arkcompiler_ets_runtime/ecmascript/compiler/llvm_codegen.cpp
zhangyukun 3b881fa018 Sync code from c_asm_interpreter to master
Use Stub to impl handlers of bytecodes and use tail call to impl
dispatch of handlers of bytecodes

Signed-off-by: zhangyukun <zhangyukun8@huawei.com>
Change-Id: I7afabdc40ddfa11345aa1029059514aa4f55b9f8
2022-02-22 17:01:18 +08:00

386 lines
15 KiB
C++

/*
* Copyright (c) 2021 Huawei Device Co., Ltd.
* 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.
*/
#include "llvm_codegen.h"
#include <string>
#include <vector>
#include "ecmascript/compiler/compiler_macros.h"
#include "ecmascript/compiler/stub-inl.h"
#include "ecmascript/ecma_macros.h"
#include "ecmascript/object_factory.h"
#include "llvm-c/Analysis.h"
#include "llvm-c/Core.h"
#include "llvm-c/Disassembler.h"
#include "llvm-c/DisassemblerTypes.h"
#include "llvm-c/Target.h"
#include "llvm-c/Transforms/PassManagerBuilder.h"
#include "llvm-c/Transforms/Scalar.h"
#include "llvm/ADT/APInt.h"
#include "llvm/CodeGen/BuiltinGCs.h"
#include "llvm/ExecutionEngine/ExecutionEngine.h"
#include "llvm/ExecutionEngine/GenericValue.h"
#include "llvm/IR/Argument.h"
#include "llvm/IR/BasicBlock.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/InstrTypes.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/LegacyPassManager.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/Type.h"
#include "llvm/IR/Verifier.h"
#include "llvm/IRReader/IRReader.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/Host.h"
#include "llvm/Support/TargetSelect.h"
#include "llvm/Transforms/Scalar.h"
#include "llvm/llvm_stackmap_parser.h"
#include "stub_descriptor.h"
using namespace panda::ecmascript;
namespace panda::ecmascript::kungfu {
void LLVMIRGeneratorImpl::GenerateCodeForStub(Circuit *circuit, const ControlFlowGraph &graph, int index,
const CompilationConfig *cfg)
{
LLVMValueRef function = module_->GetStubFunction(index);
LLVMIRBuilder builder(&graph, circuit, module_, function, cfg);
builder.Build();
}
void LLVMModuleAssembler::AssembleModule()
{
assembler_.Run();
}
void LLVMModuleAssembler::AssembleStubModule(StubModule *module)
{
auto codeBuff = reinterpret_cast<uintptr_t>(assembler_.GetCodeBuffer());
auto engine = assembler_.GetEngine();
std::map<uint64_t, std::string> addr2name;
for (int i = 0; i < ALL_STUB_MAXCOUNT; i++) {
auto stubfunction = stubmodule_->GetStubFunction(i);
#ifndef NDEBUG
COMPILER_LOG(DEBUG) << " AssembleStubModule :" << i << " th " << std::endl;
#endif
if (stubfunction != nullptr) {
uintptr_t stubEntry = reinterpret_cast<uintptr_t>(LLVMGetPointerToGlobal(engine, stubfunction));
module->SetStubEntry(i, stubEntry - codeBuff);
if (i >= FAST_STUB_MAXCOUNT) {
addr2name[stubEntry] = FastStubDescriptors::GetInstance()
.GetStubDescriptor(CallStubId::NAME_BytecodeHandler)->GetName();
} else {
addr2name[stubEntry] = FastStubDescriptors::GetInstance().GetStubDescriptor(i)->GetName();
}
#ifndef NDEBUG
COMPILER_LOG(DEBUG) << "name : " << addr2name[codeBuff] << std::endl;
#endif
}
}
module->SetHostCodeSectionAddr(codeBuff);
// stackmaps ptr and size
module->SetStackMapAddr(reinterpret_cast<uintptr_t>(assembler_.GetStackMapsSection()));
module->SetStackMapSize(assembler_.GetStackMapsSize());
#ifndef NDEBUG
assembler_.Disassemble(addr2name);
#endif
}
static uint8_t *RoundTripAllocateCodeSection(void *object, uintptr_t size, [[maybe_unused]] unsigned alignment,
[[maybe_unused]] unsigned sectionID, const char *sectionName)
{
COMPILER_LOG(DEBUG) << "RoundTripAllocateCodeSection object " << object << " - ";
struct CodeInfo& state = *static_cast<struct CodeInfo*>(object);
uint8_t *addr = state.AllocaCodeSection(size, sectionName);
COMPILER_LOG(DEBUG) << "RoundTripAllocateCodeSection addr:" << std::hex <<
reinterpret_cast<std::uintptr_t>(addr) << addr << " size:0x" << size << " + ";
return addr;
}
static uint8_t *RoundTripAllocateDataSection(void *object, uintptr_t size, [[maybe_unused]] unsigned alignment,
[[maybe_unused]] unsigned sectionID, const char *sectionName,
[[maybe_unused]] LLVMBool isReadOnly)
{
struct CodeInfo& state = *static_cast<struct CodeInfo*>(object);
return state.AllocaDataSection(size, sectionName);
}
static LLVMBool RoundTripFinalizeMemory(void *object, [[maybe_unused]] char **errMsg)
{
COMPILER_LOG(DEBUG) << "RoundTripFinalizeMemory object " << object << " - ";
return 0;
}
static void RoundTripDestroy(void *object)
{
COMPILER_LOG(DEBUG) << "RoundTripDestroy object " << object << " - ";
}
void LLVMAssembler::UseRoundTripSectionMemoryManager()
{
auto sectionMemoryManager = std::make_unique<llvm::SectionMemoryManager>();
options_.MCJMM =
LLVMCreateSimpleMCJITMemoryManager(&codeInfo_, RoundTripAllocateCodeSection,
RoundTripAllocateDataSection, RoundTripFinalizeMemory, RoundTripDestroy);
}
bool LLVMAssembler::BuildMCJITEngine()
{
COMPILER_LOG(DEBUG) << " BuildMCJITEngine - ";
LLVMBool ret = LLVMCreateMCJITCompilerForModule(&engine_, module_, &options_, sizeof(options_), &error_);
if (ret) {
LOG_ECMA(FATAL) << "error_ : " << error_;
return false;
}
COMPILER_LOG(DEBUG) << " BuildMCJITEngine + ";
return true;
}
// rewrite patchpoint id value that is planned to be stored into JS thread temporarily
void LLVMAssembler::RewritePatchPointIdStoredOnThread(LLVMValueRef instruction, uint64_t id)
{
LLVMValueRef targetInst = LLVMGetPreviousInstruction(instruction);
while (targetInst != nullptr) {
if (LLVMGetInstructionOpcode(targetInst) == LLVMStore) {
size_t len = 0;
std::string destName(LLVMGetValueName2(LLVMGetOperand(targetInst, 1), &len));
std::string targetName("getFramePatchPointIdAddr");
// find the store instruction that planned to change getFramePatchPointIdAddr and rewrite src value
if (len >= targetName.size() && destName.substr(0, targetName.size()) == targetName) {
LLVMSetOperand(targetInst, 0, LLVMConstInt(LLVMInt64Type(), id, 0));
}
}
targetInst = LLVMGetPreviousInstruction(targetInst);
}
}
void LLVMAssembler::RewritePatchPointIdOfStatePoint(LLVMValueRef instruction, uint64_t &callInsNum, uint64_t &funcNum)
{
if (LLVMIsACallInst(instruction) && !LLVMIsAIntrinsicInst(instruction)) {
const char attrName[] = "statepoint-id";
uint64_t id = (funcNum << 16) | callInsNum;
std::string patchId = std::to_string(id);
LLVMAttributeRef attr = LLVMCreateStringAttribute(LLVMGetGlobalContext(),
attrName, sizeof(attrName) - 1, patchId.c_str(), patchId.size());
LLVMAddCallSiteAttribute(instruction, LLVMAttributeFunctionIndex, attr);
callInsNum++;
if (LLVMWebKitJSCallConv == LLVMGetInstructionCallConv(instruction)) {
// 2 : 2 means patch id arguments order
LLVMSetOperand(instruction, 2, LLVMConstInt(LLVMInt64Type(), id, 0));
} else {
RewritePatchPointIdStoredOnThread(instruction, id);
}
}
}
void LLVMAssembler::FillPatchPointIDs()
{
uint64_t funcNum = 0;
LLVMValueRef func = LLVMGetFirstFunction(module_);
while (func) {
if (LLVMIsDeclaration(func)) {
func = LLVMGetNextFunction(func);
funcNum++;
continue;
}
uint64_t callInsNum = 0;
for (LLVMBasicBlockRef bb = LLVMGetFirstBasicBlock(func); bb; bb = LLVMGetNextBasicBlock(bb)) {
for (LLVMValueRef instruction = LLVMGetFirstInstruction(bb); instruction;
instruction = LLVMGetNextInstruction(instruction)) {
RewritePatchPointIdOfStatePoint(instruction, callInsNum, funcNum);
}
}
func = LLVMGetNextFunction(func);
funcNum++;
}
}
void LLVMAssembler::BuildAndRunPasses()
{
COMPILER_LOG(DEBUG) << "BuildAndRunPasses - ";
LLVMPassManagerBuilderRef pmBuilder = LLVMPassManagerBuilderCreate();
LLVMPassManagerBuilderSetOptLevel(pmBuilder, options_.OptLevel); // using O3 optimization level
LLVMPassManagerBuilderSetSizeLevel(pmBuilder, 0);
// pass manager creation:rs4gc pass is the only pass in modPass, other opt module-based pass are in modPass1
LLVMPassManagerRef funcPass = LLVMCreateFunctionPassManagerForModule(module_);
LLVMPassManagerRef modPass = LLVMCreatePassManager();
LLVMPassManagerRef modPass1 = LLVMCreatePassManager();
// add pass into pass managers
LLVMPassManagerBuilderPopulateFunctionPassManager(pmBuilder, funcPass);
llvm::unwrap(modPass)->add(llvm::createRewriteStatepointsForGCLegacyPass()); // rs4gc pass added
LLVMPassManagerBuilderPopulateModulePassManager(pmBuilder, modPass1);
// run module pass, function pass, module pass1
FillPatchPointIDs(); // add "statepoint-id" callsite attribute for statepoint ID rewrite and rewrite store value
LLVMRunPassManager(modPass, module_); // make sure rs4gc pass run first
LLVMInitializeFunctionPassManager(funcPass);
for (LLVMValueRef fn = LLVMGetFirstFunction(module_); fn; fn = LLVMGetNextFunction(fn)) {
LLVMRunFunctionPassManager(funcPass, fn);
}
LLVMFinalizeFunctionPassManager(funcPass);
LLVMRunPassManager(modPass1, module_);
LLVMPassManagerBuilderDispose(pmBuilder);
LLVMDisposePassManager(funcPass);
LLVMDisposePassManager(modPass);
LLVMDisposePassManager(modPass1);
COMPILER_LOG(DEBUG) << "BuildAndRunPasses + ";
}
LLVMAssembler::LLVMAssembler(LLVMModuleRef module)
: module_(module)
{
Initialize();
}
LLVMAssembler::~LLVMAssembler()
{
if (engine_ != nullptr) {
if (module_ != nullptr) {
char *error = nullptr;
LLVMRemoveModule(engine_, module_, &module_, &error);
if (error != nullptr) {
LLVMDisposeMessage(error);
}
}
LLVMDisposeExecutionEngine(engine_);
engine_ = nullptr;
}
module_ = nullptr;
error_ = nullptr;
}
void LLVMAssembler::Run()
{
char *error = nullptr;
#if ECMASCRIPT_ENABLE_COMPILER_LOG
char *info = LLVMPrintModuleToString(module_);
COMPILER_LOG(INFO) << "Current Module: " << info;
LLVMDisposeMessage(info);
#endif
LLVMPrintModuleToFile(module_, "stub.ll", &error);
LLVMVerifyModule(module_, LLVMAbortProcessAction, &error);
LLVMDisposeMessage(error);
UseRoundTripSectionMemoryManager();
if (!BuildMCJITEngine()) {
return;
}
BuildAndRunPasses();
LLVMPrintModuleToFile(module_, "opt_stub.ll", &error);
}
void LLVMAssembler::Initialize()
{
std::string triple(LLVMGetTarget(module_));
if (triple.compare("x86_64-unknown-linux-gnu") == 0) {
LLVMInitializeX86TargetInfo();
LLVMInitializeX86TargetMC();
LLVMInitializeX86Disassembler();
/* this method must be called, ohterwise "Target does not support MC emission" */
LLVMInitializeX86AsmPrinter();
LLVMInitializeX86AsmParser();
LLVMInitializeX86Target();
} else if (triple.compare("aarch64-unknown-linux-gnu") == 0) {
LLVMInitializeAArch64TargetInfo();
LLVMInitializeAArch64TargetMC();
LLVMInitializeAArch64Disassembler();
LLVMInitializeAArch64AsmPrinter();
LLVMInitializeAArch64AsmParser();
LLVMInitializeAArch64Target();
} else if (triple.compare("arm-unknown-linux-gnu") == 0) {
LLVMInitializeARMTargetInfo();
LLVMInitializeARMTargetMC();
LLVMInitializeARMDisassembler();
LLVMInitializeARMAsmPrinter();
LLVMInitializeARMAsmParser();
LLVMInitializeARMTarget();
} else {
UNREACHABLE();
}
llvm::linkAllBuiltinGCs();
LLVMInitializeMCJITCompilerOptions(&options_, sizeof(options_));
options_.OptLevel = 3; // opt level 2
// Just ensure that this field still exists.
#if ECMASCRIPT_ENABLE_INTERPRETER_ASM
// tmp for interpreter stub
options_.NoFramePointerElim = false;
#else
options_.NoFramePointerElim = true;
#endif
options_.CodeModel = LLVMCodeModelSmall;
}
#if ECMASCRIPT_ENABLE_COMPILER_LOG
static const char *SymbolLookupCallback([[maybe_unused]] void *disInfo, [[maybe_unused]] uint64_t referenceValue,
uint64_t *referenceType, [[maybe_unused]]uint64_t referencePC,
[[maybe_unused]] const char **referenceName)
{
*referenceType = LLVMDisassembler_ReferenceType_InOut_None;
return nullptr;
}
#endif
void LLVMAssembler::Disassemble(std::map<uint64_t, std::string> addr2name) const
{
#if ECMASCRIPT_ENABLE_COMPILER_LOG
LLVMDisasmContextRef dcr = LLVMCreateDisasm(LLVMGetTarget(module_), nullptr, 0, nullptr, SymbolLookupCallback);
std::cout << "========================================================================" << std::endl;
for (auto it : codeInfo_.GetCodeInfo()) {
uint8_t *byteSp;
uintptr_t numBytes;
byteSp = it.first;
numBytes = it.second;
std::cout << " byteSp:" << std::hex << reinterpret_cast<std::uintptr_t>(byteSp) << " numBytes:0x" << numBytes
<< std::endl;
unsigned pc = 0;
const char outStringSize = 100;
char outString[outStringSize];
while (numBytes != 0) {
size_t InstSize = LLVMDisasmInstruction(dcr, byteSp, numBytes, pc, outString, outStringSize);
if (InstSize == 0) {
std::cerr.fill('0');
std::cerr.width(8); // 8:fixed hex print width
std::cerr << std::hex << pc << ":";
std::cerr.width(8); // 8:fixed hex print width
std::cerr << std::hex << *reinterpret_cast<uint32_t *>(byteSp) << "maybe constant" << std::endl;
pc += 4; // 4 pc length
byteSp += 4; // 4 sp offset
numBytes -= 4; // 4 num bytes
}
uint64_t addr = reinterpret_cast<uint64_t>(byteSp);
if (addr2name.find(addr) != addr2name.end()) {
std::cout << addr2name[addr].c_str() << ":" << std::endl;
}
std::cerr.fill('0');
std::cerr.width(8); // 8:fixed hex print width
std::cerr << std::hex << pc << ":";
std::cerr.width(8); // 8:fixed hex print width
std::cerr << std::hex << *reinterpret_cast<uint32_t *>(byteSp) << " " << outString << std::endl;
pc += InstSize;
byteSp += InstSize;
numBytes -= InstSize;
}
}
std::cout << "========================================================================" << std::endl;
LLVMDisasmDispose(dcr);
#endif
}
} // namespace panda::ecmascript::kungfu