blob: 10bc197f172ab0e040f0d4aa6bd4cce0ccff9ea8 [file] [log] [blame]
/*
* Copyright (C) 2012 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.
*/
/*! \file LowerInvoke.cpp
\brief This file lowers the following bytecodes: INVOKE_XXX
*/
#include "libdex/DexOpcodes.h"
#include "libdex/DexFile.h"
#include "mterp/Mterp.h"
#include "Lower.h"
#include "NcgAot.h"
#include "enc_wrapper.h"
char* streamMisPred = NULL;
/* according to callee, decide the ArgsDoneType*/
ArgsDoneType convertCalleeToType(const Method* calleeMethod) {
if(calleeMethod == NULL)
return ArgsDone_Full;
if(dvmIsNativeMethod(calleeMethod))
return ArgsDone_Native;
return ArgsDone_Normal;
}
int common_invokeMethodRange(ArgsDoneType);
int common_invokeMethodNoRange(ArgsDoneType);
void gen_predicted_chain(bool isRange, u2 tmp, int IMMC, bool isInterface, int inputReg);
//inputs to common_invokeMethodRange: %ecx
// common_errNoSuchMethod: %edx
#define P_GPR_1 PhysicalReg_ESI
#define P_GPR_2 PhysicalReg_EBX
#define P_GPR_3 PhysicalReg_ECX
#define P_SCRATCH_1 PhysicalReg_EDX
#define PP_GPR_1 PhysicalReg_EBX
#define PP_GPR_2 PhysicalReg_ESI
#define PP_GPR_3 PhysicalReg_EAX
#define PP_GPR_4 PhysicalReg_EDX
#ifdef WITH_JIT_INLINING
/*
* The function here takes care the
* branch over if prediction is correct and the misprediction target for misPredBranchOver.
*/
static void genLandingPadForMispredictedCallee(MIR* mir) {
BasicBlock *fallThrough = traceCurrentBB->fallThrough;
/* Bypass the move-result block if there is one */
if (fallThrough->firstMIRInsn) {
assert(fallThrough->firstMIRInsn->OptimizationFlags & MIR_INLINED_PRED);
fallThrough = fallThrough->fallThrough;
}
/* Generate a branch over if the predicted inlining is correct */
jumpToBasicBlock(stream, fallThrough->id);
/* Hook up the target to the verification branch */
int relativeNCG = stream - streamMisPred;
unsigned instSize = encoder_get_inst_size(streamMisPred);
relativeNCG -= instSize; //size of the instruction
updateJumpInst(streamMisPred, OpndSize_8, relativeNCG);
}
#endif
//! LOWER bytecode INVOKE_VIRTUAL without usage of helper function
//!
int common_invoke_virtual_nohelper(bool isRange, u2 tmp, u2 vD) {
#ifdef WITH_JIT_INLINING
/*
* If the invoke has non-null misPredBranchOver, we need to generate
* the non-inlined version of the invoke here to handle the
* mispredicted case.
*/
if (traceCurrentMIR->meta.callsiteInfo->misPredBranchOver) {
genLandingPadForMispredictedCallee(); //cUnit, mir, bb, labelList);
}
#endif
scratchRegs[2] = PhysicalReg_Null; scratchRegs[3] = PhysicalReg_Null;
export_pc();
constVREndOfBB();
beforeCall("exception"); //dump GG, GL VRs
get_virtual_reg(vD, OpndSize_32, 5, false);
simpleNullCheck(5, false, vD);
#ifndef PREDICTED_CHAINING
move_mem_to_reg(OpndSize_32, offObject_clazz, 5, false, 6, false); //clazz of "this"
move_mem_to_reg(OpndSize_32, offClassObject_vtable, 6, false, 7, false); //vtable
/* method is already resolved in trace-based JIT */
int methodIndex =
currentMethod->clazz->pDvmDex->pResMethods[tmp]->methodIndex;
move_mem_to_reg(OpndSize_32, methodIndex*4, 7, false, PhysicalReg_ECX, true);
if(isRange) {
common_invokeMethodRange(ArgsDone_Full);
}
else {
common_invokeMethodNoRange(ArgsDone_Full);
}
#else
int methodIndex =
currentMethod->clazz->pDvmDex->pResMethods[tmp]->methodIndex;
gen_predicted_chain(isRange, tmp, methodIndex*4, false, 5/*tmp5*/);
#endif
///////////////////////////////////
return 0;
}
//! wrapper to call either common_invoke_virtual_helper or common_invoke_virtual_nohelper
//!
int common_invoke_virtual(bool isRange, u2 tmp, u2 vD) {
return common_invoke_virtual_nohelper(isRange, tmp, vD);
}
#undef P_GPR_1
#undef P_GPR_2
#undef P_GPR_3
#undef P_SCRATCH_1
#undef PP_GPR_1
#undef PP_GPR_2
#undef PP_GPR_3
#undef PP_GPR_4
#define P_GPR_1 PhysicalReg_ESI
#define P_GPR_2 PhysicalReg_EBX
#define P_GPR_3 PhysicalReg_EDX
#define PP_GPR_1 PhysicalReg_EBX
#define PP_GPR_2 PhysicalReg_ESI
#define PP_GPR_3 PhysicalReg_EAX
#define PP_GPR_4 PhysicalReg_EDX
//! common section to lower INVOKE_SUPER
//! It will use helper function if the switch is on
int common_invoke_super(bool isRange, u2 tmp) {
export_pc();
constVREndOfBB();
beforeCall("exception"); //dump GG, GL VRs
///////////////////////
scratchRegs[2] = PhysicalReg_Null; scratchRegs[3] = PhysicalReg_Null;
/* method is already resolved in trace-based JIT */
int mIndex = currentMethod->clazz->pDvmDex->pResMethods[tmp]->methodIndex;
const Method *calleeMethod =
currentMethod->clazz->super->vtable[mIndex];
move_imm_to_reg(OpndSize_32, (int) calleeMethod, PhysicalReg_ECX, true);
if(isRange) {
common_invokeMethodRange(convertCalleeToType(calleeMethod));
}
else {
common_invokeMethodNoRange(convertCalleeToType(calleeMethod));
}
///////////////////////////////
return 0;
}
#undef PP_GPR_1
#undef PP_GPR_2
#undef PP_GPR_3
#undef PP_GPR_4
//! helper function to handle no such method error
//!
int invoke_super_nsm() {
insertLabel(".invoke_super_nsm", false);
//NOTE: it seems that the name in %edx is not used in common_errNoSuchMethod
move_mem_to_reg(OpndSize_32, offMethod_name, PhysicalReg_EAX, true, PhysicalReg_EDX, true); //method name
unconditional_jump("common_errNoSuchMethod", false);
return 0;
}
#undef P_GPR_1
#undef P_GPR_2
#undef P_GPR_3
#define P_GPR_1 PhysicalReg_EBX
#define P_GPR_2 PhysicalReg_ESI
#define P_GPR_3 PhysicalReg_ECX
#define PP_GPR_1 PhysicalReg_EBX
#define PP_GPR_2 PhysicalReg_ESI
#define PP_GPR_3 PhysicalReg_EAX
#define PP_GPR_4 PhysicalReg_EDX
//! common section to lower INVOKE_DIRECT
//! It will use helper function if the switch is on
int common_invoke_direct(bool isRange, u2 tmp, u2 vD) {
//%ecx can be used as scratch when calling export_pc, get_res_methods and resolve_method
export_pc();
constVREndOfBB();
beforeCall("exception"); //dump GG, GL VRs
////////////////////////////////////
get_virtual_reg(vD, OpndSize_32, 5, false);
simpleNullCheck(5, false, vD);
/* method is already resolved in trace-based JIT */
const Method *calleeMethod =
currentMethod->clazz->pDvmDex->pResMethods[tmp];
move_imm_to_reg(OpndSize_32, (int) calleeMethod, PhysicalReg_ECX, true);
//%ecx passed to common_invokeMethod...
if(isRange) {
common_invokeMethodRange(convertCalleeToType(calleeMethod));
}
else {
common_invokeMethodNoRange(convertCalleeToType(calleeMethod));
}
////////////////////////////
return 0;
}
#undef P_GPR_1
#undef P_GPR_2
#undef P_GPR_3
#undef PP_GPR_1
#undef PP_GPR_2
#undef PP_GPR_3
#undef PP_GPR_4
#define P_GPR_1 PhysicalReg_EBX
#define P_GPR_3 PhysicalReg_ECX
#define PP_GPR_1 PhysicalReg_EBX
#define PP_GPR_2 PhysicalReg_ESI
#define PP_GPR_3 PhysicalReg_EAX
#define PP_GPR_4 PhysicalReg_EDX
//! common section to lower INVOKE_STATIC
//! It will use helper function if the switch is on
int common_invoke_static(bool isRange, u2 tmp) {
//%ecx can be used as scratch when calling export_pc, get_res_methods and resolve_method
export_pc();
constVREndOfBB();
beforeCall("exception"); //dump GG, GL VRs
////////////////////////////
/* method is already resolved in trace-based JIT */
const Method *calleeMethod =
currentMethod->clazz->pDvmDex->pResMethods[tmp];
move_imm_to_reg(OpndSize_32, (int) calleeMethod, PhysicalReg_ECX, true);
//%ecx passed to common_invokeMethod...
if(isRange) {
common_invokeMethodRange(convertCalleeToType(calleeMethod));
}
else {
common_invokeMethodNoRange(convertCalleeToType(calleeMethod));
}
////////////////////////
return 0;
}
#undef P_GPR_1
#undef PP_GPR_1
#undef PP_GPR_2
#undef PP_GPR_3
#undef PP_GPR_4
#define P_GPR_1 PhysicalReg_EBX
#define P_GPR_2 PhysicalReg_EAX //scratch
#define P_GPR_3 PhysicalReg_ECX
#define P_SCRATCH_1 PhysicalReg_ESI //clazz of object
#define PP_GPR_1 PhysicalReg_EBX
#define PP_GPR_2 PhysicalReg_ESI
#define PP_GPR_3 PhysicalReg_EAX
#define PP_GPR_4 PhysicalReg_EDX
//! common section to lower INVOKE_INTERFACE
//! It will use helper function if the switch is on
int common_invoke_interface(bool isRange, u2 tmp, u2 vD) {
#ifdef WITH_JIT_INLINING
/*
* If the invoke has non-null misPredBranchOver, we need to generate
* the non-inlined version of the invoke here to handle the
* mispredicted case.
*/
if (traceCurrentMIR->meta.callsiteInfo->misPredBranchOver) {
genLandingPadForMispredictedCallee(); //cUnit, mir, bb, labelList);
}
#endif
export_pc(); //use %edx
constVREndOfBB();
beforeCall("exception"); //dump GG, GL VRs
///////////////////////
scratchRegs[2] = PhysicalReg_Null; scratchRegs[3] = PhysicalReg_Null;
get_virtual_reg(vD, OpndSize_32, 1, false);
simpleNullCheck(1, false, vD);
#ifndef PREDICTED_CHAINING
load_effective_addr(-16, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
move_imm_to_mem(OpndSize_32, tmp, 4, PhysicalReg_ESP, true);
/* for trace-based JIT, pDvmDex is a constant at JIT time
4th argument to dvmFindInterfaceMethodInCache at -4(%esp) */
move_imm_to_mem(OpndSize_32, (int) currentMethod->clazz->pDvmDex, 12, PhysicalReg_ESP, true);
move_mem_to_reg(OpndSize_32, offObject_clazz, 1, false, 5, false);
/* for trace-based JIT, method is a constant at JIT time
3rd argument to dvmFindInterfaceMethodInCache at 8(%esp) */
move_imm_to_mem(OpndSize_32, (int) currentMethod, 8, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, 5, false, 0, PhysicalReg_ESP, true);
scratchRegs[0] = PhysicalReg_SCRATCH_3; scratchRegs[1] = PhysicalReg_Null;
call_dvmFindInterfaceMethodInCache();
load_effective_addr(16, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
compare_imm_reg(OpndSize_32, 0, PhysicalReg_EAX, true);
conditional_jump_global_API(Condition_E, "common_exceptionThrown", false);
move_reg_to_reg(OpndSize_32, PhysicalReg_EAX, true, PhysicalReg_ECX, true);
if(isRange) {
common_invokeMethodRange(ArgsDone_Full);
}
else {
common_invokeMethodNoRange(ArgsDone_Full);
}
#else
gen_predicted_chain(isRange, tmp, -1, true /*interface*/, 1/*tmp1*/);
#endif
///////////////////////
return 0;
}
#undef PP_GPR_1
#undef PP_GPR_2
#undef PP_GPR_3
#undef PP_GPR_4
#undef P_GPR_1
#undef P_GPR_2
#undef P_GPR_3
#undef P_SCRATCH_1
//! lower bytecode INVOKE_VIRTUAL by calling common_invoke_virtual
//!
int op_invoke_virtual() {
#ifdef WITH_JIT_INLINING
/* An invoke with the MIR_INLINED is effectively a no-op */
if (traceCurrentMIR->OptimizationFlags & MIR_INLINED)
return false;
#endif
//B|A|op CCCC G|F|E|D
//D: the first argument, which is the "this" pointer
//B: argument count
//D,E,F,G,A: arguments
u2 vD = FETCH(2) & 0xf;
u2 tmp = FETCH(1); //method index
int retval = common_invoke_virtual(false/*not range*/, tmp, vD);
#if defined(ENABLE_TRACING) && !defined(TRACING_OPTION2)
insertMapWorklist(offsetPC+3, stream - streamMethodStart, 1); //check when helper switch is on
#endif
rPC += 3;
return retval;
}
//! lower bytecode INVOKE_SUPER by calling common_invoke_super
//!
int op_invoke_super() {
#ifdef WITH_JIT_INLINING
/* An invoke with the MIR_INLINED is effectively a no-op */
if (traceCurrentMIR->OptimizationFlags & MIR_INLINED)
return false;
#endif
//B|A|op CCCC G|F|E|D
//D: the first argument
//B: argument count
//D,E,F,G,A: arguments
u2 tmp = FETCH(1); //method index
int retval = common_invoke_super(false/*not range*/, tmp);
#if defined(ENABLE_TRACING) && !defined(TRACING_OPTION2)
insertMapWorklist(offsetPC+3, stream - streamMethodStart, 1); //check when helper switch is on
#endif
rPC += 3;
return retval;
}
//! lower bytecode INVOKE_DIRECT by calling common_invoke_direct
//!
int op_invoke_direct() {
#ifdef WITH_JIT_INLINING
/* An invoke with the MIR_INLINED is effectively a no-op */
if (traceCurrentMIR->OptimizationFlags & MIR_INLINED)
return false;
#endif
//B|A|op CCCC G|F|E|D
//D: the first argument, which is the "this" pointer
//B: argument count
//D,E,F,G,A: arguments
u2 vD = FETCH(2) & 0xf;
u2 tmp = FETCH(1); //method index
int retval = common_invoke_direct(false/*not range*/, tmp, vD);
#if defined(ENABLE_TRACING) && !defined(TRACING_OPTION2)
insertMapWorklist(offsetPC+3, stream - streamMethodStart, 1); //check when helper switch is on
#endif
rPC += 3;
return retval;
}
//! lower bytecode INVOKE_STATIC by calling common_invoke_static
//!
int op_invoke_static() {
#ifdef WITH_JIT_INLINING
/* An invoke with the MIR_INLINED is effectively a no-op */
if (traceCurrentMIR->OptimizationFlags & MIR_INLINED)
return false;
#endif
//B|A|op CCCC G|F|E|D
//D: the first argument
//B: argument count
//D,E,F,G,A: arguments
u2 tmp = FETCH(1); //method index
int retval = common_invoke_static(false/*not range*/, tmp);
#if defined(ENABLE_TRACING) && !defined(TRACING_OPTION2)
insertMapWorklist(offsetPC+3, stream - streamMethodStart, 1); //check when helper switch is on
#endif
rPC += 3;
return retval;
}
//! lower bytecode INVOKE_INTERFACE by calling common_invoke_interface
//!
int op_invoke_interface() {
#ifdef WITH_JIT_INLINING
/* An invoke with the MIR_INLINED is effectively a no-op */
if (traceCurrentMIR->OptimizationFlags & MIR_INLINED)
return false;
#endif
//B|A|op CCCC G|F|E|D
//D: the first argument, which is the "this" pointer
//B: argument count
//D,E,F,G,A: arguments
u2 vD = FETCH(2) & 0xf;
u2 tmp = FETCH(1); //method index
int retval = common_invoke_interface(false/*not range*/, tmp, vD);
#if defined(ENABLE_TRACING) && !defined(TRACING_OPTION2)
insertMapWorklist(offsetPC+3, stream - streamMethodStart, 1); //check when helper switch is on
#endif
rPC += 3;
return retval;
}
//! lower bytecode INVOKE_VIRTUAL_RANGE by calling common_invoke_virtual
//!
int op_invoke_virtual_range() {
#ifdef WITH_JIT_INLINING
/* An invoke with the MIR_INLINED is effectively a no-op */
if (traceCurrentMIR->OptimizationFlags & MIR_INLINED)
return false;
#endif
//AA|op BBBB CCCC
//CCCC: the first argument, which is the "this" pointer
//AA: argument count
u2 tmp = FETCH(1); //BBBB, method index
u2 vD = FETCH(2); //the first argument
int retval = common_invoke_virtual(true/*range*/, tmp, vD);
#if defined(ENABLE_TRACING) && !defined(TRACING_OPTION2)
insertMapWorklist(offsetPC+3, stream - streamMethodStart, 1); //check when helper switch is on
#endif
rPC += 3;
return retval;
}
//! lower bytecode INVOKE_SUPER_RANGE by calling common_invoke_super
//!
int op_invoke_super_range() {
#ifdef WITH_JIT_INLINING
/* An invoke with the MIR_INLINED is effectively a no-op */
if (traceCurrentMIR->OptimizationFlags & MIR_INLINED)
return false;
#endif
u2 tmp = FETCH(1); //BBBB, method index
int retval = common_invoke_super(true/*range*/, tmp);
#if defined(ENABLE_TRACING) && !defined(TRACING_OPTION2)
insertMapWorklist(offsetPC+3, stream - streamMethodStart, 1); //check when helper switch is on
#endif
rPC += 3;
return retval;
}
//! lower bytecode INVOKE_DIRECT_RANGE by calling common_invoke_direct
//!
int op_invoke_direct_range() {
#ifdef WITH_JIT_INLINING
/* An invoke with the MIR_INLINED is effectively a no-op */
if (traceCurrentMIR->OptimizationFlags & MIR_INLINED)
return false;
#endif
u2 tmp = FETCH(1); //BBBB, method index
u2 vD = FETCH(2); //the first argument
int retval = common_invoke_direct(true/*range*/, tmp, vD);
#if defined(ENABLE_TRACING) && !defined(TRACING_OPTION2)
insertMapWorklist(offsetPC+3, stream - streamMethodStart, 1); //check when helper switch is on
#endif
rPC += 3;
return retval;
}
//! lower bytecode INVOKE_STATIC_RANGE by calling common_invoke_static
//!
int op_invoke_static_range() {
#ifdef WITH_JIT_INLINING
/* An invoke with the MIR_INLINED is effectively a no-op */
if (traceCurrentMIR->OptimizationFlags & MIR_INLINED)
return false;
#endif
u2 tmp = FETCH(1); //BBBB, method index
int retval = common_invoke_static(true/*range*/, tmp);
#if defined(ENABLE_TRACING) && !defined(TRACING_OPTION2)
insertMapWorklist(offsetPC+3, stream - streamMethodStart, 1); //check when helper switch is on
#endif
rPC += 3;
return retval;
}
//! lower bytecode INVOKE_INTERFACE_RANGE by calling common_invoke_interface
//!
int op_invoke_interface_range() {
#ifdef WITH_JIT_INLINING
/* An invoke with the MIR_INLINED is effectively a no-op */
if (traceCurrentMIR->OptimizationFlags & MIR_INLINED)
return false;
#endif
u2 tmp = FETCH(1); //BBBB, method index
u2 vD = FETCH(2); //the first argument
int retval = common_invoke_interface(true/*range*/, tmp, vD);
#if defined(ENABLE_TRACING) && !defined(TRACING_OPTION2)
insertMapWorklist(offsetPC+3, stream - streamMethodStart, 1); //check when helper switch is on
#endif
rPC += 3;
return retval;
}
//used %ecx, %edi, %esp %ebp
#define P_GPR_1 PhysicalReg_EBX
#define P_SCRATCH_1 PhysicalReg_ESI
#define P_SCRATCH_2 PhysicalReg_EAX
#define P_SCRATCH_3 PhysicalReg_EDX
#define P_SCRATCH_4 PhysicalReg_ESI
#define P_SCRATCH_5 PhysicalReg_EAX
//! pass the arguments for invoking method without range
//!
int common_invokeMethodNoRange_noJmp() {
u2 count = INST_B(inst);
u2 vD = FETCH(2) & 0xf;
u2 vE = (FETCH(2) >> 4) & 0xf;
u2 vF = (FETCH(2) >> 8) & 0xf;
u2 vG = (FETCH(2) >> 12) & 0xf;
u2 vA = INST_A(inst); //5th argument
int offsetFromSaveArea = -4;
if(count == 5) {
get_virtual_reg(vA, OpndSize_32, 22, false);
move_reg_to_mem(OpndSize_32, 22, false, offsetFromSaveArea-sizeofStackSaveArea, PhysicalReg_FP, true);
offsetFromSaveArea -= 4;
}
if(count >= 4) {
get_virtual_reg(vG, OpndSize_32, 23, false);
move_reg_to_mem(OpndSize_32, 23, false, offsetFromSaveArea-sizeofStackSaveArea, PhysicalReg_FP, true);
offsetFromSaveArea -= 4;
}
if(count >= 3) {
get_virtual_reg(vF, OpndSize_32, 24, false);
move_reg_to_mem(OpndSize_32, 24, false, offsetFromSaveArea-sizeofStackSaveArea, PhysicalReg_FP, true);
offsetFromSaveArea -= 4;
}
if(count >= 2) {
get_virtual_reg(vE, OpndSize_32, 25, false);
move_reg_to_mem(OpndSize_32, 25, false, offsetFromSaveArea-sizeofStackSaveArea, PhysicalReg_FP, true);
offsetFromSaveArea -= 4;
}
if(count >= 1) {
get_virtual_reg(vD, OpndSize_32, 26, false);
move_reg_to_mem(OpndSize_32, 26, false, offsetFromSaveArea-sizeofStackSaveArea, PhysicalReg_FP, true);
}
return 0;
}
int common_invokeMethod_Jmp(ArgsDoneType form) {
nextVersionOfHardReg(PhysicalReg_EDX, 1);
move_imm_to_reg(OpndSize_32, (int)rPC, PhysicalReg_EDX, true);
//arguments needed in ArgsDone:
// start of HotChainingCell for next bytecode: -4(%esp)
// start of InvokeSingletonChainingCell for callee: -8(%esp)
load_effective_addr(-8, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
insertChainingWorklist(traceCurrentBB->fallThrough->id, stream);
move_chain_to_mem(OpndSize_32, traceCurrentBB->fallThrough->id, 4, PhysicalReg_ESP, true);
// for honeycomb: JNI call doesn't need a chaining cell, so the taken branch is null
if(traceCurrentBB->taken)
insertChainingWorklist(traceCurrentBB->taken->id, stream);
int takenId = traceCurrentBB->taken ? traceCurrentBB->taken->id : 0;
move_chain_to_mem(OpndSize_32, takenId, 0, PhysicalReg_ESP, true);
if(form == ArgsDone_Full)
unconditional_jump_global_API(".invokeArgsDone_jit", false);
else if(form == ArgsDone_Native)
unconditional_jump_global_API(".invokeArgsDone_native", false);
else
unconditional_jump_global_API(".invokeArgsDone_normal", false);
return 0;
}
int common_invokeMethodNoRange(ArgsDoneType form) {
common_invokeMethodNoRange_noJmp();
common_invokeMethod_Jmp(form);
return 0;
}
#undef P_GPR_1
#undef P_SCRATCH_1
#undef P_SCRATCH_2
#undef P_SCRATCH_3
#undef P_SCRATCH_4
#undef P_SCRATCH_5
//input: %ecx (method to call)
#define P_GPR_1 PhysicalReg_EBX
#define P_GPR_2 PhysicalReg_ESI
#define P_GPR_3 PhysicalReg_EDX //not used with P_SCRATCH_2
#define P_SCRATCH_1 PhysicalReg_EAX
#define P_SCRATCH_2 PhysicalReg_EDX
#define P_SCRATCH_3 PhysicalReg_EAX
#define P_SCRATCH_4 PhysicalReg_EDX
#define P_SCRATCH_5 PhysicalReg_EAX
#define P_SCRATCH_6 PhysicalReg_EDX
#define P_SCRATCH_7 PhysicalReg_EAX
#define P_SCRATCH_8 PhysicalReg_EDX
#define P_SCRATCH_9 PhysicalReg_EAX
#define P_SCRATCH_10 PhysicalReg_EDX
//! pass the arguments for invoking method with range
//! loop is unrolled when count <= 10
int common_invokeMethodRange_noJmp() {
u2 count = INST_AA(inst);
u2 vD = FETCH(2); //the first argument
savearea_from_fp(21, false);
//vD to rFP-4*count-20
//vD+1 to rFP-4*count-20+4 = rFP-20-4*(count-1)
if(count >= 1 && count <= 10) {
get_virtual_reg(vD, OpndSize_32, 22, false);
move_reg_to_mem(OpndSize_32, 22, false, -4*count, 21, false);
}
if(count >= 2 && count <= 10) {
get_virtual_reg(vD+1, OpndSize_32, 23, false);
move_reg_to_mem(OpndSize_32, 23, false, -4*(count-1), 21, false);
}
if(count >= 3 && count <= 10) {
get_virtual_reg(vD+2, OpndSize_32, 24, false);
move_reg_to_mem(OpndSize_32, 24, false, -4*(count-2), 21, false);
}
if(count >= 4 && count <= 10) {
get_virtual_reg(vD+3, OpndSize_32, 25, false);
move_reg_to_mem(OpndSize_32, 25, false, -4*(count-3), 21, false);
}
if(count >= 5 && count <= 10) {
get_virtual_reg(vD+4, OpndSize_32, 26, false);
move_reg_to_mem(OpndSize_32, 26, false, -4*(count-4), 21, false);
}
if(count >= 6 && count <= 10) {
get_virtual_reg(vD+5, OpndSize_32, 27, false);
move_reg_to_mem(OpndSize_32, 27, false, -4*(count-5), 21, false);
}
if(count >= 7 && count <= 10) {
get_virtual_reg(vD+6, OpndSize_32, 28, false);
move_reg_to_mem(OpndSize_32, 28, false, -4*(count-6), 21, false);
}
if(count >= 8 && count <= 10) {
get_virtual_reg(vD+7, OpndSize_32, 29, false);
move_reg_to_mem(OpndSize_32, 29, false, -4*(count-7), 21, false);
}
if(count >= 9 && count <= 10) {
get_virtual_reg(vD+8, OpndSize_32, 30, false);
move_reg_to_mem(OpndSize_32, 30, false, -4*(count-8), 21, false);
}
if(count == 10) {
get_virtual_reg(vD+9, OpndSize_32, 31, false);
move_reg_to_mem(OpndSize_32, 31, false, -4*(count-9), 21, false);
}
if(count > 10) {
//dump to memory first, should we set physicalReg to Null?
//this bytecodes uses a set of virtual registers (update getVirtualInfo)
//this is necessary to correctly insert transfer points
int k;
for(k = 0; k < count; k++) {
spillVirtualReg(vD+k, LowOpndRegType_gp, true); //will update refCount
}
load_effective_addr(4*vD, PhysicalReg_FP, true, 12, false);
alu_binary_imm_reg(OpndSize_32, sub_opc, 4*count, 21, false);
move_imm_to_reg(OpndSize_32, count, 13, false);
insertLabel(".invokeMethod_1", true); //if checkDup: will perform work from ShortWorklist
rememberState(1);
move_mem_to_reg(OpndSize_32, 0, 12, false, 14, false);
move_reg_to_mem(OpndSize_32, 14, false, 0, 21, false);
load_effective_addr(4, 12, false, 12, false);
alu_binary_imm_reg(OpndSize_32, sub_opc, 1, 13, false);
load_effective_addr(4, 21, false, 21, false);
transferToState(1);
conditional_jump(Condition_NE, ".invokeMethod_1", true); //backward branch
}
return 0;
}
int common_invokeMethodRange(ArgsDoneType form) {
common_invokeMethodRange_noJmp();
common_invokeMethod_Jmp(form);
return 0;
}
#undef P_GPR_1
#undef P_GPR_2
#undef P_GPR_3
#undef P_SCRATCH_1
#undef P_SCRATCH_2
#undef P_SCRATCH_3
#undef P_SCRATCH_4
#undef P_SCRATCH_5
#undef P_SCRATCH_6
#undef P_SCRATCH_7
#undef P_SCRATCH_8
#undef P_SCRATCH_9
#undef P_SCRATCH_10
#define P_GPR_1 PhysicalReg_EBX
#define P_GPR_3 PhysicalReg_ESI
#define P_SCRATCH_1 PhysicalReg_EAX
#define P_SCRATCH_2 PhysicalReg_EDX
#define P_SCRATCH_3 PhysicalReg_EAX
#define P_SCRATCH_4 PhysicalReg_EDX
#define P_SCRATCH_5 PhysicalReg_EAX
#define P_SCRATCH_6 PhysicalReg_EDX
//! spill a register to native stack
//! decrease %esp by 4, then store a register at 0(%esp)
int spill_reg(int reg, bool isPhysical) {
load_effective_addr(-4, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, reg, isPhysical, 0, PhysicalReg_ESP, true);
return 0;
}
//! get a register from native stack
//! load a register from 0(%esp), then increase %esp by 4
int unspill_reg(int reg, bool isPhysical) {
move_mem_to_reg(OpndSize_32, 0, PhysicalReg_ESP, true, reg, isPhysical);
load_effective_addr(4, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
return 0;
}
void generate_invokeNative(bool generateForNcg); //forward declaration
void generate_stackOverflow(); //forward declaration
//! common code to invoke a method after all arguments are handled
//!
//takes one argument to generate code
// for invokeNativeSingle (form == ArgsDone_Native)
// or invokeNonNativeSingle (form == ArgsDone_Normal) when WITH_JIT is true
// to dynamically determine which one to choose (form == ArgsDone_Full)
/* common_invokeArgsDone is called at NCG time and
at execution time during relocation
generate invokeArgsDone for NCG if isJitFull is false && form == Full */
int common_invokeArgsDone(ArgsDoneType form, bool isJitFull) {
bool generateForNcg = false;
if(form == ArgsDone_Full) {
if(isJitFull)
insertLabel(".invokeArgsDone_jit", false);
else {
insertLabel(".invokeArgsDone", false);
generateForNcg = true;
}
}
else if(form == ArgsDone_Normal)
insertLabel(".invokeArgsDone_normal", false);
else if(form == ArgsDone_Native)
insertLabel(".invokeArgsDone_native", false);
//%ecx: methodToCall
movez_mem_to_reg(OpndSize_16, offMethod_registersSize, PhysicalReg_ECX, true, P_SCRATCH_1, true); //regSize
scratchRegs[0] = PhysicalReg_EBX; scratchRegs[1] = PhysicalReg_ESI;
scratchRegs[2] = PhysicalReg_EDX; scratchRegs[3] = PhysicalReg_Null;
savearea_from_fp(P_GPR_3, true);
alu_binary_imm_reg(OpndSize_32, shl_opc, 2, P_SCRATCH_1, true);
alu_binary_reg_reg(OpndSize_32, sub_opc, P_SCRATCH_1, true, P_GPR_3, true);
//update newSaveArea->savedPc, here P_GPR_3 is new FP
move_reg_to_mem(OpndSize_32, PhysicalReg_EDX, true, offStackSaveArea_savedPc-sizeofStackSaveArea, P_GPR_3, true);
movez_mem_to_reg(OpndSize_16, offMethod_outsSize, PhysicalReg_ECX, true, P_SCRATCH_2, true); //outsSize
move_reg_to_reg(OpndSize_32, P_GPR_3, true, P_GPR_1, true); //new FP
alu_binary_imm_reg(OpndSize_32, sub_opc, sizeofStackSaveArea, P_GPR_3, true);
alu_binary_imm_reg(OpndSize_32, shl_opc, 2, P_SCRATCH_2, true);
alu_binary_reg_reg(OpndSize_32, sub_opc, P_SCRATCH_2, true, P_GPR_3, true);
get_self_pointer(P_SCRATCH_3, true);
move_reg_to_mem(OpndSize_32, PhysicalReg_FP, true, offStackSaveArea_prevFrame-sizeofStackSaveArea, P_GPR_1, true); //set stack->prevFrame
compare_mem_reg(OpndSize_32, offsetof(Thread, interpStackEnd), P_SCRATCH_3, true, P_GPR_3, true);
conditional_jump(Condition_L, ".stackOverflow", true);
if(form == ArgsDone_Full) {
test_imm_mem(OpndSize_32, ACC_NATIVE, offMethod_accessFlags, PhysicalReg_ECX, true);
}
move_reg_to_mem(OpndSize_32, PhysicalReg_ECX, true, offStackSaveArea_method-sizeofStackSaveArea, P_GPR_1, true); //set stack->method
if(form == ArgsDone_Native || form == ArgsDone_Full) {
/* to correctly handle code cache reset:
update returnAddr and check returnAddr after done with the native method
if returnAddr is set to NULL during code cache reset,
the execution will correctly continue with interpreter */
//get returnAddr from 4(%esp) and update stack
move_mem_to_reg(OpndSize_32, 4, PhysicalReg_ESP, true,
PhysicalReg_EDX, true);
move_reg_to_mem(OpndSize_32, PhysicalReg_EDX, true,
offStackSaveArea_returnAddr-sizeofStackSaveArea, P_GPR_1, true);
}
if(form == ArgsDone_Native) {
generate_invokeNative(generateForNcg);
return 0;
}
if(form == ArgsDone_Full) {
conditional_jump(Condition_NE, ".invokeNative", true);
}
move_mem_to_reg(OpndSize_32, offMethod_clazz, PhysicalReg_ECX, true, P_SCRATCH_4, true); //get method->claz
move_mem_to_reg(OpndSize_32, offClassObject_pDvmDex, P_SCRATCH_4, true, P_SCRATCH_4, true); //get method->clazz->pDvmDex
move_reg_to_reg(OpndSize_32, P_GPR_1, true, PhysicalReg_FP, true); //update rFP
get_self_pointer(P_GPR_1, true);
move_reg_to_mem(OpndSize_32, PhysicalReg_ECX, true, offsetof(Thread, interpSave.method), P_GPR_1, true); //glue->method
move_reg_to_mem(OpndSize_32, P_SCRATCH_4, true, offsetof(Thread, interpSave.methodClassDex), P_GPR_1, true); //set_glue_dvmdex
move_reg_to_mem(OpndSize_32, PhysicalReg_FP, true, offThread_curFrame, P_GPR_1, true); //set glue->self->frame
if(!generateForNcg) {
/* returnAddr updated already for Full */
//get returnAddr from 4(%esp) and update stack
if(form == ArgsDone_Normal)
move_mem_to_reg(OpndSize_32, 4, PhysicalReg_ESP, true,
PhysicalReg_EDX, true);
//for JIT: starting bytecode in %ebx to invoke JitToInterp
move_mem_to_reg(OpndSize_32, offMethod_insns, PhysicalReg_ECX, true, PhysicalReg_EBX, true);
if(form == ArgsDone_Normal)
move_reg_to_mem(OpndSize_32, PhysicalReg_EDX, true,
offStackSaveArea_returnAddr-sizeofStackSaveArea, PhysicalReg_FP, true);
}
insertLabel(".invokeInterp", true);
if(!generateForNcg) {
bool callNoChain = false;
#ifdef PREDICTED_CHAINING
if(form == ArgsDone_Full) callNoChain = true;
#endif
if(callNoChain) {
scratchRegs[0] = PhysicalReg_EAX;
load_effective_addr(8, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
#if defined(WITH_JIT_TUNING)
/* Predicted chaining failed. Fall back to interpreter and indicate
* inline cache miss.
*/
move_imm_to_reg(OpndSize_32, kInlineCacheMiss, PhysicalReg_EDX, true);
#endif
call_dvmJitToInterpTraceSelectNoChain(); //input: rPC in %ebx
} else {
//jump to the stub at (%esp)
move_mem_to_reg(OpndSize_32, 0, PhysicalReg_ESP, true,
PhysicalReg_EDX, true);
load_effective_addr(8, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
unconditional_jump_reg(PhysicalReg_EDX, true);
}
}
if(form == ArgsDone_Full) generate_invokeNative(generateForNcg);
generate_stackOverflow();
return 0;
}
/* when WITH_JIT is true,
JIT'ed code invokes native method, after invoke, execution will continue
with the interpreter or with JIT'ed code if chained
*/
void generate_invokeNative(bool generateForNcg) {
insertLabel(".invokeNative", true);
//if(!generateForNcg)
// load_effective_addr(-8, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
load_effective_addr(-28, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, P_GPR_1, true, 0, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, P_GPR_1, true, 20, PhysicalReg_ESP, true);
scratchRegs[0] = PhysicalReg_EDX;
get_self_pointer(P_SCRATCH_1, true); //glue->self
move_reg_to_mem(OpndSize_32, PhysicalReg_ECX, true, 8, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, P_SCRATCH_1, true, 12, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, P_SCRATCH_1, true, 24, PhysicalReg_ESP, true);
move_mem_to_reg(OpndSize_32, offThread_jniLocal_nextEntry, P_SCRATCH_1, true, P_SCRATCH_2, true); //get self->local_next
scratchRegs[1] = PhysicalReg_EAX;
move_reg_to_mem(OpndSize_32, P_SCRATCH_2, true, offStackSaveArea_localRefTop-sizeofStackSaveArea, P_GPR_1, true); //update jniLocalRef of stack
move_reg_to_mem(OpndSize_32, P_GPR_1, true, offThread_curFrame, P_SCRATCH_1, true); //set self->curFrame
move_imm_to_mem(OpndSize_32, 0, offThread_inJitCodeCache, P_SCRATCH_1, true); //clear self->inJitCodeCache
load_effective_addr(offsetof(Thread, interpSave.retval), P_SCRATCH_1, true, P_SCRATCH_3, true); //self->retval
move_reg_to_mem(OpndSize_32, P_SCRATCH_3, true, 4, PhysicalReg_ESP, true);
//NOTE: native method checks the interpreted stack for arguments
// The immediate arguments on native stack: address of return value, new FP, self
call_mem(40, PhysicalReg_ECX, true);//*40(%ecx)
//we can't assume the argument stack is unmodified after the function call
//duplicate newFP & glue->self on stack: newFP (-28 & -8) glue->self (-16 & -4)
move_mem_to_reg(OpndSize_32, 20, PhysicalReg_ESP, true, P_GPR_3, true); //new FP
move_mem_to_reg(OpndSize_32, 24, PhysicalReg_ESP, true, P_GPR_1, true); //glue->self
load_effective_addr(28, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
move_mem_to_reg(OpndSize_32, offStackSaveArea_localRefTop-sizeofStackSaveArea, P_GPR_3, true, P_SCRATCH_1, true); //newSaveArea->jniLocal
compare_imm_mem(OpndSize_32, 0, offThread_exception, P_GPR_1, true); //self->exception
if(!generateForNcg)
load_effective_addr(8, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
//NOTE: PhysicalReg_FP should be callee-saved register
move_reg_to_mem(OpndSize_32, PhysicalReg_FP, true, offThread_curFrame, P_GPR_1, true); //set self->curFrame
move_reg_to_mem(OpndSize_32, P_SCRATCH_1, true, offThread_jniLocal_nextEntry, P_GPR_1, true); //set self->jniLocal
conditional_jump(Condition_NE, "common_exceptionThrown", false);
if(!generateForNcg) {
//get returnAddr, if it is not NULL,
// return to JIT'ed returnAddr after executing the native method
/* to correctly handle code cache reset:
update returnAddr and check returnAddr after done with the native method
if returnAddr is set to NULL during code cache reset,
the execution will correctly continue with interpreter */
move_mem_to_reg(OpndSize_32, offStackSaveArea_returnAddr-sizeofStackSaveArea, P_GPR_3, true, P_SCRATCH_2, true);
//set self->inJitCodeCache to returnAddr (P_GPR_1 is in %ebx)
move_reg_to_mem(OpndSize_32, P_SCRATCH_2, true, offThread_inJitCodeCache, P_GPR_1, true);
move_mem_to_reg(OpndSize_32, offStackSaveArea_savedPc-sizeofStackSaveArea, P_GPR_3, true, PhysicalReg_EBX, true); //savedPc
compare_imm_reg(OpndSize_32, 0, P_SCRATCH_2, true);
conditional_jump(Condition_E, ".nativeToInterp", true);
unconditional_jump_reg(P_SCRATCH_2, true);
//if returnAddr is NULL, return to interpreter after executing the native method
insertLabel(".nativeToInterp", true);
//move rPC by 6 (3 bytecode units for INVOKE)
alu_binary_imm_reg(OpndSize_32, add_opc, 6, PhysicalReg_EBX, true);
scratchRegs[0] = PhysicalReg_EAX;
#if defined(WITH_JIT_TUNING)
/* Return address not in code cache. Indicate that continuing with interpreter
*/
move_imm_to_reg(OpndSize_32, kCallsiteInterpreted, PhysicalReg_EDX, true);
#endif
call_dvmJitToInterpTraceSelectNoChain(); //rPC in %ebx
}
return;
}
void generate_stackOverflow() {
insertLabel(".stackOverflow", true);
//load_effective_addr(-8, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, PhysicalReg_ECX, true, 4, PhysicalReg_ESP, true);
get_self_pointer(P_GPR_1, true); //glue->self
move_reg_to_mem(OpndSize_32, P_GPR_1, true, 0, PhysicalReg_ESP, true);
call_dvmHandleStackOverflow();
load_effective_addr(8, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
unconditional_jump("common_exceptionThrown", false);
}
#undef P_GPR_1
#undef P_GPR_2
#undef P_GPR_3
#undef P_SCRATCH_1
#undef P_SCRATCH_2
#undef P_SCRATCH_3
#undef P_SCRATCH_4
#undef P_SCRATCH_5
#undef P_SCRATCH_6
/////////////////////////////////////////////
#define P_GPR_1 PhysicalReg_EBX
#define P_GPR_2 PhysicalReg_ECX
#define P_SCRATCH_1 PhysicalReg_ESI
#define P_SCRATCH_2 PhysicalReg_EDX
#define P_SCRATCH_3 PhysicalReg_ESI
#define P_SCRATCH_4 PhysicalReg_EDX
//! lower bytecode EXECUTE_INLINE
//!
int op_execute_inline(bool isRange) {
//tmp, vC, vD, vE, vF
int num;
if(!isRange) num = INST_B(inst);
else num = INST_AA(inst);
u2 tmp = FETCH(1);
u2 vC, vD, vE, vF;
if(!isRange) {
vC = FETCH(2) & 0xf;
vD = (FETCH(2) >> 4) & 0xf;
vE = (FETCH(2) >> 8) & 0xf;
vF = FETCH(2) >> 12;
} else {
vC = FETCH(2);
vD = vC + 1;
vE = vC + 2;
vF = vC + 3;
}
export_pc();
switch (tmp) {
case INLINE_EMPTYINLINEMETHOD:
return 0; /* Nop */
case INLINE_STRING_LENGTH:
get_virtual_reg(vC, OpndSize_32, 1, false);
compare_imm_reg(OpndSize_32, 0, 1, false);
conditional_jump(Condition_NE, ".do_inlined_string_length", true);
scratchRegs[0] = PhysicalReg_SCRATCH_1;
jumpToExceptionThrown(1/*exception number*/);
insertLabel(".do_inlined_string_length", true);
move_mem_to_reg(OpndSize_32, 0x14, 1, false, 2, false);
get_self_pointer(3, false);
move_reg_to_mem(OpndSize_32, 2, false, offsetof(Thread, interpSave.retval), 3, false);
return 0;
case INLINE_STRING_IS_EMPTY:
get_virtual_reg(vC, OpndSize_32, 1, false);
compare_imm_reg(OpndSize_32, 0, 1, false);
conditional_jump(Condition_NE, ".do_inlined_string_length", true);
scratchRegs[0] = PhysicalReg_SCRATCH_1;
jumpToExceptionThrown(1/*exception number*/);
insertLabel(".do_inlined_string_length", true);
compare_imm_mem(OpndSize_32, 0, 0x14, 1, false);
conditional_jump(Condition_E, ".inlined_string_length_return_true",
true);
get_self_pointer(2, false);
move_imm_to_mem(OpndSize_32, 0, offsetof(Thread, interpSave.retval), 2, false);
unconditional_jump(".inlined_string_length_done", true);
insertLabel(".inlined_string_length_return_true", true);
get_self_pointer(2, false);
move_imm_to_mem(OpndSize_32, 1, offsetof(Thread, interpSave.retval), 2, false);
insertLabel(".inlined_string_length_done", true);
return 0;
case INLINE_MATH_ABS_INT:
get_virtual_reg(vC, OpndSize_32, 1, false);
move_reg_to_reg(OpndSize_32, 1, false, 2, false);
alu_binary_imm_reg(OpndSize_32, sar_opc, 0x1f, 2, false);
alu_binary_reg_reg(OpndSize_32, xor_opc, 2, false, 1, false);
alu_binary_reg_reg(OpndSize_32, sub_opc, 2, false, 1, false);
get_self_pointer(3, false);
move_reg_to_mem(OpndSize_32, 1, false, offsetof(Thread, interpSave.retval), 3, false);
return 0;
case INLINE_MATH_ABS_LONG:
get_virtual_reg(vD, OpndSize_32, 1, false);
move_reg_to_reg(OpndSize_32, 1, false, 2, false);
alu_binary_imm_reg(OpndSize_32, sar_opc, 0x1f, 1, false);
move_reg_to_reg(OpndSize_32, 1, false, 3, false);
move_reg_to_reg(OpndSize_32, 1, false, 4, false);
get_virtual_reg(vC, OpndSize_32, 5, false);
alu_binary_reg_reg(OpndSize_32, xor_opc, 5, false, 1, false);
get_self_pointer(6, false);
move_reg_to_mem(OpndSize_32, 1, false, offsetof(Thread, interpSave.retval), 6, false);
alu_binary_reg_reg(OpndSize_32, xor_opc, 2, false, 3, false);
move_reg_to_mem(OpndSize_32, 3, false, 4 + offsetof(Thread, interpSave.retval), 6, false);
alu_binary_reg_mem(OpndSize_32, sub_opc, 4, false, offsetof(Thread, interpSave.retval), 6, false);
alu_binary_reg_mem(OpndSize_32, sbb_opc, 4, false, 4 + offsetof(Thread, interpSave.retval), 6, false);
return 0;
case INLINE_MATH_MAX_INT:
get_virtual_reg(vC, OpndSize_32, 1, false);
get_virtual_reg(vD, OpndSize_32, 2, false);
compare_reg_reg(1, false, 2, false);
conditional_move_reg_to_reg(OpndSize_32, Condition_GE, 2,
false/*src*/, 1, false/*dst*/);
get_self_pointer(3, false);
move_reg_to_mem(OpndSize_32, 1, false, offsetof(Thread, interpSave.retval), 3, false);
return 0;
case INLINE_MATH_ABS_FLOAT:
get_virtual_reg(vC, OpndSize_32, 1, false);
alu_binary_imm_reg(OpndSize_32, and_opc, 0x7fffffff, 1, false);
get_self_pointer(2, false);
move_reg_to_mem(OpndSize_32, 1, false, offsetof(Thread, interpSave.retval), 2, false);
return 0;
case INLINE_MATH_ABS_DOUBLE:
get_virtual_reg(vC, OpndSize_32, 1, false);
get_virtual_reg(vD, OpndSize_32, 2, false);
alu_binary_imm_reg(OpndSize_32, and_opc, 0x7fffffff, 2, false);
get_self_pointer(3, false);
move_reg_to_mem(OpndSize_32, 1, false, offsetof(Thread, interpSave.retval), 3, false);
move_reg_to_mem(OpndSize_32, 2, false, 4 + offsetof(Thread, interpSave.retval), 3, false);
return 0;
case INLINE_STRING_FASTINDEXOF_II:
#if defined(USE_GLOBAL_STRING_DEFS)
break;
#else
get_virtual_reg(vC, OpndSize_32, 1, false);
compare_imm_reg(OpndSize_32, 0, 1, false);
get_virtual_reg(vD, OpndSize_32, 2, false);
get_virtual_reg(vE, OpndSize_32, 3, false);
conditional_jump(Condition_NE, ".do_inlined_string_fastIndexof",
true);
scratchRegs[0] = PhysicalReg_SCRATCH_1;
jumpToExceptionThrown(1/*exception number*/);
insertLabel(".do_inlined_string_fastIndexof", true);
move_mem_to_reg(OpndSize_32, 0x14, 1, false, 4, false);
move_mem_to_reg(OpndSize_32, 0x8, 1, false, 5, false);
move_mem_to_reg(OpndSize_32, 0x10, 1, false, 6, false);
alu_binary_reg_reg(OpndSize_32, xor_opc, 1, false, 1, false);
compare_imm_reg(OpndSize_32, 0, 3, false);
conditional_move_reg_to_reg(OpndSize_32, Condition_NS, 3, false, 1,
false);
compare_reg_reg(4, false, 1, false);
conditional_jump(Condition_GE,
".do_inlined_string_fastIndexof_exitfalse", true);
dump_mem_scale_reg(Mnemonic_LEA, OpndSize_32, 5, false, 0xc/*disp*/,
6, false, 2, 5, false, LowOpndRegType_gp);
movez_mem_disp_scale_to_reg(OpndSize_16, 5, false, 0, 1, false, 2,
3, false);
compare_reg_reg(3, false, 2, false);
conditional_jump(Condition_E, ".do_inlined_string_fastIndexof_exit",
true);
load_effective_addr(0x1, 1, false, 3, false);
load_effective_addr_scale(5, false, 3, false, 2, 5, false);
unconditional_jump(".do_inlined_string_fastIndexof_iter", true);
insertLabel(".do_inlined_string_fastIndexof_ch_cmp", true);
if(gDvm.executionMode == kExecutionModeNcgO1) {
rememberState(1);
}
movez_mem_to_reg(OpndSize_16, 0, 5, false, 6, false);
load_effective_addr(0x2, 5, false, 5, false);
compare_reg_reg(6, false, 2, false);
conditional_jump(Condition_E, ".do_inlined_string_fastIndexof_exit",
true);
load_effective_addr(0x1, 3, false, 3, false);
insertLabel(".do_inlined_string_fastIndexof_iter", true);
compare_reg_reg(4, false, 3, false);
move_reg_to_reg(OpndSize_32, 3, false, 1, false);
if(gDvm.executionMode == kExecutionModeNcgO1) {
transferToState(1);
}
conditional_jump(Condition_NE,
".do_inlined_string_fastIndexof_ch_cmp", true);
insertLabel(".do_inlined_string_fastIndexof_exitfalse", true);
move_imm_to_reg(OpndSize_32, 0xffffffff, 1, false);
insertLabel(".do_inlined_string_fastIndexof_exit", true);
get_self_pointer(7, false);
move_reg_to_mem(OpndSize_32, 1, false, offsetof(Thread, interpSave.retval), 7, false);
return 0;
case INLINE_FLOAT_TO_RAW_INT_BITS:
get_virtual_reg(vC, OpndSize_32, 1, false);
get_self_pointer(2, false);
move_reg_to_mem(OpndSize_32, 1, false, offsetof(Thread, interpSave.retval), 2, false);
return 0;
case INLINE_INT_BITS_TO_FLOAT:
get_virtual_reg(vC, OpndSize_32, 1, false);
get_self_pointer(2, false);
move_reg_to_mem(OpndSize_32, 1, false, offsetof(Thread, interpSave.retval), 2, false);
return 0;
case INLINE_DOUBLE_TO_RAW_LONG_BITS:
get_virtual_reg(vC, OpndSize_32, 1, false);
get_self_pointer(3, false);
move_reg_to_mem(OpndSize_32, 1, false, offsetof(Thread, interpSave.retval), 3, false);
get_virtual_reg(vD, OpndSize_32, 2, false);
move_reg_to_mem(OpndSize_32, 2, false, 4 + offsetof(Thread, interpSave.retval), 3, false);
return 0;
case INLINE_LONG_BITS_TO_DOUBLE:
get_virtual_reg(vC, OpndSize_32, 1, false);
get_virtual_reg(vD, OpndSize_32, 2, false);
get_self_pointer(3, false);
move_reg_to_mem(OpndSize_32, 2, false, 4 + offsetof(Thread, interpSave.retval), 3, false);
move_reg_to_mem(OpndSize_32, 1, false, offsetof(Thread, interpSave.retval), 3, false);
return 0;
default:
break;
}
#endif
get_self_pointer(PhysicalReg_SCRATCH_1, false);
load_effective_addr(offsetof(Thread, interpSave.retval), PhysicalReg_SCRATCH_1, false, 1, false);
load_effective_addr(-24, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, 1, false, 16, PhysicalReg_ESP, true);
if(num >= 1) {
get_virtual_reg(vC, OpndSize_32, 2, false);
move_reg_to_mem(OpndSize_32, 2, false, 0, PhysicalReg_ESP, true);
}
if(num >= 2) {
get_virtual_reg(vD, OpndSize_32, 3, false);
move_reg_to_mem(OpndSize_32, 3, false, 4, PhysicalReg_ESP, true);
}
if(num >= 3) {
get_virtual_reg(vE, OpndSize_32, 4, false);
move_reg_to_mem(OpndSize_32, 4, false, 8, PhysicalReg_ESP, true);
}
if(num >= 4) {
get_virtual_reg(vF, OpndSize_32, 5, false);
move_reg_to_mem(OpndSize_32, 5, false, 12, PhysicalReg_ESP, true);
}
beforeCall("execute_inline");
load_imm_global_data_API("gDvmInlineOpsTable", OpndSize_32, 6, false);
call_mem(16*tmp, 6, false);//
afterCall("execute_inline");
compare_imm_reg(OpndSize_32, 0, PhysicalReg_EAX, true);
load_effective_addr(24, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
conditional_jump(Condition_NE, ".execute_inline_done", true);
//jump to dvmJitToExceptionThrown
scratchRegs[0] = PhysicalReg_SCRATCH_1;
jumpToExceptionThrown(1/*exception number*/);
insertLabel(".execute_inline_done", true);
rPC += 3;
return 0;
}
#undef P_GPR_1
#undef P_GPR_2
#undef P_SCRATCH_1
#undef P_SCRATCH_2
#undef P_SCRATCH_3
#undef P_SCRATCH_4
//! lower bytecode INVOKE_OBJECT_INIT_RANGE
//!
int op_invoke_object_init_range() {
return -1;
}
#define P_GPR_1 PhysicalReg_EBX
#define P_SCRATCH_1 PhysicalReg_ESI
#define P_SCRATCH_2 PhysicalReg_EDX
#define PP_GPR_1 PhysicalReg_EBX
#define PP_GPR_2 PhysicalReg_ESI
#define PP_GPR_3 PhysicalReg_EAX
#define PP_GPR_4 PhysicalReg_EDX
//! common code for INVOKE_VIRTUAL_QUICK
//! It uses helper function if the switch is on
int common_invoke_virtual_quick(bool hasRange, u2 vD, u2 IMMC) {
#ifdef WITH_JIT_INLINING
/* An invoke with the MIR_INLINED is effectively a no-op */
if (traceCurrentMIR->OptimizationFlags & MIR_INLINED)
return false;
/*
* If the invoke has non-null misPredBranchOver, we need to generate
* the non-inlined version of the invoke here to handle the
* mispredicted case.
*/
if (traceCurrentMIR->meta.callsiteInfo->misPredBranchOver) {
genLandingPadForMispredictedCallee(); //cUnit, mir, bb, labelList);
}
#endif
export_pc();
constVREndOfBB();
beforeCall("exception"); //dump GG, GL VRs
/////////////////////////////////////////////////
get_virtual_reg(vD, OpndSize_32, 1, false);
simpleNullCheck(1, false, vD);
#ifndef PREDICTED_CHAINING
move_mem_to_reg(OpndSize_32, 0, 1, false, 2, false);
move_mem_to_reg(OpndSize_32, offClassObject_vtable, 2, false, 3, false);
move_mem_to_reg(OpndSize_32, IMMC, 3, false, PhysicalReg_ECX, true);
if(hasRange) {
common_invokeMethodRange(ArgsDone_Full);
}
else {
common_invokeMethodNoRange(ArgsDone_Full);
}
#else
gen_predicted_chain(hasRange, -1, IMMC, false, 1/*tmp1*/);
#endif
////////////////////////
return 0;
}
#undef P_GPR_1
#undef P_SCRATCH_1
#undef P_SCRATCH_2
#undef PP_GPR_1
#undef PP_GPR_2
#undef PP_GPR_3
#undef PP_GPR_4
//! lower bytecode INVOKE_VIRTUAL_QUICK by calling common_invoke_virtual_quick
//!
int op_invoke_virtual_quick() {
u2 vD = FETCH(2) & 0xf;
u2 IMMC = 4*FETCH(1);
int retval = common_invoke_virtual_quick(false, vD, IMMC);
#if defined(ENABLE_TRACING) && !defined(TRACING_OPTION2)
insertMapWorklist(offsetPC+3, stream - streamMethodStart, 1); //check when helper switch is on
#endif
rPC += 3;
return retval;
}
//! lower bytecode INVOKE_VIRTUAL_QUICK_RANGE by calling common_invoke_virtual_quick
//!
int op_invoke_virtual_quick_range() {
u2 vD = FETCH(2);
u2 IMMC = 4*FETCH(1);
int retval = common_invoke_virtual_quick(true, vD, IMMC);
#if defined(ENABLE_TRACING) && !defined(TRACING_OPTION2)
insertMapWorklist(offsetPC+3, stream - streamMethodStart, 1); //check when helper switch is on
#endif
rPC += 3;
return retval;
}
#define P_GPR_1 PhysicalReg_EBX
#define P_GPR_2 PhysicalReg_ESI
#define P_SCRATCH_1 PhysicalReg_EDX
//! common code to lower INVOKE_SUPER_QUICK
//!
int common_invoke_super_quick(bool hasRange, u2 vD, u2 IMMC) {
export_pc();
constVREndOfBB();
beforeCall("exception"); //dump GG, GL VRs
compare_imm_VR(OpndSize_32, 0, vD);
conditional_jump_global_API(Condition_E, "common_errNullObject", false);
/* for trace-based JIT, callee is already resolved */
int mIndex = IMMC/4;
const Method *calleeMethod = currentMethod->clazz->super->vtable[mIndex];
move_imm_to_reg(OpndSize_32, (int) calleeMethod, PhysicalReg_ECX, true);
if(hasRange) {
common_invokeMethodRange(convertCalleeToType(calleeMethod));
}
else {
common_invokeMethodNoRange(convertCalleeToType(calleeMethod));
}
return 0;
}
#undef P_GPR_1
#undef P_GPR_2
#undef P_SCRATCH_1
//! lower bytecode INVOKE_SUPER_QUICK by calling common_invoke_super_quick
//!
int op_invoke_super_quick() {
u2 vD = FETCH(2) & 0xf;
u2 IMMC = 4*FETCH(1);
int retval = common_invoke_super_quick(false, vD, IMMC);
#if defined(ENABLE_TRACING) && !defined(TRACING_OPTION2)
insertMapWorklist(offsetPC+3, stream - streamMethodStart, 1); //check when helper switch is on
#endif
rPC += 3;
return retval;
}
//! lower bytecode INVOKE_SUPER_QUICK_RANGE by calling common_invoke_super_quick
//!
int op_invoke_super_quick_range() {
u2 vD = FETCH(2);
u2 IMMC = 4*FETCH(1);
int retval = common_invoke_super_quick(true, vD, IMMC);
#if defined(ENABLE_TRACING) && !defined(TRACING_OPTION2)
insertMapWorklist(offsetPC+3, stream - streamMethodStart, 1); //check when helper switch is on
#endif
rPC += 3;
return retval;
}
/////// code to predict the callee method for invoke_virtual & invoke_interface
#define offChainingCell_clazz 8
#define offChainingCell_method 12
#define offChainingCell_counter 16
#define P_GPR_1 PhysicalReg_EBX
#define P_GPR_2 PhysicalReg_EAX
#define P_GPR_3 PhysicalReg_ESI
#define P_SCRATCH_2 PhysicalReg_EDX
/* TODO gingerbread: implemented for O1, but not for O0:
valid input to JitToPatch & use icRechainCount */
/* update predicted method for invoke interface */
// 2 inputs: ChainingCell in P_GPR_1, current class object in P_GPR_3
void predicted_chain_interface_O0(u2 tmp) {
ALOGI("TODO chain_interface_O0");
/* set up arguments to dvmFindInterfaceMethodInCache */
load_effective_addr(-16, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
move_imm_to_mem(OpndSize_32, tmp, 4, PhysicalReg_ESP, true);
move_imm_to_mem(OpndSize_32, (int) currentMethod->clazz->pDvmDex, 12, PhysicalReg_ESP, true);
move_imm_to_mem(OpndSize_32, (int) currentMethod, 8, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, P_GPR_3, true, 0, PhysicalReg_ESP, true);
scratchRegs[0] = PhysicalReg_EDX;
call_dvmFindInterfaceMethodInCache();
load_effective_addr(16, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
/* if dvmFindInterfaceMethodInCache returns NULL, throw exception
otherwise, jump to .find_interface_done */
compare_imm_reg(OpndSize_32, 0, PhysicalReg_EAX, true);
conditional_jump(Condition_NE, ".find_interface_done", true);
scratchRegs[0] = PhysicalReg_EAX;
jumpToExceptionThrown(1/*exception number*/);
/* the interface method is found */
insertLabel(".find_interface_done", true);
/* reduce counter in chaining cell by 1 */
move_mem_to_reg(OpndSize_32, offChainingCell_counter, P_GPR_1, true, P_SCRATCH_2, true); //counter
alu_binary_imm_reg(OpndSize_32, sub_opc, 0x1, P_SCRATCH_2, true);
move_reg_to_mem(OpndSize_32, P_SCRATCH_2, true, offChainingCell_counter, P_GPR_1, true);
/* if counter is still greater than zero, skip prediction
if it is zero, update predicted method */
compare_imm_reg(OpndSize_32, 0, P_SCRATCH_2, true);
conditional_jump(Condition_G, ".skipPrediction", true);
/* call dvmJitToPatchPredictedChain to update predicted method */
//%ecx has callee method for virtual, %eax has callee for interface
/* set up arguments for dvmJitToPatchPredictedChain */
load_effective_addr(-16, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, PhysicalReg_EAX, true, 0, PhysicalReg_ESP, true);
insertChainingWorklist(traceCurrentBB->taken->id, stream);
move_chain_to_mem(OpndSize_32, traceCurrentBB->taken->id, 8, PhysicalReg_ESP, true); //predictedChainCell
move_reg_to_mem(OpndSize_32, P_GPR_3, true, 12, PhysicalReg_ESP, true);
scratchRegs[0] = PhysicalReg_EAX;
call_dvmJitToPatchPredictedChain(); //inputs: method, unused, predictedChainCell, clazz
load_effective_addr(16, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
insertLabel(".skipPrediction", true);
move_reg_to_reg(OpndSize_32, PhysicalReg_EAX, true, PhysicalReg_ECX, true);
}
// 2 inputs: ChainingCell in temp 41, current class object in temp 40
void predicted_chain_interface_O1(u2 tmp) {
/* set up arguments to dvmFindInterfaceMethodInCache */
load_effective_addr(-16, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
move_imm_to_mem(OpndSize_32, tmp, 4, PhysicalReg_ESP, true);
move_imm_to_mem(OpndSize_32, (int) currentMethod->clazz->pDvmDex, 12, PhysicalReg_ESP, true);
move_imm_to_mem(OpndSize_32, (int) currentMethod, 8, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, 40, false, 0, PhysicalReg_ESP, true);
scratchRegs[0] = PhysicalReg_SCRATCH_10;
call_dvmFindInterfaceMethodInCache();
load_effective_addr(16, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
/* if dvmFindInterfaceMethodInCache returns NULL, throw exception
otherwise, jump to .find_interface_done */
compare_imm_reg(OpndSize_32, 0, PhysicalReg_EAX, true);
conditional_jump(Condition_NE, ".find_interface_done", true);
rememberState(3);
scratchRegs[0] = PhysicalReg_SCRATCH_9;
jumpToExceptionThrown(1/*exception number*/);
goToState(3);
/* the interface method is found */
insertLabel(".find_interface_done", true);
#if 1 //
/* for gingerbread, counter is stored in glue structure
if clazz is not initialized, set icRechainCount to 0, otherwise, reduce it by 1 */
/* for gingerbread: t43 = 0 t44 = t33 t33-- cmov_ne t43 = t33 cmov_ne t44 = t33 */
move_mem_to_reg(OpndSize_32, offChainingCell_clazz, 41, false, 45, false);
move_imm_to_reg(OpndSize_32, 0, 43, false);
get_self_pointer(PhysicalReg_SCRATCH_7, isScratchPhysical);
move_mem_to_reg(OpndSize_32, offsetof(Thread, icRechainCount), PhysicalReg_SCRATCH_7, isScratchPhysical, 33, false); //counter
move_reg_to_reg(OpndSize_32, 33, false, 44, false);
alu_binary_imm_reg(OpndSize_32, sub_opc, 0x1, 33, false);
/* sub_opc will update control flags, so compare_imm_reg must happen after */
compare_imm_reg(OpndSize_32, 0, 45, false);
conditional_move_reg_to_reg(OpndSize_32, Condition_NZ, 33, false/*src*/, 43, false/*dst*/);
conditional_move_reg_to_reg(OpndSize_32, Condition_NZ, 33, false/*src*/, 44, false/*dst*/);
move_reg_to_mem(OpndSize_32, 44, false, offsetof(Thread, icRechainCount), PhysicalReg_SCRATCH_7, isScratchPhysical);
#else
/* reduce counter in chaining cell by 1 */
move_mem_to_reg(OpndSize_32, offChainingCell_counter, 41, false, 33, false); //counter
alu_binary_imm_reg(OpndSize_32, sub_opc, 0x1, 33, false);
move_reg_to_mem(OpndSize_32, 33, false, offChainingCell_counter, 41, false);
#endif
/* if counter is still greater than zero, skip prediction
if it is zero, update predicted method */
compare_imm_reg(OpndSize_32, 0, 43, false);
conditional_jump(Condition_G, ".skipPrediction", true);
rememberState(4);
/* call dvmJitToPatchPredictedChain to update predicted method */
//%ecx has callee method for virtual, %eax has callee for interface
/* set up arguments for dvmJitToPatchPredictedChain */
load_effective_addr(-16, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, PhysicalReg_EAX, true, 0, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, PhysicalReg_SCRATCH_7, isScratchPhysical, 4, PhysicalReg_ESP, true);
insertChainingWorklist(traceCurrentBB->taken->id, stream);
move_chain_to_mem(OpndSize_32, traceCurrentBB->taken->id, 8, PhysicalReg_ESP, true); //predictedChainCell
move_reg_to_mem(OpndSize_32, 40, false, 12, PhysicalReg_ESP, true);
scratchRegs[0] = PhysicalReg_SCRATCH_8;
call_dvmJitToPatchPredictedChain(); //inputs: method, unused, predictedChainCell, clazz
load_effective_addr(16, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
transferToState(4);
insertLabel(".skipPrediction", true);
move_reg_to_reg(OpndSize_32, PhysicalReg_EAX, true, PhysicalReg_ECX, true);
}
/* update predicted method for invoke virtual */
// 2 inputs: ChainingCell in P_GPR_1, current class object in P_GPR_3
void predicted_chain_virtual_O0(u2 IMMC) {
ALOGI("TODO chain_virtual_O0");
/* reduce counter in chaining cell by 1 */
move_mem_to_reg(OpndSize_32, offChainingCell_counter, P_GPR_1, true, P_GPR_2, true); //counter
move_mem_to_reg(OpndSize_32, offClassObject_vtable, P_GPR_3, true, P_SCRATCH_2, true);
alu_binary_imm_reg(OpndSize_32, sub_opc, 0x1, P_GPR_2, true);
move_mem_to_reg(OpndSize_32, IMMC, P_SCRATCH_2, true, PhysicalReg_ECX, true);
move_reg_to_mem(OpndSize_32, P_GPR_2, true, offChainingCell_counter, P_GPR_1, true);
/* if counter is still greater than zero, skip prediction
if it is zero, update predicted method */
compare_imm_reg(OpndSize_32, 0, P_GPR_2, true);
conditional_jump(Condition_G, ".skipPrediction", true);
/* call dvmJitToPatchPredictedChain to update predicted method */
//%ecx has callee method for virtual, %eax has callee for interface
/* set up arguments for dvmJitToPatchPredictedChain */
load_effective_addr(-16, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, PhysicalReg_ECX, true, 0, PhysicalReg_ESP, true);
insertChainingWorklist(traceCurrentBB->taken->id, stream);
move_chain_to_mem(OpndSize_32, traceCurrentBB->taken->id, 8, PhysicalReg_ESP, true); //predictedChainCell
move_reg_to_mem(OpndSize_32, P_GPR_3, true, 12, PhysicalReg_ESP, true);
scratchRegs[0] = PhysicalReg_EAX;
call_dvmJitToPatchPredictedChain(); //inputs: method, unused, predictedChainCell, clazz
load_effective_addr(16, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
//callee method in %ecx for invoke virtual
move_reg_to_reg(OpndSize_32, PhysicalReg_EAX, true, PhysicalReg_ECX, true);
insertLabel(".skipPrediction", true);
}
// 2 inputs: ChainingCell in temp 41, current class object in temp 40
// extra input: predicted clazz in temp 32
void predicted_chain_virtual_O1(u2 IMMC) {
/* reduce counter in chaining cell by 1 */
/* for gingerbread: t43 = 0 t44 = t33 t33-- cmov_ne t43 = t33 cmov_ne t44 = t33 */
get_self_pointer(PhysicalReg_SCRATCH_7, isScratchPhysical);
move_imm_to_reg(OpndSize_32, 0, 43, false);
move_mem_to_reg(OpndSize_32, offsetof(Thread, icRechainCount), PhysicalReg_SCRATCH_7, isScratchPhysical, 33, false); //counter
move_mem_to_reg(OpndSize_32, offClassObject_vtable, 40, false, 34, false);
move_reg_to_reg(OpndSize_32, 33, false, 44, false);
alu_binary_imm_reg(OpndSize_32, sub_opc, 0x1, 33, false);
compare_imm_reg(OpndSize_32, 0, 32, false); // after sub_opc
move_mem_to_reg(OpndSize_32, IMMC, 34, false, PhysicalReg_ECX, true);
conditional_move_reg_to_reg(OpndSize_32, Condition_NZ, 33, false/*src*/, 43, false/*dst*/);
conditional_move_reg_to_reg(OpndSize_32, Condition_NZ, 33, false/*src*/, 44, false/*dst*/);
move_reg_to_mem(OpndSize_32, 44, false, offsetof(Thread, icRechainCount), PhysicalReg_SCRATCH_7, isScratchPhysical);
/* if counter is still greater than zero, skip prediction
if it is zero, update predicted method */
compare_imm_reg(OpndSize_32, 0, 43, false);
conditional_jump(Condition_G, ".skipPrediction", true);
rememberState(2);
/* call dvmJitToPatchPredictedChain to update predicted method */
//%ecx has callee method for virtual, %eax has callee for interface
/* set up arguments for dvmJitToPatchPredictedChain */
load_effective_addr(-16, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, PhysicalReg_ECX, true, 0, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, PhysicalReg_SCRATCH_7, isScratchPhysical, 4, PhysicalReg_ESP, true);
if(traceCurrentBB->taken)
insertChainingWorklist(traceCurrentBB->taken->id, stream);
int traceTakenId = traceCurrentBB->taken ? traceCurrentBB->taken->id : 0;
move_chain_to_mem(OpndSize_32, traceTakenId, 8, PhysicalReg_ESP, true); //predictedChainCell
move_reg_to_mem(OpndSize_32, 40, false, 12, PhysicalReg_ESP, true);
scratchRegs[0] = PhysicalReg_SCRATCH_10;
call_dvmJitToPatchPredictedChain(); //inputs: method, unused, predictedChainCell, clazz
load_effective_addr(16, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
//callee method in %ecx for invoke virtual
move_reg_to_reg(OpndSize_32, PhysicalReg_EAX, true, PhysicalReg_ECX, true);
transferToState(2);
insertLabel(".skipPrediction", true);
}
static int invokeChain_inst = 0;
/* object "this" is in %ebx */
void gen_predicted_chain_O0(bool isRange, u2 tmp, int IMMC, bool isInterface, int inputReg) {
ALOGI("TODO predicted_chain_O0");
/* get current class object */
move_mem_to_reg(OpndSize_32, offObject_clazz, PhysicalReg_EBX, true,
P_GPR_3, true);
#ifdef DEBUG_CALL_STACK3
scratchRegs[0] = PhysicalReg_EAX;
call_debug_dumpSwitch(); //%ebx, %eax, %edx
move_imm_to_reg(OpndSize_32, 0xdd11, PhysicalReg_EBX, true);
call_debug_dumpSwitch();
#endif
/* get predicted clazz
get predicted method
*/
insertChainingWorklist(traceCurrentBB->taken->id, stream);
move_chain_to_reg(OpndSize_32, traceCurrentBB->taken->id, P_GPR_1, true); //predictedChainCell
move_mem_to_reg(OpndSize_32, offChainingCell_clazz, P_GPR_1, true, P_SCRATCH_2, true);//predicted clazz
move_mem_to_reg(OpndSize_32, offChainingCell_method, P_GPR_1, true, PhysicalReg_ECX, true);//predicted method
#ifdef DEBUG_CALL_STACK3
load_effective_addr(-12, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, P_GPR_1, true, 8, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, P_SCRATCH_2, true, 4, PhysicalReg_ESP, true);
move_reg_to_mem(OpndSize_32, P_GPR_3, true, 0, PhysicalReg_ESP, true);
move_reg_to_reg(OpndSize_32, P_SCRATCH_2, true, PhysicalReg_EBX, true);
call_debug_dumpSwitch();
move_imm_to_reg(OpndSize_32, 0xdd22, PhysicalReg_EBX, true);
scratchRegs[0] = PhysicalReg_EAX;
call_debug_dumpSwitch(); //%ebx, %eax, %edx
move_reg_to_reg(OpndSize_32, P_GPR_3, true, PhysicalReg_EBX, true);
call_debug_dumpSwitch();
move_reg_to_reg(OpndSize_32, PhysicalReg_ECX, true, PhysicalReg_EBX, true);
call_debug_dumpSwitch();
move_mem_to_reg(OpndSize_32, 8, PhysicalReg_ESP, true, P_GPR_1, true);
move_mem_to_reg(OpndSize_32, 4, PhysicalReg_ESP, true, P_SCRATCH_2, true);
move_mem_to_reg(OpndSize_32, 0, PhysicalReg_ESP, true, P_GPR_3, true);
load_effective_addr(12, PhysicalReg_ESP, true, PhysicalReg_ESP, true);
#endif
/* compare current class object against predicted clazz
if equal, prediction is still valid, jump to .invokeChain */
//live registers: P_GPR_1, P_GPR_3, P_SCRATCH_2
compare_reg_reg(P_GPR_3, true, P_SCRATCH_2, true);
conditional_jump(Condition_E, ".invokeChain", true);
invokeChain_inst++;
//get callee method and update predicted method if necessary
if(isInterface) {
predicted_chain_interface_O0(tmp);
} else {
predicted_chain_virtual_O0(IMMC);
}
#ifdef DEBUG_CALL_STACK3
move_imm_to_reg(OpndSize_32, 0xeeee, PhysicalReg_EBX, true);
scratchRegs[0] = PhysicalReg_EAX;
call_debug_dumpSwitch(); //%ebx, %eax, %edx
insertChainingWorklist(traceCurrentBB->taken->id, stream);
move_chain_to_reg(OpndSize_32, traceCurrentBB->taken->id, PhysicalReg_EBX, true);
call_debug_dumpSwitch();
#endif
if(isRange) {
common_invokeMethodRange(ArgsDone_Full);
}
else {
common_invokeMethodNoRange(ArgsDone_Full);
}
insertLabel(".invokeChain", true);
#ifdef DEBUG_CALL_STACK3
move_imm_to_reg(OpndSize_32, 0xdddd, PhysicalReg_EBX, true);
scratchRegs[0] = PhysicalReg_EAX;
call_debug_dumpSwitch(); //%ebx, %eax, %edx
insertChainingWorklist(traceCurrentBB->taken->id, stream);
move_chain_to_reg(OpndSize_32, traceCurrentBB->taken->id, PhysicalReg_EBX, true);
call_debug_dumpSwitch();
move_reg_to_reg(OpndSize_32, PhysicalReg_ECX, true, PhysicalReg_EBX, true);
call_debug_dumpSwitch();
#endif
if(isRange) {
common_invokeMethodRange(ArgsDone_Normal);
}
else {
common_invokeMethodNoRange(ArgsDone_Normal);
}
}
/* object "this" is in inputReg: 5 for virtual, 1 for interface, 1 for virtual_quick */
void gen_predicted_chain_O1(bool isRange, u2 tmp, int IMMC, bool isInterface, int inputReg) {
/* get current class object */
move_mem_to_reg(OpndSize_32, offObject_clazz, inputReg, false,
40, false);
/* get predicted clazz
get predicted method
*/
if(traceCurrentBB->taken)
insertChainingWorklist(traceCurrentBB->taken->id, stream);
int traceTakenId = traceCurrentBB->taken ? traceCurrentBB->taken->id : 0;
move_chain_to_reg(OpndSize_32, traceTakenId, 41, false); //predictedChainCell
move_mem_to_reg(OpndSize_32, offChainingCell_clazz, 41, false, 32, false);//predicted clazz
move_mem_to_reg(OpndSize_32, offChainingCell_method, 41, false, PhysicalReg_ECX, true);//predicted method
/* update stack with parameters first, then decide the callee */
if(isRange) common_invokeMethodRange_noJmp();
else common_invokeMethodNoRange_noJmp();
/* compare current class object against predicted clazz
if equal, prediction is still valid, jump to .invokeChain */
compare_reg_reg(40, false, 32, false);
conditional_jump(Condition_E, ".invokeChain", true);
rememberState(1);
invokeChain_inst++;
//get callee method and update predicted method if necessary
if(isInterface) {
predicted_chain_interface_O1(tmp);
} else {
predicted_chain_virtual_O1(IMMC);
}
common_invokeMethod_Jmp(ArgsDone_Full); //will touch %ecx
insertLabel(".invokeChain", true);
goToState(1);
common_invokeMethod_Jmp(ArgsDone_Normal);
}
void gen_predicted_chain(bool isRange, u2 tmp, int IMMC, bool isInterface, int inputReg) {
return gen_predicted_chain_O1(isRange, tmp, IMMC, isInterface, inputReg);
}
#undef P_GPR_1
#undef P_GPR_2
#undef P_GPR_3
#undef P_SCRATCH_2