#include "InstCombine.h"
#include "llvm/IntrinsicInst.h"
+#include "llvm/Analysis/Loads.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Transforms/Utils/Local.h"
STATISTIC(NumDeadStore, "Number of dead stores eliminated");
Instruction *InstCombiner::visitAllocaInst(AllocaInst &AI) {
+ // Ensure that the alloca array size argument has type intptr_t, so that
+ // any casting is exposed early.
+ if (TD) {
+ const Type *IntPtrTy = TD->getIntPtrType(AI.getContext());
+ if (AI.getArraySize()->getType() != IntPtrTy) {
+ Value *V = Builder->CreateIntCast(AI.getArraySize(),
+ IntPtrTy, false);
+ AI.setOperand(0, V);
+ return &AI;
+ }
+ }
+
// Convert: alloca Ty, C - where C is a constant != 1 into: alloca [C x Ty], 1
if (AI.isArrayAllocation()) { // Check C != 1
if (const ConstantInt *C = dyn_cast<ConstantInt>(AI.getArraySize())) {
Value *Idx[2];
Idx[0] = NullIdx;
Idx[1] = NullIdx;
- Value *V = GetElementPtrInst::CreateInBounds(New, Idx, Idx + 2,
- New->getName()+".sub", It);
+ Instruction *GEP =
+ GetElementPtrInst::CreateInBounds(New, Idx, Idx + 2,
+ New->getName()+".sub");
+ InsertNewInstBefore(GEP, *It);
// Now make everything use the getelementptr instead of the original
// allocation.
- return ReplaceInstUsesWith(AI, V);
+ return ReplaceInstUsesWith(AI, GEP);
} else if (isa<UndefValue>(AI.getArraySize())) {
return ReplaceInstUsesWith(AI, Constant::getNullValue(AI.getType()));
}
const Type *SrcPTy = SrcTy->getElementType();
- if (DestPTy->isInteger() || isa<PointerType>(DestPTy) ||
- isa<VectorType>(DestPTy)) {
+ if (DestPTy->isIntegerTy() || DestPTy->isPointerTy() ||
+ DestPTy->isVectorTy()) {
// If the source is an array, the code below will not succeed. Check to
// see if a trivial 'gep P, 0, 0' will help matters. Only do this for
// constants.
}
if (IC.getTargetData() &&
- (SrcPTy->isInteger() || isa<PointerType>(SrcPTy) ||
- isa<VectorType>(SrcPTy)) &&
+ (SrcPTy->isIntegerTy() || SrcPTy->isPointerTy() ||
+ SrcPTy->isVectorTy()) &&
// Do not allow turning this into a load of an integer, which is then
// casted to a pointer, this pessimizes pointer analysis a lot.
- (isa<PointerType>(SrcPTy) == isa<PointerType>(LI.getType())) &&
+ (SrcPTy->isPointerTy() == LI.getType()->isPointerTy()) &&
IC.getTargetData()->getTypeSizeInBits(SrcPTy) ==
IC.getTargetData()->getTypeSizeInBits(DestPTy)) {
// Okay, we are casting from one integer or pointer type to another of
// the same size. Instead of casting the pointer before the load, cast
// the result of the loaded value.
- Value *NewLoad =
+ LoadInst *NewLoad =
IC.Builder->CreateLoad(CastOp, LI.isVolatile(), CI->getName());
+ NewLoad->setAlignment(LI.getAlignment());
// Now cast the result of the load.
return new BitCastInst(NewLoad, LI.getType());
}
// Attempt to improve the alignment.
if (TD) {
unsigned KnownAlign =
- GetOrEnforceKnownAlignment(Op, TD->getPrefTypeAlignment(LI.getType()));
- if (KnownAlign >
- (LI.getAlignment() == 0 ? TD->getABITypeAlignment(LI.getType()) :
- LI.getAlignment()))
+ getOrEnforceKnownAlignment(Op, TD->getPrefTypeAlignment(LI.getType()),TD);
+ unsigned LoadAlign = LI.getAlignment();
+ unsigned EffectiveLoadAlign = LoadAlign != 0 ? LoadAlign :
+ TD->getABITypeAlignment(LI.getType());
+
+ if (KnownAlign > EffectiveLoadAlign)
LI.setAlignment(KnownAlign);
+ else if (LoadAlign == 0)
+ LI.setAlignment(EffectiveLoadAlign);
}
// load (cast X) --> cast (load X) iff safe.
if (LI.isVolatile()) return 0;
// Do really simple store-to-load forwarding and load CSE, to catch cases
- // where there are several consequtive memory accesses to the same location,
+ // where there are several consecutive memory accesses to the same location,
// separated by a few arithmetic operations.
BasicBlock::iterator BBI = &LI;
if (Value *AvailableVal = FindAvailableLoadedValue(Op, LI.getParent(), BBI,6))
//
if (SelectInst *SI = dyn_cast<SelectInst>(Op)) {
// load (select (Cond, &V1, &V2)) --> select(Cond, load &V1, load &V2).
- if (isSafeToLoadUnconditionally(SI->getOperand(1), SI) &&
- isSafeToLoadUnconditionally(SI->getOperand(2), SI)) {
- Value *V1 = Builder->CreateLoad(SI->getOperand(1),
- SI->getOperand(1)->getName()+".val");
- Value *V2 = Builder->CreateLoad(SI->getOperand(2),
- SI->getOperand(2)->getName()+".val");
+ unsigned Align = LI.getAlignment();
+ if (isSafeToLoadUnconditionally(SI->getOperand(1), SI, Align, TD) &&
+ isSafeToLoadUnconditionally(SI->getOperand(2), SI, Align, TD)) {
+ LoadInst *V1 = Builder->CreateLoad(SI->getOperand(1),
+ SI->getOperand(1)->getName()+".val");
+ LoadInst *V2 = Builder->CreateLoad(SI->getOperand(2),
+ SI->getOperand(2)->getName()+".val");
+ V1->setAlignment(Align);
+ V2->setAlignment(Align);
return SelectInst::Create(SI->getCondition(), V1, V2);
}
const Type *SrcPTy = SrcTy->getElementType();
- if (!DestPTy->isInteger() && !isa<PointerType>(DestPTy))
+ if (!DestPTy->isIntegerTy() && !DestPTy->isPointerTy())
return 0;
/// NewGEPIndices - If SrcPTy is an aggregate type, we can emit a "noop gep"
// If the source is an array, the code below will not succeed. Check to
// see if a trivial 'gep P, 0, 0' will help matters. Only do this for
// constants.
- if (isa<ArrayType>(SrcPTy) || isa<StructType>(SrcPTy)) {
+ if (SrcPTy->isArrayTy() || SrcPTy->isStructTy()) {
// Index through pointer.
Constant *Zero = Constant::getNullValue(Type::getInt32Ty(SI.getContext()));
NewGEPIndices.push_back(Zero);
SrcTy = PointerType::get(SrcPTy, SrcTy->getAddressSpace());
}
- if (!SrcPTy->isInteger() && !isa<PointerType>(SrcPTy))
+ if (!SrcPTy->isIntegerTy() && !SrcPTy->isPointerTy())
return 0;
// If the pointers point into different address spaces or if they point to
Instruction::CastOps opcode = Instruction::BitCast;
const Type* CastSrcTy = SIOp0->getType();
const Type* CastDstTy = SrcPTy;
- if (isa<PointerType>(CastDstTy)) {
- if (CastSrcTy->isInteger())
+ if (CastDstTy->isPointerTy()) {
+ if (CastSrcTy->isIntegerTy())
opcode = Instruction::IntToPtr;
- } else if (isa<IntegerType>(CastDstTy)) {
- if (isa<PointerType>(SIOp0->getType()))
+ } else if (CastDstTy->isIntegerTy()) {
+ if (SIOp0->getType()->isPointerTy())
opcode = Instruction::PtrToInt;
}
NewCast = IC.Builder->CreateCast(opcode, SIOp0, CastDstTy,
SIOp0->getName()+".c");
- return new StoreInst(NewCast, CastOp);
+ SI.setOperand(0, NewCast);
+ SI.setOperand(1, CastOp);
+ return &SI;
}
/// equivalentAddressValues - Test if A and B will obviously have the same
return false;
}
-// If this instruction has two uses, one of which is a llvm.dbg.declare,
-// return the llvm.dbg.declare.
-DbgDeclareInst *InstCombiner::hasOneUsePlusDeclare(Value *V) {
- if (!V->hasNUses(2))
- return 0;
- for (Value::use_iterator UI = V->use_begin(), E = V->use_end();
- UI != E; ++UI) {
- if (DbgDeclareInst *DI = dyn_cast<DbgDeclareInst>(UI))
- return DI;
- if (isa<BitCastInst>(UI) && UI->hasOneUse()) {
- if (DbgDeclareInst *DI = dyn_cast<DbgDeclareInst>(UI->use_begin()))
- return DI;
- }
- }
- return 0;
-}
-
Instruction *InstCombiner::visitStoreInst(StoreInst &SI) {
Value *Val = SI.getOperand(0);
Value *Ptr = SI.getOperand(1);
// If the RHS is an alloca with a single use, zapify the store, making the
// alloca dead.
- // If the RHS is an alloca with a two uses, the other one being a
- // llvm.dbg.declare, zapify the store and the declare, making the
- // alloca dead. We must do this to prevent declare's from affecting
- // codegen.
if (!SI.isVolatile()) {
if (Ptr->hasOneUse()) {
if (isa<AllocaInst>(Ptr))
if (isa<AllocaInst>(GEP->getOperand(0))) {
if (GEP->getOperand(0)->hasOneUse())
return EraseInstFromFunction(SI);
- if (DbgDeclareInst *DI = hasOneUsePlusDeclare(GEP->getOperand(0))) {
- EraseInstFromFunction(*DI);
- return EraseInstFromFunction(SI);
- }
}
}
}
- if (DbgDeclareInst *DI = hasOneUsePlusDeclare(Ptr)) {
- EraseInstFromFunction(*DI);
- return EraseInstFromFunction(SI);
- }
}
// Attempt to improve the alignment.
if (TD) {
unsigned KnownAlign =
- GetOrEnforceKnownAlignment(Ptr, TD->getPrefTypeAlignment(Val->getType()));
- if (KnownAlign >
- (SI.getAlignment() == 0 ? TD->getABITypeAlignment(Val->getType()) :
- SI.getAlignment()))
+ getOrEnforceKnownAlignment(Ptr, TD->getPrefTypeAlignment(Val->getType()),
+ TD);
+ unsigned StoreAlign = SI.getAlignment();
+ unsigned EffectiveStoreAlign = StoreAlign != 0 ? StoreAlign :
+ TD->getABITypeAlignment(Val->getType());
+
+ if (KnownAlign > EffectiveStoreAlign)
SI.setAlignment(KnownAlign);
+ else if (StoreAlign == 0)
+ SI.setAlignment(EffectiveStoreAlign);
}
// Do really simple DSE, to catch cases where there are several consecutive
--BBI;
// Don't count debug info directives, lest they affect codegen,
// and we skip pointer-to-pointer bitcasts, which are NOPs.
- // It is necessary for correctness to skip those that feed into a
- // llvm.dbg.declare, as these are not present when debugging is off.
if (isa<DbgInfoIntrinsic>(BBI) ||
- (isa<BitCastInst>(BBI) && isa<PointerType>(BBI->getType()))) {
+ (isa<BitCastInst>(BBI) && BBI->getType()->isPointerTy())) {
ScanInsts++;
continue;
}
// If this store is the last instruction in the basic block (possibly
- // excepting debug info instructions and the pointer bitcasts that feed
- // into them), and if the block ends with an unconditional branch, try
- // to move it to the successor block.
+ // excepting debug info instructions), and if the block ends with an
+ // unconditional branch, try to move it to the successor block.
BBI = &SI;
do {
++BBI;
} while (isa<DbgInfoIntrinsic>(BBI) ||
- (isa<BitCastInst>(BBI) && isa<PointerType>(BBI->getType())));
+ (isa<BitCastInst>(BBI) && BBI->getType()->isPointerTy()));
if (BranchInst *BI = dyn_cast<BranchInst>(BBI))
if (BI->isUnconditional())
if (SimplifyStoreAtEndOfBlock(SI))
// Determine whether Dest has exactly two predecessors and, if so, compute
// the other predecessor.
pred_iterator PI = pred_begin(DestBB);
+ BasicBlock *P = *PI;
BasicBlock *OtherBB = 0;
- if (*PI != StoreBB)
- OtherBB = *PI;
- ++PI;
- if (PI == pred_end(DestBB))
+
+ if (P != StoreBB)
+ OtherBB = P;
+
+ if (++PI == pred_end(DestBB))
return false;
- if (*PI != StoreBB) {
+ P = *PI;
+ if (P != StoreBB) {
if (OtherBB)
return false;
- OtherBB = *PI;
+ OtherBB = P;
}
if (++PI != pred_end(DestBB))
return false;
--BBI;
// Skip over debugging info.
while (isa<DbgInfoIntrinsic>(BBI) ||
- (isa<BitCastInst>(BBI) && isa<PointerType>(BBI->getType()))) {
+ (isa<BitCastInst>(BBI) && BBI->getType()->isPointerTy())) {
if (BBI==OtherBB->begin())
return false;
--BBI;
// Insert a PHI node now if we need it.
Value *MergedVal = OtherStore->getOperand(0);
if (MergedVal != SI.getOperand(0)) {
- PHINode *PN = PHINode::Create(MergedVal->getType(), "storemerge");
- PN->reserveOperandSpace(2);
+ PHINode *PN = PHINode::Create(MergedVal->getType(), 2, "storemerge");
PN->addIncoming(SI.getOperand(0), SI.getParent());
PN->addIncoming(OtherStore->getOperand(0), OtherBB);
MergedVal = InsertNewInstBefore(PN, DestBB->front());
// Advance to a place where it is safe to insert the new store and
// insert it.
BBI = DestBB->getFirstNonPHI();
- InsertNewInstBefore(new StoreInst(MergedVal, SI.getOperand(1),
- OtherStore->isVolatile(),
- SI.getAlignment()), *BBI);
-
+ StoreInst *NewSI = new StoreInst(MergedVal, SI.getOperand(1),
+ OtherStore->isVolatile(),
+ SI.getAlignment());
+ InsertNewInstBefore(NewSI, *BBI);
+ NewSI->setDebugLoc(OtherStore->getDebugLoc());
+
// Nuke the old stores.
EraseInstFromFunction(SI);
EraseInstFromFunction(*OtherStore);
projects / oota-llvm.git / blobdiff