diff options
Diffstat (limited to 'lib/ExecutionEngine')
22 files changed, 1999 insertions, 1241 deletions
diff --git a/lib/ExecutionEngine/CMakeLists.txt b/lib/ExecutionEngine/CMakeLists.txt index cb11bfe..3102c7b 100644 --- a/lib/ExecutionEngine/CMakeLists.txt +++ b/lib/ExecutionEngine/CMakeLists.txt @@ -3,6 +3,7 @@ add_llvm_library(LLVMExecutionEngine ExecutionEngine.cpp ExecutionEngineBindings.cpp + RTDyldMemoryManager.cpp TargetSelect.cpp ) diff --git a/lib/ExecutionEngine/ExecutionEngine.cpp b/lib/ExecutionEngine/ExecutionEngine.cpp index 3d59d25..c463e9f 100644 --- a/lib/ExecutionEngine/ExecutionEngine.cpp +++ b/lib/ExecutionEngine/ExecutionEngine.cpp @@ -14,6 +14,7 @@ #define DEBUG_TYPE "jit" #include "llvm/ExecutionEngine/ExecutionEngine.h" +#include "llvm/ExecutionEngine/JITMemoryManager.h" #include "llvm/ADT/SmallString.h" #include "llvm/ADT/Statistic.h" #include "llvm/ExecutionEngine/GenericValue.h" @@ -47,7 +48,7 @@ ExecutionEngine *(*ExecutionEngine::JITCtor)( ExecutionEngine *(*ExecutionEngine::MCJITCtor)( Module *M, std::string *ErrorStr, - JITMemoryManager *JMM, + RTDyldMemoryManager *MCJMM, bool GVsWithCode, TargetMachine *TM) = 0; ExecutionEngine *(*ExecutionEngine::InterpCtor)(Module *M, @@ -117,7 +118,7 @@ char *ExecutionEngine::getMemoryForGV(const GlobalVariable *GV) { } bool ExecutionEngine::removeModule(Module *M) { - for(SmallVector<Module *, 1>::iterator I = Modules.begin(), + for(SmallVectorImpl<Module *>::iterator I = Modules.begin(), E = Modules.end(); I != E; ++I) { Module *Found = *I; if (Found == M) { @@ -455,10 +456,12 @@ ExecutionEngine *EngineBuilder::create(TargetMachine *TM) { if (sys::DynamicLibrary::LoadLibraryPermanently(0, ErrorStr)) return 0; + assert(!(JMM && MCJMM)); + // If the user specified a memory manager but didn't specify which engine to // create, we assume they only want the JIT, and we fail if they only want // the interpreter. - if (JMM) { + if (JMM || MCJMM) { if (WhichEngine & EngineKind::JIT) WhichEngine = EngineKind::JIT; else { @@ -467,6 +470,14 @@ ExecutionEngine *EngineBuilder::create(TargetMachine *TM) { return 0; } } + + if (MCJMM && ! UseMCJIT) { + if (ErrorStr) + *ErrorStr = + "Cannot create a legacy JIT with a runtime dyld memory " + "manager."; + return 0; + } // Unless the interpreter was explicitly selected or the JIT is not linked, // try making a JIT. @@ -480,7 +491,7 @@ ExecutionEngine *EngineBuilder::create(TargetMachine *TM) { if (UseMCJIT && ExecutionEngine::MCJITCtor) { ExecutionEngine *EE = - ExecutionEngine::MCJITCtor(M, ErrorStr, JMM, + ExecutionEngine::MCJITCtor(M, ErrorStr, MCJMM ? MCJMM : JMM, AllocateGVsWithCode, TheTM.take()); if (EE) return EE; } else if (ExecutionEngine::JITCtor) { @@ -535,6 +546,8 @@ GenericValue ExecutionEngine::getConstantValue(const Constant *C) { if (isa<UndefValue>(C)) { GenericValue Result; switch (C->getType()->getTypeID()) { + default: + break; case Type::IntegerTyID: case Type::X86_FP80TyID: case Type::FP128TyID: @@ -543,7 +556,16 @@ GenericValue ExecutionEngine::getConstantValue(const Constant *C) { // with the correct bit width. Result.IntVal = APInt(C->getType()->getPrimitiveSizeInBits(), 0); break; - default: + case Type::VectorTyID: + // if the whole vector is 'undef' just reserve memory for the value. + const VectorType* VTy = dyn_cast<VectorType>(C->getType()); + const Type *ElemTy = VTy->getElementType(); + unsigned int elemNum = VTy->getNumElements(); + Result.AggregateVal.resize(elemNum); + if (ElemTy->isIntegerTy()) + for (unsigned int i = 0; i < elemNum; ++i) + Result.AggregateVal[i].IntVal = + APInt(ElemTy->getPrimitiveSizeInBits(), 0); break; } return Result; @@ -825,6 +847,101 @@ GenericValue ExecutionEngine::getConstantValue(const Constant *C) { else llvm_unreachable("Unknown constant pointer type!"); break; + case Type::VectorTyID: { + unsigned elemNum; + Type* ElemTy; + const ConstantDataVector *CDV = dyn_cast<ConstantDataVector>(C); + const ConstantVector *CV = dyn_cast<ConstantVector>(C); + const ConstantAggregateZero *CAZ = dyn_cast<ConstantAggregateZero>(C); + + if (CDV) { + elemNum = CDV->getNumElements(); + ElemTy = CDV->getElementType(); + } else if (CV || CAZ) { + VectorType* VTy = dyn_cast<VectorType>(C->getType()); + elemNum = VTy->getNumElements(); + ElemTy = VTy->getElementType(); + } else { + llvm_unreachable("Unknown constant vector type!"); + } + + Result.AggregateVal.resize(elemNum); + // Check if vector holds floats. + if(ElemTy->isFloatTy()) { + if (CAZ) { + GenericValue floatZero; + floatZero.FloatVal = 0.f; + std::fill(Result.AggregateVal.begin(), Result.AggregateVal.end(), + floatZero); + break; + } + if(CV) { + for (unsigned i = 0; i < elemNum; ++i) + if (!isa<UndefValue>(CV->getOperand(i))) + Result.AggregateVal[i].FloatVal = cast<ConstantFP>( + CV->getOperand(i))->getValueAPF().convertToFloat(); + break; + } + if(CDV) + for (unsigned i = 0; i < elemNum; ++i) + Result.AggregateVal[i].FloatVal = CDV->getElementAsFloat(i); + + break; + } + // Check if vector holds doubles. + if (ElemTy->isDoubleTy()) { + if (CAZ) { + GenericValue doubleZero; + doubleZero.DoubleVal = 0.0; + std::fill(Result.AggregateVal.begin(), Result.AggregateVal.end(), + doubleZero); + break; + } + if(CV) { + for (unsigned i = 0; i < elemNum; ++i) + if (!isa<UndefValue>(CV->getOperand(i))) + Result.AggregateVal[i].DoubleVal = cast<ConstantFP>( + CV->getOperand(i))->getValueAPF().convertToDouble(); + break; + } + if(CDV) + for (unsigned i = 0; i < elemNum; ++i) + Result.AggregateVal[i].DoubleVal = CDV->getElementAsDouble(i); + + break; + } + // Check if vector holds integers. + if (ElemTy->isIntegerTy()) { + if (CAZ) { + GenericValue intZero; + intZero.IntVal = APInt(ElemTy->getScalarSizeInBits(), 0ull); + std::fill(Result.AggregateVal.begin(), Result.AggregateVal.end(), + intZero); + break; + } + if(CV) { + for (unsigned i = 0; i < elemNum; ++i) + if (!isa<UndefValue>(CV->getOperand(i))) + Result.AggregateVal[i].IntVal = cast<ConstantInt>( + CV->getOperand(i))->getValue(); + else { + Result.AggregateVal[i].IntVal = + APInt(CV->getOperand(i)->getType()->getPrimitiveSizeInBits(), 0); + } + break; + } + if(CDV) + for (unsigned i = 0; i < elemNum; ++i) + Result.AggregateVal[i].IntVal = APInt( + CDV->getElementType()->getPrimitiveSizeInBits(), + CDV->getElementAsInteger(i)); + + break; + } + llvm_unreachable("Unknown constant pointer type!"); + } + break; + default: SmallString<256> Msg; raw_svector_ostream OS(Msg); @@ -842,7 +959,7 @@ static void StoreIntToMemory(const APInt &IntVal, uint8_t *Dst, assert((IntVal.getBitWidth()+7)/8 >= StoreBytes && "Integer too small!"); const uint8_t *Src = (const uint8_t *)IntVal.getRawData(); - if (sys::isLittleEndianHost()) { + if (sys::IsLittleEndianHost) { // Little-endian host - the source is ordered from LSB to MSB. Order the // destination from LSB to MSB: Do a straight copy. memcpy(Dst, Src, StoreBytes); @@ -866,6 +983,9 @@ void ExecutionEngine::StoreValueToMemory(const GenericValue &Val, const unsigned StoreBytes = getDataLayout()->getTypeStoreSize(Ty); switch (Ty->getTypeID()) { + default: + dbgs() << "Cannot store value of type " << *Ty << "!\n"; + break; case Type::IntegerTyID: StoreIntToMemory(Val.IntVal, (uint8_t*)Ptr, StoreBytes); break; @@ -885,11 +1005,22 @@ void ExecutionEngine::StoreValueToMemory(const GenericValue &Val, *((PointerTy*)Ptr) = Val.PointerVal; break; - default: - dbgs() << "Cannot store value of type " << *Ty << "!\n"; + case Type::VectorTyID: + for (unsigned i = 0; i < Val.AggregateVal.size(); ++i) { + if (cast<VectorType>(Ty)->getElementType()->isDoubleTy()) + *(((double*)Ptr)+i) = Val.AggregateVal[i].DoubleVal; + if (cast<VectorType>(Ty)->getElementType()->isFloatTy()) + *(((float*)Ptr)+i) = Val.AggregateVal[i].FloatVal; + if (cast<VectorType>(Ty)->getElementType()->isIntegerTy()) { + unsigned numOfBytes =(Val.AggregateVal[i].IntVal.getBitWidth()+7)/8; + StoreIntToMemory(Val.AggregateVal[i].IntVal, + (uint8_t*)Ptr + numOfBytes*i, numOfBytes); + } + } + break; } - if (sys::isLittleEndianHost() != getDataLayout()->isLittleEndian()) + if (sys::IsLittleEndianHost != getDataLayout()->isLittleEndian()) // Host and target are different endian - reverse the stored bytes. std::reverse((uint8_t*)Ptr, StoreBytes + (uint8_t*)Ptr); } @@ -901,7 +1032,7 @@ static void LoadIntFromMemory(APInt &IntVal, uint8_t *Src, unsigned LoadBytes) { uint8_t *Dst = reinterpret_cast<uint8_t *>( const_cast<uint64_t *>(IntVal.getRawData())); - if (sys::isLittleEndianHost()) + if (sys::IsLittleEndianHost) // Little-endian host - the destination must be ordered from LSB to MSB. // The source is ordered from LSB to MSB: Do a straight copy. memcpy(Dst, Src, LoadBytes); @@ -951,6 +1082,31 @@ void ExecutionEngine::LoadValueFromMemory(GenericValue &Result, Result.IntVal = APInt(80, y); break; } + case Type::VectorTyID: { + const VectorType *VT = cast<VectorType>(Ty); + const Type *ElemT = VT->getElementType(); + const unsigned numElems = VT->getNumElements(); + if (ElemT->isFloatTy()) { + Result.AggregateVal.resize(numElems); + for (unsigned i = 0; i < numElems; ++i) + Result.AggregateVal[i].FloatVal = *((float*)Ptr+i); + } + if (ElemT->isDoubleTy()) { + Result.AggregateVal.resize(numElems); + for (unsigned i = 0; i < numElems; ++i) + Result.AggregateVal[i].DoubleVal = *((double*)Ptr+i); + } + if (ElemT->isIntegerTy()) { + GenericValue intZero; + const unsigned elemBitWidth = cast<IntegerType>(ElemT)->getBitWidth(); + intZero.IntVal = APInt(elemBitWidth, 0); + Result.AggregateVal.resize(numElems, intZero); + for (unsigned i = 0; i < numElems; ++i) + LoadIntFromMemory(Result.AggregateVal[i].IntVal, + (uint8_t*)Ptr+((elemBitWidth+7)/8)*i, (elemBitWidth+7)/8); + } + break; + } default: SmallString<256> Msg; raw_svector_ostream OS(Msg); diff --git a/lib/ExecutionEngine/ExecutionEngineBindings.cpp b/lib/ExecutionEngine/ExecutionEngineBindings.cpp index f4e8246..88e73bf 100644 --- a/lib/ExecutionEngine/ExecutionEngineBindings.cpp +++ b/lib/ExecutionEngine/ExecutionEngineBindings.cpp @@ -15,11 +15,34 @@ #include "llvm-c/ExecutionEngine.h" #include "llvm/ExecutionEngine/ExecutionEngine.h" #include "llvm/ExecutionEngine/GenericValue.h" +#include "llvm/ExecutionEngine/RTDyldMemoryManager.h" +#include "llvm/IR/DerivedTypes.h" +#include "llvm/IR/Module.h" #include "llvm/Support/ErrorHandling.h" #include <cstring> using namespace llvm; +// Wrapping the C bindings types. +DEFINE_SIMPLE_CONVERSION_FUNCTIONS(GenericValue, LLVMGenericValueRef) + +inline DataLayout *unwrap(LLVMTargetDataRef P) { + return reinterpret_cast<DataLayout*>(P); +} + +inline LLVMTargetDataRef wrap(const DataLayout *P) { + return reinterpret_cast<LLVMTargetDataRef>(const_cast<DataLayout*>(P)); +} + +inline TargetLibraryInfo *unwrap(LLVMTargetLibraryInfoRef P) { + return reinterpret_cast<TargetLibraryInfo*>(P); +} + +inline LLVMTargetLibraryInfoRef wrap(const TargetLibraryInfo *P) { + TargetLibraryInfo *X = const_cast<TargetLibraryInfo*>(P); + return reinterpret_cast<LLVMTargetLibraryInfoRef>(X); +} + /*===-- Operations on generic values --------------------------------------===*/ LLVMGenericValueRef LLVMCreateGenericValueOfInt(LLVMTypeRef Ty, @@ -132,6 +155,59 @@ LLVMBool LLVMCreateJITCompilerForModule(LLVMExecutionEngineRef *OutJIT, return 1; } +void LLVMInitializeMCJITCompilerOptions(LLVMMCJITCompilerOptions *PassedOptions, + size_t SizeOfPassedOptions) { + LLVMMCJITCompilerOptions options; + memset(&options, 0, sizeof(options)); // Most fields are zero by default. + options.CodeModel = LLVMCodeModelJITDefault; + + memcpy(PassedOptions, &options, + std::min(sizeof(options), SizeOfPassedOptions)); +} + +LLVMBool LLVMCreateMCJITCompilerForModule( + LLVMExecutionEngineRef *OutJIT, LLVMModuleRef M, + LLVMMCJITCompilerOptions *PassedOptions, size_t SizeOfPassedOptions, + char **OutError) { + LLVMMCJITCompilerOptions options; + // If the user passed a larger sized options struct, then they were compiled + // against a newer LLVM. Tell them that something is wrong. + if (SizeOfPassedOptions > sizeof(options)) { + *OutError = strdup( + "Refusing to use options struct that is larger than my own; assuming " + "LLVM library mismatch."); + return 1; + } + + // Defend against the user having an old version of the API by ensuring that + // any fields they didn't see are cleared. We must defend against fields being + // set to the bitwise equivalent of zero, and assume that this means "do the + // default" as if that option hadn't been available. + LLVMInitializeMCJITCompilerOptions(&options, sizeof(options)); + memcpy(&options, PassedOptions, SizeOfPassedOptions); + + TargetOptions targetOptions; + targetOptions.NoFramePointerElim = options.NoFramePointerElim; + targetOptions.EnableFastISel = options.EnableFastISel; + + std::string Error; + EngineBuilder builder(unwrap(M)); + builder.setEngineKind(EngineKind::JIT) + .setErrorStr(&Error) + .setUseMCJIT(true) + .setOptLevel((CodeGenOpt::Level)options.OptLevel) + .setCodeModel(unwrap(options.CodeModel)) + .setTargetOptions(targetOptions); + if (options.MCJMM) + builder.setMCJITMemoryManager(unwrap(options.MCJMM)); + if (ExecutionEngine *JIT = builder.create()) { + *OutJIT = wrap(JIT); + return 0; + } + *OutError = strdup(Error.c_str()); + return 1; +} + LLVMBool LLVMCreateExecutionEngine(LLVMExecutionEngineRef *OutEE, LLVMModuleProviderRef MP, char **OutError) { @@ -176,6 +252,8 @@ void LLVMRunStaticDestructors(LLVMExecutionEngineRef EE) { int LLVMRunFunctionAsMain(LLVMExecutionEngineRef EE, LLVMValueRef F, unsigned ArgC, const char * const *ArgV, const char * const *EnvP) { + unwrap(EE)->finalizeObject(); + std::vector<std::string> ArgVec; for (unsigned I = 0; I != ArgC; ++I) ArgVec.push_back(ArgV[I]); @@ -186,6 +264,8 @@ int LLVMRunFunctionAsMain(LLVMExecutionEngineRef EE, LLVMValueRef F, LLVMGenericValueRef LLVMRunFunction(LLVMExecutionEngineRef EE, LLVMValueRef F, unsigned NumArgs, LLVMGenericValueRef *Args) { + unwrap(EE)->finalizeObject(); + std::vector<GenericValue> ArgVec; ArgVec.reserve(NumArgs); for (unsigned I = 0; I != NumArgs; ++I) @@ -234,7 +314,8 @@ LLVMBool LLVMFindFunction(LLVMExecutionEngineRef EE, const char *Name, return 1; } -void *LLVMRecompileAndRelinkFunction(LLVMExecutionEngineRef EE, LLVMValueRef Fn) { +void *LLVMRecompileAndRelinkFunction(LLVMExecutionEngineRef EE, + LLVMValueRef Fn) { return unwrap(EE)->recompileAndRelinkFunction(unwrap<Function>(Fn)); } @@ -248,5 +329,114 @@ void LLVMAddGlobalMapping(LLVMExecutionEngineRef EE, LLVMValueRef Global, } void *LLVMGetPointerToGlobal(LLVMExecutionEngineRef EE, LLVMValueRef Global) { + unwrap(EE)->finalizeObject(); + return unwrap(EE)->getPointerToGlobal(unwrap<GlobalValue>(Global)); } + +/*===-- Operations on memory managers -------------------------------------===*/ + +namespace { + +struct SimpleBindingMMFunctions { + uint8_t *(*AllocateCodeSection)(void *Opaque, + uintptr_t Size, unsigned Alignment, + unsigned SectionID); + uint8_t *(*AllocateDataSection)(void *Opaque, + uintptr_t Size, unsigned Alignment, + unsigned SectionID, LLVMBool IsReadOnly); + LLVMBool (*FinalizeMemory)(void *Opaque, char **ErrMsg); + void (*Destroy)(void *Opaque); +}; + +class SimpleBindingMemoryManager : public RTDyldMemoryManager { +public: + SimpleBindingMemoryManager(const SimpleBindingMMFunctions& Functions, + void *Opaque); + virtual ~SimpleBindingMemoryManager(); + + virtual uint8_t *allocateCodeSection(uintptr_t Size, unsigned Alignment, + unsigned SectionID); + + virtual uint8_t *allocateDataSection(uintptr_t Size, unsigned Alignment, + unsigned SectionID, + bool isReadOnly); + + virtual bool finalizeMemory(std::string *ErrMsg); + +private: + SimpleBindingMMFunctions Functions; + void *Opaque; +}; + +SimpleBindingMemoryManager::SimpleBindingMemoryManager( + const SimpleBindingMMFunctions& Functions, + void *Opaque) + : Functions(Functions), Opaque(Opaque) { + assert(Functions.AllocateCodeSection && + "No AllocateCodeSection function provided!"); + assert(Functions.AllocateDataSection && + "No AllocateDataSection function provided!"); + assert(Functions.FinalizeMemory && + "No FinalizeMemory function provided!"); + assert(Functions.Destroy && + "No Destroy function provided!"); +} + +SimpleBindingMemoryManager::~SimpleBindingMemoryManager() { + Functions.Destroy(Opaque); +} + +uint8_t *SimpleBindingMemoryManager::allocateCodeSection( + uintptr_t Size, unsigned Alignment, unsigned SectionID) { + return Functions.AllocateCodeSection(Opaque, Size, Alignment, SectionID); +} + +uint8_t *SimpleBindingMemoryManager::allocateDataSection( + uintptr_t Size, unsigned Alignment, unsigned SectionID, bool isReadOnly) { + return Functions.AllocateDataSection(Opaque, Size, Alignment, SectionID, + isReadOnly); +} + +bool SimpleBindingMemoryManager::finalizeMemory(std::string *ErrMsg) { + char *errMsgCString = 0; + bool result = Functions.FinalizeMemory(Opaque, &errMsgCString); + assert((result || !errMsgCString) && + "Did not expect an error message if FinalizeMemory succeeded"); + if (errMsgCString) { + if (ErrMsg) + *ErrMsg = errMsgCString; + free(errMsgCString); + } + return result; +} + +} // anonymous namespace + +LLVMMCJITMemoryManagerRef LLVMCreateSimpleMCJITMemoryManager( + void *Opaque, + uint8_t *(*AllocateCodeSection)(void *Opaque, + uintptr_t Size, unsigned Alignment, + unsigned SectionID), + uint8_t *(*AllocateDataSection)(void *Opaque, + uintptr_t Size, unsigned Alignment, + unsigned SectionID, LLVMBool IsReadOnly), + LLVMBool (*FinalizeMemory)(void *Opaque, char **ErrMsg), + void (*Destroy)(void *Opaque)) { + + if (!AllocateCodeSection || !AllocateDataSection || !FinalizeMemory || + !Destroy) + return NULL; + + SimpleBindingMMFunctions functions; + functions.AllocateCodeSection = AllocateCodeSection; + functions.AllocateDataSection = AllocateDataSection; + functions.FinalizeMemory = FinalizeMemory; + functions.Destroy = Destroy; + return wrap(new SimpleBindingMemoryManager(functions, Opaque)); +} + +void LLVMDisposeMCJITMemoryManager(LLVMMCJITMemoryManagerRef MM) { + delete unwrap(MM); +} + diff --git a/lib/ExecutionEngine/Interpreter/Execution.cpp b/lib/ExecutionEngine/Interpreter/Execution.cpp index ec4f7f6..fc3d579 100644 --- a/lib/ExecutionEngine/Interpreter/Execution.cpp +++ b/lib/ExecutionEngine/Interpreter/Execution.cpp @@ -114,6 +114,15 @@ static void executeFRemInst(GenericValue &Dest, GenericValue Src1, Dest.IntVal = APInt(1,Src1.IntVal.OP(Src2.IntVal)); \ break; +#define IMPLEMENT_VECTOR_INTEGER_ICMP(OP, TY) \ + case Type::VectorTyID: { \ + assert(Src1.AggregateVal.size() == Src2.AggregateVal.size()); \ + Dest.AggregateVal.resize( Src1.AggregateVal.size() ); \ + for( uint32_t _i=0;_i<Src1.AggregateVal.size();_i++) \ + Dest.AggregateVal[_i].IntVal = APInt(1, \ + Src1.AggregateVal[_i].IntVal.OP(Src2.AggregateVal[_i].IntVal));\ + } break; + // Handle pointers specially because they must be compared with only as much // width as the host has. We _do not_ want to be comparing 64 bit values when // running on a 32-bit target, otherwise the upper 32 bits might mess up @@ -129,6 +138,7 @@ static GenericValue executeICMP_EQ(GenericValue Src1, GenericValue Src2, GenericValue Dest; switch (Ty->getTypeID()) { IMPLEMENT_INTEGER_ICMP(eq,Ty); + IMPLEMENT_VECTOR_INTEGER_ICMP(eq,Ty); IMPLEMENT_POINTER_ICMP(==); default: dbgs() << "Unhandled type for ICMP_EQ predicate: " << *Ty << "\n"; @@ -142,6 +152,7 @@ static GenericValue executeICMP_NE(GenericValue Src1, GenericValue Src2, GenericValue Dest; switch (Ty->getTypeID()) { IMPLEMENT_INTEGER_ICMP(ne,Ty); + IMPLEMENT_VECTOR_INTEGER_ICMP(ne,Ty); IMPLEMENT_POINTER_ICMP(!=); default: dbgs() << "Unhandled type for ICMP_NE predicate: " << *Ty << "\n"; @@ -155,6 +166,7 @@ static GenericValue executeICMP_ULT(GenericValue Src1, GenericValue Src2, GenericValue Dest; switch (Ty->getTypeID()) { IMPLEMENT_INTEGER_ICMP(ult,Ty); + IMPLEMENT_VECTOR_INTEGER_ICMP(ult,Ty); IMPLEMENT_POINTER_ICMP(<); default: dbgs() << "Unhandled type for ICMP_ULT predicate: " << *Ty << "\n"; @@ -168,6 +180,7 @@ static GenericValue executeICMP_SLT(GenericValue Src1, GenericValue Src2, GenericValue Dest; switch (Ty->getTypeID()) { IMPLEMENT_INTEGER_ICMP(slt,Ty); + IMPLEMENT_VECTOR_INTEGER_ICMP(slt,Ty); IMPLEMENT_POINTER_ICMP(<); default: dbgs() << "Unhandled type for ICMP_SLT predicate: " << *Ty << "\n"; @@ -181,6 +194,7 @@ static GenericValue executeICMP_UGT(GenericValue Src1, GenericValue Src2, GenericValue Dest; switch (Ty->getTypeID()) { IMPLEMENT_INTEGER_ICMP(ugt,Ty); + IMPLEMENT_VECTOR_INTEGER_ICMP(ugt,Ty); IMPLEMENT_POINTER_ICMP(>); default: dbgs() << "Unhandled type for ICMP_UGT predicate: " << *Ty << "\n"; @@ -194,6 +208,7 @@ static GenericValue executeICMP_SGT(GenericValue Src1, GenericValue Src2, GenericValue Dest; switch (Ty->getTypeID()) { IMPLEMENT_INTEGER_ICMP(sgt,Ty); + IMPLEMENT_VECTOR_INTEGER_ICMP(sgt,Ty); IMPLEMENT_POINTER_ICMP(>); default: dbgs() << "Unhandled type for ICMP_SGT predicate: " << *Ty << "\n"; @@ -207,6 +222,7 @@ static GenericValue executeICMP_ULE(GenericValue Src1, GenericValue Src2, GenericValue Dest; switch (Ty->getTypeID()) { IMPLEMENT_INTEGER_ICMP(ule,Ty); + IMPLEMENT_VECTOR_INTEGER_ICMP(ule,Ty); IMPLEMENT_POINTER_ICMP(<=); default: dbgs() << "Unhandled type for ICMP_ULE predicate: " << *Ty << "\n"; @@ -220,6 +236,7 @@ static GenericValue executeICMP_SLE(GenericValue Src1, GenericValue Src2, GenericValue Dest; switch (Ty->getTypeID()) { IMPLEMENT_INTEGER_ICMP(sle,Ty); + IMPLEMENT_VECTOR_INTEGER_ICMP(sle,Ty); IMPLEMENT_POINTER_ICMP(<=); default: dbgs() << "Unhandled type for ICMP_SLE predicate: " << *Ty << "\n"; @@ -233,6 +250,7 @@ static GenericValue executeICMP_UGE(GenericValue Src1, GenericValue Src2, GenericValue Dest; switch (Ty->getTypeID()) { IMPLEMENT_INTEGER_ICMP(uge,Ty); + IMPLEMENT_VECTOR_INTEGER_ICMP(uge,Ty); IMPLEMENT_POINTER_ICMP(>=); default: dbgs() << "Unhandled type for ICMP_UGE predicate: " << *Ty << "\n"; @@ -246,6 +264,7 @@ static GenericValue executeICMP_SGE(GenericValue Src1, GenericValue Src2, GenericValue Dest; switch (Ty->getTypeID()) { IMPLEMENT_INTEGER_ICMP(sge,Ty); + IMPLEMENT_VECTOR_INTEGER_ICMP(sge,Ty); IMPLEMENT_POINTER_ICMP(>=); default: dbgs() << "Unhandled type for ICMP_SGE predicate: " << *Ty << "\n"; @@ -285,12 +304,29 @@ void Interpreter::visitICmpInst(ICmpInst &I) { Dest.IntVal = APInt(1,Src1.TY##Val OP Src2.TY##Val); \ break +#define IMPLEMENT_VECTOR_FCMP_T(OP, TY) \ + assert(Src1.AggregateVal.size() == Src2.AggregateVal.size()); \ + Dest.AggregateVal.resize( Src1.AggregateVal.size() ); \ + for( uint32_t _i=0;_i<Src1.AggregateVal.size();_i++) \ + Dest.AggregateVal[_i].IntVal = APInt(1, \ + Src1.AggregateVal[_i].TY##Val OP Src2.AggregateVal[_i].TY##Val);\ + break; + +#define IMPLEMENT_VECTOR_FCMP(OP) \ + case Type::VectorTyID: \ + if(dyn_cast<VectorType>(Ty)->getElementType()->isFloatTy()) { \ + IMPLEMENT_VECTOR_FCMP_T(OP, Float); \ + } else { \ + IMPLEMENT_VECTOR_FCMP_T(OP, Double); \ + } + static GenericValue executeFCMP_OEQ(GenericValue Src1, GenericValue Src2, Type *Ty) { GenericValue Dest; switch (Ty->getTypeID()) { IMPLEMENT_FCMP(==, Float); IMPLEMENT_FCMP(==, Double); + IMPLEMENT_VECTOR_FCMP(==); default: dbgs() << "Unhandled type for FCmp EQ instruction: " << *Ty << "\n"; llvm_unreachable(0); @@ -298,17 +334,65 @@ static GenericValue executeFCMP_OEQ(GenericValue Src1, GenericValue Src2, return Dest; } +#define IMPLEMENT_SCALAR_NANS(TY, X,Y) \ + if (TY->isFloatTy()) { \ + if (X.FloatVal != X.FloatVal || Y.FloatVal != Y.FloatVal) { \ + Dest.IntVal = APInt(1,false); \ + return Dest; \ + } \ + } else { \ + if (X.DoubleVal != X.DoubleVal || Y.DoubleVal != Y.DoubleVal) { \ + Dest.IntVal = APInt(1,false); \ + return Dest; \ + } \ + } + +#define MASK_VECTOR_NANS_T(X,Y, TZ, FLAG) \ + assert(X.AggregateVal.size() == Y.AggregateVal.size()); \ + Dest.AggregateVal.resize( X.AggregateVal.size() ); \ + for( uint32_t _i=0;_i<X.AggregateVal.size();_i++) { \ + if (X.AggregateVal[_i].TZ##Val != X.AggregateVal[_i].TZ##Val || \ + Y.AggregateVal[_i].TZ##Val != Y.AggregateVal[_i].TZ##Val) \ + Dest.AggregateVal[_i].IntVal = APInt(1,FLAG); \ + else { \ + Dest.AggregateVal[_i].IntVal = APInt(1,!FLAG); \ + } \ + } + +#define MASK_VECTOR_NANS(TY, X,Y, FLAG) \ + if (TY->isVectorTy()) { \ + if (dyn_cast<VectorType>(TY)->getElementType()->isFloatTy()) { \ + MASK_VECTOR_NANS_T(X, Y, Float, FLAG) \ + } else { \ + MASK_VECTOR_NANS_T(X, Y, Double, FLAG) \ + } \ + } \ + + + static GenericValue executeFCMP_ONE(GenericValue Src1, GenericValue Src2, - Type *Ty) { + Type *Ty) +{ GenericValue Dest; + // if input is scalar value and Src1 or Src2 is NaN return false + IMPLEMENT_SCALAR_NANS(Ty, Src1, Src2) + // if vector input detect NaNs and fill mask + MASK_VECTOR_NANS(Ty, Src1, Src2, false) + GenericValue DestMask = Dest; switch (Ty->getTypeID()) { IMPLEMENT_FCMP(!=, Float); IMPLEMENT_FCMP(!=, Double); - - default: - dbgs() << "Unhandled type for FCmp NE instruction: " << *Ty << "\n"; - llvm_unreachable(0); + IMPLEMENT_VECTOR_FCMP(!=); + default: + dbgs() << "Unhandled type for FCmp NE instruction: " << *Ty << "\n"; + llvm_unreachable(0); } + // in vector case mask out NaN elements + if (Ty->isVectorTy()) + for( size_t _i=0; _i<Src1.AggregateVal.size(); _i++) + if (DestMask.AggregateVal[_i].IntVal == false) + Dest.AggregateVal[_i].IntVal = APInt(1,false); + return Dest; } @@ -318,6 +402,7 @@ static GenericValue executeFCMP_OLE(GenericValue Src1, GenericValue Src2, switch (Ty->getTypeID()) { IMPLEMENT_FCMP(<=, Float); IMPLEMENT_FCMP(<=, Double); + IMPLEMENT_VECTOR_FCMP(<=); default: dbgs() << "Unhandled type for FCmp LE instruction: " << *Ty << "\n"; llvm_unreachable(0); @@ -331,6 +416,7 @@ static GenericValue executeFCMP_OGE(GenericValue Src1, GenericValue Src2, switch (Ty->getTypeID()) { IMPLEMENT_FCMP(>=, Float); IMPLEMENT_FCMP(>=, Double); + IMPLEMENT_VECTOR_FCMP(>=); default: dbgs() << "Unhandled type for FCmp GE instruction: " << *Ty << "\n"; llvm_unreachable(0); @@ -344,6 +430,7 @@ static GenericValue executeFCMP_OLT(GenericValue Src1, GenericValue Src2, switch (Ty->getTypeID()) { IMPLEMENT_FCMP(<, Float); IMPLEMENT_FCMP(<, Double); + IMPLEMENT_VECTOR_FCMP(<); default: dbgs() << "Unhandled type for FCmp LT instruction: " << *Ty << "\n"; llvm_unreachable(0); @@ -357,6 +444,7 @@ static GenericValue executeFCMP_OGT(GenericValue Src1, GenericValue Src2, switch (Ty->getTypeID()) { IMPLEMENT_FCMP(>, Float); IMPLEMENT_FCMP(>, Double); + IMPLEMENT_VECTOR_FCMP(>); default: dbgs() << "Unhandled type for FCmp GT instruction: " << *Ty << "\n"; llvm_unreachable(0); @@ -375,18 +463,32 @@ static GenericValue executeFCMP_OGT(GenericValue Src1, GenericValue Src2, return Dest; \ } +#define IMPLEMENT_VECTOR_UNORDERED(TY, X,Y, _FUNC) \ + if (TY->isVectorTy()) { \ + GenericValue DestMask = Dest; \ + Dest = _FUNC(Src1, Src2, Ty); \ + for( size_t _i=0; _i<Src1.AggregateVal.size(); _i++) \ + if (DestMask.AggregateVal[_i].IntVal == true) \ + Dest.AggregateVal[_i].IntVal = APInt(1,true); \ + return Dest; \ + } static GenericValue executeFCMP_UEQ(GenericValue Src1, GenericValue Src2, Type *Ty) { GenericValue Dest; IMPLEMENT_UNORDERED(Ty, Src1, Src2) + MASK_VECTOR_NANS(Ty, Src1, Src2, true) + IMPLEMENT_VECTOR_UNORDERED(Ty, Src1, Src2, executeFCMP_OEQ) return executeFCMP_OEQ(Src1, Src2, Ty); + } static GenericValue executeFCMP_UNE(GenericValue Src1, GenericValue Src2, Type *Ty) { GenericValue Dest; IMPLEMENT_UNORDERED(Ty, Src1, Src2) + MASK_VECTOR_NANS(Ty, Src1, Src2, true) + IMPLEMENT_VECTOR_UNORDERED(Ty, Src1, Src2, executeFCMP_ONE) return executeFCMP_ONE(Src1, Src2, Ty); } @@ -394,6 +496,8 @@ static GenericValue executeFCMP_ULE(GenericValue Src1, GenericValue Src2, Type *Ty) { GenericValue Dest; IMPLEMENT_UNORDERED(Ty, Src1, Src2) + MASK_VECTOR_NANS(Ty, Src1, Src2, true) + IMPLEMENT_VECTOR_UNORDERED(Ty, Src1, Src2, executeFCMP_OLE) return executeFCMP_OLE(Src1, Src2, Ty); } @@ -401,6 +505,8 @@ static GenericValue executeFCMP_UGE(GenericValue Src1, GenericValue Src2, Type *Ty) { GenericValue Dest; IMPLEMENT_UNORDERED(Ty, Src1, Src2) + MASK_VECTOR_NANS(Ty, Src1, Src2, true) + IMPLEMENT_VECTOR_UNORDERED(Ty, Src1, Src2, executeFCMP_OGE) return executeFCMP_OGE(Src1, Src2, Ty); } @@ -408,6 +514,8 @@ static GenericValue executeFCMP_ULT(GenericValue Src1, GenericValue Src2, Type *Ty) { GenericValue Dest; IMPLEMENT_UNORDERED(Ty, Src1, Src2) + MASK_VECTOR_NANS(Ty, Src1, Src2, true) + IMPLEMENT_VECTOR_UNORDERED(Ty, Src1, Src2, executeFCMP_OLT) return executeFCMP_OLT(Src1, Src2, Ty); } @@ -415,33 +523,88 @@ static GenericValue executeFCMP_UGT(GenericValue Src1, GenericValue Src2, Type *Ty) { GenericValue Dest; IMPLEMENT_UNORDERED(Ty, Src1, Src2) + MASK_VECTOR_NANS(Ty, Src1, Src2, true) + IMPLEMENT_VECTOR_UNORDERED(Ty, Src1, Src2, executeFCMP_OGT) return executeFCMP_OGT(Src1, Src2, Ty); } static GenericValue executeFCMP_ORD(GenericValue Src1, GenericValue Src2, Type *Ty) { GenericValue Dest; - if (Ty->isFloatTy()) + if(Ty->isVectorTy()) { + assert(Src1.AggregateVal.size() == Src2.AggregateVal.size()); + Dest.AggregateVal.resize( Src1.AggregateVal.size() ); + if(dyn_cast<VectorType>(Ty)->getElementType()->isFloatTy()) { + for( size_t _i=0;_i<Src1.AggregateVal.size();_i++) + Dest.AggregateVal[_i].IntVal = APInt(1, + ( (Src1.AggregateVal[_i].FloatVal == + Src1.AggregateVal[_i].FloatVal) && + (Src2.AggregateVal[_i].FloatVal == + Src2.AggregateVal[_i].FloatVal))); + } else { + for( size_t _i=0;_i<Src1.AggregateVal.size();_i++) + Dest.AggregateVal[_i].IntVal = APInt(1, + ( (Src1.AggregateVal[_i].DoubleVal == + Src1.AggregateVal[_i].DoubleVal) && + (Src2.AggregateVal[_i].DoubleVal == + Src2.AggregateVal[_i].DoubleVal))); + } + } else if (Ty->isFloatTy()) Dest.IntVal = APInt(1,(Src1.FloatVal == Src1.FloatVal && Src2.FloatVal == Src2.FloatVal)); - else + else { Dest.IntVal = APInt(1,(Src1.DoubleVal == Src1.DoubleVal && Src2.DoubleVal == Src2.DoubleVal)); + } return Dest; } static GenericValue executeFCMP_UNO(GenericValue Src1, GenericValue Src2, Type *Ty) { GenericValue Dest; - if (Ty->isFloatTy()) + if(Ty->isVectorTy()) { + assert(Src1.AggregateVal.size() == Src2.AggregateVal.size()); + Dest.AggregateVal.resize( Src1.AggregateVal.size() ); + if(dyn_cast<VectorType>(Ty)->getElementType()->isFloatTy()) { + for( size_t _i=0;_i<Src1.AggregateVal.size();_i++) + Dest.AggregateVal[_i].IntVal = APInt(1, + ( (Src1.AggregateVal[_i].FloatVal != + Src1.AggregateVal[_i].FloatVal) || + (Src2.AggregateVal[_i].FloatVal != + Src2.AggregateVal[_i].FloatVal))); + } else { + for( size_t _i=0;_i<Src1.AggregateVal.size();_i++) + Dest.AggregateVal[_i].IntVal = APInt(1, + ( (Src1.AggregateVal[_i].DoubleVal != + Src1.AggregateVal[_i].DoubleVal) || + (Src2.AggregateVal[_i].DoubleVal != + Src2.AggregateVal[_i].DoubleVal))); + } + } else if (Ty->isFloatTy()) Dest.IntVal = APInt(1,(Src1.FloatVal != Src1.FloatVal || Src2.FloatVal != Src2.FloatVal)); - else + else { Dest.IntVal = APInt(1,(Src1.DoubleVal != Src1.DoubleVal || Src2.DoubleVal != Src2.DoubleVal)); + } return Dest; } +static GenericValue executeFCMP_BOOL(GenericValue Src1, GenericValue Src2, + const Type *Ty, const bool val) { + GenericValue Dest; + if(Ty->isVectorTy()) { + assert(Src1.AggregateVal.size() == Src2.AggregateVal.size()); + Dest.AggregateVal.resize( Src1.AggregateVal.size() ); + for( size_t _i=0; _i<Src1.AggregateVal.size(); _i++) + Dest.AggregateVal[_i].IntVal = APInt(1,val); + } else { + Dest.IntVal = APInt(1, val); + } + + return Dest; +} + void Interpreter::visitFCmpInst(FCmpInst &I) { ExecutionContext &SF = ECStack.back(); Type *Ty = I.getOperand(0)->getType(); @@ -450,8 +613,14 @@ void Interpreter::visitFCmpInst(FCmpInst &I) { GenericValue R; // Result switch (I.getPredicate()) { - case FCmpInst::FCMP_FALSE: R.IntVal = APInt(1,false); break; - case FCmpInst::FCMP_TRUE: R.IntVal = APInt(1,true); break; + default: + dbgs() << "Don't know how to handle this FCmp predicate!\n-->" << I; + llvm_unreachable(0); + break; + case FCmpInst::FCMP_FALSE: R = executeFCMP_BOOL(Src1, Src2, Ty, false); + break; + case FCmpInst::FCMP_TRUE: R = executeFCMP_BOOL(Src1, Src2, Ty, true); + break; case FCmpInst::FCMP_ORD: R = executeFCMP_ORD(Src1, Src2, Ty); break; case FCmpInst::FCMP_UNO: R = executeFCMP_UNO(Src1, Src2, Ty); break; case FCmpInst::FCMP_UEQ: R = executeFCMP_UEQ(Src1, Src2, Ty); break; @@ -466,9 +635,6 @@ void Interpreter::visitFCmpInst(FCmpInst &I) { case FCmpInst::FCMP_OLE: R = executeFCMP_OLE(Src1, Src2, Ty); break; case FCmpInst::FCMP_UGE: R = executeFCMP_UGE(Src1, Src2, Ty); break; case FCmpInst::FCMP_OGE: R = executeFCMP_OGE(Src1, Src2, Ty); break; - default: - dbgs() << "Don't know how to handle this FCmp predicate!\n-->" << I; - llvm_unreachable(0); } SetValue(&I, R, SF); @@ -502,16 +668,8 @@ static GenericValue executeCmpInst(unsigned predicate, GenericValue Src1, case FCmpInst::FCMP_ULE: return executeFCMP_ULE(Src1, Src2, Ty); case FCmpInst::FCMP_OGE: return executeFCMP_OGE(Src1, Src2, Ty); case FCmpInst::FCMP_UGE: return executeFCMP_UGE(Src1, Src2, Ty); - case FCmpInst::FCMP_FALSE: { - GenericValue Result; - Result.IntVal = APInt(1, false); - return Result; - } - case FCmpInst::FCMP_TRUE: { - GenericValue Result; - Result.IntVal = APInt(1, true); - return Result; - } + case FCmpInst::FCMP_FALSE: return executeFCMP_BOOL(Src1, Src2, Ty, false); + case FCmpInst::FCMP_TRUE: return executeFCMP_BOOL(Src1, Src2, Ty, true); default: dbgs() << "Unhandled Cmp predicate\n"; llvm_unreachable(0); @@ -525,27 +683,105 @@ void Interpreter::visitBinaryOperator(BinaryOperator &I) { GenericValue Src2 = getOperandValue(I.getOperand(1), SF); GenericValue R; // Result - switch (I.getOpcode()) { - case Instruction::Add: R.IntVal = Src1.IntVal + Src2.IntVal; break; - case Instruction::Sub: R.IntVal = Src1.IntVal - Src2.IntVal; break; - case Instruction::Mul: R.IntVal = Src1.IntVal * Src2.IntVal; break; - case Instruction::FAdd: executeFAddInst(R, Src1, Src2, Ty); break; - case Instruction::FSub: executeFSubInst(R, Src1, Src2, Ty); break; - case Instruction::FMul: executeFMulInst(R, Src1, Src2, Ty); break; - case Instruction::FDiv: executeFDivInst(R, Src1, Src2, Ty); break; - case Instruction::FRem: executeFRemInst(R, Src1, Src2, Ty); break; - case Instruction::UDiv: R.IntVal = Src1.IntVal.udiv(Src2.IntVal); break; - case Instruction::SDiv: R.IntVal = Src1.IntVal.sdiv(Src2.IntVal); break; - case Instruction::URem: R.IntVal = Src1.IntVal.urem(Src2.IntVal); break; - case Instruction::SRem: R.IntVal = Src1.IntVal.srem(Src2.IntVal); break; - case Instruction::And: R.IntVal = Src1.IntVal & Src2.IntVal; break; - case Instruction::Or: R.IntVal = Src1.IntVal | Src2.IntVal; break; - case Instruction::Xor: R.IntVal = Src1.IntVal ^ Src2.IntVal; break; - default: - dbgs() << "Don't know how to handle this binary operator!\n-->" << I; - llvm_unreachable(0); + // First process vector operation + if (Ty->isVectorTy()) { + assert(Src1.AggregateVal.size() == Src2.AggregateVal.size()); + R.AggregateVal.resize(Src1.AggregateVal.size()); + + // Macros to execute binary operation 'OP' over integer vectors +#define INTEGER_VECTOR_OPERATION(OP) \ + for (unsigned i = 0; i < R.AggregateVal.size(); ++i) \ + R.AggregateVal[i].IntVal = \ + Src1.AggregateVal[i].IntVal OP Src2.AggregateVal[i].IntVal; + + // Additional macros to execute binary operations udiv/sdiv/urem/srem since + // they have different notation. +#define INTEGER_VECTOR_FUNCTION(OP) \ + for (unsigned i = 0; i < R.AggregateVal.size(); ++i) \ + R.AggregateVal[i].IntVal = \ + Src1.AggregateVal[i].IntVal.OP(Src2.AggregateVal[i].IntVal); + + // Macros to execute binary operation 'OP' over floating point type TY + // (float or double) vectors +#define FLOAT_VECTOR_FUNCTION(OP, TY) \ + for (unsigned i = 0; i < R.AggregateVal.size(); ++i) \ + R.AggregateVal[i].TY = \ + Src1.AggregateVal[i].TY OP Src2.AggregateVal[i].TY; + + // Macros to choose appropriate TY: float or double and run operation + // execution +#define FLOAT_VECTOR_OP(OP) { \ + if (dyn_cast<VectorType>(Ty)->getElementType()->isFloatTy()) \ + FLOAT_VECTOR_FUNCTION(OP, FloatVal) \ + else { \ + if (dyn_cast<VectorType>(Ty)->getElementType()->isDoubleTy()) \ + FLOAT_VECTOR_FUNCTION(OP, DoubleVal) \ + else { \ + dbgs() << "Unhandled type for OP instruction: " << *Ty << "\n"; \ + llvm_unreachable(0); \ + } \ + } \ +} + + switch(I.getOpcode()){ + default: + dbgs() << "Don't know how to handle this binary operator!\n-->" << I; + llvm_unreachable(0); + break; + case Instruction::Add: INTEGER_VECTOR_OPERATION(+) break; + case Instruction::Sub: INTEGER_VECTOR_OPERATION(-) break; + case Instruction::Mul: INTEGER_VECTOR_OPERATION(*) break; + case Instruction::UDiv: INTEGER_VECTOR_FUNCTION(udiv) break; + case Instruction::SDiv: INTEGER_VECTOR_FUNCTION(sdiv) break; + case Instruction::URem: INTEGER_VECTOR_FUNCTION(urem) break; + case Instruction::SRem: INTEGER_VECTOR_FUNCTION(srem) break; + case Instruction::And: INTEGER_VECTOR_OPERATION(&) break; + case Instruction::Or: INTEGER_VECTOR_OPERATION(|) break; + case Instruction::Xor: INTEGER_VECTOR_OPERATION(^) break; + case Instruction::FAdd: FLOAT_VECTOR_OP(+) break; + case Instruction::FSub: FLOAT_VECTOR_OP(-) break; + case Instruction::FMul: FLOAT_VECTOR_OP(*) break; + case Instruction::FDiv: FLOAT_VECTOR_OP(/) break; + case Instruction::FRem: + if (dyn_cast<VectorType>(Ty)->getElementType()->isFloatTy()) + for (unsigned i = 0; i < R.AggregateVal.size(); ++i) + R.AggregateVal[i].FloatVal = + fmod(Src1.AggregateVal[i].FloatVal, Src2.AggregateVal[i].FloatVal); + else { + if (dyn_cast<VectorType>(Ty)->getElementType()->isDoubleTy()) + for (unsigned i = 0; i < R.AggregateVal.size(); ++i) + R.AggregateVal[i].DoubleVal = + fmod(Src1.AggregateVal[i].DoubleVal, Src2.AggregateVal[i].DoubleVal); + else { + dbgs() << "Unhandled type for Rem instruction: " << *Ty << "\n"; + llvm_unreachable(0); + } + } + break; + } + } else { + switch (I.getOpcode()) { + default: + dbgs() << "Don't know how to handle this binary operator!\n-->" << I; + llvm_unreachable(0); + break; + case Instruction::Add: R.IntVal = Src1.IntVal + Src2.IntVal; break; + case Instruction::Sub: R.IntVal = Src1.IntVal - Src2.IntVal; break; + case Instruction::Mul: R.IntVal = Src1.IntVal * Src2.IntVal; break; + case Instruction::FAdd: executeFAddInst(R, Src1, Src2, Ty); break; + case Instruction::FSub: executeFSubInst(R, Src1, Src2, Ty); break; + case Instruction::FMul: executeFMulInst(R, Src1, Src2, Ty); break; + case Instruction::FDiv: executeFDivInst(R, Src1, Src2, Ty); break; + case Instruction::FRem: executeFRemInst(R, Src1, Src2, Ty); break; + case Instruction::UDiv: R.IntVal = Src1.IntVal.udiv(Src2.IntVal); break; + case Instruction::SDiv: R.IntVal = Src1.IntVal.sdiv(Src2.IntVal); break; + case Instruction::URem: R.IntVal = Src1.IntVal.urem(Src2.IntVal); break; + case Instruction::SRem: R.IntVal = Src1.IntVal.srem(Src2.IntVal); break; + case Instruction::And: R.IntVal = Src1.IntVal & Src2.IntVal; break; + case Instruction::Or: R.IntVal = Src1.IntVal | Src2.IntVal; break; + case Instruction::Xor: R.IntVal = Src1.IntVal ^ Src2.IntVal; break; + } } - SetValue(&I, R, SF); } @@ -902,16 +1138,42 @@ void Interpreter::visitCallSite(CallSite CS) { callFunction((Function*)GVTOP(SRC), ArgVals); } +// auxilary function for shift operations +static unsigned getShiftAmount(uint64_t orgShiftAmount, + llvm::APInt valueToShift) { + unsigned valueWidth = valueToShift.getBitWidth(); + if (orgShiftAmount < (uint64_t)valueWidth) + return orgShiftAmount; + // according to the llvm documentation, if orgShiftAmount > valueWidth, + // the result is undfeined. but we do shift by this rule: + return (NextPowerOf2(valueWidth-1) - 1) & orgShiftAmount; +} + + void Interpreter::visitShl(BinaryOperator &I) { ExecutionContext &SF = ECStack.back(); GenericValue Src1 = getOperandValue(I.getOperand(0), SF); GenericValue Src2 = getOperandValue(I.getOperand(1), SF); GenericValue Dest; - if (Src2.IntVal.getZExtValue() < Src1.IntVal.getBitWidth()) - Dest.IntVal = Src1.IntVal.shl(Src2.IntVal.getZExtValue()); - else - Dest.IntVal = Src1.IntVal; - + const Type *Ty = I.getType(); + + if (Ty->isVectorTy()) { + uint32_t src1Size = uint32_t(Src1.AggregateVal.size()); + assert(src1Size == Src2.AggregateVal.size()); + for (unsigned i = 0; i < src1Size; i++) { + GenericValue Result; + uint64_t shiftAmount = Src2.AggregateVal[i].IntVal.getZExtValue(); + llvm::APInt valueToShift = Src1.AggregateVal[i].IntVal; + Result.IntVal = valueToShift.shl(getShiftAmount(shiftAmount, valueToShift)); + Dest.AggregateVal.push_back(Result); + } + } else { + // scalar + uint64_t shiftAmount = Src2.IntVal.getZExtValue(); + llvm::APInt valueToShift = Src1.IntVal; + Dest.IntVal = valueToShift.shl(getShiftAmount(shiftAmount, valueToShift)); + } + SetValue(&I, Dest, SF); } @@ -920,11 +1182,25 @@ void Interpreter::visitLShr(BinaryOperator &I) { GenericValue Src1 = getOperandValue(I.getOperand(0), SF); GenericValue Src2 = getOperandValue(I.getOperand(1), SF); GenericValue Dest; - if (Src2.IntVal.getZExtValue() < Src1.IntVal.getBitWidth()) - Dest.IntVal = Src1.IntVal.lshr(Src2.IntVal.getZExtValue()); - else - Dest.IntVal = Src1.IntVal; - + const Type *Ty = I.getType(); + + if (Ty->isVectorTy()) { + uint32_t src1Size = uint32_t(Src1.AggregateVal.size()); + assert(src1Size == Src2.AggregateVal.size()); + for (unsigned i = 0; i < src1Size; i++) { + GenericValue Result; + uint64_t shiftAmount = Src2.AggregateVal[i].IntVal.getZExtValue(); + llvm::APInt valueToShift = Src1.AggregateVal[i].IntVal; + Result.IntVal = valueToShift.lshr(getShiftAmount(shiftAmount, valueToShift)); + Dest.AggregateVal.push_back(Result); + } + } else { + // scalar + uint64_t shiftAmount = Src2.IntVal.getZExtValue(); + llvm::APInt valueToShift = Src1.IntVal; + Dest.IntVal = valueToShift.lshr(getShiftAmount(shiftAmount, valueToShift)); + } + SetValue(&I, Dest, SF); } @@ -933,110 +1209,273 @@ void Interpreter::visitAShr(BinaryOperator &I) { GenericValue Src1 = getOperandValue(I.getOperand(0), SF); GenericValue Src2 = getOperandValue(I.getOperand(1), SF); GenericValue Dest; - if (Src2.IntVal.getZExtValue() < Src1.IntVal.getBitWidth()) - Dest.IntVal = Src1.IntVal.ashr(Src2.IntVal.getZExtValue()); - else - Dest.IntVal = Src1.IntVal; - + const Type *Ty = I.getType(); + + if (Ty->isVectorTy()) { + size_t src1Size = Src1.AggregateVal.size(); + assert(src1Size == Src2.AggregateVal.size()); + for (unsigned i = 0; i < src1Size; i++) { + GenericValue Result; + uint64_t shiftAmount = Src2.AggregateVal[i].IntVal.getZExtValue(); + llvm::APInt valueToShift = Src1.AggregateVal[i].IntVal; + Result.IntVal = valueToShift.ashr(getShiftAmount(shiftAmount, valueToShift)); + Dest.AggregateVal.push_back(Result); + } + } else { + // scalar + uint64_t shiftAmount = Src2.IntVal.getZExtValue(); + llvm::APInt valueToShift = Src1.IntVal; + Dest.IntVal = valueToShift.ashr(getShiftAmount(shiftAmount, valueToShift)); + } + SetValue(&I, Dest, SF); } GenericValue Interpreter::executeTruncInst(Value *SrcVal, Type *DstTy, ExecutionContext &SF) { GenericValue Dest, Src = getOperandValue(SrcVal, SF); - IntegerType *DITy = cast<IntegerType>(DstTy); - unsigned DBitWidth = DITy->getBitWidth(); - Dest.IntVal = Src.IntVal.trunc(DBitWidth); + Type *SrcTy = SrcVal->getType(); + if (SrcTy->isVectorTy()) { + Type *DstVecTy = DstTy->getScalarType(); + unsigned DBitWidth = cast<IntegerType>(DstVecTy)->getBitWidth(); + unsigned NumElts = Src.AggregateVal.size(); + // the sizes of src and dst vectors must be equal + Dest.AggregateVal.resize(NumElts); + for (unsigned i = 0; i < NumElts; i++) + Dest.AggregateVal[i].IntVal = Src.AggregateVal[i].IntVal.trunc(DBitWidth); + } else { + IntegerType *DITy = cast<IntegerType>(DstTy); + unsigned DBitWidth = DITy->getBitWidth(); + Dest.IntVal = Src.IntVal.trunc(DBitWidth); + } return Dest; } GenericValue Interpreter::executeSExtInst(Value *SrcVal, Type *DstTy, ExecutionContext &SF) { + const Type *SrcTy = SrcVal->getType(); GenericValue Dest, Src = getOperandValue(SrcVal, SF); - IntegerType *DITy = cast<IntegerType>(DstTy); - unsigned DBitWidth = DITy->getBitWidth(); - Dest.IntVal = Src.IntVal.sext(DBitWidth); + if (SrcTy->isVectorTy()) { + const Type *DstVecTy = DstTy->getScalarType(); + unsigned DBitWidth = cast<IntegerType>(DstVecTy)->getBitWidth(); + unsigned size = Src.AggregateVal.size(); + // the sizes of src and dst vectors must be equal. + Dest.AggregateVal.resize(size); + for (unsigned i = 0; i < size; i++) + Dest.AggregateVal[i].IntVal = Src.AggregateVal[i].IntVal.sext(DBitWidth); + } else { + const IntegerType *DITy = cast<IntegerType>(DstTy); + unsigned DBitWidth = DITy->getBitWidth(); + Dest.IntVal = Src.IntVal.sext(DBitWidth); + } return Dest; } GenericValue Interpreter::executeZExtInst(Value *SrcVal, Type *DstTy, ExecutionContext &SF) { + const Type *SrcTy = SrcVal->getType(); GenericValue Dest, Src = getOperandValue(SrcVal, SF); - IntegerType *DITy = cast<IntegerType>(DstTy); - unsigned DBitWidth = DITy->getBitWidth(); - Dest.IntVal = Src.IntVal.zext(DBitWidth); + if (SrcTy->isVectorTy()) { + const Type *DstVecTy = DstTy->getScalarType(); + unsigned DBitWidth = cast<IntegerType>(DstVecTy)->getBitWidth(); + + unsigned size = Src.AggregateVal.size(); + // the sizes of src and dst vectors must be equal. + Dest.AggregateVal.resize(size); + for (unsigned i = 0; i < size; i++) + Dest.AggregateVal[i].IntVal = Src.AggregateVal[i].IntVal.zext(DBitWidth); + } else { + const IntegerType *DITy = cast<IntegerType>(DstTy); + unsigned DBitWidth = DITy->getBitWidth(); + Dest.IntVal = Src.IntVal.zext(DBitWidth); + } return Dest; } GenericValue Interpreter::executeFPTruncInst(Value *SrcVal, Type *DstTy, ExecutionContext &SF) { GenericValue Dest, Src = getOperandValue(SrcVal, SF); - assert(SrcVal->getType()->isDoubleTy() && DstTy->isFloatTy() && - "Invalid FPTrunc instruction"); - Dest.FloatVal = (float) Src.DoubleVal; + + if (SrcVal->getType()->getTypeID() == Type::VectorTyID) { + assert(SrcVal->getType()->getScalarType()->isDoubleTy() && + DstTy->getScalarType()->isFloatTy() && + "Invalid FPTrunc instruction"); + + unsigned size = Src.AggregateVal.size(); + // the sizes of src and dst vectors must be equal. + Dest.AggregateVal.resize(size); + for (unsigned i = 0; i < size; i++) + Dest.AggregateVal[i].FloatVal = (float)Src.AggregateVal[i].DoubleVal; + } else { + assert(SrcVal->getType()->isDoubleTy() && DstTy->isFloatTy() && + "Invalid FPTrunc instruction"); + Dest.FloatVal = (float)Src.DoubleVal; + } + return Dest; } GenericValue Interpreter::executeFPExtInst(Value *SrcVal, Type *DstTy, ExecutionContext &SF) { GenericValue Dest, Src = getOperandValue(SrcVal, SF); - assert(SrcVal->getType()->isFloatTy() && DstTy->isDoubleTy() && - "Invalid FPTrunc instruction"); - Dest.DoubleVal = (double) Src.FloatVal; + + if (SrcVal->getType()->getTypeID() == Type::VectorTyID) { + assert(SrcVal->getType()->getScalarType()->isFloatTy() && + DstTy->getScalarType()->isDoubleTy() && "Invalid FPExt instruction"); + + unsigned size = Src.AggregateVal.size(); + // the sizes of src and dst vectors must be equal. + Dest.AggregateVal.resize(size); + for (unsigned i = 0; i < size; i++) + Dest.AggregateVal[i].DoubleVal = (double)Src.AggregateVal[i].FloatVal; + } else { + assert(SrcVal->getType()->isFloatTy() && DstTy->isDoubleTy() && + "Invalid FPExt instruction"); + Dest.DoubleVal = (double)Src.FloatVal; + } + return Dest; } GenericValue Interpreter::executeFPToUIInst(Value *SrcVal, Type *DstTy, ExecutionContext &SF) { Type *SrcTy = SrcVal->getType(); - uint32_t DBitWidth = cast<IntegerType>(DstTy)->getBitWidth(); GenericValue Dest, Src = getOperandValue(SrcVal, SF); - assert(SrcTy->isFloatingPointTy() && "Invalid FPToUI instruction"); - if (SrcTy->getTypeID() == Type::FloatTyID) - Dest.IntVal = APIntOps::RoundFloatToAPInt(Src.FloatVal, DBitWidth); - else - Dest.IntVal = APIntOps::RoundDoubleToAPInt(Src.DoubleVal, DBitWidth); + if (SrcTy->getTypeID() == Type::VectorTyID) { + const Type *DstVecTy = DstTy->getScalarType(); + const Type *SrcVecTy = SrcTy->getScalarType(); + uint32_t DBitWidth = cast<IntegerType>(DstVecTy)->getBitWidth(); + unsigned size = Src.AggregateVal.size(); + // the sizes of src and dst vectors must be equal. + Dest.AggregateVal.resize(size); + + if (SrcVecTy->getTypeID() == Type::FloatTyID) { + assert(SrcVecTy->isFloatingPointTy() && "Invalid FPToUI instruction"); + for (unsigned i = 0; i < size; i++) + Dest.AggregateVal[i].IntVal = APIntOps::RoundFloatToAPInt( + Src.AggregateVal[i].FloatVal, DBitWidth); + } else { + for (unsigned i = 0; i < size; i++) + Dest.AggregateVal[i].IntVal = APIntOps::RoundDoubleToAPInt( + Src.AggregateVal[i].DoubleVal, DBitWidth); + } + } else { + // scalar + uint32_t DBitWidth = cast<IntegerType>(DstTy)->getBitWidth(); + assert(SrcTy->isFloatingPointTy() && "Invalid FPToUI instruction"); + + if (SrcTy->getTypeID() == Type::FloatTyID) + Dest.IntVal = APIntOps::RoundFloatToAPInt(Src.FloatVal, DBitWidth); + else { + Dest.IntVal = APIntOps::RoundDoubleToAPInt(Src.DoubleVal, DBitWidth); + } + } + return Dest; } GenericValue Interpreter::executeFPToSIInst(Value *SrcVal, Type *DstTy, ExecutionContext &SF) { Type *SrcTy = SrcVal->getType(); - uint32_t DBitWidth = cast<IntegerType>(DstTy)->getBitWidth(); GenericValue Dest, Src = getOperandValue(SrcVal, SF); - assert(SrcTy->isFloatingPointTy() && "Invalid FPToSI instruction"); - if (SrcTy->getTypeID() == Type::FloatTyID) - Dest.IntVal = APIntOps::RoundFloatToAPInt(Src.FloatVal, DBitWidth); - else - Dest.IntVal = APIntOps::RoundDoubleToAPInt(Src.DoubleVal, DBitWidth); + if (SrcTy->getTypeID() == Type::VectorTyID) { + const Type *DstVecTy = DstTy->getScalarType(); + const Type *SrcVecTy = SrcTy->getScalarType(); + uint32_t DBitWidth = cast<IntegerType>(DstVecTy)->getBitWidth(); + unsigned size = Src.AggregateVal.size(); + // the sizes of src and dst vectors must be equal + Dest.AggregateVal.resize(size); + + if (SrcVecTy->getTypeID() == Type::FloatTyID) { + assert(SrcVecTy->isFloatingPointTy() && "Invalid FPToSI instruction"); + for (unsigned i = 0; i < size; i++) + Dest.AggregateVal[i].IntVal = APIntOps::RoundFloatToAPInt( + Src.AggregateVal[i].FloatVal, DBitWidth); + } else { + for (unsigned i = 0; i < size; i++) + Dest.AggregateVal[i].IntVal = APIntOps::RoundDoubleToAPInt( + Src.AggregateVal[i].DoubleVal, DBitWidth); + } + } else { + // scalar + unsigned DBitWidth = cast<IntegerType>(DstTy)->getBitWidth(); + assert(SrcTy->isFloatingPointTy() && "Invalid FPToSI instruction"); + + if (SrcTy->getTypeID() == Type::FloatTyID) + Dest.IntVal = APIntOps::RoundFloatToAPInt(Src.FloatVal, DBitWidth); + else { + Dest.IntVal = APIntOps::RoundDoubleToAPInt(Src.DoubleVal, DBitWidth); + } + } return Dest; } GenericValue Interpreter::executeUIToFPInst(Value *SrcVal, Type *DstTy, ExecutionContext &SF) { GenericValue Dest, Src = getOperandValue(SrcVal, SF); - assert(DstTy->isFloatingPointTy() && "Invalid UIToFP instruction"); - if (DstTy->getTypeID() == Type::FloatTyID) - Dest.FloatVal = APIntOps::RoundAPIntToFloat(Src.IntVal); - else - Dest.DoubleVal = APIntOps::RoundAPIntToDouble(Src.IntVal); + if (SrcVal->getType()->getTypeID() == Type::VectorTyID) { + const Type *DstVecTy = DstTy->getScalarType(); + unsigned size = Src.AggregateVal.size(); + // the sizes of src and dst vectors must be equal + Dest.AggregateVal.resize(size); + + if (DstVecTy->getTypeID() == Type::FloatTyID) { + assert(DstVecTy->isFloatingPointTy() && "Invalid UIToFP instruction"); + for (unsigned i = 0; i < size; i++) + Dest.AggregateVal[i].FloatVal = + APIntOps::RoundAPIntToFloat(Src.AggregateVal[i].IntVal); + } else { + for (unsigned i = 0; i < size; i++) + Dest.AggregateVal[i].DoubleVal = + APIntOps::RoundAPIntToDouble(Src.AggregateVal[i].IntVal); + } + } else { + // scalar + assert(DstTy->isFloatingPointTy() && "Invalid UIToFP instruction"); + if (DstTy->getTypeID() == Type::FloatTyID) + Dest.FloatVal = APIntOps::RoundAPIntToFloat(Src.IntVal); + else { + Dest.DoubleVal = APIntOps::RoundAPIntToDouble(Src.IntVal); + } + } return Dest; } GenericValue Interpreter::executeSIToFPInst(Value *SrcVal, Type *DstTy, ExecutionContext &SF) { GenericValue Dest, Src = getOperandValue(SrcVal, SF); - assert(DstTy->isFloatingPointTy() && "Invalid SIToFP instruction"); - if (DstTy->getTypeID() == Type::FloatTyID) - Dest.FloatVal = APIntOps::RoundSignedAPIntToFloat(Src.IntVal); - else - Dest.DoubleVal = APIntOps::RoundSignedAPIntToDouble(Src.IntVal); - return Dest; + if (SrcVal->getType()->getTypeID() == Type::VectorTyID) { + const Type *DstVecTy = DstTy->getScalarType(); + unsigned size = Src.AggregateVal.size(); + // the sizes of src and dst vectors must be equal + Dest.AggregateVal.resize(size); + + if (DstVecTy->getTypeID() == Type::FloatTyID) { + assert(DstVecTy->isFloatingPointTy() && "Invalid SIToFP instruction"); + for (unsigned i = 0; i < size; i++) + Dest.AggregateVal[i].FloatVal = + APIntOps::RoundSignedAPIntToFloat(Src.AggregateVal[i].IntVal); + } else { + for (unsigned i = 0; i < size; i++) + Dest.AggregateVal[i].DoubleVal = + APIntOps::RoundSignedAPIntToDouble(Src.AggregateVal[i].IntVal); + } + } else { + // scalar + assert(DstTy->isFloatingPointTy() && "Invalid SIToFP instruction"); + if (DstTy->getTypeID() == Type::FloatTyID) + Dest.FloatVal = APIntOps::RoundSignedAPIntToFloat(Src.IntVal); + else { + Dest.DoubleVal = APIntOps::RoundSignedAPIntToDouble(Src.IntVal); + } + } + + return Dest; } GenericValue Interpreter::executePtrToIntInst(Value *SrcVal, Type *DstTy, @@ -1064,33 +1503,167 @@ GenericValue Interpreter::executeIntToPtrInst(Value *SrcVal, Type *DstTy, GenericValue Interpreter::executeBitCastInst(Value *SrcVal, Type *DstTy, ExecutionContext &SF) { - + + // This instruction supports bitwise conversion of vectors to integers and + // to vectors of other types (as long as they have the same size) Type *SrcTy = SrcVal->getType(); GenericValue Dest, Src = getOperandValue(SrcVal, SF); - if (DstTy->isPointerTy()) { - assert(SrcTy->isPointerTy() && "Invalid BitCast"); - Dest.PointerVal = Src.PointerVal; - } else if (DstTy->isIntegerTy()) { - if (SrcTy->isFloatTy()) { - Dest.IntVal = APInt::floatToBits(Src.FloatVal); - } else if (SrcTy->isDoubleTy()) { - Dest.IntVal = APInt::doubleToBits(Src.DoubleVal); - } else if (SrcTy->isIntegerTy()) { - Dest.IntVal = Src.IntVal; - } else + + if ((SrcTy->getTypeID() == Type::VectorTyID) || + (DstTy->getTypeID() == Type::VectorTyID)) { + // vector src bitcast to vector dst or vector src bitcast to scalar dst or + // scalar src bitcast to vector dst + bool isLittleEndian = TD.isLittleEndian(); + GenericValue TempDst, TempSrc, SrcVec; + const Type *SrcElemTy; + const Type *DstElemTy; + unsigned SrcBitSize; + unsigned DstBitSize; + unsigned SrcNum; + unsigned DstNum; + + if (SrcTy->getTypeID() == Type::VectorTyID) { + SrcElemTy = SrcTy->getScalarType(); + SrcBitSize = SrcTy->getScalarSizeInBits(); + SrcNum = Src.AggregateVal.size(); + SrcVec = Src; + } else { + // if src is scalar value, make it vector <1 x type> + SrcElemTy = SrcTy; + SrcBitSize = SrcTy->getPrimitiveSizeInBits(); + SrcNum = 1; + SrcVec.AggregateVal.push_back(Src); + } + + if (DstTy->getTypeID() == Type::VectorTyID) { + DstElemTy = DstTy->getScalarType(); + DstBitSize = DstTy->getScalarSizeInBits(); + DstNum = (SrcNum * SrcBitSize) / DstBitSize; + } else { + DstElemTy = DstTy; + DstBitSize = DstTy->getPrimitiveSizeInBits(); + DstNum = 1; + } + + if (SrcNum * SrcBitSize != DstNum * DstBitSize) llvm_unreachable("Invalid BitCast"); - } else if (DstTy->isFloatTy()) { - if (SrcTy->isIntegerTy()) - Dest.FloatVal = Src.IntVal.bitsToFloat(); - else - Dest.FloatVal = Src.FloatVal; - } else if (DstTy->isDoubleTy()) { - if (SrcTy->isIntegerTy()) - Dest.DoubleVal = Src.IntVal.bitsToDouble(); - else - Dest.DoubleVal = Src.DoubleVal; - } else - llvm_unreachable("Invalid Bitcast"); + + // If src is floating point, cast to integer first. + TempSrc.AggregateVal.resize(SrcNum); + if (SrcElemTy->isFloatTy()) { + for (unsigned i = 0; i < SrcNum; i++) + TempSrc.AggregateVal[i].IntVal = + APInt::floatToBits(SrcVec.AggregateVal[i].FloatVal); + + } else if (SrcElemTy->isDoubleTy()) { + for (unsigned i = 0; i < SrcNum; i++) + TempSrc.AggregateVal[i].IntVal = + APInt::doubleToBits(SrcVec.AggregateVal[i].DoubleVal); + } else if (SrcElemTy->isIntegerTy()) { + for (unsigned i = 0; i < SrcNum; i++) + TempSrc.AggregateVal[i].IntVal = SrcVec.AggregateVal[i].IntVal; + } else { + // Pointers are not allowed as the element type of vector. + llvm_unreachable("Invalid Bitcast"); + } + + // now TempSrc is integer type vector + if (DstNum < SrcNum) { + // Example: bitcast <4 x i32> <i32 0, i32 1, i32 2, i32 3> to <2 x i64> + unsigned Ratio = SrcNum / DstNum; + unsigned SrcElt = 0; + for (unsigned i = 0; i < DstNum; i++) { + GenericValue Elt; + Elt.IntVal = 0; + Elt.IntVal = Elt.IntVal.zext(DstBitSize); + unsigned ShiftAmt = isLittleEndian ? 0 : SrcBitSize * (Ratio - 1); + for (unsigned j = 0; j < Ratio; j++) { + APInt Tmp; + Tmp = Tmp.zext(SrcBitSize); + Tmp = TempSrc.AggregateVal[SrcElt++].IntVal; + Tmp = Tmp.zext(DstBitSize); + Tmp = Tmp.shl(ShiftAmt); + ShiftAmt += isLittleEndian ? SrcBitSize : -SrcBitSize; + Elt.IntVal |= Tmp; + } + TempDst.AggregateVal.push_back(Elt); + } + } else { + // Example: bitcast <2 x i64> <i64 0, i64 1> to <4 x i32> + unsigned Ratio = DstNum / SrcNum; + for (unsigned i = 0; i < SrcNum; i++) { + unsigned ShiftAmt = isLittleEndian ? 0 : DstBitSize * (Ratio - 1); + for (unsigned j = 0; j < Ratio; j++) { + GenericValue Elt; + Elt.IntVal = Elt.IntVal.zext(SrcBitSize); + Elt.IntVal = TempSrc.AggregateVal[i].IntVal; + Elt.IntVal = Elt.IntVal.lshr(ShiftAmt); + // it could be DstBitSize == SrcBitSize, so check it + if (DstBitSize < SrcBitSize) + Elt.IntVal = Elt.IntVal.trunc(DstBitSize); + ShiftAmt += isLittleEndian ? DstBitSize : -DstBitSize; + TempDst.AggregateVal.push_back(Elt); + } + } + } + + // convert result from integer to specified type + if (DstTy->getTypeID() == Type::VectorTyID) { + if (DstElemTy->isDoubleTy()) { + Dest.AggregateVal.resize(DstNum); + for (unsigned i = 0; i < DstNum; i++) + Dest.AggregateVal[i].DoubleVal = + TempDst.AggregateVal[i].IntVal.bitsToDouble(); + } else if (DstElemTy->isFloatTy()) { + Dest.AggregateVal.resize(DstNum); + for (unsigned i = 0; i < DstNum; i++) + Dest.AggregateVal[i].FloatVal = + TempDst.AggregateVal[i].IntVal.bitsToFloat(); + } else { + Dest = TempDst; + } + } else { + if (DstElemTy->isDoubleTy()) + Dest.DoubleVal = TempDst.AggregateVal[0].IntVal.bitsToDouble(); + else if (DstElemTy->isFloatTy()) { + Dest.FloatVal = TempDst.AggregateVal[0].IntVal.bitsToFloat(); + } else { + Dest.IntVal = TempDst.AggregateVal[0].IntVal; + } + } + } else { // if ((SrcTy->getTypeID() == Type::VectorTyID) || + // (DstTy->getTypeID() == Type::VectorTyID)) + + // scalar src bitcast to scalar dst + if (DstTy->isPointerTy()) { + assert(SrcTy->isPointerTy() && "Invalid BitCast"); + Dest.PointerVal = Src.PointerVal; + } else if (DstTy->isIntegerTy()) { + if (SrcTy->isFloatTy()) + Dest.IntVal = APInt::floatToBits(Src.FloatVal); + else if (SrcTy->isDoubleTy()) { + Dest.IntVal = APInt::doubleToBits(Src.DoubleVal); + } else if (SrcTy->isIntegerTy()) { + Dest.IntVal = Src.IntVal; + } else { + llvm_unreachable("Invalid BitCast"); + } + } else if (DstTy->isFloatTy()) { + if (SrcTy->isIntegerTy()) + Dest.FloatVal = Src.IntVal.bitsToFloat(); + else { + Dest.FloatVal = Src.FloatVal; + } + } else if (DstTy->isDoubleTy()) { + if (SrcTy->isIntegerTy()) + Dest.DoubleVal = Src.IntVal.bitsToDouble(); + else { + Dest.DoubleVal = Src.DoubleVal; + } + } else { + llvm_unreachable("Invalid Bitcast"); + } + } return Dest; } @@ -1187,6 +1760,39 @@ void Interpreter::visitVAArgInst(VAArgInst &I) { ++VAList.UIntPairVal.second; } +void Interpreter::visitExtractElementInst(ExtractElementInst &I) { + ExecutionContext &SF = ECStack.back(); + GenericValue Src1 = getOperandValue(I.getOperand(0), SF); + GenericValue Src2 = getOperandValue(I.getOperand(1), SF); + GenericValue Dest; + + Type *Ty = I.getType(); + const unsigned indx = unsigned(Src2.IntVal.getZExtValue()); + + if(Src1.AggregateVal.size() > indx) { + switch (Ty->getTypeID()) { + default: + dbgs() << "Unhandled destination type for extractelement instruction: " + << *Ty << "\n"; + llvm_unreachable(0); + break; + case Type::IntegerTyID: + Dest.IntVal = Src1.AggregateVal[indx].IntVal; + break; + case Type::FloatTyID: + Dest.FloatVal = Src1.AggregateVal[indx].FloatVal; + break; + case Type::DoubleTyID: + Dest.DoubleVal = Src1.AggregateVal[indx].DoubleVal; + break; + } + } else { + dbgs() << "Invalid index in extractelement instruction\n"; + } + + SetValue(&I, Dest, SF); +} + GenericValue Interpreter::getConstantExprValue (ConstantExpr *CE, ExecutionContext &SF) { switch (CE->getOpcode()) { diff --git a/lib/ExecutionEngine/Interpreter/Interpreter.h b/lib/ExecutionEngine/Interpreter/Interpreter.h index e95db2f..2952d7e 100644 --- a/lib/ExecutionEngine/Interpreter/Interpreter.h +++ b/lib/ExecutionEngine/Interpreter/Interpreter.h @@ -178,6 +178,7 @@ public: void visitAShr(BinaryOperator &I); void visitVAArgInst(VAArgInst &I); + void visitExtractElementInst(ExtractElementInst &I); void visitInstruction(Instruction &I) { errs() << I << "\n"; llvm_unreachable("Instruction not interpretable yet!"); diff --git a/lib/ExecutionEngine/JIT/Android.mk b/lib/ExecutionEngine/JIT/Android.mk index 02cef81..0466ba0 100644 --- a/lib/ExecutionEngine/JIT/Android.mk +++ b/lib/ExecutionEngine/JIT/Android.mk @@ -6,7 +6,6 @@ include $(CLEAR_VARS) LOCAL_SRC_FILES := \ JIT.cpp \ - JITDwarfEmitter.cpp \ JITEmitter.cpp \ JITMemoryManager.cpp diff --git a/lib/ExecutionEngine/JIT/CMakeLists.txt b/lib/ExecutionEngine/JIT/CMakeLists.txt index 52bb389..e16baed 100644 --- a/lib/ExecutionEngine/JIT/CMakeLists.txt +++ b/lib/ExecutionEngine/JIT/CMakeLists.txt @@ -3,7 +3,6 @@ add_definitions(-DENABLE_X86_JIT) add_llvm_library(LLVMJIT JIT.cpp - JITDwarfEmitter.cpp JITEmitter.cpp JITMemoryManager.cpp ) diff --git a/lib/ExecutionEngine/JIT/JITDwarfEmitter.cpp b/lib/ExecutionEngine/JIT/JITDwarfEmitter.cpp deleted file mode 100644 index 35d2b8b..0000000 --- a/lib/ExecutionEngine/JIT/JITDwarfEmitter.cpp +++ /dev/null @@ -1,596 +0,0 @@ -//===----- JITDwarfEmitter.cpp - Write dwarf tables into memory -----------===// -// -// The LLVM Compiler Infrastructure -// -// This file is distributed under the University of Illinois Open Source -// License. See LICENSE.TXT for details. -// -//===----------------------------------------------------------------------===// -// -// This file defines a JITDwarfEmitter object that is used by the JIT to -// write dwarf tables to memory. -// -//===----------------------------------------------------------------------===// - -#include "JITDwarfEmitter.h" -#include "JIT.h" -#include "llvm/ADT/DenseMap.h" -#include "llvm/CodeGen/JITCodeEmitter.h" -#include "llvm/CodeGen/MachineFunction.h" -#include "llvm/CodeGen/MachineModuleInfo.h" -#include "llvm/ExecutionEngine/JITMemoryManager.h" -#include "llvm/IR/DataLayout.h" -#include "llvm/IR/Function.h" -#include "llvm/MC/MCAsmInfo.h" -#include "llvm/MC/MCSymbol.h" -#include "llvm/MC/MachineLocation.h" -#include "llvm/Support/ErrorHandling.h" -#include "llvm/Target/TargetFrameLowering.h" -#include "llvm/Target/TargetInstrInfo.h" -#include "llvm/Target/TargetMachine.h" -#include "llvm/Target/TargetRegisterInfo.h" -using namespace llvm; - -JITDwarfEmitter::JITDwarfEmitter(JIT& theJit) : MMI(0), Jit(theJit) {} - - -unsigned char* JITDwarfEmitter::EmitDwarfTable(MachineFunction& F, - JITCodeEmitter& jce, - unsigned char* StartFunction, - unsigned char* EndFunction, - unsigned char* &EHFramePtr) { - assert(MMI && "MachineModuleInfo not registered!"); - - const TargetMachine& TM = F.getTarget(); - TD = TM.getDataLayout(); - stackGrowthDirection = TM.getFrameLowering()->getStackGrowthDirection(); - RI = TM.getRegisterInfo(); - MAI = TM.getMCAsmInfo(); - JCE = &jce; - - unsigned char* ExceptionTable = EmitExceptionTable(&F, StartFunction, - EndFunction); - - unsigned char* Result = 0; - - const std::vector<const Function *> Personalities = MMI->getPersonalities(); - EHFramePtr = EmitCommonEHFrame(Personalities[MMI->getPersonalityIndex()]); - - Result = EmitEHFrame(Personalities[MMI->getPersonalityIndex()], EHFramePtr, - StartFunction, EndFunction, ExceptionTable); - - return Result; -} - - -void -JITDwarfEmitter::EmitFrameMoves(intptr_t BaseLabelPtr, - const std::vector<MachineMove> &Moves) const { - unsigned PointerSize = TD->getPointerSize(); - int stackGrowth = stackGrowthDirection == TargetFrameLowering::StackGrowsUp ? - PointerSize : -PointerSize; - MCSymbol *BaseLabel = 0; - - for (unsigned i = 0, N = Moves.size(); i < N; ++i) { - const MachineMove &Move = Moves[i]; - MCSymbol *Label = Move.getLabel(); - - // Throw out move if the label is invalid. - if (Label && (*JCE->getLabelLocations())[Label] == 0) - continue; - - intptr_t LabelPtr = 0; - if (Label) LabelPtr = JCE->getLabelAddress(Label); - - const MachineLocation &Dst = Move.getDestination(); - const MachineLocation &Src = Move.getSource(); - - // Advance row if new location. - if (BaseLabelPtr && Label && BaseLabel != Label) { - JCE->emitByte(dwarf::DW_CFA_advance_loc4); - JCE->emitInt32(LabelPtr - BaseLabelPtr); - - BaseLabel = Label; - BaseLabelPtr = LabelPtr; - } - - // If advancing cfa. - if (Dst.isReg() && Dst.getReg() == MachineLocation::VirtualFP) { - if (!Src.isReg()) { - if (Src.getReg() == MachineLocation::VirtualFP) { - JCE->emitByte(dwarf::DW_CFA_def_cfa_offset); - } else { - JCE->emitByte(dwarf::DW_CFA_def_cfa); - JCE->emitULEB128Bytes(RI->getDwarfRegNum(Src.getReg(), true)); - } - - JCE->emitULEB128Bytes(-Src.getOffset()); - } else { - llvm_unreachable("Machine move not supported yet."); - } - } else if (Src.isReg() && - Src.getReg() == MachineLocation::VirtualFP) { - if (Dst.isReg()) { - JCE->emitByte(dwarf::DW_CFA_def_cfa_register); - JCE->emitULEB128Bytes(RI->getDwarfRegNum(Dst.getReg(), true)); - } else { - llvm_unreachable("Machine move not supported yet."); - } - } else { - unsigned Reg = RI->getDwarfRegNum(Src.getReg(), true); - int Offset = Dst.getOffset() / stackGrowth; - - if (Offset < 0) { - JCE->emitByte(dwarf::DW_CFA_offset_extended_sf); - JCE->emitULEB128Bytes(Reg); - JCE->emitSLEB128Bytes(Offset); - } else if (Reg < 64) { - JCE->emitByte(dwarf::DW_CFA_offset + Reg); - JCE->emitULEB128Bytes(Offset); - } else { - JCE->emitByte(dwarf::DW_CFA_offset_extended); - JCE->emitULEB128Bytes(Reg); - JCE->emitULEB128Bytes(Offset); - } - } - } -} - -/// SharedTypeIds - How many leading type ids two landing pads have in common. -static unsigned SharedTypeIds(const LandingPadInfo *L, - const LandingPadInfo *R) { - const std::vector<int> &LIds = L->TypeIds, &RIds = R->TypeIds; - unsigned LSize = LIds.size(), RSize = RIds.size(); - unsigned MinSize = LSize < RSize ? LSize : RSize; - unsigned Count = 0; - - for (; Count != MinSize; ++Count) - if (LIds[Count] != RIds[Count]) - return Count; - - return Count; -} - - -/// PadLT - Order landing pads lexicographically by type id. -static bool PadLT(const LandingPadInfo *L, const LandingPadInfo *R) { - const std::vector<int> &LIds = L->TypeIds, &RIds = R->TypeIds; - unsigned LSize = LIds.size(), RSize = RIds.size(); - unsigned MinSize = LSize < RSize ? LSize : RSize; - - for (unsigned i = 0; i != MinSize; ++i) - if (LIds[i] != RIds[i]) - return LIds[i] < RIds[i]; - - return LSize < RSize; -} - -namespace { - -/// ActionEntry - Structure describing an entry in the actions table. -struct ActionEntry { - int ValueForTypeID; // The value to write - may not be equal to the type id. - int NextAction; - struct ActionEntry *Previous; -}; - -/// PadRange - Structure holding a try-range and the associated landing pad. -struct PadRange { - // The index of the landing pad. - unsigned PadIndex; - // The index of the begin and end labels in the landing pad's label lists. - unsigned RangeIndex; -}; - -typedef DenseMap<MCSymbol*, PadRange> RangeMapType; - -/// CallSiteEntry - Structure describing an entry in the call-site table. -struct CallSiteEntry { - MCSymbol *BeginLabel; // zero indicates the start of the function. - MCSymbol *EndLabel; // zero indicates the end of the function. - MCSymbol *PadLabel; // zero indicates that there is no landing pad. - unsigned Action; -}; - -} - -unsigned char* JITDwarfEmitter::EmitExceptionTable(MachineFunction* MF, - unsigned char* StartFunction, - unsigned char* EndFunction) const { - assert(MMI && "MachineModuleInfo not registered!"); - - // Map all labels and get rid of any dead landing pads. - MMI->TidyLandingPads(JCE->getLabelLocations()); - - const std::vector<const GlobalVariable *> &TypeInfos = MMI->getTypeInfos(); - const std::vector<unsigned> &FilterIds = MMI->getFilterIds(); - const std::vector<LandingPadInfo> &PadInfos = MMI->getLandingPads(); - if (PadInfos.empty()) return 0; - - // Sort the landing pads in order of their type ids. This is used to fold - // duplicate actions. - SmallVector<const LandingPadInfo *, 64> LandingPads; - LandingPads.reserve(PadInfos.size()); - for (unsigned i = 0, N = PadInfos.size(); i != N; ++i) - LandingPads.push_back(&PadInfos[i]); - std::sort(LandingPads.begin(), LandingPads.end(), PadLT); - - // Negative type ids index into FilterIds, positive type ids index into - // TypeInfos. The value written for a positive type id is just the type - // id itself. For a negative type id, however, the value written is the - // (negative) byte offset of the corresponding FilterIds entry. The byte - // offset is usually equal to the type id, because the FilterIds entries - // are written using a variable width encoding which outputs one byte per - // entry as long as the value written is not too large, but can differ. - // This kind of complication does not occur for positive type ids because - // type infos are output using a fixed width encoding. - // FilterOffsets[i] holds the byte offset corresponding to FilterIds[i]. - SmallVector<int, 16> FilterOffsets; - FilterOffsets.reserve(FilterIds.size()); - int Offset = -1; - for(std::vector<unsigned>::const_iterator I = FilterIds.begin(), - E = FilterIds.end(); I != E; ++I) { - FilterOffsets.push_back(Offset); - Offset -= MCAsmInfo::getULEB128Size(*I); - } - - // Compute the actions table and gather the first action index for each - // landing pad site. - SmallVector<ActionEntry, 32> Actions; - SmallVector<unsigned, 64> FirstActions; - FirstActions.reserve(LandingPads.size()); - - int FirstAction = 0; - unsigned SizeActions = 0; - for (unsigned i = 0, N = LandingPads.size(); i != N; ++i) { - const LandingPadInfo *LP = LandingPads[i]; - const std::vector<int> &TypeIds = LP->TypeIds; - const unsigned NumShared = i ? SharedTypeIds(LP, LandingPads[i-1]) : 0; - unsigned SizeSiteActions = 0; - - if (NumShared < TypeIds.size()) { - unsigned SizeAction = 0; - ActionEntry *PrevAction = 0; - - if (NumShared) { - const unsigned SizePrevIds = LandingPads[i-1]->TypeIds.size(); - assert(Actions.size()); - PrevAction = &Actions.back(); - SizeAction = MCAsmInfo::getSLEB128Size(PrevAction->NextAction) + - MCAsmInfo::getSLEB128Size(PrevAction->ValueForTypeID); - for (unsigned j = NumShared; j != SizePrevIds; ++j) { - SizeAction -= MCAsmInfo::getSLEB128Size(PrevAction->ValueForTypeID); - SizeAction += -PrevAction->NextAction; - PrevAction = PrevAction->Previous; - } - } - - // Compute the actions. - for (unsigned I = NumShared, M = TypeIds.size(); I != M; ++I) { - int TypeID = TypeIds[I]; - assert(-1-TypeID < (int)FilterOffsets.size() && "Unknown filter id!"); - int ValueForTypeID = TypeID < 0 ? FilterOffsets[-1 - TypeID] : TypeID; - unsigned SizeTypeID = MCAsmInfo::getSLEB128Size(ValueForTypeID); - - int NextAction = SizeAction ? -(SizeAction + SizeTypeID) : 0; - SizeAction = SizeTypeID + MCAsmInfo::getSLEB128Size(NextAction); - SizeSiteActions += SizeAction; - - ActionEntry Action = {ValueForTypeID, NextAction, PrevAction}; - Actions.push_back(Action); - - PrevAction = &Actions.back(); - } - - // Record the first action of the landing pad site. - FirstAction = SizeActions + SizeSiteActions - SizeAction + 1; - } // else identical - re-use previous FirstAction - - FirstActions.push_back(FirstAction); - - // Compute this sites contribution to size. - SizeActions += SizeSiteActions; - } - - // Compute the call-site table. Entries must be ordered by address. - SmallVector<CallSiteEntry, 64> CallSites; - - RangeMapType PadMap; - for (unsigned i = 0, N = LandingPads.size(); i != N; ++i) { - const LandingPadInfo *LandingPad = LandingPads[i]; - for (unsigned j=0, E = LandingPad->BeginLabels.size(); j != E; ++j) { - MCSymbol *BeginLabel = LandingPad->BeginLabels[j]; - assert(!PadMap.count(BeginLabel) && "Duplicate landing pad labels!"); - PadRange P = { i, j }; - PadMap[BeginLabel] = P; - } - } - - bool MayThrow = false; - MCSymbol *LastLabel = 0; - for (MachineFunction::const_iterator I = MF->begin(), E = MF->end(); - I != E; ++I) { - for (MachineBasicBlock::const_iterator MI = I->begin(), E = I->end(); - MI != E; ++MI) { - if (!MI->isLabel()) { - MayThrow |= MI->isCall(); - continue; - } - - MCSymbol *BeginLabel = MI->getOperand(0).getMCSymbol(); - assert(BeginLabel && "Invalid label!"); - - if (BeginLabel == LastLabel) - MayThrow = false; - - RangeMapType::iterator L = PadMap.find(BeginLabel); - - if (L == PadMap.end()) - continue; - - PadRange P = L->second; - const LandingPadInfo *LandingPad = LandingPads[P.PadIndex]; - - assert(BeginLabel == LandingPad->BeginLabels[P.RangeIndex] && - "Inconsistent landing pad map!"); - - // If some instruction between the previous try-range and this one may - // throw, create a call-site entry with no landing pad for the region - // between the try-ranges. - if (MayThrow) { - CallSiteEntry Site = {LastLabel, BeginLabel, 0, 0}; - CallSites.push_back(Site); - } - - LastLabel = LandingPad->EndLabels[P.RangeIndex]; - CallSiteEntry Site = {BeginLabel, LastLabel, - LandingPad->LandingPadLabel, FirstActions[P.PadIndex]}; - - assert(Site.BeginLabel && Site.EndLabel && Site.PadLabel && - "Invalid landing pad!"); - - // Try to merge with the previous call-site. - if (CallSites.size()) { - CallSiteEntry &Prev = CallSites.back(); - if (Site.PadLabel == Prev.PadLabel && Site.Action == Prev.Action) { - // Extend the range of the previous entry. - Prev.EndLabel = Site.EndLabel; - continue; - } - } - - // Otherwise, create a new call-site. - CallSites.push_back(Site); - } - } - // If some instruction between the previous try-range and the end of the - // function may throw, create a call-site entry with no landing pad for the - // region following the try-range. - if (MayThrow) { - CallSiteEntry Site = {LastLabel, 0, 0, 0}; - CallSites.push_back(Site); - } - - // Final tallies. - unsigned SizeSites = CallSites.size() * (sizeof(int32_t) + // Site start. - sizeof(int32_t) + // Site length. - sizeof(int32_t)); // Landing pad. - for (unsigned i = 0, e = CallSites.size(); i < e; ++i) - SizeSites += MCAsmInfo::getULEB128Size(CallSites[i].Action); - - unsigned SizeTypes = TypeInfos.size() * TD->getPointerSize(); - - unsigned TypeOffset = sizeof(int8_t) + // Call site format - // Call-site table length - MCAsmInfo::getULEB128Size(SizeSites) + - SizeSites + SizeActions + SizeTypes; - - // Begin the exception table. - JCE->emitAlignmentWithFill(4, 0); - // Asm->EOL("Padding"); - - unsigned char* DwarfExceptionTable = (unsigned char*)JCE->getCurrentPCValue(); - - // Emit the header. - JCE->emitByte(dwarf::DW_EH_PE_omit); - // Asm->EOL("LPStart format (DW_EH_PE_omit)"); - JCE->emitByte(dwarf::DW_EH_PE_absptr); - // Asm->EOL("TType format (DW_EH_PE_absptr)"); - JCE->emitULEB128Bytes(TypeOffset); - // Asm->EOL("TType base offset"); - JCE->emitByte(dwarf::DW_EH_PE_udata4); - // Asm->EOL("Call site format (DW_EH_PE_udata4)"); - JCE->emitULEB128Bytes(SizeSites); - // Asm->EOL("Call-site table length"); - - // Emit the landing pad site information. - for (unsigned i = 0; i < CallSites.size(); ++i) { - CallSiteEntry &S = CallSites[i]; - intptr_t BeginLabelPtr = 0; - intptr_t EndLabelPtr = 0; - - if (!S.BeginLabel) { - BeginLabelPtr = (intptr_t)StartFunction; - JCE->emitInt32(0); - } else { - BeginLabelPtr = JCE->getLabelAddress(S.BeginLabel); - JCE->emitInt32(BeginLabelPtr - (intptr_t)StartFunction); - } - - // Asm->EOL("Region start"); - - if (!S.EndLabel) - EndLabelPtr = (intptr_t)EndFunction; - else - EndLabelPtr = JCE->getLabelAddress(S.EndLabel); - - JCE->emitInt32(EndLabelPtr - BeginLabelPtr); - //Asm->EOL("Region length"); - - if (!S.PadLabel) { - JCE->emitInt32(0); - } else { - unsigned PadLabelPtr = JCE->getLabelAddress(S.PadLabel); - JCE->emitInt32(PadLabelPtr - (intptr_t)StartFunction); - } - // Asm->EOL("Landing pad"); - - JCE->emitULEB128Bytes(S.Action); - // Asm->EOL("Action"); - } - - // Emit the actions. - for (unsigned I = 0, N = Actions.size(); I != N; ++I) { - ActionEntry &Action = Actions[I]; - - JCE->emitSLEB128Bytes(Action.ValueForTypeID); - //Asm->EOL("TypeInfo index"); - JCE->emitSLEB128Bytes(Action.NextAction); - //Asm->EOL("Next action"); - } - - // Emit the type ids. - for (unsigned M = TypeInfos.size(); M; --M) { - const GlobalVariable *GV = TypeInfos[M - 1]; - - if (GV) { - if (TD->getPointerSize() == sizeof(int32_t)) - JCE->emitInt32((intptr_t)Jit.getOrEmitGlobalVariable(GV)); - else - JCE->emitInt64((intptr_t)Jit.getOrEmitGlobalVariable(GV)); - } else { - if (TD->getPointerSize() == sizeof(int32_t)) - JCE->emitInt32(0); - else - JCE->emitInt64(0); - } - // Asm->EOL("TypeInfo"); - } - - // Emit the filter typeids. - for (unsigned j = 0, M = FilterIds.size(); j < M; ++j) { - unsigned TypeID = FilterIds[j]; - JCE->emitULEB128Bytes(TypeID); - //Asm->EOL("Filter TypeInfo index"); - } - - JCE->emitAlignmentWithFill(4, 0); - - return DwarfExceptionTable; -} - -unsigned char* -JITDwarfEmitter::EmitCommonEHFrame(const Function* Personality) const { - unsigned PointerSize = TD->getPointerSize(); - int stackGrowth = stackGrowthDirection == TargetFrameLowering::StackGrowsUp ? - PointerSize : -PointerSize; - - unsigned char* StartCommonPtr = (unsigned char*)JCE->getCurrentPCValue(); - // EH Common Frame header - JCE->allocateSpace(4, 0); - unsigned char* FrameCommonBeginPtr = (unsigned char*)JCE->getCurrentPCValue(); - JCE->emitInt32((int)0); - JCE->emitByte(dwarf::DW_CIE_VERSION); - JCE->emitString(Personality ? "zPLR" : "zR"); - JCE->emitULEB128Bytes(1); - JCE->emitSLEB128Bytes(stackGrowth); - JCE->emitByte(RI->getDwarfRegNum(RI->getRARegister(), true)); - - if (Personality) { - // Augmentation Size: 3 small ULEBs of one byte each, and the personality - // function which size is PointerSize. - JCE->emitULEB128Bytes(3 + PointerSize); - - // We set the encoding of the personality as direct encoding because we use - // the function pointer. The encoding is not relative because the current - // PC value may be bigger than the personality function pointer. - if (PointerSize == 4) { - JCE->emitByte(dwarf::DW_EH_PE_sdata4); - JCE->emitInt32(((intptr_t)Jit.getPointerToGlobal(Personality))); - } else { - JCE->emitByte(dwarf::DW_EH_PE_sdata8); - JCE->emitInt64(((intptr_t)Jit.getPointerToGlobal(Personality))); - } - - // LSDA encoding: This must match the encoding used in EmitEHFrame () - if (PointerSize == 4) - JCE->emitULEB128Bytes(dwarf::DW_EH_PE_pcrel | dwarf::DW_EH_PE_sdata4); - else - JCE->emitULEB128Bytes(dwarf::DW_EH_PE_pcrel | dwarf::DW_EH_PE_sdata8); - JCE->emitULEB128Bytes(dwarf::DW_EH_PE_pcrel | dwarf::DW_EH_PE_sdata4); - } else { - JCE->emitULEB128Bytes(1); - JCE->emitULEB128Bytes(dwarf::DW_EH_PE_pcrel | dwarf::DW_EH_PE_sdata4); - } - - EmitFrameMoves(0, MAI->getInitialFrameState()); - - JCE->emitAlignmentWithFill(PointerSize, dwarf::DW_CFA_nop); - - JCE->emitInt32At((uintptr_t*)StartCommonPtr, - (uintptr_t)((unsigned char*)JCE->getCurrentPCValue() - - FrameCommonBeginPtr)); - - return StartCommonPtr; -} - - -unsigned char* -JITDwarfEmitter::EmitEHFrame(const Function* Personality, - unsigned char* StartCommonPtr, - unsigned char* StartFunction, - unsigned char* EndFunction, - unsigned char* ExceptionTable) const { - unsigned PointerSize = TD->getPointerSize(); - - // EH frame header. - unsigned char* StartEHPtr = (unsigned char*)JCE->getCurrentPCValue(); - JCE->allocateSpace(4, 0); - unsigned char* FrameBeginPtr = (unsigned char*)JCE->getCurrentPCValue(); - // FDE CIE Offset - JCE->emitInt32(FrameBeginPtr - StartCommonPtr); - JCE->emitInt32(StartFunction - (unsigned char*)JCE->getCurrentPCValue()); - JCE->emitInt32(EndFunction - StartFunction); - - // If there is a personality and landing pads then point to the language - // specific data area in the exception table. - if (Personality) { - JCE->emitULEB128Bytes(PointerSize == 4 ? 4 : 8); - - if (PointerSize == 4) { - if (!MMI->getLandingPads().empty()) - JCE->emitInt32(ExceptionTable-(unsigned char*)JCE->getCurrentPCValue()); - else - JCE->emitInt32((int)0); - } else { - if (!MMI->getLandingPads().empty()) - JCE->emitInt64(ExceptionTable-(unsigned char*)JCE->getCurrentPCValue()); - else - JCE->emitInt64((int)0); - } - } else { - JCE->emitULEB128Bytes(0); - } - - // Indicate locations of function specific callee saved registers in - // frame. - EmitFrameMoves((intptr_t)StartFunction, MMI->getFrameMoves()); - - JCE->emitAlignmentWithFill(PointerSize, dwarf::DW_CFA_nop); - - // Indicate the size of the table - JCE->emitInt32At((uintptr_t*)StartEHPtr, - (uintptr_t)((unsigned char*)JCE->getCurrentPCValue() - - StartEHPtr)); - - // Double zeroes for the unwind runtime - if (PointerSize == 8) { - JCE->emitInt64(0); - JCE->emitInt64(0); - } else { - JCE->emitInt32(0); - JCE->emitInt32(0); - } - - return StartEHPtr; -} diff --git a/lib/ExecutionEngine/JIT/JITDwarfEmitter.h b/lib/ExecutionEngine/JIT/JITDwarfEmitter.h deleted file mode 100644 index 98ac340..0000000 --- a/lib/ExecutionEngine/JIT/JITDwarfEmitter.h +++ /dev/null @@ -1,77 +0,0 @@ -//===------ JITDwarfEmitter.h - Write dwarf tables into memory ------------===// -// -// The LLVM Compiler Infrastructure -// -// This file is distributed under the University of Illinois Open Source -// License. See LICENSE.TXT for details. -// -//===----------------------------------------------------------------------===// -// -// This file defines a JITDwarfEmitter object that is used by the JIT to -// write dwarf tables to memory. -// -//===----------------------------------------------------------------------===// - -#ifndef LLVM_EXECUTION_ENGINE_JIT_DWARFEMITTER_H -#define LLVM_EXECUTION_ENGINE_JIT_DWARFEMITTER_H - -#include "llvm/Support/DataTypes.h" -#include <vector> - -namespace llvm { - -class Function; -class JIT; -class JITCodeEmitter; -class MachineFunction; -class MachineModuleInfo; -class MachineMove; -class MCAsmInfo; -class DataLayout; -class TargetMachine; -class TargetRegisterInfo; - -class JITDwarfEmitter { - const DataLayout* TD; - JITCodeEmitter* JCE; - const TargetRegisterInfo* RI; - const MCAsmInfo *MAI; - MachineModuleInfo* MMI; - JIT& Jit; - bool stackGrowthDirection; - - unsigned char* EmitExceptionTable(MachineFunction* MF, - unsigned char* StartFunction, - unsigned char* EndFunction) const; - - void EmitFrameMoves(intptr_t BaseLabelPtr, - const std::vector<MachineMove> &Moves) const; - - unsigned char* EmitCommonEHFrame(const Function* Personality) const; - - unsigned char* EmitEHFrame(const Function* Personality, - unsigned char* StartBufferPtr, - unsigned char* StartFunction, - unsigned char* EndFunction, - unsigned char* ExceptionTable) const; - -public: - - JITDwarfEmitter(JIT& jit); - - unsigned char* EmitDwarfTable(MachineFunction& F, - JITCodeEmitter& JCE, - unsigned char* StartFunction, - unsigned char* EndFunction, - unsigned char* &EHFramePtr); - - - void setModuleInfo(MachineModuleInfo* Info) { - MMI = Info; - } -}; - - -} // end namespace llvm - -#endif // LLVM_EXECUTION_ENGINE_JIT_DWARFEMITTER_H diff --git a/lib/ExecutionEngine/JIT/JITEmitter.cpp b/lib/ExecutionEngine/JIT/JITEmitter.cpp index c273876..acbbfa1 100644 --- a/lib/ExecutionEngine/JIT/JITEmitter.cpp +++ b/lib/ExecutionEngine/JIT/JITEmitter.cpp @@ -14,7 +14,6 @@ #define DEBUG_TYPE "jit" #include "JIT.h" -#include "JITDwarfEmitter.h" #include "llvm/ADT/DenseMap.h" #include "llvm/ADT/OwningPtr.h" #include "llvm/ADT/SmallPtrSet.h" @@ -325,9 +324,6 @@ namespace { /// Resolver - This contains info about the currently resolved functions. JITResolver Resolver; - /// DE - The dwarf emitter for the jit. - OwningPtr<JITDwarfEmitter> DE; - /// LabelLocations - This vector is a mapping from Label ID's to their /// address. DenseMap<MCSymbol*, uintptr_t> LabelLocations; @@ -363,22 +359,16 @@ namespace { /// Instance of the JIT JIT *TheJIT; - bool JITExceptionHandling; - public: JITEmitter(JIT &jit, JITMemoryManager *JMM, TargetMachine &TM) : SizeEstimate(0), Resolver(jit, *this), MMI(0), CurFn(0), - EmittedFunctions(this), TheJIT(&jit), - JITExceptionHandling(TM.Options.JITExceptionHandling) { + EmittedFunctions(this), TheJIT(&jit) { MemMgr = JMM ? JMM : JITMemoryManager::CreateDefaultMemManager(); if (jit.getJITInfo().needsGOT()) { MemMgr->AllocateGOT(); DEBUG(dbgs() << "JIT is managing a GOT\n"); } - if (JITExceptionHandling) { - DE.reset(new JITDwarfEmitter(jit)); - } } ~JITEmitter() { delete MemMgr; @@ -460,7 +450,6 @@ namespace { virtual void setModuleInfo(MachineModuleInfo* Info) { MMI = Info; - if (DE.get()) DE->setModuleInfo(Info); } private: @@ -964,40 +953,6 @@ bool JITEmitter::finishFunction(MachineFunction &F) { } }); - if (JITExceptionHandling) { - uintptr_t ActualSize = 0; - SavedBufferBegin = BufferBegin; - SavedBufferEnd = BufferEnd; - SavedCurBufferPtr = CurBufferPtr; - uint8_t *FrameRegister; - - while (true) { - BufferBegin = CurBufferPtr = MemMgr->startExceptionTable(F.getFunction(), - ActualSize); - BufferEnd = BufferBegin+ActualSize; - EmittedFunctions[F.getFunction()].ExceptionTable = BufferBegin; - uint8_t *EhStart; - FrameRegister = DE->EmitDwarfTable(F, *this, FnStart, FnEnd, EhStart); - - // If the buffer was large enough to hold the table then we are done. - if (CurBufferPtr != BufferEnd) - break; - - // Try again with twice as much space. - ActualSize = (CurBufferPtr - BufferBegin) * 2; - MemMgr->deallocateExceptionTable(BufferBegin); - } - MemMgr->endExceptionTable(F.getFunction(), BufferBegin, CurBufferPtr, - FrameRegister); - BufferBegin = SavedBufferBegin; - BufferEnd = SavedBufferEnd; - CurBufferPtr = SavedCurBufferPtr; - - if (JITExceptionHandling) { - TheJIT->RegisterTable(F.getFunction(), FrameRegister); - } - } - if (MMI) MMI->EndFunction(); @@ -1027,15 +982,10 @@ void JITEmitter::deallocateMemForFunction(const Function *F) { Emitted = EmittedFunctions.find(F); if (Emitted != EmittedFunctions.end()) { MemMgr->deallocateFunctionBody(Emitted->second.FunctionBody); - MemMgr->deallocateExceptionTable(Emitted->second.ExceptionTable); TheJIT->NotifyFreeingMachineCode(Emitted->second.Code); EmittedFunctions.erase(Emitted); } - - if (JITExceptionHandling) { - TheJIT->DeregisterTable(F); - } } diff --git a/lib/ExecutionEngine/JIT/JITMemoryManager.cpp b/lib/ExecutionEngine/JIT/JITMemoryManager.cpp index 66aeb77..94db245 100644 --- a/lib/ExecutionEngine/JIT/JITMemoryManager.cpp +++ b/lib/ExecutionEngine/JIT/JITMemoryManager.cpp @@ -468,7 +468,11 @@ namespace { // Grow the required block size to account for the block header Size += sizeof(*CurBlock); - // FIXME: Alignement handling. + // Alignment handling. + if (!Alignment) + Alignment = 16; + Size += Alignment - 1; + FreeRangeHeader* candidateBlock = FreeMemoryList; FreeRangeHeader* head = FreeMemoryList; FreeRangeHeader* iter = head->Next; @@ -500,7 +504,8 @@ namespace { FreeMemoryList = candidateBlock->AllocateBlock(); // Release the memory at the end of this block that isn't needed. FreeMemoryList = CurBlock->TrimAllocationToSize(FreeMemoryList, Size); - return (uint8_t *)(CurBlock + 1); + uintptr_t unalignedAddr = (uintptr_t)CurBlock + sizeof(*CurBlock); + return (uint8_t*)RoundUpToAlignment((uint64_t)unalignedAddr, Alignment); } /// allocateDataSection - Allocate memory for a data section. @@ -509,30 +514,10 @@ namespace { return (uint8_t*)DataAllocator.Allocate(Size, Alignment); } - bool applyPermissions(std::string *ErrMsg) { + bool finalizeMemory(std::string *ErrMsg) { return false; } - /// startExceptionTable - Use startFunctionBody to allocate memory for the - /// function's exception table. - uint8_t* startExceptionTable(const Function* F, uintptr_t &ActualSize) { - return startFunctionBody(F, ActualSize); - } - - /// endExceptionTable - The exception table of F is now allocated, - /// and takes the memory in the range [TableStart,TableEnd). - void endExceptionTable(const Function *F, uint8_t *TableStart, - uint8_t *TableEnd, uint8_t* FrameRegister) { - assert(TableEnd > TableStart); - assert(TableStart == (uint8_t *)(CurBlock+1) && - "Mismatched table start/end!"); - - uintptr_t BlockSize = TableEnd - (uint8_t *)CurBlock; - - // Release the memory at the end of this block that isn't needed. - FreeMemoryList =CurBlock->TrimAllocationToSize(FreeMemoryList, BlockSize); - } - uint8_t *getGOTBase() const { return GOTBase; } @@ -557,12 +542,6 @@ namespace { if (Body) deallocateBlock(Body); } - /// deallocateExceptionTable - Deallocate memory for the specified - /// exception table. - void deallocateExceptionTable(void *ET) { - if (ET) deallocateBlock(ET); - } - /// setMemoryWritable - When code generation is in progress, /// the code pages may need permissions changed. void setMemoryWritable() diff --git a/lib/ExecutionEngine/MCJIT/MCJIT.cpp b/lib/ExecutionEngine/MCJIT/MCJIT.cpp index fee10e1..09dd924 100644 --- a/lib/ExecutionEngine/MCJIT/MCJIT.cpp +++ b/lib/ExecutionEngine/MCJIT/MCJIT.cpp @@ -14,6 +14,7 @@ #include "llvm/ExecutionEngine/MCJIT.h" #include "llvm/ExecutionEngine/ObjectBuffer.h" #include "llvm/ExecutionEngine/ObjectImage.h" +#include "llvm/ExecutionEngine/SectionMemoryManager.h" #include "llvm/IR/DataLayout.h" #include "llvm/IR/DerivedTypes.h" #include "llvm/IR/Function.h" @@ -38,7 +39,7 @@ extern "C" void LLVMLinkInMCJIT() { ExecutionEngine *MCJIT::createJIT(Module *M, std::string *ErrorStr, - JITMemoryManager *JMM, + RTDyldMemoryManager *MemMgr, bool GVsWithCode, TargetMachine *TM) { // Try to register the program as a source of symbols to resolve against. @@ -46,13 +47,14 @@ ExecutionEngine *MCJIT::createJIT(Module *M, // FIXME: Don't do this here. sys::DynamicLibrary::LoadLibraryPermanently(0, NULL); - return new MCJIT(M, TM, JMM, GVsWithCode); + return new MCJIT(M, TM, MemMgr ? MemMgr : new SectionMemoryManager(), + GVsWithCode); } MCJIT::MCJIT(Module *m, TargetMachine *tm, RTDyldMemoryManager *MM, bool AllocateGVsWithCode) : ExecutionEngine(m), TM(tm), Ctx(0), MemMgr(MM), Dyld(MM), - isCompiled(false), M(m) { + IsLoaded(false), M(m), ObjCache(0) { setDataLayout(TM->getDataLayout()); } @@ -64,7 +66,11 @@ MCJIT::~MCJIT() { delete TM; } -void MCJIT::emitObject(Module *m) { +void MCJIT::setObjectCache(ObjectCache* NewCache) { + ObjCache = NewCache; +} + +ObjectBufferStream* MCJIT::emitObject(Module *m) { /// Currently, MCJIT only supports a single module and the module passed to /// this function call is expected to be the contained module. The module /// is passed as a parameter here to prepare for multiple module support in @@ -77,30 +83,66 @@ void MCJIT::emitObject(Module *m) { // FIXME: Track compilation state on a per-module basis when multiple modules // are supported. // Re-compilation is not supported - if (isCompiled) - return; + assert(!IsLoaded); PassManager PM; PM.add(new DataLayout(*TM->getDataLayout())); // The RuntimeDyld will take ownership of this shortly - OwningPtr<ObjectBufferStream> Buffer(new ObjectBufferStream()); + OwningPtr<ObjectBufferStream> CompiledObject(new ObjectBufferStream()); // Turn the machine code intermediate representation into bytes in memory // that may be executed. - if (TM->addPassesToEmitMC(PM, Ctx, Buffer->getOStream(), false)) { + if (TM->addPassesToEmitMC(PM, Ctx, CompiledObject->getOStream(), false)) { report_fatal_error("Target does not support MC emission!"); } // Initialize passes. PM.run(*m); // Flush the output buffer to get the generated code into memory - Buffer->flush(); + CompiledObject->flush(); + + // If we have an object cache, tell it about the new object. + // Note that we're using the compiled image, not the loaded image (as below). + if (ObjCache) { + // MemoryBuffer is a thin wrapper around the actual memory, so it's OK + // to create a temporary object here and delete it after the call. + OwningPtr<MemoryBuffer> MB(CompiledObject->getMemBuffer()); + ObjCache->notifyObjectCompiled(m, MB.get()); + } + + return CompiledObject.take(); +} + +void MCJIT::loadObject(Module *M) { + + // Get a thread lock to make sure we aren't trying to load multiple times + MutexGuard locked(lock); + + // FIXME: Track compilation state on a per-module basis when multiple modules + // are supported. + // Re-compilation is not supported + if (IsLoaded) + return; + + OwningPtr<ObjectBuffer> ObjectToLoad; + // Try to load the pre-compiled object from cache if possible + if (0 != ObjCache) { + OwningPtr<MemoryBuffer> PreCompiledObject(ObjCache->getObject(M)); + if (0 != PreCompiledObject.get()) + ObjectToLoad.reset(new ObjectBuffer(PreCompiledObject.take())); + } + + // If the cache did not contain a suitable object, compile the object + if (!ObjectToLoad) { + ObjectToLoad.reset(emitObject(M)); + assert(ObjectToLoad.get() && "Compilation did not produce an object."); + } // Load the object into the dynamic linker. // handing off ownership of the buffer - LoadedObject.reset(Dyld.loadObject(Buffer.take())); + LoadedObject.reset(Dyld.loadObject(ObjectToLoad.take())); if (!LoadedObject) report_fatal_error(Dyld.getErrorString()); @@ -113,7 +155,7 @@ void MCJIT::emitObject(Module *m) { NotifyObjectEmitted(*LoadedObject); // FIXME: Add support for per-module compilation state - isCompiled = true; + IsLoaded = true; } // FIXME: Add a parameter to identify which object is being finalized when @@ -122,22 +164,21 @@ void MCJIT::emitObject(Module *m) { // protection in the interface. void MCJIT::finalizeObject() { // If the module hasn't been compiled, just do that. - if (!isCompiled) { - // If the call to Dyld.resolveRelocations() is removed from emitObject() + if (!IsLoaded) { + // If the call to Dyld.resolveRelocations() is removed from loadObject() // we'll need to do that here. - emitObject(M); - - // Set page permissions. - MemMgr->applyPermissions(); - - return; + loadObject(M); + } else { + // Resolve any relocations. + Dyld.resolveRelocations(); } - // Resolve any relocations. - Dyld.resolveRelocations(); + StringRef EHData = Dyld.getEHFrameSection(); + if (!EHData.empty()) + MemMgr->registerEHFrames(EHData); // Set page permissions. - MemMgr->applyPermissions(); + MemMgr->finalizeMemory(); } void *MCJIT::getPointerToBasicBlock(BasicBlock *BB) { @@ -151,8 +192,8 @@ void *MCJIT::getPointerToFunction(Function *F) { // dies. // FIXME: Add support for per-module compilation state - if (!isCompiled) - emitObject(M); + if (!IsLoaded) + loadObject(M); if (F->isDeclaration() || F->hasAvailableExternallyLinkage()) { bool AbortOnFailure = !F->hasExternalWeakLinkage(); @@ -284,8 +325,8 @@ GenericValue MCJIT::runFunction(Function *F, void *MCJIT::getPointerToNamedFunction(const std::string &Name, bool AbortOnFailure) { // FIXME: Add support for per-module compilation state - if (!isCompiled) - emitObject(M); + if (!IsLoaded) + loadObject(M); if (!isSymbolSearchingDisabled() && MemMgr) { void *ptr = MemMgr->getPointerToNamedFunction(Name, false); diff --git a/lib/ExecutionEngine/MCJIT/MCJIT.h b/lib/ExecutionEngine/MCJIT/MCJIT.h index 283a8e5..a899d4f 100644 --- a/lib/ExecutionEngine/MCJIT/MCJIT.h +++ b/lib/ExecutionEngine/MCJIT/MCJIT.h @@ -12,6 +12,7 @@ #include "llvm/ADT/SmallVector.h" #include "llvm/ExecutionEngine/ExecutionEngine.h" +#include "llvm/ExecutionEngine/ObjectCache.h" #include "llvm/ExecutionEngine/RuntimeDyld.h" #include "llvm/PassManager.h" @@ -34,16 +35,30 @@ class MCJIT : public ExecutionEngine { SmallVector<JITEventListener*, 2> EventListeners; // FIXME: Add support for multiple modules - bool isCompiled; + bool IsLoaded; Module *M; OwningPtr<ObjectImage> LoadedObject; + // An optional ObjectCache to be notified of compiled objects and used to + // perform lookup of pre-compiled code to avoid re-compilation. + ObjectCache *ObjCache; + public: ~MCJIT(); /// @name ExecutionEngine interface implementation /// @{ + /// Sets the object manager that MCJIT should use to avoid compilation. + virtual void setObjectCache(ObjectCache *manager); + + /// finalizeObject - ensure the module is fully processed and is usable. + /// + /// It is the user-level function for completing the process of making the + /// object usable for execution. It should be called after sections within an + /// object have been relocated using mapSectionAddress. When this method is + /// called the MCJIT execution engine will reapply relocations for a loaded + /// object. virtual void finalizeObject(); virtual void *getPointerToBasicBlock(BasicBlock *BB); @@ -90,7 +105,7 @@ public: static ExecutionEngine *createJIT(Module *M, std::string *ErrorStr, - JITMemoryManager *JMM, + RTDyldMemoryManager *MemMgr, bool GVsWithCode, TargetMachine *TM); @@ -102,7 +117,9 @@ protected: /// this function call is expected to be the contained module. The module /// is passed as a parameter here to prepare for multiple module support in /// the future. - void emitObject(Module *M); + ObjectBufferStream* emitObject(Module *M); + + void loadObject(Module *M); void NotifyObjectEmitted(const ObjectImage& Obj); void NotifyFreeingObject(const ObjectImage& Obj); diff --git a/lib/ExecutionEngine/MCJIT/SectionMemoryManager.cpp b/lib/ExecutionEngine/MCJIT/SectionMemoryManager.cpp index fa35acd..650832e 100644 --- a/lib/ExecutionEngine/MCJIT/SectionMemoryManager.cpp +++ b/lib/ExecutionEngine/MCJIT/SectionMemoryManager.cpp @@ -14,19 +14,8 @@ #include "llvm/Config/config.h" #include "llvm/ExecutionEngine/SectionMemoryManager.h" -#include "llvm/Support/DynamicLibrary.h" #include "llvm/Support/MathExtras.h" -#ifdef __linux__ - // These includes used by SectionMemoryManager::getPointerToNamedFunction() - // for Glibc trickery. See comments in this function for more information. - #ifdef HAVE_SYS_STAT_H - #include <sys/stat.h> - #endif - #include <fcntl.h> - #include <unistd.h> -#endif - namespace llvm { uint8_t *SectionMemoryManager::allocateDataSection(uintptr_t Size, @@ -111,7 +100,7 @@ uint8_t *SectionMemoryManager::allocateSection(MemoryGroup &MemGroup, return (uint8_t*)Addr; } -bool SectionMemoryManager::applyPermissions(std::string *ErrMsg) +bool SectionMemoryManager::finalizeMemory(std::string *ErrMsg) { // FIXME: Should in-progress permissions be reverted if an error occurs? error_code ec; @@ -138,6 +127,11 @@ bool SectionMemoryManager::applyPermissions(std::string *ErrMsg) // Read-write data memory already has the correct permissions + // Some platforms with separate data cache and instruction cache require + // explicit cache flush, otherwise JIT code manipulations (like resolved + // relocations) will get to the data cache but not to the instruction cache. + invalidateInstructionCache(); + return false; } @@ -162,57 +156,6 @@ void SectionMemoryManager::invalidateInstructionCache() { CodeMem.AllocatedMem[i].size()); } -static int jit_noop() { - return 0; -} - -void *SectionMemoryManager::getPointerToNamedFunction(const std::string &Name, - bool AbortOnFailure) { -#if defined(__linux__) - //===--------------------------------------------------------------------===// - // Function stubs that are invoked instead of certain library calls - // - // Force the following functions to be linked in to anything that uses the - // JIT. This is a hack designed to work around the all-too-clever Glibc - // strategy of making these functions work differently when inlined vs. when - // not inlined, and hiding their real definitions in a separate archive file - // that the dynamic linker can't see. For more info, search for - // 'libc_nonshared.a' on Google, or read http://llvm.org/PR274. - if (Name == "stat") return (void*)(intptr_t)&stat; - if (Name == "fstat") return (void*)(intptr_t)&fstat; - if (Name == "lstat") return (void*)(intptr_t)&lstat; - if (Name == "stat64") return (void*)(intptr_t)&stat64; - if (Name == "fstat64") return (void*)(intptr_t)&fstat64; - if (Name == "lstat64") return (void*)(intptr_t)&lstat64; - if (Name == "atexit") return (void*)(intptr_t)&atexit; - if (Name == "mknod") return (void*)(intptr_t)&mknod; -#endif // __linux__ - - // We should not invoke parent's ctors/dtors from generated main()! - // On Mingw and Cygwin, the symbol __main is resolved to - // callee's(eg. tools/lli) one, to invoke wrong duplicated ctors - // (and register wrong callee's dtors with atexit(3)). - // We expect ExecutionEngine::runStaticConstructorsDestructors() - // is called before ExecutionEngine::runFunctionAsMain() is called. - if (Name == "__main") return (void*)(intptr_t)&jit_noop; - - const char *NameStr = Name.c_str(); - void *Ptr = sys::DynamicLibrary::SearchForAddressOfSymbol(NameStr); - if (Ptr) return Ptr; - - // If it wasn't found and if it starts with an underscore ('_') character, - // try again without the underscore. - if (NameStr[0] == '_') { - Ptr = sys::DynamicLibrary::SearchForAddressOfSymbol(NameStr+1); - if (Ptr) return Ptr; - } - - if (AbortOnFailure) - report_fatal_error("Program used external function '" + Name + - "' which could not be resolved!"); - return 0; -} - SectionMemoryManager::~SectionMemoryManager() { for (unsigned i = 0, e = CodeMem.AllocatedMem.size(); i != e; ++i) sys::Memory::releaseMappedMemory(CodeMem.AllocatedMem[i]); diff --git a/lib/ExecutionEngine/RTDyldMemoryManager.cpp b/lib/ExecutionEngine/RTDyldMemoryManager.cpp new file mode 100644 index 0000000..4e76457 --- /dev/null +++ b/lib/ExecutionEngine/RTDyldMemoryManager.cpp @@ -0,0 +1,118 @@ +//===-- RTDyldMemoryManager.cpp - Memory manager for MC-JIT -----*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Implementation of the runtime dynamic memory manager base class. +// +//===----------------------------------------------------------------------===// + +#include "llvm/Config/config.h" +#include "llvm/ExecutionEngine/RTDyldMemoryManager.h" +#include "llvm/Support/DynamicLibrary.h" +#include "llvm/Support/ErrorHandling.h" + +#include <cstdlib> + +#ifdef __linux__ + // These includes used by RTDyldMemoryManager::getPointerToNamedFunction() + // for Glibc trickery. See comments in this function for more information. + #ifdef HAVE_SYS_STAT_H + #include <sys/stat.h> + #endif + #include <fcntl.h> + #include <unistd.h> +#endif + +namespace llvm { + +RTDyldMemoryManager::~RTDyldMemoryManager() {} + +// Determine whether we can register EH tables. +#if (defined(__GNUC__) && !defined(__ARM_EABI__) && \ + !defined(__USING_SJLJ_EXCEPTIONS__)) +#define HAVE_EHTABLE_SUPPORT 1 +#else +#define HAVE_EHTABLE_SUPPORT 0 +#endif + +#if HAVE_EHTABLE_SUPPORT +extern "C" void __register_frame(void*); + +static const char *processFDE(const char *Entry) { + const char *P = Entry; + uint32_t Length = *((const uint32_t *)P); + P += 4; + uint32_t Offset = *((const uint32_t *)P); + if (Offset != 0) + __register_frame(const_cast<char *>(Entry)); + return P + Length; +} +#endif + +void RTDyldMemoryManager::registerEHFrames(StringRef SectionData) { +#if HAVE_EHTABLE_SUPPORT + const char *P = SectionData.data(); + const char *End = SectionData.data() + SectionData.size(); + do { + P = processFDE(P); + } while(P != End); +#endif +} + +static int jit_noop() { + return 0; +} + +void *RTDyldMemoryManager::getPointerToNamedFunction(const std::string &Name, + bool AbortOnFailure) { +#if defined(__linux__) + //===--------------------------------------------------------------------===// + // Function stubs that are invoked instead of certain library calls + // + // Force the following functions to be linked in to anything that uses the + // JIT. This is a hack designed to work around the all-too-clever Glibc + // strategy of making these functions work differently when inlined vs. when + // not inlined, and hiding their real definitions in a separate archive file + // that the dynamic linker can't see. For more info, search for + // 'libc_nonshared.a' on Google, or read http://llvm.org/PR274. + if (Name == "stat") return (void*)(intptr_t)&stat; + if (Name == "fstat") return (void*)(intptr_t)&fstat; + if (Name == "lstat") return (void*)(intptr_t)&lstat; + if (Name == "stat64") return (void*)(intptr_t)&stat64; + if (Name == "fstat64") return (void*)(intptr_t)&fstat64; + if (Name == "lstat64") return (void*)(intptr_t)&lstat64; + if (Name == "atexit") return (void*)(intptr_t)&atexit; + if (Name == "mknod") return (void*)(intptr_t)&mknod; +#endif // __linux__ + + // We should not invoke parent's ctors/dtors from generated main()! + // On Mingw and Cygwin, the symbol __main is resolved to + // callee's(eg. tools/lli) one, to invoke wrong duplicated ctors + // (and register wrong callee's dtors with atexit(3)). + // We expect ExecutionEngine::runStaticConstructorsDestructors() + // is called before ExecutionEngine::runFunctionAsMain() is called. + if (Name == "__main") return (void*)(intptr_t)&jit_noop; + + const char *NameStr = Name.c_str(); + void *Ptr = sys::DynamicLibrary::SearchForAddressOfSymbol(NameStr); + if (Ptr) return Ptr; + + // If it wasn't found and if it starts with an underscore ('_') character, + // try again without the underscore. + if (NameStr[0] == '_') { + Ptr = sys::DynamicLibrary::SearchForAddressOfSymbol(NameStr+1); + if (Ptr) return Ptr; + } + + if (AbortOnFailure) + report_fatal_error("Program used external function '" + Name + + "' which could not be resolved!"); + return 0; +} + +} // namespace llvm diff --git a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp index 409b25f..943622f 100644 --- a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp +++ b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp @@ -17,18 +17,22 @@ #include "RuntimeDyldELF.h" #include "RuntimeDyldImpl.h" #include "RuntimeDyldMachO.h" +#include "llvm/Support/FileSystem.h" #include "llvm/Support/MathExtras.h" -#include "llvm/Support/Path.h" +#include "llvm/Object/ELF.h" using namespace llvm; using namespace llvm::object; // Empty out-of-line virtual destructor as the key function. -RTDyldMemoryManager::~RTDyldMemoryManager() {} RuntimeDyldImpl::~RuntimeDyldImpl() {} namespace llvm { +StringRef RuntimeDyldImpl::getEHFrameSection() { + return StringRef(); +} + // Resolve the relocations for all symbols we currently know about. void RuntimeDyldImpl::resolveRelocations() { // First, resolve relocations associated with external symbols. @@ -96,7 +100,8 @@ ObjectImage *RuntimeDyldImpl::loadObject(ObjectBuffer *InputBuffer) { bool isCommon = flags & SymbolRef::SF_Common; if (isCommon) { // Add the common symbols to a list. We'll allocate them all below. - uint64_t Align = getCommonSymbolAlignment(*i); + uint32_t Align; + Check(i->getAlignment(Align)); uint64_t Size = 0; Check(i->getSize(Size)); CommonSize += Size + Align; @@ -142,6 +147,7 @@ ObjectImage *RuntimeDyldImpl::loadObject(ObjectBuffer *InputBuffer) { bool isFirstRelocation = true; unsigned SectionID = 0; StubMap Stubs; + section_iterator RelocatedSection = si->getRelocatedSection(); for (relocation_iterator i = si->begin_relocations(), e = si->end_relocations(); i != e; i.increment(err)) { @@ -149,23 +155,14 @@ ObjectImage *RuntimeDyldImpl::loadObject(ObjectBuffer *InputBuffer) { // If it's the first relocation in this section, find its SectionID if (isFirstRelocation) { - SectionID = findOrEmitSection(*obj, *si, true, LocalSections); + SectionID = + findOrEmitSection(*obj, *RelocatedSection, true, LocalSections); DEBUG(dbgs() << "\tSectionID: " << SectionID << "\n"); isFirstRelocation = false; } - ObjRelocationInfo RI; - RI.SectionID = SectionID; - Check(i->getAdditionalInfo(RI.AdditionalInfo)); - Check(i->getOffset(RI.Offset)); - Check(i->getSymbol(RI.Symbol)); - Check(i->getType(RI.Type)); - - DEBUG(dbgs() << "\t\tAddend: " << RI.AdditionalInfo - << " Offset: " << format("%p", (uintptr_t)RI.Offset) - << " Type: " << (uint32_t)(RI.Type & 0xffffffffL) - << "\n"); - processRelocationRef(RI, *obj, LocalSections, LocalSymbols, Stubs); + processRelocationRef(SectionID, *i, *obj, LocalSections, LocalSymbols, + Stubs); } } @@ -183,7 +180,7 @@ void RuntimeDyldImpl::emitCommonSymbols(ObjectImage &Obj, if (!Addr) report_fatal_error("Unable to allocate memory for common symbols!"); uint64_t Offset = 0; - Sections.push_back(SectionEntry(StringRef(), Addr, TotalSize, TotalSize, 0)); + Sections.push_back(SectionEntry(StringRef(), Addr, TotalSize, 0)); memset(Addr, 0, TotalSize); DEBUG(dbgs() << "emitCommonSection SectionID: " << SectionID @@ -220,11 +217,25 @@ unsigned RuntimeDyldImpl::emitSection(ObjectImage &Obj, unsigned StubBufSize = 0, StubSize = getMaxStubSize(); error_code err; + const ObjectFile *ObjFile = Obj.getObjectFile(); + // FIXME: this is an inefficient way to handle this. We should computed the + // necessary section allocation size in loadObject by walking all the sections + // once. if (StubSize > 0) { - for (relocation_iterator i = Section.begin_relocations(), - e = Section.end_relocations(); i != e; i.increment(err), Check(err)) - StubBufSize += StubSize; + for (section_iterator SI = ObjFile->begin_sections(), + SE = ObjFile->end_sections(); + SI != SE; SI.increment(err), Check(err)) { + section_iterator RelSecI = SI->getRelocatedSection(); + if (!(RelSecI == Section)) + continue; + + for (relocation_iterator I = SI->begin_relocations(), + E = SI->end_relocations(); I != E; I.increment(err), Check(err)) { + StubBufSize += StubSize; + } + } } + StringRef data; uint64_t Alignment64; Check(Section.getContents(data)); @@ -243,6 +254,12 @@ unsigned RuntimeDyldImpl::emitSection(ObjectImage &Obj, Check(Section.isReadOnlyData(IsReadOnly)); Check(Section.getSize(DataSize)); Check(Section.getName(Name)); + if (StubSize > 0) { + unsigned StubAlignment = getStubAlignment(); + unsigned EndAlignment = (DataSize | Alignment) & -(DataSize | Alignment); + if (StubAlignment > EndAlignment) + StubBufSize += StubAlignment - EndAlignment; + } unsigned Allocate; unsigned SectionID = Sections.size(); @@ -295,8 +312,7 @@ unsigned RuntimeDyldImpl::emitSection(ObjectImage &Obj, << "\n"); } - Sections.push_back(SectionEntry(Name, Addr, Allocate, DataSize, - (uintptr_t)pData)); + Sections.push_back(SectionEntry(Name, Addr, DataSize, (uintptr_t)pData)); return SectionID; } @@ -339,7 +355,25 @@ void RuntimeDyldImpl::addRelocationForSymbol(const RelocationEntry &RE, } uint8_t *RuntimeDyldImpl::createStubFunction(uint8_t *Addr) { - if (Arch == Triple::arm) { + if (Arch == Triple::aarch64) { + // This stub has to be able to access the full address space, + // since symbol lookup won't necessarily find a handy, in-range, + // PLT stub for functions which could be anywhere. + uint32_t *StubAddr = (uint32_t*)Addr; + + // Stub can use ip0 (== x16) to calculate address + *StubAddr = 0xd2e00010; // movz ip0, #:abs_g3:<addr> + StubAddr++; + *StubAddr = 0xf2c00010; // movk ip0, #:abs_g2_nc:<addr> + StubAddr++; + *StubAddr = 0xf2a00010; // movk ip0, #:abs_g1_nc:<addr> + StubAddr++; + *StubAddr = 0xf2800010; // movk ip0, #:abs_g0_nc:<addr> + StubAddr++; + *StubAddr = 0xd61f0200; // br ip0 + + return Addr; + } else if (Arch == Triple::arm) { // TODO: There is only ARM far stub now. We should add the Thumb stub, // and stubs for branches Thumb - ARM and ARM - Thumb. uint32_t *StubAddr = (uint32_t*)Addr; @@ -362,7 +396,7 @@ uint8_t *RuntimeDyldImpl::createStubFunction(uint8_t *Addr) { StubAddr++; *StubAddr = NopInstr; return Addr; - } else if (Arch == Triple::ppc64) { + } else if (Arch == Triple::ppc64 || Arch == Triple::ppc64le) { // PowerPC64 stub: the address points to a function descriptor // instead of the function itself. Load the function address // on r11 and sets it to control register. Also loads the function @@ -380,6 +414,13 @@ uint8_t *RuntimeDyldImpl::createStubFunction(uint8_t *Addr) { writeInt32BE(Addr+40, 0x4E800420); // bctr return Addr; + } else if (Arch == Triple::systemz) { + writeInt16BE(Addr, 0xC418); // lgrl %r1,.+8 + writeInt16BE(Addr+2, 0x0000); + writeInt16BE(Addr+4, 0x0004); + writeInt16BE(Addr+6, 0x07F1); // brc 15,%r1 + // 8-byte address stored at Addr + 8 + return Addr; } return Addr; } @@ -401,26 +442,14 @@ void RuntimeDyldImpl::reassignSectionAddress(unsigned SectionID, Sections[SectionID].LoadAddress = Addr; } -void RuntimeDyldImpl::resolveRelocationEntry(const RelocationEntry &RE, - uint64_t Value) { - // Ignore relocations for sections that were not loaded - if (Sections[RE.SectionID].Address != 0) { - DEBUG(dbgs() << "\tSectionID: " << RE.SectionID - << " + " << RE.Offset << " (" - << format("%p", Sections[RE.SectionID].Address + RE.Offset) << ")" - << " RelType: " << RE.RelType - << " Addend: " << RE.Addend - << "\n"); - - resolveRelocation(Sections[RE.SectionID], RE.Offset, - Value, RE.RelType, RE.Addend); - } -} - void RuntimeDyldImpl::resolveRelocationList(const RelocationList &Relocs, uint64_t Value) { for (unsigned i = 0, e = Relocs.size(); i != e; ++i) { - resolveRelocationEntry(Relocs[i], Value); + const RelocationEntry &RE = Relocs[i]; + // Ignore relocations for sections that were not loaded + if (Sections[RE.SectionID].Address == 0) + continue; + resolveRelocation(RE, Value); } } @@ -472,33 +501,34 @@ RuntimeDyld::~RuntimeDyld() { ObjectImage *RuntimeDyld::loadObject(ObjectBuffer *InputBuffer) { if (!Dyld) { - sys::LLVMFileType type = sys::IdentifyFileType( - InputBuffer->getBufferStart(), - static_cast<unsigned>(InputBuffer->getBufferSize())); - switch (type) { - case sys::ELF_Relocatable_FileType: - case sys::ELF_Executable_FileType: - case sys::ELF_SharedObject_FileType: - case sys::ELF_Core_FileType: - Dyld = new RuntimeDyldELF(MM); - break; - case sys::Mach_O_Object_FileType: - case sys::Mach_O_Executable_FileType: - case sys::Mach_O_FixedVirtualMemorySharedLib_FileType: - case sys::Mach_O_Core_FileType: - case sys::Mach_O_PreloadExecutable_FileType: - case sys::Mach_O_DynamicallyLinkedSharedLib_FileType: - case sys::Mach_O_DynamicLinker_FileType: - case sys::Mach_O_Bundle_FileType: - case sys::Mach_O_DynamicallyLinkedSharedLibStub_FileType: - case sys::Mach_O_DSYMCompanion_FileType: - Dyld = new RuntimeDyldMachO(MM); - break; - case sys::Unknown_FileType: - case sys::Bitcode_FileType: - case sys::Archive_FileType: - case sys::COFF_FileType: - report_fatal_error("Incompatible object format!"); + sys::fs::file_magic Type = + sys::fs::identify_magic(InputBuffer->getBuffer()); + switch (Type) { + case sys::fs::file_magic::elf_relocatable: + case sys::fs::file_magic::elf_executable: + case sys::fs::file_magic::elf_shared_object: + case sys::fs::file_magic::elf_core: + Dyld = new RuntimeDyldELF(MM); + break; + case sys::fs::file_magic::macho_object: + case sys::fs::file_magic::macho_executable: + case sys::fs::file_magic::macho_fixed_virtual_memory_shared_lib: + case sys::fs::file_magic::macho_core: + case sys::fs::file_magic::macho_preload_executable: + case sys::fs::file_magic::macho_dynamically_linked_shared_lib: + case sys::fs::file_magic::macho_dynamic_linker: + case sys::fs::file_magic::macho_bundle: + case sys::fs::file_magic::macho_dynamically_linked_shared_lib_stub: + case sys::fs::file_magic::macho_dsym_companion: + Dyld = new RuntimeDyldMachO(MM); + break; + case sys::fs::file_magic::unknown: + case sys::fs::file_magic::bitcode: + case sys::fs::file_magic::archive: + case sys::fs::file_magic::coff_object: + case sys::fs::file_magic::pecoff_executable: + case sys::fs::file_magic::macho_universal_binary: + report_fatal_error("Incompatible object format!"); } } else { if (!Dyld->isCompatibleFormat(InputBuffer)) @@ -534,4 +564,8 @@ StringRef RuntimeDyld::getErrorString() { return Dyld->getErrorString(); } +StringRef RuntimeDyld::getEHFrameSection() { + return Dyld->getEHFrameSection(); +} + } // end namespace llvm diff --git a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp index b8537b1..cd99c3c 100644 --- a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp +++ b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp @@ -41,7 +41,7 @@ error_code check(error_code Err) { template<class ELFT> class DyldELFObject : public ELFObjectFile<ELFT> { - LLVM_ELF_IMPORT_TYPES(ELFT) + LLVM_ELF_IMPORT_TYPES_ELFT(ELFT) typedef Elf_Shdr_Impl<ELFT> Elf_Shdr; typedef Elf_Sym_Impl<ELFT> Elf_Sym; @@ -151,6 +151,14 @@ void DyldELFObject<ELFT>::updateSymbolAddress(const SymbolRef &SymRef, namespace llvm { +StringRef RuntimeDyldELF::getEHFrameSection() { + for (int i = 0, e = Sections.size(); i != e; ++i) { + if (Sections[i].Name == ".eh_frame") + return StringRef((const char*)Sections[i].Address, Sections[i].Size); + } + return StringRef(); +} + ObjectImage *RuntimeDyldELF::createObjectImage(ObjectBuffer *Buffer) { if (Buffer->getBufferSize() < ELF::EI_NIDENT) llvm_unreachable("Unexpected ELF object size"); @@ -269,12 +277,115 @@ void RuntimeDyldELF::resolveX86Relocation(const SectionEntry &Section, } } +void RuntimeDyldELF::resolveAArch64Relocation(const SectionEntry &Section, + uint64_t Offset, + uint64_t Value, + uint32_t Type, + int64_t Addend) { + uint32_t *TargetPtr = reinterpret_cast<uint32_t*>(Section.Address + Offset); + uint64_t FinalAddress = Section.LoadAddress + Offset; + + DEBUG(dbgs() << "resolveAArch64Relocation, LocalAddress: 0x" + << format("%llx", Section.Address + Offset) + << " FinalAddress: 0x" << format("%llx",FinalAddress) + << " Value: 0x" << format("%llx",Value) + << " Type: 0x" << format("%x",Type) + << " Addend: 0x" << format("%llx",Addend) + << "\n"); + + switch (Type) { + default: + llvm_unreachable("Relocation type not implemented yet!"); + break; + case ELF::R_AARCH64_ABS64: { + uint64_t *TargetPtr = reinterpret_cast<uint64_t*>(Section.Address + Offset); + *TargetPtr = Value + Addend; + break; + } + case ELF::R_AARCH64_PREL32: { + uint64_t Result = Value + Addend - FinalAddress; + assert(static_cast<int64_t>(Result) >= INT32_MIN && + static_cast<int64_t>(Result) <= UINT32_MAX); + *TargetPtr = static_cast<uint32_t>(Result & 0xffffffffU); + break; + } + case ELF::R_AARCH64_CALL26: // fallthrough + case ELF::R_AARCH64_JUMP26: { + // Operation: S+A-P. Set Call or B immediate value to bits fff_fffc of the + // calculation. + uint64_t BranchImm = Value + Addend - FinalAddress; + + // "Check that -2^27 <= result < 2^27". + assert(-(1LL << 27) <= static_cast<int64_t>(BranchImm) && + static_cast<int64_t>(BranchImm) < (1LL << 27)); + + // AArch64 code is emitted with .rela relocations. The data already in any + // bits affected by the relocation on entry is garbage. + *TargetPtr &= 0xfc000000U; + // Immediate goes in bits 25:0 of B and BL. + *TargetPtr |= static_cast<uint32_t>(BranchImm & 0xffffffcU) >> 2; + break; + } + case ELF::R_AARCH64_MOVW_UABS_G3: { + uint64_t Result = Value + Addend; + + // AArch64 code is emitted with .rela relocations. The data already in any + // bits affected by the relocation on entry is garbage. + *TargetPtr &= 0xffe0001fU; + // Immediate goes in bits 20:5 of MOVZ/MOVK instruction + *TargetPtr |= Result >> (48 - 5); + // Shift must be "lsl #48", in bits 22:21 + assert((*TargetPtr >> 21 & 0x3) == 3 && "invalid shift for relocation"); + break; + } + case ELF::R_AARCH64_MOVW_UABS_G2_NC: { + uint64_t Result = Value + Addend; + + + // AArch64 code is emitted with .rela relocations. The data already in any + // bits affected by the relocation on entry is garbage. + *TargetPtr &= 0xffe0001fU; + // Immediate goes in bits 20:5 of MOVZ/MOVK instruction + *TargetPtr |= ((Result & 0xffff00000000ULL) >> (32 - 5)); + // Shift must be "lsl #32", in bits 22:21 + assert((*TargetPtr >> 21 & 0x3) == 2 && "invalid shift for relocation"); + break; + } + case ELF::R_AARCH64_MOVW_UABS_G1_NC: { + uint64_t Result = Value + Addend; + + // AArch64 code is emitted with .rela relocations. The data already in any + // bits affected by the relocation on entry is garbage. + *TargetPtr &= 0xffe0001fU; + // Immediate goes in bits 20:5 of MOVZ/MOVK instruction + *TargetPtr |= ((Result & 0xffff0000U) >> (16 - 5)); + // Shift must be "lsl #16", in bits 22:2 + assert((*TargetPtr >> 21 & 0x3) == 1 && "invalid shift for relocation"); + break; + } + case ELF::R_AARCH64_MOVW_UABS_G0_NC: { + uint64_t Result = Value + Addend; + + // AArch64 code is emitted with .rela relocations. The data already in any + // bits affected by the relocation on entry is garbage. + *TargetPtr &= 0xffe0001fU; + // Immediate goes in bits 20:5 of MOVZ/MOVK instruction + *TargetPtr |= ((Result & 0xffffU) << 5); + // Shift must be "lsl #0", in bits 22:21. + assert((*TargetPtr >> 21 & 0x3) == 0 && "invalid shift for relocation"); + break; + } + } +} + void RuntimeDyldELF::resolveARMRelocation(const SectionEntry &Section, uint64_t Offset, uint32_t Value, uint32_t Type, int32_t Addend) { // TODO: Add Thumb relocations. + uint32_t *Placeholder = reinterpret_cast<uint32_t*>(Section.ObjAddress + + Offset); uint32_t* TargetPtr = (uint32_t*)(Section.Address + Offset); uint32_t FinalAddress = ((Section.LoadAddress + Offset) & 0xFFFFFFFF); Value += Addend; @@ -293,44 +404,51 @@ void RuntimeDyldELF::resolveARMRelocation(const SectionEntry &Section, // Write a 32bit value to relocation address, taking into account the // implicit addend encoded in the target. - case ELF::R_ARM_TARGET1 : - case ELF::R_ARM_ABS32 : - *TargetPtr += Value; + case ELF::R_ARM_TARGET1: + case ELF::R_ARM_ABS32: + *TargetPtr = *Placeholder + Value; break; - // Write first 16 bit of 32 bit value to the mov instruction. // Last 4 bit should be shifted. - case ELF::R_ARM_MOVW_ABS_NC : + case ELF::R_ARM_MOVW_ABS_NC: // We are not expecting any other addend in the relocation address. // Using 0x000F0FFF because MOVW has its 16 bit immediate split into 2 // non-contiguous fields. - assert((*TargetPtr & 0x000F0FFF) == 0); + assert((*Placeholder & 0x000F0FFF) == 0); Value = Value & 0xFFFF; - *TargetPtr |= Value & 0xFFF; + *TargetPtr = *Placeholder | (Value & 0xFFF); *TargetPtr |= ((Value >> 12) & 0xF) << 16; break; - // Write last 16 bit of 32 bit value to the mov instruction. // Last 4 bit should be shifted. - case ELF::R_ARM_MOVT_ABS : + case ELF::R_ARM_MOVT_ABS: // We are not expecting any other addend in the relocation address. // Use 0x000F0FFF for the same reason as R_ARM_MOVW_ABS_NC. - assert((*TargetPtr & 0x000F0FFF) == 0); + assert((*Placeholder & 0x000F0FFF) == 0); + Value = (Value >> 16) & 0xFFFF; - *TargetPtr |= Value & 0xFFF; + *TargetPtr = *Placeholder | (Value & 0xFFF); *TargetPtr |= ((Value >> 12) & 0xF) << 16; break; - // Write 24 bit relative value to the branch instruction. case ELF::R_ARM_PC24 : // Fall through. case ELF::R_ARM_CALL : // Fall through. - case ELF::R_ARM_JUMP24 : + case ELF::R_ARM_JUMP24: { int32_t RelValue = static_cast<int32_t>(Value - FinalAddress - 8); RelValue = (RelValue & 0x03FFFFFC) >> 2; + assert((*TargetPtr & 0xFFFFFF) == 0xFFFFFE); *TargetPtr &= 0xFF000000; *TargetPtr |= RelValue; break; } + case ELF::R_ARM_PRIVATE_0: + // This relocation is reserved by the ARM ELF ABI for internal use. We + // appropriate it here to act as an R_ARM_ABS32 without any addend for use + // in the stubs created during JIT (which can't put an addend into the + // original object file). + *TargetPtr = Value; + break; + } } void RuntimeDyldELF::resolveMIPSRelocation(const SectionEntry &Section, @@ -338,6 +456,8 @@ void RuntimeDyldELF::resolveMIPSRelocation(const SectionEntry &Section, uint32_t Value, uint32_t Type, int32_t Addend) { + uint32_t *Placeholder = reinterpret_cast<uint32_t*>(Section.ObjAddress + + Offset); uint32_t* TargetPtr = (uint32_t*)(Section.Address + Offset); Value += Addend; @@ -355,19 +475,30 @@ void RuntimeDyldELF::resolveMIPSRelocation(const SectionEntry &Section, llvm_unreachable("Not implemented relocation type!"); break; case ELF::R_MIPS_32: - *TargetPtr = Value + (*TargetPtr); + *TargetPtr = Value + (*Placeholder); break; case ELF::R_MIPS_26: - *TargetPtr = ((*TargetPtr) & 0xfc000000) | (( Value & 0x0fffffff) >> 2); + *TargetPtr = ((*Placeholder) & 0xfc000000) | (( Value & 0x0fffffff) >> 2); break; case ELF::R_MIPS_HI16: // Get the higher 16-bits. Also add 1 if bit 15 is 1. - Value += ((*TargetPtr) & 0x0000ffff) << 16; + Value += ((*Placeholder) & 0x0000ffff) << 16; + *TargetPtr = ((*Placeholder) & 0xffff0000) | + (((Value + 0x8000) >> 16) & 0xffff); + break; + case ELF::R_MIPS_LO16: + Value += ((*Placeholder) & 0x0000ffff); + *TargetPtr = ((*Placeholder) & 0xffff0000) | (Value & 0xffff); + break; + case ELF::R_MIPS_UNUSED1: + // Similar to ELF::R_ARM_PRIVATE_0, R_MIPS_UNUSED1 and R_MIPS_UNUSED2 + // are used for internal JIT purpose. These relocations are similar to + // R_MIPS_HI16 and R_MIPS_LO16, but they do not take any addend into + // account. *TargetPtr = ((*TargetPtr) & 0xffff0000) | (((Value + 0x8000) >> 16) & 0xffff); break; - case ELF::R_MIPS_LO16: - Value += ((*TargetPtr) & 0x0000ffff); + case ELF::R_MIPS_UNUSED2: *TargetPtr = ((*TargetPtr) & 0xffff0000) | (Value & 0xffff); break; } @@ -412,9 +543,13 @@ void RuntimeDyldELF::findOPDEntrySection(ObjectImage &Obj, error_code err; for (section_iterator si = Obj.begin_sections(), se = Obj.end_sections(); si != se; si.increment(err)) { - StringRef SectionName; - check(si->getName(SectionName)); - if (SectionName != ".opd") + section_iterator RelSecI = si->getRelocatedSection(); + if (RelSecI == Obj.end_sections()) + continue; + + StringRef RelSectionName; + check(RelSecI->getName(RelSectionName)); + if (RelSectionName != ".opd") continue; for (relocation_iterator i = si->begin_relocations(), @@ -430,12 +565,11 @@ void RuntimeDyldELF::findOPDEntrySection(ObjectImage &Obj, continue; } - SymbolRef TargetSymbol; uint64_t TargetSymbolOffset; - int64_t TargetAdditionalInfo; - check(i->getSymbol(TargetSymbol)); + symbol_iterator TargetSymbol = i->getSymbol(); check(i->getOffset(TargetSymbolOffset)); - check(i->getAdditionalInfo(TargetAdditionalInfo)); + int64_t Addend; + check(getELFRelocationAddend(*i, Addend)); i = i.increment(err); if (i == e) @@ -455,9 +589,9 @@ void RuntimeDyldELF::findOPDEntrySection(ObjectImage &Obj, continue; section_iterator tsi(Obj.end_sections()); - check(TargetSymbol.getSection(tsi)); + check(TargetSymbol->getSection(tsi)); Rel.SectionID = findOrEmitSection(Obj, (*tsi), true, LocalSections); - Rel.Addend = (intptr_t)TargetAdditionalInfo; + Rel.Addend = (intptr_t)Addend; return; } } @@ -541,6 +675,11 @@ void RuntimeDyldELF::resolvePPC64Relocation(const SectionEntry &Section, llvm_unreachable("Relocation R_PPC64_REL32 overflow"); writeInt32BE(LocalAddress, delta); } break; + case ELF::R_PPC64_REL64: { + uint64_t FinalAddress = (Section.LoadAddress + Offset); + uint64_t Delta = Value - FinalAddress + Addend; + writeInt64BE(LocalAddress, Delta); + } break; case ELF::R_PPC64_ADDR64 : writeInt64BE(LocalAddress, Value + Addend); break; @@ -560,6 +699,48 @@ void RuntimeDyldELF::resolvePPC64Relocation(const SectionEntry &Section, } } +void RuntimeDyldELF::resolveSystemZRelocation(const SectionEntry &Section, + uint64_t Offset, + uint64_t Value, + uint32_t Type, + int64_t Addend) { + uint8_t *LocalAddress = Section.Address + Offset; + switch (Type) { + default: + llvm_unreachable("Relocation type not implemented yet!"); + break; + case ELF::R_390_PC16DBL: + case ELF::R_390_PLT16DBL: { + int64_t Delta = (Value + Addend) - (Section.LoadAddress + Offset); + assert(int16_t(Delta / 2) * 2 == Delta && "R_390_PC16DBL overflow"); + writeInt16BE(LocalAddress, Delta / 2); + break; + } + case ELF::R_390_PC32DBL: + case ELF::R_390_PLT32DBL: { + int64_t Delta = (Value + Addend) - (Section.LoadAddress + Offset); + assert(int32_t(Delta / 2) * 2 == Delta && "R_390_PC32DBL overflow"); + writeInt32BE(LocalAddress, Delta / 2); + break; + } + case ELF::R_390_PC32: { + int64_t Delta = (Value + Addend) - (Section.LoadAddress + Offset); + assert(int32_t(Delta) == Delta && "R_390_PC32 overflow"); + writeInt32BE(LocalAddress, Delta); + break; + } + case ELF::R_390_64: + writeInt64BE(LocalAddress, Value + Addend); + break; + } +} + +void RuntimeDyldELF::resolveRelocation(const RelocationEntry &RE, + uint64_t Value) { + const SectionEntry &Section = Sections[RE.SectionID]; + return resolveRelocation(Section, RE.Offset, Value, RE.RelType, RE.Addend); +} + void RuntimeDyldELF::resolveRelocation(const SectionEntry &Section, uint64_t Offset, uint64_t Value, @@ -574,6 +755,9 @@ void RuntimeDyldELF::resolveRelocation(const SectionEntry &Section, (uint32_t)(Value & 0xffffffffL), Type, (uint32_t)(Addend & 0xffffffffL)); break; + case Triple::aarch64: + resolveAArch64Relocation(Section, Offset, Value, Type, Addend); + break; case Triple::arm: // Fall through. case Triple::thumb: resolveARMRelocation(Section, Offset, @@ -586,45 +770,56 @@ void RuntimeDyldELF::resolveRelocation(const SectionEntry &Section, (uint32_t)(Value & 0xffffffffL), Type, (uint32_t)(Addend & 0xffffffffL)); break; - case Triple::ppc64: + case Triple::ppc64: // Fall through. + case Triple::ppc64le: resolvePPC64Relocation(Section, Offset, Value, Type, Addend); break; + case Triple::systemz: + resolveSystemZRelocation(Section, Offset, Value, Type, Addend); + break; default: llvm_unreachable("Unsupported CPU type!"); } } -void RuntimeDyldELF::processRelocationRef(const ObjRelocationInfo &Rel, +void RuntimeDyldELF::processRelocationRef(unsigned SectionID, + RelocationRef RelI, ObjectImage &Obj, ObjSectionToIDMap &ObjSectionToID, const SymbolTableMap &Symbols, StubMap &Stubs) { - - uint32_t RelType = (uint32_t)(Rel.Type & 0xffffffffL); - intptr_t Addend = (intptr_t)Rel.AdditionalInfo; - const SymbolRef &Symbol = Rel.Symbol; + uint64_t RelType; + Check(RelI.getType(RelType)); + int64_t Addend; + Check(getELFRelocationAddend(RelI, Addend)); + symbol_iterator Symbol = RelI.getSymbol(); // Obtain the symbol name which is referenced in the relocation StringRef TargetName; - Symbol.getName(TargetName); + if (Symbol != Obj.end_symbols()) + Symbol->getName(TargetName); DEBUG(dbgs() << "\t\tRelType: " << RelType << " Addend: " << Addend << " TargetName: " << TargetName << "\n"); RelocationValueRef Value; // First search for the symbol in the local symbol table - SymbolTableMap::const_iterator lsi = Symbols.find(TargetName.data()); - SymbolRef::Type SymType; - Symbol.getType(SymType); + SymbolTableMap::const_iterator lsi = Symbols.end(); + SymbolRef::Type SymType = SymbolRef::ST_Unknown; + if (Symbol != Obj.end_symbols()) { + lsi = Symbols.find(TargetName.data()); + Symbol->getType(SymType); + } if (lsi != Symbols.end()) { Value.SectionID = lsi->second.first; - Value.Addend = lsi->second.second; + Value.Addend = lsi->second.second + Addend; } else { // Search for the symbol in the global symbol table - SymbolTableMap::const_iterator gsi = - GlobalSymbolTable.find(TargetName.data()); + SymbolTableMap::const_iterator gsi = GlobalSymbolTable.end(); + if (Symbol != Obj.end_symbols()) + gsi = GlobalSymbolTable.find(TargetName.data()); if (gsi != GlobalSymbolTable.end()) { Value.SectionID = gsi->second.first; - Value.Addend = gsi->second.second; + Value.Addend = gsi->second.second + Addend; } else { switch (SymType) { case SymbolRef::ST_Debug: { @@ -632,7 +827,7 @@ void RuntimeDyldELF::processRelocationRef(const ObjRelocationInfo &Rel, // and can be changed by another developers. Maybe best way is add // a new symbol type ST_Section to SymbolRef and use it. section_iterator si(Obj.end_sections()); - Symbol.getSection(si); + Symbol->getSection(si); if (si == Obj.end_sections()) llvm_unreachable("Symbol section not found, bad object file format!"); DEBUG(dbgs() << "\t\tThis is section symbol\n"); @@ -657,21 +852,73 @@ void RuntimeDyldELF::processRelocationRef(const ObjRelocationInfo &Rel, } } } - DEBUG(dbgs() << "\t\tRel.SectionID: " << Rel.SectionID - << " Rel.Offset: " << Rel.Offset + uint64_t Offset; + Check(RelI.getOffset(Offset)); + + DEBUG(dbgs() << "\t\tSectionID: " << SectionID + << " Offset: " << Offset << "\n"); - if (Arch == Triple::arm && + if (Arch == Triple::aarch64 && + (RelType == ELF::R_AARCH64_CALL26 || + RelType == ELF::R_AARCH64_JUMP26)) { + // This is an AArch64 branch relocation, need to use a stub function. + DEBUG(dbgs() << "\t\tThis is an AArch64 branch relocation."); + SectionEntry &Section = Sections[SectionID]; + + // Look for an existing stub. + StubMap::const_iterator i = Stubs.find(Value); + if (i != Stubs.end()) { + resolveRelocation(Section, Offset, + (uint64_t)Section.Address + i->second, RelType, 0); + DEBUG(dbgs() << " Stub function found\n"); + } else { + // Create a new stub function. + DEBUG(dbgs() << " Create a new stub function\n"); + Stubs[Value] = Section.StubOffset; + uint8_t *StubTargetAddr = createStubFunction(Section.Address + + Section.StubOffset); + + RelocationEntry REmovz_g3(SectionID, + StubTargetAddr - Section.Address, + ELF::R_AARCH64_MOVW_UABS_G3, Value.Addend); + RelocationEntry REmovk_g2(SectionID, + StubTargetAddr - Section.Address + 4, + ELF::R_AARCH64_MOVW_UABS_G2_NC, Value.Addend); + RelocationEntry REmovk_g1(SectionID, + StubTargetAddr - Section.Address + 8, + ELF::R_AARCH64_MOVW_UABS_G1_NC, Value.Addend); + RelocationEntry REmovk_g0(SectionID, + StubTargetAddr - Section.Address + 12, + ELF::R_AARCH64_MOVW_UABS_G0_NC, Value.Addend); + + if (Value.SymbolName) { + addRelocationForSymbol(REmovz_g3, Value.SymbolName); + addRelocationForSymbol(REmovk_g2, Value.SymbolName); + addRelocationForSymbol(REmovk_g1, Value.SymbolName); + addRelocationForSymbol(REmovk_g0, Value.SymbolName); + } else { + addRelocationForSection(REmovz_g3, Value.SectionID); + addRelocationForSection(REmovk_g2, Value.SectionID); + addRelocationForSection(REmovk_g1, Value.SectionID); + addRelocationForSection(REmovk_g0, Value.SectionID); + } + resolveRelocation(Section, Offset, + (uint64_t)Section.Address + Section.StubOffset, + RelType, 0); + Section.StubOffset += getMaxStubSize(); + } + } else if (Arch == Triple::arm && (RelType == ELF::R_ARM_PC24 || RelType == ELF::R_ARM_CALL || RelType == ELF::R_ARM_JUMP24)) { // This is an ARM branch relocation, need to use a stub function. DEBUG(dbgs() << "\t\tThis is an ARM branch relocation."); - SectionEntry &Section = Sections[Rel.SectionID]; + SectionEntry &Section = Sections[SectionID]; // Look for an existing stub. StubMap::const_iterator i = Stubs.find(Value); if (i != Stubs.end()) { - resolveRelocation(Section, Rel.Offset, + resolveRelocation(Section, Offset, (uint64_t)Section.Address + i->second, RelType, 0); DEBUG(dbgs() << " Stub function found\n"); } else { @@ -680,14 +927,14 @@ void RuntimeDyldELF::processRelocationRef(const ObjRelocationInfo &Rel, Stubs[Value] = Section.StubOffset; uint8_t *StubTargetAddr = createStubFunction(Section.Address + Section.StubOffset); - RelocationEntry RE(Rel.SectionID, StubTargetAddr - Section.Address, - ELF::R_ARM_ABS32, Value.Addend); + RelocationEntry RE(SectionID, StubTargetAddr - Section.Address, + ELF::R_ARM_PRIVATE_0, Value.Addend); if (Value.SymbolName) addRelocationForSymbol(RE, Value.SymbolName); else addRelocationForSection(RE, Value.SectionID); - resolveRelocation(Section, Rel.Offset, + resolveRelocation(Section, Offset, (uint64_t)Section.Address + Section.StubOffset, RelType, 0); Section.StubOffset += getMaxStubSize(); @@ -696,8 +943,8 @@ void RuntimeDyldELF::processRelocationRef(const ObjRelocationInfo &Rel, RelType == ELF::R_MIPS_26) { // This is an Mips branch relocation, need to use a stub function. DEBUG(dbgs() << "\t\tThis is a Mips branch relocation."); - SectionEntry &Section = Sections[Rel.SectionID]; - uint8_t *Target = Section.Address + Rel.Offset; + SectionEntry &Section = Sections[SectionID]; + uint8_t *Target = Section.Address + Offset; uint32_t *TargetAddress = (uint32_t *)Target; // Extract the addend from the instruction. @@ -708,7 +955,7 @@ void RuntimeDyldELF::processRelocationRef(const ObjRelocationInfo &Rel, // Look up for existing stub. StubMap::const_iterator i = Stubs.find(Value); if (i != Stubs.end()) { - resolveRelocation(Section, Rel.Offset, + resolveRelocation(Section, Offset, (uint64_t)Section.Address + i->second, RelType, 0); DEBUG(dbgs() << " Stub function found\n"); } else { @@ -719,12 +966,12 @@ void RuntimeDyldELF::processRelocationRef(const ObjRelocationInfo &Rel, Section.StubOffset); // Creating Hi and Lo relocations for the filled stub instructions. - RelocationEntry REHi(Rel.SectionID, + RelocationEntry REHi(SectionID, StubTargetAddr - Section.Address, - ELF::R_MIPS_HI16, Value.Addend); - RelocationEntry RELo(Rel.SectionID, + ELF::R_MIPS_UNUSED1, Value.Addend); + RelocationEntry RELo(SectionID, StubTargetAddr - Section.Address + 4, - ELF::R_MIPS_LO16, Value.Addend); + ELF::R_MIPS_UNUSED2, Value.Addend); if (Value.SymbolName) { addRelocationForSymbol(REHi, Value.SymbolName); @@ -734,18 +981,18 @@ void RuntimeDyldELF::processRelocationRef(const ObjRelocationInfo &Rel, addRelocationForSection(RELo, Value.SectionID); } - resolveRelocation(Section, Rel.Offset, + resolveRelocation(Section, Offset, (uint64_t)Section.Address + Section.StubOffset, RelType, 0); Section.StubOffset += getMaxStubSize(); } - } else if (Arch == Triple::ppc64) { + } else if (Arch == Triple::ppc64 || Arch == Triple::ppc64le) { if (RelType == ELF::R_PPC64_REL24) { // A PPC branch relocation will need a stub function if the target is // an external symbol (Symbol::ST_Unknown) or if the target address // is not within the signed 24-bits branch address. - SectionEntry &Section = Sections[Rel.SectionID]; - uint8_t *Target = Section.Address + Rel.Offset; + SectionEntry &Section = Sections[SectionID]; + uint8_t *Target = Section.Address + Offset; bool RangeOverflow = false; if (SymType != SymbolRef::ST_Unknown) { // A function call may points to the .opd entry, so the final symbol value @@ -755,7 +1002,7 @@ void RuntimeDyldELF::processRelocationRef(const ObjRelocationInfo &Rel, int32_t delta = static_cast<int32_t>(Target - RelocTarget); // If it is within 24-bits branch range, just set the branch target if (SignExtend32<24>(delta) == delta) { - RelocationEntry RE(Rel.SectionID, Rel.Offset, RelType, Value.Addend); + RelocationEntry RE(SectionID, Offset, RelType, Value.Addend); if (Value.SymbolName) addRelocationForSymbol(RE, Value.SymbolName); else @@ -770,7 +1017,7 @@ void RuntimeDyldELF::processRelocationRef(const ObjRelocationInfo &Rel, StubMap::const_iterator i = Stubs.find(Value); if (i != Stubs.end()) { // Symbol function stub already created, just relocate to it - resolveRelocation(Section, Rel.Offset, + resolveRelocation(Section, Offset, (uint64_t)Section.Address + i->second, RelType, 0); DEBUG(dbgs() << " Stub function found\n"); } else { @@ -779,21 +1026,21 @@ void RuntimeDyldELF::processRelocationRef(const ObjRelocationInfo &Rel, Stubs[Value] = Section.StubOffset; uint8_t *StubTargetAddr = createStubFunction(Section.Address + Section.StubOffset); - RelocationEntry RE(Rel.SectionID, StubTargetAddr - Section.Address, + RelocationEntry RE(SectionID, StubTargetAddr - Section.Address, ELF::R_PPC64_ADDR64, Value.Addend); // Generates the 64-bits address loads as exemplified in section // 4.5.1 in PPC64 ELF ABI. - RelocationEntry REhst(Rel.SectionID, + RelocationEntry REhst(SectionID, StubTargetAddr - Section.Address + 2, ELF::R_PPC64_ADDR16_HIGHEST, Value.Addend); - RelocationEntry REhr(Rel.SectionID, + RelocationEntry REhr(SectionID, StubTargetAddr - Section.Address + 6, ELF::R_PPC64_ADDR16_HIGHER, Value.Addend); - RelocationEntry REh(Rel.SectionID, + RelocationEntry REh(SectionID, StubTargetAddr - Section.Address + 14, ELF::R_PPC64_ADDR16_HI, Value.Addend); - RelocationEntry REl(Rel.SectionID, + RelocationEntry REl(SectionID, StubTargetAddr - Section.Address + 18, ELF::R_PPC64_ADDR16_LO, Value.Addend); @@ -809,7 +1056,7 @@ void RuntimeDyldELF::processRelocationRef(const ObjRelocationInfo &Rel, addRelocationForSection(REl, Value.SectionID); } - resolveRelocation(Section, Rel.Offset, + resolveRelocation(Section, Offset, (uint64_t)Section.Address + Section.StubOffset, RelType, 0); if (SymType == SymbolRef::ST_Unknown) @@ -819,7 +1066,7 @@ void RuntimeDyldELF::processRelocationRef(const ObjRelocationInfo &Rel, } } } else { - RelocationEntry RE(Rel.SectionID, Rel.Offset, RelType, Value.Addend); + RelocationEntry RE(SectionID, Offset, RelType, Value.Addend); // Extra check to avoid relocation againt empty symbols (usually // the R_PPC64_TOC). if (Value.SymbolName && !TargetName.empty()) @@ -827,8 +1074,55 @@ void RuntimeDyldELF::processRelocationRef(const ObjRelocationInfo &Rel, else addRelocationForSection(RE, Value.SectionID); } + } else if (Arch == Triple::systemz && + (RelType == ELF::R_390_PLT32DBL || + RelType == ELF::R_390_GOTENT)) { + // Create function stubs for both PLT and GOT references, regardless of + // whether the GOT reference is to data or code. The stub contains the + // full address of the symbol, as needed by GOT references, and the + // executable part only adds an overhead of 8 bytes. + // + // We could try to conserve space by allocating the code and data + // parts of the stub separately. However, as things stand, we allocate + // a stub for every relocation, so using a GOT in JIT code should be + // no less space efficient than using an explicit constant pool. + DEBUG(dbgs() << "\t\tThis is a SystemZ indirect relocation."); + SectionEntry &Section = Sections[SectionID]; + + // Look for an existing stub. + StubMap::const_iterator i = Stubs.find(Value); + uintptr_t StubAddress; + if (i != Stubs.end()) { + StubAddress = uintptr_t(Section.Address) + i->second; + DEBUG(dbgs() << " Stub function found\n"); + } else { + // Create a new stub function. + DEBUG(dbgs() << " Create a new stub function\n"); + + uintptr_t BaseAddress = uintptr_t(Section.Address); + uintptr_t StubAlignment = getStubAlignment(); + StubAddress = (BaseAddress + Section.StubOffset + + StubAlignment - 1) & -StubAlignment; + unsigned StubOffset = StubAddress - BaseAddress; + + Stubs[Value] = StubOffset; + createStubFunction((uint8_t *)StubAddress); + RelocationEntry RE(SectionID, StubOffset + 8, + ELF::R_390_64, Value.Addend - Addend); + if (Value.SymbolName) + addRelocationForSymbol(RE, Value.SymbolName); + else + addRelocationForSection(RE, Value.SectionID); + Section.StubOffset = StubOffset + getMaxStubSize(); + } + + if (RelType == ELF::R_390_GOTENT) + resolveRelocation(Section, Offset, StubAddress + 8, + ELF::R_390_PC32DBL, Addend); + else + resolveRelocation(Section, Offset, StubAddress, RelType, Addend); } else { - RelocationEntry RE(Rel.SectionID, Rel.Offset, RelType, Value.Addend); + RelocationEntry RE(SectionID, Offset, RelType, Value.Addend); if (Value.SymbolName) addRelocationForSymbol(RE, Value.SymbolName); else @@ -836,13 +1130,6 @@ void RuntimeDyldELF::processRelocationRef(const ObjRelocationInfo &Rel, } } -unsigned RuntimeDyldELF::getCommonSymbolAlignment(const SymbolRef &Sym) { - // In ELF, the value of an SHN_COMMON symbol is its alignment requirement. - uint64_t Align; - Check(Sym.getValue(Align)); - return Align; -} - bool RuntimeDyldELF::isCompatibleFormat(const ObjectBuffer *Buffer) const { if (Buffer->getBufferSize() < strlen(ELF::ElfMagic)) return false; diff --git a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.h b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.h index 07e704b..794c7ec 100644 --- a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.h +++ b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.h @@ -31,7 +31,12 @@ namespace { } // end anonymous namespace class RuntimeDyldELF : public RuntimeDyldImpl { -protected: + void resolveRelocation(const SectionEntry &Section, + uint64_t Offset, + uint64_t Value, + uint32_t Type, + int64_t Addend); + void resolveX86_64Relocation(const SectionEntry &Section, uint64_t Offset, uint64_t Value, @@ -44,6 +49,12 @@ protected: uint32_t Type, int32_t Addend); + void resolveAArch64Relocation(const SectionEntry &Section, + uint64_t Offset, + uint64_t Value, + uint32_t Type, + int64_t Addend); + void resolveARMRelocation(const SectionEntry &Section, uint64_t Offset, uint32_t Value, @@ -62,21 +73,11 @@ protected: uint32_t Type, int64_t Addend); - virtual void resolveRelocation(const SectionEntry &Section, - uint64_t Offset, - uint64_t Value, - uint32_t Type, - int64_t Addend); - - virtual void processRelocationRef(const ObjRelocationInfo &Rel, - ObjectImage &Obj, - ObjSectionToIDMap &ObjSectionToID, - const SymbolTableMap &Symbols, - StubMap &Stubs); - - unsigned getCommonSymbolAlignment(const SymbolRef &Sym); - - virtual ObjectImage *createObjectImage(ObjectBuffer *InputBuffer); + void resolveSystemZRelocation(const SectionEntry &Section, + uint64_t Offset, + uint64_t Value, + uint32_t Type, + int64_t Addend); uint64_t findPPC64TOC() const; void findOPDEntrySection(ObjectImage &Obj, @@ -84,12 +85,19 @@ protected: RelocationValueRef &Rel); public: - RuntimeDyldELF(RTDyldMemoryManager *mm) - : RuntimeDyldImpl(mm) {} + RuntimeDyldELF(RTDyldMemoryManager *mm) : RuntimeDyldImpl(mm) {} + virtual void resolveRelocation(const RelocationEntry &RE, uint64_t Value); + virtual void processRelocationRef(unsigned SectionID, + RelocationRef RelI, + ObjectImage &Obj, + ObjSectionToIDMap &ObjSectionToID, + const SymbolTableMap &Symbols, + StubMap &Stubs); + virtual bool isCompatibleFormat(const ObjectBuffer *Buffer) const; + virtual ObjectImage *createObjectImage(ObjectBuffer *InputBuffer); + virtual StringRef getEHFrameSection(); virtual ~RuntimeDyldELF(); - - bool isCompatibleFormat(const ObjectBuffer *Buffer) const; }; } // end namespace llvm diff --git a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldImpl.h b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldImpl.h index f100994..14d945b 100644 --- a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldImpl.h +++ b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldImpl.h @@ -49,7 +49,7 @@ public: /// Address - address in the linker's memory where the section resides. uint8_t *Address; - /// Size - section size. + /// Size - section size. Doesn't include the stubs. size_t Size; /// LoadAddress - the address of the section in the target process's memory. @@ -67,9 +67,9 @@ public: uintptr_t ObjAddress; SectionEntry(StringRef name, uint8_t *address, size_t size, - uintptr_t stubOffset, uintptr_t objAddress) + uintptr_t objAddress) : Name(name), Address(address), Size(size), LoadAddress((uintptr_t)address), - StubOffset(stubOffset), ObjAddress(objAddress) {} + StubOffset(size), ObjAddress(objAddress) {} }; /// RelocationEntry - used to represent relocations internally in the dynamic @@ -89,20 +89,20 @@ public: /// used to make a relocation section relative instead of symbol relative. intptr_t Addend; + /// True if this is a PCRel relocation (MachO specific). + bool IsPCRel; + + /// The size of this relocation (MachO specific). + unsigned Size; + RelocationEntry(unsigned id, uint64_t offset, uint32_t type, int64_t addend) - : SectionID(id), Offset(offset), RelType(type), Addend(addend) {} -}; + : SectionID(id), Offset(offset), RelType(type), Addend(addend), + IsPCRel(false), Size(0) {} -/// ObjRelocationInfo - relocation information as read from the object file. -/// Used to pass around data taken from object::RelocationRef, together with -/// the section to which the relocation points (represented by a SectionID). -class ObjRelocationInfo { -public: - unsigned SectionID; - uint64_t Offset; - SymbolRef Symbol; - uint64_t Type; - int64_t AdditionalInfo; + RelocationEntry(unsigned id, uint64_t offset, uint32_t type, int64_t addend, + bool IsPCRel, unsigned Size) + : SectionID(id), Offset(offset), RelType(type), Addend(addend), + IsPCRel(IsPCRel), Size(Size) {} }; class RelocationValueRef { @@ -166,16 +166,29 @@ protected: Triple::ArchType Arch; inline unsigned getMaxStubSize() { + if (Arch == Triple::aarch64) + return 20; // movz; movk; movk; movk; br if (Arch == Triple::arm || Arch == Triple::thumb) return 8; // 32-bit instruction and 32-bit address else if (Arch == Triple::mipsel || Arch == Triple::mips) return 16; - else if (Arch == Triple::ppc64) + else if (Arch == Triple::ppc64 || Arch == Triple::ppc64le) return 44; + else if (Arch == Triple::x86_64) + return 8; // GOT + else if (Arch == Triple::systemz) + return 16; else return 0; } + inline unsigned getStubAlignment() { + if (Arch == Triple::systemz) + return 8; + else + return 1; + } + bool HasError; std::string ErrorStr; @@ -194,22 +207,15 @@ protected: return (uint8_t*)Sections[SectionID].Address; } - // Subclasses can override this method to get the alignment requirement of - // a common symbol. Returns no alignment requirement if not implemented. - virtual unsigned getCommonSymbolAlignment(const SymbolRef &Sym) { - return 0; - } - - void writeInt16BE(uint8_t *Addr, uint16_t Value) { - if (sys::isLittleEndianHost()) + if (sys::IsLittleEndianHost) Value = sys::SwapByteOrder(Value); *Addr = (Value >> 8) & 0xFF; *(Addr+1) = Value & 0xFF; } void writeInt32BE(uint8_t *Addr, uint32_t Value) { - if (sys::isLittleEndianHost()) + if (sys::IsLittleEndianHost) Value = sys::SwapByteOrder(Value); *Addr = (Value >> 24) & 0xFF; *(Addr+1) = (Value >> 16) & 0xFF; @@ -218,7 +224,7 @@ protected: } void writeInt64BE(uint8_t *Addr, uint64_t Value) { - if (sys::isLittleEndianHost()) + if (sys::IsLittleEndianHost) Value = sys::SwapByteOrder(Value); *Addr = (Value >> 56) & 0xFF; *(Addr+1) = (Value >> 48) & 0xFF; @@ -269,24 +275,16 @@ protected: /// \brief Resolves relocations from Relocs list with address from Value. void resolveRelocationList(const RelocationList &Relocs, uint64_t Value); - void resolveRelocationEntry(const RelocationEntry &RE, uint64_t Value); /// \brief A object file specific relocation resolver - /// \param Section The section where the relocation is being applied - /// \param Offset The offset into the section for this relocation + /// \param RE The relocation to be resolved /// \param Value Target symbol address to apply the relocation action - /// \param Type object file specific relocation type - /// \param Addend A constant addend used to compute the value to be stored - /// into the relocatable field - virtual void resolveRelocation(const SectionEntry &Section, - uint64_t Offset, - uint64_t Value, - uint32_t Type, - int64_t Addend) = 0; + virtual void resolveRelocation(const RelocationEntry &RE, uint64_t Value) = 0; /// \brief Parses the object file relocation and stores it to Relocations /// or SymbolRelocations (this depends on the object file type). - virtual void processRelocationRef(const ObjRelocationInfo &Rel, + virtual void processRelocationRef(unsigned SectionID, + RelocationRef RelI, ObjectImage &Obj, ObjSectionToIDMap &ObjSectionToID, const SymbolTableMap &Symbols, @@ -336,6 +334,8 @@ public: StringRef getErrorString() { return ErrorStr; } virtual bool isCompatibleFormat(const ObjectBuffer *Buffer) const = 0; + + virtual StringRef getEHFrameSection(); }; } // end namespace llvm diff --git a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.cpp b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.cpp index bcc3df1..0384b32 100644 --- a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.cpp +++ b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.cpp @@ -21,16 +21,87 @@ using namespace llvm::object; namespace llvm { +static unsigned char *processFDE(unsigned char *P, intptr_t DeltaForText, intptr_t DeltaForEH) { + uint32_t Length = *((uint32_t*)P); + P += 4; + unsigned char *Ret = P + Length; + uint32_t Offset = *((uint32_t*)P); + if (Offset == 0) // is a CIE + return Ret; + + P += 4; + intptr_t FDELocation = *((intptr_t*)P); + intptr_t NewLocation = FDELocation - DeltaForText; + *((intptr_t*)P) = NewLocation; + P += sizeof(intptr_t); + + // Skip the FDE address range + P += sizeof(intptr_t); + + uint8_t Augmentationsize = *P; + P += 1; + if (Augmentationsize != 0) { + intptr_t LSDA = *((intptr_t*)P); + intptr_t NewLSDA = LSDA - DeltaForEH; + *((intptr_t*)P) = NewLSDA; + } + + return Ret; +} + +static intptr_t computeDelta(SectionEntry *A, SectionEntry *B) { + intptr_t ObjDistance = A->ObjAddress - B->ObjAddress; + intptr_t MemDistance = A->LoadAddress - B->LoadAddress; + return ObjDistance - MemDistance; +} + +StringRef RuntimeDyldMachO::getEHFrameSection() { + SectionEntry *Text = NULL; + SectionEntry *EHFrame = NULL; + SectionEntry *ExceptTab = NULL; + for (int i = 0, e = Sections.size(); i != e; ++i) { + if (Sections[i].Name == "__eh_frame") + EHFrame = &Sections[i]; + else if (Sections[i].Name == "__text") + Text = &Sections[i]; + else if (Sections[i].Name == "__gcc_except_tab") + ExceptTab = &Sections[i]; + } + if (Text == NULL || EHFrame == NULL) + return StringRef(); + + intptr_t DeltaForText = computeDelta(Text, EHFrame); + intptr_t DeltaForEH = 0; + if (ExceptTab) + DeltaForEH = computeDelta(ExceptTab, EHFrame); + + unsigned char *P = EHFrame->Address; + unsigned char *End = P + EHFrame->Size; + do { + P = processFDE(P, DeltaForText, DeltaForEH); + } while(P != End); + + return StringRef((char*)EHFrame->Address, EHFrame->Size); +} + +void RuntimeDyldMachO::resolveRelocation(const RelocationEntry &RE, + uint64_t Value) { + const SectionEntry &Section = Sections[RE.SectionID]; + return resolveRelocation(Section, RE.Offset, Value, RE.RelType, RE.Addend, + RE.IsPCRel, RE.Size); +} + void RuntimeDyldMachO::resolveRelocation(const SectionEntry &Section, uint64_t Offset, uint64_t Value, uint32_t Type, - int64_t Addend) { + int64_t Addend, + bool isPCRel, + unsigned LogSize) { uint8_t *LocalAddress = Section.Address + Offset; uint64_t FinalAddress = Section.LoadAddress + Offset; - bool isPCRel = (Type >> 24) & 1; - unsigned MachoType = (Type >> 28) & 0xf; - unsigned Size = 1 << ((Type >> 25) & 3); + unsigned MachoType = Type; + unsigned Size = 1 << LogSize; DEBUG(dbgs() << "resolveRelocation LocalAddress: " << format("%p", LocalAddress) @@ -205,89 +276,110 @@ bool RuntimeDyldMachO::resolveARMRelocation(uint8_t *LocalAddress, return false; } -void RuntimeDyldMachO::processRelocationRef(const ObjRelocationInfo &Rel, +void RuntimeDyldMachO::processRelocationRef(unsigned SectionID, + RelocationRef RelI, ObjectImage &Obj, ObjSectionToIDMap &ObjSectionToID, const SymbolTableMap &Symbols, StubMap &Stubs) { + const ObjectFile *OF = Obj.getObjectFile(); + const MachOObjectFile *MachO = static_cast<const MachOObjectFile*>(OF); + macho::RelocationEntry RE = MachO->getRelocation(RelI.getRawDataRefImpl()); - uint32_t RelType = (uint32_t) (Rel.Type & 0xffffffffL); + uint32_t RelType = MachO->getAnyRelocationType(RE); RelocationValueRef Value; - SectionEntry &Section = Sections[Rel.SectionID]; + SectionEntry &Section = Sections[SectionID]; + + bool isExtern = MachO->getPlainRelocationExternal(RE); + bool IsPCRel = MachO->getAnyRelocationPCRel(RE); + unsigned Size = MachO->getAnyRelocationLength(RE); + uint64_t Offset; + RelI.getOffset(Offset); + uint8_t *LocalAddress = Section.Address + Offset; + unsigned NumBytes = 1 << Size; + uint64_t Addend = 0; + memcpy(&Addend, LocalAddress, NumBytes); - bool isExtern = (RelType >> 27) & 1; if (isExtern) { // Obtain the symbol name which is referenced in the relocation + symbol_iterator Symbol = RelI.getSymbol(); StringRef TargetName; - const SymbolRef &Symbol = Rel.Symbol; - Symbol.getName(TargetName); + Symbol->getName(TargetName); // First search for the symbol in the local symbol table SymbolTableMap::const_iterator lsi = Symbols.find(TargetName.data()); if (lsi != Symbols.end()) { Value.SectionID = lsi->second.first; - Value.Addend = lsi->second.second; + Value.Addend = lsi->second.second + Addend; } else { // Search for the symbol in the global symbol table SymbolTableMap::const_iterator gsi = GlobalSymbolTable.find(TargetName.data()); if (gsi != GlobalSymbolTable.end()) { Value.SectionID = gsi->second.first; - Value.Addend = gsi->second.second; - } else + Value.Addend = gsi->second.second + Addend; + } else { Value.SymbolName = TargetName.data(); + Value.Addend = Addend; + } } } else { - error_code err; - uint8_t sectionIndex = static_cast<uint8_t>(RelType & 0xFF); - section_iterator si = Obj.begin_sections(), - se = Obj.end_sections(); - for (uint8_t i = 1; i < sectionIndex; i++) { - error_code err; - si.increment(err); - if (si == se) - break; - } - assert(si != se && "No section containing relocation!"); - Value.SectionID = findOrEmitSection(Obj, *si, true, ObjSectionToID); - Value.Addend = 0; - // FIXME: The size and type of the relocation determines if we can - // encode an Addend in the target location itself, and if so, how many - // bytes we should read in order to get it. We don't yet support doing - // that, and just assuming it's sizeof(intptr_t) is blatantly wrong. - //Value.Addend = *(const intptr_t *)Target; - if (Value.Addend) { - // The MachO addend is an offset from the current section. We need it - // to be an offset from the destination section - Value.Addend += Section.ObjAddress - Sections[Value.SectionID].ObjAddress; - } + SectionRef Sec = MachO->getRelocationSection(RE); + Value.SectionID = findOrEmitSection(Obj, Sec, true, ObjSectionToID); + uint64_t Addr; + Sec.getAddress(Addr); + Value.Addend = Addend - Addr; } - if (Arch == Triple::arm && (RelType & 0xf) == macho::RIT_ARM_Branch24Bit) { + if (Arch == Triple::x86_64 && RelType == macho::RIT_X86_64_GOT) { + assert(IsPCRel); + assert(Size == 2); + StubMap::const_iterator i = Stubs.find(Value); + uint8_t *Addr; + if (i != Stubs.end()) { + Addr = Section.Address + i->second; + } else { + Stubs[Value] = Section.StubOffset; + uint8_t *GOTEntry = Section.Address + Section.StubOffset; + RelocationEntry RE(SectionID, Section.StubOffset, + macho::RIT_X86_64_Unsigned, Value.Addend - 4, false, + 3); + if (Value.SymbolName) + addRelocationForSymbol(RE, Value.SymbolName); + else + addRelocationForSection(RE, Value.SectionID); + Section.StubOffset += 8; + Addr = GOTEntry; + } + resolveRelocation(Section, Offset, (uint64_t)Addr, + macho::RIT_X86_64_Unsigned, 4, true, 2); + } else if (Arch == Triple::arm && + (RelType & 0xf) == macho::RIT_ARM_Branch24Bit) { // This is an ARM branch relocation, need to use a stub function. // Look up for existing stub. StubMap::const_iterator i = Stubs.find(Value); if (i != Stubs.end()) - resolveRelocation(Section, Rel.Offset, + resolveRelocation(Section, Offset, (uint64_t)Section.Address + i->second, - RelType, 0); + RelType, 0, IsPCRel, Size); else { // Create a new stub function. Stubs[Value] = Section.StubOffset; uint8_t *StubTargetAddr = createStubFunction(Section.Address + Section.StubOffset); - RelocationEntry RE(Rel.SectionID, StubTargetAddr - Section.Address, + RelocationEntry RE(SectionID, StubTargetAddr - Section.Address, macho::RIT_Vanilla, Value.Addend); if (Value.SymbolName) addRelocationForSymbol(RE, Value.SymbolName); else addRelocationForSection(RE, Value.SectionID); - resolveRelocation(Section, Rel.Offset, + resolveRelocation(Section, Offset, (uint64_t)Section.Address + Section.StubOffset, - RelType, 0); + RelType, 0, IsPCRel, Size); Section.StubOffset += getMaxStubSize(); } } else { - RelocationEntry RE(Rel.SectionID, Rel.Offset, RelType, Value.Addend); + RelocationEntry RE(SectionID, Offset, RelType, Value.Addend, + IsPCRel, Size); if (Value.SymbolName) addRelocationForSymbol(RE, Value.SymbolName); else diff --git a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.h b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.h index 62d8487..df8d3bb 100644 --- a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.h +++ b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldMachO.h @@ -16,7 +16,7 @@ #include "RuntimeDyldImpl.h" #include "llvm/ADT/IndexedMap.h" -#include "llvm/Object/MachOObject.h" +#include "llvm/Object/MachO.h" #include "llvm/Support/Format.h" using namespace llvm; @@ -25,7 +25,6 @@ using namespace llvm::object; namespace llvm { class RuntimeDyldMachO : public RuntimeDyldImpl { -protected: bool resolveI386Relocation(uint8_t *LocalAddress, uint64_t FinalAddress, uint64_t Value, @@ -48,22 +47,25 @@ protected: unsigned Size, int64_t Addend); - virtual void processRelocationRef(const ObjRelocationInfo &Rel, + void resolveRelocation(const SectionEntry &Section, + uint64_t Offset, + uint64_t Value, + uint32_t Type, + int64_t Addend, + bool isPCRel, + unsigned Size); +public: + RuntimeDyldMachO(RTDyldMemoryManager *mm) : RuntimeDyldImpl(mm) {} + + virtual void resolveRelocation(const RelocationEntry &RE, uint64_t Value); + virtual void processRelocationRef(unsigned SectionID, + RelocationRef RelI, ObjectImage &Obj, ObjSectionToIDMap &ObjSectionToID, const SymbolTableMap &Symbols, StubMap &Stubs); - -public: - virtual void resolveRelocation(const SectionEntry &Section, - uint64_t Offset, - uint64_t Value, - uint32_t Type, - int64_t Addend); - - RuntimeDyldMachO(RTDyldMemoryManager *mm) : RuntimeDyldImpl(mm) {} - - bool isCompatibleFormat(const ObjectBuffer *Buffer) const; + virtual bool isCompatibleFormat(const ObjectBuffer *Buffer) const; + virtual StringRef getEHFrameSection(); }; } // end namespace llvm diff --git a/lib/ExecutionEngine/TargetSelect.cpp b/lib/ExecutionEngine/TargetSelect.cpp index ca4330f..558d8b3 100644 --- a/lib/ExecutionEngine/TargetSelect.cpp +++ b/lib/ExecutionEngine/TargetSelect.cpp @@ -88,6 +88,14 @@ TargetMachine *EngineBuilder::selectTarget(const Triple &TargetTriple, FeaturesStr = Features.getString(); } + // FIXME: non-iOS ARM FastISel is broken with MCJIT. + if (UseMCJIT && + TheTriple.getArch() == Triple::arm && + TheTriple.getOS() != Triple::IOS && + OptLevel == CodeGenOpt::None) { + OptLevel = CodeGenOpt::Less; + } + // Allocate a target... TargetMachine *Target = TheTarget->createTargetMachine(TheTriple.getTriple(), MCPU, FeaturesStr, |