2ADDR */
+
+.LOP_USHR_LONG_2ADDR_finish:
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    stmia   r9, {r0-r1}                 @ vAA/vAA+1<- r0/r1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/* continuation for OP_EXECUTE_INLINE */
+
+    /*
+     * Extract args, call function.
+     *  r0 = #of args (0-4)
+     *  r10 = call index
+     *  lr = return addr, above  [DO NOT bl out of here w/o preserving LR]
+     *
+     * Other ideas:
+     * - Use a jump table from the main piece to jump directly into the
+     *   AND/LDR pairs.  Costs a data load, saves a branch.
+     * - Have five separate pieces that do the loading, so we can work the
+     *   interleave a little better.  Increases code size.
+     */
+.LOP_EXECUTE_INLINE_continue:
+    rsb     r0, r0, #4                  @ r0<- 4-r0
+    FETCH(r9, 2)                        @ r9<- FEDC
+    add     pc, pc, r0, lsl #3          @ computed goto, 2 instrs each
+    bl      common_abort                @ (skipped due to ARM prefetch)
+4:  and     ip, r9, #0xf000             @ isolate F
+    ldr     r3, [rFP, ip, lsr #10]      @ r3<- vF (shift right 12, left 2)
+3:  and     ip, r9, #0x0f00             @ isolate E
+    ldr     r2, [rFP, ip, lsr #6]       @ r2<- vE
+2:  and     ip, r9, #0x00f0             @ isolate D
+    ldr     r1, [rFP, ip, lsr #2]       @ r1<- vD
+1:  and     ip, r9, #0x000f             @ isolate C
+    ldr     r0, [rFP, ip, lsl #2]       @ r0<- vC
+0:
+    ldr     r9, .LOP_EXECUTE_INLINE_table       @ table of InlineOperation
+    LDR_PC  "[r9, r10, lsl #4]"         @ sizeof=16, "func" is first entry
+    @ (not reached)
+
+.LOP_EXECUTE_INLINE_table:
+    .word   gDvmInlineOpsTable
+
+
+    .size   dvmAsmSisterStart, .-dvmAsmSisterStart
+    .global dvmAsmSisterEnd
+dvmAsmSisterEnd:
+
+/* File: armv5te/footer.S */
+/*
+ * ===========================================================================
+ *  Common subroutines and data
+ * ===========================================================================
+ */
+
+    .text
+    .align  2
+
+/*
+ * Common code when a backward branch is taken.
+ *
+ * On entry:
+ *  r9 is PC adjustment *in bytes*
+ */
+common_backwardBranch:
+    mov     r0, #kInterpEntryInstr
+    bl      common_periodicChecks
+    FETCH_ADVANCE_INST_RB(r9)           @ update rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+
+/*
+ * Need to see if the thread needs to be suspended or debugger/profiler
+ * activity has begun.
+ *
+ * TODO: if JDWP isn't running, zero out pDebuggerActive pointer so we don't
+ * have to do the second ldr.
+ *
+ * TODO: reduce this so we're just checking a single location.
+ *
+ * On entry:
+ *  r0 is reentry type, e.g. kInterpEntryInstr
+ *  r9 is trampoline PC adjustment *in bytes*
+ */
+common_periodicChecks:
+    ldr     r3, [rGLUE, #offGlue_pSelfSuspendCount] @ r3<- &suspendCount
+
+#if defined(WITH_DEBUGGER)
+    ldr     r1, [rGLUE, #offGlue_pDebuggerActive]   @ r1<- &debuggerActive
+#endif
+#if defined(WITH_PROFILER)
+    ldr     r2, [rGLUE, #offGlue_pActiveProfilers]  @ r2<- &activeProfilers
+#endif
+
+    ldr     r3, [r3]                    @ r3<- suspendCount (int)
+
+#if defined(WITH_DEBUGGER)
+    ldrb    r1, [r1]                    @ r1<- debuggerActive (boolean)
+#endif
+#if defined (WITH_PROFILER)
+    ldr     r2, [r2]                    @ r2<- activeProfilers (int)
+#endif
+
+    cmp     r3, #0                      @ suspend pending?
+    bne     2f                          @ yes, do full suspension check
+
+#if defined(WITH_DEBUGGER) || defined(WITH_PROFILER)
+# if defined(WITH_DEBUGGER) && defined(WITH_PROFILER)
+    orrs    r1, r1, r2                  @ r1<- r1 | r2
+    cmp     r1, #0                      @ debugger attached or profiler started?
+# elif defined(WITH_DEBUGGER)
+    cmp     r1, #0                      @ debugger attached?
+# elif defined(WITH_PROFILER)
+    cmp     r2, #0                      @ profiler started?
+# endif
+    bne     3f                          @ debugger/profiler, switch interp
+#endif
+
+    bx      lr                          @ nothing to do, return
+
+2:  @ check suspend
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- glue->self
+    EXPORT_PC()                         @ need for precise GC
+    b       dvmCheckSuspendPending      @ suspend if necessary, then return
+
+3:  @ debugger/profiler enabled, bail out
+    add     rPC, rPC, r9                @ update rPC
+    str     r0, [rGLUE, #offGlue_entryPoint]
+    mov     r1, #1                      @ "want switch" = true
+    b       common_gotoBail
+
+
+/*
+ * The equivalent of "goto bail", this calls through the "bail handler".
+ *
+ * State registers will be saved to the "glue" area before bailing.
+ *
+ * On entry:
+ *  r1 is "bool changeInterp", indicating if we want to switch to the
+ *     other interpreter or just bail all the way out
+ */
+common_gotoBail:
+    SAVE_PC_FP_TO_GLUE()                @ export state to "glue"
+    mov     r0, rGLUE                   @ r0<- glue ptr
+    b       dvmMterpStdBail             @ call(glue, changeInterp)
+
+    @add     r1, r1, #1                  @ using (boolean+1)
+    @add     r0, rGLUE, #offGlue_jmpBuf  @ r0<- &glue->jmpBuf
+    @bl      _longjmp                    @ does not return
+    @bl      common_abort
+
+
+/*
+ * Common code for method invocation with range.
+ *
+ * On entry:
+ *  r0 is "Method* methodToCall", the method we're trying to call
+ */
+common_invokeMethodRange:
+.LinvokeNewRange:
+    @ prepare to copy args to "outs" area of current frame
+    movs    r2, rINST, lsr #8           @ r2<- AA (arg count) -- test for zero
+    SAVEAREA_FROM_FP(r10, rFP)          @ r10<- stack save area
+    beq     .LinvokeArgsDone            @ if no args, skip the rest
+    FETCH(r1, 2)                        @ r1<- CCCC
+
+    @ r0=methodToCall, r1=CCCC, r2=count, r10=outs
+    @ (very few methods have > 10 args; could unroll for common cases)
+    add     r3, rFP, r1, lsl #2         @ r3<- &fp[CCCC]
+    sub     r10, r10, r2, lsl #2        @ r10<- "outs" area, for call args
+    ldrh    r9, [r0, #offMethod_registersSize]  @ r9<- methodToCall->regsSize
+1:  ldr     r1, [r3], #4                @ val = *fp++
+    subs    r2, r2, #1                  @ count--
+    str     r1, [r10], #4               @ *outs++ = val
+    bne     1b                          @ ...while count != 0
+    ldrh    r3, [r0, #offMethod_outsSize]   @ r3<- methodToCall->outsSize
+    b       .LinvokeArgsDone
+
+/*
+ * Common code for method invocation without range.
+ *
+ * On entry:
+ *  r0 is "Method* methodToCall", the method we're trying to call
+ */
+common_invokeMethodNoRange:
+.LinvokeNewNoRange:
+    @ prepare to copy args to "outs" area of current frame
+    movs    r2, rINST, lsr #12          @ r2<- B (arg count) -- test for zero
+    SAVEAREA_FROM_FP(r10, rFP)          @ r10<- stack save area
+    FETCH(r1, 2)                        @ r1<- GFED (load here to hide latency)
+    ldrh    r9, [r0, #offMethod_registersSize]  @ r9<- methodToCall->regsSize
+    ldrh    r3, [r0, #offMethod_outsSize]  @ r3<- methodToCall->outsSize
+    beq     .LinvokeArgsDone
+
+    @ r0=methodToCall, r1=GFED, r3=outSize, r2=count, r9=regSize, r10=outs
+.LinvokeNonRange:
+    rsb     r2, r2, #5                  @ r2<- 5-r2
+    add     pc, pc, r2, lsl #4          @ computed goto, 4 instrs each
+    bl      common_abort                @ (skipped due to ARM prefetch)
+5:  and     ip, rINST, #0x0f00          @ isolate A
+    ldr     r2, [rFP, ip, lsr #6]       @ r2<- vA (shift right 8, left 2)
+    mov     r0, r0                      @ nop
+    str     r2, [r10, #-4]!             @ *--outs = vA
+4:  and     ip, r1, #0xf000             @ isolate G
+    ldr     r2, [rFP, ip, lsr #10]      @ r2<- vG (shift right 12, left 2)
+    mov     r0, r0                      @ nop
+    str     r2, [r10, #-4]!             @ *--outs = vG
+3:  and     ip, r1, #0x0f00             @ isolate F
+    ldr     r2, [rFP, ip, lsr #6]       @ r2<- vF
+    mov     r0, r0                      @ nop
+    str     r2, [r10, #-4]!             @ *--outs = vF
+2:  and     ip, r1, #0x00f0             @ isolate E
+    ldr     r2, [rFP, ip, lsr #2]       @ r2<- vE
+    mov     r0, r0                      @ nop
+    str     r2, [r10, #-4]!             @ *--outs = vE
+1:  and     ip, r1, #0x000f             @ isolate D
+    ldr     r2, [rFP, ip, lsl #2]       @ r2<- vD
+    mov     r0, r0                      @ nop
+    str     r2, [r10, #-4]!             @ *--outs = vD
+0:  @ fall through to .LinvokeArgsDone
+
+.LinvokeArgsDone: @ r0=methodToCall, r3=outSize, r9=regSize
+    ldr     r2, [r0, #offMethod_insns]  @ r2<- method->insns
+    ldr     rINST, [r0, #offMethod_clazz]  @ rINST<- method->clazz
+    @ find space for the new stack frame, check for overflow
+    SAVEAREA_FROM_FP(r1, rFP)           @ r1<- stack save area
+    sub     r1, r1, r9, lsl #2          @ r1<- newFp (old savearea - regsSize)
+    SAVEAREA_FROM_FP(r10, r1)           @ r10<- newSaveArea
+@    bl      common_dumpRegs
+    ldr     r9, [rGLUE, #offGlue_interpStackEnd]    @ r9<- interpStackEnd
+    sub     r3, r10, r3, lsl #2         @ r3<- bottom (newsave - outsSize)
+    cmp     r3, r9                      @ bottom < interpStackEnd?
+    ldr     r3, [r0, #offMethod_accessFlags] @ r3<- methodToCall->accessFlags
+    blt     .LstackOverflow             @ yes, this frame will overflow stack
+
+    @ set up newSaveArea
+#ifdef EASY_GDB
+    SAVEAREA_FROM_FP(ip, rFP)           @ ip<- stack save area
+    str     ip, [r10, #offStackSaveArea_prevSave]
+#endif
+    str     rFP, [r10, #offStackSaveArea_prevFrame]
+    str     rPC, [r10, #offStackSaveArea_savedPc]
+    str     r0, [r10, #offStackSaveArea_method]
+    tst     r3, #ACC_NATIVE
+    bne     .LinvokeNative
+
+    /*
+    stmfd   sp!, {r0-r3}
+    bl      common_printNewline
+    mov     r0, rFP
+    mov     r1, #0
+    bl      dvmDumpFp
+    ldmfd   sp!, {r0-r3}
+    stmfd   sp!, {r0-r3}
+    mov     r0, r1
+    mov     r1, r10
+    bl      dvmDumpFp
+    bl      common_printNewline
+    ldmfd   sp!, {r0-r3}
+    */
+
+    ldrh    r9, [r2]                        @ r9 <- load INST from new PC
+    ldr     r3, [rINST, #offClassObject_pDvmDex] @ r3<- method->clazz->pDvmDex
+    mov     rPC, r2                         @ publish new rPC
+    ldr     r2, [rGLUE, #offGlue_self]      @ r2<- glue->self
+
+    @ Update "glue" values for the new method
+    @ r0=methodToCall, r1=newFp, r2=self, r3=newMethodClass, r9=newINST
+    str     r0, [rGLUE, #offGlue_method]    @ glue->method = methodToCall
+    str     r3, [rGLUE, #offGlue_methodClassDex] @ glue->methodClassDex = ...
+    mov     rFP, r1                         @ fp = newFp
+    GET_PREFETCHED_OPCODE(ip, r9)           @ extract prefetched opcode from r9
+    mov     rINST, r9                       @ publish new rINST
+    str     r1, [r2, #offThread_curFrame]   @ self->curFrame = newFp
+    GOTO_OPCODE(ip)                         @ jump to next instruction
+
+.LinvokeNative:
+    @ Prep for the native call
+    @ r0=methodToCall, r1=newFp, r10=newSaveArea
+    ldr     r3, [rGLUE, #offGlue_self]      @ r3<- glue->self
+    ldr     r9, [r3, #offThread_jniLocal_nextEntry] @ r9<- thread->refNext
+    str     r1, [r3, #offThread_curFrame]   @ self->curFrame = newFp
+    str     r9, [r10, #offStackSaveArea_localRefTop] @newFp->localRefTop=refNext
+    mov     r9, r3                      @ r9<- glue->self (preserve)
+
+    mov     r2, r0                      @ r2<- methodToCall
+    mov     r0, r1                      @ r0<- newFp (points to args)
+    add     r1, rGLUE, #offGlue_retval  @ r1<- &retval
+
+#ifdef ASSIST_DEBUGGER
+    /* insert fake function header to help gdb find the stack frame */
+    b       .Lskip
+    .type   dalvik_mterp, %function
+dalvik_mterp:
+    .fnstart
+    MTERP_ENTRY1
+    MTERP_ENTRY2
+.Lskip:
+#endif
+
+    @mov     lr, pc                      @ set return addr
+    @ldr     pc, [r2, #offMethod_nativeFunc] @ pc<- methodToCall->nativeFunc
+    LDR_PC_LR "[r2, #offMethod_nativeFunc]"
+
+    @ native return; r9=self, r10=newSaveArea
+    @ equivalent to dvmPopJniLocals
+    ldr     r0, [r10, #offStackSaveArea_localRefTop] @ r0<- newSave->localRefTop
+    ldr     r1, [r9, #offThread_exception] @ check for exception
+    str     rFP, [r9, #offThread_curFrame]  @ self->curFrame = fp
+    cmp     r1, #0                      @ null?
+    str     r0, [r9, #offThread_jniLocal_nextEntry] @ self->refNext<- r0
+    bne     common_exceptionThrown      @ no, handle exception
+
+    FETCH_ADVANCE_INST(3)               @ advance rPC, load rINST
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+.LstackOverflow:
+    ldr     r0, [rGLUE, #offGlue_self]  @ r0<- self
+    bl      dvmHandleStackOverflow
+    b       common_exceptionThrown
+#ifdef ASSIST_DEBUGGER
+    .fnend
+#endif
+
+
+    /*
+     * Common code for method invocation, calling through "glue code".
+     *
+     * TODO: now that we have range and non-range invoke handlers, this
+     *       needs to be split into two.  Maybe just create entry points
+     *       that set r9 and jump here?
+     *
+     * On entry:
+     *  r0 is "Method* methodToCall", the method we're trying to call
+     *  r9 is "bool methodCallRange", indicating if this is a /range variant
+     */
+     .if    0
+.LinvokeOld:
+    sub     sp, sp, #8                  @ space for args + pad
+    FETCH(ip, 2)                        @ ip<- FEDC or CCCC
+    mov     r2, r0                      @ A2<- methodToCall
+    mov     r0, rGLUE                   @ A0<- glue
+    SAVE_PC_FP_TO_GLUE()                @ export state to "glue"
+    mov     r1, r9                      @ A1<- methodCallRange
+    mov     r3, rINST, lsr #8           @ A3<- AA
+    str     ip, [sp, #0]                @ A4<- ip
+    bl      dvmMterp_invokeMethod       @ call the C invokeMethod
+    add     sp, sp, #8                  @ remove arg area
+    b       common_resumeAfterGlueCall  @ continue to next instruction
+    .endif
+
+
+
+/*
+ * Common code for handling a return instruction.
+ *
+ * This does not return.
+ */
+common_returnFromMethod:
+.LreturnNew:
+    mov     r0, #kInterpEntryReturn
+    mov     r9, #0
+    bl      common_periodicChecks
+
+    SAVEAREA_FROM_FP(r0, rFP)           @ r0<- saveArea (old)
+    ldr     rFP, [r0, #offStackSaveArea_prevFrame] @ fp = saveArea->prevFrame
+    ldr     r9, [r0, #offStackSaveArea_savedPc] @ r9 = saveArea->savedPc
+    ldr     r2, [rFP, #(offStackSaveArea_method - sizeofStackSaveArea)]
+                                        @ r2<- method we're returning to
+    ldr     r3, [rGLUE, #offGlue_self]  @ r3<- glue->self
+    cmp     r2, #0                      @ is this a break frame?
+    ldrne   r10, [r2, #offMethod_clazz] @ r10<- method->clazz
+    mov     r1, #0                      @ "want switch" = false
+    beq     common_gotoBail             @ break frame, bail out completely
+
+    PREFETCH_ADVANCE_INST(rINST, r9, 3) @ advance r9, update new rINST
+    str     r2, [rGLUE, #offGlue_method]@ glue->method = newSave->method
+    ldr     r1, [r10, #offClassObject_pDvmDex]   @ r1<- method->clazz->pDvmDex
+    str     rFP, [r3, #offThread_curFrame]  @ self->curFrame = fp
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    mov     rPC, r9                     @ publish new rPC
+    str     r1, [rGLUE, #offGlue_methodClassDex]
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+    /*
+     * Return handling, calls through "glue code".
+     */
+     .if    0
+.LreturnOld:
+    SAVE_PC_FP_TO_GLUE()                @ export state
+    mov     r0, rGLUE                   @ arg to function
+    bl      dvmMterp_returnFromMethod
+    b       common_resumeAfterGlueCall
+    .endif
+
+
+/*
+ * Somebody has thrown an exception.  Handle it.
+ *
+ * If the exception processing code returns to us (instead of falling
+ * out of the interpreter), continue with whatever the next instruction
+ * now happens to be.
+ *
+ * This does not return.
+ */
+common_exceptionThrown:
+.LexceptionNew:
+    mov     r0, #kInterpEntryThrow
+    mov     r9, #0
+    bl      common_periodicChecks
+
+    ldr     r10, [rGLUE, #offGlue_self] @ r10<- glue->self
+    ldr     r9, [r10, #offThread_exception] @ r9<- self->exception
+    mov     r1, r10                     @ r1<- self
+    mov     r0, r9                      @ r0<- exception
+    bl      dvmAddTrackedAlloc          @ don't let the exception be GCed
+    mov     r3, #0                      @ r3<- NULL
+    str     r3, [r10, #offThread_exception] @ self->exception = NULL
+
+    /* set up args and a local for "&fp" */
+    /* (str sp, [sp, #-4]!  would be perfect here, but is discouraged) */
+    str     rFP, [sp, #-4]!             @ *--sp = fp
+    mov     ip, sp                      @ ip<- &fp
+    mov     r3, #0                      @ r3<- false
+    str     ip, [sp, #-4]!              @ *--sp = &fp
+    ldr     r1, [rGLUE, #offGlue_method] @ r1<- glue->method
+    mov     r0, r10                     @ r0<- self
+    ldr     r1, [r1, #offMethod_insns]  @ r1<- method->insns
+    mov     r2, r9                      @ r2<- exception
+    sub     r1, rPC, r1                 @ r1<- pc - method->insns
+    mov     r1, r1, asr #1              @ r1<- offset in code units
+
+    /* call, r0 gets catchRelPc (a code-unit offset) */
+    bl      dvmFindCatchBlock           @ call(self, relPc, exc, scan?, &fp)
+
+    /* fix earlier stack overflow if necessary; may trash rFP */
+    ldrb    r1, [r10, #offThread_stackOverflowed]
+    cmp     r1, #0                      @ did we overflow earlier?
+    beq     1f                          @ no, skip ahead
+    mov     rFP, r0                     @ save relPc result in rFP
+    mov     r0, r10                     @ r0<- self
+    bl      dvmCleanupStackOverflow     @ call(self)
+    mov     r0, rFP                     @ restore result
+1:
+
+    /* update frame pointer and check result from dvmFindCatchBlock */
+    ldr     rFP, [sp, #4]               @ retrieve the updated rFP
+    cmp     r0, #0                      @ is catchRelPc < 0?
+    add     sp, sp, #8                  @ restore stack
+    bmi     .LnotCaughtLocally
+
+    /* adjust locals to match self->curFrame and updated PC */
+    SAVEAREA_FROM_FP(r1, rFP)           @ r1<- new save area
+    ldr     r1, [r1, #offStackSaveArea_method] @ r1<- new method
+    str     r1, [rGLUE, #offGlue_method]    @ glue->method = new method
+    ldr     r2, [r1, #offMethod_clazz]      @ r2<- method->clazz
+    ldr     r3, [r1, #offMethod_insns]      @ r3<- method->insns
+    ldr     r2, [r2, #offClassObject_pDvmDex] @ r2<- method->clazz->pDvmDex
+    add     rPC, r3, r0, asl #1             @ rPC<- method->insns + catchRelPc
+    str     r2, [rGLUE, #offGlue_methodClassDex] @ glue->pDvmDex = meth...
+
+    /* release the tracked alloc on the exception */
+    mov     r0, r9                      @ r0<- exception
+    mov     r1, r10                     @ r1<- self
+    bl      dvmReleaseTrackedAlloc      @ release the exception
+
+    /* restore the exception if the handler wants it */
+    FETCH_INST()                        @ load rINST from rPC
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    cmp     ip, #OP_MOVE_EXCEPTION      @ is it "move-exception"?
+    streq   r9, [r10, #offThread_exception] @ yes, restore the exception
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+.LnotCaughtLocally: @ r9=exception, r10=self
+    /* fix stack overflow if necessary */
+    ldrb    r1, [r10, #offThread_stackOverflowed]
+    cmp     r1, #0                      @ did we overflow earlier?
+    movne   r0, r10                     @ if yes: r0<- self
+    blne    dvmCleanupStackOverflow     @ if yes: call(self)
+
+    @ may want to show "not caught locally" debug messages here
+#if DVM_SHOW_EXCEPTION >= 2
+    /* call __android_log_print(prio, tag, format, ...) */
+    /* "Exception %s from %s:%d not caught locally" */
+    @ dvmLineNumFromPC(method, pc - method->insns)
+    ldr     r0, [rGLUE, #offGlue_method]
+    ldr     r1, [r0, #offMethod_insns]
+    sub     r1, rPC, r1
+    asr     r1, r1, #1
+    bl      dvmLineNumFromPC
+    str     r0, [sp, #-4]!
+    @ dvmGetMethodSourceFile(method)
+    ldr     r0, [rGLUE, #offGlue_method]
+    bl      dvmGetMethodSourceFile
+    str     r0, [sp, #-4]!
+    @ exception->clazz->descriptor
+    ldr     r3, [r9, #offObject_clazz]
+    ldr     r3, [r3, #offClassObject_descriptor]
+    @
+    ldr     r2, strExceptionNotCaughtLocally
+    ldr     r1, strLogTag
+    mov     r0, #3                      @ LOG_DEBUG
+    bl      __android_log_print
+#endif
+    str     r9, [r10, #offThread_exception] @ restore exception
+    mov     r0, r9                      @ r0<- exception
+    mov     r1, r10                     @ r1<- self
+    bl      dvmReleaseTrackedAlloc      @ release the exception
+    mov     r1, #0                      @ "want switch" = false
+    b       common_gotoBail             @ bail out
+
+
+    /*
+     * Exception handling, calls through "glue code".
+     */
+    .if     0
+.LexceptionOld:
+    SAVE_PC_FP_TO_GLUE()                @ export state
+    mov     r0, rGLUE                   @ arg to function
+    bl      dvmMterp_exceptionThrown
+    b       common_resumeAfterGlueCall
+    .endif
+
+
+/*
+ * After returning from a "glued" function, pull out the updated
+ * values and start executing at the next instruction.
+ */
+common_resumeAfterGlueCall:
+    LOAD_PC_FP_FROM_GLUE()              @ pull rPC and rFP out of glue
+    FETCH_INST()                        @ load rINST from rPC
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
+/*
+ * Invalid array index.
+ */
+common_errArrayIndex:
+    EXPORT_PC()
+    ldr     r0, strArrayIndexException
+    mov     r1, #0
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+/*
+ * Invalid array value.
+ */
+common_errArrayStore:
+    EXPORT_PC()
+    ldr     r0, strArrayStoreException
+    mov     r1, #0
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+/*
+ * Integer divide or mod by zero.
+ */
+common_errDivideByZero:
+    EXPORT_PC()
+    ldr     r0, strArithmeticException
+    ldr     r1, strDivideByZero
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+/*
+ * Attempt to allocate an array with a negative size.
+ */
+common_errNegativeArraySize:
+    EXPORT_PC()
+    ldr     r0, strNegativeArraySizeException
+    mov     r1, #0
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+/*
+ * Invocation of a non-existent method.
+ */
+common_errNoSuchMethod:
+    EXPORT_PC()
+    ldr     r0, strNoSuchMethodError
+    mov     r1, #0
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+/*
+ * We encountered a null object when we weren't expecting one.  We
+ * export the PC, throw a NullPointerException, and goto the exception
+ * processing code.
+ */
+common_errNullObject:
+    EXPORT_PC()
+    ldr     r0, strNullPointerException
+    mov     r1, #0
+    bl      dvmThrowException
+    b       common_exceptionThrown
+
+/*
+ * For debugging, cause an immediate fault.  The source address will
+ * be in lr (use a bl instruction to jump here).
+ */
+common_abort:
+    ldr     pc, .LdeadFood
+.LdeadFood:
+    .word   0xdeadf00d
+
+/*
+ * Spit out a "we were here", preserving all registers.  (The attempt
+ * to save ip won't work, but we need to save an even number of
+ * registers for EABI 64-bit stack alignment.)
+ */
+    .macro  SQUEAK num
+common_squeak\num:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    ldr     r0, strSqueak
+    mov     r1, #\num
+    bl      printf
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+    .endm
+
+    SQUEAK  0
+    SQUEAK  1
+    SQUEAK  2
+    SQUEAK  3
+    SQUEAK  4
+    SQUEAK  5
+
+/*
+ * Spit out the number in r0, preserving registers.
+ */
+common_printNum:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    mov     r1, r0
+    ldr     r0, strSqueak
+    bl      printf
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+
+/*
+ * Print a newline, preserving registers.
+ */
+common_printNewline:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    ldr     r0, strNewline
+    bl      printf
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+
+    /*
+     * Print the 32-bit quantity in r0 as a hex value, preserving registers.
+     */
+common_printHex:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    mov     r1, r0
+    ldr     r0, strPrintHex
+    bl      printf
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+
+/*
+ * Print the 64-bit quantity in r0-r1, preserving registers.
+ */
+common_printLong:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    mov     r3, r1
+    mov     r2, r0
+    ldr     r0, strPrintLong
+    bl      printf
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+
+/*
+ * Print full method info.  Pass the Method* in r0.  Preserves regs.
+ */
+common_printMethod:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bl      dvmMterpPrintMethod
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+
+/*
+ * Call a C helper function that dumps regs and possibly some
+ * additional info.  Requires the C function to be compiled in.
+ */
+    .if     0
+common_dumpRegs:
+    stmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bl      dvmMterpDumpArmRegs
+    ldmfd   sp!, {r0, r1, r2, r3, ip, lr}
+    bx      lr
+    .endif
+
+
+/*
+ * String references, must be close to the code that uses them.
+ */
+    .align  2
+strArithmeticException:
+    .word   .LstrArithmeticException
+strArrayIndexException:
+    .word   .LstrArrayIndexException
+strArrayStoreException:
+    .word   .LstrArrayStoreException
+strDivideByZero:
+    .word   .LstrDivideByZero
+strNegativeArraySizeException:
+    .word   .LstrNegativeArraySizeException
+strNoSuchMethodError:
+    .word   .LstrNoSuchMethodError
+strNullPointerException:
+    .word   .LstrNullPointerException
+
+strLogTag:
+    .word   .LstrLogTag
+strExceptionNotCaughtLocally:
+    .word   .LstrExceptionNotCaughtLocally
+
+strNewline:
+    .word   .LstrNewline
+strSqueak:
+    .word   .LstrSqueak
+strPrintHex:
+    .word   .LstrPrintHex
+strPrintLong:
+    .word   .LstrPrintLong
+
+/*
+ * Zero-terminated ASCII string data.
+ *
+ * On ARM we have two choices: do like gcc does, and LDR from a .word
+ * with the address, or use an ADR pseudo-op to get the address
+ * directly.  ADR saves 4 bytes and an indirection, but it's using a
+ * PC-relative addressing mode and hence has a limited range, which
+ * makes it not work well with mergeable string sections.
+ */
+    .section .rodata.str1.4,"aMS",%progbits,1
+
+.LstrBadEntryPoint:
+    .asciz  "Bad entry point %d\n"
+.LstrArithmeticException:
+    .asciz  "Ljava/lang/ArithmeticException;"
+.LstrArrayIndexException:
+    .asciz  "Ljava/lang/ArrayIndexOutOfBoundsException;"
+.LstrArrayStoreException:
+    .asciz  "Ljava/lang/ArrayStoreException;"
+.LstrClassCastException:
+    .asciz  "Ljava/lang/ClassCastException;"
+.LstrDivideByZero:
+    .asciz  "divide by zero"
+.LstrFilledNewArrayNotImpl:
+    .asciz  "filled-new-array only implemented for objects and 'int'"
+.LstrInternalError:
+    .asciz  "Ljava/lang/InternalError;"
+.LstrInstantiationError:
+    .asciz  "Ljava/lang/InstantiationError;"
+.LstrNegativeArraySizeException:
+    .asciz  "Ljava/lang/NegativeArraySizeException;"
+.LstrNoSuchMethodError:
+    .asciz  "Ljava/lang/NoSuchMethodError;"
+.LstrNullPointerException:
+    .asciz  "Ljava/lang/NullPointerException;"
+
+.LstrLogTag:
+    .asciz  "mterp"
+.LstrExceptionNotCaughtLocally:
+    .asciz  "Exception %s from %s:%d not caught locally\n"
+
+.LstrNewline:
+    .asciz  "\n"
+.LstrSqueak:
+    .asciz  "<%d>"
+.LstrPrintHex:
+    .asciz  "<0x%x>"
+.LstrPrintLong:
+    .asciz  "<%lld>"
+
+
diff --git a/vm/mterp/out/InterpAsm-armv5te.S b/vm/mterp/out/InterpAsm-armv5te.S
index e7efe25..8deb4aa 100644
--- a/vm/mterp/out/InterpAsm-armv5te.S
+++ b/vm/mterp/out/InterpAsm-armv5te.S
@@ -40,7 +40,9 @@
 r0 holds returns of <= 4 bytes
 r0-r1 hold returns of 8 bytes, low word in r0
 
-Callee must save/restore r4+ (except r12) if it modifies them.
+Callee must save/restore r4+ (except r12) if it modifies them.  If VFP
+is present, registers s16-s31 (a/k/a d8-d15, a/k/a q4-q7) must be preserved,
+s0-s15 (d0-d7, q0-a3) do not need to be.
 
 Stack is "full descending".  Only the arguments that don't fit in the first 4
 registers are placed on the stack.  "sp" points at the first stacked argument
@@ -181,6 +183,12 @@
 #define SET_VREG(_reg, _vreg)   str     _reg, [rFP, _vreg, lsl #2]
 
 /*
+ * Convert a virtual register index into an address.
+ */
+#define VREG_INDEX_TO_ADDR(_reg, _vreg) \
+        add     _reg, rFP, _vreg, lsl #2
+
+/*
  * This is a #include, not a %include, because we want the C pre-processor
  * to expand the macros into assembler assignment statements.
  */
diff --git a/vm/mterp/out/InterpC-armv4t.c b/vm/mterp/out/InterpC-armv4t.c
index 39cc993..2f4b2e8 100644
--- a/vm/mterp/out/InterpC-armv4t.c
+++ b/vm/mterp/out/InterpC-armv4t.c
@@ -1270,12 +1270,12 @@
      * It is a direct (non-virtual) method if it is static, private,
      * or a constructor.
      */
-    bool isDirect = 
+    bool isDirect =
         ((method->accessFlags & (ACC_STATIC|ACC_PRIVATE)) != 0) ||
         (method->name[0] == '<');
 
     char* desc = dexProtoCopyMethodDescriptor(&method->prototype);
-        
+
     printf("<%c:%s.%s %s> ",
             isDirect ? 'D' : 'V',
             method->clazz->descriptor,
diff --git a/vm/mterp/out/InterpC-armv5te-vfp.c b/vm/mterp/out/InterpC-armv5te-vfp.c
new file mode 100644
index 0000000..92c29fb
--- /dev/null
+++ b/vm/mterp/out/InterpC-armv5te-vfp.c
@@ -0,0 +1,1287 @@
+/*
+ * This file was generated automatically by gen-mterp.py for 'armv5te-vfp'.
+ *
+ * --> DO NOT EDIT <--
+ */
+
+/* File: c/header.c */
+/*
+ * Copyright (C) 2008 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/* common includes */
+#include "Dalvik.h"
+#include "interp/InterpDefs.h"
+#include "mterp/Mterp.h"
+#include <math.h>                   // needed for fmod, fmodf
+
+/*
+ * Configuration defines.  These affect the C implementations, i.e. the
+ * portable interpreter(s) and C stubs.
+ *
+ * Some defines are controlled by the Makefile, e.g.:
+ *   WITH_PROFILER
+ *   WITH_DEBUGGER
+ *   WITH_INSTR_CHECKS
+ *   WITH_TRACKREF_CHECKS
+ *   EASY_GDB
+ *   NDEBUG
+ *
+ * If THREADED_INTERP is not defined, we use a classic "while true / switch"
+ * interpreter.  If it is defined, then the tail end of each instruction
+ * handler fetches the next instruction and jumps directly to the handler.
+ * This increases the size of the "Std" interpreter by about 10%, but
+ * provides a speedup of about the same magnitude.
+ *
+ * There's a "hybrid" approach that uses a goto table instead of a switch
+ * statement, avoiding the "is the opcode in range" tests required for switch.
+ * The performance is close to the threaded version, and without the 10%
+ * size increase, but the benchmark results are off enough that it's not
+ * worth adding as a third option.
+ */
+#define THREADED_INTERP             /* threaded vs. while-loop interpreter */
+
+#ifdef WITH_INSTR_CHECKS            /* instruction-level paranoia (slow!) */
+# define CHECK_BRANCH_OFFSETS
+# define CHECK_REGISTER_INDICES
+#endif
+
+/*
+ * ARM EABI requires 64-bit alignment for access to 64-bit data types.  We
+ * can't just use pointers to copy 64-bit values out of our interpreted
+ * register set, because gcc will generate ldrd/strd.
+ *
+ * The __UNION version copies data in and out of a union.  The __MEMCPY
+ * version uses a memcpy() call to do the transfer; gcc is smart enough to
+ * not actually call memcpy().  The __UNION version is very bad on ARM;
+ * it only uses one more instruction than __MEMCPY, but for some reason
+ * gcc thinks it needs separate storage for every instance of the union.
+ * On top of that, it feels the need to zero them out at the start of the
+ * method.  Net result is we zero out ~700 bytes of stack space at the top
+ * of the interpreter using ARM STM instructions.
+ */
+#if defined(__ARM_EABI__)
+//# define NO_UNALIGN_64__UNION
+# define NO_UNALIGN_64__MEMCPY
+#endif
+
+//#define LOG_INSTR                   /* verbose debugging */
+/* set and adjust ANDROID_LOG_TAGS='*:i jdwp:i dalvikvm:i dalvikvmi:i' */
+
+/*
+ * Keep a tally of accesses to fields.  Currently only works if full DEX
+ * optimization is disabled.
+ */
+#ifdef PROFILE_FIELD_ACCESS
+# define UPDATE_FIELD_GET(_field) { (_field)->gets++; }
+# define UPDATE_FIELD_PUT(_field) { (_field)->puts++; }
+#else
+# define UPDATE_FIELD_GET(_field) ((void)0)
+# define UPDATE_FIELD_PUT(_field) ((void)0)
+#endif
+
+/*
+ * Export another copy of the PC on every instruction; this is largely
+ * redundant with EXPORT_PC and the debugger code.  This value can be
+ * compared against what we have stored on the stack with EXPORT_PC to
+ * help ensure that we aren't missing any export calls.
+ */
+#if WITH_EXTRA_GC_CHECKS > 1
+# define EXPORT_EXTRA_PC() (self->currentPc2 = pc)
+#else
+# define EXPORT_EXTRA_PC()
+#endif
+
+/*
+ * Adjust the program counter.  "_offset" is a signed int, in 16-bit units.
+ *
+ * Assumes the existence of "const u2* pc" and "const u2* curMethod->insns".
+ *
+ * We don't advance the program counter until we finish an instruction or
+ * branch, because we do want to have to unroll the PC if there's an
+ * exception.
+ */
+#ifdef CHECK_BRANCH_OFFSETS
+# define ADJUST_PC(_offset) do {                                            \
+        int myoff = _offset;        /* deref only once */                   \
+        if (pc + myoff < curMethod->insns ||                                \
+            pc + myoff >= curMethod->insns + dvmGetMethodInsnsSize(curMethod)) \
+        {                                                                   \
+            char* desc;                                                     \
+            desc = dexProtoCopyMethodDescriptor(&curMethod->prototype);     \
+            LOGE("Invalid branch %d at 0x%04x in %s.%s %s\n",               \
+                myoff, (int) (pc - curMethod->insns),                       \
+                curMethod->clazz->descriptor, curMethod->name, desc);       \
+            free(desc);                                                     \
+            dvmAbort();                                                     \
+        }                                                                   \
+        pc += myoff;                                                        \
+        EXPORT_EXTRA_PC();                                                  \
+    } while (false)
+#else
+# define ADJUST_PC(_offset) do {                                            \
+        pc += _offset;                                                      \
+        EXPORT_EXTRA_PC();                                                  \
+    } while (false)
+#endif
+
+/*
+ * If enabled, log instructions as we execute them.
+ */
+#ifdef LOG_INSTR
+# define ILOGD(...) ILOG(LOG_DEBUG, __VA_ARGS__)
+# define ILOGV(...) ILOG(LOG_VERBOSE, __VA_ARGS__)
+# define ILOG(_level, ...) do {                                             \
+        char debugStrBuf[128];                                              \
+        snprintf(debugStrBuf, sizeof(debugStrBuf), __VA_ARGS__);            \
+        if (curMethod != NULL)                                                 \
+            LOG(_level, LOG_TAG"i", "%-2d|%04x%s\n",                        \
+                self->threadId, (int)(pc - curMethod->insns), debugStrBuf); \
+        else                                                                \
+            LOG(_level, LOG_TAG"i", "%-2d|####%s\n",                        \
+                self->threadId, debugStrBuf);                               \
+    } while(false)
+void dvmDumpRegs(const Method* method, const u4* framePtr, bool inOnly);
+# define DUMP_REGS(_meth, _frame, _inOnly) dvmDumpRegs(_meth, _frame, _inOnly)
+static const char kSpacing[] = "            ";
+#else
+# define ILOGD(...) ((void)0)
+# define ILOGV(...) ((void)0)
+# define DUMP_REGS(_meth, _frame, _inOnly) ((void)0)
+#endif
+
+/* get a long from an array of u4 */
+static inline s8 getLongFromArray(const u4* ptr, int idx)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { s8 ll; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.parts[0] = ptr[0];
+    conv.parts[1] = ptr[1];
+    return conv.ll;
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    s8 val;
+    memcpy(&val, &ptr[idx], 8);
+    return val;
+#else
+    return *((s8*) &ptr[idx]);
+#endif
+}
+
+/* store a long into an array of u4 */
+static inline void putLongToArray(u4* ptr, int idx, s8 val)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { s8 ll; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.ll = val;
+    ptr[0] = conv.parts[0];
+    ptr[1] = conv.parts[1];
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    memcpy(&ptr[idx], &val, 8);
+#else
+    *((s8*) &ptr[idx]) = val;
+#endif
+}
+
+/* get a double from an array of u4 */
+static inline double getDoubleFromArray(const u4* ptr, int idx)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { double d; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.parts[0] = ptr[0];
+    conv.parts[1] = ptr[1];
+    return conv.d;
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    double dval;
+    memcpy(&dval, &ptr[idx], 8);
+    return dval;
+#else
+    return *((double*) &ptr[idx]);
+#endif
+}
+
+/* store a double into an array of u4 */
+static inline void putDoubleToArray(u4* ptr, int idx, double dval)
+{
+#if defined(NO_UNALIGN_64__UNION)
+    union { double d; u4 parts[2]; } conv;
+
+    ptr += idx;
+    conv.d = dval;
+    ptr[0] = conv.parts[0];
+    ptr[1] = conv.parts[1];
+#elif defined(NO_UNALIGN_64__MEMCPY)
+    memcpy(&ptr[idx], &dval, 8);
+#else
+    *((double*) &ptr[idx]) = dval;
+#endif
+}
+
+/*
+ * If enabled, validate the register number on every access.  Otherwise,
+ * just do an array access.
+ *
+ * Assumes the existence of "u4* fp".
+ *
+ * "_idx" may be referenced more than once.
+ */
+#ifdef CHECK_REGISTER_INDICES
+# define GET_REGISTER(_idx) \
+    ( (_idx) < curMethod->registersSize ? \
+        (fp[(_idx)]) : (assert(!"bad reg"),1969) )
+# define SET_REGISTER(_idx, _val) \
+    ( (_idx) < curMethod->registersSize ? \
+        (fp[(_idx)] = (u4)(_val)) : (assert(!"bad reg"),1969) )
+# define GET_REGISTER_AS_OBJECT(_idx)       ((Object *)GET_REGISTER(_idx))
+# define SET_REGISTER_AS_OBJECT(_idx, _val) SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_INT(_idx) ((s4) GET_REGISTER(_idx))
+# define SET_REGISTER_INT(_idx, _val) SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_WIDE(_idx) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        getLongFromArray(fp, (_idx)) : (assert(!"bad reg"),1969) )
+# define SET_REGISTER_WIDE(_idx, _val) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        putLongToArray(fp, (_idx), (_val)) : (assert(!"bad reg"),1969) )
+# define GET_REGISTER_FLOAT(_idx) \
+    ( (_idx) < curMethod->registersSize ? \
+        (*((float*) &fp[(_idx)])) : (assert(!"bad reg"),1969.0f) )
+# define SET_REGISTER_FLOAT(_idx, _val) \
+    ( (_idx) < curMethod->registersSize ? \
+        (*((float*) &fp[(_idx)]) = (_val)) : (assert(!"bad reg"),1969.0f) )
+# define GET_REGISTER_DOUBLE(_idx) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        getDoubleFromArray(fp, (_idx)) : (assert(!"bad reg"),1969.0) )
+# define SET_REGISTER_DOUBLE(_idx, _val) \
+    ( (_idx) < curMethod->registersSize-1 ? \
+        putDoubleToArray(fp, (_idx), (_val)) : (assert(!"bad reg"),1969.0) )
+#else
+# define GET_REGISTER(_idx)                 (fp[(_idx)])
+# define SET_REGISTER(_idx, _val)           (fp[(_idx)] = (_val))
+# define GET_REGISTER_AS_OBJECT(_idx)       ((Object*) fp[(_idx)])
+# define SET_REGISTER_AS_OBJECT(_idx, _val) (fp[(_idx)] = (u4)(_val))
+# define GET_REGISTER_INT(_idx)             ((s4)GET_REGISTER(_idx))
+# define SET_REGISTER_INT(_idx, _val)       SET_REGISTER(_idx, (s4)_val)
+# define GET_REGISTER_WIDE(_idx)            getLongFromArray(fp, (_idx))
+# define SET_REGISTER_WIDE(_idx, _val)      putLongToArray(fp, (_idx), (_val))
+# define GET_REGISTER_FLOAT(_idx)           (*((float*) &fp[(_idx)]))
+# define SET_REGISTER_FLOAT(_idx, _val)     (*((float*) &fp[(_idx)]) = (_val))
+# define GET_REGISTER_DOUBLE(_idx)          getDoubleFromArray(fp, (_idx))
+# define SET_REGISTER_DOUBLE(_idx, _val)    putDoubleToArray(fp, (_idx), (_val))
+#endif
+
+/*
+ * Get 16 bits from the specified offset of the program counter.  We always
+ * want to load 16 bits at a time from the instruction stream -- it's more
+ * efficient than 8 and won't have the alignment problems that 32 might.
+ *
+ * Assumes existence of "const u2* pc".
+ */
+#define FETCH(_offset)     (pc[(_offset)])
+
+/*
+ * Extract instruction byte from 16-bit fetch (_inst is a u2).
+ */
+#define INST_INST(_inst)    ((_inst) & 0xff)
+
+/*
+ * Extract the "vA, vB" 4-bit registers from the instruction word (_inst is u2).
+ */
+#define INST_A(_inst)       (((_inst) >> 8) & 0x0f)
+#define INST_B(_inst)       ((_inst) >> 12)
+
+/*
+ * Get the 8-bit "vAA" 8-bit register index from the instruction word.
+ * (_inst is u2)
+ */
+#define INST_AA(_inst)      ((_inst) >> 8)
+
+/*
+ * The current PC must be available to Throwable constructors, e.g.
+ * those created by dvmThrowException(), so that the exception stack
+ * trace can be generated correctly.  If we don't do this, the offset
+ * within the current method won't be shown correctly.  See the notes
+ * in Exception.c.
+ *
+ * This is also used to determine the address for precise GC.
+ *
+ * Assumes existence of "u4* fp" and "const u2* pc".
+ */
+#define EXPORT_PC()         (SAVEAREA_FROM_FP(fp)->xtra.currentPc = pc)
+
+/*
+ * Determine if we need to switch to a different interpreter.  "_current"
+ * is either INTERP_STD or INTERP_DBG.  It should be fixed for a given
+ * interpreter generation file, which should remove the outer conditional
+ * from the following.
+ *
+ * If we're building without debug and profiling support, we never switch.
+ */
+#if defined(WITH_PROFILER) || defined(WITH_DEBUGGER)
+# define NEED_INTERP_SWITCH(_current) (                                     \
+    (_current == INTERP_STD) ?                                              \
+        dvmDebuggerOrProfilerActive() : !dvmDebuggerOrProfilerActive() )
+#else
+# define NEED_INTERP_SWITCH(_current) (false)
+#endif
+
+/*
+ * Look up an interface on a class using the cache.
+ */
+INLINE Method* dvmFindInterfaceMethodInCache(ClassObject* thisClass,
+    u4 methodIdx, const Method* method, DvmDex* methodClassDex)
+{
+#define ATOMIC_CACHE_CALC \
+    dvmInterpFindInterfaceMethod(thisClass, methodIdx, method, methodClassDex)
+
+    return (Method*) ATOMIC_CACHE_LOOKUP(methodClassDex->pInterfaceCache,
+                DEX_INTERFACE_CACHE_SIZE, thisClass, methodIdx);
+
+#undef ATOMIC_CACHE_CALC
+}
+
+/*
+ * Check to see if "obj" is NULL.  If so, throw an exception.  Assumes the
+ * pc has already been exported to the stack.
+ *
+ * Perform additional checks on debug builds.
+ *
+ * Use this to check for NULL when the instruction handler calls into
+ * something that could throw an exception (so we have already called
+ * EXPORT_PC at the top).
+ */
+static inline bool checkForNull(Object* obj)
+{
+    if (obj == NULL) {
+        dvmThrowException("Ljava/lang/NullPointerException;", NULL);
+        return false;
+    }
+#ifdef WITH_EXTRA_OBJECT_VALIDATION
+    if (!dvmIsValidObject(obj)) {
+        LOGE("Invalid object %p\n", obj);
+        dvmAbort();
+    }
+#endif
+#ifndef NDEBUG
+    if (obj->clazz == NULL || ((u4) obj->clazz) <= 65536) {
+        /* probable heap corruption */
+        LOGE("Invalid object class %p (in %p)\n", obj->clazz, obj);
+        dvmAbort();
+    }
+#endif
+    return true;
+}
+
+/*
+ * Check to see if "obj" is NULL.  If so, export the PC into the stack
+ * frame and throw an exception.
+ *
+ * Perform additional checks on debug builds.
+ *
+ * Use this to check for NULL when the instruction handler doesn't do
+ * anything else that can throw an exception.
+ */
+static inline bool checkForNullExportPC(Object* obj, u4* fp, const u2* pc)
+{
+    if (obj == NULL) {
+        EXPORT_PC();
+        dvmThrowException("Ljava/lang/NullPointerException;", NULL);
+        return false;
+    }
+#ifdef WITH_EXTRA_OBJECT_VALIDATION
+    if (!dvmIsValidObject(obj)) {
+        LOGE("Invalid object %p\n", obj);
+        dvmAbort();
+    }
+#endif
+#ifndef NDEBUG
+    if (obj->clazz == NULL || ((u4) obj->clazz) <= 65536) {
+        /* probable heap corruption */
+        LOGE("Invalid object class %p (in %p)\n", obj->clazz, obj);
+        dvmAbort();
+    }
+#endif
+    return true;
+}
+
+
+/* File: cstubs/stubdefs.c */
+/* this is a standard (no debug support) interpreter */
+#define INTERP_TYPE INTERP_STD
+#define CHECK_DEBUG_AND_PROF() ((void)0)
+# define CHECK_TRACKED_REFS() ((void)0)
+
+/*
+ * In the C mterp stubs, "goto" is a function call followed immediately
+ * by a return.
+ */
+
+#define GOTO_TARGET_DECL(_target, ...)                                      \
+    void dvmMterp_##_target(MterpGlue* glue, ## __VA_ARGS__);
+
+#define GOTO_TARGET(_target, ...)                                           \
+    void dvmMterp_##_target(MterpGlue* glue, ## __VA_ARGS__) {              \
+        u2 ref, vsrc1, vsrc2, vdst;                                         \
+        u2 inst = FETCH(0);                                                 \
+        const Method* methodToCall;                                         \
+        StackSaveArea* debugSaveArea;
+
+#define GOTO_TARGET_END }
+
+/*
+ * Redefine what used to be local variable accesses into MterpGlue struct
+ * references.  (These are undefined down in "footer.c".)
+ */
+#define retval                  glue->retval
+#define pc                      glue->pc
+#define fp                      glue->fp
+#define curMethod               glue->method
+#define methodClassDex          glue->methodClassDex
+#define self                    glue->self
+#define debugTrackedRefStart    glue->debugTrackedRefStart
+
+/* ugh */
+#define STUB_HACK(x) x
+
+
+/*
+ * Opcode handler framing macros.  Here, each opcode is a separate function
+ * that takes a "glue" argument and returns void.  We can't declare
+ * these "static" because they may be called from an assembly stub.
+ */
+#define HANDLE_OPCODE(_op)                                                  \
+    void dvmMterp_##_op(MterpGlue* glue) {                                  \
+        u2 ref, vsrc1, vsrc2, vdst;                                         \
+        u2 inst = FETCH(0);
+
+#define OP_END }
+
+/*
+ * Like the "portable" FINISH, but don't reload "inst", and return to caller
+ * when done.
+ */
+#define FINISH(_offset) {                                                   \
+        ADJUST_PC(_offset);                                                 \
+        CHECK_DEBUG_AND_PROF();                                             \
+        CHECK_TRACKED_REFS();                                               \
+        return;                                                             \
+    }
+
+
+/*
+ * The "goto label" statements turn into function calls followed by
+ * return statements.  Some of the functions take arguments, which in the
+ * portable interpreter are handled by assigning values to globals.
+ */
+
+#define GOTO_exceptionThrown()                                              \
+    do {                                                                    \
+        dvmMterp_exceptionThrown(glue);                                     \
+        return;                                                             \
+    } while(false)
+
+#define GOTO_returnFromMethod()                                             \
+    do {                                                                    \
+        dvmMterp_returnFromMethod(glue);                                    \
+        return;                                                             \
+    } while(false)
+
+#define GOTO_invoke(_target, _methodCallRange)                              \
+    do {                                                                    \
+        dvmMterp_##_target(glue, _methodCallRange);                         \
+        return;                                                             \
+    } while(false)
+
+#define GOTO_invokeMethod(_methodCallRange, _methodToCall, _vsrc1, _vdst)   \
+    do {                                                                    \
+        dvmMterp_invokeMethod(glue, _methodCallRange, _methodToCall,        \
+            _vsrc1, _vdst);                                                 \
+        return;                                                             \
+    } while(false)
+
+/*
+ * As a special case, "goto bail" turns into a longjmp.  Use "bail_switch"
+ * if we need to switch to the other interpreter upon our return.
+ */
+#define GOTO_bail()                                                         \
+    dvmMterpStdBail(glue, false);
+#define GOTO_bail_switch()                                                  \
+    dvmMterpStdBail(glue, true);
+
+/*
+ * Periodically check for thread suspension.
+ *
+ * While we're at it, see if a debugger has attached or the profiler has
+ * started.  If so, switch to a different "goto" table.
+ */
+#define PERIODIC_CHECKS(_entryPoint, _pcadj) {                              \
+        if (dvmCheckSuspendQuick(self)) {                                   \
+            EXPORT_PC();  /* need for precise GC */                         \
+            dvmCheckSuspendPending(self);                                   \
+        }                                                                   \
+        if (NEED_INTERP_SWITCH(INTERP_TYPE)) {                              \
+            ADJUST_PC(_pcadj);                                              \
+            glue->entryPoint = _entryPoint;                                 \
+            LOGVV("threadid=%d: switch to STD ep=%d adj=%d\n",              \
+                glue->self->threadId, (_entryPoint), (_pcadj));             \
+            GOTO_bail_switch();                                             \
+        }                                                                   \
+    }
+
+
+/* File: c/opcommon.c */
+/* forward declarations of goto targets */
+GOTO_TARGET_DECL(filledNewArray, bool methodCallRange);
+GOTO_TARGET_DECL(invokeVirtual, bool methodCallRange);
+GOTO_TARGET_DECL(invokeSuper, bool methodCallRange);
+GOTO_TARGET_DECL(invokeInterface, bool methodCallRange);
+GOTO_TARGET_DECL(invokeDirect, bool methodCallRange);
+GOTO_TARGET_DECL(invokeStatic, bool methodCallRange);
+GOTO_TARGET_DECL(invokeVirtualQuick, bool methodCallRange);
+GOTO_TARGET_DECL(invokeSuperQuick, bool methodCallRange);
+GOTO_TARGET_DECL(invokeMethod, bool methodCallRange, const Method* methodToCall,
+    u2 count, u2 regs);
+GOTO_TARGET_DECL(returnFromMethod);
+GOTO_TARGET_DECL(exceptionThrown);
+
+/*
+ * ===========================================================================
+ *
+ * What follows are opcode definitions shared between multiple opcodes with
+ * minor substitutions handled by the C pre-processor.  These should probably
+ * use the mterp substitution mechanism instead, with the code here moved
+ * into common fragment files (like the asm "binop.S"), although it's hard
+ * to give up the C preprocessor in favor of the much simpler text subst.
+ *
+ * ===========================================================================
+ */
+
+#define HANDLE_NUMCONV(_opcode, _opname, _fromtype, _totype)                \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        SET_REGISTER##_totype(vdst,                                         \
+            GET_REGISTER##_fromtype(vsrc1));                                \
+        FINISH(1);
+
+#define HANDLE_FLOAT_TO_INT(_opcode, _opname, _fromvtype, _fromrtype,       \
+        _tovtype, _tortype)                                                 \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+    {                                                                       \
+        /* spec defines specific handling for +/- inf and NaN values */     \
+        _fromvtype val;                                                     \
+        _tovtype intMin, intMax, result;                                    \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        val = GET_REGISTER##_fromrtype(vsrc1);                              \
+        intMin = (_tovtype) 1 << (sizeof(_tovtype) * 8 -1);                 \
+        intMax = ~intMin;                                                   \
+        result = (_tovtype) val;                                            \
+        if (val >= intMax)          /* +inf */                              \
+            result = intMax;                                                \
+        else if (val <= intMin)     /* -inf */                              \
+            result = intMin;                                                \
+        else if (val != val)        /* NaN */                               \
+            result = 0;                                                     \
+        else                                                                \
+            result = (_tovtype) val;                                        \
+        SET_REGISTER##_tortype(vdst, result);                               \
+    }                                                                       \
+    FINISH(1);
+
+#define HANDLE_INT_TO_SMALL(_opcode, _opname, _type)                        \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|int-to-%s v%d,v%d", (_opname), vdst, vsrc1);                \
+        SET_REGISTER(vdst, (_type) GET_REGISTER(vsrc1));                    \
+        FINISH(1);
+
+/* NOTE: the comparison result is always a signed 4-byte integer */
+#define HANDLE_OP_CMPX(_opcode, _opname, _varType, _type, _nanVal)          \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        int result;                                                         \
+        u2 regs;                                                            \
+        _varType val1, val2;                                                \
+        vdst = INST_AA(inst);                                               \
+        regs = FETCH(1);                                                    \
+        vsrc1 = regs & 0xff;                                                \
+        vsrc2 = regs >> 8;                                                  \
+        ILOGV("|cmp%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);         \
+        val1 = GET_REGISTER##_type(vsrc1);                                  \
+        val2 = GET_REGISTER##_type(vsrc2);                                  \
+        if (val1 == val2)                                                   \
+            result = 0;                                                     \
+        else if (val1 < val2)                                               \
+            result = -1;                                                    \
+        else if (val1 > val2)                                               \
+            result = 1;                                                     \
+        else                                                                \
+            result = (_nanVal);                                             \
+        ILOGV("+ result=%d\n", result);                                     \
+        SET_REGISTER(vdst, result);                                         \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_IF_XX(_opcode, _opname, _cmp)                             \
+    HANDLE_OPCODE(_opcode /*vA, vB, +CCCC*/)                                \
+        vsrc1 = INST_A(inst);                                               \
+        vsrc2 = INST_B(inst);                                               \
+        if ((s4) GET_REGISTER(vsrc1) _cmp (s4) GET_REGISTER(vsrc2)) {       \
+            int branchOffset = (s2)FETCH(1);    /* sign-extended */         \
+            ILOGV("|if-%s v%d,v%d,+0x%04x", (_opname), vsrc1, vsrc2,        \
+                branchOffset);                                              \
+            ILOGV("> branch taken");                                        \
+            if (branchOffset < 0)                                           \
+                PERIODIC_CHECKS(kInterpEntryInstr, branchOffset);           \
+            FINISH(branchOffset);                                           \
+        } else {                                                            \
+            ILOGV("|if-%s v%d,v%d,-", (_opname), vsrc1, vsrc2);             \
+            FINISH(2);                                                      \
+        }
+
+#define HANDLE_OP_IF_XXZ(_opcode, _opname, _cmp)                            \
+    HANDLE_OPCODE(_opcode /*vAA, +BBBB*/)                                   \
+        vsrc1 = INST_AA(inst);                                              \
+        if ((s4) GET_REGISTER(vsrc1) _cmp 0) {                              \
+            int branchOffset = (s2)FETCH(1);    /* sign-extended */         \
+            ILOGV("|if-%s v%d,+0x%04x", (_opname), vsrc1, branchOffset);    \
+            ILOGV("> branch taken");                                        \
+            if (branchOffset < 0)                                           \
+                PERIODIC_CHECKS(kInterpEntryInstr, branchOffset);           \
+            FINISH(branchOffset);                                           \
+        } else {                                                            \
+            ILOGV("|if-%s v%d,-", (_opname), vsrc1);                        \
+            FINISH(2);                                                      \
+        }
+
+#define HANDLE_UNOP(_opcode, _opname, _pfx, _sfx, _type)                    \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s v%d,v%d", (_opname), vdst, vsrc1);                       \
+        SET_REGISTER##_type(vdst, _pfx GET_REGISTER##_type(vsrc1) _sfx);    \
+        FINISH(1);
+
+#define HANDLE_OP_X_INT(_opcode, _opname, _op, _chkdiv)                     \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-int v%d,v%d", (_opname), vdst, vsrc1);                   \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            secondVal = GET_REGISTER(vsrc2);                                \
+            if (secondVal == 0) {                                           \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && secondVal == -1) {            \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            /* non-div/rem case */                                          \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vsrc1) _op (s4) GET_REGISTER(vsrc2));     \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_INT(_opcode, _opname, _cast, _op)                     \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-int v%d,v%d", (_opname), vdst, vsrc1);                   \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vsrc1) _op (GET_REGISTER(vsrc2) & 0x1f));    \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_INT_LIT16(_opcode, _opname, _op, _chkdiv)               \
+    HANDLE_OPCODE(_opcode /*vA, vB, #+CCCC*/)                               \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        vsrc2 = FETCH(1);                                                   \
+        ILOGV("|%s-int/lit16 v%d,v%d,#+0x%04x",                             \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, result;                                            \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            if ((s2) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                      \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && ((s2) vsrc2) == -1) {         \
+                /* won't generate /lit16 instr for this; check anyway */    \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op (s2) vsrc2;                           \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            /* non-div/rem case */                                          \
+            SET_REGISTER(vdst, GET_REGISTER(vsrc1) _op (s2) vsrc2);         \
+        }                                                                   \
+        FINISH(2);
+
+#define HANDLE_OP_X_INT_LIT8(_opcode, _opname, _op, _chkdiv)                \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, #+CC*/)                               \
+    {                                                                       \
+        u2 litInfo;                                                         \
+        vdst = INST_AA(inst);                                               \
+        litInfo = FETCH(1);                                                 \
+        vsrc1 = litInfo & 0xff;                                             \
+        vsrc2 = litInfo >> 8;       /* constant */                          \
+        ILOGV("|%s-int/lit8 v%d,v%d,#+0x%02x",                              \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, result;                                            \
+            firstVal = GET_REGISTER(vsrc1);                                 \
+            if ((s1) vsrc2 == 0) {                                          \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && ((s1) vsrc2) == -1) {         \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op ((s1) vsrc2);                         \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vsrc1) _op (s1) vsrc2);                   \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_INT_LIT8(_opcode, _opname, _cast, _op)                \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, #+CC*/)                               \
+    {                                                                       \
+        u2 litInfo;                                                         \
+        vdst = INST_AA(inst);                                               \
+        litInfo = FETCH(1);                                                 \
+        vsrc1 = litInfo & 0xff;                                             \
+        vsrc2 = litInfo >> 8;       /* constant */                          \
+        ILOGV("|%s-int/lit8 v%d,v%d,#+0x%02x",                              \
+            (_opname), vdst, vsrc1, vsrc2);                                 \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vsrc1) _op (vsrc2 & 0x1f));                  \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_INT_2ADDR(_opcode, _opname, _op, _chkdiv)               \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-int-2addr v%d,v%d", (_opname), vdst, vsrc1);             \
+        if (_chkdiv != 0) {                                                 \
+            s4 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER(vdst);                                  \
+            secondVal = GET_REGISTER(vsrc1);                                \
+            if (secondVal == 0) {                                           \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u4)firstVal == 0x80000000 && secondVal == -1) {            \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER(vdst, result);                                     \
+        } else {                                                            \
+            SET_REGISTER(vdst,                                              \
+                (s4) GET_REGISTER(vdst) _op (s4) GET_REGISTER(vsrc1));      \
+        }                                                                   \
+        FINISH(1);
+
+#define HANDLE_OP_SHX_INT_2ADDR(_opcode, _opname, _cast, _op)               \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-int-2addr v%d,v%d", (_opname), vdst, vsrc1);             \
+        SET_REGISTER(vdst,                                                  \
+            _cast GET_REGISTER(vdst) _op (GET_REGISTER(vsrc1) & 0x1f));     \
+        FINISH(1);
+
+#define HANDLE_OP_X_LONG(_opcode, _opname, _op, _chkdiv)                    \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-long v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);       \
+        if (_chkdiv != 0) {                                                 \
+            s8 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER_WIDE(vsrc1);                            \
+            secondVal = GET_REGISTER_WIDE(vsrc2);                           \
+            if (secondVal == 0LL) {                                         \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u8)firstVal == 0x8000000000000000ULL &&                    \
+                secondVal == -1LL)                                          \
+            {                                                               \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER_WIDE(vdst, result);                                \
+        } else {                                                            \
+            SET_REGISTER_WIDE(vdst,                                         \
+                (s8) GET_REGISTER_WIDE(vsrc1) _op (s8) GET_REGISTER_WIDE(vsrc2)); \
+        }                                                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_SHX_LONG(_opcode, _opname, _cast, _op)                    \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-long v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);       \
+        SET_REGISTER_WIDE(vdst,                                             \
+            _cast GET_REGISTER_WIDE(vsrc1) _op (GET_REGISTER(vsrc2) & 0x3f)); \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_LONG_2ADDR(_opcode, _opname, _op, _chkdiv)              \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-long-2addr v%d,v%d", (_opname), vdst, vsrc1);            \
+        if (_chkdiv != 0) {                                                 \
+            s8 firstVal, secondVal, result;                                 \
+            firstVal = GET_REGISTER_WIDE(vdst);                             \
+            secondVal = GET_REGISTER_WIDE(vsrc1);                           \
+            if (secondVal == 0LL) {                                         \
+                EXPORT_PC();                                                \
+                dvmThrowException("Ljava/lang/ArithmeticException;",        \
+                    "divide by zero");                                      \
+                GOTO_exceptionThrown();                                     \
+            }                                                               \
+            if ((u8)firstVal == 0x8000000000000000ULL &&                    \
+                secondVal == -1LL)                                          \
+            {                                                               \
+                if (_chkdiv == 1)                                           \
+                    result = firstVal;  /* division */                      \
+                else                                                        \
+                    result = 0;         /* remainder */                     \
+            } else {                                                        \
+                result = firstVal _op secondVal;                            \
+            }                                                               \
+            SET_REGISTER_WIDE(vdst, result);                                \
+        } else {                                                            \
+            SET_REGISTER_WIDE(vdst,                                         \
+                (s8) GET_REGISTER_WIDE(vdst) _op (s8)GET_REGISTER_WIDE(vsrc1));\
+        }                                                                   \
+        FINISH(1);
+
+#define HANDLE_OP_SHX_LONG_2ADDR(_opcode, _opname, _cast, _op)              \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-long-2addr v%d,v%d", (_opname), vdst, vsrc1);            \
+        SET_REGISTER_WIDE(vdst,                                             \
+            _cast GET_REGISTER_WIDE(vdst) _op (GET_REGISTER(vsrc1) & 0x3f)); \
+        FINISH(1);
+
+#define HANDLE_OP_X_FLOAT(_opcode, _opname, _op)                            \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-float v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);      \
+        SET_REGISTER_FLOAT(vdst,                                            \
+            GET_REGISTER_FLOAT(vsrc1) _op GET_REGISTER_FLOAT(vsrc2));       \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_DOUBLE(_opcode, _opname, _op)                           \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        u2 srcRegs;                                                         \
+        vdst = INST_AA(inst);                                               \
+        srcRegs = FETCH(1);                                                 \
+        vsrc1 = srcRegs & 0xff;                                             \
+        vsrc2 = srcRegs >> 8;                                               \
+        ILOGV("|%s-double v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);     \
+        SET_REGISTER_DOUBLE(vdst,                                           \
+            GET_REGISTER_DOUBLE(vsrc1) _op GET_REGISTER_DOUBLE(vsrc2));     \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_X_FLOAT_2ADDR(_opcode, _opname, _op)                      \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-float-2addr v%d,v%d", (_opname), vdst, vsrc1);           \
+        SET_REGISTER_FLOAT(vdst,                                            \
+            GET_REGISTER_FLOAT(vdst) _op GET_REGISTER_FLOAT(vsrc1));        \
+        FINISH(1);
+
+#define HANDLE_OP_X_DOUBLE_2ADDR(_opcode, _opname, _op)                     \
+    HANDLE_OPCODE(_opcode /*vA, vB*/)                                       \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);                                               \
+        ILOGV("|%s-double-2addr v%d,v%d", (_opname), vdst, vsrc1);          \
+        SET_REGISTER_DOUBLE(vdst,                                           \
+            GET_REGISTER_DOUBLE(vdst) _op GET_REGISTER_DOUBLE(vsrc1));      \
+        FINISH(1);
+
+#define HANDLE_OP_AGET(_opcode, _opname, _type, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        ArrayObject* arrayObj;                                              \
+        u2 arrayInfo;                                                       \
+        EXPORT_PC();                                                        \
+        vdst = INST_AA(inst);                                               \
+        arrayInfo = FETCH(1);                                               \
+        vsrc1 = arrayInfo & 0xff;    /* array ptr */                        \
+        vsrc2 = arrayInfo >> 8;      /* index */                            \
+        ILOGV("|aget%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);        \
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);                      \
+        if (!checkForNull((Object*) arrayObj))                              \
+            GOTO_exceptionThrown();                                         \
+        if (GET_REGISTER(vsrc2) >= arrayObj->length) {                      \
+            LOGV("Invalid array access: %p %d (len=%d)\n",                  \
+                arrayObj, vsrc2, arrayObj->length);                         \
+            dvmThrowException("Ljava/lang/ArrayIndexOutOfBoundsException;", \
+                NULL);                                                      \
+            GOTO_exceptionThrown();                                         \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst,                                        \
+            ((_type*) arrayObj->contents)[GET_REGISTER(vsrc2)]);            \
+        ILOGV("+ AGET[%d]=0x%x", GET_REGISTER(vsrc2), GET_REGISTER(vdst));  \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_OP_APUT(_opcode, _opname, _type, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, vBB, vCC*/)                                \
+    {                                                                       \
+        ArrayObject* arrayObj;                                              \
+        u2 arrayInfo;                                                       \
+        EXPORT_PC();                                                        \
+        vdst = INST_AA(inst);       /* AA: source value */                  \
+        arrayInfo = FETCH(1);                                               \
+        vsrc1 = arrayInfo & 0xff;   /* BB: array ptr */                     \
+        vsrc2 = arrayInfo >> 8;     /* CC: index */                         \
+        ILOGV("|aput%s v%d,v%d,v%d", (_opname), vdst, vsrc1, vsrc2);        \
+        arrayObj = (ArrayObject*) GET_REGISTER(vsrc1);                      \
+        if (!checkForNull((Object*) arrayObj))                              \
+            GOTO_exceptionThrown();                                         \
+        if (GET_REGISTER(vsrc2) >= arrayObj->length) {                      \
+            dvmThrowException("Ljava/lang/ArrayIndexOutOfBoundsException;", \
+                NULL);                                                      \
+            GOTO_exceptionThrown();                                         \
+        }                                                                   \
+        ILOGV("+ APUT[%d]=0x%08x", GET_REGISTER(vsrc2), GET_REGISTER(vdst));\
+        ((_type*) arrayObj->contents)[GET_REGISTER(vsrc2)] =                \
+            GET_REGISTER##_regsize(vdst);                                   \
+    }                                                                       \
+    FINISH(2);
+
+/*
+ * It's possible to get a bad value out of a field with sub-32-bit stores
+ * because the -quick versions always operate on 32 bits.  Consider:
+ *   short foo = -1  (sets a 32-bit register to 0xffffffff)
+ *   iput-quick foo  (writes all 32 bits to the field)
+ *   short bar = 1   (sets a 32-bit register to 0x00000001)
+ *   iput-short      (writes the low 16 bits to the field)
+ *   iget-quick foo  (reads all 32 bits from the field, yielding 0xffff0001)
+ * This can only happen when optimized and non-optimized code has interleaved
+ * access to the same field.  This is unlikely but possible.
+ *
+ * The easiest way to fix this is to always read/write 32 bits at a time.  On
+ * a device with a 16-bit data bus this is sub-optimal.  (The alternative
+ * approach is to have sub-int versions of iget-quick, but now we're wasting
+ * Dalvik instruction space and making it less likely that handler code will
+ * already be in the CPU i-cache.)
+ */
+#define HANDLE_IGET_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        InstField* ifield;                                                  \
+        Object* obj;                                                        \
+        EXPORT_PC();                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|iget%s v%d,v%d,field@0x%04x", (_opname), vdst, vsrc1, ref); \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNull(obj))                                             \
+            GOTO_exceptionThrown();                                         \
+        ifield = (InstField*) dvmDexGetResolvedField(methodClassDex, ref);  \
+        if (ifield == NULL) {                                               \
+            ifield = dvmResolveInstField(curMethod->clazz, ref);            \
+            if (ifield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst,                                        \
+            dvmGetField##_ftype(obj, ifield->byteOffset));                  \
+        ILOGV("+ IGET '%s'=0x%08llx", ifield->field.name,                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+        UPDATE_FIELD_GET(&ifield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IGET_X_QUICK(_opcode, _opname, _ftype, _regsize)             \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        Object* obj;                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field offset */                          \
+        ILOGV("|iget%s-quick v%d,v%d,field@+%u",                            \
+            (_opname), vdst, vsrc1, ref);                                   \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNullExportPC(obj, fp, pc))                             \
+            GOTO_exceptionThrown();                                         \
+        SET_REGISTER##_regsize(vdst, dvmGetField##_ftype(obj, ref));        \
+        ILOGV("+ IGETQ %d=0x%08llx", ref,                                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IPUT_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        InstField* ifield;                                                  \
+        Object* obj;                                                        \
+        EXPORT_PC();                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|iput%s v%d,v%d,field@0x%04x", (_opname), vdst, vsrc1, ref); \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNull(obj))                                             \
+            GOTO_exceptionThrown();                                         \
+        ifield = (InstField*) dvmDexGetResolvedField(methodClassDex, ref);  \
+        if (ifield == NULL) {                                               \
+            ifield = dvmResolveInstField(curMethod->clazz, ref);            \
+            if (ifield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        dvmSetField##_ftype(obj, ifield->byteOffset,                        \
+            GET_REGISTER##_regsize(vdst));                                  \
+        ILOGV("+ IPUT '%s'=0x%08llx", ifield->field.name,                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+        UPDATE_FIELD_PUT(&ifield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_IPUT_X_QUICK(_opcode, _opname, _ftype, _regsize)             \
+    HANDLE_OPCODE(_opcode /*vA, vB, field@CCCC*/)                           \
+    {                                                                       \
+        Object* obj;                                                        \
+        vdst = INST_A(inst);                                                \
+        vsrc1 = INST_B(inst);   /* object ptr */                            \
+        ref = FETCH(1);         /* field offset */                          \
+        ILOGV("|iput%s-quick v%d,v%d,field@0x%04x",                         \
+            (_opname), vdst, vsrc1, ref);                                   \
+        obj = (Object*) GET_REGISTER(vsrc1);                                \
+        if (!checkForNullExportPC(obj, fp, pc))                             \
+            GOTO_exceptionThrown();                                         \
+        dvmSetField##_ftype(obj, ref, GET_REGISTER##_regsize(vdst));        \
+        ILOGV("+ IPUTQ %d=0x%08llx", ref,                                   \
+            (u8) GET_REGISTER##_regsize(vdst));                             \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_SGET_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, field@BBBB*/)                              \
+    {                                                                       \
+        StaticField* sfield;                                                \
+        vdst = INST_AA(inst);                                               \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|sget%s v%d,sfield@0x%04x", (_opname), vdst, ref);           \
+        sfield = (StaticField*)dvmDexGetResolvedField(methodClassDex, ref); \
+        if (sfield == NULL) {                                               \
+            EXPORT_PC();                                                    \
+            sfield = dvmResolveStaticField(curMethod->clazz, ref);          \
+            if (sfield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        SET_REGISTER##_regsize(vdst, dvmGetStaticField##_ftype(sfield));    \
+        ILOGV("+ SGET '%s'=0x%08llx",                                       \
+            sfield->field.name, (u8)GET_REGISTER##_regsize(vdst));          \
+        UPDATE_FIELD_GET(&sfield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+#define HANDLE_SPUT_X(_opcode, _opname, _ftype, _regsize)                   \
+    HANDLE_OPCODE(_opcode /*vAA, field@BBBB*/)                              \
+    {                                                                       \
+        StaticField* sfield;                                                \
+        vdst = INST_AA(inst);                                               \
+        ref = FETCH(1);         /* field ref */                             \
+        ILOGV("|sput%s v%d,sfield@0x%04x", (_opname), vdst, ref);           \
+        sfield = (StaticField*)dvmDexGetResolvedField(methodClassDex, ref); \
+        if (sfield == NULL) {                                               \
+            EXPORT_PC();                                                    \
+            sfield = dvmResolveStaticField(curMethod->clazz, ref);          \
+            if (sfield == NULL)                                             \
+                GOTO_exceptionThrown();                                     \
+        }                                                                   \
+        dvmSetStaticField##_ftype(sfield, GET_REGISTER##_regsize(vdst));    \
+        ILOGV("+ SPUT '%s'=0x%08llx",                                       \
+            sfield->field.name, (u8)GET_REGISTER##_regsize(vdst));          \
+        UPDATE_FIELD_PUT(&sfield->field);                                   \
+    }                                                                       \
+    FINISH(2);
+
+
+/* File: cstubs/enddefs.c */
+
+/* undefine "magic" name remapping */
+#undef retval
+#undef pc
+#undef fp
+#undef curMethod
+#undef methodClassDex
+#undef self
+#undef debugTrackedRefStart
+
+/* File: armv5te/debug.c */
+#include <inttypes.h>
+
+/*
+ * Dump the fixed-purpose ARM registers, along with some other info.
+ *
+ * This function MUST be compiled in ARM mode -- THUMB will yield bogus
+ * results.
+ *
+ * This will NOT preserve r0-r3/ip.
+ */
+void dvmMterpDumpArmRegs(uint32_t r0, uint32_t r1, uint32_t r2, uint32_t r3)
+{
+    register uint32_t rPC       asm("r4");
+    register uint32_t rFP       asm("r5");
+    register uint32_t rGLUE     asm("r6");
+    register uint32_t rIBASE    asm("r7");
+    register uint32_t rINST     asm("r8");
+    register uint32_t r9        asm("r9");
+    register uint32_t r10       asm("r10");
+
+    extern char dvmAsmInstructionStart[];
+
+    printf("REGS: r0=%08x r1=%08x r2=%08x r3=%08x\n", r0, r1, r2, r3);
+    printf("    : rPC=%08x rFP=%08x rGLUE=%08x rIBASE=%08x\n",
+        rPC, rFP, rGLUE, rIBASE);
+    printf("    : rINST=%08x r9=%08x r10=%08x\n", rINST, r9, r10);
+
+    MterpGlue* glue = (MterpGlue*) rGLUE;
+    const Method* method = glue->method;
+    printf("    + self is %p\n", dvmThreadSelf());
+    //printf("    + currently in %s.%s %s\n",
+    //    method->clazz->descriptor, method->name, method->signature);
+    //printf("    + dvmAsmInstructionStart = %p\n", dvmAsmInstructionStart);
+    //printf("    + next handler for 0x%02x = %p\n",
+    //    rINST & 0xff, dvmAsmInstructionStart + (rINST & 0xff) * 64);
+}
+
+/*
+ * Dump the StackSaveArea for the specified frame pointer.
+ */
+void dvmDumpFp(void* fp, StackSaveArea* otherSaveArea)
+{
+    StackSaveArea* saveArea = SAVEAREA_FROM_FP(fp);
+    printf("StackSaveArea for fp %p [%p/%p]:\n", fp, saveArea, otherSaveArea);
+#ifdef EASY_GDB
+    printf("  prevSave=%p, prevFrame=%p savedPc=%p meth=%p curPc=%p\n",
+        saveArea->prevSave, saveArea->prevFrame, saveArea->savedPc,
+        saveArea->method, saveArea->xtra.currentPc);
+#else
+    printf("  prevFrame=%p savedPc=%p meth=%p curPc=%p fp[0]=0x%08x\n",
+        saveArea->prevFrame, saveArea->savedPc,
+        saveArea->method, saveArea->xtra.currentPc,
+        *(u4*)fp);
+#endif
+}
+
+/*
+ * Does the bulk of the work for common_printMethod().
+ */
+void dvmMterpPrintMethod(Method* method)
+{
+    /*
+     * It is a direct (non-virtual) method if it is static, private,
+     * or a constructor.
+     */
+    bool isDirect =
+        ((method->accessFlags & (ACC_STATIC|ACC_PRIVATE)) != 0) ||
+        (method->name[0] == '<');
+
+    char* desc = dexProtoCopyMethodDescriptor(&method->prototype);
+
+    printf("<%c:%s.%s %s> ",
+            isDirect ? 'D' : 'V',
+            method->clazz->descriptor,
+            method->name,
+            desc);
+
+    free(desc);
+}
+
diff --git a/vm/mterp/out/InterpC-armv5te.c b/vm/mterp/out/InterpC-armv5te.c
index c8f4ced..70794cc 100644
--- a/vm/mterp/out/InterpC-armv5te.c
+++ b/vm/mterp/out/InterpC-armv5te.c
@@ -1270,12 +1270,12 @@
      * It is a direct (non-virtual) method if it is static, private,
      * or a constructor.
      */
-    bool isDirect = 
+    bool isDirect =
         ((method->accessFlags & (ACC_STATIC|ACC_PRIVATE)) != 0) ||
         (method->name[0] == '<');
 
     char* desc = dexProtoCopyMethodDescriptor(&method->prototype);
-        
+
     printf("<%c:%s.%s %s> ",
             isDirect ? 'D' : 'V',
             method->clazz->descriptor,
diff --git a/vm/mterp/rebuild.sh b/vm/mterp/rebuild.sh
index 1380f69..1faed70 100755
--- a/vm/mterp/rebuild.sh
+++ b/vm/mterp/rebuild.sh
@@ -19,7 +19,7 @@
 # generated as part of the build.
 #
 set -e
-for arch in portstd portdbg allstubs armv4t armv5te x86; do TARGET_ARCH_EXT=$arch make -f Makefile-mterp; done
+for arch in portstd portdbg allstubs armv4t armv5te armv5te-vfp x86; do TARGET_ARCH_EXT=$arch make -f Makefile-mterp; done
 
 # These aren't actually used, so just go ahead and remove them.  The correct
 # approach is to prevent them from being generated in the first place, but
diff --git a/vm/mterp/vfp/OP_ADD_DOUBLE.S b/vm/mterp/vfp/OP_ADD_DOUBLE.S
new file mode 100644
index 0000000..4d4bd64
--- /dev/null
+++ b/vm/mterp/vfp/OP_ADD_DOUBLE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/fbinopWide.S" {"instr":"faddd   d2, d0, d1"}
diff --git a/vm/mterp/vfp/OP_ADD_DOUBLE_2ADDR.S b/vm/mterp/vfp/OP_ADD_DOUBLE_2ADDR.S
new file mode 100644
index 0000000..012f93a
--- /dev/null
+++ b/vm/mterp/vfp/OP_ADD_DOUBLE_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/fbinopWide2addr.S" {"instr":"faddd   d2, d0, d1"}
diff --git a/vm/mterp/vfp/OP_ADD_FLOAT.S b/vm/mterp/vfp/OP_ADD_FLOAT.S
new file mode 100644
index 0000000..c30ce00
--- /dev/null
+++ b/vm/mterp/vfp/OP_ADD_FLOAT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/fbinop.S" {"instr":"fadds   s2, s0, s1"}
diff --git a/vm/mterp/vfp/OP_ADD_FLOAT_2ADDR.S b/vm/mterp/vfp/OP_ADD_FLOAT_2ADDR.S
new file mode 100644
index 0000000..e3f38bb
--- /dev/null
+++ b/vm/mterp/vfp/OP_ADD_FLOAT_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/fbinop2addr.S" {"instr":"fadds   s2, s0, s1"}
diff --git a/vm/mterp/vfp/OP_CMPG_DOUBLE.S b/vm/mterp/vfp/OP_CMPG_DOUBLE.S
new file mode 100644
index 0000000..1a377cf
--- /dev/null
+++ b/vm/mterp/vfp/OP_CMPG_DOUBLE.S
@@ -0,0 +1,43 @@
+%verify "executed"
+%verify "basic lt, gt, eq */
+%verify "left arg NaN"
+%verify "right arg NaN"
+    /*
+     * Compare two floating-point values.  Puts 0, 1, or -1 into the
+     * destination register based on the results of the comparison.
+     *
+     * int compare(x, y) {
+     *     if (x == y) {
+     *         return 0;
+     *     } else if (x < y) {
+     *         return -1;
+     *     } else if (x > y) {
+     *         return 1;
+     *     } else {
+     *         return 1;
+     *     }
+     * }
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    VREG_INDEX_TO_ADDR(r2, r2)          @ r2<- &vBB
+    fldd    d0, [r2]                    @ d0<- vBB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vCC
+    fldd    d1, [r3]                    @ d1<- vCC
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    fcmped  d0, d1                      @ compare (vBB, vCC)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    mov     r0, #1                      @ r0<- 1 (default)
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    fmstat                              @ export status flags
+    mvnmi   r0, #0                      @ (less than) r1<- -1
+    moveq   r0, #0                      @ (equal) r1<- 0
+    bl      .L${opcode}_finish          @ argh
+
+%break
+.L${opcode}_finish:
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
diff --git a/vm/mterp/vfp/OP_CMPG_FLOAT.S b/vm/mterp/vfp/OP_CMPG_FLOAT.S
new file mode 100644
index 0000000..4dc142d
--- /dev/null
+++ b/vm/mterp/vfp/OP_CMPG_FLOAT.S
@@ -0,0 +1,43 @@
+%verify "executed"
+%verify "basic lt, gt, eq */
+%verify "left arg NaN"
+%verify "right arg NaN"
+    /*
+     * Compare two floating-point values.  Puts 0, 1, or -1 into the
+     * destination register based on the results of the comparison.
+     *
+     * int compare(x, y) {
+     *     if (x == y) {
+     *         return 0;
+     *     } else if (x < y) {
+     *         return -1;
+     *     } else if (x > y) {
+     *         return 1;
+     *     } else {
+     *         return 1;
+     *     }
+     * }
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    VREG_INDEX_TO_ADDR(r2, r2)          @ r2<- &vBB
+    flds    s0, [r2]                    @ s0<- vBB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vCC
+    flds    s1, [r3]                    @ s1<- vCC
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    fcmpes  s0, s1                      @ compare (vBB, vCC)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    mov     r0, #1                      @ r0<- 1 (default)
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    fmstat                              @ export status flags
+    mvnmi   r0, #0                      @ (less than) r1<- -1
+    moveq   r0, #0                      @ (equal) r1<- 0
+    bl      .L${opcode}_finish          @ argh
+
+%break
+.L${opcode}_finish:
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
diff --git a/vm/mterp/vfp/OP_CMPL_DOUBLE.S b/vm/mterp/vfp/OP_CMPL_DOUBLE.S
new file mode 100644
index 0000000..94d9102
--- /dev/null
+++ b/vm/mterp/vfp/OP_CMPL_DOUBLE.S
@@ -0,0 +1,43 @@
+%verify "executed"
+%verify "basic lt, gt, eq */
+%verify "left arg NaN"
+%verify "right arg NaN"
+    /*
+     * Compare two floating-point values.  Puts 0, 1, or -1 into the
+     * destination register based on the results of the comparison.
+     *
+     * int compare(x, y) {
+     *     if (x == y) {
+     *         return 0;
+     *     } else if (x > y) {
+     *         return 1;
+     *     } else if (x < y) {
+     *         return -1;
+     *     } else {
+     *         return -1;
+     *     }
+     * }
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    VREG_INDEX_TO_ADDR(r2, r2)          @ r2<- &vBB
+    fldd    d0, [r2]                    @ d0<- vBB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vCC
+    fldd    d1, [r3]                    @ d1<- vCC
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    fcmped  d0, d1                      @ compare (vBB, vCC)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    mvn     r0, #0                      @ r0<- -1 (default)
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    fmstat                              @ export status flags
+    movgt   r0, #1                      @ (greater than) r1<- 1
+    moveq   r0, #0                      @ (equal) r1<- 0
+    bl      .L${opcode}_finish          @ argh
+
+%break
+.L${opcode}_finish:
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
diff --git a/vm/mterp/vfp/OP_CMPL_FLOAT.S b/vm/mterp/vfp/OP_CMPL_FLOAT.S
new file mode 100644
index 0000000..4e72e13
--- /dev/null
+++ b/vm/mterp/vfp/OP_CMPL_FLOAT.S
@@ -0,0 +1,43 @@
+%verify "executed"
+%verify "basic lt, gt, eq */
+%verify "left arg NaN"
+%verify "right arg NaN"
+    /*
+     * Compare two floating-point values.  Puts 0, 1, or -1 into the
+     * destination register based on the results of the comparison.
+     *
+     * int compare(x, y) {
+     *     if (x == y) {
+     *         return 0;
+     *     } else if (x > y) {
+     *         return 1;
+     *     } else if (x < y) {
+     *         return -1;
+     *     } else {
+     *         return -1;
+     *     }
+     * }
+     */
+    /* op vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    and     r2, r0, #255                @ r2<- BB
+    VREG_INDEX_TO_ADDR(r2, r2)          @ r2<- &vBB
+    flds    s0, [r2]                    @ s0<- vBB
+    mov     r3, r0, lsr #8              @ r3<- CC
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vCC
+    flds    s1, [r3]                    @ s1<- vCC
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    fcmpes  s0, s1                      @ compare (vBB, vCC)
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    mvn     r0, #0                      @ r0<- -1 (default)
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    fmstat                              @ export status flags
+    movgt   r0, #1                      @ (greater than) r1<- 1
+    moveq   r0, #0                      @ (equal) r1<- 0
+    bl      .L${opcode}_finish          @ argh
+
+%break
+.L${opcode}_finish:
+    SET_VREG(r0, r9)                    @ vAA<- r0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
+
diff --git a/vm/mterp/vfp/OP_DIV_DOUBLE.S b/vm/mterp/vfp/OP_DIV_DOUBLE.S
new file mode 100644
index 0000000..346afab
--- /dev/null
+++ b/vm/mterp/vfp/OP_DIV_DOUBLE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/fbinopWide.S" {"instr":"fdivd   d2, d0, d1"}
diff --git a/vm/mterp/vfp/OP_DIV_DOUBLE_2ADDR.S b/vm/mterp/vfp/OP_DIV_DOUBLE_2ADDR.S
new file mode 100644
index 0000000..5b21994
--- /dev/null
+++ b/vm/mterp/vfp/OP_DIV_DOUBLE_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/fbinopWide2addr.S" {"instr":"fdivd   d2, d0, d1"}
diff --git a/vm/mterp/vfp/OP_DIV_FLOAT.S b/vm/mterp/vfp/OP_DIV_FLOAT.S
new file mode 100644
index 0000000..e8b6e16
--- /dev/null
+++ b/vm/mterp/vfp/OP_DIV_FLOAT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/fbinop.S" {"instr":"fdivs   s2, s0, s1"}
diff --git a/vm/mterp/vfp/OP_DIV_FLOAT_2ADDR.S b/vm/mterp/vfp/OP_DIV_FLOAT_2ADDR.S
new file mode 100644
index 0000000..edc611c
--- /dev/null
+++ b/vm/mterp/vfp/OP_DIV_FLOAT_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/fbinop2addr.S" {"instr":"fdivs   s2, s0, s1"}
diff --git a/vm/mterp/vfp/OP_DOUBLE_TO_FLOAT.S b/vm/mterp/vfp/OP_DOUBLE_TO_FLOAT.S
new file mode 100644
index 0000000..7987183
--- /dev/null
+++ b/vm/mterp/vfp/OP_DOUBLE_TO_FLOAT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/funopNarrower.S" {"instr":"fcvtsd  s0, d0"}
diff --git a/vm/mterp/vfp/OP_DOUBLE_TO_INT.S b/vm/mterp/vfp/OP_DOUBLE_TO_INT.S
new file mode 100644
index 0000000..8887858
--- /dev/null
+++ b/vm/mterp/vfp/OP_DOUBLE_TO_INT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/funopNarrower.S" {"instr":"ftosizd  s0, d0"}
diff --git a/vm/mterp/vfp/OP_FLOAT_TO_DOUBLE.S b/vm/mterp/vfp/OP_FLOAT_TO_DOUBLE.S
new file mode 100644
index 0000000..79f7eac
--- /dev/null
+++ b/vm/mterp/vfp/OP_FLOAT_TO_DOUBLE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/funopWider.S" {"instr":"fcvtds  d0, s0"}
diff --git a/vm/mterp/vfp/OP_FLOAT_TO_INT.S b/vm/mterp/vfp/OP_FLOAT_TO_INT.S
new file mode 100644
index 0000000..ef397d9
--- /dev/null
+++ b/vm/mterp/vfp/OP_FLOAT_TO_INT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/funop.S" {"instr":"ftosizs s1, s0"}
diff --git a/vm/mterp/vfp/OP_INT_TO_DOUBLE.S b/vm/mterp/vfp/OP_INT_TO_DOUBLE.S
new file mode 100644
index 0000000..02f278a
--- /dev/null
+++ b/vm/mterp/vfp/OP_INT_TO_DOUBLE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/funopWider.S" {"instr":"fsitod  d0, s0"}
diff --git a/vm/mterp/vfp/OP_INT_TO_FLOAT.S b/vm/mterp/vfp/OP_INT_TO_FLOAT.S
new file mode 100644
index 0000000..ec8e08e
--- /dev/null
+++ b/vm/mterp/vfp/OP_INT_TO_FLOAT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/funop.S" {"instr":"fsitos  s1, s0"}
diff --git a/vm/mterp/vfp/OP_MUL_DOUBLE.S b/vm/mterp/vfp/OP_MUL_DOUBLE.S
new file mode 100644
index 0000000..4257684
--- /dev/null
+++ b/vm/mterp/vfp/OP_MUL_DOUBLE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/fbinopWide.S" {"instr":"fmuld   d2, d0, d1"}
diff --git a/vm/mterp/vfp/OP_MUL_DOUBLE_2ADDR.S b/vm/mterp/vfp/OP_MUL_DOUBLE_2ADDR.S
new file mode 100644
index 0000000..ddcc328
--- /dev/null
+++ b/vm/mterp/vfp/OP_MUL_DOUBLE_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/fbinopWide2addr.S" {"instr":"fmuld   d2, d0, d1"}
diff --git a/vm/mterp/vfp/OP_MUL_FLOAT.S b/vm/mterp/vfp/OP_MUL_FLOAT.S
new file mode 100644
index 0000000..5de44c5
--- /dev/null
+++ b/vm/mterp/vfp/OP_MUL_FLOAT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/fbinop.S" {"instr":"fmuls   s2, s0, s1"}
diff --git a/vm/mterp/vfp/OP_MUL_FLOAT_2ADDR.S b/vm/mterp/vfp/OP_MUL_FLOAT_2ADDR.S
new file mode 100644
index 0000000..ea925f3
--- /dev/null
+++ b/vm/mterp/vfp/OP_MUL_FLOAT_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/fbinop2addr.S" {"instr":"fmuls   s2, s0, s1"}
diff --git a/vm/mterp/vfp/OP_SUB_DOUBLE.S b/vm/mterp/vfp/OP_SUB_DOUBLE.S
new file mode 100644
index 0000000..5ec55a0
--- /dev/null
+++ b/vm/mterp/vfp/OP_SUB_DOUBLE.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/fbinopWide.S" {"instr":"fsubd   d2, d0, d1"}
diff --git a/vm/mterp/vfp/OP_SUB_DOUBLE_2ADDR.S b/vm/mterp/vfp/OP_SUB_DOUBLE_2ADDR.S
new file mode 100644
index 0000000..13d47d9
--- /dev/null
+++ b/vm/mterp/vfp/OP_SUB_DOUBLE_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/fbinopWide2addr.S" {"instr":"fsubd   d2, d0, d1"}
diff --git a/vm/mterp/vfp/OP_SUB_FLOAT.S b/vm/mterp/vfp/OP_SUB_FLOAT.S
new file mode 100644
index 0000000..7967132
--- /dev/null
+++ b/vm/mterp/vfp/OP_SUB_FLOAT.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/fbinop.S" {"instr":"fsubs   s2, s0, s1"}
diff --git a/vm/mterp/vfp/OP_SUB_FLOAT_2ADDR.S b/vm/mterp/vfp/OP_SUB_FLOAT_2ADDR.S
new file mode 100644
index 0000000..1fed95b
--- /dev/null
+++ b/vm/mterp/vfp/OP_SUB_FLOAT_2ADDR.S
@@ -0,0 +1,2 @@
+%verify "executed"
+%include "vfp/fbinop2addr.S" {"instr":"fsubs   s2, s0, s1"}
diff --git a/vm/mterp/vfp/fbinop.S b/vm/mterp/vfp/fbinop.S
new file mode 100644
index 0000000..0613231
--- /dev/null
+++ b/vm/mterp/vfp/fbinop.S
@@ -0,0 +1,23 @@
+    /*
+     * Generic 32-bit floating-point operation.  Provide an "instr" line that
+     * specifies an instruction that performs "s2 = s0 op s1".  Because we
+     * use the "softfp" ABI, this must be an instruction, not a function call.
+     *
+     * For: add-float, sub-float, mul-float, div-float
+     */
+    /* floatop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vCC
+    flds    s1, [r3]                    @ s1<- vCC
+    and     r2, r0, #255                @ r2<- BB
+    VREG_INDEX_TO_ADDR(r2, r2)          @ r2<- &vBB
+    flds    s0, [r2]                    @ s0<- vBB
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    $instr                              @ s2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vAA
+    fsts    s2, [r9]                    @ vAA<- s2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
diff --git a/vm/mterp/vfp/fbinop2addr.S b/vm/mterp/vfp/fbinop2addr.S
new file mode 100644
index 0000000..3593c33
--- /dev/null
+++ b/vm/mterp/vfp/fbinop2addr.S
@@ -0,0 +1,21 @@
+    /*
+     * Generic 32-bit floating point "/2addr" binary operation.  Provide
+     * an "instr" line that specifies an instruction that performs
+     * "s2 = s0 op s1".
+     *
+     * For: add-float/2addr, sub-float/2addr, mul-float/2addr, div-float/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    flds    s1, [r3]                    @ s1<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    and     r9, r9, #15                 @ r9<- A
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    flds    s0, [r9]                    @ s0<- vA
+
+    $instr                              @ s2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    fsts    s2, [r9]                    @ vAA<- s2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
diff --git a/vm/mterp/vfp/fbinopWide.S b/vm/mterp/vfp/fbinopWide.S
new file mode 100644
index 0000000..997814f
--- /dev/null
+++ b/vm/mterp/vfp/fbinopWide.S
@@ -0,0 +1,23 @@
+    /*
+     * Generic 64-bit double-precision floating point binary operation.
+     * Provide an "instr" line that specifies an instruction that performs
+     * "d2 = d0 op d1".
+     *
+     * for: add-double, sub-double, mul-double, div-double
+     */
+    /* doubleop vAA, vBB, vCC */
+    FETCH(r0, 1)                        @ r0<- CCBB
+    mov     r9, rINST, lsr #8           @ r9<- AA
+    mov     r3, r0, lsr #8              @ r3<- CC
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vCC
+    fldd    d1, [r3]                    @ d1<- vCC
+    and     r2, r0, #255                @ r2<- BB
+    VREG_INDEX_TO_ADDR(r2, r2)          @ r2<- &vBB
+    fldd    d0, [r2]                    @ d0<- vBB
+
+    FETCH_ADVANCE_INST(2)               @ advance rPC, load rINST
+    $instr                              @ s2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vAA
+    fstd    d2, [r9]                    @ vAA<- d2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
diff --git a/vm/mterp/vfp/fbinopWide2addr.S b/vm/mterp/vfp/fbinopWide2addr.S
new file mode 100644
index 0000000..5a9792f
--- /dev/null
+++ b/vm/mterp/vfp/fbinopWide2addr.S
@@ -0,0 +1,22 @@
+    /*
+     * Generic 64-bit floating point "/2addr" binary operation.  Provide
+     * an "instr" line that specifies an instruction that performs
+     * "d2 = d0 op d1".
+     *
+     * For: add-double/2addr, sub-double/2addr, mul-double/2addr,
+     *      div-double/2addr
+     */
+    /* binop/2addr vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    fldd    d1, [r3]                    @ d1<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    and     r9, r9, #15                 @ r9<- A
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    fldd    d0, [r9]                    @ d0<- vA
+
+    $instr                              @ d2<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    fstd    d2, [r9]                    @ vAA<- d2
+    GOTO_OPCODE(ip)                     @ jump to next instruction
diff --git a/vm/mterp/vfp/funop.S b/vm/mterp/vfp/funop.S
new file mode 100644
index 0000000..d88457b
--- /dev/null
+++ b/vm/mterp/vfp/funop.S
@@ -0,0 +1,18 @@
+    /*
+     * Generic 32-bit unary floating-point operation.  Provide an "instr"
+     * line that specifies an instruction that performs "s1 = op s0".
+     *
+     * for: int-to-float, float-to-int
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    flds    s0, [r3]                    @ s0<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    and     r9, r9, #15                 @ r9<- A
+    $instr                              @ s1<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    fsts    s1, [r9]                    @ vA<- s1
+    GOTO_OPCODE(ip)                     @ jump to next instruction
diff --git a/vm/mterp/vfp/funopNarrower.S b/vm/mterp/vfp/funopNarrower.S
new file mode 100644
index 0000000..1d0e54a
--- /dev/null
+++ b/vm/mterp/vfp/funopNarrower.S
@@ -0,0 +1,18 @@
+    /*
+     * Generic 64bit-to-32bit unary floating point operation.  Provide an
+     * "instr" line that specifies an instruction that performs "s0 = op d0".
+     *
+     * For: double-to-int, double-to-float
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    fldd    d0, [r3]                    @ d0<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    and     r9, r9, #15                 @ r9<- A
+    $instr                              @ s0<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    fsts    s0, [r9]                    @ vA<- s0
+    GOTO_OPCODE(ip)                     @ jump to next instruction
diff --git a/vm/mterp/vfp/funopWider.S b/vm/mterp/vfp/funopWider.S
new file mode 100644
index 0000000..985a35c
--- /dev/null
+++ b/vm/mterp/vfp/funopWider.S
@@ -0,0 +1,18 @@
+    /*
+     * Generic 32bit-to-64bit floating point unary operation.  Provide an
+     * "instr" line that specifies an instruction that performs "d0 = op s0".
+     *
+     * For: int-to-double, float-to-double
+     */
+    /* unop vA, vB */
+    mov     r3, rINST, lsr #12          @ r3<- B
+    VREG_INDEX_TO_ADDR(r3, r3)          @ r3<- &vB
+    flds    s0, [r3]                    @ s0<- vB
+    mov     r9, rINST, lsr #8           @ r9<- A+
+    FETCH_ADVANCE_INST(1)               @ advance rPC, load rINST
+    and     r9, r9, #15                 @ r9<- A
+    $instr                              @ d0<- op
+    GET_INST_OPCODE(ip)                 @ extract opcode from rINST
+    VREG_INDEX_TO_ADDR(r9, r9)          @ r9<- &vA
+    fstd    d0, [r9]                    @ vA<- d0
+    GOTO_OPCODE(ip)                     @ jump to next instruction