Bug Summary

File:lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
Warning:line 1551, column 28
Called C++ object pointer is null

Annotated Source Code

1//===- InstCombineLoadStoreAlloca.cpp -------------------------------------===//
2//
3// The LLVM Compiler Infrastructure
4//
5// This file is distributed under the University of Illinois Open Source
6// License. See LICENSE.TXT for details.
7//
8//===----------------------------------------------------------------------===//
9//
10// This file implements the visit functions for load, store and alloca.
11//
12//===----------------------------------------------------------------------===//
13
14#include "InstCombineInternal.h"
15#include "llvm/ADT/MapVector.h"
16#include "llvm/ADT/SmallString.h"
17#include "llvm/ADT/Statistic.h"
18#include "llvm/Analysis/Loads.h"
19#include "llvm/IR/ConstantRange.h"
20#include "llvm/IR/DataLayout.h"
21#include "llvm/IR/IntrinsicInst.h"
22#include "llvm/IR/LLVMContext.h"
23#include "llvm/IR/MDBuilder.h"
24#include "llvm/IR/PatternMatch.h"
25#include "llvm/Transforms/Utils/BasicBlockUtils.h"
26#include "llvm/Transforms/Utils/Local.h"
27using namespace llvm;
28using namespace PatternMatch;
29
30#define DEBUG_TYPE"instcombine" "instcombine"
31
32STATISTIC(NumDeadStore, "Number of dead stores eliminated")static llvm::Statistic NumDeadStore = {"instcombine", "NumDeadStore"
, "Number of dead stores eliminated", {0}, false}
;
33STATISTIC(NumGlobalCopies, "Number of allocas copied from constant global")static llvm::Statistic NumGlobalCopies = {"instcombine", "NumGlobalCopies"
, "Number of allocas copied from constant global", {0}, false
}
;
34
35/// pointsToConstantGlobal - Return true if V (possibly indirectly) points to
36/// some part of a constant global variable. This intentionally only accepts
37/// constant expressions because we can't rewrite arbitrary instructions.
38static bool pointsToConstantGlobal(Value *V) {
39 if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V))
40 return GV->isConstant();
41
42 if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) {
43 if (CE->getOpcode() == Instruction::BitCast ||
44 CE->getOpcode() == Instruction::AddrSpaceCast ||
45 CE->getOpcode() == Instruction::GetElementPtr)
46 return pointsToConstantGlobal(CE->getOperand(0));
47 }
48 return false;
49}
50
51/// isOnlyCopiedFromConstantGlobal - Recursively walk the uses of a (derived)
52/// pointer to an alloca. Ignore any reads of the pointer, return false if we
53/// see any stores or other unknown uses. If we see pointer arithmetic, keep
54/// track of whether it moves the pointer (with IsOffset) but otherwise traverse
55/// the uses. If we see a memcpy/memmove that targets an unoffseted pointer to
56/// the alloca, and if the source pointer is a pointer to a constant global, we
57/// can optimize this.
58static bool
59isOnlyCopiedFromConstantGlobal(Value *V, MemTransferInst *&TheCopy,
60 SmallVectorImpl<Instruction *> &ToDelete) {
61 // We track lifetime intrinsics as we encounter them. If we decide to go
62 // ahead and replace the value with the global, this lets the caller quickly
63 // eliminate the markers.
64
65 SmallVector<std::pair<Value *, bool>, 35> ValuesToInspect;
66 ValuesToInspect.emplace_back(V, false);
67 while (!ValuesToInspect.empty()) {
68 auto ValuePair = ValuesToInspect.pop_back_val();
69 const bool IsOffset = ValuePair.second;
70 for (auto &U : ValuePair.first->uses()) {
71 auto *I = cast<Instruction>(U.getUser());
72
73 if (auto *LI = dyn_cast<LoadInst>(I)) {
74 // Ignore non-volatile loads, they are always ok.
75 if (!LI->isSimple()) return false;
76 continue;
77 }
78
79 if (isa<BitCastInst>(I) || isa<AddrSpaceCastInst>(I)) {
80 // If uses of the bitcast are ok, we are ok.
81 ValuesToInspect.emplace_back(I, IsOffset);
82 continue;
83 }
84 if (auto *GEP = dyn_cast<GetElementPtrInst>(I)) {
85 // If the GEP has all zero indices, it doesn't offset the pointer. If it
86 // doesn't, it does.
87 ValuesToInspect.emplace_back(I, IsOffset || !GEP->hasAllZeroIndices());
88 continue;
89 }
90
91 if (auto CS = CallSite(I)) {
92 // If this is the function being called then we treat it like a load and
93 // ignore it.
94 if (CS.isCallee(&U))
95 continue;
96
97 unsigned DataOpNo = CS.getDataOperandNo(&U);
98 bool IsArgOperand = CS.isArgOperand(&U);
99
100 // Inalloca arguments are clobbered by the call.
101 if (IsArgOperand && CS.isInAllocaArgument(DataOpNo))
102 return false;
103
104 // If this is a readonly/readnone call site, then we know it is just a
105 // load (but one that potentially returns the value itself), so we can
106 // ignore it if we know that the value isn't captured.
107 if (CS.onlyReadsMemory() &&
108 (CS.getInstruction()->use_empty() || CS.doesNotCapture(DataOpNo)))
109 continue;
110
111 // If this is being passed as a byval argument, the caller is making a
112 // copy, so it is only a read of the alloca.
113 if (IsArgOperand && CS.isByValArgument(DataOpNo))
114 continue;
115 }
116
117 // Lifetime intrinsics can be handled by the caller.
118 if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
119 if (II->getIntrinsicID() == Intrinsic::lifetime_start ||
120 II->getIntrinsicID() == Intrinsic::lifetime_end) {
121 assert(II->use_empty() && "Lifetime markers have no result to use!")(static_cast <bool> (II->use_empty() && "Lifetime markers have no result to use!"
) ? void (0) : __assert_fail ("II->use_empty() && \"Lifetime markers have no result to use!\""
, "/build/llvm-toolchain-snapshot-6.0~svn320940/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp"
, 121, __extension__ __PRETTY_FUNCTION__))
;
122 ToDelete.push_back(II);
123 continue;
124 }
125 }
126
127 // If this is isn't our memcpy/memmove, reject it as something we can't
128 // handle.
129 MemTransferInst *MI = dyn_cast<MemTransferInst>(I);
130 if (!MI)
131 return false;
132
133 // If the transfer is using the alloca as a source of the transfer, then
134 // ignore it since it is a load (unless the transfer is volatile).
135 if (U.getOperandNo() == 1) {
136 if (MI->isVolatile()) return false;
137 continue;
138 }
139
140 // If we already have seen a copy, reject the second one.
141 if (TheCopy) return false;
142
143 // If the pointer has been offset from the start of the alloca, we can't
144 // safely handle this.
145 if (IsOffset) return false;
146
147 // If the memintrinsic isn't using the alloca as the dest, reject it.
148 if (U.getOperandNo() != 0) return false;
149
150 // If the source of the memcpy/move is not a constant global, reject it.
151 if (!pointsToConstantGlobal(MI->getSource()))
152 return false;
153
154 // Otherwise, the transform is safe. Remember the copy instruction.
155 TheCopy = MI;
156 }
157 }
158 return true;
159}
160
161/// isOnlyCopiedFromConstantGlobal - Return true if the specified alloca is only
162/// modified by a copy from a constant global. If we can prove this, we can
163/// replace any uses of the alloca with uses of the global directly.
164static MemTransferInst *
165isOnlyCopiedFromConstantGlobal(AllocaInst *AI,
166 SmallVectorImpl<Instruction *> &ToDelete) {
167 MemTransferInst *TheCopy = nullptr;
168 if (isOnlyCopiedFromConstantGlobal(AI, TheCopy, ToDelete))
169 return TheCopy;
170 return nullptr;
171}
172
173/// Returns true if V is dereferenceable for size of alloca.
174static bool isDereferenceableForAllocaSize(const Value *V, const AllocaInst *AI,
175 const DataLayout &DL) {
176 if (AI->isArrayAllocation())
177 return false;
178 uint64_t AllocaSize = DL.getTypeStoreSize(AI->getAllocatedType());
179 if (!AllocaSize)
180 return false;
181 return isDereferenceableAndAlignedPointer(V, AI->getAlignment(),
182 APInt(64, AllocaSize), DL);
183}
184
185static Instruction *simplifyAllocaArraySize(InstCombiner &IC, AllocaInst &AI) {
186 // Check for array size of 1 (scalar allocation).
187 if (!AI.isArrayAllocation()) {
188 // i32 1 is the canonical array size for scalar allocations.
189 if (AI.getArraySize()->getType()->isIntegerTy(32))
190 return nullptr;
191
192 // Canonicalize it.
193 Value *V = IC.Builder.getInt32(1);
194 AI.setOperand(0, V);
195 return &AI;
196 }
197
198 // Convert: alloca Ty, C - where C is a constant != 1 into: alloca [C x Ty], 1
199 if (const ConstantInt *C = dyn_cast<ConstantInt>(AI.getArraySize())) {
200 Type *NewTy = ArrayType::get(AI.getAllocatedType(), C->getZExtValue());
201 AllocaInst *New = IC.Builder.CreateAlloca(NewTy, nullptr, AI.getName());
202 New->setAlignment(AI.getAlignment());
203
204 // Scan to the end of the allocation instructions, to skip over a block of
205 // allocas if possible...also skip interleaved debug info
206 //
207 BasicBlock::iterator It(New);
208 while (isa<AllocaInst>(*It) || isa<DbgInfoIntrinsic>(*It))
209 ++It;
210
211 // Now that I is pointing to the first non-allocation-inst in the block,
212 // insert our getelementptr instruction...
213 //
214 Type *IdxTy = IC.getDataLayout().getIntPtrType(AI.getType());
215 Value *NullIdx = Constant::getNullValue(IdxTy);
216 Value *Idx[2] = {NullIdx, NullIdx};
217 Instruction *GEP =
218 GetElementPtrInst::CreateInBounds(New, Idx, New->getName() + ".sub");
219 IC.InsertNewInstBefore(GEP, *It);
220
221 // Now make everything use the getelementptr instead of the original
222 // allocation.
223 return IC.replaceInstUsesWith(AI, GEP);
224 }
225
226 if (isa<UndefValue>(AI.getArraySize()))
227 return IC.replaceInstUsesWith(AI, Constant::getNullValue(AI.getType()));
228
229 // Ensure that the alloca array size argument has type intptr_t, so that
230 // any casting is exposed early.
231 Type *IntPtrTy = IC.getDataLayout().getIntPtrType(AI.getType());
232 if (AI.getArraySize()->getType() != IntPtrTy) {
233 Value *V = IC.Builder.CreateIntCast(AI.getArraySize(), IntPtrTy, false);
234 AI.setOperand(0, V);
235 return &AI;
236 }
237
238 return nullptr;
239}
240
241namespace {
242// If I and V are pointers in different address space, it is not allowed to
243// use replaceAllUsesWith since I and V have different types. A
244// non-target-specific transformation should not use addrspacecast on V since
245// the two address space may be disjoint depending on target.
246//
247// This class chases down uses of the old pointer until reaching the load
248// instructions, then replaces the old pointer in the load instructions with
249// the new pointer. If during the chasing it sees bitcast or GEP, it will
250// create new bitcast or GEP with the new pointer and use them in the load
251// instruction.
252class PointerReplacer {
253public:
254 PointerReplacer(InstCombiner &IC) : IC(IC) {}
255 void replacePointer(Instruction &I, Value *V);
256
257private:
258 void findLoadAndReplace(Instruction &I);
259 void replace(Instruction *I);
260 Value *getReplacement(Value *I);
261
262 SmallVector<Instruction *, 4> Path;
263 MapVector<Value *, Value *> WorkMap;
264 InstCombiner &IC;
265};
266} // end anonymous namespace
267
268void PointerReplacer::findLoadAndReplace(Instruction &I) {
269 for (auto U : I.users()) {
270 auto *Inst = dyn_cast<Instruction>(&*U);
271 if (!Inst)
272 return;
273 DEBUG(dbgs() << "Found pointer user: " << *U << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("instcombine")) { dbgs() << "Found pointer user: " <<
*U << '\n'; } } while (false)
;
274 if (isa<LoadInst>(Inst)) {
275 for (auto P : Path)
276 replace(P);
277 replace(Inst);
278 } else if (isa<GetElementPtrInst>(Inst) || isa<BitCastInst>(Inst)) {
279 Path.push_back(Inst);
280 findLoadAndReplace(*Inst);
281 Path.pop_back();
282 } else {
283 return;
284 }
285 }
286}
287
288Value *PointerReplacer::getReplacement(Value *V) {
289 auto Loc = WorkMap.find(V);
290 if (Loc != WorkMap.end())
291 return Loc->second;
292 return nullptr;
293}
294
295void PointerReplacer::replace(Instruction *I) {
296 if (getReplacement(I))
297 return;
298
299 if (auto *LT = dyn_cast<LoadInst>(I)) {
300 auto *V = getReplacement(LT->getPointerOperand());
301 assert(V && "Operand not replaced")(static_cast <bool> (V && "Operand not replaced"
) ? void (0) : __assert_fail ("V && \"Operand not replaced\""
, "/build/llvm-toolchain-snapshot-6.0~svn320940/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp"
, 301, __extension__ __PRETTY_FUNCTION__))
;
302 auto *NewI = new LoadInst(V);
303 NewI->takeName(LT);
304 IC.InsertNewInstWith(NewI, *LT);
305 IC.replaceInstUsesWith(*LT, NewI);
306 WorkMap[LT] = NewI;
307 } else if (auto *GEP = dyn_cast<GetElementPtrInst>(I)) {
308 auto *V = getReplacement(GEP->getPointerOperand());
309 assert(V && "Operand not replaced")(static_cast <bool> (V && "Operand not replaced"
) ? void (0) : __assert_fail ("V && \"Operand not replaced\""
, "/build/llvm-toolchain-snapshot-6.0~svn320940/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp"
, 309, __extension__ __PRETTY_FUNCTION__))
;
310 SmallVector<Value *, 8> Indices;
311 Indices.append(GEP->idx_begin(), GEP->idx_end());
312 auto *NewI = GetElementPtrInst::Create(
313 V->getType()->getPointerElementType(), V, Indices);
314 IC.InsertNewInstWith(NewI, *GEP);
315 NewI->takeName(GEP);
316 WorkMap[GEP] = NewI;
317 } else if (auto *BC = dyn_cast<BitCastInst>(I)) {
318 auto *V = getReplacement(BC->getOperand(0));
319 assert(V && "Operand not replaced")(static_cast <bool> (V && "Operand not replaced"
) ? void (0) : __assert_fail ("V && \"Operand not replaced\""
, "/build/llvm-toolchain-snapshot-6.0~svn320940/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp"
, 319, __extension__ __PRETTY_FUNCTION__))
;
320 auto *NewT = PointerType::get(BC->getType()->getPointerElementType(),
321 V->getType()->getPointerAddressSpace());
322 auto *NewI = new BitCastInst(V, NewT);
323 IC.InsertNewInstWith(NewI, *BC);
324 NewI->takeName(BC);
325 WorkMap[BC] = NewI;
326 } else {
327 llvm_unreachable("should never reach here")::llvm::llvm_unreachable_internal("should never reach here", "/build/llvm-toolchain-snapshot-6.0~svn320940/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp"
, 327)
;
328 }
329}
330
331void PointerReplacer::replacePointer(Instruction &I, Value *V) {
332#ifndef NDEBUG
333 auto *PT = cast<PointerType>(I.getType());
334 auto *NT = cast<PointerType>(V->getType());
335 assert(PT != NT && PT->getElementType() == NT->getElementType() &&(static_cast <bool> (PT != NT && PT->getElementType
() == NT->getElementType() && "Invalid usage") ? void
(0) : __assert_fail ("PT != NT && PT->getElementType() == NT->getElementType() && \"Invalid usage\""
, "/build/llvm-toolchain-snapshot-6.0~svn320940/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp"
, 336, __extension__ __PRETTY_FUNCTION__))
336 "Invalid usage")(static_cast <bool> (PT != NT && PT->getElementType
() == NT->getElementType() && "Invalid usage") ? void
(0) : __assert_fail ("PT != NT && PT->getElementType() == NT->getElementType() && \"Invalid usage\""
, "/build/llvm-toolchain-snapshot-6.0~svn320940/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp"
, 336, __extension__ __PRETTY_FUNCTION__))
;
337#endif
338 WorkMap[&I] = V;
339 findLoadAndReplace(I);
340}
341
342Instruction *InstCombiner::visitAllocaInst(AllocaInst &AI) {
343 if (auto *I = simplifyAllocaArraySize(*this, AI))
344 return I;
345
346 if (AI.getAllocatedType()->isSized()) {
347 // If the alignment is 0 (unspecified), assign it the preferred alignment.
348 if (AI.getAlignment() == 0)
349 AI.setAlignment(DL.getPrefTypeAlignment(AI.getAllocatedType()));
350
351 // Move all alloca's of zero byte objects to the entry block and merge them
352 // together. Note that we only do this for alloca's, because malloc should
353 // allocate and return a unique pointer, even for a zero byte allocation.
354 if (DL.getTypeAllocSize(AI.getAllocatedType()) == 0) {
355 // For a zero sized alloca there is no point in doing an array allocation.
356 // This is helpful if the array size is a complicated expression not used
357 // elsewhere.
358 if (AI.isArrayAllocation()) {
359 AI.setOperand(0, ConstantInt::get(AI.getArraySize()->getType(), 1));
360 return &AI;
361 }
362
363 // Get the first instruction in the entry block.
364 BasicBlock &EntryBlock = AI.getParent()->getParent()->getEntryBlock();
365 Instruction *FirstInst = EntryBlock.getFirstNonPHIOrDbg();
366 if (FirstInst != &AI) {
367 // If the entry block doesn't start with a zero-size alloca then move
368 // this one to the start of the entry block. There is no problem with
369 // dominance as the array size was forced to a constant earlier already.
370 AllocaInst *EntryAI = dyn_cast<AllocaInst>(FirstInst);
371 if (!EntryAI || !EntryAI->getAllocatedType()->isSized() ||
372 DL.getTypeAllocSize(EntryAI->getAllocatedType()) != 0) {
373 AI.moveBefore(FirstInst);
374 return &AI;
375 }
376
377 // If the alignment of the entry block alloca is 0 (unspecified),
378 // assign it the preferred alignment.
379 if (EntryAI->getAlignment() == 0)
380 EntryAI->setAlignment(
381 DL.getPrefTypeAlignment(EntryAI->getAllocatedType()));
382 // Replace this zero-sized alloca with the one at the start of the entry
383 // block after ensuring that the address will be aligned enough for both
384 // types.
385 unsigned MaxAlign = std::max(EntryAI->getAlignment(),
386 AI.getAlignment());
387 EntryAI->setAlignment(MaxAlign);
388 if (AI.getType() != EntryAI->getType())
389 return new BitCastInst(EntryAI, AI.getType());
390 return replaceInstUsesWith(AI, EntryAI);
391 }
392 }
393 }
394
395 if (AI.getAlignment()) {
396 // Check to see if this allocation is only modified by a memcpy/memmove from
397 // a constant global whose alignment is equal to or exceeds that of the
398 // allocation. If this is the case, we can change all users to use
399 // the constant global instead. This is commonly produced by the CFE by
400 // constructs like "void foo() { int A[] = {1,2,3,4,5,6,7,8,9...}; }" if 'A'
401 // is only subsequently read.
402 SmallVector<Instruction *, 4> ToDelete;
403 if (MemTransferInst *Copy = isOnlyCopiedFromConstantGlobal(&AI, ToDelete)) {
404 unsigned SourceAlign = getOrEnforceKnownAlignment(
405 Copy->getSource(), AI.getAlignment(), DL, &AI, &AC, &DT);
406 if (AI.getAlignment() <= SourceAlign &&
407 isDereferenceableForAllocaSize(Copy->getSource(), &AI, DL)) {
408 DEBUG(dbgs() << "Found alloca equal to global: " << AI << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("instcombine")) { dbgs() << "Found alloca equal to global: "
<< AI << '\n'; } } while (false)
;
409 DEBUG(dbgs() << " memcpy = " << *Copy << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("instcombine")) { dbgs() << " memcpy = " << *Copy
<< '\n'; } } while (false)
;
410 for (unsigned i = 0, e = ToDelete.size(); i != e; ++i)
411 eraseInstFromFunction(*ToDelete[i]);
412 Constant *TheSrc = cast<Constant>(Copy->getSource());
413 auto *SrcTy = TheSrc->getType();
414 auto *DestTy = PointerType::get(AI.getType()->getPointerElementType(),
415 SrcTy->getPointerAddressSpace());
416 Constant *Cast =
417 ConstantExpr::getPointerBitCastOrAddrSpaceCast(TheSrc, DestTy);
418 if (AI.getType()->getPointerAddressSpace() ==
419 SrcTy->getPointerAddressSpace()) {
420 Instruction *NewI = replaceInstUsesWith(AI, Cast);
421 eraseInstFromFunction(*Copy);
422 ++NumGlobalCopies;
423 return NewI;
424 } else {
425 PointerReplacer PtrReplacer(*this);
426 PtrReplacer.replacePointer(AI, Cast);
427 ++NumGlobalCopies;
428 }
429 }
430 }
431 }
432
433 // At last, use the generic allocation site handler to aggressively remove
434 // unused allocas.
435 return visitAllocSite(AI);
436}
437
438// Are we allowed to form a atomic load or store of this type?
439static bool isSupportedAtomicType(Type *Ty) {
440 return Ty->isIntegerTy() || Ty->isPointerTy() || Ty->isFloatingPointTy();
441}
442
443/// \brief Helper to combine a load to a new type.
444///
445/// This just does the work of combining a load to a new type. It handles
446/// metadata, etc., and returns the new instruction. The \c NewTy should be the
447/// loaded *value* type. This will convert it to a pointer, cast the operand to
448/// that pointer type, load it, etc.
449///
450/// Note that this will create all of the instructions with whatever insert
451/// point the \c InstCombiner currently is using.
452static LoadInst *combineLoadToNewType(InstCombiner &IC, LoadInst &LI, Type *NewTy,
453 const Twine &Suffix = "") {
454 assert((!LI.isAtomic() || isSupportedAtomicType(NewTy)) &&(static_cast <bool> ((!LI.isAtomic() || isSupportedAtomicType
(NewTy)) && "can't fold an atomic load to requested type"
) ? void (0) : __assert_fail ("(!LI.isAtomic() || isSupportedAtomicType(NewTy)) && \"can't fold an atomic load to requested type\""
, "/build/llvm-toolchain-snapshot-6.0~svn320940/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp"
, 455, __extension__ __PRETTY_FUNCTION__))
455 "can't fold an atomic load to requested type")(static_cast <bool> ((!LI.isAtomic() || isSupportedAtomicType
(NewTy)) && "can't fold an atomic load to requested type"
) ? void (0) : __assert_fail ("(!LI.isAtomic() || isSupportedAtomicType(NewTy)) && \"can't fold an atomic load to requested type\""
, "/build/llvm-toolchain-snapshot-6.0~svn320940/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp"
, 455, __extension__ __PRETTY_FUNCTION__))
;
456
457 Value *Ptr = LI.getPointerOperand();
458 unsigned AS = LI.getPointerAddressSpace();
459 SmallVector<std::pair<unsigned, MDNode *>, 8> MD;
460 LI.getAllMetadata(MD);
461
462 LoadInst *NewLoad = IC.Builder.CreateAlignedLoad(
463 IC.Builder.CreateBitCast(Ptr, NewTy->getPointerTo(AS)),
464 LI.getAlignment(), LI.isVolatile(), LI.getName() + Suffix);
465 NewLoad->setAtomic(LI.getOrdering(), LI.getSyncScopeID());
466 MDBuilder MDB(NewLoad->getContext());
467 for (const auto &MDPair : MD) {
468 unsigned ID = MDPair.first;
469 MDNode *N = MDPair.second;
470 // Note, essentially every kind of metadata should be preserved here! This
471 // routine is supposed to clone a load instruction changing *only its type*.
472 // The only metadata it makes sense to drop is metadata which is invalidated
473 // when the pointer type changes. This should essentially never be the case
474 // in LLVM, but we explicitly switch over only known metadata to be
475 // conservatively correct. If you are adding metadata to LLVM which pertains
476 // to loads, you almost certainly want to add it here.
477 switch (ID) {
478 case LLVMContext::MD_dbg:
479 case LLVMContext::MD_tbaa:
480 case LLVMContext::MD_prof:
481 case LLVMContext::MD_fpmath:
482 case LLVMContext::MD_tbaa_struct:
483 case LLVMContext::MD_invariant_load:
484 case LLVMContext::MD_alias_scope:
485 case LLVMContext::MD_noalias:
486 case LLVMContext::MD_nontemporal:
487 case LLVMContext::MD_mem_parallel_loop_access:
488 // All of these directly apply.
489 NewLoad->setMetadata(ID, N);
490 break;
491
492 case LLVMContext::MD_nonnull:
493 copyNonnullMetadata(LI, N, *NewLoad);
494 break;
495 case LLVMContext::MD_align:
496 case LLVMContext::MD_dereferenceable:
497 case LLVMContext::MD_dereferenceable_or_null:
498 // These only directly apply if the new type is also a pointer.
499 if (NewTy->isPointerTy())
500 NewLoad->setMetadata(ID, N);
501 break;
502 case LLVMContext::MD_range:
503 copyRangeMetadata(IC.getDataLayout(), LI, N, *NewLoad);
504 break;
505 }
506 }
507 return NewLoad;
508}
509
510/// \brief Combine a store to a new type.
511///
512/// Returns the newly created store instruction.
513static StoreInst *combineStoreToNewValue(InstCombiner &IC, StoreInst &SI, Value *V) {
514 assert((!SI.isAtomic() || isSupportedAtomicType(V->getType())) &&(static_cast <bool> ((!SI.isAtomic() || isSupportedAtomicType
(V->getType())) && "can't fold an atomic store of requested type"
) ? void (0) : __assert_fail ("(!SI.isAtomic() || isSupportedAtomicType(V->getType())) && \"can't fold an atomic store of requested type\""
, "/build/llvm-toolchain-snapshot-6.0~svn320940/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp"
, 515, __extension__ __PRETTY_FUNCTION__))
515 "can't fold an atomic store of requested type")(static_cast <bool> ((!SI.isAtomic() || isSupportedAtomicType
(V->getType())) && "can't fold an atomic store of requested type"
) ? void (0) : __assert_fail ("(!SI.isAtomic() || isSupportedAtomicType(V->getType())) && \"can't fold an atomic store of requested type\""
, "/build/llvm-toolchain-snapshot-6.0~svn320940/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp"
, 515, __extension__ __PRETTY_FUNCTION__))
;
516
517 Value *Ptr = SI.getPointerOperand();
518 unsigned AS = SI.getPointerAddressSpace();
519 SmallVector<std::pair<unsigned, MDNode *>, 8> MD;
520 SI.getAllMetadata(MD);
521
522 StoreInst *NewStore = IC.Builder.CreateAlignedStore(
523 V, IC.Builder.CreateBitCast(Ptr, V->getType()->getPointerTo(AS)),
524 SI.getAlignment(), SI.isVolatile());
525 NewStore->setAtomic(SI.getOrdering(), SI.getSyncScopeID());
526 for (const auto &MDPair : MD) {
527 unsigned ID = MDPair.first;
528 MDNode *N = MDPair.second;
529 // Note, essentially every kind of metadata should be preserved here! This
530 // routine is supposed to clone a store instruction changing *only its
531 // type*. The only metadata it makes sense to drop is metadata which is
532 // invalidated when the pointer type changes. This should essentially
533 // never be the case in LLVM, but we explicitly switch over only known
534 // metadata to be conservatively correct. If you are adding metadata to
535 // LLVM which pertains to stores, you almost certainly want to add it
536 // here.
537 switch (ID) {
538 case LLVMContext::MD_dbg:
539 case LLVMContext::MD_tbaa:
540 case LLVMContext::MD_prof:
541 case LLVMContext::MD_fpmath:
542 case LLVMContext::MD_tbaa_struct:
543 case LLVMContext::MD_alias_scope:
544 case LLVMContext::MD_noalias:
545 case LLVMContext::MD_nontemporal:
546 case LLVMContext::MD_mem_parallel_loop_access:
547 // All of these directly apply.
548 NewStore->setMetadata(ID, N);
549 break;
550
551 case LLVMContext::MD_invariant_load:
552 case LLVMContext::MD_nonnull:
553 case LLVMContext::MD_range:
554 case LLVMContext::MD_align:
555 case LLVMContext::MD_dereferenceable:
556 case LLVMContext::MD_dereferenceable_or_null:
557 // These don't apply for stores.
558 break;
559 }
560 }
561
562 return NewStore;
563}
564
565/// Returns true if instruction represent minmax pattern like:
566/// select ((cmp load V1, load V2), V1, V2).
567static bool isMinMaxWithLoads(Value *V) {
568 assert(V->getType()->isPointerTy() && "Expected pointer type.")(static_cast <bool> (V->getType()->isPointerTy() &&
"Expected pointer type.") ? void (0) : __assert_fail ("V->getType()->isPointerTy() && \"Expected pointer type.\""
, "/build/llvm-toolchain-snapshot-6.0~svn320940/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp"
, 568, __extension__ __PRETTY_FUNCTION__))
;
569 // Ignore possible ty* to ixx* bitcast.
570 V = peekThroughBitcast(V);
571 // Check that select is select ((cmp load V1, load V2), V1, V2) - minmax
572 // pattern.
573 CmpInst::Predicate Pred;
574 Instruction *L1;
575 Instruction *L2;
576 Value *LHS;
577 Value *RHS;
578 if (!match(V, m_Select(m_Cmp(Pred, m_Instruction(L1), m_Instruction(L2)),
579 m_Value(LHS), m_Value(RHS))))
580 return false;
581 return (match(L1, m_Load(m_Specific(LHS))) &&
582 match(L2, m_Load(m_Specific(RHS)))) ||
583 (match(L1, m_Load(m_Specific(RHS))) &&
584 match(L2, m_Load(m_Specific(LHS))));
585}
586
587/// \brief Combine loads to match the type of their uses' value after looking
588/// through intervening bitcasts.
589///
590/// The core idea here is that if the result of a load is used in an operation,
591/// we should load the type most conducive to that operation. For example, when
592/// loading an integer and converting that immediately to a pointer, we should
593/// instead directly load a pointer.
594///
595/// However, this routine must never change the width of a load or the number of
596/// loads as that would introduce a semantic change. This combine is expected to
597/// be a semantic no-op which just allows loads to more closely model the types
598/// of their consuming operations.
599///
600/// Currently, we also refuse to change the precise type used for an atomic load
601/// or a volatile load. This is debatable, and might be reasonable to change
602/// later. However, it is risky in case some backend or other part of LLVM is
603/// relying on the exact type loaded to select appropriate atomic operations.
604static Instruction *combineLoadToOperationType(InstCombiner &IC, LoadInst &LI) {
605 // FIXME: We could probably with some care handle both volatile and ordered
606 // atomic loads here but it isn't clear that this is important.
607 if (!LI.isUnordered())
608 return nullptr;
609
610 if (LI.use_empty())
611 return nullptr;
612
613 // swifterror values can't be bitcasted.
614 if (LI.getPointerOperand()->isSwiftError())
615 return nullptr;
616
617 Type *Ty = LI.getType();
618 const DataLayout &DL = IC.getDataLayout();
619
620 // Try to canonicalize loads which are only ever stored to operate over
621 // integers instead of any other type. We only do this when the loaded type
622 // is sized and has a size exactly the same as its store size and the store
623 // size is a legal integer type.
624 // Do not perform canonicalization if minmax pattern is found (to avoid
625 // infinite loop).
626 if (!Ty->isIntegerTy() && Ty->isSized() &&
627 DL.isLegalInteger(DL.getTypeStoreSizeInBits(Ty)) &&
628 DL.getTypeStoreSizeInBits(Ty) == DL.getTypeSizeInBits(Ty) &&
629 !DL.isNonIntegralPointerType(Ty) &&
630 !isMinMaxWithLoads(
631 peekThroughBitcast(LI.getPointerOperand(), /*OneUseOnly=*/true))) {
632 if (all_of(LI.users(), [&LI](User *U) {
633 auto *SI = dyn_cast<StoreInst>(U);
634 return SI && SI->getPointerOperand() != &LI &&
635 !SI->getPointerOperand()->isSwiftError();
636 })) {
637 LoadInst *NewLoad = combineLoadToNewType(
638 IC, LI,
639 Type::getIntNTy(LI.getContext(), DL.getTypeStoreSizeInBits(Ty)));
640 // Replace all the stores with stores of the newly loaded value.
641 for (auto UI = LI.user_begin(), UE = LI.user_end(); UI != UE;) {
642 auto *SI = cast<StoreInst>(*UI++);
643 IC.Builder.SetInsertPoint(SI);
644 combineStoreToNewValue(IC, *SI, NewLoad);
645 IC.eraseInstFromFunction(*SI);
646 }
647 assert(LI.use_empty() && "Failed to remove all users of the load!")(static_cast <bool> (LI.use_empty() && "Failed to remove all users of the load!"
) ? void (0) : __assert_fail ("LI.use_empty() && \"Failed to remove all users of the load!\""
, "/build/llvm-toolchain-snapshot-6.0~svn320940/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp"
, 647, __extension__ __PRETTY_FUNCTION__))
;
648 // Return the old load so the combiner can delete it safely.
649 return &LI;
650 }
651 }
652
653 // Fold away bit casts of the loaded value by loading the desired type.
654 // We can do this for BitCastInsts as well as casts from and to pointer types,
655 // as long as those are noops (i.e., the source or dest type have the same
656 // bitwidth as the target's pointers).
657 if (LI.hasOneUse())
658 if (auto* CI = dyn_cast<CastInst>(LI.user_back()))
659 if (CI->isNoopCast(DL))
660 if (!LI.isAtomic() || isSupportedAtomicType(CI->getDestTy())) {
661 LoadInst *NewLoad = combineLoadToNewType(IC, LI, CI->getDestTy());
662 CI->replaceAllUsesWith(NewLoad);
663 IC.eraseInstFromFunction(*CI);
664 return &LI;
665 }
666
667 // FIXME: We should also canonicalize loads of vectors when their elements are
668 // cast to other types.
669 return nullptr;
670}
671
672static Instruction *unpackLoadToAggregate(InstCombiner &IC, LoadInst &LI) {
673 // FIXME: We could probably with some care handle both volatile and atomic
674 // stores here but it isn't clear that this is important.
675 if (!LI.isSimple())
676 return nullptr;
677
678 Type *T = LI.getType();
679 if (!T->isAggregateType())
680 return nullptr;
681
682 StringRef Name = LI.getName();
683 assert(LI.getAlignment() && "Alignment must be set at this point")(static_cast <bool> (LI.getAlignment() && "Alignment must be set at this point"
) ? void (0) : __assert_fail ("LI.getAlignment() && \"Alignment must be set at this point\""
, "/build/llvm-toolchain-snapshot-6.0~svn320940/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp"
, 683, __extension__ __PRETTY_FUNCTION__))
;
684
685 if (auto *ST = dyn_cast<StructType>(T)) {
686 // If the struct only have one element, we unpack.
687 auto NumElements = ST->getNumElements();
688 if (NumElements == 1) {
689 LoadInst *NewLoad = combineLoadToNewType(IC, LI, ST->getTypeAtIndex(0U),
690 ".unpack");
691 AAMDNodes AAMD;
692 LI.getAAMetadata(AAMD);
693 NewLoad->setAAMetadata(AAMD);
694 return IC.replaceInstUsesWith(LI, IC.Builder.CreateInsertValue(
695 UndefValue::get(T), NewLoad, 0, Name));
696 }
697
698 // We don't want to break loads with padding here as we'd loose
699 // the knowledge that padding exists for the rest of the pipeline.
700 const DataLayout &DL = IC.getDataLayout();
701 auto *SL = DL.getStructLayout(ST);
702 if (SL->hasPadding())
703 return nullptr;
704
705 auto Align = LI.getAlignment();
706 if (!Align)
707 Align = DL.getABITypeAlignment(ST);
708
709 auto *Addr = LI.getPointerOperand();
710 auto *IdxType = Type::getInt32Ty(T->getContext());
711 auto *Zero = ConstantInt::get(IdxType, 0);
712
713 Value *V = UndefValue::get(T);
714 for (unsigned i = 0; i < NumElements; i++) {
715 Value *Indices[2] = {
716 Zero,
717 ConstantInt::get(IdxType, i),
718 };
719 auto *Ptr = IC.Builder.CreateInBoundsGEP(ST, Addr, makeArrayRef(Indices),
720 Name + ".elt");
721 auto EltAlign = MinAlign(Align, SL->getElementOffset(i));
722 auto *L = IC.Builder.CreateAlignedLoad(Ptr, EltAlign, Name + ".unpack");
723 // Propagate AA metadata. It'll still be valid on the narrowed load.
724 AAMDNodes AAMD;
725 LI.getAAMetadata(AAMD);
726 L->setAAMetadata(AAMD);
727 V = IC.Builder.CreateInsertValue(V, L, i);
728 }
729
730 V->setName(Name);
731 return IC.replaceInstUsesWith(LI, V);
732 }
733
734 if (auto *AT = dyn_cast<ArrayType>(T)) {
735 auto *ET = AT->getElementType();
736 auto NumElements = AT->getNumElements();
737 if (NumElements == 1) {
738 LoadInst *NewLoad = combineLoadToNewType(IC, LI, ET, ".unpack");
739 AAMDNodes AAMD;
740 LI.getAAMetadata(AAMD);
741 NewLoad->setAAMetadata(AAMD);
742 return IC.replaceInstUsesWith(LI, IC.Builder.CreateInsertValue(
743 UndefValue::get(T), NewLoad, 0, Name));
744 }
745
746 // Bail out if the array is too large. Ideally we would like to optimize
747 // arrays of arbitrary size but this has a terrible impact on compile time.
748 // The threshold here is chosen arbitrarily, maybe needs a little bit of
749 // tuning.
750 if (NumElements > IC.MaxArraySizeForCombine)
751 return nullptr;
752
753 const DataLayout &DL = IC.getDataLayout();
754 auto EltSize = DL.getTypeAllocSize(ET);
755 auto Align = LI.getAlignment();
756 if (!Align)
757 Align = DL.getABITypeAlignment(T);
758
759 auto *Addr = LI.getPointerOperand();
760 auto *IdxType = Type::getInt64Ty(T->getContext());
761 auto *Zero = ConstantInt::get(IdxType, 0);
762
763 Value *V = UndefValue::get(T);
764 uint64_t Offset = 0;
765 for (uint64_t i = 0; i < NumElements; i++) {
766 Value *Indices[2] = {
767 Zero,
768 ConstantInt::get(IdxType, i),
769 };
770 auto *Ptr = IC.Builder.CreateInBoundsGEP(AT, Addr, makeArrayRef(Indices),
771 Name + ".elt");
772 auto *L = IC.Builder.CreateAlignedLoad(Ptr, MinAlign(Align, Offset),
773 Name + ".unpack");
774 AAMDNodes AAMD;
775 LI.getAAMetadata(AAMD);
776 L->setAAMetadata(AAMD);
777 V = IC.Builder.CreateInsertValue(V, L, i);
778 Offset += EltSize;
779 }
780
781 V->setName(Name);
782 return IC.replaceInstUsesWith(LI, V);
783 }
784
785 return nullptr;
786}
787
788// If we can determine that all possible objects pointed to by the provided
789// pointer value are, not only dereferenceable, but also definitively less than
790// or equal to the provided maximum size, then return true. Otherwise, return
791// false (constant global values and allocas fall into this category).
792//
793// FIXME: This should probably live in ValueTracking (or similar).
794static bool isObjectSizeLessThanOrEq(Value *V, uint64_t MaxSize,
795 const DataLayout &DL) {
796 SmallPtrSet<Value *, 4> Visited;
797 SmallVector<Value *, 4> Worklist(1, V);
798
799 do {
800 Value *P = Worklist.pop_back_val();
801 P = P->stripPointerCasts();
802
803 if (!Visited.insert(P).second)
804 continue;
805
806 if (SelectInst *SI = dyn_cast<SelectInst>(P)) {
807 Worklist.push_back(SI->getTrueValue());
808 Worklist.push_back(SI->getFalseValue());
809 continue;
810 }
811
812 if (PHINode *PN = dyn_cast<PHINode>(P)) {
813 for (Value *IncValue : PN->incoming_values())
814 Worklist.push_back(IncValue);
815 continue;
816 }
817
818 if (GlobalAlias *GA = dyn_cast<GlobalAlias>(P)) {
819 if (GA->isInterposable())
820 return false;
821 Worklist.push_back(GA->getAliasee());
822 continue;
823 }
824
825 // If we know how big this object is, and it is less than MaxSize, continue
826 // searching. Otherwise, return false.
827 if (AllocaInst *AI = dyn_cast<AllocaInst>(P)) {
828 if (!AI->getAllocatedType()->isSized())
829 return false;
830
831 ConstantInt *CS = dyn_cast<ConstantInt>(AI->getArraySize());
832 if (!CS)
833 return false;
834
835 uint64_t TypeSize = DL.getTypeAllocSize(AI->getAllocatedType());
836 // Make sure that, even if the multiplication below would wrap as an
837 // uint64_t, we still do the right thing.
838 if ((CS->getValue().zextOrSelf(128)*APInt(128, TypeSize)).ugt(MaxSize))
839 return false;
840 continue;
841 }
842
843 if (GlobalVariable *GV = dyn_cast<GlobalVariable>(P)) {
844 if (!GV->hasDefinitiveInitializer() || !GV->isConstant())
845 return false;
846
847 uint64_t InitSize = DL.getTypeAllocSize(GV->getValueType());
848 if (InitSize > MaxSize)
849 return false;
850 continue;
851 }
852
853 return false;
854 } while (!Worklist.empty());
855
856 return true;
857}
858
859// If we're indexing into an object of a known size, and the outer index is
860// not a constant, but having any value but zero would lead to undefined
861// behavior, replace it with zero.
862//
863// For example, if we have:
864// @f.a = private unnamed_addr constant [1 x i32] [i32 12], align 4
865// ...
866// %arrayidx = getelementptr inbounds [1 x i32]* @f.a, i64 0, i64 %x
867// ... = load i32* %arrayidx, align 4
868// Then we know that we can replace %x in the GEP with i64 0.
869//
870// FIXME: We could fold any GEP index to zero that would cause UB if it were
871// not zero. Currently, we only handle the first such index. Also, we could
872// also search through non-zero constant indices if we kept track of the
873// offsets those indices implied.
874static bool canReplaceGEPIdxWithZero(InstCombiner &IC, GetElementPtrInst *GEPI,
875 Instruction *MemI, unsigned &Idx) {
876 if (GEPI->getNumOperands() < 2)
877 return false;
878
879 // Find the first non-zero index of a GEP. If all indices are zero, return
880 // one past the last index.
881 auto FirstNZIdx = [](const GetElementPtrInst *GEPI) {
882 unsigned I = 1;
883 for (unsigned IE = GEPI->getNumOperands(); I != IE; ++I) {
884 Value *V = GEPI->getOperand(I);
885 if (const ConstantInt *CI = dyn_cast<ConstantInt>(V))
886 if (CI->isZero())
887 continue;
888
889 break;
890 }
891
892 return I;
893 };
894
895 // Skip through initial 'zero' indices, and find the corresponding pointer
896 // type. See if the next index is not a constant.
897 Idx = FirstNZIdx(GEPI);
898 if (Idx == GEPI->getNumOperands())
899 return false;
900 if (isa<Constant>(GEPI->getOperand(Idx)))
901 return false;
902
903 SmallVector<Value *, 4> Ops(GEPI->idx_begin(), GEPI->idx_begin() + Idx);
904 Type *AllocTy =
905 GetElementPtrInst::getIndexedType(GEPI->getSourceElementType(), Ops);
906 if (!AllocTy || !AllocTy->isSized())
907 return false;
908 const DataLayout &DL = IC.getDataLayout();
909 uint64_t TyAllocSize = DL.getTypeAllocSize(AllocTy);
910
911 // If there are more indices after the one we might replace with a zero, make
912 // sure they're all non-negative. If any of them are negative, the overall
913 // address being computed might be before the base address determined by the
914 // first non-zero index.
915 auto IsAllNonNegative = [&]() {
916 for (unsigned i = Idx+1, e = GEPI->getNumOperands(); i != e; ++i) {
917 KnownBits Known = IC.computeKnownBits(GEPI->getOperand(i), 0, MemI);
918 if (Known.isNonNegative())
919 continue;
920 return false;
921 }
922
923 return true;
924 };
925
926 // FIXME: If the GEP is not inbounds, and there are extra indices after the
927 // one we'll replace, those could cause the address computation to wrap
928 // (rendering the IsAllNonNegative() check below insufficient). We can do
929 // better, ignoring zero indices (and other indices we can prove small
930 // enough not to wrap).
931 if (Idx+1 != GEPI->getNumOperands() && !GEPI->isInBounds())
932 return false;
933
934 // Note that isObjectSizeLessThanOrEq will return true only if the pointer is
935 // also known to be dereferenceable.
936 return isObjectSizeLessThanOrEq(GEPI->getOperand(0), TyAllocSize, DL) &&
937 IsAllNonNegative();
938}
939
940// If we're indexing into an object with a variable index for the memory
941// access, but the object has only one element, we can assume that the index
942// will always be zero. If we replace the GEP, return it.
943template <typename T>
944static Instruction *replaceGEPIdxWithZero(InstCombiner &IC, Value *Ptr,
945 T &MemI) {
946 if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(Ptr)) {
947 unsigned Idx;
948 if (canReplaceGEPIdxWithZero(IC, GEPI, &MemI, Idx)) {
949 Instruction *NewGEPI = GEPI->clone();
950 NewGEPI->setOperand(Idx,
951 ConstantInt::get(GEPI->getOperand(Idx)->getType(), 0));
952 NewGEPI->insertBefore(GEPI);
953 MemI.setOperand(MemI.getPointerOperandIndex(), NewGEPI);
954 return NewGEPI;
955 }
956 }
957
958 return nullptr;
959}
960
961static bool canSimplifyNullStoreOrGEP(StoreInst &SI) {
962 if (SI.getPointerAddressSpace() != 0)
963 return false;
964
965 auto *Ptr = SI.getPointerOperand();
966 if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(Ptr))
967 Ptr = GEPI->getOperand(0);
968 return isa<ConstantPointerNull>(Ptr);
969}
970
971static bool canSimplifyNullLoadOrGEP(LoadInst &LI, Value *Op) {
972 if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(Op)) {
973 const Value *GEPI0 = GEPI->getOperand(0);
974 if (isa<ConstantPointerNull>(GEPI0) && GEPI->getPointerAddressSpace() == 0)
975 return true;
976 }
977 if (isa<UndefValue>(Op) ||
978 (isa<ConstantPointerNull>(Op) && LI.getPointerAddressSpace() == 0))
979 return true;
980 return false;
981}
982
983Instruction *InstCombiner::visitLoadInst(LoadInst &LI) {
984 Value *Op = LI.getOperand(0);
985
986 // Try to canonicalize the loaded type.
987 if (Instruction *Res = combineLoadToOperationType(*this, LI))
988 return Res;
989
990 // Attempt to improve the alignment.
991 unsigned KnownAlign = getOrEnforceKnownAlignment(
992 Op, DL.getPrefTypeAlignment(LI.getType()), DL, &LI, &AC, &DT);
993 unsigned LoadAlign = LI.getAlignment();
994 unsigned EffectiveLoadAlign =
995 LoadAlign != 0 ? LoadAlign : DL.getABITypeAlignment(LI.getType());
996
997 if (KnownAlign > EffectiveLoadAlign)
998 LI.setAlignment(KnownAlign);
999 else if (LoadAlign == 0)
1000 LI.setAlignment(EffectiveLoadAlign);
1001
1002 // Replace GEP indices if possible.
1003 if (Instruction *NewGEPI = replaceGEPIdxWithZero(*this, Op, LI)) {
1004 Worklist.Add(NewGEPI);
1005 return &LI;
1006 }
1007
1008 if (Instruction *Res = unpackLoadToAggregate(*this, LI))
1009 return Res;
1010
1011 // Do really simple store-to-load forwarding and load CSE, to catch cases
1012 // where there are several consecutive memory accesses to the same location,
1013 // separated by a few arithmetic operations.
1014 BasicBlock::iterator BBI(LI);
1015 bool IsLoadCSE = false;
1016 if (Value *AvailableVal = FindAvailableLoadedValue(
1017 &LI, LI.getParent(), BBI, DefMaxInstsToScan, AA, &IsLoadCSE)) {
1018 if (IsLoadCSE)
1019 combineMetadataForCSE(cast<LoadInst>(AvailableVal), &LI);
1020
1021 return replaceInstUsesWith(
1022 LI, Builder.CreateBitOrPointerCast(AvailableVal, LI.getType(),
1023 LI.getName() + ".cast"));
1024 }
1025
1026 // None of the following transforms are legal for volatile/ordered atomic
1027 // loads. Most of them do apply for unordered atomics.
1028 if (!LI.isUnordered()) return nullptr;
1029
1030 // load(gep null, ...) -> unreachable
1031 // load null/undef -> unreachable
1032 // TODO: Consider a target hook for valid address spaces for this xforms.
1033 if (canSimplifyNullLoadOrGEP(LI, Op)) {
1034 // Insert a new store to null instruction before the load to indicate
1035 // that this code is not reachable. We do this instead of inserting
1036 // an unreachable instruction directly because we cannot modify the
1037 // CFG.
1038 StoreInst *SI = new StoreInst(UndefValue::get(LI.getType()),
1039 Constant::getNullValue(Op->getType()), &LI);
1040 SI->setDebugLoc(LI.getDebugLoc());
1041 return replaceInstUsesWith(LI, UndefValue::get(LI.getType()));
1042 }
1043
1044 if (Op->hasOneUse()) {
1045 // Change select and PHI nodes to select values instead of addresses: this
1046 // helps alias analysis out a lot, allows many others simplifications, and
1047 // exposes redundancy in the code.
1048 //
1049 // Note that we cannot do the transformation unless we know that the
1050 // introduced loads cannot trap! Something like this is valid as long as
1051 // the condition is always false: load (select bool %C, int* null, int* %G),
1052 // but it would not be valid if we transformed it to load from null
1053 // unconditionally.
1054 //
1055 if (SelectInst *SI = dyn_cast<SelectInst>(Op)) {
1056 // load (select (Cond, &V1, &V2)) --> select(Cond, load &V1, load &V2).
1057 unsigned Align = LI.getAlignment();
1058 if (isSafeToLoadUnconditionally(SI->getOperand(1), Align, DL, SI) &&
1059 isSafeToLoadUnconditionally(SI->getOperand(2), Align, DL, SI)) {
1060 LoadInst *V1 = Builder.CreateLoad(SI->getOperand(1),
1061 SI->getOperand(1)->getName()+".val");
1062 LoadInst *V2 = Builder.CreateLoad(SI->getOperand(2),
1063 SI->getOperand(2)->getName()+".val");
1064 assert(LI.isUnordered() && "implied by above")(static_cast <bool> (LI.isUnordered() && "implied by above"
) ? void (0) : __assert_fail ("LI.isUnordered() && \"implied by above\""
, "/build/llvm-toolchain-snapshot-6.0~svn320940/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp"
, 1064, __extension__ __PRETTY_FUNCTION__))
;
1065 V1->setAlignment(Align);
1066 V1->setAtomic(LI.getOrdering(), LI.getSyncScopeID());
1067 V2->setAlignment(Align);
1068 V2->setAtomic(LI.getOrdering(), LI.getSyncScopeID());
1069 return SelectInst::Create(SI->getCondition(), V1, V2);
1070 }
1071
1072 // load (select (cond, null, P)) -> load P
1073 if (isa<ConstantPointerNull>(SI->getOperand(1)) &&
1074 LI.getPointerAddressSpace() == 0) {
1075 LI.setOperand(0, SI->getOperand(2));
1076 return &LI;
1077 }
1078
1079 // load (select (cond, P, null)) -> load P
1080 if (isa<ConstantPointerNull>(SI->getOperand(2)) &&
1081 LI.getPointerAddressSpace() == 0) {
1082 LI.setOperand(0, SI->getOperand(1));
1083 return &LI;
1084 }
1085 }
1086 }
1087 return nullptr;
1088}
1089
1090/// \brief Look for extractelement/insertvalue sequence that acts like a bitcast.
1091///
1092/// \returns underlying value that was "cast", or nullptr otherwise.
1093///
1094/// For example, if we have:
1095///
1096/// %E0 = extractelement <2 x double> %U, i32 0
1097/// %V0 = insertvalue [2 x double] undef, double %E0, 0
1098/// %E1 = extractelement <2 x double> %U, i32 1
1099/// %V1 = insertvalue [2 x double] %V0, double %E1, 1
1100///
1101/// and the layout of a <2 x double> is isomorphic to a [2 x double],
1102/// then %V1 can be safely approximated by a conceptual "bitcast" of %U.
1103/// Note that %U may contain non-undef values where %V1 has undef.
1104static Value *likeBitCastFromVector(InstCombiner &IC, Value *V) {
1105 Value *U = nullptr;
1106 while (auto *IV = dyn_cast<InsertValueInst>(V)) {
1107 auto *E = dyn_cast<ExtractElementInst>(IV->getInsertedValueOperand());
1108 if (!E)
1109 return nullptr;
1110 auto *W = E->getVectorOperand();
1111 if (!U)
1112 U = W;
1113 else if (U != W)
1114 return nullptr;
1115 auto *CI = dyn_cast<ConstantInt>(E->getIndexOperand());
1116 if (!CI || IV->getNumIndices() != 1 || CI->getZExtValue() != *IV->idx_begin())
1117 return nullptr;
1118 V = IV->getAggregateOperand();
1119 }
1120 if (!isa<UndefValue>(V) ||!U)
1121 return nullptr;
1122
1123 auto *UT = cast<VectorType>(U->getType());
1124 auto *VT = V->getType();
1125 // Check that types UT and VT are bitwise isomorphic.
1126 const auto &DL = IC.getDataLayout();
1127 if (DL.getTypeStoreSizeInBits(UT) != DL.getTypeStoreSizeInBits(VT)) {
1128 return nullptr;
1129 }
1130 if (auto *AT = dyn_cast<ArrayType>(VT)) {
1131 if (AT->getNumElements() != UT->getNumElements())
1132 return nullptr;
1133 } else {
1134 auto *ST = cast<StructType>(VT);
1135 if (ST->getNumElements() != UT->getNumElements())
1136 return nullptr;
1137 for (const auto *EltT : ST->elements()) {
1138 if (EltT != UT->getElementType())
1139 return nullptr;
1140 }
1141 }
1142 return U;
1143}
1144
1145/// \brief Combine stores to match the type of value being stored.
1146///
1147/// The core idea here is that the memory does not have any intrinsic type and
1148/// where we can we should match the type of a store to the type of value being
1149/// stored.
1150///
1151/// However, this routine must never change the width of a store or the number of
1152/// stores as that would introduce a semantic change. This combine is expected to
1153/// be a semantic no-op which just allows stores to more closely model the types
1154/// of their incoming values.
1155///
1156/// Currently, we also refuse to change the precise type used for an atomic or
1157/// volatile store. This is debatable, and might be reasonable to change later.
1158/// However, it is risky in case some backend or other part of LLVM is relying
1159/// on the exact type stored to select appropriate atomic operations.
1160///
1161/// \returns true if the store was successfully combined away. This indicates
1162/// the caller must erase the store instruction. We have to let the caller erase
1163/// the store instruction as otherwise there is no way to signal whether it was
1164/// combined or not: IC.EraseInstFromFunction returns a null pointer.
1165static bool combineStoreToValueType(InstCombiner &IC, StoreInst &SI) {
1166 // FIXME: We could probably with some care handle both volatile and ordered
1167 // atomic stores here but it isn't clear that this is important.
1168 if (!SI.isUnordered())
1169 return false;
1170
1171 // swifterror values can't be bitcasted.
1172 if (SI.getPointerOperand()->isSwiftError())
1173 return false;
1174
1175 Value *V = SI.getValueOperand();
1176
1177 // Fold away bit casts of the stored value by storing the original type.
1178 if (auto *BC = dyn_cast<BitCastInst>(V)) {
1179 V = BC->getOperand(0);
1180 if (!SI.isAtomic() || isSupportedAtomicType(V->getType())) {
1181 combineStoreToNewValue(IC, SI, V);
1182 return true;
1183 }
1184 }
1185
1186 if (Value *U = likeBitCastFromVector(IC, V))
1187 if (!SI.isAtomic() || isSupportedAtomicType(U->getType())) {
1188 combineStoreToNewValue(IC, SI, U);
1189 return true;
1190 }
1191
1192 // FIXME: We should also canonicalize stores of vectors when their elements
1193 // are cast to other types.
1194 return false;
1195}
1196
1197static bool unpackStoreToAggregate(InstCombiner &IC, StoreInst &SI) {
1198 // FIXME: We could probably with some care handle both volatile and atomic
1199 // stores here but it isn't clear that this is important.
1200 if (!SI.isSimple())
1201 return false;
1202
1203 Value *V = SI.getValueOperand();
1204 Type *T = V->getType();
1205
1206 if (!T->isAggregateType())
1207 return false;
1208
1209 if (auto *ST = dyn_cast<StructType>(T)) {
1210 // If the struct only have one element, we unpack.
1211 unsigned Count = ST->getNumElements();
1212 if (Count == 1) {
1213 V = IC.Builder.CreateExtractValue(V, 0);
1214 combineStoreToNewValue(IC, SI, V);
1215 return true;
1216 }
1217
1218 // We don't want to break loads with padding here as we'd loose
1219 // the knowledge that padding exists for the rest of the pipeline.
1220 const DataLayout &DL = IC.getDataLayout();
1221 auto *SL = DL.getStructLayout(ST);
1222 if (SL->hasPadding())
1223 return false;
1224
1225 auto Align = SI.getAlignment();
1226 if (!Align)
1227 Align = DL.getABITypeAlignment(ST);
1228
1229 SmallString<16> EltName = V->getName();
1230 EltName += ".elt";
1231 auto *Addr = SI.getPointerOperand();
1232 SmallString<16> AddrName = Addr->getName();
1233 AddrName += ".repack";
1234
1235 auto *IdxType = Type::getInt32Ty(ST->getContext());
1236 auto *Zero = ConstantInt::get(IdxType, 0);
1237 for (unsigned i = 0; i < Count; i++) {
1238 Value *Indices[2] = {
1239 Zero,
1240 ConstantInt::get(IdxType, i),
1241 };
1242 auto *Ptr = IC.Builder.CreateInBoundsGEP(ST, Addr, makeArrayRef(Indices),
1243 AddrName);
1244 auto *Val = IC.Builder.CreateExtractValue(V, i, EltName);
1245 auto EltAlign = MinAlign(Align, SL->getElementOffset(i));
1246 llvm::Instruction *NS = IC.Builder.CreateAlignedStore(Val, Ptr, EltAlign);
1247 AAMDNodes AAMD;
1248 SI.getAAMetadata(AAMD);
1249 NS->setAAMetadata(AAMD);
1250 }
1251
1252 return true;
1253 }
1254
1255 if (auto *AT = dyn_cast<ArrayType>(T)) {
1256 // If the array only have one element, we unpack.
1257 auto NumElements = AT->getNumElements();
1258 if (NumElements == 1) {
1259 V = IC.Builder.CreateExtractValue(V, 0);
1260 combineStoreToNewValue(IC, SI, V);
1261 return true;
1262 }
1263
1264 // Bail out if the array is too large. Ideally we would like to optimize
1265 // arrays of arbitrary size but this has a terrible impact on compile time.
1266 // The threshold here is chosen arbitrarily, maybe needs a little bit of
1267 // tuning.
1268 if (NumElements > IC.MaxArraySizeForCombine)
1269 return false;
1270
1271 const DataLayout &DL = IC.getDataLayout();
1272 auto EltSize = DL.getTypeAllocSize(AT->getElementType());
1273 auto Align = SI.getAlignment();
1274 if (!Align)
1275 Align = DL.getABITypeAlignment(T);
1276
1277 SmallString<16> EltName = V->getName();
1278 EltName += ".elt";
1279 auto *Addr = SI.getPointerOperand();
1280 SmallString<16> AddrName = Addr->getName();
1281 AddrName += ".repack";
1282
1283 auto *IdxType = Type::getInt64Ty(T->getContext());
1284 auto *Zero = ConstantInt::get(IdxType, 0);
1285
1286 uint64_t Offset = 0;
1287 for (uint64_t i = 0; i < NumElements; i++) {
1288 Value *Indices[2] = {
1289 Zero,
1290 ConstantInt::get(IdxType, i),
1291 };
1292 auto *Ptr = IC.Builder.CreateInBoundsGEP(AT, Addr, makeArrayRef(Indices),
1293 AddrName);
1294 auto *Val = IC.Builder.CreateExtractValue(V, i, EltName);
1295 auto EltAlign = MinAlign(Align, Offset);
1296 Instruction *NS = IC.Builder.CreateAlignedStore(Val, Ptr, EltAlign);
1297 AAMDNodes AAMD;
1298 SI.getAAMetadata(AAMD);
1299 NS->setAAMetadata(AAMD);
1300 Offset += EltSize;
1301 }
1302
1303 return true;
1304 }
1305
1306 return false;
1307}
1308
1309/// equivalentAddressValues - Test if A and B will obviously have the same
1310/// value. This includes recognizing that %t0 and %t1 will have the same
1311/// value in code like this:
1312/// %t0 = getelementptr \@a, 0, 3
1313/// store i32 0, i32* %t0
1314/// %t1 = getelementptr \@a, 0, 3
1315/// %t2 = load i32* %t1
1316///
1317static bool equivalentAddressValues(Value *A, Value *B) {
1318 // Test if the values are trivially equivalent.
1319 if (A == B) return true;
1320
1321 // Test if the values come form identical arithmetic instructions.
1322 // This uses isIdenticalToWhenDefined instead of isIdenticalTo because
1323 // its only used to compare two uses within the same basic block, which
1324 // means that they'll always either have the same value or one of them
1325 // will have an undefined value.
1326 if (isa<BinaryOperator>(A) ||
1327 isa<CastInst>(A) ||
1328 isa<PHINode>(A) ||
1329 isa<GetElementPtrInst>(A))
1330 if (Instruction *BI = dyn_cast<Instruction>(B))
1331 if (cast<Instruction>(A)->isIdenticalToWhenDefined(BI))
1332 return true;
1333
1334 // Otherwise they may not be equivalent.
1335 return false;
1336}
1337
1338/// Converts store (bitcast (load (bitcast (select ...)))) to
1339/// store (load (select ...)), where select is minmax:
1340/// select ((cmp load V1, load V2), V1, V2).
1341static bool removeBitcastsFromLoadStoreOnMinMax(InstCombiner &IC,
1342 StoreInst &SI) {
1343 // bitcast?
1344 if (!match(SI.getPointerOperand(), m_BitCast(m_Value())))
1345 return false;
1346 // load? integer?
1347 Value *LoadAddr;
1348 if (!match(SI.getValueOperand(), m_Load(m_BitCast(m_Value(LoadAddr)))))
1349 return false;
1350 auto *LI = cast<LoadInst>(SI.getValueOperand());
1351 if (!LI->getType()->isIntegerTy())
1352 return false;
1353 if (!isMinMaxWithLoads(LoadAddr))
1354 return false;
1355
1356 if (!all_of(LI->users(), [LI, LoadAddr](User *U) {
1357 auto *SI = dyn_cast<StoreInst>(U);
1358 return SI && SI->getPointerOperand() != LI &&
1359 peekThroughBitcast(SI->getPointerOperand()) != LoadAddr &&
1360 !SI->getPointerOperand()->isSwiftError();
1361 }))
1362 return false;
1363
1364 IC.Builder.SetInsertPoint(LI);
1365 LoadInst *NewLI = combineLoadToNewType(
1366 IC, *LI, LoadAddr->getType()->getPointerElementType());
1367 // Replace all the stores with stores of the newly loaded value.
1368 for (auto *UI : LI->users()) {
1369 auto *USI = cast<StoreInst>(UI);
1370 IC.Builder.SetInsertPoint(USI);
1371 combineStoreToNewValue(IC, *USI, NewLI);
1372 }
1373 IC.replaceInstUsesWith(*LI, UndefValue::get(LI->getType()));
1374 IC.eraseInstFromFunction(*LI);
1375 return true;
1376}
1377
1378Instruction *InstCombiner::visitStoreInst(StoreInst &SI) {
1379 Value *Val = SI.getOperand(0);
1380 Value *Ptr = SI.getOperand(1);
1381
1382 // Try to canonicalize the stored type.
1383 if (combineStoreToValueType(*this, SI))
1
Taking false branch
1384 return eraseInstFromFunction(SI);
1385
1386 // Attempt to improve the alignment.
1387 unsigned KnownAlign = getOrEnforceKnownAlignment(
1388 Ptr, DL.getPrefTypeAlignment(Val->getType()), DL, &SI, &AC, &DT);
1389 unsigned StoreAlign = SI.getAlignment();
1390 unsigned EffectiveStoreAlign =
1391 StoreAlign != 0 ? StoreAlign : DL.getABITypeAlignment(Val->getType());
2
Assuming 'StoreAlign' is equal to 0
3
'?' condition is false
1392
1393 if (KnownAlign > EffectiveStoreAlign)
4
Taking false branch
1394 SI.setAlignment(KnownAlign);
1395 else if (StoreAlign == 0)
5
Taking true branch
1396 SI.setAlignment(EffectiveStoreAlign);
1397
1398 // Try to canonicalize the stored type.
1399 if (unpackStoreToAggregate(*this, SI))
6
Taking false branch
1400 return eraseInstFromFunction(SI);
1401
1402 if (removeBitcastsFromLoadStoreOnMinMax(*this, SI))
7
Taking false branch
1403 return eraseInstFromFunction(SI);
1404
1405 // Replace GEP indices if possible.
1406 if (Instruction *NewGEPI = replaceGEPIdxWithZero(*this, Ptr, SI)) {
8
Taking false branch
1407 Worklist.Add(NewGEPI);
1408 return &SI;
1409 }
1410
1411 // Don't hack volatile/ordered stores.
1412 // FIXME: Some bits are legal for ordered atomic stores; needs refactoring.
1413 if (!SI.isUnordered()) return nullptr;
9
Taking false branch
1414
1415 // If the RHS is an alloca with a single use, zapify the store, making the
1416 // alloca dead.
1417 if (Ptr->hasOneUse()) {
10
Assuming the condition is false
11
Taking false branch
1418 if (isa<AllocaInst>(Ptr))
1419 return eraseInstFromFunction(SI);
1420 if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Ptr)) {
1421 if (isa<AllocaInst>(GEP->getOperand(0))) {
1422 if (GEP->getOperand(0)->hasOneUse())
1423 return eraseInstFromFunction(SI);
1424 }
1425 }
1426 }
1427
1428 // Do really simple DSE, to catch cases where there are several consecutive
1429 // stores to the same location, separated by a few arithmetic operations. This
1430 // situation often occurs with bitfield accesses.
1431 BasicBlock::iterator BBI(SI);
1432 for (unsigned ScanInsts = 6; BBI != SI.getParent()->begin() && ScanInsts;
12
Loop condition is false. Execution continues on line 1476
1433 --ScanInsts) {
1434 --BBI;
1435 // Don't count debug info directives, lest they affect codegen,
1436 // and we skip pointer-to-pointer bitcasts, which are NOPs.
1437 if (isa<DbgInfoIntrinsic>(BBI) ||
1438 (isa<BitCastInst>(BBI) && BBI->getType()->isPointerTy())) {
1439 ScanInsts++;
1440 continue;
1441 }
1442
1443 if (StoreInst *PrevSI = dyn_cast<StoreInst>(BBI)) {
1444 // Prev store isn't volatile, and stores to the same location?
1445 if (PrevSI->isUnordered() && equivalentAddressValues(PrevSI->getOperand(1),
1446 SI.getOperand(1))) {
1447 ++NumDeadStore;
1448 ++BBI;
1449 eraseInstFromFunction(*PrevSI);
1450 continue;
1451 }
1452 break;
1453 }
1454
1455 // If this is a load, we have to stop. However, if the loaded value is from
1456 // the pointer we're loading and is producing the pointer we're storing,
1457 // then *this* store is dead (X = load P; store X -> P).
1458 if (LoadInst *LI = dyn_cast<LoadInst>(BBI)) {
1459 if (LI == Val && equivalentAddressValues(LI->getOperand(0), Ptr)) {
1460 assert(SI.isUnordered() && "can't eliminate ordering operation")(static_cast <bool> (SI.isUnordered() && "can't eliminate ordering operation"
) ? void (0) : __assert_fail ("SI.isUnordered() && \"can't eliminate ordering operation\""
, "/build/llvm-toolchain-snapshot-6.0~svn320940/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp"
, 1460, __extension__ __PRETTY_FUNCTION__))
;
1461 return eraseInstFromFunction(SI);
1462 }
1463
1464 // Otherwise, this is a load from some other location. Stores before it
1465 // may not be dead.
1466 break;
1467 }
1468
1469 // Don't skip over loads, throws or things that can modify memory.
1470 if (BBI->mayWriteToMemory() || BBI->mayReadFromMemory() || BBI->mayThrow())
1471 break;
1472 }
1473
1474 // store X, null -> turns into 'unreachable' in SimplifyCFG
1475 // store X, GEP(null, Y) -> turns into 'unreachable' in SimplifyCFG
1476 if (canSimplifyNullStoreOrGEP(SI)) {
13
Taking false branch
1477 if (!isa<UndefValue>(Val)) {
1478 SI.setOperand(0, UndefValue::get(Val->getType()));
1479 if (Instruction *U = dyn_cast<Instruction>(Val))
1480 Worklist.Add(U); // Dropped a use.
1481 }
1482 return nullptr; // Do not modify these!
1483 }
1484
1485 // store undef, Ptr -> noop
1486 if (isa<UndefValue>(Val))
14
Taking false branch
1487 return eraseInstFromFunction(SI);
1488
1489 // If this store is the last instruction in the basic block (possibly
1490 // excepting debug info instructions), and if the block ends with an
1491 // unconditional branch, try to move it to the successor block.
1492 BBI = SI.getIterator();
1493 do {
15
Loop condition is false. Exiting loop
1494 ++BBI;
1495 } while (isa<DbgInfoIntrinsic>(BBI) ||
1496 (isa<BitCastInst>(BBI) && BBI->getType()->isPointerTy()));
1497 if (BranchInst *BI = dyn_cast<BranchInst>(BBI))
16
Assuming 'BI' is non-null
17
Taking true branch
1498 if (BI->isUnconditional())
18
Taking true branch
1499 if (SimplifyStoreAtEndOfBlock(SI))
19
Calling 'InstCombiner::SimplifyStoreAtEndOfBlock'
1500 return nullptr; // xform done!
1501
1502 return nullptr;
1503}
1504
1505/// SimplifyStoreAtEndOfBlock - Turn things like:
1506/// if () { *P = v1; } else { *P = v2 }
1507/// into a phi node with a store in the successor.
1508///
1509/// Simplify things like:
1510/// *P = v1; if () { *P = v2; }
1511/// into a phi node with a store in the successor.
1512///
1513bool InstCombiner::SimplifyStoreAtEndOfBlock(StoreInst &SI) {
1514 assert(SI.isUnordered() &&(static_cast <bool> (SI.isUnordered() && "this code has not been auditted for volatile or ordered store case"
) ? void (0) : __assert_fail ("SI.isUnordered() && \"this code has not been auditted for volatile or ordered store case\""
, "/build/llvm-toolchain-snapshot-6.0~svn320940/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp"
, 1515, __extension__ __PRETTY_FUNCTION__))
1515 "this code has not been auditted for volatile or ordered store case")(static_cast <bool> (SI.isUnordered() && "this code has not been auditted for volatile or ordered store case"
) ? void (0) : __assert_fail ("SI.isUnordered() && \"this code has not been auditted for volatile or ordered store case\""
, "/build/llvm-toolchain-snapshot-6.0~svn320940/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp"
, 1515, __extension__ __PRETTY_FUNCTION__))
;
1516
1517 BasicBlock *StoreBB = SI.getParent();
1518
1519 // Check to see if the successor block has exactly two incoming edges. If
1520 // so, see if the other predecessor contains a store to the same location.
1521 // if so, insert a PHI node (if needed) and move the stores down.
1522 BasicBlock *DestBB = StoreBB->getTerminator()->getSuccessor(0);
1523
1524 // Determine whether Dest has exactly two predecessors and, if so, compute
1525 // the other predecessor.
1526 pred_iterator PI = pred_begin(DestBB);
1527 BasicBlock *P = *PI;
1528 BasicBlock *OtherBB = nullptr;
20
'OtherBB' initialized to a null pointer value
1529
1530 if (P != StoreBB)
21
Assuming 'P' is equal to 'StoreBB'
22
Taking false branch
1531 OtherBB = P;
1532
1533 if (++PI == pred_end(DestBB))
23
Assuming the condition is false
24
Taking false branch
1534 return false;
1535
1536 P = *PI;
1537 if (P != StoreBB) {
25
Assuming 'P' is equal to 'StoreBB'
26
Taking false branch
1538 if (OtherBB)
1539 return false;
1540 OtherBB = P;
1541 }
1542 if (++PI != pred_end(DestBB))
27
Assuming the condition is false
28
Taking false branch
1543 return false;
1544
1545 // Bail out if all the relevant blocks aren't distinct (this can happen,
1546 // for example, if SI is in an infinite loop)
1547 if (StoreBB == DestBB || OtherBB == DestBB)
29
Assuming 'StoreBB' is not equal to 'DestBB'
30
Assuming 'OtherBB' is not equal to 'DestBB'
31
Taking false branch
1548 return false;
1549
1550 // Verify that the other block ends in a branch and is not otherwise empty.
1551 BasicBlock::iterator BBI(OtherBB->getTerminator());
32
Called C++ object pointer is null
1552 BranchInst *OtherBr = dyn_cast<BranchInst>(BBI);
1553 if (!OtherBr || BBI == OtherBB->begin())
1554 return false;
1555
1556 // If the other block ends in an unconditional branch, check for the 'if then
1557 // else' case. there is an instruction before the branch.
1558 StoreInst *OtherStore = nullptr;
1559 if (OtherBr->isUnconditional()) {
1560 --BBI;
1561 // Skip over debugging info.
1562 while (isa<DbgInfoIntrinsic>(BBI) ||
1563 (isa<BitCastInst>(BBI) && BBI->getType()->isPointerTy())) {
1564 if (BBI==OtherBB->begin())
1565 return false;
1566 --BBI;
1567 }
1568 // If this isn't a store, isn't a store to the same location, or is not the
1569 // right kind of store, bail out.
1570 OtherStore = dyn_cast<StoreInst>(BBI);
1571 if (!OtherStore || OtherStore->getOperand(1) != SI.getOperand(1) ||
1572 !SI.isSameOperationAs(OtherStore))
1573 return false;
1574 } else {
1575 // Otherwise, the other block ended with a conditional branch. If one of the
1576 // destinations is StoreBB, then we have the if/then case.
1577 if (OtherBr->getSuccessor(0) != StoreBB &&
1578 OtherBr->getSuccessor(1) != StoreBB)
1579 return false;
1580
1581 // Okay, we know that OtherBr now goes to Dest and StoreBB, so this is an
1582 // if/then triangle. See if there is a store to the same ptr as SI that
1583 // lives in OtherBB.
1584 for (;; --BBI) {
1585 // Check to see if we find the matching store.
1586 if ((OtherStore = dyn_cast<StoreInst>(BBI))) {
1587 if (OtherStore->getOperand(1) != SI.getOperand(1) ||
1588 !SI.isSameOperationAs(OtherStore))
1589 return false;
1590 break;
1591 }
1592 // If we find something that may be using or overwriting the stored
1593 // value, or if we run out of instructions, we can't do the xform.
1594 if (BBI->mayReadFromMemory() || BBI->mayThrow() ||
1595 BBI->mayWriteToMemory() || BBI == OtherBB->begin())
1596 return false;
1597 }
1598
1599 // In order to eliminate the store in OtherBr, we have to
1600 // make sure nothing reads or overwrites the stored value in
1601 // StoreBB.
1602 for (BasicBlock::iterator I = StoreBB->begin(); &*I != &SI; ++I) {
1603 // FIXME: This should really be AA driven.
1604 if (I->mayReadFromMemory() || I->mayThrow() || I->mayWriteToMemory())
1605 return false;
1606 }
1607 }
1608
1609 // Insert a PHI node now if we need it.
1610 Value *MergedVal = OtherStore->getOperand(0);
1611 if (MergedVal != SI.getOperand(0)) {
1612 PHINode *PN = PHINode::Create(MergedVal->getType(), 2, "storemerge");
1613 PN->addIncoming(SI.getOperand(0), SI.getParent());
1614 PN->addIncoming(OtherStore->getOperand(0), OtherBB);
1615 MergedVal = InsertNewInstBefore(PN, DestBB->front());
1616 }
1617
1618 // Advance to a place where it is safe to insert the new store and
1619 // insert it.
1620 BBI = DestBB->getFirstInsertionPt();
1621 StoreInst *NewSI = new StoreInst(MergedVal, SI.getOperand(1),
1622 SI.isVolatile(),
1623 SI.getAlignment(),
1624 SI.getOrdering(),
1625 SI.getSyncScopeID());
1626 InsertNewInstBefore(NewSI, *BBI);
1627 // The debug locations of the original instructions might differ; merge them.
1628 NewSI->applyMergedLocation(SI.getDebugLoc(), OtherStore->getDebugLoc());
1629
1630 // If the two stores had AA tags, merge them.
1631 AAMDNodes AATags;
1632 SI.getAAMetadata(AATags);
1633 if (AATags) {
1634 OtherStore->getAAMetadata(AATags, /* Merge = */ true);
1635 NewSI->setAAMetadata(AATags);
1636 }
1637
1638 // Nuke the old stores.
1639 eraseInstFromFunction(SI);
1640 eraseInstFromFunction(*OtherStore);
1641 return true;
1642}