X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FIR%2FAutoUpgrade.cpp;h=12c354c89b206d272e7ca3b57edb018f1f21266c;hb=65711ad4e36f91de64aa3dde76fde143cec85c83;hp=aeefa38f74d6cb6ebe4f1092020915339ce33700;hpb=ad0ddd8e0190ef7ebb162921e53115f9d969f939;p=oota-llvm.git diff --git a/lib/IR/AutoUpgrade.cpp b/lib/IR/AutoUpgrade.cpp index aeefa38f74d..12c354c89b2 100644 --- a/lib/IR/AutoUpgrade.cpp +++ b/lib/IR/AutoUpgrade.cpp @@ -27,6 +27,7 @@ #include "llvm/IR/LLVMContext.h" #include "llvm/IR/Module.h" #include "llvm/Support/ErrorHandling.h" +#include "llvm/Support/Regex.h" #include using namespace llvm; @@ -92,8 +93,42 @@ static bool UpgradeIntrinsicFunction1(Function *F, Function *&NewFn) { F->arg_begin()->getType()); return true; } + Regex vldRegex("^arm\\.neon\\.vld([1234]|[234]lane)\\.v[a-z0-9]*$"); + if (vldRegex.match(Name)) { + auto fArgs = F->getFunctionType()->params(); + SmallVector Tys(fArgs.begin(), fArgs.end()); + // Can't use Intrinsic::getDeclaration here as the return types might + // then only be structurally equal. + FunctionType* fType = FunctionType::get(F->getReturnType(), Tys, false); + NewFn = Function::Create(fType, F->getLinkage(), + "llvm." + Name + ".p0i8", F->getParent()); + return true; + } + Regex vstRegex("^arm\\.neon\\.vst([1234]|[234]lane)\\.v[a-z0-9]*$"); + if (vstRegex.match(Name)) { + static const Intrinsic::ID StoreInts[] = {Intrinsic::arm_neon_vst1, + Intrinsic::arm_neon_vst2, + Intrinsic::arm_neon_vst3, + Intrinsic::arm_neon_vst4}; + + static const Intrinsic::ID StoreLaneInts[] = { + Intrinsic::arm_neon_vst2lane, Intrinsic::arm_neon_vst3lane, + Intrinsic::arm_neon_vst4lane + }; + + auto fArgs = F->getFunctionType()->params(); + Type *Tys[] = {fArgs[0], fArgs[1]}; + if (Name.find("lane") == StringRef::npos) + NewFn = Intrinsic::getDeclaration(F->getParent(), + StoreInts[fArgs.size() - 3], Tys); + else + NewFn = Intrinsic::getDeclaration(F->getParent(), + StoreLaneInts[fArgs.size() - 5], Tys); + return true; + } break; } + case 'c': { if (Name.startswith("ctlz.") && F->arg_size() == 1) { F->setName(Name + ".old"); @@ -132,6 +167,7 @@ static bool UpgradeIntrinsicFunction1(Function *F, Function *&NewFn) { Name.startswith("x86.avx2.vbroadcast") || Name.startswith("x86.avx2.pbroadcast") || Name.startswith("x86.avx.vpermil.") || + Name.startswith("x86.sse41.pmovsx") || Name == "x86.avx.vinsertf128.pd.256" || Name == "x86.avx.vinsertf128.ps.256" || Name == "x86.avx.vinsertf128.si.256" || @@ -164,6 +200,7 @@ static bool UpgradeIntrinsicFunction1(Function *F, Function *&NewFn) { Name == "x86.avx2.pblendd.128" || Name == "x86.avx2.pblendd.256" || Name == "x86.avx2.vbroadcasti128" || + Name == "x86.xop.vpcmov" || (Name.startswith("x86.xop.vpcom") && F->arg_size() == 2)) { NewFn = nullptr; return true; @@ -327,7 +364,7 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) { Function *F = CI->getCalledFunction(); LLVMContext &C = CI->getContext(); IRBuilder<> Builder(C); - Builder.SetInsertPoint(CI->getParent(), CI); + Builder.SetInsertPoint(CI->getParent(), CI->getIterator()); assert(F && "Intrinsic call is not direct?"); @@ -353,7 +390,7 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) { Name == "llvm.x86.avx.movnt.ps.256" || Name == "llvm.x86.avx.movnt.pd.256") { IRBuilder<> Builder(C); - Builder.SetInsertPoint(CI->getParent(), CI); + Builder.SetInsertPoint(CI->getParent(), CI->getIterator()); Module *M = F->getParent(); SmallVector Elts; @@ -370,7 +407,7 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) { "cast"); StoreInst *SI = Builder.CreateStore(Arg1, BC); SI->setMetadata(M->getMDKindID("nontemporal"), Node); - SI->setAlignment(16); + SI->setAlignment(32); // Remove intrinsic. CI->eraseFromParent(); @@ -421,6 +458,16 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) { Rep = Builder.CreateCall(VPCOM, {CI->getArgOperand(0), CI->getArgOperand(1), Builder.getInt8(Imm)}); + } else if (Name == "llvm.x86.xop.vpcmov") { + Value *Arg0 = CI->getArgOperand(0); + Value *Arg1 = CI->getArgOperand(1); + Value *Sel = CI->getArgOperand(2); + unsigned NumElts = CI->getType()->getVectorNumElements(); + Constant *MinusOne = ConstantVector::getSplat(NumElts, Builder.getInt64(-1)); + Value *NotSel = Builder.CreateXor(Sel, MinusOne); + Value *Sel0 = Builder.CreateAnd(Arg0, Sel); + Value *Sel1 = Builder.CreateAnd(Arg1, NotSel); + Rep = Builder.CreateOr(Sel0, Sel1); } else if (Name == "llvm.x86.sse42.crc32.64.8") { Function *CRC32 = Intrinsic::getDeclaration(F->getParent(), Intrinsic::x86_sse42_crc32_32_8); @@ -440,6 +487,19 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) { for (unsigned I = 0; I < EltNum; ++I) Rep = Builder.CreateInsertElement(Rep, Load, ConstantInt::get(I32Ty, I)); + } else if (Name.startswith("llvm.x86.sse41.pmovsx")) { + VectorType *SrcTy = cast(CI->getArgOperand(0)->getType()); + VectorType *DstTy = cast(CI->getType()); + unsigned NumDstElts = DstTy->getNumElements(); + + // Extract a subvector of the first NumDstElts lanes and sign extend. + SmallVector ShuffleMask; + for (int i = 0; i != (int)NumDstElts; ++i) + ShuffleMask.push_back(i); + + Value *SV = Builder.CreateShuffleVector( + CI->getArgOperand(0), UndefValue::get(SrcTy), ShuffleMask); + Rep = Builder.CreateSExt(SV, DstTy); } else if (Name == "llvm.x86.avx2.vbroadcasti128") { // Replace vbroadcasts with a vector shuffle. Type *VT = VectorType::get(Type::getInt64Ty(C), 2); @@ -527,10 +587,10 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) { unsigned Imm = cast(CI->getArgOperand(2))->getZExtValue(); VectorType *VecTy = cast(CI->getType()); unsigned NumElts = VecTy->getNumElements(); - + // Mask off the high bits of the immediate value; hardware ignores those. Imm = Imm & 1; - + // Extend the second operand into a vector that is twice as big. Value *UndefV = UndefValue::get(Op1->getType()); SmallVector Idxs; @@ -572,7 +632,7 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) { unsigned Imm = cast(CI->getArgOperand(1))->getZExtValue(); VectorType *VecTy = cast(CI->getType()); unsigned NumElts = VecTy->getNumElements(); - + // Mask off the high bits of the immediate value; hardware ignores those. Imm = Imm & 1; @@ -637,6 +697,27 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) { default: llvm_unreachable("Unknown function for CallInst upgrade."); + case Intrinsic::arm_neon_vld1: + case Intrinsic::arm_neon_vld2: + case Intrinsic::arm_neon_vld3: + case Intrinsic::arm_neon_vld4: + case Intrinsic::arm_neon_vld2lane: + case Intrinsic::arm_neon_vld3lane: + case Intrinsic::arm_neon_vld4lane: + case Intrinsic::arm_neon_vst1: + case Intrinsic::arm_neon_vst2: + case Intrinsic::arm_neon_vst3: + case Intrinsic::arm_neon_vst4: + case Intrinsic::arm_neon_vst2lane: + case Intrinsic::arm_neon_vst3lane: + case Intrinsic::arm_neon_vst4lane: { + SmallVector Args(CI->arg_operands().begin(), + CI->arg_operands().end()); + CI->replaceAllUsesWith(Builder.CreateCall(NewFn, Args)); + CI->eraseFromParent(); + return; + } + case Intrinsic::ctlz: case Intrinsic::cttz: assert(CI->getNumArgOperands() == 1 &&