X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FCodeGen%2FIntrinsicLowering.cpp;h=67e4e6f7626e825019875b4fa694208487e37e7b;hb=67d894ea64bc52245abf3f2ba89122a749d99c53;hp=c47a107cc33310403b15ff67ff9f4806f1a06989;hpb=0b118206bf3411722707f2e5cab8fd2eedcd50d6;p=oota-llvm.git diff --git a/lib/CodeGen/IntrinsicLowering.cpp b/lib/CodeGen/IntrinsicLowering.cpp index c47a107cc33..67e4e6f7626 100644 --- a/lib/CodeGen/IntrinsicLowering.cpp +++ b/lib/CodeGen/IntrinsicLowering.cpp @@ -2,35 +2,57 @@ // // The LLVM Compiler Infrastructure // -// This file was developed by the LLVM research group and is distributed under -// the University of Illinois Open Source License. See LICENSE.TXT for details. +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// // -// This file implements the default intrinsic lowering implementation. +// This file implements the IntrinsicLowering class. // //===----------------------------------------------------------------------===// -#include "llvm/CodeGen/IntrinsicLowering.h" #include "llvm/Constants.h" #include "llvm/DerivedTypes.h" #include "llvm/Module.h" -#include "llvm/Instructions.h" #include "llvm/Type.h" -#include - +#include "llvm/CodeGen/IntrinsicLowering.h" +#include "llvm/Support/IRBuilder.h" +#include "llvm/Support/ErrorHandling.h" +#include "llvm/Target/TargetData.h" +#include "llvm/ADT/SmallVector.h" using namespace llvm; template -static Function *EnsureFunctionExists(Module &M, const char *Name, - ArgIt ArgBegin, ArgIt ArgEnd, - const Type *RetTy) { - if (Function *F = M.getNamedFunction(Name)) return F; - // It doesn't already exist in the program, insert a new definition now. +static void EnsureFunctionExists(Module &M, const char *Name, + ArgIt ArgBegin, ArgIt ArgEnd, + const Type *RetTy) { + // Insert a correctly-typed definition now. std::vector ParamTys; for (ArgIt I = ArgBegin; I != ArgEnd; ++I) ParamTys.push_back(I->getType()); - return M.getOrInsertFunction(Name, FunctionType::get(RetTy, ParamTys, false)); + M.getOrInsertFunction(Name, FunctionType::get(RetTy, ParamTys, false)); +} + +static void EnsureFPIntrinsicsExist(Module &M, Function *Fn, + const char *FName, + const char *DName, const char *LDName) { + // Insert definitions for all the floating point types. + switch((int)Fn->arg_begin()->getType()->getTypeID()) { + case Type::FloatTyID: + EnsureFunctionExists(M, FName, Fn->arg_begin(), Fn->arg_end(), + Type::FloatTy); + break; + case Type::DoubleTyID: + EnsureFunctionExists(M, DName, Fn->arg_begin(), Fn->arg_end(), + Type::DoubleTy); + break; + case Type::X86_FP80TyID: + case Type::FP128TyID: + case Type::PPC_FP128TyID: + EnsureFunctionExists(M, LDName, Fn->arg_begin(), Fn->arg_end(), + Fn->arg_begin()->getType()); + break; + } } /// ReplaceCallWith - This function is used when we want to lower an intrinsic @@ -40,56 +62,34 @@ static Function *EnsureFunctionExists(Module &M, const char *Name, template static CallInst *ReplaceCallWith(const char *NewFn, CallInst *CI, ArgIt ArgBegin, ArgIt ArgEnd, - const Type *RetTy, Function *&FCache) { - if (!FCache) { - // If we haven't already looked up this function, check to see if the - // program already contains a function with this name. - Module *M = CI->getParent()->getParent()->getParent(); - FCache = M->getNamedFunction(NewFn); - if (!FCache) { - // It doesn't already exist in the program, insert a new definition now. - std::vector ParamTys; - for (ArgIt I = ArgBegin; I != ArgEnd; ++I) - ParamTys.push_back((*I)->getType()); - FCache = M->getOrInsertFunction(NewFn, - FunctionType::get(RetTy, ParamTys, false)); - } - } - - const FunctionType *FT = FCache->getFunctionType(); - std::vector Operands; - unsigned ArgNo = 0; - for (ArgIt I = ArgBegin; I != ArgEnd && ArgNo != FT->getNumParams(); - ++I, ++ArgNo) { - Value *Arg = *I; - if (Arg->getType() != FT->getParamType(ArgNo)) - Arg = new CastInst(Arg, FT->getParamType(ArgNo), Arg->getName(), CI); - Operands.push_back(Arg); - } - // Pass nulls into any additional arguments... - for (; ArgNo != FT->getNumParams(); ++ArgNo) - Operands.push_back(Constant::getNullValue(FT->getParamType(ArgNo))); - - std::string Name = CI->getName(); CI->setName(""); - if (FT->getReturnType() == Type::VoidTy) Name.clear(); - CallInst *NewCI = new CallInst(FCache, Operands, Name, CI); - if (!CI->use_empty()) { - Value *V = NewCI; - if (CI->getType() != NewCI->getType()) - V = new CastInst(NewCI, CI->getType(), Name, CI); - CI->replaceAllUsesWith(V); - } + const Type *RetTy) { + // If we haven't already looked up this function, check to see if the + // program already contains a function with this name. + Module *M = CI->getParent()->getParent()->getParent(); + // Get or insert the definition now. + std::vector ParamTys; + for (ArgIt I = ArgBegin; I != ArgEnd; ++I) + ParamTys.push_back((*I)->getType()); + Constant* FCache = M->getOrInsertFunction(NewFn, + FunctionType::get(RetTy, ParamTys, false)); + + IRBuilder<> Builder(CI->getParent(), CI); + SmallVector Args(ArgBegin, ArgEnd); + CallInst *NewCI = Builder.CreateCall(FCache, Args.begin(), Args.end()); + NewCI->setName(CI->getName()); + if (!CI->use_empty()) + CI->replaceAllUsesWith(NewCI); return NewCI; } -void DefaultIntrinsicLowering::AddPrototypes(Module &M) { +void IntrinsicLowering::AddPrototypes(Module &M) { for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) - if (I->isExternal() && !I->use_empty()) + if (I->isDeclaration() && !I->use_empty()) switch (I->getIntrinsicID()) { default: break; case Intrinsic::setjmp: EnsureFunctionExists(M, "setjmp", I->arg_begin(), I->arg_end(), - Type::IntTy); + Type::Int32Ty); break; case Intrinsic::longjmp: EnsureFunctionExists(M, "longjmp", I->arg_begin(), I->arg_end(), @@ -100,133 +100,148 @@ void DefaultIntrinsicLowering::AddPrototypes(Module &M) { Type::VoidTy); break; case Intrinsic::memcpy: - EnsureFunctionExists(M, "memcpy", I->arg_begin(), --I->arg_end(), - I->arg_begin()->getType()); + M.getOrInsertFunction("memcpy", PointerType::getUnqual(Type::Int8Ty), + PointerType::getUnqual(Type::Int8Ty), + PointerType::getUnqual(Type::Int8Ty), + TD.getIntPtrType(), (Type *)0); break; case Intrinsic::memmove: - EnsureFunctionExists(M, "memmove", I->arg_begin(), --I->arg_end(), - I->arg_begin()->getType()); + M.getOrInsertFunction("memmove", PointerType::getUnqual(Type::Int8Ty), + PointerType::getUnqual(Type::Int8Ty), + PointerType::getUnqual(Type::Int8Ty), + TD.getIntPtrType(), (Type *)0); break; case Intrinsic::memset: - M.getOrInsertFunction("memset", PointerType::get(Type::SByteTy), - PointerType::get(Type::SByteTy), - Type::IntTy, (--(--I->arg_end()))->getType(), - (Type *)0); + M.getOrInsertFunction("memset", PointerType::getUnqual(Type::Int8Ty), + PointerType::getUnqual(Type::Int8Ty), + Type::Int32Ty, + TD.getIntPtrType(), (Type *)0); break; - case Intrinsic::isunordered_f32: - case Intrinsic::isunordered_f64: - EnsureFunctionExists(M, "isunordered", I->arg_begin(), I->arg_end(), - Type::BoolTy); + case Intrinsic::sqrt: + EnsureFPIntrinsicsExist(M, I, "sqrtf", "sqrt", "sqrtl"); break; - case Intrinsic::sqrt_f32: - case Intrinsic::sqrt_f64: - if(I->arg_begin()->getType() == Type::FloatTy) - EnsureFunctionExists(M, "sqrtf", I->arg_begin(), I->arg_end(), - Type::FloatTy); - else - EnsureFunctionExists(M, "sqrt", I->arg_begin(), I->arg_end(), - Type::DoubleTy); + case Intrinsic::sin: + EnsureFPIntrinsicsExist(M, I, "sinf", "sin", "sinl"); + break; + case Intrinsic::cos: + EnsureFPIntrinsicsExist(M, I, "cosf", "cos", "cosl"); + break; + case Intrinsic::pow: + EnsureFPIntrinsicsExist(M, I, "powf", "pow", "powl"); + break; + case Intrinsic::log: + EnsureFPIntrinsicsExist(M, I, "logf", "log", "logl"); + break; + case Intrinsic::log2: + EnsureFPIntrinsicsExist(M, I, "log2f", "log2", "log2l"); + break; + case Intrinsic::log10: + EnsureFPIntrinsicsExist(M, I, "log10f", "log10", "log10l"); + break; + case Intrinsic::exp: + EnsureFPIntrinsicsExist(M, I, "expf", "exp", "expl"); + break; + case Intrinsic::exp2: + EnsureFPIntrinsicsExist(M, I, "exp2f", "exp2", "exp2l"); break; } } /// LowerBSWAP - Emit the code to lower bswap of V before the specified /// instruction IP. -static Value *LowerBSWAP(Value *V, Instruction *IP) { +static Value *LowerBSWAP(LLVMContext &Context, Value *V, Instruction *IP) { assert(V->getType()->isInteger() && "Can't bswap a non-integer type!"); - const Type *DestTy = V->getType(); - - // Force to unsigned so that the shift rights are logical. - if (DestTy->isSigned()) - V = new CastInst(V, DestTy->getUnsignedVersion(), V->getName(), IP); - unsigned BitSize = V->getType()->getPrimitiveSizeInBits(); + IRBuilder<> Builder(IP->getParent(), IP); + switch(BitSize) { - default: assert(0 && "Unhandled type size of value to byteswap!"); + default: llvm_unreachable("Unhandled type size of value to byteswap!"); case 16: { - Value *Tmp1 = new ShiftInst(Instruction::Shl, V, - ConstantInt::get(Type::UByteTy,8),"bswap.2",IP); - Value *Tmp2 = new ShiftInst(Instruction::Shr, V, - ConstantInt::get(Type::UByteTy,8),"bswap.1",IP); - V = BinaryOperator::createOr(Tmp1, Tmp2, "bswap.i16", IP); + Value *Tmp1 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 8), + "bswap.2"); + Value *Tmp2 = Builder.CreateLShr(V, ConstantInt::get(V->getType(), 8), + "bswap.1"); + V = Builder.CreateOr(Tmp1, Tmp2, "bswap.i16"); break; } case 32: { - Value *Tmp4 = new ShiftInst(Instruction::Shl, V, - ConstantInt::get(Type::UByteTy,24),"bswap.4", IP); - Value *Tmp3 = new ShiftInst(Instruction::Shl, V, - ConstantInt::get(Type::UByteTy,8),"bswap.3",IP); - Value *Tmp2 = new ShiftInst(Instruction::Shr, V, - ConstantInt::get(Type::UByteTy,8),"bswap.2",IP); - Value *Tmp1 = new ShiftInst(Instruction::Shr, V, - ConstantInt::get(Type::UByteTy,24),"bswap.1", IP); - Tmp3 = BinaryOperator::createAnd(Tmp3, - ConstantUInt::get(Type::UIntTy, 0xFF0000), - "bswap.and3", IP); - Tmp2 = BinaryOperator::createAnd(Tmp2, - ConstantUInt::get(Type::UIntTy, 0xFF00), - "bswap.and2", IP); - Tmp4 = BinaryOperator::createOr(Tmp4, Tmp3, "bswap.or1", IP); - Tmp2 = BinaryOperator::createOr(Tmp2, Tmp1, "bswap.or2", IP); - V = BinaryOperator::createOr(Tmp4, Tmp3, "bswap.i32", IP); + Value *Tmp4 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 24), + "bswap.4"); + Value *Tmp3 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 8), + "bswap.3"); + Value *Tmp2 = Builder.CreateLShr(V, ConstantInt::get(V->getType(), 8), + "bswap.2"); + Value *Tmp1 = Builder.CreateLShr(V,ConstantInt::get(V->getType(), 24), + "bswap.1"); + Tmp3 = Builder.CreateAnd(Tmp3, + ConstantInt::get(Type::Int32Ty, 0xFF0000), + "bswap.and3"); + Tmp2 = Builder.CreateAnd(Tmp2, + ConstantInt::get(Type::Int32Ty, 0xFF00), + "bswap.and2"); + Tmp4 = Builder.CreateOr(Tmp4, Tmp3, "bswap.or1"); + Tmp2 = Builder.CreateOr(Tmp2, Tmp1, "bswap.or2"); + V = Builder.CreateOr(Tmp4, Tmp2, "bswap.i32"); break; } case 64: { - Value *Tmp8 = new ShiftInst(Instruction::Shl, V, - ConstantInt::get(Type::UByteTy,56),"bswap.8", IP); - Value *Tmp7 = new ShiftInst(Instruction::Shl, V, - ConstantInt::get(Type::UByteTy,40),"bswap.7", IP); - Value *Tmp6 = new ShiftInst(Instruction::Shl, V, - ConstantInt::get(Type::UByteTy,24),"bswap.6", IP); - Value *Tmp5 = new ShiftInst(Instruction::Shl, V, - ConstantInt::get(Type::UByteTy,8),"bswap.5",IP); - Value *Tmp4 = new ShiftInst(Instruction::Shr, V, - ConstantInt::get(Type::UByteTy,8),"bswap.4",IP); - Value *Tmp3 = new ShiftInst(Instruction::Shr, V, - ConstantInt::get(Type::UByteTy,24),"bswap.3", IP); - Value *Tmp2 = new ShiftInst(Instruction::Shr, V, - ConstantInt::get(Type::UByteTy,40),"bswap.2", IP); - Value *Tmp1 = new ShiftInst(Instruction::Shr, V, - ConstantInt::get(Type::UByteTy,56),"bswap.1", IP); - Tmp7 = BinaryOperator::createAnd(Tmp7, - ConstantUInt::get(Type::ULongTy, 0xFF000000000000ULL), - "bswap.and7", IP); - Tmp6 = BinaryOperator::createAnd(Tmp6, - ConstantUInt::get(Type::ULongTy, 0xFF0000000000ULL), - "bswap.and6", IP); - Tmp5 = BinaryOperator::createAnd(Tmp5, - ConstantUInt::get(Type::ULongTy, 0xFF00000000ULL), - "bswap.and5", IP); - Tmp4 = BinaryOperator::createAnd(Tmp4, - ConstantUInt::get(Type::ULongTy, 0xFF000000ULL), - "bswap.and4", IP); - Tmp3 = BinaryOperator::createAnd(Tmp3, - ConstantUInt::get(Type::ULongTy, 0xFF0000ULL), - "bswap.and3", IP); - Tmp2 = BinaryOperator::createAnd(Tmp2, - ConstantUInt::get(Type::ULongTy, 0xFF00ULL), - "bswap.and2", IP); - Tmp8 = BinaryOperator::createOr(Tmp8, Tmp7, "bswap.or1", IP); - Tmp6 = BinaryOperator::createOr(Tmp6, Tmp5, "bswap.or2", IP); - Tmp4 = BinaryOperator::createOr(Tmp4, Tmp3, "bswap.or3", IP); - Tmp2 = BinaryOperator::createOr(Tmp2, Tmp1, "bswap.or4", IP); - Tmp8 = BinaryOperator::createOr(Tmp8, Tmp6, "bswap.or5", IP); - Tmp4 = BinaryOperator::createOr(Tmp4, Tmp2, "bswap.or6", IP); - V = BinaryOperator::createOr(Tmp8, Tmp4, "bswap.i64", IP); + Value *Tmp8 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 56), + "bswap.8"); + Value *Tmp7 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 40), + "bswap.7"); + Value *Tmp6 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 24), + "bswap.6"); + Value *Tmp5 = Builder.CreateShl(V, ConstantInt::get(V->getType(), 8), + "bswap.5"); + Value* Tmp4 = Builder.CreateLShr(V, ConstantInt::get(V->getType(), 8), + "bswap.4"); + Value* Tmp3 = Builder.CreateLShr(V, + ConstantInt::get(V->getType(), 24), + "bswap.3"); + Value* Tmp2 = Builder.CreateLShr(V, + ConstantInt::get(V->getType(), 40), + "bswap.2"); + Value* Tmp1 = Builder.CreateLShr(V, + ConstantInt::get(V->getType(), 56), + "bswap.1"); + Tmp7 = Builder.CreateAnd(Tmp7, + ConstantInt::get(Type::Int64Ty, + 0xFF000000000000ULL), + "bswap.and7"); + Tmp6 = Builder.CreateAnd(Tmp6, + ConstantInt::get(Type::Int64Ty, + 0xFF0000000000ULL), + "bswap.and6"); + Tmp5 = Builder.CreateAnd(Tmp5, + ConstantInt::get(Type::Int64Ty, 0xFF00000000ULL), + "bswap.and5"); + Tmp4 = Builder.CreateAnd(Tmp4, + ConstantInt::get(Type::Int64Ty, 0xFF000000ULL), + "bswap.and4"); + Tmp3 = Builder.CreateAnd(Tmp3, + ConstantInt::get(Type::Int64Ty, 0xFF0000ULL), + "bswap.and3"); + Tmp2 = Builder.CreateAnd(Tmp2, + ConstantInt::get(Type::Int64Ty, 0xFF00ULL), + "bswap.and2"); + Tmp8 = Builder.CreateOr(Tmp8, Tmp7, "bswap.or1"); + Tmp6 = Builder.CreateOr(Tmp6, Tmp5, "bswap.or2"); + Tmp4 = Builder.CreateOr(Tmp4, Tmp3, "bswap.or3"); + Tmp2 = Builder.CreateOr(Tmp2, Tmp1, "bswap.or4"); + Tmp8 = Builder.CreateOr(Tmp8, Tmp6, "bswap.or5"); + Tmp4 = Builder.CreateOr(Tmp4, Tmp2, "bswap.or6"); + V = Builder.CreateOr(Tmp8, Tmp4, "bswap.i64"); break; } } - - if (V->getType() != DestTy) - V = new CastInst(V, DestTy, V->getName(), IP); return V; } /// LowerCTPOP - Emit the code to lower ctpop of V before the specified /// instruction IP. -static Value *LowerCTPOP(Value *V, Instruction *IP) { +static Value *LowerCTPOP(LLVMContext &Context, Value *V, Instruction *IP) { assert(V->getType()->isInteger() && "Can't ctpop a non-integer type!"); static const uint64_t MaskValues[6] = { @@ -235,76 +250,96 @@ static Value *LowerCTPOP(Value *V, Instruction *IP) { 0x0000FFFF0000FFFFULL, 0x00000000FFFFFFFFULL }; - const Type *DestTy = V->getType(); - - // Force to unsigned so that the shift rights are logical. - if (DestTy->isSigned()) - V = new CastInst(V, DestTy->getUnsignedVersion(), V->getName(), IP); + IRBuilder<> Builder(IP->getParent(), IP); unsigned BitSize = V->getType()->getPrimitiveSizeInBits(); - for (unsigned i = 1, ct = 0; i != BitSize; i <<= 1, ++ct) { - Value *MaskCst = - ConstantExpr::getCast(ConstantUInt::get(Type::ULongTy, - MaskValues[ct]), V->getType()); - Value *LHS = BinaryOperator::createAnd(V, MaskCst, "cppop.and1", IP); - Value *VShift = new ShiftInst(Instruction::Shr, V, - ConstantInt::get(Type::UByteTy, i), "ctpop.sh", IP); - Value *RHS = BinaryOperator::createAnd(VShift, MaskCst, "cppop.and2", IP); - V = BinaryOperator::createAdd(LHS, RHS, "ctpop.step", IP); + unsigned WordSize = (BitSize + 63) / 64; + Value *Count = ConstantInt::get(V->getType(), 0); + + for (unsigned n = 0; n < WordSize; ++n) { + Value *PartValue = V; + for (unsigned i = 1, ct = 0; i < (BitSize>64 ? 64 : BitSize); + i <<= 1, ++ct) { + Value *MaskCst = ConstantInt::get(V->getType(), MaskValues[ct]); + Value *LHS = Builder.CreateAnd(PartValue, MaskCst, "cppop.and1"); + Value *VShift = Builder.CreateLShr(PartValue, + ConstantInt::get(V->getType(), i), + "ctpop.sh"); + Value *RHS = Builder.CreateAnd(VShift, MaskCst, "cppop.and2"); + PartValue = Builder.CreateAdd(LHS, RHS, "ctpop.step"); + } + Count = Builder.CreateAdd(PartValue, Count, "ctpop.part"); + if (BitSize > 64) { + V = Builder.CreateLShr(V, ConstantInt::get(V->getType(), 64), + "ctpop.part.sh"); + BitSize -= 64; + } } - if (V->getType() != DestTy) - V = new CastInst(V, DestTy, V->getName(), IP); - return V; + return Count; } /// LowerCTLZ - Emit the code to lower ctlz of V before the specified /// instruction IP. -static Value *LowerCTLZ(Value *V, Instruction *IP) { - const Type *DestTy = V->getType(); +static Value *LowerCTLZ(LLVMContext &Context, Value *V, Instruction *IP) { - // Force to unsigned so that the shift rights are logical. - if (DestTy->isSigned()) - V = new CastInst(V, DestTy->getUnsignedVersion(), V->getName(), IP); + IRBuilder<> Builder(IP->getParent(), IP); unsigned BitSize = V->getType()->getPrimitiveSizeInBits(); - for (unsigned i = 1; i != BitSize; i <<= 1) { - Value *ShVal = ConstantInt::get(Type::UByteTy, i); - ShVal = new ShiftInst(Instruction::Shr, V, ShVal, "ctlz.sh", IP); - V = BinaryOperator::createOr(V, ShVal, "ctlz.step", IP); + for (unsigned i = 1; i < BitSize; i <<= 1) { + Value *ShVal = ConstantInt::get(V->getType(), i); + ShVal = Builder.CreateLShr(V, ShVal, "ctlz.sh"); + V = Builder.CreateOr(V, ShVal, "ctlz.step"); } - if (V->getType() != DestTy) - V = new CastInst(V, DestTy, V->getName(), IP); - - V = BinaryOperator::createNot(V, "", IP); - return LowerCTPOP(V, IP); + V = Builder.CreateNot(V); + return LowerCTPOP(Context, V, IP); } +static void ReplaceFPIntrinsicWithCall(CallInst *CI, const char *Fname, + const char *Dname, + const char *LDname) { + switch (CI->getOperand(1)->getType()->getTypeID()) { + default: llvm_unreachable("Invalid type in intrinsic"); + case Type::FloatTyID: + ReplaceCallWith(Fname, CI, CI->op_begin() + 1, CI->op_end(), + Type::FloatTy); + break; + case Type::DoubleTyID: + ReplaceCallWith(Dname, CI, CI->op_begin() + 1, CI->op_end(), + Type::DoubleTy); + break; + case Type::X86_FP80TyID: + case Type::FP128TyID: + case Type::PPC_FP128TyID: + ReplaceCallWith(LDname, CI, CI->op_begin() + 1, CI->op_end(), + CI->getOperand(1)->getType()); + break; + } +} +void IntrinsicLowering::LowerIntrinsicCall(CallInst *CI) { + IRBuilder<> Builder(CI->getParent(), CI); + LLVMContext &Context = CI->getContext(); -void DefaultIntrinsicLowering::LowerIntrinsicCall(CallInst *CI) { Function *Callee = CI->getCalledFunction(); assert(Callee && "Cannot lower an indirect call!"); switch (Callee->getIntrinsicID()) { case Intrinsic::not_intrinsic: - std::cerr << "Cannot lower a call to a non-intrinsic function '" - << Callee->getName() << "'!\n"; - abort(); + llvm_report_error("Cannot lower a call to a non-intrinsic function '"+ + Callee->getName() + "'!"); default: - std::cerr << "Error: Code generator does not support intrinsic function '" - << Callee->getName() << "'!\n"; - abort(); + llvm_report_error("Code generator does not support intrinsic function '"+ + Callee->getName()+"'!"); // The setjmp/longjmp intrinsics should only exist in the code if it was // never optimized (ie, right out of the CFE), or if it has been hacked on // by the lowerinvoke pass. In both cases, the right thing to do is to // convert the call to an explicit setjmp or longjmp call. case Intrinsic::setjmp: { - static Function *SetjmpFCache = 0; - Value *V = ReplaceCallWith("setjmp", CI, CI->op_begin()+1, CI->op_end(), - Type::IntTy, SetjmpFCache); + Value *V = ReplaceCallWith("setjmp", CI, CI->op_begin() + 1, CI->op_end(), + Type::Int32Ty); if (CI->getType() != Type::VoidTy) CI->replaceAllUsesWith(V); break; @@ -315,60 +350,47 @@ void DefaultIntrinsicLowering::LowerIntrinsicCall(CallInst *CI) { break; case Intrinsic::longjmp: { - static Function *LongjmpFCache = 0; - ReplaceCallWith("longjmp", CI, CI->op_begin()+1, CI->op_end(), - Type::VoidTy, LongjmpFCache); + ReplaceCallWith("longjmp", CI, CI->op_begin() + 1, CI->op_end(), + Type::VoidTy); break; } case Intrinsic::siglongjmp: { // Insert the call to abort - static Function *AbortFCache = 0; - ReplaceCallWith("abort", CI, CI->op_end(), CI->op_end(), Type::VoidTy, - AbortFCache); + ReplaceCallWith("abort", CI, CI->op_end(), CI->op_end(), + Type::VoidTy); break; } - case Intrinsic::ctpop_i8: - case Intrinsic::ctpop_i16: - case Intrinsic::ctpop_i32: - case Intrinsic::ctpop_i64: - CI->replaceAllUsesWith(LowerCTPOP(CI->getOperand(1), CI)); + case Intrinsic::ctpop: + CI->replaceAllUsesWith(LowerCTPOP(Context, CI->getOperand(1), CI)); break; - case Intrinsic::bswap_i16: - case Intrinsic::bswap_i32: - case Intrinsic::bswap_i64: - CI->replaceAllUsesWith(LowerBSWAP(CI->getOperand(1), CI)); + case Intrinsic::bswap: + CI->replaceAllUsesWith(LowerBSWAP(Context, CI->getOperand(1), CI)); break; - case Intrinsic::ctlz_i8: - case Intrinsic::ctlz_i16: - case Intrinsic::ctlz_i32: - case Intrinsic::ctlz_i64: - CI->replaceAllUsesWith(LowerCTLZ(CI->getOperand(1), CI)); + case Intrinsic::ctlz: + CI->replaceAllUsesWith(LowerCTLZ(Context, CI->getOperand(1), CI)); break; - case Intrinsic::cttz_i8: - case Intrinsic::cttz_i16: - case Intrinsic::cttz_i32: - case Intrinsic::cttz_i64: { + case Intrinsic::cttz: { // cttz(x) -> ctpop(~X & (X-1)) Value *Src = CI->getOperand(1); - Value *NotSrc = BinaryOperator::createNot(Src, Src->getName()+".not", CI); - Value *SrcM1 = ConstantInt::get(Src->getType(), 1); - SrcM1 = BinaryOperator::createSub(Src, SrcM1, "", CI); - Src = LowerCTPOP(BinaryOperator::createAnd(NotSrc, SrcM1, "", CI), CI); + Value *NotSrc = Builder.CreateNot(Src); + NotSrc->setName(Src->getName() + ".not"); + Value *SrcM1 = ConstantInt::get(Src->getType(), 1); + SrcM1 = Builder.CreateSub(Src, SrcM1); + Src = LowerCTPOP(Context, Builder.CreateAnd(NotSrc, SrcM1), CI); CI->replaceAllUsesWith(Src); break; } case Intrinsic::stacksave: case Intrinsic::stackrestore: { - static bool Warned = false; if (!Warned) - std::cerr << "WARNING: this target does not support the llvm.stack" - << (Callee->getIntrinsicID() == Intrinsic::stacksave ? - "save" : "restore") << " intrinsic.\n"; + cerr << "WARNING: this target does not support the llvm.stack" + << (Callee->getIntrinsicID() == Intrinsic::stacksave ? + "save" : "restore") << " intrinsic.\n"; Warned = true; if (Callee->getIntrinsicID() == Intrinsic::stacksave) CI->replaceAllUsesWith(Constant::getNullValue(CI->getType())); @@ -377,9 +399,9 @@ void DefaultIntrinsicLowering::LowerIntrinsicCall(CallInst *CI) { case Intrinsic::returnaddress: case Intrinsic::frameaddress: - std::cerr << "WARNING: this target does not support the llvm." - << (Callee->getIntrinsicID() == Intrinsic::returnaddress ? - "return" : "frame") << "address intrinsic.\n"; + cerr << "WARNING: this target does not support the llvm." + << (Callee->getIntrinsicID() == Intrinsic::returnaddress ? + "return" : "frame") << "address intrinsic.\n"; CI->replaceAllUsesWith(ConstantPointerNull::get( cast(CI->getType()))); break; @@ -390,69 +412,102 @@ void DefaultIntrinsicLowering::LowerIntrinsicCall(CallInst *CI) { case Intrinsic::pcmarker: break; // Simply strip out pcmarker on unsupported architectures case Intrinsic::readcyclecounter: { - std::cerr << "WARNING: this target does not support the llvm.readcyclecoun" - << "ter intrinsic. It is being lowered to a constant 0\n"; - CI->replaceAllUsesWith(ConstantUInt::get(Type::ULongTy, 0)); + cerr << "WARNING: this target does not support the llvm.readcyclecoun" + << "ter intrinsic. It is being lowered to a constant 0\n"; + CI->replaceAllUsesWith(ConstantInt::get(Type::Int64Ty, 0)); break; } case Intrinsic::dbg_stoppoint: case Intrinsic::dbg_region_start: case Intrinsic::dbg_region_end: - case Intrinsic::dbg_declare: case Intrinsic::dbg_func_start: - if (CI->getType() != Type::VoidTy) - CI->replaceAllUsesWith(Constant::getNullValue(CI->getType())); + case Intrinsic::dbg_declare: break; // Simply strip out debugging intrinsics + case Intrinsic::eh_exception: + case Intrinsic::eh_selector_i32: + case Intrinsic::eh_selector_i64: + CI->replaceAllUsesWith(Constant::getNullValue(CI->getType())); + break; + + case Intrinsic::eh_typeid_for_i32: + case Intrinsic::eh_typeid_for_i64: + // Return something different to eh_selector. + CI->replaceAllUsesWith(ConstantInt::get(CI->getType(), 1)); + break; + + case Intrinsic::var_annotation: + break; // Strip out annotate intrinsic + case Intrinsic::memcpy: { - // The memcpy intrinsic take an extra alignment argument that the memcpy - // libc function does not. - static Function *MemcpyFCache = 0; - ReplaceCallWith("memcpy", CI, CI->op_begin()+1, CI->op_end()-1, - (*(CI->op_begin()+1))->getType(), MemcpyFCache); + const IntegerType *IntPtr = TD.getIntPtrType(); + Value *Size = Builder.CreateIntCast(CI->getOperand(3), IntPtr, + /* isSigned */ false); + Value *Ops[3]; + Ops[0] = CI->getOperand(1); + Ops[1] = CI->getOperand(2); + Ops[2] = Size; + ReplaceCallWith("memcpy", CI, Ops, Ops+3, CI->getOperand(1)->getType()); break; } case Intrinsic::memmove: { - // The memmove intrinsic take an extra alignment argument that the memmove - // libc function does not. - static Function *MemmoveFCache = 0; - ReplaceCallWith("memmove", CI, CI->op_begin()+1, CI->op_end()-1, - (*(CI->op_begin()+1))->getType(), MemmoveFCache); + const IntegerType *IntPtr = TD.getIntPtrType(); + Value *Size = Builder.CreateIntCast(CI->getOperand(3), IntPtr, + /* isSigned */ false); + Value *Ops[3]; + Ops[0] = CI->getOperand(1); + Ops[1] = CI->getOperand(2); + Ops[2] = Size; + ReplaceCallWith("memmove", CI, Ops, Ops+3, CI->getOperand(1)->getType()); break; } case Intrinsic::memset: { - // The memset intrinsic take an extra alignment argument that the memset - // libc function does not. - static Function *MemsetFCache = 0; - ReplaceCallWith("memset", CI, CI->op_begin()+1, CI->op_end()-1, - (*(CI->op_begin()+1))->getType(), MemsetFCache); + const IntegerType *IntPtr = TD.getIntPtrType(); + Value *Size = Builder.CreateIntCast(CI->getOperand(3), IntPtr, + /* isSigned */ false); + Value *Ops[3]; + Ops[0] = CI->getOperand(1); + // Extend the amount to i32. + Ops[1] = Builder.CreateIntCast(CI->getOperand(2), Type::Int32Ty, + /* isSigned */ false); + Ops[2] = Size; + ReplaceCallWith("memset", CI, Ops, Ops+3, CI->getOperand(1)->getType()); + break; + } + case Intrinsic::sqrt: { + ReplaceFPIntrinsicWithCall(CI, "sqrtf", "sqrt", "sqrtl"); + break; + } + case Intrinsic::log: { + ReplaceFPIntrinsicWithCall(CI, "logf", "log", "logl"); + break; + } + case Intrinsic::log2: { + ReplaceFPIntrinsicWithCall(CI, "log2f", "log2", "log2l"); + break; + } + case Intrinsic::log10: { + ReplaceFPIntrinsicWithCall(CI, "log10f", "log10", "log10l"); break; } - case Intrinsic::isunordered_f32: - case Intrinsic::isunordered_f64: { - Value *L = CI->getOperand(1); - Value *R = CI->getOperand(2); - - Value *LIsNan = new SetCondInst(Instruction::SetNE, L, L, "LIsNan", CI); - Value *RIsNan = new SetCondInst(Instruction::SetNE, R, R, "RIsNan", CI); - CI->replaceAllUsesWith( - BinaryOperator::create(Instruction::Or, LIsNan, RIsNan, - "isunordered", CI)); + case Intrinsic::exp: { + ReplaceFPIntrinsicWithCall(CI, "expf", "exp", "expl"); break; } - case Intrinsic::sqrt_f32: - case Intrinsic::sqrt_f64: { - static Function *sqrtFCache = 0; - static Function *sqrtfFCache = 0; - if(CI->getType() == Type::FloatTy) - ReplaceCallWith("sqrtf", CI, CI->op_begin()+1, CI->op_end(), - Type::FloatTy, sqrtfFCache); - else - ReplaceCallWith("sqrt", CI, CI->op_begin()+1, CI->op_end(), - Type::DoubleTy, sqrtFCache); + case Intrinsic::exp2: { + ReplaceFPIntrinsicWithCall(CI, "exp2f", "exp2", "exp2l"); break; } + case Intrinsic::pow: { + ReplaceFPIntrinsicWithCall(CI, "powf", "pow", "powl"); + break; + } + case Intrinsic::flt_rounds: + // Lower to "round to the nearest" + if (CI->getType() != Type::VoidTy) + CI->replaceAllUsesWith(ConstantInt::get(CI->getType(), 1)); + break; } assert(CI->use_empty() &&