#include "InstCombine.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/MemoryBuiltins.h"
+#include "llvm/IR/CallSite.h"
#include "llvm/IR/DataLayout.h"
-#include "llvm/Support/CallSite.h"
-#include "llvm/Support/PatternMatch.h"
+#include "llvm/IR/PatternMatch.h"
#include "llvm/Transforms/Utils/BuildLibCalls.h"
#include "llvm/Transforms/Utils/Local.h"
using namespace llvm;
using namespace PatternMatch;
+#define DEBUG_TYPE "instcombine"
+
STATISTIC(NumSimplified, "Number of library calls simplified");
/// getPromotedType - Return the specified type promoted as it would be to pass
}
Instruction *InstCombiner::SimplifyMemTransfer(MemIntrinsic *MI) {
- unsigned DstAlign = getKnownAlignment(MI->getArgOperand(0), TD);
- unsigned SrcAlign = getKnownAlignment(MI->getArgOperand(1), TD);
+ unsigned DstAlign = getKnownAlignment(MI->getArgOperand(0), DL);
+ unsigned SrcAlign = getKnownAlignment(MI->getArgOperand(1), DL);
unsigned MinAlign = std::min(DstAlign, SrcAlign);
unsigned CopyAlign = MI->getAlignment();
if (StrippedDest != MI->getArgOperand(0)) {
Type *SrcETy = cast<PointerType>(StrippedDest->getType())
->getElementType();
- if (TD && SrcETy->isSized() && TD->getTypeStoreSize(SrcETy) == Size) {
+ if (DL && SrcETy->isSized() && DL->getTypeStoreSize(SrcETy) == Size) {
// The SrcETy might be something like {{{double}}} or [1 x double]. Rip
// down through these levels if so.
SrcETy = reduceToSingleValueType(SrcETy);
}
Instruction *InstCombiner::SimplifyMemSet(MemSetInst *MI) {
- unsigned Alignment = getKnownAlignment(MI->getDest(), TD);
+ unsigned Alignment = getKnownAlignment(MI->getDest(), DL);
if (MI->getAlignment() < Alignment) {
MI->setAlignment(ConstantInt::get(MI->getAlignmentType(),
Alignment, false));
default: break;
case Intrinsic::objectsize: {
uint64_t Size;
- if (getObjectSize(II->getArgOperand(0), Size, TD, TLI))
+ if (getObjectSize(II->getArgOperand(0), Size, DL, TLI))
return ReplaceInstUsesWith(CI, ConstantInt::get(CI.getType(), Size));
return 0;
}
case Intrinsic::ppc_altivec_lvx:
case Intrinsic::ppc_altivec_lvxl:
// Turn PPC lvx -> load if the pointer is known aligned.
- if (getOrEnforceKnownAlignment(II->getArgOperand(0), 16, TD) >= 16) {
+ if (getOrEnforceKnownAlignment(II->getArgOperand(0), 16, DL) >= 16) {
Value *Ptr = Builder->CreateBitCast(II->getArgOperand(0),
PointerType::getUnqual(II->getType()));
return new LoadInst(Ptr);
case Intrinsic::ppc_altivec_stvx:
case Intrinsic::ppc_altivec_stvxl:
// Turn stvx -> store if the pointer is known aligned.
- if (getOrEnforceKnownAlignment(II->getArgOperand(1), 16, TD) >= 16) {
+ if (getOrEnforceKnownAlignment(II->getArgOperand(1), 16, DL) >= 16) {
Type *OpPtrTy =
PointerType::getUnqual(II->getArgOperand(0)->getType());
Value *Ptr = Builder->CreateBitCast(II->getArgOperand(1), OpPtrTy);
case Intrinsic::x86_sse2_storeu_pd:
case Intrinsic::x86_sse2_storeu_dq:
// Turn X86 storeu -> store if the pointer is known aligned.
- if (getOrEnforceKnownAlignment(II->getArgOperand(0), 16, TD) >= 16) {
+ if (getOrEnforceKnownAlignment(II->getArgOperand(0), 16, DL) >= 16) {
Type *OpPtrTy =
PointerType::getUnqual(II->getArgOperand(1)->getType());
Value *Ptr = Builder->CreateBitCast(II->getArgOperand(0), OpPtrTy);
break;
}
+ case Intrinsic::x86_avx_vpermilvar_ps:
+ case Intrinsic::x86_avx_vpermilvar_ps_256:
+ case Intrinsic::x86_avx_vpermilvar_pd:
+ case Intrinsic::x86_avx_vpermilvar_pd_256: {
+ // Convert vpermil* to shufflevector if the mask is constant.
+ Value *V = II->getArgOperand(1);
+ if (auto C = dyn_cast<ConstantDataVector>(V)) {
+ auto V1 = II->getArgOperand(0);
+ auto V2 = UndefValue::get(V1->getType());
+ auto Shuffle = Builder->CreateShuffleVector(V1, V2, C);
+ return ReplaceInstUsesWith(CI, Shuffle);
+ }
+ break;
+ }
+
case Intrinsic::ppc_altivec_vperm:
// Turn vperm(V1,V2,mask) -> shuffle(V1,V2,mask) if mask is a constant.
if (Constant *Mask = dyn_cast<Constant>(II->getArgOperand(2))) {
case Intrinsic::arm_neon_vst2lane:
case Intrinsic::arm_neon_vst3lane:
case Intrinsic::arm_neon_vst4lane: {
- unsigned MemAlign = getKnownAlignment(II->getArgOperand(0), TD);
+ unsigned MemAlign = getKnownAlignment(II->getArgOperand(0), DL);
unsigned AlignArg = II->getNumArgOperands() - 1;
ConstantInt *IntrAlign = dyn_cast<ConstantInt>(II->getArgOperand(AlignArg));
if (IntrAlign && IntrAlign->getZExtValue() < MemAlign) {
}
case Intrinsic::arm_neon_vmulls:
- case Intrinsic::arm_neon_vmullu: {
+ case Intrinsic::arm_neon_vmullu:
+ case Intrinsic::arm64_neon_smull:
+ case Intrinsic::arm64_neon_umull: {
Value *Arg0 = II->getArgOperand(0);
Value *Arg1 = II->getArgOperand(1);
}
// Check for constant LHS & RHS - in this case we just simplify.
- bool Zext = (II->getIntrinsicID() == Intrinsic::arm_neon_vmullu);
+ bool Zext = (II->getIntrinsicID() == Intrinsic::arm_neon_vmullu ||
+ II->getIntrinsicID() == Intrinsic::arm64_neon_umull);
VectorType *NewVT = cast<VectorType>(II->getType());
- unsigned NewWidth = NewVT->getElementType()->getIntegerBitWidth();
- if (ConstantDataVector *CV0 = dyn_cast<ConstantDataVector>(Arg0)) {
- if (ConstantDataVector *CV1 = dyn_cast<ConstantDataVector>(Arg1)) {
- VectorType* VT = cast<VectorType>(CV0->getType());
- SmallVector<Constant*, 4> NewElems;
- for (unsigned i = 0; i < VT->getNumElements(); ++i) {
- APInt CV0E =
- (cast<ConstantInt>(CV0->getAggregateElement(i)))->getValue();
- CV0E = Zext ? CV0E.zext(NewWidth) : CV0E.sext(NewWidth);
- APInt CV1E =
- (cast<ConstantInt>(CV1->getAggregateElement(i)))->getValue();
- CV1E = Zext ? CV1E.zext(NewWidth) : CV1E.sext(NewWidth);
- NewElems.push_back(
- ConstantInt::get(NewVT->getElementType(), CV0E * CV1E));
- }
- return ReplaceInstUsesWith(CI, ConstantVector::get(NewElems));
+ if (Constant *CV0 = dyn_cast<Constant>(Arg0)) {
+ if (Constant *CV1 = dyn_cast<Constant>(Arg1)) {
+ CV0 = ConstantExpr::getIntegerCast(CV0, NewVT, /*isSigned=*/!Zext);
+ CV1 = ConstantExpr::getIntegerCast(CV1, NewVT, /*isSigned=*/!Zext);
+
+ return ReplaceInstUsesWith(CI, ConstantExpr::getMul(CV0, CV1));
}
// Couldn't simplify - canonicalize constant to the RHS.
}
// Handle mul by one:
- if (ConstantDataVector *CV1 = dyn_cast<ConstantDataVector>(Arg1)) {
+ if (Constant *CV1 = dyn_cast<Constant>(Arg1))
if (ConstantInt *Splat =
- dyn_cast_or_null<ConstantInt>(CV1->getSplatValue())) {
- if (Splat->isOne()) {
- if (Zext)
- return CastInst::CreateZExtOrBitCast(Arg0, II->getType());
- // else
- return CastInst::CreateSExtOrBitCast(Arg0, II->getType());
- }
- }
- }
+ dyn_cast_or_null<ConstantInt>(CV1->getSplatValue()))
+ if (Splat->isOne())
+ return CastInst::CreateIntegerCast(Arg0, II->getType(),
+ /*isSigned=*/!Zext);
break;
}
/// passed through the varargs area, we can eliminate the use of the cast.
static bool isSafeToEliminateVarargsCast(const CallSite CS,
const CastInst * const CI,
- const DataLayout * const TD,
+ const DataLayout * const DL,
const int ix) {
if (!CI->isLosslessCast())
return false;
Type* DstTy = cast<PointerType>(CI->getType())->getElementType();
if (!SrcTy->isSized() || !DstTy->isSized())
return false;
- if (!TD || TD->getTypeAllocSize(SrcTy) != TD->getTypeAllocSize(DstTy))
+ if (!DL || DL->getTypeAllocSize(SrcTy) != DL->getTypeAllocSize(DstTy))
return false;
return true;
}
// Currently we're only working with the checking functions, memcpy_chk,
// mempcpy_chk, memmove_chk, memset_chk, strcpy_chk, stpcpy_chk, strncpy_chk,
// strcat_chk and strncat_chk.
-Instruction *InstCombiner::tryOptimizeCall(CallInst *CI, const DataLayout *TD) {
+Instruction *InstCombiner::tryOptimizeCall(CallInst *CI, const DataLayout *DL) {
if (CI->getCalledFunction() == 0) return 0;
if (Value *With = Simplifier->optimizeCall(CI)) {
// is good enough in practice and simpler than handling any number of casts.
Value *Underlying = TrampMem->stripPointerCasts();
if (Underlying != TrampMem &&
- (!Underlying->hasOneUse() || *Underlying->use_begin() != TrampMem))
+ (!Underlying->hasOneUse() || Underlying->user_back() != TrampMem))
return 0;
if (!isa<AllocaInst>(Underlying))
return 0;
IntrinsicInst *InitTrampoline = 0;
- for (Value::use_iterator I = TrampMem->use_begin(), E = TrampMem->use_end();
- I != E; I++) {
- IntrinsicInst *II = dyn_cast<IntrinsicInst>(*I);
+ for (User *U : TrampMem->users()) {
+ IntrinsicInst *II = dyn_cast<IntrinsicInst>(U);
if (!II)
return 0;
if (II->getIntrinsicID() == Intrinsic::init_trampoline) {
for (CallSite::arg_iterator I = CS.arg_begin() + FTy->getNumParams(),
E = CS.arg_end(); I != E; ++I, ++ix) {
CastInst *CI = dyn_cast<CastInst>(*I);
- if (CI && isSafeToEliminateVarargsCast(CS, CI, TD, ix)) {
+ if (CI && isSafeToEliminateVarargsCast(CS, CI, DL, ix)) {
*I = CI->getOperand(0);
Changed = true;
}
// this. None of these calls are seen as possibly dead so go ahead and
// delete the instruction now.
if (CallInst *CI = dyn_cast<CallInst>(CS.getInstruction())) {
- Instruction *I = tryOptimizeCall(CI, TD);
+ Instruction *I = tryOptimizeCall(CI, DL);
// If we changed something return the result, etc. Otherwise let
// the fallthrough check.
if (I) return EraseInstFromFunction(*I);
// the critical edge). Bail out in this case.
if (!Caller->use_empty())
if (InvokeInst *II = dyn_cast<InvokeInst>(Caller))
- for (Value::use_iterator UI = II->use_begin(), E = II->use_end();
- UI != E; ++UI)
- if (PHINode *PN = dyn_cast<PHINode>(*UI))
+ for (User *U : II->users())
+ if (PHINode *PN = dyn_cast<PHINode>(U))
if (PN->getParent() == II->getNormalDest() ||
PN->getParent() == II->getUnwindDest())
return false;
CallerPAL.getParamAttributes(i + 1).hasAttribute(i + 1,
Attribute::ByVal)) {
PointerType *ParamPTy = dyn_cast<PointerType>(ParamTy);
- if (ParamPTy == 0 || !ParamPTy->getElementType()->isSized() || TD == 0)
+ if (ParamPTy == 0 || !ParamPTy->getElementType()->isSized() || DL == 0)
return false;
Type *CurElTy = ActTy->getPointerElementType();
- if (TD->getTypeAllocSize(CurElTy) !=
- TD->getTypeAllocSize(ParamPTy->getElementType()))
+ if (DL->getTypeAllocSize(CurElTy) !=
+ DL->getTypeAllocSize(ParamPTy->getElementType()))
return false;
}
}
if (!Caller->use_empty())
ReplaceInstUsesWith(*Caller, NV);
+ else if (Caller->hasValueHandle())
+ ValueHandleBase::ValueIsRAUWd(Caller, NV);
EraseInstFromFunction(*Caller);
return true;
projects / oota-llvm.git / blobdiff