#include "llvm/Support/CallSite.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Analysis/MemoryBuiltins.h"
+#include "llvm/Transforms/Utils/BuildLibCalls.h"
+#include "llvm/Transforms/Utils/Local.h"
using namespace llvm;
/// getPromotedType - Return the specified type promoted as it would be to pass
return Ty;
}
-/// EnforceKnownAlignment - If the specified pointer points to an object that
-/// we control, modify the object's alignment to PrefAlign. This isn't
-/// often possible though. If alignment is important, a more reliable approach
-/// is to simply align all global variables and allocation instructions to
-/// their preferred alignment from the beginning.
-///
-static unsigned EnforceKnownAlignment(Value *V,
- unsigned Align, unsigned PrefAlign) {
-
- User *U = dyn_cast<User>(V);
- if (!U) return Align;
-
- switch (Operator::getOpcode(U)) {
- default: break;
- case Instruction::BitCast:
- return EnforceKnownAlignment(U->getOperand(0), Align, PrefAlign);
- case Instruction::GetElementPtr: {
- // If all indexes are zero, it is just the alignment of the base pointer.
- bool AllZeroOperands = true;
- for (User::op_iterator i = U->op_begin() + 1, e = U->op_end(); i != e; ++i)
- if (!isa<Constant>(*i) ||
- !cast<Constant>(*i)->isNullValue()) {
- AllZeroOperands = false;
- break;
- }
-
- if (AllZeroOperands) {
- // Treat this like a bitcast.
- return EnforceKnownAlignment(U->getOperand(0), Align, PrefAlign);
- }
- break;
- }
- }
-
- if (GlobalValue *GV = dyn_cast<GlobalValue>(V)) {
- // If there is a large requested alignment and we can, bump up the alignment
- // of the global.
- if (!GV->isDeclaration()) {
- if (GV->getAlignment() >= PrefAlign)
- Align = GV->getAlignment();
- else {
- GV->setAlignment(PrefAlign);
- Align = PrefAlign;
- }
- }
- } else if (AllocaInst *AI = dyn_cast<AllocaInst>(V)) {
- // If there is a requested alignment and if this is an alloca, round up.
- if (AI->getAlignment() >= PrefAlign)
- Align = AI->getAlignment();
- else {
- AI->setAlignment(PrefAlign);
- Align = PrefAlign;
- }
- }
-
- return Align;
-}
-
-/// GetOrEnforceKnownAlignment - If the specified pointer has an alignment that
-/// we can determine, return it, otherwise return 0. If PrefAlign is specified,
-/// and it is more than the alignment of the ultimate object, see if we can
-/// increase the alignment of the ultimate object, making this check succeed.
-unsigned InstCombiner::GetOrEnforceKnownAlignment(Value *V,
- unsigned PrefAlign) {
- unsigned BitWidth = TD ? TD->getTypeSizeInBits(V->getType()) :
- sizeof(PrefAlign) * CHAR_BIT;
- APInt Mask = APInt::getAllOnesValue(BitWidth);
- APInt KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
- ComputeMaskedBits(V, Mask, KnownZero, KnownOne);
- unsigned TrailZ = KnownZero.countTrailingOnes();
- unsigned Align = 1u << std::min(BitWidth - 1, TrailZ);
-
- if (PrefAlign > Align)
- Align = EnforceKnownAlignment(V, Align, PrefAlign);
-
- // We don't need to make any adjustment.
- return Align;
-}
Instruction *InstCombiner::SimplifyMemTransfer(MemIntrinsic *MI) {
- unsigned DstAlign = GetOrEnforceKnownAlignment(MI->getOperand(1));
- unsigned SrcAlign = GetOrEnforceKnownAlignment(MI->getOperand(2));
+ unsigned DstAlign = getKnownAlignment(MI->getArgOperand(0), TD);
+ unsigned SrcAlign = getKnownAlignment(MI->getArgOperand(1), TD);
unsigned MinAlign = std::min(DstAlign, SrcAlign);
unsigned CopyAlign = MI->getAlignment();
// If MemCpyInst length is 1/2/4/8 bytes then replace memcpy with
// load/store.
- ConstantInt *MemOpLength = dyn_cast<ConstantInt>(MI->getOperand(3));
+ ConstantInt *MemOpLength = dyn_cast<ConstantInt>(MI->getArgOperand(2));
if (MemOpLength == 0) return 0;
// Source and destination pointer types are always "i8*" for intrinsic. See
return 0; // If not 1/2/4/8 bytes, exit.
// Use an integer load+store unless we can find something better.
- Type *NewPtrTy =
- PointerType::getUnqual(IntegerType::get(MI->getContext(), Size<<3));
+ unsigned SrcAddrSp =
+ cast<PointerType>(MI->getArgOperand(1)->getType())->getAddressSpace();
+ unsigned DstAddrSp =
+ cast<PointerType>(MI->getArgOperand(0)->getType())->getAddressSpace();
+
+ const IntegerType* IntType = IntegerType::get(MI->getContext(), Size<<3);
+ Type *NewSrcPtrTy = PointerType::get(IntType, SrcAddrSp);
+ Type *NewDstPtrTy = PointerType::get(IntType, DstAddrSp);
// Memcpy forces the use of i8* for the source and destination. That means
// that if you're using memcpy to move one double around, you'll get a cast
// an i64 load+store, here because this improves the odds that the source or
// dest address will be promotable. See if we can find a better type than the
// integer datatype.
- Value *StrippedDest = MI->getOperand(1)->stripPointerCasts();
- if (StrippedDest != MI->getOperand(1)) {
+ Value *StrippedDest = MI->getArgOperand(0)->stripPointerCasts();
+ if (StrippedDest != MI->getArgOperand(0)) {
const Type *SrcETy = cast<PointerType>(StrippedDest->getType())
->getElementType();
if (TD && SrcETy->isSized() && TD->getTypeStoreSize(SrcETy) == Size) {
break;
}
- if (SrcETy->isSingleValueType())
- NewPtrTy = PointerType::getUnqual(SrcETy);
+ if (SrcETy->isSingleValueType()) {
+ NewSrcPtrTy = PointerType::get(SrcETy, SrcAddrSp);
+ NewDstPtrTy = PointerType::get(SrcETy, DstAddrSp);
+ }
}
}
SrcAlign = std::max(SrcAlign, CopyAlign);
DstAlign = std::max(DstAlign, CopyAlign);
- Value *Src = Builder->CreateBitCast(MI->getOperand(2), NewPtrTy);
- Value *Dest = Builder->CreateBitCast(MI->getOperand(1), NewPtrTy);
- Instruction *L = new LoadInst(Src, "tmp", false, SrcAlign);
+ Value *Src = Builder->CreateBitCast(MI->getArgOperand(1), NewSrcPtrTy);
+ Value *Dest = Builder->CreateBitCast(MI->getArgOperand(0), NewDstPtrTy);
+ Instruction *L = new LoadInst(Src, "tmp", MI->isVolatile(), SrcAlign);
InsertNewInstBefore(L, *MI);
- InsertNewInstBefore(new StoreInst(L, Dest, false, DstAlign), *MI);
+ InsertNewInstBefore(new StoreInst(L, Dest, MI->isVolatile(), DstAlign),
+ *MI);
// Set the size of the copy to 0, it will be deleted on the next iteration.
- MI->setOperand(3, Constant::getNullValue(MemOpLength->getType()));
+ MI->setArgOperand(2, Constant::getNullValue(MemOpLength->getType()));
return MI;
}
Instruction *InstCombiner::SimplifyMemSet(MemSetInst *MI) {
- unsigned Alignment = GetOrEnforceKnownAlignment(MI->getDest());
+ unsigned Alignment = getKnownAlignment(MI->getDest(), TD);
if (MI->getAlignment() < Alignment) {
MI->setAlignment(ConstantInt::get(MI->getAlignmentType(),
Alignment, false));
// Extract the length and alignment and fill if they are constant.
ConstantInt *LenC = dyn_cast<ConstantInt>(MI->getLength());
ConstantInt *FillC = dyn_cast<ConstantInt>(MI->getValue());
- if (!LenC || !FillC || !FillC->getType()->isInteger(8))
+ if (!LenC || !FillC || !FillC->getType()->isIntegerTy(8))
return 0;
uint64_t Len = LenC->getZExtValue();
Alignment = MI->getAlignment();
const Type *ITy = IntegerType::get(MI->getContext(), Len*8); // n=1 -> i8.
Value *Dest = MI->getDest();
- Dest = Builder->CreateBitCast(Dest, PointerType::getUnqual(ITy));
+ unsigned DstAddrSp = cast<PointerType>(Dest->getType())->getAddressSpace();
+ Type *NewDstPtrTy = PointerType::get(ITy, DstAddrSp);
+ Dest = Builder->CreateBitCast(Dest, NewDstPtrTy);
// Alignment 0 is identity for alignment 1 for memset, but not store.
if (Alignment == 0) Alignment = 1;
return 0;
}
-
/// visitCallInst - CallInst simplification. This mostly only handles folding
/// of intrinsic instructions. For normal calls, it allows visitCallSite to do
/// the heavy lifting.
Instruction *InstCombiner::visitCallInst(CallInst &CI) {
if (isFreeCall(&CI))
return visitFree(CI);
+ if (isMalloc(&CI))
+ return visitMalloc(CI);
// If the caller function is nounwind, mark the call as nounwind, even if the
// callee isn't.
IntrinsicInst *II = dyn_cast<IntrinsicInst>(&CI);
if (!II) return visitCallSite(&CI);
-
+
// Intrinsics cannot occur in an invoke, so handle them here instead of in
// visitCallSite.
if (MemIntrinsic *MI = dyn_cast<MemIntrinsic>(II)) {
// memmove/cpy/set of zero bytes is a noop.
if (Constant *NumBytes = dyn_cast<Constant>(MI->getLength())) {
- if (NumBytes->isNullValue()) return EraseInstFromFunction(CI);
+ if (NumBytes->isNullValue())
+ return EraseInstFromFunction(CI);
if (ConstantInt *CI = dyn_cast<ConstantInt>(NumBytes))
if (CI->getZExtValue() == 1) {
// alignment is sufficient.
}
}
+
+ // No other transformations apply to volatile transfers.
+ if (MI->isVolatile())
+ return 0;
// If we have a memmove and the source operation is a constant global,
// then the source and dest pointers can't alias, so we can change this
if (GVSrc->isConstant()) {
Module *M = CI.getParent()->getParent()->getParent();
Intrinsic::ID MemCpyID = Intrinsic::memcpy;
- const Type *Tys[1];
- Tys[0] = CI.getOperand(3)->getType();
- CI.setOperand(0,
- Intrinsic::getDeclaration(M, MemCpyID, Tys, 1));
+ const Type *Tys[3] = { CI.getArgOperand(0)->getType(),
+ CI.getArgOperand(1)->getType(),
+ CI.getArgOperand(2)->getType() };
+ CI.setCalledFunction(Intrinsic::getDeclaration(M, MemCpyID, Tys, 3));
Changed = true;
}
}
if (Instruction *I = SimplifyMemSet(MSI))
return I;
}
-
+
if (Changed) return II;
}
switch (II->getIntrinsicID()) {
default: break;
+ case Intrinsic::objectsize: {
+ // We need target data for just about everything so depend on it.
+ if (!TD) break;
+
+ const Type *ReturnTy = CI.getType();
+ uint64_t DontKnow = II->getArgOperand(1) == Builder->getTrue() ? 0 : -1ULL;
+
+ // Get to the real allocated thing and offset as fast as possible.
+ Value *Op1 = II->getArgOperand(0)->stripPointerCasts();
+
+ uint64_t Offset = 0;
+ uint64_t Size = -1ULL;
+
+ // Try to look through constant GEPs.
+ if (GEPOperator *GEP = dyn_cast<GEPOperator>(Op1)) {
+ if (!GEP->hasAllConstantIndices()) break;
+
+ // Get the current byte offset into the thing. Use the original
+ // operand in case we're looking through a bitcast.
+ SmallVector<Value*, 8> Ops(GEP->idx_begin(), GEP->idx_end());
+ Offset = TD->getIndexedOffset(GEP->getPointerOperandType(),
+ Ops.data(), Ops.size());
+
+ Op1 = GEP->getPointerOperand()->stripPointerCasts();
+
+ // Make sure we're not a constant offset from an external
+ // global.
+ if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Op1))
+ if (!GV->hasDefinitiveInitializer()) break;
+ }
+
+ // If we've stripped down to a single global variable that we
+ // can know the size of then just return that.
+ if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Op1)) {
+ if (GV->hasDefinitiveInitializer()) {
+ Constant *C = GV->getInitializer();
+ Size = TD->getTypeAllocSize(C->getType());
+ } else {
+ // Can't determine size of the GV.
+ Constant *RetVal = ConstantInt::get(ReturnTy, DontKnow);
+ return ReplaceInstUsesWith(CI, RetVal);
+ }
+ } else if (AllocaInst *AI = dyn_cast<AllocaInst>(Op1)) {
+ // Get alloca size.
+ if (AI->getAllocatedType()->isSized()) {
+ Size = TD->getTypeAllocSize(AI->getAllocatedType());
+ if (AI->isArrayAllocation()) {
+ const ConstantInt *C = dyn_cast<ConstantInt>(AI->getArraySize());
+ if (!C) break;
+ Size *= C->getZExtValue();
+ }
+ }
+ } else if (CallInst *MI = extractMallocCall(Op1)) {
+ // Get allocation size.
+ const Type* MallocType = getMallocAllocatedType(MI);
+ if (MallocType && MallocType->isSized())
+ if (Value *NElems = getMallocArraySize(MI, TD, true))
+ if (ConstantInt *NElements = dyn_cast<ConstantInt>(NElems))
+ Size = NElements->getZExtValue() * TD->getTypeAllocSize(MallocType);
+ }
+
+ // Do not return "I don't know" here. Later optimization passes could
+ // make it possible to evaluate objectsize to a constant.
+ if (Size == -1ULL)
+ break;
+
+ if (Size < Offset) {
+ // Out of bound reference? Negative index normalized to large
+ // index? Just return "I don't know".
+ return ReplaceInstUsesWith(CI, ConstantInt::get(ReturnTy, DontKnow));
+ }
+ return ReplaceInstUsesWith(CI, ConstantInt::get(ReturnTy, Size-Offset));
+ }
case Intrinsic::bswap:
// bswap(bswap(x)) -> x
- if (IntrinsicInst *Operand = dyn_cast<IntrinsicInst>(II->getOperand(1)))
+ if (IntrinsicInst *Operand = dyn_cast<IntrinsicInst>(II->getArgOperand(0)))
if (Operand->getIntrinsicID() == Intrinsic::bswap)
- return ReplaceInstUsesWith(CI, Operand->getOperand(1));
+ return ReplaceInstUsesWith(CI, Operand->getArgOperand(0));
// bswap(trunc(bswap(x))) -> trunc(lshr(x, c))
- if (TruncInst *TI = dyn_cast<TruncInst>(II->getOperand(1))) {
+ if (TruncInst *TI = dyn_cast<TruncInst>(II->getArgOperand(0))) {
if (IntrinsicInst *Operand = dyn_cast<IntrinsicInst>(TI->getOperand(0)))
if (Operand->getIntrinsicID() == Intrinsic::bswap) {
unsigned C = Operand->getType()->getPrimitiveSizeInBits() -
TI->getType()->getPrimitiveSizeInBits();
Value *CV = ConstantInt::get(Operand->getType(), C);
- Value *V = Builder->CreateLShr(Operand->getOperand(1), CV);
+ Value *V = Builder->CreateLShr(Operand->getArgOperand(0), CV);
return new TruncInst(V, TI->getType());
}
}
break;
case Intrinsic::powi:
- if (ConstantInt *Power = dyn_cast<ConstantInt>(II->getOperand(2))) {
+ if (ConstantInt *Power = dyn_cast<ConstantInt>(II->getArgOperand(1))) {
// powi(x, 0) -> 1.0
if (Power->isZero())
return ReplaceInstUsesWith(CI, ConstantFP::get(CI.getType(), 1.0));
// powi(x, 1) -> x
if (Power->isOne())
- return ReplaceInstUsesWith(CI, II->getOperand(1));
+ return ReplaceInstUsesWith(CI, II->getArgOperand(0));
// powi(x, -1) -> 1/x
if (Power->isAllOnesValue())
return BinaryOperator::CreateFDiv(ConstantFP::get(CI.getType(), 1.0),
- II->getOperand(1));
+ II->getArgOperand(0));
}
break;
case Intrinsic::cttz: {
// If all bits below the first known one are known zero,
// this value is constant.
- const IntegerType *IT = cast<IntegerType>(II->getOperand(1)->getType());
+ const IntegerType *IT = cast<IntegerType>(II->getArgOperand(0)->getType());
uint32_t BitWidth = IT->getBitWidth();
APInt KnownZero(BitWidth, 0);
APInt KnownOne(BitWidth, 0);
- ComputeMaskedBits(II->getOperand(1), APInt::getAllOnesValue(BitWidth),
+ ComputeMaskedBits(II->getArgOperand(0), APInt::getAllOnesValue(BitWidth),
KnownZero, KnownOne);
unsigned TrailingZeros = KnownOne.countTrailingZeros();
APInt Mask(APInt::getLowBitsSet(BitWidth, TrailingZeros));
case Intrinsic::ctlz: {
// If all bits above the first known one are known zero,
// this value is constant.
- const IntegerType *IT = cast<IntegerType>(II->getOperand(1)->getType());
+ const IntegerType *IT = cast<IntegerType>(II->getArgOperand(0)->getType());
uint32_t BitWidth = IT->getBitWidth();
APInt KnownZero(BitWidth, 0);
APInt KnownOne(BitWidth, 0);
- ComputeMaskedBits(II->getOperand(1), APInt::getAllOnesValue(BitWidth),
+ ComputeMaskedBits(II->getArgOperand(0), APInt::getAllOnesValue(BitWidth),
KnownZero, KnownOne);
unsigned LeadingZeros = KnownOne.countLeadingZeros();
APInt Mask(APInt::getHighBitsSet(BitWidth, LeadingZeros));
}
break;
case Intrinsic::uadd_with_overflow: {
- Value *LHS = II->getOperand(1), *RHS = II->getOperand(2);
- const IntegerType *IT = cast<IntegerType>(II->getOperand(1)->getType());
+ Value *LHS = II->getArgOperand(0), *RHS = II->getArgOperand(1);
+ const IntegerType *IT = cast<IntegerType>(II->getArgOperand(0)->getType());
uint32_t BitWidth = IT->getBitWidth();
APInt Mask = APInt::getSignBit(BitWidth);
APInt LHSKnownZero(BitWidth, 0);
// FALL THROUGH uadd into sadd
case Intrinsic::sadd_with_overflow:
// Canonicalize constants into the RHS.
- if (isa<Constant>(II->getOperand(1)) &&
- !isa<Constant>(II->getOperand(2))) {
- Value *LHS = II->getOperand(1);
- II->setOperand(1, II->getOperand(2));
- II->setOperand(2, LHS);
+ if (isa<Constant>(II->getArgOperand(0)) &&
+ !isa<Constant>(II->getArgOperand(1))) {
+ Value *LHS = II->getArgOperand(0);
+ II->setArgOperand(0, II->getArgOperand(1));
+ II->setArgOperand(1, LHS);
return II;
}
// X + undef -> undef
- if (isa<UndefValue>(II->getOperand(2)))
+ if (isa<UndefValue>(II->getArgOperand(1)))
return ReplaceInstUsesWith(CI, UndefValue::get(II->getType()));
- if (ConstantInt *RHS = dyn_cast<ConstantInt>(II->getOperand(2))) {
+ if (ConstantInt *RHS = dyn_cast<ConstantInt>(II->getArgOperand(1))) {
// X + 0 -> {X, false}
if (RHS->isZero()) {
Constant *V[] = {
- UndefValue::get(II->getOperand(0)->getType()),
+ UndefValue::get(II->getArgOperand(0)->getType()),
ConstantInt::getFalse(II->getContext())
};
Constant *Struct = ConstantStruct::get(II->getContext(), V, 2, false);
- return InsertValueInst::Create(Struct, II->getOperand(1), 0);
+ return InsertValueInst::Create(Struct, II->getArgOperand(0), 0);
}
}
break;
case Intrinsic::ssub_with_overflow:
// undef - X -> undef
// X - undef -> undef
- if (isa<UndefValue>(II->getOperand(1)) ||
- isa<UndefValue>(II->getOperand(2)))
+ if (isa<UndefValue>(II->getArgOperand(0)) ||
+ isa<UndefValue>(II->getArgOperand(1)))
return ReplaceInstUsesWith(CI, UndefValue::get(II->getType()));
- if (ConstantInt *RHS = dyn_cast<ConstantInt>(II->getOperand(2))) {
+ if (ConstantInt *RHS = dyn_cast<ConstantInt>(II->getArgOperand(1))) {
// X - 0 -> {X, false}
if (RHS->isZero()) {
Constant *V[] = {
- UndefValue::get(II->getOperand(1)->getType()),
+ UndefValue::get(II->getArgOperand(0)->getType()),
ConstantInt::getFalse(II->getContext())
};
Constant *Struct = ConstantStruct::get(II->getContext(), V, 2, false);
- return InsertValueInst::Create(Struct, II->getOperand(1), 0);
+ return InsertValueInst::Create(Struct, II->getArgOperand(0), 0);
}
}
break;
case Intrinsic::umul_with_overflow:
case Intrinsic::smul_with_overflow:
// Canonicalize constants into the RHS.
- if (isa<Constant>(II->getOperand(1)) &&
- !isa<Constant>(II->getOperand(2))) {
- Value *LHS = II->getOperand(1);
- II->setOperand(1, II->getOperand(2));
- II->setOperand(2, LHS);
+ if (isa<Constant>(II->getArgOperand(0)) &&
+ !isa<Constant>(II->getArgOperand(1))) {
+ Value *LHS = II->getArgOperand(0);
+ II->setArgOperand(0, II->getArgOperand(1));
+ II->setArgOperand(1, LHS);
return II;
}
// X * undef -> undef
- if (isa<UndefValue>(II->getOperand(2)))
+ if (isa<UndefValue>(II->getArgOperand(1)))
return ReplaceInstUsesWith(CI, UndefValue::get(II->getType()));
- if (ConstantInt *RHSI = dyn_cast<ConstantInt>(II->getOperand(2))) {
+ if (ConstantInt *RHSI = dyn_cast<ConstantInt>(II->getArgOperand(1))) {
// X*0 -> {0, false}
if (RHSI->isZero())
return ReplaceInstUsesWith(CI, Constant::getNullValue(II->getType()));
// X * 1 -> {X, false}
if (RHSI->equalsInt(1)) {
Constant *V[] = {
- UndefValue::get(II->getOperand(1)->getType()),
+ UndefValue::get(II->getArgOperand(0)->getType()),
ConstantInt::getFalse(II->getContext())
};
Constant *Struct = ConstantStruct::get(II->getContext(), V, 2, false);
- return InsertValueInst::Create(Struct, II->getOperand(1), 0);
+ return InsertValueInst::Create(Struct, II->getArgOperand(0), 0);
}
}
break;
case Intrinsic::x86_sse2_loadu_dq:
// Turn PPC lvx -> load if the pointer is known aligned.
// Turn X86 loadups -> load if the pointer is known aligned.
- if (GetOrEnforceKnownAlignment(II->getOperand(1), 16) >= 16) {
- Value *Ptr = Builder->CreateBitCast(II->getOperand(1),
+ if (getOrEnforceKnownAlignment(II->getArgOperand(0), 16, TD) >= 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->getOperand(2), 16) >= 16) {
+ if (getOrEnforceKnownAlignment(II->getArgOperand(1), 16, TD) >= 16) {
const Type *OpPtrTy =
- PointerType::getUnqual(II->getOperand(1)->getType());
- Value *Ptr = Builder->CreateBitCast(II->getOperand(2), OpPtrTy);
- return new StoreInst(II->getOperand(1), Ptr);
+ PointerType::getUnqual(II->getArgOperand(0)->getType());
+ Value *Ptr = Builder->CreateBitCast(II->getArgOperand(1), OpPtrTy);
+ return new StoreInst(II->getArgOperand(0), Ptr);
}
break;
case Intrinsic::x86_sse_storeu_ps:
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->getOperand(1), 16) >= 16) {
+ if (getOrEnforceKnownAlignment(II->getArgOperand(0), 16, TD) >= 16) {
const Type *OpPtrTy =
- PointerType::getUnqual(II->getOperand(2)->getType());
- Value *Ptr = Builder->CreateBitCast(II->getOperand(1), OpPtrTy);
- return new StoreInst(II->getOperand(2), Ptr);
+ PointerType::getUnqual(II->getArgOperand(1)->getType());
+ Value *Ptr = Builder->CreateBitCast(II->getArgOperand(0), OpPtrTy);
+ return new StoreInst(II->getArgOperand(1), Ptr);
}
break;
-
- case Intrinsic::x86_sse_cvttss2si: {
- // These intrinsics only demands the 0th element of its input vector. If
+
+ case Intrinsic::x86_sse_cvtss2si:
+ case Intrinsic::x86_sse_cvtss2si64:
+ case Intrinsic::x86_sse_cvttss2si:
+ case Intrinsic::x86_sse_cvttss2si64:
+ case Intrinsic::x86_sse2_cvtsd2si:
+ case Intrinsic::x86_sse2_cvtsd2si64:
+ case Intrinsic::x86_sse2_cvttsd2si:
+ case Intrinsic::x86_sse2_cvttsd2si64: {
+ // These intrinsics only demand the 0th element of their input vectors. If
// we can simplify the input based on that, do so now.
unsigned VWidth =
- cast<VectorType>(II->getOperand(1)->getType())->getNumElements();
+ cast<VectorType>(II->getArgOperand(0)->getType())->getNumElements();
APInt DemandedElts(VWidth, 1);
APInt UndefElts(VWidth, 0);
- if (Value *V = SimplifyDemandedVectorElts(II->getOperand(1), DemandedElts,
- UndefElts)) {
- II->setOperand(1, V);
+ if (Value *V = SimplifyDemandedVectorElts(II->getArgOperand(0),
+ DemandedElts, UndefElts)) {
+ II->setArgOperand(0, V);
return II;
}
break;
}
-
+
case Intrinsic::ppc_altivec_vperm:
// Turn vperm(V1,V2,mask) -> shuffle(V1,V2,mask) if mask is a constant.
- if (ConstantVector *Mask = dyn_cast<ConstantVector>(II->getOperand(3))) {
+ if (ConstantVector *Mask = dyn_cast<ConstantVector>(II->getArgOperand(2))) {
assert(Mask->getNumOperands() == 16 && "Bad type for intrinsic!");
// Check that all of the elements are integer constants or undefs.
if (AllEltsOk) {
// Cast the input vectors to byte vectors.
- Value *Op0 = Builder->CreateBitCast(II->getOperand(1), Mask->getType());
- Value *Op1 = Builder->CreateBitCast(II->getOperand(2), Mask->getType());
+ Value *Op0 = Builder->CreateBitCast(II->getArgOperand(0),
+ Mask->getType());
+ Value *Op1 = Builder->CreateBitCast(II->getArgOperand(1),
+ Mask->getType());
Value *Result = UndefValue::get(Op0->getType());
// Only extract each element once.
}
break;
+ 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: {
+ unsigned MemAlign = getKnownAlignment(II->getArgOperand(0), TD);
+ unsigned AlignArg = II->getNumArgOperands() - 1;
+ ConstantInt *IntrAlign = dyn_cast<ConstantInt>(II->getArgOperand(AlignArg));
+ if (IntrAlign && IntrAlign->getZExtValue() < MemAlign) {
+ II->setArgOperand(AlignArg,
+ ConstantInt::get(Type::getInt32Ty(II->getContext()),
+ MemAlign, false));
+ return II;
+ }
+ break;
+ }
+
case Intrinsic::stackrestore: {
// If the save is right next to the restore, remove the restore. This can
// happen when variable allocas are DCE'd.
- if (IntrinsicInst *SS = dyn_cast<IntrinsicInst>(II->getOperand(1))) {
+ if (IntrinsicInst *SS = dyn_cast<IntrinsicInst>(II->getArgOperand(0))) {
if (SS->getIntrinsicID() == Intrinsic::stacksave) {
BasicBlock::iterator BI = SS;
if (&*++BI == II)
return EraseInstFromFunction(CI);
break;
}
- case Intrinsic::objectsize: {
- ConstantInt *Const = dyn_cast<ConstantInt>(II->getOperand(2));
-
- if (!Const) return 0;
-
- const Type *Ty = CI.getType();
-
- if (Const->getZExtValue() == 0)
- return ReplaceInstUsesWith(CI, Constant::getAllOnesValue(Ty));
- else
- return ReplaceInstUsesWith(CI, ConstantInt::get(Ty, 0));
- }
}
return visitCallSite(II);
return true;
}
+namespace {
+class InstCombineFortifiedLibCalls : public SimplifyFortifiedLibCalls {
+ InstCombiner *IC;
+protected:
+ void replaceCall(Value *With) {
+ NewInstruction = IC->ReplaceInstUsesWith(*CI, With);
+ }
+ bool isFoldable(unsigned SizeCIOp, unsigned SizeArgOp, bool isString) const {
+ if (CI->getArgOperand(SizeCIOp) == CI->getArgOperand(SizeArgOp))
+ return true;
+ if (ConstantInt *SizeCI =
+ dyn_cast<ConstantInt>(CI->getArgOperand(SizeCIOp))) {
+ if (SizeCI->isAllOnesValue())
+ return true;
+ if (isString)
+ return SizeCI->getZExtValue() >=
+ GetStringLength(CI->getArgOperand(SizeArgOp));
+ if (ConstantInt *Arg = dyn_cast<ConstantInt>(
+ CI->getArgOperand(SizeArgOp)))
+ return SizeCI->getZExtValue() >= Arg->getZExtValue();
+ }
+ return false;
+ }
+public:
+ InstCombineFortifiedLibCalls(InstCombiner *IC) : IC(IC), NewInstruction(0) { }
+ Instruction *NewInstruction;
+};
+} // end anonymous namespace
+
+// Try to fold some different type of calls here.
+// 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 TargetData *TD) {
+ if (CI->getCalledFunction() == 0) return 0;
+
+ InstCombineFortifiedLibCalls Simplifier(this);
+ Simplifier.fold(CI, TD);
+ return Simplifier.NewInstruction;
+}
+
// visitCallSite - Improvements for call and invoke instructions.
//
Instruction *InstCombiner::visitCallSite(CallSite CS) {
bool Changed = false;
- // If the callee is a constexpr cast of a function, attempt to move the cast
- // to the arguments of the call/invoke.
- if (transformConstExprCastCall(CS)) return 0;
-
+ // If the callee is a pointer to a function, attempt to move any casts to the
+ // arguments of the call/invoke.
Value *Callee = CS.getCalledValue();
+ if (!isa<Function>(Callee) && transformConstExprCastCall(CS))
+ return 0;
if (Function *CalleeF = dyn_cast<Function>(Callee))
- if (CalleeF->getCallingConv() != CS.getCallingConv()) {
+ // If the call and callee calling conventions don't match, this call must
+ // be unreachable, as the call is undefined.
+ if (CalleeF->getCallingConv() != CS.getCallingConv() &&
+ // Only do this for calls to a function with a body. A prototype may
+ // not actually end up matching the implementation's calling conv for a
+ // variety of reasons (e.g. it may be written in assembly).
+ !CalleeF->isDeclaration()) {
Instruction *OldCall = CS.getInstruction();
- // If the call and callee calling conventions don't match, this call must
- // be unreachable, as the call is undefined.
new StoreInst(ConstantInt::getTrue(Callee->getContext()),
UndefValue::get(Type::getInt1PtrTy(Callee->getContext())),
OldCall);
// This allows ValueHandlers and custom metadata to adjust itself.
if (!OldCall->getType()->isVoidTy())
OldCall->replaceAllUsesWith(UndefValue::get(OldCall->getType()));
- if (isa<CallInst>(OldCall)) // Not worth removing an invoke here.
+ if (isa<CallInst>(OldCall))
return EraseInstFromFunction(*OldCall);
+
+ // We cannot remove an invoke, because it would change the CFG, just
+ // change the callee to a null pointer.
+ cast<InvokeInst>(OldCall)->setCalledFunction(
+ Constant::getNullValue(CalleeF->getType()));
return 0;
}
UndefValue::get(Type::getInt1PtrTy(Callee->getContext())),
CS.getInstruction());
- // If CS dues not return void then replaceAllUsesWith undef.
+ // If CS does not return void then replaceAllUsesWith undef.
// This allows ValueHandlers and custom metadata to adjust itself.
if (!CS.getInstruction()->getType()->isVoidTy())
CS.getInstruction()->
Changed = true;
}
+ // Try to optimize the call if possible, we require TargetData for most of
+ // 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);
+ // If we changed something return the result, etc. Otherwise let
+ // the fallthrough check.
+ if (I) return EraseInstFromFunction(*I);
+ }
+
return Changed ? CS.getInstruction() : 0;
}
// attempt to move the cast to the arguments of the call/invoke.
//
bool InstCombiner::transformConstExprCastCall(CallSite CS) {
- if (!isa<ConstantExpr>(CS.getCalledValue())) return false;
- ConstantExpr *CE = cast<ConstantExpr>(CS.getCalledValue());
- if (CE->getOpcode() != Instruction::BitCast ||
- !isa<Function>(CE->getOperand(0)))
+ Function *Callee =
+ dyn_cast<Function>(CS.getCalledValue()->stripPointerCasts());
+ if (Callee == 0)
return false;
- Function *Callee = cast<Function>(CE->getOperand(0));
Instruction *Caller = CS.getInstruction();
const AttrListPtr &CallerPAL = CS.getAttributes();
const Type *OldRetTy = Caller->getType();
const Type *NewRetTy = FT->getReturnType();
- if (isa<StructType>(NewRetTy))
+ if (NewRetTy->isStructTy())
return false; // TODO: Handle multiple return values.
// Check to see if we are changing the return type...
if (Callee->isDeclaration() &&
// Conversion is ok if changing from one pointer type to another or from
// a pointer to an integer of the same size.
- !((isa<PointerType>(OldRetTy) || !TD ||
+ !((OldRetTy->isPointerTy() || !TD ||
OldRetTy == TD->getIntPtrType(Caller->getContext())) &&
- (isa<PointerType>(NewRetTy) || !TD ||
+ (NewRetTy->isPointerTy() || !TD ||
NewRetTy == TD->getIntPtrType(Caller->getContext()))))
return false; // Cannot transform this return value.
if (!CastInst::isCastable(ActTy, ParamTy))
return false; // Cannot transform this parameter value.
- if (CallerPAL.getParamAttributes(i + 1)
- & Attribute::typeIncompatible(ParamTy))
+ unsigned Attrs = CallerPAL.getParamAttributes(i + 1);
+ if (Attrs & Attribute::typeIncompatible(ParamTy))
return false; // Attribute not compatible with transformed value.
+
+ // If the parameter is passed as a byval argument, then we have to have a
+ // sized type and the sized type has to have the same size as the old type.
+ if (ParamTy != ActTy && (Attrs & Attribute::ByVal)) {
+ const PointerType *ParamPTy = dyn_cast<PointerType>(ParamTy);
+ if (ParamPTy == 0 || !ParamPTy->getElementType()->isSized() || TD == 0)
+ return false;
+
+ const Type *CurElTy = cast<PointerType>(ActTy)->getElementType();
+ if (TD->getTypeAllocSize(CurElTy) !=
+ TD->getTypeAllocSize(ParamPTy->getElementType()))
+ return false;
+ }
// Converting from one pointer type to another or between a pointer and an
// integer of the same size is safe even if we do not have a body.
bool isConvertible = ActTy == ParamTy ||
- (TD && ((isa<PointerType>(ParamTy) ||
+ (TD && ((ParamTy->isPointerTy() ||
ParamTy == TD->getIntPtrType(Caller->getContext())) &&
- (isa<PointerType>(ActTy) ||
+ (ActTy->isPointerTy() ||
ActTy == TD->getIntPtrType(Caller->getContext()))));
if (Callee->isDeclaration() && !isConvertible) return false;
}
Value *NV = NC;
if (OldRetTy != NV->getType() && !Caller->use_empty()) {
if (!NV->getType()->isVoidTy()) {
- Instruction::CastOps opcode = CastInst::getCastOpcode(NC, false,
- OldRetTy, false);
+ Instruction::CastOps opcode =
+ CastInst::getCastOpcode(NC, false, OldRetTy, false);
NV = NC = CastInst::Create(opcode, NC, OldRetTy, "tmp");
// If this is an invoke instruction, we should insert it after the first
BasicBlock::iterator I = II->getNormalDest()->getFirstNonPHI();
InsertNewInstBefore(NC, *I);
} else {
- // Otherwise, it's a call, just insert cast right after the call instr
+ // Otherwise, it's a call, just insert cast right after the call.
InsertNewInstBefore(NC, *Caller);
}
Worklist.AddUsersToWorkList(*Caller);
}
}
-
if (!Caller->use_empty())
Caller->replaceAllUsesWith(NV);
IntrinsicInst *Tramp =
cast<IntrinsicInst>(cast<BitCastInst>(Callee)->getOperand(0));
- Function *NestF = cast<Function>(Tramp->getOperand(2)->stripPointerCasts());
+ Function *NestF =cast<Function>(Tramp->getArgOperand(1)->stripPointerCasts());
const PointerType *NestFPTy = cast<PointerType>(NestF->getType());
const FunctionType *NestFTy = cast<FunctionType>(NestFPTy->getElementType());
do {
if (Idx == NestIdx) {
// Add the chain argument and attributes.
- Value *NestVal = Tramp->getOperand(3);
+ Value *NestVal = Tramp->getArgOperand(2);
if (NestVal->getType() != NestTy)
NestVal = new BitCastInst(NestVal, NestTy, "nest", Caller);
NewArgs.push_back(NestVal);