1 //===- InstCombineLoadStoreAlloca.cpp -------------------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file implements the visit functions for load, store and alloca.
12 //===----------------------------------------------------------------------===//
14 #include "InstCombine.h"
15 #include "llvm/ADT/Statistic.h"
16 #include "llvm/Analysis/Loads.h"
17 #include "llvm/IR/DataLayout.h"
18 #include "llvm/IR/IntrinsicInst.h"
19 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
20 #include "llvm/Transforms/Utils/Local.h"
23 #define DEBUG_TYPE "instcombine"
25 STATISTIC(NumDeadStore, "Number of dead stores eliminated");
26 STATISTIC(NumGlobalCopies, "Number of allocas copied from constant global");
28 /// pointsToConstantGlobal - Return true if V (possibly indirectly) points to
29 /// some part of a constant global variable. This intentionally only accepts
30 /// constant expressions because we can't rewrite arbitrary instructions.
31 static bool pointsToConstantGlobal(Value *V) {
32 if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V))
33 return GV->isConstant();
35 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) {
36 if (CE->getOpcode() == Instruction::BitCast ||
37 CE->getOpcode() == Instruction::AddrSpaceCast ||
38 CE->getOpcode() == Instruction::GetElementPtr)
39 return pointsToConstantGlobal(CE->getOperand(0));
44 /// isOnlyCopiedFromConstantGlobal - Recursively walk the uses of a (derived)
45 /// pointer to an alloca. Ignore any reads of the pointer, return false if we
46 /// see any stores or other unknown uses. If we see pointer arithmetic, keep
47 /// track of whether it moves the pointer (with IsOffset) but otherwise traverse
48 /// the uses. If we see a memcpy/memmove that targets an unoffseted pointer to
49 /// the alloca, and if the source pointer is a pointer to a constant global, we
50 /// can optimize this.
52 isOnlyCopiedFromConstantGlobal(Value *V, MemTransferInst *&TheCopy,
53 SmallVectorImpl<Instruction *> &ToDelete,
54 bool IsOffset = false) {
55 // We track lifetime intrinsics as we encounter them. If we decide to go
56 // ahead and replace the value with the global, this lets the caller quickly
57 // eliminate the markers.
59 for (Use &U : V->uses()) {
60 Instruction *I = cast<Instruction>(U.getUser());
62 if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
63 // Ignore non-volatile loads, they are always ok.
64 if (!LI->isSimple()) return false;
68 if (isa<BitCastInst>(I) || isa<AddrSpaceCastInst>(I)) {
69 // If uses of the bitcast are ok, we are ok.
70 if (!isOnlyCopiedFromConstantGlobal(I, TheCopy, ToDelete, IsOffset))
74 if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(I)) {
75 // If the GEP has all zero indices, it doesn't offset the pointer. If it
77 if (!isOnlyCopiedFromConstantGlobal(
78 GEP, TheCopy, ToDelete, IsOffset || !GEP->hasAllZeroIndices()))
83 if (CallSite CS = I) {
84 // If this is the function being called then we treat it like a load and
89 // Inalloca arguments are clobbered by the call.
90 unsigned ArgNo = CS.getArgumentNo(&U);
91 if (CS.isInAllocaArgument(ArgNo))
94 // If this is a readonly/readnone call site, then we know it is just a
95 // load (but one that potentially returns the value itself), so we can
96 // ignore it if we know that the value isn't captured.
97 if (CS.onlyReadsMemory() &&
98 (CS.getInstruction()->use_empty() || CS.doesNotCapture(ArgNo)))
101 // If this is being passed as a byval argument, the caller is making a
102 // copy, so it is only a read of the alloca.
103 if (CS.isByValArgument(ArgNo))
107 // Lifetime intrinsics can be handled by the caller.
108 if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
109 if (II->getIntrinsicID() == Intrinsic::lifetime_start ||
110 II->getIntrinsicID() == Intrinsic::lifetime_end) {
111 assert(II->use_empty() && "Lifetime markers have no result to use!");
112 ToDelete.push_back(II);
117 // If this is isn't our memcpy/memmove, reject it as something we can't
119 MemTransferInst *MI = dyn_cast<MemTransferInst>(I);
123 // If the transfer is using the alloca as a source of the transfer, then
124 // ignore it since it is a load (unless the transfer is volatile).
125 if (U.getOperandNo() == 1) {
126 if (MI->isVolatile()) return false;
130 // If we already have seen a copy, reject the second one.
131 if (TheCopy) return false;
133 // If the pointer has been offset from the start of the alloca, we can't
134 // safely handle this.
135 if (IsOffset) return false;
137 // If the memintrinsic isn't using the alloca as the dest, reject it.
138 if (U.getOperandNo() != 0) return false;
140 // If the source of the memcpy/move is not a constant global, reject it.
141 if (!pointsToConstantGlobal(MI->getSource()))
144 // Otherwise, the transform is safe. Remember the copy instruction.
150 /// isOnlyCopiedFromConstantGlobal - Return true if the specified alloca is only
151 /// modified by a copy from a constant global. If we can prove this, we can
152 /// replace any uses of the alloca with uses of the global directly.
153 static MemTransferInst *
154 isOnlyCopiedFromConstantGlobal(AllocaInst *AI,
155 SmallVectorImpl<Instruction *> &ToDelete) {
156 MemTransferInst *TheCopy = nullptr;
157 if (isOnlyCopiedFromConstantGlobal(AI, TheCopy, ToDelete))
162 Instruction *InstCombiner::visitAllocaInst(AllocaInst &AI) {
163 // Ensure that the alloca array size argument has type intptr_t, so that
164 // any casting is exposed early.
166 Type *IntPtrTy = DL->getIntPtrType(AI.getType());
167 if (AI.getArraySize()->getType() != IntPtrTy) {
168 Value *V = Builder->CreateIntCast(AI.getArraySize(),
175 // Convert: alloca Ty, C - where C is a constant != 1 into: alloca [C x Ty], 1
176 if (AI.isArrayAllocation()) { // Check C != 1
177 if (const ConstantInt *C = dyn_cast<ConstantInt>(AI.getArraySize())) {
179 ArrayType::get(AI.getAllocatedType(), C->getZExtValue());
180 AllocaInst *New = Builder->CreateAlloca(NewTy, nullptr, AI.getName());
181 New->setAlignment(AI.getAlignment());
183 // Scan to the end of the allocation instructions, to skip over a block of
184 // allocas if possible...also skip interleaved debug info
186 BasicBlock::iterator It = New;
187 while (isa<AllocaInst>(*It) || isa<DbgInfoIntrinsic>(*It)) ++It;
189 // Now that I is pointing to the first non-allocation-inst in the block,
190 // insert our getelementptr instruction...
193 ? DL->getIntPtrType(AI.getType())
194 : Type::getInt64Ty(AI.getContext());
195 Value *NullIdx = Constant::getNullValue(IdxTy);
196 Value *Idx[2] = { NullIdx, NullIdx };
198 GetElementPtrInst::CreateInBounds(New, Idx, New->getName() + ".sub");
199 InsertNewInstBefore(GEP, *It);
201 // Now make everything use the getelementptr instead of the original
203 return ReplaceInstUsesWith(AI, GEP);
204 } else if (isa<UndefValue>(AI.getArraySize())) {
205 return ReplaceInstUsesWith(AI, Constant::getNullValue(AI.getType()));
209 if (DL && AI.getAllocatedType()->isSized()) {
210 // If the alignment is 0 (unspecified), assign it the preferred alignment.
211 if (AI.getAlignment() == 0)
212 AI.setAlignment(DL->getPrefTypeAlignment(AI.getAllocatedType()));
214 // Move all alloca's of zero byte objects to the entry block and merge them
215 // together. Note that we only do this for alloca's, because malloc should
216 // allocate and return a unique pointer, even for a zero byte allocation.
217 if (DL->getTypeAllocSize(AI.getAllocatedType()) == 0) {
218 // For a zero sized alloca there is no point in doing an array allocation.
219 // This is helpful if the array size is a complicated expression not used
221 if (AI.isArrayAllocation()) {
222 AI.setOperand(0, ConstantInt::get(AI.getArraySize()->getType(), 1));
226 // Get the first instruction in the entry block.
227 BasicBlock &EntryBlock = AI.getParent()->getParent()->getEntryBlock();
228 Instruction *FirstInst = EntryBlock.getFirstNonPHIOrDbg();
229 if (FirstInst != &AI) {
230 // If the entry block doesn't start with a zero-size alloca then move
231 // this one to the start of the entry block. There is no problem with
232 // dominance as the array size was forced to a constant earlier already.
233 AllocaInst *EntryAI = dyn_cast<AllocaInst>(FirstInst);
234 if (!EntryAI || !EntryAI->getAllocatedType()->isSized() ||
235 DL->getTypeAllocSize(EntryAI->getAllocatedType()) != 0) {
236 AI.moveBefore(FirstInst);
240 // If the alignment of the entry block alloca is 0 (unspecified),
241 // assign it the preferred alignment.
242 if (EntryAI->getAlignment() == 0)
243 EntryAI->setAlignment(
244 DL->getPrefTypeAlignment(EntryAI->getAllocatedType()));
245 // Replace this zero-sized alloca with the one at the start of the entry
246 // block after ensuring that the address will be aligned enough for both
248 unsigned MaxAlign = std::max(EntryAI->getAlignment(),
250 EntryAI->setAlignment(MaxAlign);
251 if (AI.getType() != EntryAI->getType())
252 return new BitCastInst(EntryAI, AI.getType());
253 return ReplaceInstUsesWith(AI, EntryAI);
258 if (AI.getAlignment()) {
259 // Check to see if this allocation is only modified by a memcpy/memmove from
260 // a constant global whose alignment is equal to or exceeds that of the
261 // allocation. If this is the case, we can change all users to use
262 // the constant global instead. This is commonly produced by the CFE by
263 // constructs like "void foo() { int A[] = {1,2,3,4,5,6,7,8,9...}; }" if 'A'
264 // is only subsequently read.
265 SmallVector<Instruction *, 4> ToDelete;
266 if (MemTransferInst *Copy = isOnlyCopiedFromConstantGlobal(&AI, ToDelete)) {
267 unsigned SourceAlign = getOrEnforceKnownAlignment(Copy->getSource(),
268 AI.getAlignment(), DL);
269 if (AI.getAlignment() <= SourceAlign) {
270 DEBUG(dbgs() << "Found alloca equal to global: " << AI << '\n');
271 DEBUG(dbgs() << " memcpy = " << *Copy << '\n');
272 for (unsigned i = 0, e = ToDelete.size(); i != e; ++i)
273 EraseInstFromFunction(*ToDelete[i]);
274 Constant *TheSrc = cast<Constant>(Copy->getSource());
276 = ConstantExpr::getPointerBitCastOrAddrSpaceCast(TheSrc, AI.getType());
277 Instruction *NewI = ReplaceInstUsesWith(AI, Cast);
278 EraseInstFromFunction(*Copy);
285 // At last, use the generic allocation site handler to aggressively remove
287 return visitAllocSite(AI);
291 /// InstCombineLoadCast - Fold 'load (cast P)' -> cast (load P)' when possible.
292 static Instruction *InstCombineLoadCast(InstCombiner &IC, LoadInst &LI,
293 const DataLayout *DL) {
294 User *CI = cast<User>(LI.getOperand(0));
295 Value *CastOp = CI->getOperand(0);
297 PointerType *DestTy = cast<PointerType>(CI->getType());
298 Type *DestPTy = DestTy->getElementType();
299 if (PointerType *SrcTy = dyn_cast<PointerType>(CastOp->getType())) {
301 // If the address spaces don't match, don't eliminate the cast.
302 if (DestTy->getAddressSpace() != SrcTy->getAddressSpace())
305 Type *SrcPTy = SrcTy->getElementType();
307 if (DestPTy->isIntegerTy() || DestPTy->isPointerTy() ||
308 DestPTy->isVectorTy()) {
309 // If the source is an array, the code below will not succeed. Check to
310 // see if a trivial 'gep P, 0, 0' will help matters. Only do this for
312 if (ArrayType *ASrcTy = dyn_cast<ArrayType>(SrcPTy))
313 if (Constant *CSrc = dyn_cast<Constant>(CastOp))
314 if (ASrcTy->getNumElements() != 0) {
316 ? DL->getIntPtrType(SrcTy)
317 : Type::getInt64Ty(SrcTy->getContext());
318 Value *Idx = Constant::getNullValue(IdxTy);
319 Value *Idxs[2] = { Idx, Idx };
320 CastOp = ConstantExpr::getGetElementPtr(CSrc, Idxs);
321 SrcTy = cast<PointerType>(CastOp->getType());
322 SrcPTy = SrcTy->getElementType();
325 if (IC.getDataLayout() &&
326 (SrcPTy->isIntegerTy() || SrcPTy->isPointerTy() ||
327 SrcPTy->isVectorTy()) &&
328 // Do not allow turning this into a load of an integer, which is then
329 // casted to a pointer, this pessimizes pointer analysis a lot.
330 (SrcPTy->isPtrOrPtrVectorTy() ==
331 LI.getType()->isPtrOrPtrVectorTy()) &&
332 IC.getDataLayout()->getTypeSizeInBits(SrcPTy) ==
333 IC.getDataLayout()->getTypeSizeInBits(DestPTy)) {
335 // Okay, we are casting from one integer or pointer type to another of
336 // the same size. Instead of casting the pointer before the load, cast
337 // the result of the loaded value.
339 IC.Builder->CreateLoad(CastOp, LI.isVolatile(), CI->getName());
340 NewLoad->setAlignment(LI.getAlignment());
341 NewLoad->setAtomic(LI.getOrdering(), LI.getSynchScope());
342 // Now cast the result of the load.
343 PointerType *OldTy = dyn_cast<PointerType>(NewLoad->getType());
344 PointerType *NewTy = dyn_cast<PointerType>(LI.getType());
345 if (OldTy && NewTy &&
346 OldTy->getAddressSpace() != NewTy->getAddressSpace()) {
347 return new AddrSpaceCastInst(NewLoad, LI.getType());
350 return new BitCastInst(NewLoad, LI.getType());
357 Instruction *InstCombiner::visitLoadInst(LoadInst &LI) {
358 Value *Op = LI.getOperand(0);
360 // Attempt to improve the alignment.
362 unsigned KnownAlign =
363 getOrEnforceKnownAlignment(Op, DL->getPrefTypeAlignment(LI.getType()),DL);
364 unsigned LoadAlign = LI.getAlignment();
365 unsigned EffectiveLoadAlign = LoadAlign != 0 ? LoadAlign :
366 DL->getABITypeAlignment(LI.getType());
368 if (KnownAlign > EffectiveLoadAlign)
369 LI.setAlignment(KnownAlign);
370 else if (LoadAlign == 0)
371 LI.setAlignment(EffectiveLoadAlign);
374 // load (cast X) --> cast (load X) iff safe.
375 if (isa<CastInst>(Op))
376 if (Instruction *Res = InstCombineLoadCast(*this, LI, DL))
379 // None of the following transforms are legal for volatile/atomic loads.
380 // FIXME: Some of it is okay for atomic loads; needs refactoring.
381 if (!LI.isSimple()) return nullptr;
383 // Do really simple store-to-load forwarding and load CSE, to catch cases
384 // where there are several consecutive memory accesses to the same location,
385 // separated by a few arithmetic operations.
386 BasicBlock::iterator BBI = &LI;
387 if (Value *AvailableVal = FindAvailableLoadedValue(Op, LI.getParent(), BBI,6))
388 return ReplaceInstUsesWith(LI, AvailableVal);
390 // load(gep null, ...) -> unreachable
391 if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(Op)) {
392 const Value *GEPI0 = GEPI->getOperand(0);
393 // TODO: Consider a target hook for valid address spaces for this xform.
394 if (isa<ConstantPointerNull>(GEPI0) && GEPI->getPointerAddressSpace() == 0){
395 // Insert a new store to null instruction before the load to indicate
396 // that this code is not reachable. We do this instead of inserting
397 // an unreachable instruction directly because we cannot modify the
399 new StoreInst(UndefValue::get(LI.getType()),
400 Constant::getNullValue(Op->getType()), &LI);
401 return ReplaceInstUsesWith(LI, UndefValue::get(LI.getType()));
405 // load null/undef -> unreachable
406 // TODO: Consider a target hook for valid address spaces for this xform.
407 if (isa<UndefValue>(Op) ||
408 (isa<ConstantPointerNull>(Op) && LI.getPointerAddressSpace() == 0)) {
409 // Insert a new store to null instruction before the load to indicate that
410 // this code is not reachable. We do this instead of inserting an
411 // unreachable instruction directly because we cannot modify the CFG.
412 new StoreInst(UndefValue::get(LI.getType()),
413 Constant::getNullValue(Op->getType()), &LI);
414 return ReplaceInstUsesWith(LI, UndefValue::get(LI.getType()));
417 // Instcombine load (constantexpr_cast global) -> cast (load global)
418 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Op))
420 if (Instruction *Res = InstCombineLoadCast(*this, LI, DL))
423 if (Op->hasOneUse()) {
424 // Change select and PHI nodes to select values instead of addresses: this
425 // helps alias analysis out a lot, allows many others simplifications, and
426 // exposes redundancy in the code.
428 // Note that we cannot do the transformation unless we know that the
429 // introduced loads cannot trap! Something like this is valid as long as
430 // the condition is always false: load (select bool %C, int* null, int* %G),
431 // but it would not be valid if we transformed it to load from null
434 if (SelectInst *SI = dyn_cast<SelectInst>(Op)) {
435 // load (select (Cond, &V1, &V2)) --> select(Cond, load &V1, load &V2).
436 unsigned Align = LI.getAlignment();
437 if (isSafeToLoadUnconditionally(SI->getOperand(1), SI, Align, DL) &&
438 isSafeToLoadUnconditionally(SI->getOperand(2), SI, Align, DL)) {
439 LoadInst *V1 = Builder->CreateLoad(SI->getOperand(1),
440 SI->getOperand(1)->getName()+".val");
441 LoadInst *V2 = Builder->CreateLoad(SI->getOperand(2),
442 SI->getOperand(2)->getName()+".val");
443 V1->setAlignment(Align);
444 V2->setAlignment(Align);
445 return SelectInst::Create(SI->getCondition(), V1, V2);
448 // load (select (cond, null, P)) -> load P
449 if (Constant *C = dyn_cast<Constant>(SI->getOperand(1)))
450 if (C->isNullValue()) {
451 LI.setOperand(0, SI->getOperand(2));
455 // load (select (cond, P, null)) -> load P
456 if (Constant *C = dyn_cast<Constant>(SI->getOperand(2)))
457 if (C->isNullValue()) {
458 LI.setOperand(0, SI->getOperand(1));
466 /// InstCombineStoreToCast - Fold store V, (cast P) -> store (cast V), P
467 /// when possible. This makes it generally easy to do alias analysis and/or
468 /// SROA/mem2reg of the memory object.
469 static Instruction *InstCombineStoreToCast(InstCombiner &IC, StoreInst &SI) {
470 User *CI = cast<User>(SI.getOperand(1));
471 Value *CastOp = CI->getOperand(0);
473 Type *DestPTy = cast<PointerType>(CI->getType())->getElementType();
474 PointerType *SrcTy = dyn_cast<PointerType>(CastOp->getType());
475 if (!SrcTy) return nullptr;
477 Type *SrcPTy = SrcTy->getElementType();
479 if (!DestPTy->isIntegerTy() && !DestPTy->isPointerTy())
482 /// NewGEPIndices - If SrcPTy is an aggregate type, we can emit a "noop gep"
483 /// to its first element. This allows us to handle things like:
484 /// store i32 xxx, (bitcast {foo*, float}* %P to i32*)
486 SmallVector<Value*, 4> NewGEPIndices;
488 // If the source is an array, the code below will not succeed. Check to
489 // see if a trivial 'gep P, 0, 0' will help matters. Only do this for
491 if (SrcPTy->isArrayTy() || SrcPTy->isStructTy()) {
492 // Index through pointer.
493 Constant *Zero = Constant::getNullValue(Type::getInt32Ty(SI.getContext()));
494 NewGEPIndices.push_back(Zero);
497 if (StructType *STy = dyn_cast<StructType>(SrcPTy)) {
498 if (!STy->getNumElements()) /* Struct can be empty {} */
500 NewGEPIndices.push_back(Zero);
501 SrcPTy = STy->getElementType(0);
502 } else if (ArrayType *ATy = dyn_cast<ArrayType>(SrcPTy)) {
503 NewGEPIndices.push_back(Zero);
504 SrcPTy = ATy->getElementType();
510 SrcTy = PointerType::get(SrcPTy, SrcTy->getAddressSpace());
513 if (!SrcPTy->isIntegerTy() && !SrcPTy->isPointerTy())
516 // If the pointers point into different address spaces don't do the
518 if (SrcTy->getAddressSpace() !=
519 cast<PointerType>(CI->getType())->getAddressSpace())
522 // If the pointers point to values of different sizes don't do the
524 if (!IC.getDataLayout() ||
525 IC.getDataLayout()->getTypeSizeInBits(SrcPTy) !=
526 IC.getDataLayout()->getTypeSizeInBits(DestPTy))
529 // If the pointers point to pointers to different address spaces don't do the
530 // transformation. It is not safe to introduce an addrspacecast instruction in
531 // this case since, depending on the target, addrspacecast may not be a no-op
533 if (SrcPTy->isPointerTy() && DestPTy->isPointerTy() &&
534 SrcPTy->getPointerAddressSpace() != DestPTy->getPointerAddressSpace())
537 // Okay, we are casting from one integer or pointer type to another of
538 // the same size. Instead of casting the pointer before
539 // the store, cast the value to be stored.
541 Instruction::CastOps opcode = Instruction::BitCast;
542 Type* CastSrcTy = DestPTy;
543 Type* CastDstTy = SrcPTy;
544 if (CastDstTy->isPointerTy()) {
545 if (CastSrcTy->isIntegerTy())
546 opcode = Instruction::IntToPtr;
547 } else if (CastDstTy->isIntegerTy()) {
548 if (CastSrcTy->isPointerTy())
549 opcode = Instruction::PtrToInt;
552 // SIOp0 is a pointer to aggregate and this is a store to the first field,
553 // emit a GEP to index into its first field.
554 if (!NewGEPIndices.empty())
555 CastOp = IC.Builder->CreateInBoundsGEP(CastOp, NewGEPIndices);
557 Value *SIOp0 = SI.getOperand(0);
558 NewCast = IC.Builder->CreateCast(opcode, SIOp0, CastDstTy,
559 SIOp0->getName()+".c");
560 SI.setOperand(0, NewCast);
561 SI.setOperand(1, CastOp);
565 /// equivalentAddressValues - Test if A and B will obviously have the same
566 /// value. This includes recognizing that %t0 and %t1 will have the same
567 /// value in code like this:
568 /// %t0 = getelementptr \@a, 0, 3
569 /// store i32 0, i32* %t0
570 /// %t1 = getelementptr \@a, 0, 3
571 /// %t2 = load i32* %t1
573 static bool equivalentAddressValues(Value *A, Value *B) {
574 // Test if the values are trivially equivalent.
575 if (A == B) return true;
577 // Test if the values come form identical arithmetic instructions.
578 // This uses isIdenticalToWhenDefined instead of isIdenticalTo because
579 // its only used to compare two uses within the same basic block, which
580 // means that they'll always either have the same value or one of them
581 // will have an undefined value.
582 if (isa<BinaryOperator>(A) ||
585 isa<GetElementPtrInst>(A))
586 if (Instruction *BI = dyn_cast<Instruction>(B))
587 if (cast<Instruction>(A)->isIdenticalToWhenDefined(BI))
590 // Otherwise they may not be equivalent.
594 Instruction *InstCombiner::visitStoreInst(StoreInst &SI) {
595 Value *Val = SI.getOperand(0);
596 Value *Ptr = SI.getOperand(1);
598 // Attempt to improve the alignment.
600 unsigned KnownAlign =
601 getOrEnforceKnownAlignment(Ptr, DL->getPrefTypeAlignment(Val->getType()),
603 unsigned StoreAlign = SI.getAlignment();
604 unsigned EffectiveStoreAlign = StoreAlign != 0 ? StoreAlign :
605 DL->getABITypeAlignment(Val->getType());
607 if (KnownAlign > EffectiveStoreAlign)
608 SI.setAlignment(KnownAlign);
609 else if (StoreAlign == 0)
610 SI.setAlignment(EffectiveStoreAlign);
613 // Don't hack volatile/atomic stores.
614 // FIXME: Some bits are legal for atomic stores; needs refactoring.
615 if (!SI.isSimple()) return nullptr;
617 // If the RHS is an alloca with a single use, zapify the store, making the
619 if (Ptr->hasOneUse()) {
620 if (isa<AllocaInst>(Ptr))
621 return EraseInstFromFunction(SI);
622 if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Ptr)) {
623 if (isa<AllocaInst>(GEP->getOperand(0))) {
624 if (GEP->getOperand(0)->hasOneUse())
625 return EraseInstFromFunction(SI);
630 // Do really simple DSE, to catch cases where there are several consecutive
631 // stores to the same location, separated by a few arithmetic operations. This
632 // situation often occurs with bitfield accesses.
633 BasicBlock::iterator BBI = &SI;
634 for (unsigned ScanInsts = 6; BBI != SI.getParent()->begin() && ScanInsts;
637 // Don't count debug info directives, lest they affect codegen,
638 // and we skip pointer-to-pointer bitcasts, which are NOPs.
639 if (isa<DbgInfoIntrinsic>(BBI) ||
640 (isa<BitCastInst>(BBI) && BBI->getType()->isPointerTy())) {
645 if (StoreInst *PrevSI = dyn_cast<StoreInst>(BBI)) {
646 // Prev store isn't volatile, and stores to the same location?
647 if (PrevSI->isSimple() && equivalentAddressValues(PrevSI->getOperand(1),
651 EraseInstFromFunction(*PrevSI);
657 // If this is a load, we have to stop. However, if the loaded value is from
658 // the pointer we're loading and is producing the pointer we're storing,
659 // then *this* store is dead (X = load P; store X -> P).
660 if (LoadInst *LI = dyn_cast<LoadInst>(BBI)) {
661 if (LI == Val && equivalentAddressValues(LI->getOperand(0), Ptr) &&
663 return EraseInstFromFunction(SI);
665 // Otherwise, this is a load from some other location. Stores before it
670 // Don't skip over loads or things that can modify memory.
671 if (BBI->mayWriteToMemory() || BBI->mayReadFromMemory())
675 // store X, null -> turns into 'unreachable' in SimplifyCFG
676 if (isa<ConstantPointerNull>(Ptr) && SI.getPointerAddressSpace() == 0) {
677 if (!isa<UndefValue>(Val)) {
678 SI.setOperand(0, UndefValue::get(Val->getType()));
679 if (Instruction *U = dyn_cast<Instruction>(Val))
680 Worklist.Add(U); // Dropped a use.
682 return nullptr; // Do not modify these!
685 // store undef, Ptr -> noop
686 if (isa<UndefValue>(Val))
687 return EraseInstFromFunction(SI);
689 // If the pointer destination is a cast, see if we can fold the cast into the
691 if (isa<CastInst>(Ptr))
692 if (Instruction *Res = InstCombineStoreToCast(*this, SI))
694 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(Ptr))
696 if (Instruction *Res = InstCombineStoreToCast(*this, SI))
700 // If this store is the last instruction in the basic block (possibly
701 // excepting debug info instructions), and if the block ends with an
702 // unconditional branch, try to move it to the successor block.
706 } while (isa<DbgInfoIntrinsic>(BBI) ||
707 (isa<BitCastInst>(BBI) && BBI->getType()->isPointerTy()));
708 if (BranchInst *BI = dyn_cast<BranchInst>(BBI))
709 if (BI->isUnconditional())
710 if (SimplifyStoreAtEndOfBlock(SI))
711 return nullptr; // xform done!
716 /// SimplifyStoreAtEndOfBlock - Turn things like:
717 /// if () { *P = v1; } else { *P = v2 }
718 /// into a phi node with a store in the successor.
720 /// Simplify things like:
721 /// *P = v1; if () { *P = v2; }
722 /// into a phi node with a store in the successor.
724 bool InstCombiner::SimplifyStoreAtEndOfBlock(StoreInst &SI) {
725 BasicBlock *StoreBB = SI.getParent();
727 // Check to see if the successor block has exactly two incoming edges. If
728 // so, see if the other predecessor contains a store to the same location.
729 // if so, insert a PHI node (if needed) and move the stores down.
730 BasicBlock *DestBB = StoreBB->getTerminator()->getSuccessor(0);
732 // Determine whether Dest has exactly two predecessors and, if so, compute
733 // the other predecessor.
734 pred_iterator PI = pred_begin(DestBB);
736 BasicBlock *OtherBB = nullptr;
741 if (++PI == pred_end(DestBB))
750 if (++PI != pred_end(DestBB))
753 // Bail out if all the relevant blocks aren't distinct (this can happen,
754 // for example, if SI is in an infinite loop)
755 if (StoreBB == DestBB || OtherBB == DestBB)
758 // Verify that the other block ends in a branch and is not otherwise empty.
759 BasicBlock::iterator BBI = OtherBB->getTerminator();
760 BranchInst *OtherBr = dyn_cast<BranchInst>(BBI);
761 if (!OtherBr || BBI == OtherBB->begin())
764 // If the other block ends in an unconditional branch, check for the 'if then
765 // else' case. there is an instruction before the branch.
766 StoreInst *OtherStore = nullptr;
767 if (OtherBr->isUnconditional()) {
769 // Skip over debugging info.
770 while (isa<DbgInfoIntrinsic>(BBI) ||
771 (isa<BitCastInst>(BBI) && BBI->getType()->isPointerTy())) {
772 if (BBI==OtherBB->begin())
776 // If this isn't a store, isn't a store to the same location, or is not the
777 // right kind of store, bail out.
778 OtherStore = dyn_cast<StoreInst>(BBI);
779 if (!OtherStore || OtherStore->getOperand(1) != SI.getOperand(1) ||
780 !SI.isSameOperationAs(OtherStore))
783 // Otherwise, the other block ended with a conditional branch. If one of the
784 // destinations is StoreBB, then we have the if/then case.
785 if (OtherBr->getSuccessor(0) != StoreBB &&
786 OtherBr->getSuccessor(1) != StoreBB)
789 // Okay, we know that OtherBr now goes to Dest and StoreBB, so this is an
790 // if/then triangle. See if there is a store to the same ptr as SI that
793 // Check to see if we find the matching store.
794 if ((OtherStore = dyn_cast<StoreInst>(BBI))) {
795 if (OtherStore->getOperand(1) != SI.getOperand(1) ||
796 !SI.isSameOperationAs(OtherStore))
800 // If we find something that may be using or overwriting the stored
801 // value, or if we run out of instructions, we can't do the xform.
802 if (BBI->mayReadFromMemory() || BBI->mayWriteToMemory() ||
803 BBI == OtherBB->begin())
807 // In order to eliminate the store in OtherBr, we have to
808 // make sure nothing reads or overwrites the stored value in
810 for (BasicBlock::iterator I = StoreBB->begin(); &*I != &SI; ++I) {
811 // FIXME: This should really be AA driven.
812 if (I->mayReadFromMemory() || I->mayWriteToMemory())
817 // Insert a PHI node now if we need it.
818 Value *MergedVal = OtherStore->getOperand(0);
819 if (MergedVal != SI.getOperand(0)) {
820 PHINode *PN = PHINode::Create(MergedVal->getType(), 2, "storemerge");
821 PN->addIncoming(SI.getOperand(0), SI.getParent());
822 PN->addIncoming(OtherStore->getOperand(0), OtherBB);
823 MergedVal = InsertNewInstBefore(PN, DestBB->front());
826 // Advance to a place where it is safe to insert the new store and
828 BBI = DestBB->getFirstInsertionPt();
829 StoreInst *NewSI = new StoreInst(MergedVal, SI.getOperand(1),
834 InsertNewInstBefore(NewSI, *BBI);
835 NewSI->setDebugLoc(OtherStore->getDebugLoc());
837 // If the two stores had the same TBAA tag, preserve it.
838 if (MDNode *TBAATag = SI.getMetadata(LLVMContext::MD_tbaa))
839 if ((TBAATag = MDNode::getMostGenericTBAA(TBAATag,
840 OtherStore->getMetadata(LLVMContext::MD_tbaa))))
841 NewSI->setMetadata(LLVMContext::MD_tbaa, TBAATag);
844 // Nuke the old stores.
845 EraseInstFromFunction(SI);
846 EraseInstFromFunction(*OtherStore);