X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FCodeGen%2FIntrinsicLowering.cpp;h=5c0484f2a7ffa96923b885d586769078058acbb8;hb=e984e504b5f3090ab270cbdab02638ac3a2afb21;hp=e75291dffa4c241001db836ea7ed36b60a3666da;hpb=774b864f0e7f3812e7daddf8553649e995ce040a;p=oota-llvm.git diff --git a/lib/CodeGen/IntrinsicLowering.cpp b/lib/CodeGen/IntrinsicLowering.cpp index e75291dffa4..5c0484f2a7f 100644 --- a/lib/CodeGen/IntrinsicLowering.cpp +++ b/lib/CodeGen/IntrinsicLowering.cpp @@ -2,8 +2,8 @@ // // 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. // //===----------------------------------------------------------------------===// // @@ -20,6 +20,7 @@ #include "llvm/Support/Streams.h" #include "llvm/Target/TargetData.h" #include "llvm/ADT/SmallVector.h" +#include "llvm/ADT/STLExtras.h" using namespace llvm; template @@ -53,8 +54,8 @@ static CallInst *ReplaceCallWith(const char *NewFn, CallInst *CI, FunctionType::get(RetTy, ParamTys, false)); } - SmallVector Operands(ArgBegin, ArgEnd); - CallInst *NewCI = new CallInst(FCache, &Operands[0], Operands.size(), + SmallVector Args(ArgBegin, ArgEnd); + CallInst *NewCI = new CallInst(FCache, Args.begin(), Args.end(), CI->getName(), CI); if (!CI->use_empty()) CI->replaceAllUsesWith(NewCI); @@ -80,32 +81,84 @@ void IntrinsicLowering::AddPrototypes(Module &M) { break; case Intrinsic::memcpy_i32: case Intrinsic::memcpy_i64: - M.getOrInsertFunction("memcpy", PointerType::get(Type::Int8Ty), - PointerType::get(Type::Int8Ty), - PointerType::get(Type::Int8Ty), + M.getOrInsertFunction("memcpy", PointerType::getUnqual(Type::Int8Ty), + PointerType::getUnqual(Type::Int8Ty), + PointerType::getUnqual(Type::Int8Ty), TD.getIntPtrType(), (Type *)0); break; case Intrinsic::memmove_i32: case Intrinsic::memmove_i64: - M.getOrInsertFunction("memmove", PointerType::get(Type::Int8Ty), - PointerType::get(Type::Int8Ty), - PointerType::get(Type::Int8Ty), + M.getOrInsertFunction("memmove", PointerType::getUnqual(Type::Int8Ty), + PointerType::getUnqual(Type::Int8Ty), + PointerType::getUnqual(Type::Int8Ty), TD.getIntPtrType(), (Type *)0); break; case Intrinsic::memset_i32: case Intrinsic::memset_i64: - M.getOrInsertFunction("memset", PointerType::get(Type::Int8Ty), - PointerType::get(Type::Int8Ty), Type::Int32Ty, + M.getOrInsertFunction("memset", PointerType::getUnqual(Type::Int8Ty), + PointerType::getUnqual(Type::Int8Ty), + Type::Int32Ty, TD.getIntPtrType(), (Type *)0); break; - case Intrinsic::sqrt_f32: - case Intrinsic::sqrt_f64: - if(I->arg_begin()->getType() == Type::FloatTy) + case Intrinsic::sqrt: + switch((int)I->arg_begin()->getType()->getTypeID()) { + case Type::FloatTyID: EnsureFunctionExists(M, "sqrtf", I->arg_begin(), I->arg_end(), Type::FloatTy); - else + case Type::DoubleTyID: EnsureFunctionExists(M, "sqrt", I->arg_begin(), I->arg_end(), Type::DoubleTy); + case Type::X86_FP80TyID: + case Type::FP128TyID: + case Type::PPC_FP128TyID: + EnsureFunctionExists(M, "sqrtl", I->arg_begin(), I->arg_end(), + I->arg_begin()->getType()); + } + break; + case Intrinsic::sin: + switch((int)I->arg_begin()->getType()->getTypeID()) { + case Type::FloatTyID: + EnsureFunctionExists(M, "sinf", I->arg_begin(), I->arg_end(), + Type::FloatTy); + case Type::DoubleTyID: + EnsureFunctionExists(M, "sin", I->arg_begin(), I->arg_end(), + Type::DoubleTy); + case Type::X86_FP80TyID: + case Type::FP128TyID: + case Type::PPC_FP128TyID: + EnsureFunctionExists(M, "sinl", I->arg_begin(), I->arg_end(), + I->arg_begin()->getType()); + } + break; + case Intrinsic::cos: + switch((int)I->arg_begin()->getType()->getTypeID()) { + case Type::FloatTyID: + EnsureFunctionExists(M, "cosf", I->arg_begin(), I->arg_end(), + Type::FloatTy); + case Type::DoubleTyID: + EnsureFunctionExists(M, "cos", I->arg_begin(), I->arg_end(), + Type::DoubleTy); + case Type::X86_FP80TyID: + case Type::FP128TyID: + case Type::PPC_FP128TyID: + EnsureFunctionExists(M, "cosl", I->arg_begin(), I->arg_end(), + I->arg_begin()->getType()); + } + break; + case Intrinsic::pow: + switch((int)I->arg_begin()->getType()->getTypeID()) { + case Type::FloatTyID: + EnsureFunctionExists(M, "powf", I->arg_begin(), I->arg_end(), + Type::FloatTy); + case Type::DoubleTyID: + EnsureFunctionExists(M, "pow", I->arg_begin(), I->arg_end(), + Type::DoubleTy); + case Type::X86_FP80TyID: + case Type::FP128TyID: + case Type::PPC_FP128TyID: + EnsureFunctionExists(M, "powl", I->arg_begin(), I->arg_end(), + I->arg_begin()->getType()); + } break; } } @@ -144,7 +197,7 @@ static Value *LowerBSWAP(Value *V, Instruction *IP) { "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); + V = BinaryOperator::createOr(Tmp4, Tmp2, "bswap.i32", IP); break; } case 64: { @@ -208,17 +261,30 @@ static Value *LowerCTPOP(Value *V, Instruction *IP) { }; unsigned BitSize = V->getType()->getPrimitiveSizeInBits(); + unsigned WordSize = (BitSize + 63) / 64; + Value *Count = ConstantInt::get(V->getType(), 0); - for (unsigned i = 1, ct = 0; i != BitSize; i <<= 1, ++ct) { - Value *MaskCst = ConstantInt::get(V->getType(), MaskValues[ct]); - Value *LHS = BinaryOperator::createAnd(V, MaskCst, "cppop.and1", IP); - Value *VShift = BinaryOperator::createLShr(V, - ConstantInt::get(V->getType(), i), "ctpop.sh", IP); - Value *RHS = BinaryOperator::createAnd(VShift, MaskCst, "cppop.and2", IP); - V = BinaryOperator::createAdd(LHS, RHS, "ctpop.step", IP); + 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 = BinaryOperator::createAnd( + PartValue, MaskCst, "cppop.and1", IP); + Value *VShift = BinaryOperator::createLShr(PartValue, + ConstantInt::get(V->getType(), i), "ctpop.sh", IP); + Value *RHS = BinaryOperator::createAnd(VShift, MaskCst, "cppop.and2", IP); + PartValue = BinaryOperator::createAdd(LHS, RHS, "ctpop.step", IP); + } + Count = BinaryOperator::createAdd(PartValue, Count, "ctpop.part", IP); + if (BitSize > 64) { + V = BinaryOperator::createLShr(V, ConstantInt::get(V->getType(), 64), + "ctpop.part.sh", IP); + BitSize -= 64; + } } - return V; + return Count; } /// LowerCTLZ - Emit the code to lower ctlz of V before the specified @@ -226,7 +292,7 @@ static Value *LowerCTPOP(Value *V, Instruction *IP) { static Value *LowerCTLZ(Value *V, Instruction *IP) { unsigned BitSize = V->getType()->getPrimitiveSizeInBits(); - for (unsigned i = 1; i != BitSize; i <<= 1) { + for (unsigned i = 1; i < BitSize; i <<= 1) { Value *ShVal = ConstantInt::get(V->getType(), i); ShVal = BinaryOperator::createLShr(V, ShVal, "ctlz.sh", IP); V = BinaryOperator::createOr(V, ShVal, "ctlz.step", IP); @@ -236,6 +302,346 @@ static Value *LowerCTLZ(Value *V, Instruction *IP) { return LowerCTPOP(V, IP); } +/// Convert the llvm.part.select.iX.iY intrinsic. This intrinsic takes +/// three integer arguments. The first argument is the Value from which the +/// bits will be selected. It may be of any bit width. The second and third +/// arguments specify a range of bits to select with the second argument +/// specifying the low bit and the third argument specifying the high bit. Both +/// must be type i32. The result is the corresponding selected bits from the +/// Value in the same width as the Value (first argument). If the low bit index +/// is higher than the high bit index then the inverse selection is done and +/// the bits are returned in inverse order. +/// @brief Lowering of llvm.part.select intrinsic. +static Instruction *LowerPartSelect(CallInst *CI) { + // Make sure we're dealing with a part select intrinsic here + Function *F = CI->getCalledFunction(); + const FunctionType *FT = F->getFunctionType(); + if (!F->isDeclaration() || !FT->getReturnType()->isInteger() || + FT->getNumParams() != 3 || !FT->getParamType(0)->isInteger() || + !FT->getParamType(1)->isInteger() || !FT->getParamType(2)->isInteger()) + return CI; + + // Get the intrinsic implementation function by converting all the . to _ + // in the intrinsic's function name and then reconstructing the function + // declaration. + std::string Name(F->getName()); + for (unsigned i = 4; i < Name.length(); ++i) + if (Name[i] == '.') + Name[i] = '_'; + Module* M = F->getParent(); + F = cast(M->getOrInsertFunction(Name, FT)); + F->setLinkage(GlobalValue::WeakLinkage); + + // If we haven't defined the impl function yet, do so now + if (F->isDeclaration()) { + + // Get the arguments to the function + Function::arg_iterator args = F->arg_begin(); + Value* Val = args++; Val->setName("Val"); + Value* Lo = args++; Lo->setName("Lo"); + Value* Hi = args++; Hi->setName("High"); + + // We want to select a range of bits here such that [Hi, Lo] is shifted + // down to the low bits. However, it is quite possible that Hi is smaller + // than Lo in which case the bits have to be reversed. + + // Create the blocks we will need for the two cases (forward, reverse) + BasicBlock* CurBB = new BasicBlock("entry", F); + BasicBlock *RevSize = new BasicBlock("revsize", CurBB->getParent()); + BasicBlock *FwdSize = new BasicBlock("fwdsize", CurBB->getParent()); + BasicBlock *Compute = new BasicBlock("compute", CurBB->getParent()); + BasicBlock *Reverse = new BasicBlock("reverse", CurBB->getParent()); + BasicBlock *RsltBlk = new BasicBlock("result", CurBB->getParent()); + + // Cast Hi and Lo to the size of Val so the widths are all the same + if (Hi->getType() != Val->getType()) + Hi = CastInst::createIntegerCast(Hi, Val->getType(), false, + "tmp", CurBB); + if (Lo->getType() != Val->getType()) + Lo = CastInst::createIntegerCast(Lo, Val->getType(), false, + "tmp", CurBB); + + // Compute a few things that both cases will need, up front. + Constant* Zero = ConstantInt::get(Val->getType(), 0); + Constant* One = ConstantInt::get(Val->getType(), 1); + Constant* AllOnes = ConstantInt::getAllOnesValue(Val->getType()); + + // Compare the Hi and Lo bit positions. This is used to determine + // which case we have (forward or reverse) + ICmpInst *Cmp = new ICmpInst(ICmpInst::ICMP_ULT, Hi, Lo, "less",CurBB); + new BranchInst(RevSize, FwdSize, Cmp, CurBB); + + // First, copmute the number of bits in the forward case. + Instruction* FBitSize = + BinaryOperator::createSub(Hi, Lo,"fbits", FwdSize); + new BranchInst(Compute, FwdSize); + + // Second, compute the number of bits in the reverse case. + Instruction* RBitSize = + BinaryOperator::createSub(Lo, Hi, "rbits", RevSize); + new BranchInst(Compute, RevSize); + + // Now, compute the bit range. Start by getting the bitsize and the shift + // amount (either Hi or Lo) from PHI nodes. Then we compute a mask for + // the number of bits we want in the range. We shift the bits down to the + // least significant bits, apply the mask to zero out unwanted high bits, + // and we have computed the "forward" result. It may still need to be + // reversed. + + // Get the BitSize from one of the two subtractions + PHINode *BitSize = new PHINode(Val->getType(), "bits", Compute); + BitSize->reserveOperandSpace(2); + BitSize->addIncoming(FBitSize, FwdSize); + BitSize->addIncoming(RBitSize, RevSize); + + // Get the ShiftAmount as the smaller of Hi/Lo + PHINode *ShiftAmt = new PHINode(Val->getType(), "shiftamt", Compute); + ShiftAmt->reserveOperandSpace(2); + ShiftAmt->addIncoming(Lo, FwdSize); + ShiftAmt->addIncoming(Hi, RevSize); + + // Increment the bit size + Instruction *BitSizePlusOne = + BinaryOperator::createAdd(BitSize, One, "bits", Compute); + + // Create a Mask to zero out the high order bits. + Instruction* Mask = + BinaryOperator::createShl(AllOnes, BitSizePlusOne, "mask", Compute); + Mask = BinaryOperator::createNot(Mask, "mask", Compute); + + // Shift the bits down and apply the mask + Instruction* FRes = + BinaryOperator::createLShr(Val, ShiftAmt, "fres", Compute); + FRes = BinaryOperator::createAnd(FRes, Mask, "fres", Compute); + new BranchInst(Reverse, RsltBlk, Cmp, Compute); + + // In the Reverse block we have the mask already in FRes but we must reverse + // it by shifting FRes bits right and putting them in RRes by shifting them + // in from left. + + // First set up our loop counters + PHINode *Count = new PHINode(Val->getType(), "count", Reverse); + Count->reserveOperandSpace(2); + Count->addIncoming(BitSizePlusOne, Compute); + + // Next, get the value that we are shifting. + PHINode *BitsToShift = new PHINode(Val->getType(), "val", Reverse); + BitsToShift->reserveOperandSpace(2); + BitsToShift->addIncoming(FRes, Compute); + + // Finally, get the result of the last computation + PHINode *RRes = new PHINode(Val->getType(), "rres", Reverse); + RRes->reserveOperandSpace(2); + RRes->addIncoming(Zero, Compute); + + // Decrement the counter + Instruction *Decr = BinaryOperator::createSub(Count, One, "decr", Reverse); + Count->addIncoming(Decr, Reverse); + + // Compute the Bit that we want to move + Instruction *Bit = + BinaryOperator::createAnd(BitsToShift, One, "bit", Reverse); + + // Compute the new value for next iteration. + Instruction *NewVal = + BinaryOperator::createLShr(BitsToShift, One, "rshift", Reverse); + BitsToShift->addIncoming(NewVal, Reverse); + + // Shift the bit into the low bits of the result. + Instruction *NewRes = + BinaryOperator::createShl(RRes, One, "lshift", Reverse); + NewRes = BinaryOperator::createOr(NewRes, Bit, "addbit", Reverse); + RRes->addIncoming(NewRes, Reverse); + + // Terminate loop if we've moved all the bits. + ICmpInst *Cond = + new ICmpInst(ICmpInst::ICMP_EQ, Decr, Zero, "cond", Reverse); + new BranchInst(RsltBlk, Reverse, Cond, Reverse); + + // Finally, in the result block, select one of the two results with a PHI + // node and return the result; + CurBB = RsltBlk; + PHINode *BitSelect = new PHINode(Val->getType(), "part_select", CurBB); + BitSelect->reserveOperandSpace(2); + BitSelect->addIncoming(FRes, Compute); + BitSelect->addIncoming(NewRes, Reverse); + new ReturnInst(BitSelect, CurBB); + } + + // Return a call to the implementation function + Value *Args[] = { + CI->getOperand(1), + CI->getOperand(2), + CI->getOperand(3) + }; + return new CallInst(F, Args, array_endof(Args), CI->getName(), CI); +} + +/// Convert the llvm.part.set.iX.iY.iZ intrinsic. This intrinsic takes +/// four integer arguments (iAny %Value, iAny %Replacement, i32 %Low, i32 %High) +/// The first two arguments can be any bit width. The result is the same width +/// as %Value. The operation replaces bits between %Low and %High with the value +/// in %Replacement. If %Replacement is not the same width, it is truncated or +/// zero extended as appropriate to fit the bits being replaced. If %Low is +/// greater than %High then the inverse set of bits are replaced. +/// @brief Lowering of llvm.bit.part.set intrinsic. +static Instruction *LowerPartSet(CallInst *CI) { + // Make sure we're dealing with a part select intrinsic here + Function *F = CI->getCalledFunction(); + const FunctionType *FT = F->getFunctionType(); + if (!F->isDeclaration() || !FT->getReturnType()->isInteger() || + FT->getNumParams() != 4 || !FT->getParamType(0)->isInteger() || + !FT->getParamType(1)->isInteger() || !FT->getParamType(2)->isInteger() || + !FT->getParamType(3)->isInteger()) + return CI; + + // Get the intrinsic implementation function by converting all the . to _ + // in the intrinsic's function name and then reconstructing the function + // declaration. + std::string Name(F->getName()); + for (unsigned i = 4; i < Name.length(); ++i) + if (Name[i] == '.') + Name[i] = '_'; + Module* M = F->getParent(); + F = cast(M->getOrInsertFunction(Name, FT)); + F->setLinkage(GlobalValue::WeakLinkage); + + // If we haven't defined the impl function yet, do so now + if (F->isDeclaration()) { + // Get the arguments for the function. + Function::arg_iterator args = F->arg_begin(); + Value* Val = args++; Val->setName("Val"); + Value* Rep = args++; Rep->setName("Rep"); + Value* Lo = args++; Lo->setName("Lo"); + Value* Hi = args++; Hi->setName("Hi"); + + // Get some types we need + const IntegerType* ValTy = cast(Val->getType()); + const IntegerType* RepTy = cast(Rep->getType()); + uint32_t ValBits = ValTy->getBitWidth(); + uint32_t RepBits = RepTy->getBitWidth(); + + // Constant Definitions + ConstantInt* RepBitWidth = ConstantInt::get(Type::Int32Ty, RepBits); + ConstantInt* RepMask = ConstantInt::getAllOnesValue(RepTy); + ConstantInt* ValMask = ConstantInt::getAllOnesValue(ValTy); + ConstantInt* One = ConstantInt::get(Type::Int32Ty, 1); + ConstantInt* ValOne = ConstantInt::get(ValTy, 1); + ConstantInt* Zero = ConstantInt::get(Type::Int32Ty, 0); + ConstantInt* ValZero = ConstantInt::get(ValTy, 0); + + // Basic blocks we fill in below. + BasicBlock* entry = new BasicBlock("entry", F, 0); + BasicBlock* large = new BasicBlock("large", F, 0); + BasicBlock* small = new BasicBlock("small", F, 0); + BasicBlock* reverse = new BasicBlock("reverse", F, 0); + BasicBlock* result = new BasicBlock("result", F, 0); + + // BASIC BLOCK: entry + // First, get the number of bits that we're placing as an i32 + ICmpInst* is_forward = + new ICmpInst(ICmpInst::ICMP_ULT, Lo, Hi, "", entry); + SelectInst* Hi_pn = new SelectInst(is_forward, Hi, Lo, "", entry); + SelectInst* Lo_pn = new SelectInst(is_forward, Lo, Hi, "", entry); + BinaryOperator* NumBits = BinaryOperator::createSub(Hi_pn, Lo_pn, "",entry); + NumBits = BinaryOperator::createAdd(NumBits, One, "", entry); + // Now, convert Lo and Hi to ValTy bit width + if (ValBits > 32) { + Lo = new ZExtInst(Lo_pn, ValTy, "", entry); + } else if (ValBits < 32) { + Lo = new TruncInst(Lo_pn, ValTy, "", entry); + } + // Determine if the replacement bits are larger than the number of bits we + // are replacing and deal with it. + ICmpInst* is_large = + new ICmpInst(ICmpInst::ICMP_ULT, NumBits, RepBitWidth, "", entry); + new BranchInst(large, small, is_large, entry); + + // BASIC BLOCK: large + Instruction* MaskBits = + BinaryOperator::createSub(RepBitWidth, NumBits, "", large); + MaskBits = CastInst::createIntegerCast(MaskBits, RepMask->getType(), + false, "", large); + BinaryOperator* Mask1 = + BinaryOperator::createLShr(RepMask, MaskBits, "", large); + BinaryOperator* Rep2 = BinaryOperator::createAnd(Mask1, Rep, "", large); + new BranchInst(small, large); + + // BASIC BLOCK: small + PHINode* Rep3 = new PHINode(RepTy, "", small); + Rep3->reserveOperandSpace(2); + Rep3->addIncoming(Rep2, large); + Rep3->addIncoming(Rep, entry); + Value* Rep4 = Rep3; + if (ValBits > RepBits) + Rep4 = new ZExtInst(Rep3, ValTy, "", small); + else if (ValBits < RepBits) + Rep4 = new TruncInst(Rep3, ValTy, "", small); + new BranchInst(result, reverse, is_forward, small); + + // BASIC BLOCK: reverse (reverses the bits of the replacement) + // Set up our loop counter as a PHI so we can decrement on each iteration. + // We will loop for the number of bits in the replacement value. + PHINode *Count = new PHINode(Type::Int32Ty, "count", reverse); + Count->reserveOperandSpace(2); + Count->addIncoming(NumBits, small); + + // Get the value that we are shifting bits out of as a PHI because + // we'll change this with each iteration. + PHINode *BitsToShift = new PHINode(Val->getType(), "val", reverse); + BitsToShift->reserveOperandSpace(2); + BitsToShift->addIncoming(Rep4, small); + + // Get the result of the last computation or zero on first iteration + PHINode *RRes = new PHINode(Val->getType(), "rres", reverse); + RRes->reserveOperandSpace(2); + RRes->addIncoming(ValZero, small); + + // Decrement the loop counter by one + Instruction *Decr = BinaryOperator::createSub(Count, One, "", reverse); + Count->addIncoming(Decr, reverse); + + // Get the bit that we want to move into the result + Value *Bit = BinaryOperator::createAnd(BitsToShift, ValOne, "", reverse); + + // Compute the new value of the bits to shift for the next iteration. + Value *NewVal = BinaryOperator::createLShr(BitsToShift, ValOne,"", reverse); + BitsToShift->addIncoming(NewVal, reverse); + + // Shift the bit we extracted into the low bit of the result. + Instruction *NewRes = BinaryOperator::createShl(RRes, ValOne, "", reverse); + NewRes = BinaryOperator::createOr(NewRes, Bit, "", reverse); + RRes->addIncoming(NewRes, reverse); + + // Terminate loop if we've moved all the bits. + ICmpInst *Cond = new ICmpInst(ICmpInst::ICMP_EQ, Decr, Zero, "", reverse); + new BranchInst(result, reverse, Cond, reverse); + + // BASIC BLOCK: result + PHINode *Rplcmnt = new PHINode(Val->getType(), "", result); + Rplcmnt->reserveOperandSpace(2); + Rplcmnt->addIncoming(NewRes, reverse); + Rplcmnt->addIncoming(Rep4, small); + Value* t0 = CastInst::createIntegerCast(NumBits,ValTy,false,"",result); + Value* t1 = BinaryOperator::createShl(ValMask, Lo, "", result); + Value* t2 = BinaryOperator::createNot(t1, "", result); + Value* t3 = BinaryOperator::createShl(t1, t0, "", result); + Value* t4 = BinaryOperator::createOr(t2, t3, "", result); + Value* t5 = BinaryOperator::createAnd(t4, Val, "", result); + Value* t6 = BinaryOperator::createShl(Rplcmnt, Lo, "", result); + Value* Rslt = BinaryOperator::createOr(t5, t6, "part_set", result); + new ReturnInst(Rslt, result); + } + + // Return a call to the implementation function + Value *Args[] = { + CI->getOperand(1), + CI->getOperand(2), + CI->getOperand(3), + CI->getOperand(4) + }; + return new CallInst(F, Args, array_endof(Args), CI->getName(), CI); +} void IntrinsicLowering::LowerIntrinsicCall(CallInst *CI) { @@ -283,30 +689,19 @@ void IntrinsicLowering::LowerIntrinsicCall(CallInst *CI) { Type::VoidTy, AbortFCache); break; } - case Intrinsic::ctpop_i8: - case Intrinsic::ctpop_i16: - case Intrinsic::ctpop_i32: - case Intrinsic::ctpop_i64: + case Intrinsic::ctpop: CI->replaceAllUsesWith(LowerCTPOP(CI->getOperand(1), CI)); break; - case Intrinsic::bswap_i16: - case Intrinsic::bswap_i32: - case Intrinsic::bswap_i64: + case Intrinsic::bswap: CI->replaceAllUsesWith(LowerBSWAP(CI->getOperand(1), CI)); break; - case Intrinsic::ctlz_i8: - case Intrinsic::ctlz_i16: - case Intrinsic::ctlz_i32: - case Intrinsic::ctlz_i64: + case Intrinsic::ctlz: CI->replaceAllUsesWith(LowerCTLZ(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); @@ -317,6 +712,14 @@ void IntrinsicLowering::LowerIntrinsicCall(CallInst *CI) { break; } + case Intrinsic::part_select: + CI->replaceAllUsesWith(LowerPartSelect(CI)); + break; + + case Intrinsic::part_set: + CI->replaceAllUsesWith(LowerPartSet(CI)); + break; + case Intrinsic::stacksave: case Intrinsic::stackrestore: { static bool Warned = false; @@ -356,10 +759,23 @@ void IntrinsicLowering::LowerIntrinsicCall(CallInst *CI) { case Intrinsic::dbg_region_end: case Intrinsic::dbg_func_start: case Intrinsic::dbg_declare: + break; // Simply strip out debugging intrinsics + case Intrinsic::eh_exception: - case Intrinsic::eh_handlers: - break; // Simply strip out debugging and eh intrinsics + 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_i32: case Intrinsic::memcpy_i64: { static Constant *MemcpyFCache = 0; @@ -418,18 +834,34 @@ void IntrinsicLowering::LowerIntrinsicCall(CallInst *CI) { MemsetFCache); break; } - case Intrinsic::sqrt_f32: { + case Intrinsic::sqrt: { static Constant *sqrtfFCache = 0; - ReplaceCallWith("sqrtf", CI, CI->op_begin()+1, CI->op_end(), - Type::FloatTy, sqrtfFCache); - break; - } - case Intrinsic::sqrt_f64: { static Constant *sqrtFCache = 0; - ReplaceCallWith("sqrt", CI, CI->op_begin()+1, CI->op_end(), + static Constant *sqrtLDCache = 0; + switch (CI->getOperand(1)->getType()->getTypeID()) { + default: assert(0 && "Invalid type in sqrt"); abort(); + case Type::FloatTyID: + ReplaceCallWith("sqrtf", CI, CI->op_begin()+1, CI->op_end(), + Type::FloatTy, sqrtfFCache); + break; + case Type::DoubleTyID: + ReplaceCallWith("sqrt", CI, CI->op_begin()+1, CI->op_end(), Type::DoubleTy, sqrtFCache); + break; + case Type::X86_FP80TyID: + case Type::FP128TyID: + case Type::PPC_FP128TyID: + ReplaceCallWith("sqrtl", CI, CI->op_begin()+1, CI->op_end(), + CI->getOperand(1)->getType(), sqrtLDCache); + break; + } 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() &&