//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "instcombine"
#include "InstCombine.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/Loads.h"
#include "llvm/Transforms/Utils/Local.h"
using namespace llvm;
+#define DEBUG_TYPE "instcombine"
+
STATISTIC(NumDeadStore, "Number of dead stores eliminated");
STATISTIC(NumGlobalCopies, "Number of allocas copied from constant global");
static bool pointsToConstantGlobal(Value *V) {
if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V))
return GV->isConstant();
- if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
+
+ if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) {
if (CE->getOpcode() == Instruction::BitCast ||
+ CE->getOpcode() == Instruction::AddrSpaceCast ||
CE->getOpcode() == Instruction::GetElementPtr)
return pointsToConstantGlobal(CE->getOperand(0));
+ }
return false;
}
/// can optimize this.
static bool
isOnlyCopiedFromConstantGlobal(Value *V, MemTransferInst *&TheCopy,
- SmallVectorImpl<Instruction *> &ToDelete,
- bool IsOffset = false) {
+ SmallVectorImpl<Instruction *> &ToDelete) {
// We track lifetime intrinsics as we encounter them. If we decide to go
// ahead and replace the value with the global, this lets the caller quickly
// eliminate the markers.
- for (Use &U : V->uses()) {
- Instruction *I = cast<Instruction>(U.getUser());
-
- if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
- // Ignore non-volatile loads, they are always ok.
- if (!LI->isSimple()) return false;
- continue;
- }
-
- if (BitCastInst *BCI = dyn_cast<BitCastInst>(I)) {
- // If uses of the bitcast are ok, we are ok.
- if (!isOnlyCopiedFromConstantGlobal(BCI, TheCopy, ToDelete, IsOffset))
- return false;
- continue;
- }
- if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(I)) {
- // If the GEP has all zero indices, it doesn't offset the pointer. If it
- // doesn't, it does.
- if (!isOnlyCopiedFromConstantGlobal(
- GEP, TheCopy, ToDelete, IsOffset || !GEP->hasAllZeroIndices()))
- return false;
- continue;
- }
+ SmallVector<std::pair<Value *, bool>, 35> ValuesToInspect;
+ ValuesToInspect.push_back(std::make_pair(V, false));
+ while (!ValuesToInspect.empty()) {
+ auto ValuePair = ValuesToInspect.pop_back_val();
+ const bool IsOffset = ValuePair.second;
+ for (auto &U : ValuePair.first->uses()) {
+ Instruction *I = cast<Instruction>(U.getUser());
+
+ if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
+ // Ignore non-volatile loads, they are always ok.
+ if (!LI->isSimple()) return false;
+ continue;
+ }
- if (CallSite CS = I) {
- // If this is the function being called then we treat it like a load and
- // ignore it.
- if (CS.isCallee(&U))
+ if (isa<BitCastInst>(I) || isa<AddrSpaceCastInst>(I)) {
+ // If uses of the bitcast are ok, we are ok.
+ ValuesToInspect.push_back(std::make_pair(I, IsOffset));
continue;
+ }
+ if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(I)) {
+ // If the GEP has all zero indices, it doesn't offset the pointer. If it
+ // doesn't, it does.
+ ValuesToInspect.push_back(
+ std::make_pair(I, IsOffset || !GEP->hasAllZeroIndices()));
+ continue;
+ }
- // Inalloca arguments are clobbered by the call.
- unsigned ArgNo = CS.getArgumentNo(&U);
- if (CS.isInAllocaArgument(ArgNo))
- return false;
+ if (CallSite CS = I) {
+ // If this is the function being called then we treat it like a load and
+ // ignore it.
+ if (CS.isCallee(&U))
+ continue;
- // If this is a readonly/readnone call site, then we know it is just a
- // load (but one that potentially returns the value itself), so we can
- // ignore it if we know that the value isn't captured.
- if (CS.onlyReadsMemory() &&
- (CS.getInstruction()->use_empty() || CS.doesNotCapture(ArgNo)))
- continue;
+ // Inalloca arguments are clobbered by the call.
+ unsigned ArgNo = CS.getArgumentNo(&U);
+ if (CS.isInAllocaArgument(ArgNo))
+ return false;
- // If this is being passed as a byval argument, the caller is making a
- // copy, so it is only a read of the alloca.
- if (CS.isByValArgument(ArgNo))
- continue;
- }
+ // If this is a readonly/readnone call site, then we know it is just a
+ // load (but one that potentially returns the value itself), so we can
+ // ignore it if we know that the value isn't captured.
+ if (CS.onlyReadsMemory() &&
+ (CS.getInstruction()->use_empty() || CS.doesNotCapture(ArgNo)))
+ continue;
+
+ // If this is being passed as a byval argument, the caller is making a
+ // copy, so it is only a read of the alloca.
+ if (CS.isByValArgument(ArgNo))
+ continue;
+ }
- // Lifetime intrinsics can be handled by the caller.
- if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
- if (II->getIntrinsicID() == Intrinsic::lifetime_start ||
- II->getIntrinsicID() == Intrinsic::lifetime_end) {
- assert(II->use_empty() && "Lifetime markers have no result to use!");
- ToDelete.push_back(II);
- continue;
+ // Lifetime intrinsics can be handled by the caller.
+ if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
+ if (II->getIntrinsicID() == Intrinsic::lifetime_start ||
+ II->getIntrinsicID() == Intrinsic::lifetime_end) {
+ assert(II->use_empty() && "Lifetime markers have no result to use!");
+ ToDelete.push_back(II);
+ continue;
+ }
}
- }
- // If this is isn't our memcpy/memmove, reject it as something we can't
- // handle.
- MemTransferInst *MI = dyn_cast<MemTransferInst>(I);
- if (MI == 0)
- return false;
+ // If this is isn't our memcpy/memmove, reject it as something we can't
+ // handle.
+ MemTransferInst *MI = dyn_cast<MemTransferInst>(I);
+ if (!MI)
+ return false;
- // If the transfer is using the alloca as a source of the transfer, then
- // ignore it since it is a load (unless the transfer is volatile).
- if (U.getOperandNo() == 1) {
- if (MI->isVolatile()) return false;
- continue;
- }
+ // If the transfer is using the alloca as a source of the transfer, then
+ // ignore it since it is a load (unless the transfer is volatile).
+ if (U.getOperandNo() == 1) {
+ if (MI->isVolatile()) return false;
+ continue;
+ }
- // If we already have seen a copy, reject the second one.
- if (TheCopy) return false;
+ // If we already have seen a copy, reject the second one.
+ if (TheCopy) return false;
- // If the pointer has been offset from the start of the alloca, we can't
- // safely handle this.
- if (IsOffset) return false;
+ // If the pointer has been offset from the start of the alloca, we can't
+ // safely handle this.
+ if (IsOffset) return false;
- // If the memintrinsic isn't using the alloca as the dest, reject it.
- if (U.getOperandNo() != 0) return false;
+ // If the memintrinsic isn't using the alloca as the dest, reject it.
+ if (U.getOperandNo() != 0) return false;
- // If the source of the memcpy/move is not a constant global, reject it.
- if (!pointsToConstantGlobal(MI->getSource()))
- return false;
+ // If the source of the memcpy/move is not a constant global, reject it.
+ if (!pointsToConstantGlobal(MI->getSource()))
+ return false;
- // Otherwise, the transform is safe. Remember the copy instruction.
- TheCopy = MI;
+ // Otherwise, the transform is safe. Remember the copy instruction.
+ TheCopy = MI;
+ }
}
return true;
}
static MemTransferInst *
isOnlyCopiedFromConstantGlobal(AllocaInst *AI,
SmallVectorImpl<Instruction *> &ToDelete) {
- MemTransferInst *TheCopy = 0;
+ MemTransferInst *TheCopy = nullptr;
if (isOnlyCopiedFromConstantGlobal(AI, TheCopy, ToDelete))
return TheCopy;
- return 0;
+ return nullptr;
}
Instruction *InstCombiner::visitAllocaInst(AllocaInst &AI) {
if (const ConstantInt *C = dyn_cast<ConstantInt>(AI.getArraySize())) {
Type *NewTy =
ArrayType::get(AI.getAllocatedType(), C->getZExtValue());
- AllocaInst *New = Builder->CreateAlloca(NewTy, 0, AI.getName());
+ AllocaInst *New = Builder->CreateAlloca(NewTy, nullptr, AI.getName());
New->setAlignment(AI.getAlignment());
// Scan to the end of the allocation instructions, to skip over a block of
SmallVector<Instruction *, 4> ToDelete;
if (MemTransferInst *Copy = isOnlyCopiedFromConstantGlobal(&AI, ToDelete)) {
unsigned SourceAlign = getOrEnforceKnownAlignment(Copy->getSource(),
- AI.getAlignment(), DL);
+ AI.getAlignment(),
+ DL, AT, &AI, DT);
if (AI.getAlignment() <= SourceAlign) {
DEBUG(dbgs() << "Found alloca equal to global: " << AI << '\n');
DEBUG(dbgs() << " memcpy = " << *Copy << '\n');
// If the address spaces don't match, don't eliminate the cast.
if (DestTy->getAddressSpace() != SrcTy->getAddressSpace())
- return 0;
+ return nullptr;
Type *SrcPTy = SrcTy->getElementType();
}
}
}
- return 0;
+ return nullptr;
}
Instruction *InstCombiner::visitLoadInst(LoadInst &LI) {
// Attempt to improve the alignment.
if (DL) {
unsigned KnownAlign =
- getOrEnforceKnownAlignment(Op, DL->getPrefTypeAlignment(LI.getType()),DL);
+ getOrEnforceKnownAlignment(Op, DL->getPrefTypeAlignment(LI.getType()),
+ DL, AT, &LI, DT);
unsigned LoadAlign = LI.getAlignment();
unsigned EffectiveLoadAlign = LoadAlign != 0 ? LoadAlign :
DL->getABITypeAlignment(LI.getType());
// None of the following transforms are legal for volatile/atomic loads.
// FIXME: Some of it is okay for atomic loads; needs refactoring.
- if (!LI.isSimple()) return 0;
+ if (!LI.isSimple()) return nullptr;
// Do really simple store-to-load forwarding and load CSE, to catch cases
// where there are several consecutive memory accesses to the same location,
}
}
}
- return 0;
+ return nullptr;
}
/// InstCombineStoreToCast - Fold store V, (cast P) -> store (cast V), P
User *CI = cast<User>(SI.getOperand(1));
Value *CastOp = CI->getOperand(0);
- Type *DestPTy = cast<PointerType>(CI->getType())->getElementType();
+ Type *DestPTy = CI->getType()->getPointerElementType();
PointerType *SrcTy = dyn_cast<PointerType>(CastOp->getType());
- if (SrcTy == 0) return 0;
+ if (!SrcTy) return nullptr;
Type *SrcPTy = SrcTy->getElementType();
if (!DestPTy->isIntegerTy() && !DestPTy->isPointerTy())
- return 0;
+ return nullptr;
/// NewGEPIndices - If SrcPTy is an aggregate type, we can emit a "noop gep"
/// to its first element. This allows us to handle things like:
}
if (!SrcPTy->isIntegerTy() && !SrcPTy->isPointerTy())
- return 0;
+ return nullptr;
// If the pointers point into different address spaces don't do the
// transformation.
- if (SrcTy->getAddressSpace() !=
- cast<PointerType>(CI->getType())->getAddressSpace())
- return 0;
+ if (SrcTy->getAddressSpace() != CI->getType()->getPointerAddressSpace())
+ return nullptr;
// If the pointers point to values of different sizes don't do the
// transformation.
if (!IC.getDataLayout() ||
IC.getDataLayout()->getTypeSizeInBits(SrcPTy) !=
IC.getDataLayout()->getTypeSizeInBits(DestPTy))
- return 0;
+ return nullptr;
// If the pointers point to pointers to different address spaces don't do the
// transformation. It is not safe to introduce an addrspacecast instruction in
// cast.
if (SrcPTy->isPointerTy() && DestPTy->isPointerTy() &&
SrcPTy->getPointerAddressSpace() != DestPTy->getPointerAddressSpace())
- return 0;
+ return nullptr;
// Okay, we are casting from one integer or pointer type to another of
// the same size. Instead of casting the pointer before
if (DL) {
unsigned KnownAlign =
getOrEnforceKnownAlignment(Ptr, DL->getPrefTypeAlignment(Val->getType()),
- DL);
+ DL, AT, &SI, DT);
unsigned StoreAlign = SI.getAlignment();
unsigned EffectiveStoreAlign = StoreAlign != 0 ? StoreAlign :
DL->getABITypeAlignment(Val->getType());
// Don't hack volatile/atomic stores.
// FIXME: Some bits are legal for atomic stores; needs refactoring.
- if (!SI.isSimple()) return 0;
+ if (!SI.isSimple()) return nullptr;
// If the RHS is an alloca with a single use, zapify the store, making the
// alloca dead.
if (Instruction *U = dyn_cast<Instruction>(Val))
Worklist.Add(U); // Dropped a use.
}
- return 0; // Do not modify these!
+ return nullptr; // Do not modify these!
}
// store undef, Ptr -> noop
if (BranchInst *BI = dyn_cast<BranchInst>(BBI))
if (BI->isUnconditional())
if (SimplifyStoreAtEndOfBlock(SI))
- return 0; // xform done!
+ return nullptr; // xform done!
- return 0;
+ return nullptr;
}
/// SimplifyStoreAtEndOfBlock - Turn things like:
// the other predecessor.
pred_iterator PI = pred_begin(DestBB);
BasicBlock *P = *PI;
- BasicBlock *OtherBB = 0;
+ BasicBlock *OtherBB = nullptr;
if (P != StoreBB)
OtherBB = P;
// If the other block ends in an unconditional branch, check for the 'if then
// else' case. there is an instruction before the branch.
- StoreInst *OtherStore = 0;
+ StoreInst *OtherStore = nullptr;
if (OtherBr->isUnconditional()) {
--BBI;
// Skip over debugging info.
InsertNewInstBefore(NewSI, *BBI);
NewSI->setDebugLoc(OtherStore->getDebugLoc());
- // If the two stores had the same TBAA tag, preserve it.
- if (MDNode *TBAATag = SI.getMetadata(LLVMContext::MD_tbaa))
- if ((TBAATag = MDNode::getMostGenericTBAA(TBAATag,
- OtherStore->getMetadata(LLVMContext::MD_tbaa))))
- NewSI->setMetadata(LLVMContext::MD_tbaa, TBAATag);
-
+ // If the two stores had AA tags, merge them.
+ AAMDNodes AATags;
+ SI.getAAMetadata(AATags);
+ if (AATags) {
+ OtherStore->getAAMetadata(AATags, /* Merge = */ true);
+ NewSI->setAAMetadata(AATags);
+ }
// Nuke the old stores.
EraseInstFromFunction(SI);
projects / oota-llvm.git / blobdiff