X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FCodeGen%2FIntrinsicLowering.cpp;h=bd3c8c71f5b221c1b5efc707b85a2a939c29ba39;hb=7cbd8a3e92221437048b484d5ef9c0a22d0f8c58;hp=12d392c09f0f6f17fbacda2693f5f3e115641778;hpb=5156f5b214bafc24e20ed49114b35ffbd55dc623;p=oota-llvm.git diff --git a/lib/CodeGen/IntrinsicLowering.cpp b/lib/CodeGen/IntrinsicLowering.cpp index 12d392c09f0..bd3c8c71f5b 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 @@ -53,9 +54,9 @@ 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(), - CI->getName(), CI); + SmallVector Args(ArgBegin, ArgEnd); + CallInst *NewCI = CallInst::Create(FCache, Args.begin(), Args.end(), + CI->getName(), CI); if (!CI->use_empty()) CI->replaceAllUsesWith(NewCI); return 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; } } @@ -120,76 +173,76 @@ static Value *LowerBSWAP(Value *V, Instruction *IP) { switch(BitSize) { default: assert(0 && "Unhandled type size of value to byteswap!"); case 16: { - Value *Tmp1 = BinaryOperator::createShl(V, + Value *Tmp1 = BinaryOperator::CreateShl(V, ConstantInt::get(V->getType(),8),"bswap.2",IP); - Value *Tmp2 = BinaryOperator::createLShr(V, + Value *Tmp2 = BinaryOperator::CreateLShr(V, ConstantInt::get(V->getType(),8),"bswap.1",IP); - V = BinaryOperator::createOr(Tmp1, Tmp2, "bswap.i16", IP); + V = BinaryOperator::CreateOr(Tmp1, Tmp2, "bswap.i16", IP); break; } case 32: { - Value *Tmp4 = BinaryOperator::createShl(V, + Value *Tmp4 = BinaryOperator::CreateShl(V, ConstantInt::get(V->getType(),24),"bswap.4", IP); - Value *Tmp3 = BinaryOperator::createShl(V, + Value *Tmp3 = BinaryOperator::CreateShl(V, ConstantInt::get(V->getType(),8),"bswap.3",IP); - Value *Tmp2 = BinaryOperator::createLShr(V, + Value *Tmp2 = BinaryOperator::CreateLShr(V, ConstantInt::get(V->getType(),8),"bswap.2",IP); - Value *Tmp1 = BinaryOperator::createLShr(V, + Value *Tmp1 = BinaryOperator::CreateLShr(V, ConstantInt::get(V->getType(),24),"bswap.1", IP); - Tmp3 = BinaryOperator::createAnd(Tmp3, + Tmp3 = BinaryOperator::CreateAnd(Tmp3, ConstantInt::get(Type::Int32Ty, 0xFF0000), "bswap.and3", IP); - Tmp2 = BinaryOperator::createAnd(Tmp2, + Tmp2 = BinaryOperator::CreateAnd(Tmp2, ConstantInt::get(Type::Int32Ty, 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); + Tmp4 = BinaryOperator::CreateOr(Tmp4, Tmp3, "bswap.or1", IP); + Tmp2 = BinaryOperator::CreateOr(Tmp2, Tmp1, "bswap.or2", IP); + V = BinaryOperator::CreateOr(Tmp4, Tmp2, "bswap.i32", IP); break; } case 64: { - Value *Tmp8 = BinaryOperator::createShl(V, + Value *Tmp8 = BinaryOperator::CreateShl(V, ConstantInt::get(V->getType(),56),"bswap.8", IP); - Value *Tmp7 = BinaryOperator::createShl(V, + Value *Tmp7 = BinaryOperator::CreateShl(V, ConstantInt::get(V->getType(),40),"bswap.7", IP); - Value *Tmp6 = BinaryOperator::createShl(V, + Value *Tmp6 = BinaryOperator::CreateShl(V, ConstantInt::get(V->getType(),24),"bswap.6", IP); - Value *Tmp5 = BinaryOperator::createShl(V, + Value *Tmp5 = BinaryOperator::CreateShl(V, ConstantInt::get(V->getType(),8),"bswap.5", IP); - Value* Tmp4 = BinaryOperator::createLShr(V, + Value* Tmp4 = BinaryOperator::CreateLShr(V, ConstantInt::get(V->getType(),8),"bswap.4", IP); - Value* Tmp3 = BinaryOperator::createLShr(V, + Value* Tmp3 = BinaryOperator::CreateLShr(V, ConstantInt::get(V->getType(),24),"bswap.3", IP); - Value* Tmp2 = BinaryOperator::createLShr(V, + Value* Tmp2 = BinaryOperator::CreateLShr(V, ConstantInt::get(V->getType(),40),"bswap.2", IP); - Value* Tmp1 = BinaryOperator::createLShr(V, + Value* Tmp1 = BinaryOperator::CreateLShr(V, ConstantInt::get(V->getType(),56),"bswap.1", IP); - Tmp7 = BinaryOperator::createAnd(Tmp7, + Tmp7 = BinaryOperator::CreateAnd(Tmp7, ConstantInt::get(Type::Int64Ty, 0xFF000000000000ULL), "bswap.and7", IP); - Tmp6 = BinaryOperator::createAnd(Tmp6, + Tmp6 = BinaryOperator::CreateAnd(Tmp6, ConstantInt::get(Type::Int64Ty, 0xFF0000000000ULL), "bswap.and6", IP); - Tmp5 = BinaryOperator::createAnd(Tmp5, + Tmp5 = BinaryOperator::CreateAnd(Tmp5, ConstantInt::get(Type::Int64Ty, 0xFF00000000ULL), "bswap.and5", IP); - Tmp4 = BinaryOperator::createAnd(Tmp4, + Tmp4 = BinaryOperator::CreateAnd(Tmp4, ConstantInt::get(Type::Int64Ty, 0xFF000000ULL), "bswap.and4", IP); - Tmp3 = BinaryOperator::createAnd(Tmp3, + Tmp3 = BinaryOperator::CreateAnd(Tmp3, ConstantInt::get(Type::Int64Ty, 0xFF0000ULL), "bswap.and3", IP); - Tmp2 = BinaryOperator::createAnd(Tmp2, + Tmp2 = BinaryOperator::CreateAnd(Tmp2, ConstantInt::get(Type::Int64Ty, 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); + 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); break; } } @@ -208,17 +261,30 @@ static Value *LowerCTPOP(Value *V, Instruction *IP) { }; unsigned BitSize = V->getType()->getPrimitiveSizeInBits(); - - 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); + 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 = 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 CastInst::createIntegerCast(V, Type::Int32Ty, false, "ctpop", IP); + return Count; } /// LowerCTLZ - Emit the code to lower ctlz of V before the specified @@ -226,13 +292,13 @@ 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); + ShVal = BinaryOperator::CreateLShr(V, ShVal, "ctlz.sh", IP); + V = BinaryOperator::CreateOr(V, ShVal, "ctlz.step", IP); } - V = BinaryOperator::createNot(V, "", IP); + V = BinaryOperator::CreateNot(V, "", IP); return LowerCTPOP(V, IP); } @@ -273,26 +339,26 @@ static Instruction *LowerPartSelect(CallInst *CI) { 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"); + 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()); + BasicBlock* CurBB = BasicBlock::Create("entry", F); + BasicBlock *RevSize = BasicBlock::Create("revsize", CurBB->getParent()); + BasicBlock *FwdSize = BasicBlock::Create("fwdsize", CurBB->getParent()); + BasicBlock *Compute = BasicBlock::Create("compute", CurBB->getParent()); + BasicBlock *Reverse = BasicBlock::Create("reverse", CurBB->getParent()); + BasicBlock *RsltBlk = BasicBlock::Create("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, + Hi = CastInst::CreateIntegerCast(Hi, Val->getType(), false, "tmp", CurBB); if (Lo->getType() != Val->getType()) - Lo = CastInst::createIntegerCast(Lo, Val->getType(), false, + Lo = CastInst::CreateIntegerCast(Lo, Val->getType(), false, "tmp", CurBB); // Compute a few things that both cases will need, up front. @@ -303,17 +369,17 @@ static Instruction *LowerPartSelect(CallInst *CI) { // 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); + BranchInst::Create(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); + BinaryOperator::CreateSub(Hi, Lo,"fbits", FwdSize); + BranchInst::Create(Compute, FwdSize); // Second, compute the number of bits in the reverse case. Instruction* RBitSize = - BinaryOperator::createSub(Lo, Hi, "rbits", RevSize); - new BranchInst(Compute, RevSize); + BinaryOperator::CreateSub(Lo, Hi, "rbits", RevSize); + BranchInst::Create(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 @@ -323,83 +389,83 @@ static Instruction *LowerPartSelect(CallInst *CI) { // reversed. // Get the BitSize from one of the two subtractions - PHINode *BitSize = new PHINode(Val->getType(), "bits", Compute); + PHINode *BitSize = PHINode::Create(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); + PHINode *ShiftAmt = PHINode::Create(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); + 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); + 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); + BinaryOperator::CreateLShr(Val, ShiftAmt, "fres", Compute); + FRes = BinaryOperator::CreateAnd(FRes, Mask, "fres", Compute); + BranchInst::Create(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); + PHINode *Count = PHINode::Create(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); + PHINode *BitsToShift = PHINode::Create(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); + PHINode *RRes = PHINode::Create(Val->getType(), "rres", Reverse); RRes->reserveOperandSpace(2); RRes->addIncoming(Zero, Compute); // Decrement the counter - Instruction *Decr = BinaryOperator::createSub(Count, One, "decr", Reverse); + 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); + BinaryOperator::CreateAnd(BitsToShift, One, "bit", Reverse); // Compute the new value for next iteration. Instruction *NewVal = - BinaryOperator::createLShr(BitsToShift, One, "rshift", Reverse); + 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); + 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); + BranchInst::Create(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); + PHINode *BitSelect = PHINode::Create(Val->getType(), "part_select", CurBB); BitSelect->reserveOperandSpace(2); BitSelect->addIncoming(FRes, Compute); BitSelect->addIncoming(NewRes, Reverse); - new ReturnInst(BitSelect, CurBB); + ReturnInst::Create(BitSelect, CurBB); } // Return a call to the implementation function @@ -408,7 +474,7 @@ static Instruction *LowerPartSelect(CallInst *CI) { CI->getOperand(2), CI->getOperand(3) }; - return new CallInst(F, Args, sizeof(Args)/sizeof(Args[0]), CI->getName(), CI); + return CallInst::Create(F, Args, array_endof(Args), CI->getName(), CI); } /// Convert the llvm.part.set.iX.iY.iZ intrinsic. This intrinsic takes @@ -442,57 +508,6 @@ static Instruction *LowerPartSet(CallInst *CI) { // If we haven't defined the impl function yet, do so now if (F->isDeclaration()) { - // Note: the following code is based on code generated by llvm2cpp with - // the following input. This is just *one* example of a generated function. - // The functions vary by bit width of result and first two arguments. - // The generated code has been changed to deal with any bit width not just - // the 32/64 bitwidths used in the above sample. - // - // define i64 @part_set(i64 %Val, i32 %Rep, i32 %Lo, i32 %Hi) { - // entry: - // %is_forward = icmp ult i32 %Lo, %Hi - // %Lo.pn = select i1 %is_forward, i32 %Hi, i32 %Lo - // %Hi.pn = select i1 %is_forward, i32 %Lo, i32 %Hi - // %iftmp.16.0 = sub i32 %Lo.pn, %Hi.pn - // icmp ult i32 %iftmp.16.0, 32 - // br i1 %1, label %cond_true11, label %cond_next19 - // cond_true11: - // %tmp13 = sub i32 32, %iftmp.16.0 - // %tmp14 = lshr i32 -1, %tmp13 - // %tmp16 = and i32 %tmp14, %Rep - // br label %cond_next19 - // cond_next19: - // %iftmp.17.0 = phi i32 [ %tmp16, %cond_true11 ], [ %Rep, %entry ] - // %tmp2021 = zext i32 %iftmp.17.0 to i64 - // icmp ugt i32 %Lo, %Hi - // br i1 %2, label %cond_next60, label %cond_true24 - // cond_true24: - // %tmp25.cast = zext i32 %Hi to i64 - // %tmp26 = lshr i64 -1, %tmp25.cast - // %tmp27.cast = zext i32 %Lo to i64 - // %tmp28 = shl i64 %tmp26, %tmp27.cast - // %tmp28not = xor i64 %tmp28, -1 - // %tmp31 = shl i64 %tmp2021, %tmp27.cast - // %tmp34 = and i64 %tmp28not, %Val - // %Val_addr.064 = or i64 %tmp31, %tmp34 - // ret i64 %Val_addr.064 - // cond_next60: - // %tmp39.cast = zext i32 %Lo to i64 - // %tmp40 = shl i64 -1, %tmp39.cast - // %tmp41.cast = zext i32 %Hi to i64 - // %tmp42 = shl i64 -1, %tmp41.cast - // %tmp45.demorgan = or i64 %tmp42, %tmp40 - // %tmp45 = xor i64 %tmp45.demorgan, -1 - // %tmp47 = and i64 %tmp45, %Val - // %tmp50 = shl i64 %tmp2021, %tmp39.cast - // %tmp52 = sub i32 32, %Hi - // %tmp52.cast = zext i32 %tmp52 to i64 - // %tmp54 = lshr i64 %tmp2021, %tmp52.cast - // %tmp57 = or i64 %tmp50, %tmp47 - // %Val_addr.0 = or i64 %tmp57, %tmp54 - // ret i64 %Val_addr.0 - // } - // Get the arguments for the function. Function::arg_iterator args = F->arg_begin(); Value* Val = args++; Val->setName("Val"); @@ -510,46 +525,50 @@ static Instruction *LowerPartSet(CallInst *CI) { ConstantInt* RepBitWidth = ConstantInt::get(Type::Int32Ty, RepBits); ConstantInt* RepMask = ConstantInt::getAllOnesValue(RepTy); ConstantInt* ValMask = ConstantInt::getAllOnesValue(ValTy); - - BasicBlock* entry = new BasicBlock("entry",F,0); - BasicBlock* large = new BasicBlock("large",F,0); - BasicBlock* small = new BasicBlock("small",F,0); - BasicBlock* forward = new BasicBlock("forward",F,0); - BasicBlock* reverse = new BasicBlock("reverse",F,0); - - // Block entry (entry) + 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 = BasicBlock::Create("entry", F, 0); + BasicBlock* large = BasicBlock::Create("large", F, 0); + BasicBlock* small = BasicBlock::Create("small", F, 0); + BasicBlock* reverse = BasicBlock::Create("reverse", F, 0); + BasicBlock* result = BasicBlock::Create("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* Lo_pn = new SelectInst(is_forward, Hi, Lo, "", entry); - SelectInst* Hi_pn = new SelectInst(is_forward, Lo, Hi, "", entry); - BinaryOperator* NumBits = BinaryOperator::createSub(Lo_pn, Hi_pn, "",entry); + SelectInst* Hi_pn = SelectInst::Create(is_forward, Hi, Lo, "", entry); + SelectInst* Lo_pn = SelectInst::Create(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) { - Hi = new ZExtInst(Hi, ValTy, "", entry); - Lo = new ZExtInst(Lo, ValTy, "", entry); + Lo = new ZExtInst(Lo_pn, ValTy, "", entry); } else if (ValBits < 32) { - Hi = new TruncInst(Hi, ValTy, "", entry); - Lo = new TruncInst(Lo, ValTy, "", entry); + 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); + BranchInst::Create(large, small, is_large, entry); - // Block "large" + // BASIC BLOCK: large Instruction* MaskBits = - BinaryOperator::createSub(RepBitWidth, NumBits, "", large); - MaskBits = CastInst::createIntegerCast(MaskBits, RepMask->getType(), + 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); + BinaryOperator::CreateLShr(RepMask, MaskBits, "", large); + BinaryOperator* Rep2 = BinaryOperator::CreateAnd(Mask1, Rep, "", large); + BranchInst::Create(small, large); - // Block "small" - PHINode* Rep3 = new PHINode(RepTy, "", small); + // BASIC BLOCK: small + PHINode* Rep3 = PHINode::Create(RepTy, "", small); Rep3->reserveOperandSpace(2); Rep3->addIncoming(Rep2, large); Rep3->addIncoming(Rep, entry); @@ -558,37 +577,60 @@ static Instruction *LowerPartSet(CallInst *CI) { Rep4 = new ZExtInst(Rep3, ValTy, "", small); else if (ValBits < RepBits) Rep4 = new TruncInst(Rep3, ValTy, "", small); - ICmpInst* is_reverse = - new ICmpInst(ICmpInst::ICMP_UGT, Lo, Hi, "", small); - new BranchInst(reverse, forward, is_reverse, small); - - // Block "forward" - Value* t1 = BinaryOperator::createLShr(ValMask, Hi, "", forward); - Value* t2 = BinaryOperator::createShl(t1, Lo, "", forward); - Value* nott2 = BinaryOperator::createXor(t2, ValMask, "", forward); - Value* t3 = BinaryOperator::createShl(Rep4, Lo, "", forward); - Value* t4 = BinaryOperator::createAnd(nott2, Val, "", forward); - Value* FRslt = BinaryOperator::createOr(t3, t4, "", forward); - new ReturnInst(FRslt, forward); - - // Block "reverse" - Value* t5 = BinaryOperator::createShl(ValMask, Lo, "", reverse); - Value* t6 = BinaryOperator::createShl(ValMask, Hi, "", reverse); - Value* t7 = BinaryOperator::createOr(t6, t5, "", reverse); - Value* t8 = BinaryOperator::createXor(t7, ValMask, "", reverse); - Value* t9 = BinaryOperator::createAnd(t8, Val, "", reverse); - Value* t10 = BinaryOperator::createShl(Rep4, Lo, "", reverse); - if (32 < ValBits) - RepBitWidth = - cast(ConstantExpr::getZExt(RepBitWidth, ValTy)); - else if (32 > ValBits) - RepBitWidth = - cast(ConstantExpr::getTrunc(RepBitWidth, ValTy)); - Value* t11 = BinaryOperator::createSub(RepBitWidth, Hi, "", reverse); - Value* t13 = BinaryOperator::createLShr(Rep4, t11, "",reverse); - Value* t14 = BinaryOperator::createOr(t10, t9, "", reverse); - Value* RRslt = BinaryOperator::createOr(t14, t13, "", reverse); - new ReturnInst(RRslt, reverse); + BranchInst::Create(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 = PHINode::Create(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 = PHINode::Create(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 = PHINode::Create(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); + BranchInst::Create(result, reverse, Cond, reverse); + + // BASIC BLOCK: result + PHINode *Rplcmnt = PHINode::Create(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); + ReturnInst::Create(Rslt, result); } // Return a call to the implementation function @@ -598,7 +640,7 @@ static Instruction *LowerPartSet(CallInst *CI) { CI->getOperand(3), CI->getOperand(4) }; - return new CallInst(F, Args, sizeof(Args)/sizeof(Args[0]), CI->getName(), CI); + return CallInst::Create(F, Args, array_endof(Args), CI->getName(), CI); } @@ -662,10 +704,10 @@ void IntrinsicLowering::LowerIntrinsicCall(CallInst *CI) { 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 = 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); CI->replaceAllUsesWith(Src); break; } @@ -717,11 +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_selector: - case Intrinsic::eh_filter: - 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; @@ -780,18 +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() &&