Bug Summary

File:lib/Transforms/Vectorize/LoadStoreVectorizer.cpp
Warning:line 985, column 17
1st function call argument is an uninitialized value

Annotated Source Code

Press '?' to see keyboard shortcuts

clang -cc1 -triple x86_64-pc-linux-gnu -analyze -disable-free -disable-llvm-verifier -discard-value-names -main-file-name LoadStoreVectorizer.cpp -analyzer-store=region -analyzer-opt-analyze-nested-blocks -analyzer-eagerly-assume -analyzer-checker=core -analyzer-checker=apiModeling -analyzer-checker=unix -analyzer-checker=deadcode -analyzer-checker=cplusplus -analyzer-checker=security.insecureAPI.UncheckedReturn -analyzer-checker=security.insecureAPI.getpw -analyzer-checker=security.insecureAPI.gets -analyzer-checker=security.insecureAPI.mktemp -analyzer-checker=security.insecureAPI.mkstemp -analyzer-checker=security.insecureAPI.vfork -analyzer-checker=nullability.NullPassedToNonnull -analyzer-checker=nullability.NullReturnedFromNonnull -analyzer-output plist -w -mrelocation-model pic -pic-level 2 -mthread-model posix -fmath-errno -masm-verbose -mconstructor-aliases -munwind-tables -fuse-init-array -target-cpu x86-64 -dwarf-column-info -debugger-tuning=gdb -momit-leaf-frame-pointer -ffunction-sections -fdata-sections -resource-dir /usr/lib/llvm-7/lib/clang/7.0.0 -D _DEBUG -D _GNU_SOURCE -D __STDC_CONSTANT_MACROS -D __STDC_FORMAT_MACROS -D __STDC_LIMIT_MACROS -I /build/llvm-toolchain-snapshot-7~svn329677/build-llvm/lib/Transforms/Vectorize -I /build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize -I /build/llvm-toolchain-snapshot-7~svn329677/build-llvm/include -I /build/llvm-toolchain-snapshot-7~svn329677/include -U NDEBUG -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/x86_64-linux-gnu/c++/7.3.0 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/x86_64-linux-gnu/c++/7.3.0 -internal-isystem /usr/lib/gcc/x86_64-linux-gnu/7.3.0/../../../../include/c++/7.3.0/backward -internal-isystem /usr/include/clang/7.0.0/include/ -internal-isystem /usr/local/include -internal-isystem /usr/lib/llvm-7/lib/clang/7.0.0/include -internal-externc-isystem /usr/include/x86_64-linux-gnu -internal-externc-isystem /include -internal-externc-isystem /usr/include -O2 -Wno-unused-parameter -Wwrite-strings -Wno-missing-field-initializers -Wno-long-long -Wno-maybe-uninitialized -Wno-comment -std=c++11 -fdeprecated-macro -fdebug-compilation-dir /build/llvm-toolchain-snapshot-7~svn329677/build-llvm/lib/Transforms/Vectorize -ferror-limit 19 -fmessage-length 0 -fvisibility-inlines-hidden -fobjc-runtime=gcc -fdiagnostics-show-option -vectorize-loops -vectorize-slp -analyzer-checker optin.performance.Padding -analyzer-output=html -analyzer-config stable-report-filename=true -o /tmp/scan-build-2018-04-11-031539-24776-1 -x c++ /build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp
1//===- LoadStoreVectorizer.cpp - GPU Load & Store Vectorizer --------------===//
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 pass merges loads/stores to/from sequential memory addresses into vector
11// loads/stores. Although there's nothing GPU-specific in here, this pass is
12// motivated by the microarchitectural quirks of nVidia and AMD GPUs.
13//
14// (For simplicity below we talk about loads only, but everything also applies
15// to stores.)
16//
17// This pass is intended to be run late in the pipeline, after other
18// vectorization opportunities have been exploited. So the assumption here is
19// that immediately following our new vector load we'll need to extract out the
20// individual elements of the load, so we can operate on them individually.
21//
22// On CPUs this transformation is usually not beneficial, because extracting the
23// elements of a vector register is expensive on most architectures. It's
24// usually better just to load each element individually into its own scalar
25// register.
26//
27// However, nVidia and AMD GPUs don't have proper vector registers. Instead, a
28// "vector load" loads directly into a series of scalar registers. In effect,
29// extracting the elements of the vector is free. It's therefore always
30// beneficial to vectorize a sequence of loads on these architectures.
31//
32// Vectorizing (perhaps a better name might be "coalescing") loads can have
33// large performance impacts on GPU kernels, and opportunities for vectorizing
34// are common in GPU code. This pass tries very hard to find such
35// opportunities; its runtime is quadratic in the number of loads in a BB.
36//
37// Some CPU architectures, such as ARM, have instructions that load into
38// multiple scalar registers, similar to a GPU vectorized load. In theory ARM
39// could use this pass (with some modifications), but currently it implements
40// its own pass to do something similar to what we do here.
41
42#include "llvm/ADT/APInt.h"
43#include "llvm/ADT/ArrayRef.h"
44#include "llvm/ADT/MapVector.h"
45#include "llvm/ADT/PostOrderIterator.h"
46#include "llvm/ADT/STLExtras.h"
47#include "llvm/ADT/SmallPtrSet.h"
48#include "llvm/ADT/SmallVector.h"
49#include "llvm/ADT/Statistic.h"
50#include "llvm/ADT/iterator_range.h"
51#include "llvm/Analysis/AliasAnalysis.h"
52#include "llvm/Analysis/MemoryLocation.h"
53#include "llvm/Analysis/OrderedBasicBlock.h"
54#include "llvm/Analysis/ScalarEvolution.h"
55#include "llvm/Analysis/TargetTransformInfo.h"
56#include "llvm/Analysis/Utils/Local.h"
57#include "llvm/Analysis/ValueTracking.h"
58#include "llvm/Analysis/VectorUtils.h"
59#include "llvm/IR/Attributes.h"
60#include "llvm/IR/BasicBlock.h"
61#include "llvm/IR/Constants.h"
62#include "llvm/IR/DataLayout.h"
63#include "llvm/IR/DerivedTypes.h"
64#include "llvm/IR/Dominators.h"
65#include "llvm/IR/Function.h"
66#include "llvm/IR/IRBuilder.h"
67#include "llvm/IR/InstrTypes.h"
68#include "llvm/IR/Instruction.h"
69#include "llvm/IR/Instructions.h"
70#include "llvm/IR/IntrinsicInst.h"
71#include "llvm/IR/Module.h"
72#include "llvm/IR/Type.h"
73#include "llvm/IR/User.h"
74#include "llvm/IR/Value.h"
75#include "llvm/Pass.h"
76#include "llvm/Support/Casting.h"
77#include "llvm/Support/Debug.h"
78#include "llvm/Support/KnownBits.h"
79#include "llvm/Support/MathExtras.h"
80#include "llvm/Support/raw_ostream.h"
81#include "llvm/Transforms/Vectorize.h"
82#include <algorithm>
83#include <cassert>
84#include <cstdlib>
85#include <tuple>
86#include <utility>
87
88using namespace llvm;
89
90#define DEBUG_TYPE"load-store-vectorizer" "load-store-vectorizer"
91
92STATISTIC(NumVectorInstructions, "Number of vector accesses generated")static llvm::Statistic NumVectorInstructions = {"load-store-vectorizer"
, "NumVectorInstructions", "Number of vector accesses generated"
, {0}, {false}}
;
93STATISTIC(NumScalarsVectorized, "Number of scalar accesses vectorized")static llvm::Statistic NumScalarsVectorized = {"load-store-vectorizer"
, "NumScalarsVectorized", "Number of scalar accesses vectorized"
, {0}, {false}}
;
94
95// FIXME: Assuming stack alignment of 4 is always good enough
96static const unsigned StackAdjustedAlignment = 4;
97
98namespace {
99
100using InstrList = SmallVector<Instruction *, 8>;
101using InstrListMap = MapVector<Value *, InstrList>;
102
103class Vectorizer {
104 Function &F;
105 AliasAnalysis &AA;
106 DominatorTree &DT;
107 ScalarEvolution &SE;
108 TargetTransformInfo &TTI;
109 const DataLayout &DL;
110 IRBuilder<> Builder;
111
112public:
113 Vectorizer(Function &F, AliasAnalysis &AA, DominatorTree &DT,
114 ScalarEvolution &SE, TargetTransformInfo &TTI)
115 : F(F), AA(AA), DT(DT), SE(SE), TTI(TTI),
116 DL(F.getParent()->getDataLayout()), Builder(SE.getContext()) {}
117
118 bool run();
119
120private:
121 GetElementPtrInst *getSourceGEP(Value *Src) const;
122
123 unsigned getPointerAddressSpace(Value *I);
124
125 unsigned getAlignment(LoadInst *LI) const {
126 unsigned Align = LI->getAlignment();
127 if (Align != 0)
128 return Align;
129
130 return DL.getABITypeAlignment(LI->getType());
131 }
132
133 unsigned getAlignment(StoreInst *SI) const {
134 unsigned Align = SI->getAlignment();
135 if (Align != 0)
136 return Align;
137
138 return DL.getABITypeAlignment(SI->getValueOperand()->getType());
139 }
140
141 bool isConsecutiveAccess(Value *A, Value *B);
142
143 /// After vectorization, reorder the instructions that I depends on
144 /// (the instructions defining its operands), to ensure they dominate I.
145 void reorder(Instruction *I);
146
147 /// Returns the first and the last instructions in Chain.
148 std::pair<BasicBlock::iterator, BasicBlock::iterator>
149 getBoundaryInstrs(ArrayRef<Instruction *> Chain);
150
151 /// Erases the original instructions after vectorizing.
152 void eraseInstructions(ArrayRef<Instruction *> Chain);
153
154 /// "Legalize" the vector type that would be produced by combining \p
155 /// ElementSizeBits elements in \p Chain. Break into two pieces such that the
156 /// total size of each piece is 1, 2 or a multiple of 4 bytes. \p Chain is
157 /// expected to have more than 4 elements.
158 std::pair<ArrayRef<Instruction *>, ArrayRef<Instruction *>>
159 splitOddVectorElts(ArrayRef<Instruction *> Chain, unsigned ElementSizeBits);
160
161 /// Finds the largest prefix of Chain that's vectorizable, checking for
162 /// intervening instructions which may affect the memory accessed by the
163 /// instructions within Chain.
164 ///
165 /// The elements of \p Chain must be all loads or all stores and must be in
166 /// address order.
167 ArrayRef<Instruction *> getVectorizablePrefix(ArrayRef<Instruction *> Chain);
168
169 /// Collects load and store instructions to vectorize.
170 std::pair<InstrListMap, InstrListMap> collectInstructions(BasicBlock *BB);
171
172 /// Processes the collected instructions, the \p Map. The values of \p Map
173 /// should be all loads or all stores.
174 bool vectorizeChains(InstrListMap &Map);
175
176 /// Finds the load/stores to consecutive memory addresses and vectorizes them.
177 bool vectorizeInstructions(ArrayRef<Instruction *> Instrs);
178
179 /// Vectorizes the load instructions in Chain.
180 bool
181 vectorizeLoadChain(ArrayRef<Instruction *> Chain,
182 SmallPtrSet<Instruction *, 16> *InstructionsProcessed);
183
184 /// Vectorizes the store instructions in Chain.
185 bool
186 vectorizeStoreChain(ArrayRef<Instruction *> Chain,
187 SmallPtrSet<Instruction *, 16> *InstructionsProcessed);
188
189 /// Check if this load/store access is misaligned accesses.
190 bool accessIsMisaligned(unsigned SzInBytes, unsigned AddressSpace,
191 unsigned Alignment);
192};
193
194class LoadStoreVectorizer : public FunctionPass {
195public:
196 static char ID;
197
198 LoadStoreVectorizer() : FunctionPass(ID) {
199 initializeLoadStoreVectorizerPass(*PassRegistry::getPassRegistry());
200 }
201
202 bool runOnFunction(Function &F) override;
203
204 StringRef getPassName() const override {
205 return "GPU Load and Store Vectorizer";
206 }
207
208 void getAnalysisUsage(AnalysisUsage &AU) const override {
209 AU.addRequired<AAResultsWrapperPass>();
210 AU.addRequired<ScalarEvolutionWrapperPass>();
211 AU.addRequired<DominatorTreeWrapperPass>();
212 AU.addRequired<TargetTransformInfoWrapperPass>();
213 AU.setPreservesCFG();
214 }
215};
216
217} // end anonymous namespace
218
219char LoadStoreVectorizer::ID = 0;
220
221INITIALIZE_PASS_BEGIN(LoadStoreVectorizer, DEBUG_TYPE,static void *initializeLoadStoreVectorizerPassOnce(PassRegistry
&Registry) {
222 "Vectorize load and Store instructions", false, false)static void *initializeLoadStoreVectorizerPassOnce(PassRegistry
&Registry) {
223INITIALIZE_PASS_DEPENDENCY(SCEVAAWrapperPass)initializeSCEVAAWrapperPassPass(Registry);
224INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)initializeDominatorTreeWrapperPassPass(Registry);
225INITIALIZE_PASS_DEPENDENCY(AAResultsWrapperPass)initializeAAResultsWrapperPassPass(Registry);
226INITIALIZE_PASS_DEPENDENCY(GlobalsAAWrapperPass)initializeGlobalsAAWrapperPassPass(Registry);
227INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)initializeTargetTransformInfoWrapperPassPass(Registry);
228INITIALIZE_PASS_END(LoadStoreVectorizer, DEBUG_TYPE,PassInfo *PI = new PassInfo( "Vectorize load and store instructions"
, "load-store-vectorizer", &LoadStoreVectorizer::ID, PassInfo
::NormalCtor_t(callDefaultCtor<LoadStoreVectorizer>), false
, false); Registry.registerPass(*PI, true); return PI; } static
llvm::once_flag InitializeLoadStoreVectorizerPassFlag; void llvm
::initializeLoadStoreVectorizerPass(PassRegistry &Registry
) { llvm::call_once(InitializeLoadStoreVectorizerPassFlag, initializeLoadStoreVectorizerPassOnce
, std::ref(Registry)); }
229 "Vectorize load and store instructions", false, false)PassInfo *PI = new PassInfo( "Vectorize load and store instructions"
, "load-store-vectorizer", &LoadStoreVectorizer::ID, PassInfo
::NormalCtor_t(callDefaultCtor<LoadStoreVectorizer>), false
, false); Registry.registerPass(*PI, true); return PI; } static
llvm::once_flag InitializeLoadStoreVectorizerPassFlag; void llvm
::initializeLoadStoreVectorizerPass(PassRegistry &Registry
) { llvm::call_once(InitializeLoadStoreVectorizerPassFlag, initializeLoadStoreVectorizerPassOnce
, std::ref(Registry)); }
230
231Pass *llvm::createLoadStoreVectorizerPass() {
232 return new LoadStoreVectorizer();
233}
234
235// The real propagateMetadata expects a SmallVector<Value*>, but we deal in
236// vectors of Instructions.
237static void propagateMetadata(Instruction *I, ArrayRef<Instruction *> IL) {
238 SmallVector<Value *, 8> VL(IL.begin(), IL.end());
239 propagateMetadata(I, VL);
240}
241
242bool LoadStoreVectorizer::runOnFunction(Function &F) {
243 // Don't vectorize when the attribute NoImplicitFloat is used.
244 if (skipFunction(F) || F.hasFnAttribute(Attribute::NoImplicitFloat))
245 return false;
246
247 AliasAnalysis &AA = getAnalysis<AAResultsWrapperPass>().getAAResults();
248 DominatorTree &DT = getAnalysis<DominatorTreeWrapperPass>().getDomTree();
249 ScalarEvolution &SE = getAnalysis<ScalarEvolutionWrapperPass>().getSE();
250 TargetTransformInfo &TTI =
251 getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F);
252
253 Vectorizer V(F, AA, DT, SE, TTI);
254 return V.run();
255}
256
257// Vectorizer Implementation
258bool Vectorizer::run() {
259 bool Changed = false;
260
261 // Scan the blocks in the function in post order.
262 for (BasicBlock *BB : post_order(&F)) {
263 InstrListMap LoadRefs, StoreRefs;
264 std::tie(LoadRefs, StoreRefs) = collectInstructions(BB);
265 Changed |= vectorizeChains(LoadRefs);
266 Changed |= vectorizeChains(StoreRefs);
267 }
268
269 return Changed;
270}
271
272unsigned Vectorizer::getPointerAddressSpace(Value *I) {
273 if (LoadInst *L = dyn_cast<LoadInst>(I))
274 return L->getPointerAddressSpace();
275 if (StoreInst *S = dyn_cast<StoreInst>(I))
276 return S->getPointerAddressSpace();
277 return -1;
278}
279
280GetElementPtrInst *Vectorizer::getSourceGEP(Value *Src) const {
281 // First strip pointer bitcasts. Make sure pointee size is the same with
282 // and without casts.
283 // TODO: a stride set by the add instruction below can match the difference
284 // in pointee type size here. Currently it will not be vectorized.
285 Value *SrcPtr = getLoadStorePointerOperand(Src);
286 Value *SrcBase = SrcPtr->stripPointerCasts();
287 if (DL.getTypeStoreSize(SrcPtr->getType()->getPointerElementType()) ==
288 DL.getTypeStoreSize(SrcBase->getType()->getPointerElementType()))
289 SrcPtr = SrcBase;
290 return dyn_cast<GetElementPtrInst>(SrcPtr);
291}
292
293// FIXME: Merge with llvm::isConsecutiveAccess
294bool Vectorizer::isConsecutiveAccess(Value *A, Value *B) {
295 Value *PtrA = getLoadStorePointerOperand(A);
296 Value *PtrB = getLoadStorePointerOperand(B);
297 unsigned ASA = getPointerAddressSpace(A);
298 unsigned ASB = getPointerAddressSpace(B);
299
300 // Check that the address spaces match and that the pointers are valid.
301 if (!PtrA || !PtrB || (ASA != ASB))
302 return false;
303
304 // Make sure that A and B are different pointers of the same size type.
305 unsigned PtrBitWidth = DL.getPointerSizeInBits(ASA);
306 Type *PtrATy = PtrA->getType()->getPointerElementType();
307 Type *PtrBTy = PtrB->getType()->getPointerElementType();
308 if (PtrA == PtrB ||
309 PtrATy->isVectorTy() != PtrBTy->isVectorTy() ||
310 DL.getTypeStoreSize(PtrATy) != DL.getTypeStoreSize(PtrBTy) ||
311 DL.getTypeStoreSize(PtrATy->getScalarType()) !=
312 DL.getTypeStoreSize(PtrBTy->getScalarType()))
313 return false;
314
315 APInt Size(PtrBitWidth, DL.getTypeStoreSize(PtrATy));
316
317 unsigned IdxWidth = DL.getIndexSizeInBits(ASA);
318 APInt OffsetA(IdxWidth, 0), OffsetB(IdxWidth, 0);
319 PtrA = PtrA->stripAndAccumulateInBoundsConstantOffsets(DL, OffsetA);
320 PtrB = PtrB->stripAndAccumulateInBoundsConstantOffsets(DL, OffsetB);
321
322 APInt OffsetDelta = OffsetB - OffsetA;
323
324 // Check if they are based on the same pointer. That makes the offsets
325 // sufficient.
326 if (PtrA == PtrB)
327 return OffsetDelta == Size;
328
329 // Compute the necessary base pointer delta to have the necessary final delta
330 // equal to the size.
331 APInt BaseDelta = Size - OffsetDelta;
332
333 // Compute the distance with SCEV between the base pointers.
334 const SCEV *PtrSCEVA = SE.getSCEV(PtrA);
335 const SCEV *PtrSCEVB = SE.getSCEV(PtrB);
336 const SCEV *C = SE.getConstant(BaseDelta);
337 const SCEV *X = SE.getAddExpr(PtrSCEVA, C);
338 if (X == PtrSCEVB)
339 return true;
340
341 // Sometimes even this doesn't work, because SCEV can't always see through
342 // patterns that look like (gep (ext (add (shl X, C1), C2))). Try checking
343 // things the hard way.
344
345 // Look through GEPs after checking they're the same except for the last
346 // index.
347 GetElementPtrInst *GEPA = getSourceGEP(A);
348 GetElementPtrInst *GEPB = getSourceGEP(B);
349 if (!GEPA || !GEPB || GEPA->getNumOperands() != GEPB->getNumOperands())
350 return false;
351 unsigned FinalIndex = GEPA->getNumOperands() - 1;
352 for (unsigned i = 0; i < FinalIndex; i++)
353 if (GEPA->getOperand(i) != GEPB->getOperand(i))
354 return false;
355
356 Instruction *OpA = dyn_cast<Instruction>(GEPA->getOperand(FinalIndex));
357 Instruction *OpB = dyn_cast<Instruction>(GEPB->getOperand(FinalIndex));
358 if (!OpA || !OpB || OpA->getOpcode() != OpB->getOpcode() ||
359 OpA->getType() != OpB->getType())
360 return false;
361
362 // Only look through a ZExt/SExt.
363 if (!isa<SExtInst>(OpA) && !isa<ZExtInst>(OpA))
364 return false;
365
366 bool Signed = isa<SExtInst>(OpA);
367
368 OpA = dyn_cast<Instruction>(OpA->getOperand(0));
369 OpB = dyn_cast<Instruction>(OpB->getOperand(0));
370 if (!OpA || !OpB || OpA->getType() != OpB->getType())
371 return false;
372
373 // Now we need to prove that adding 1 to OpA won't overflow.
374 bool Safe = false;
375 // First attempt: if OpB is an add with NSW/NUW, and OpB is 1 added to OpA,
376 // we're okay.
377 if (OpB->getOpcode() == Instruction::Add &&
378 isa<ConstantInt>(OpB->getOperand(1)) &&
379 cast<ConstantInt>(OpB->getOperand(1))->getSExtValue() > 0) {
380 if (Signed)
381 Safe = cast<BinaryOperator>(OpB)->hasNoSignedWrap();
382 else
383 Safe = cast<BinaryOperator>(OpB)->hasNoUnsignedWrap();
384 }
385
386 unsigned BitWidth = OpA->getType()->getScalarSizeInBits();
387
388 // Second attempt:
389 // If any bits are known to be zero other than the sign bit in OpA, we can
390 // add 1 to it while guaranteeing no overflow of any sort.
391 if (!Safe) {
392 KnownBits Known(BitWidth);
393 computeKnownBits(OpA, Known, DL, 0, nullptr, OpA, &DT);
394 if (Known.countMaxTrailingOnes() < (BitWidth - 1))
395 Safe = true;
396 }
397
398 if (!Safe)
399 return false;
400
401 const SCEV *OffsetSCEVA = SE.getSCEV(OpA);
402 const SCEV *OffsetSCEVB = SE.getSCEV(OpB);
403 const SCEV *One = SE.getConstant(APInt(BitWidth, 1));
404 const SCEV *X2 = SE.getAddExpr(OffsetSCEVA, One);
405 return X2 == OffsetSCEVB;
406}
407
408void Vectorizer::reorder(Instruction *I) {
409 OrderedBasicBlock OBB(I->getParent());
410 SmallPtrSet<Instruction *, 16> InstructionsToMove;
411 SmallVector<Instruction *, 16> Worklist;
412
413 Worklist.push_back(I);
414 while (!Worklist.empty()) {
415 Instruction *IW = Worklist.pop_back_val();
416 int NumOperands = IW->getNumOperands();
417 for (int i = 0; i < NumOperands; i++) {
418 Instruction *IM = dyn_cast<Instruction>(IW->getOperand(i));
419 if (!IM || IM->getOpcode() == Instruction::PHI)
420 continue;
421
422 // If IM is in another BB, no need to move it, because this pass only
423 // vectorizes instructions within one BB.
424 if (IM->getParent() != I->getParent())
425 continue;
426
427 if (!OBB.dominates(IM, I)) {
428 InstructionsToMove.insert(IM);
429 Worklist.push_back(IM);
430 }
431 }
432 }
433
434 // All instructions to move should follow I. Start from I, not from begin().
435 for (auto BBI = I->getIterator(), E = I->getParent()->end(); BBI != E;
436 ++BBI) {
437 if (!InstructionsToMove.count(&*BBI))
438 continue;
439 Instruction *IM = &*BBI;
440 --BBI;
441 IM->removeFromParent();
442 IM->insertBefore(I);
443 }
444}
445
446std::pair<BasicBlock::iterator, BasicBlock::iterator>
447Vectorizer::getBoundaryInstrs(ArrayRef<Instruction *> Chain) {
448 Instruction *C0 = Chain[0];
449 BasicBlock::iterator FirstInstr = C0->getIterator();
450 BasicBlock::iterator LastInstr = C0->getIterator();
451
452 BasicBlock *BB = C0->getParent();
453 unsigned NumFound = 0;
454 for (Instruction &I : *BB) {
455 if (!is_contained(Chain, &I))
456 continue;
457
458 ++NumFound;
459 if (NumFound == 1) {
460 FirstInstr = I.getIterator();
461 }
462 if (NumFound == Chain.size()) {
463 LastInstr = I.getIterator();
464 break;
465 }
466 }
467
468 // Range is [first, last).
469 return std::make_pair(FirstInstr, ++LastInstr);
470}
471
472void Vectorizer::eraseInstructions(ArrayRef<Instruction *> Chain) {
473 SmallVector<Instruction *, 16> Instrs;
474 for (Instruction *I : Chain) {
475 Value *PtrOperand = getLoadStorePointerOperand(I);
476 assert(PtrOperand && "Instruction must have a pointer operand.")(static_cast <bool> (PtrOperand && "Instruction must have a pointer operand."
) ? void (0) : __assert_fail ("PtrOperand && \"Instruction must have a pointer operand.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 476, __extension__ __PRETTY_FUNCTION__))
;
477 Instrs.push_back(I);
478 if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(PtrOperand))
479 Instrs.push_back(GEP);
480 }
481
482 // Erase instructions.
483 for (Instruction *I : Instrs)
484 if (I->use_empty())
485 I->eraseFromParent();
486}
487
488std::pair<ArrayRef<Instruction *>, ArrayRef<Instruction *>>
489Vectorizer::splitOddVectorElts(ArrayRef<Instruction *> Chain,
490 unsigned ElementSizeBits) {
491 unsigned ElementSizeBytes = ElementSizeBits / 8;
492 unsigned SizeBytes = ElementSizeBytes * Chain.size();
493 unsigned NumLeft = (SizeBytes - (SizeBytes % 4)) / ElementSizeBytes;
494 if (NumLeft == Chain.size()) {
495 if ((NumLeft & 1) == 0)
496 NumLeft /= 2; // Split even in half
497 else
498 --NumLeft; // Split off last element
499 } else if (NumLeft == 0)
500 NumLeft = 1;
501 return std::make_pair(Chain.slice(0, NumLeft), Chain.slice(NumLeft));
502}
503
504ArrayRef<Instruction *>
505Vectorizer::getVectorizablePrefix(ArrayRef<Instruction *> Chain) {
506 // These are in BB order, unlike Chain, which is in address order.
507 SmallVector<Instruction *, 16> MemoryInstrs;
508 SmallVector<Instruction *, 16> ChainInstrs;
509
510 bool IsLoadChain = isa<LoadInst>(Chain[0]);
511 DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { for (Instruction *I : Chain) {
if (IsLoadChain) (static_cast <bool> (isa<LoadInst>
(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<LoadInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 515, __extension__ __PRETTY_FUNCTION__)); else (static_cast
<bool> (isa<StoreInst>(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<StoreInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 518, __extension__ __PRETTY_FUNCTION__)); } }; } } while (false
)
512 for (Instruction *I : Chain) {do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { for (Instruction *I : Chain) {
if (IsLoadChain) (static_cast <bool> (isa<LoadInst>
(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<LoadInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 515, __extension__ __PRETTY_FUNCTION__)); else (static_cast
<bool> (isa<StoreInst>(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<StoreInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 518, __extension__ __PRETTY_FUNCTION__)); } }; } } while (false
)
513 if (IsLoadChain)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { for (Instruction *I : Chain) {
if (IsLoadChain) (static_cast <bool> (isa<LoadInst>
(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<LoadInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 515, __extension__ __PRETTY_FUNCTION__)); else (static_cast
<bool> (isa<StoreInst>(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<StoreInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 518, __extension__ __PRETTY_FUNCTION__)); } }; } } while (false
)
514 assert(isa<LoadInst>(I) &&do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { for (Instruction *I : Chain) {
if (IsLoadChain) (static_cast <bool> (isa<LoadInst>
(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<LoadInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 515, __extension__ __PRETTY_FUNCTION__)); else (static_cast
<bool> (isa<StoreInst>(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<StoreInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 518, __extension__ __PRETTY_FUNCTION__)); } }; } } while (false
)
515 "All elements of Chain must be loads, or all must be stores.");do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { for (Instruction *I : Chain) {
if (IsLoadChain) (static_cast <bool> (isa<LoadInst>
(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<LoadInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 515, __extension__ __PRETTY_FUNCTION__)); else (static_cast
<bool> (isa<StoreInst>(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<StoreInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 518, __extension__ __PRETTY_FUNCTION__)); } }; } } while (false
)
516 elsedo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { for (Instruction *I : Chain) {
if (IsLoadChain) (static_cast <bool> (isa<LoadInst>
(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<LoadInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 515, __extension__ __PRETTY_FUNCTION__)); else (static_cast
<bool> (isa<StoreInst>(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<StoreInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 518, __extension__ __PRETTY_FUNCTION__)); } }; } } while (false
)
517 assert(isa<StoreInst>(I) &&do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { for (Instruction *I : Chain) {
if (IsLoadChain) (static_cast <bool> (isa<LoadInst>
(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<LoadInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 515, __extension__ __PRETTY_FUNCTION__)); else (static_cast
<bool> (isa<StoreInst>(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<StoreInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 518, __extension__ __PRETTY_FUNCTION__)); } }; } } while (false
)
518 "All elements of Chain must be loads, or all must be stores.");do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { for (Instruction *I : Chain) {
if (IsLoadChain) (static_cast <bool> (isa<LoadInst>
(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<LoadInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 515, __extension__ __PRETTY_FUNCTION__)); else (static_cast
<bool> (isa<StoreInst>(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<StoreInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 518, __extension__ __PRETTY_FUNCTION__)); } }; } } while (false
)
519 }do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { for (Instruction *I : Chain) {
if (IsLoadChain) (static_cast <bool> (isa<LoadInst>
(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<LoadInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 515, __extension__ __PRETTY_FUNCTION__)); else (static_cast
<bool> (isa<StoreInst>(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<StoreInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 518, __extension__ __PRETTY_FUNCTION__)); } }; } } while (false
)
520 })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { for (Instruction *I : Chain) {
if (IsLoadChain) (static_cast <bool> (isa<LoadInst>
(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<LoadInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 515, __extension__ __PRETTY_FUNCTION__)); else (static_cast
<bool> (isa<StoreInst>(I) && "All elements of Chain must be loads, or all must be stores."
) ? void (0) : __assert_fail ("isa<StoreInst>(I) && \"All elements of Chain must be loads, or all must be stores.\""
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 518, __extension__ __PRETTY_FUNCTION__)); } }; } } while (false
)
;
521
522 for (Instruction &I : make_range(getBoundaryInstrs(Chain))) {
523 if (isa<LoadInst>(I) || isa<StoreInst>(I)) {
524 if (!is_contained(Chain, &I))
525 MemoryInstrs.push_back(&I);
526 else
527 ChainInstrs.push_back(&I);
528 } else if (isa<IntrinsicInst>(&I) &&
529 cast<IntrinsicInst>(&I)->getIntrinsicID() ==
530 Intrinsic::sideeffect) {
531 // Ignore llvm.sideeffect calls.
532 } else if (IsLoadChain && (I.mayWriteToMemory() || I.mayThrow())) {
533 DEBUG(dbgs() << "LSV: Found may-write/throw operation: " << I << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { dbgs() << "LSV: Found may-write/throw operation: "
<< I << '\n'; } } while (false)
;
534 break;
535 } else if (!IsLoadChain && (I.mayReadOrWriteMemory() || I.mayThrow())) {
536 DEBUG(dbgs() << "LSV: Found may-read/write/throw operation: " << Ido { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { dbgs() << "LSV: Found may-read/write/throw operation: "
<< I << '\n'; } } while (false)
537 << '\n')do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { dbgs() << "LSV: Found may-read/write/throw operation: "
<< I << '\n'; } } while (false)
;
538 break;
539 }
540 }
541
542 OrderedBasicBlock OBB(Chain[0]->getParent());
543
544 // Loop until we find an instruction in ChainInstrs that we can't vectorize.
545 unsigned ChainInstrIdx = 0;
546 Instruction *BarrierMemoryInstr = nullptr;
547
548 for (unsigned E = ChainInstrs.size(); ChainInstrIdx < E; ++ChainInstrIdx) {
549 Instruction *ChainInstr = ChainInstrs[ChainInstrIdx];
550
551 // If a barrier memory instruction was found, chain instructions that follow
552 // will not be added to the valid prefix.
553 if (BarrierMemoryInstr && OBB.dominates(BarrierMemoryInstr, ChainInstr))
554 break;
555
556 // Check (in BB order) if any instruction prevents ChainInstr from being
557 // vectorized. Find and store the first such "conflicting" instruction.
558 for (Instruction *MemInstr : MemoryInstrs) {
559 // If a barrier memory instruction was found, do not check past it.
560 if (BarrierMemoryInstr && OBB.dominates(BarrierMemoryInstr, MemInstr))
561 break;
562
563 if (isa<LoadInst>(MemInstr) && isa<LoadInst>(ChainInstr))
564 continue;
565
566 // We can ignore the alias as long as the load comes before the store,
567 // because that means we won't be moving the load past the store to
568 // vectorize it (the vectorized load is inserted at the location of the
569 // first load in the chain).
570 if (isa<StoreInst>(MemInstr) && isa<LoadInst>(ChainInstr) &&
571 OBB.dominates(ChainInstr, MemInstr))
572 continue;
573
574 // Same case, but in reverse.
575 if (isa<LoadInst>(MemInstr) && isa<StoreInst>(ChainInstr) &&
576 OBB.dominates(MemInstr, ChainInstr))
577 continue;
578
579 if (!AA.isNoAlias(MemoryLocation::get(MemInstr),
580 MemoryLocation::get(ChainInstr))) {
581 DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Found alias:\n"
" Aliasing instruction and pointer:\n" << " " <<
*MemInstr << '\n' << " " << *getLoadStorePointerOperand
(MemInstr) << '\n' << " Aliased instruction and pointer:\n"
<< " " << *ChainInstr << '\n' << " "
<< *getLoadStorePointerOperand(ChainInstr) << '\n'
; }; } } while (false)
582 dbgs() << "LSV: Found alias:\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Found alias:\n"
" Aliasing instruction and pointer:\n" << " " <<
*MemInstr << '\n' << " " << *getLoadStorePointerOperand
(MemInstr) << '\n' << " Aliased instruction and pointer:\n"
<< " " << *ChainInstr << '\n' << " "
<< *getLoadStorePointerOperand(ChainInstr) << '\n'
; }; } } while (false)
583 " Aliasing instruction and pointer:\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Found alias:\n"
" Aliasing instruction and pointer:\n" << " " <<
*MemInstr << '\n' << " " << *getLoadStorePointerOperand
(MemInstr) << '\n' << " Aliased instruction and pointer:\n"
<< " " << *ChainInstr << '\n' << " "
<< *getLoadStorePointerOperand(ChainInstr) << '\n'
; }; } } while (false)
584 << " " << *MemInstr << '\n'do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Found alias:\n"
" Aliasing instruction and pointer:\n" << " " <<
*MemInstr << '\n' << " " << *getLoadStorePointerOperand
(MemInstr) << '\n' << " Aliased instruction and pointer:\n"
<< " " << *ChainInstr << '\n' << " "
<< *getLoadStorePointerOperand(ChainInstr) << '\n'
; }; } } while (false)
585 << " " << *getLoadStorePointerOperand(MemInstr) << '\n'do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Found alias:\n"
" Aliasing instruction and pointer:\n" << " " <<
*MemInstr << '\n' << " " << *getLoadStorePointerOperand
(MemInstr) << '\n' << " Aliased instruction and pointer:\n"
<< " " << *ChainInstr << '\n' << " "
<< *getLoadStorePointerOperand(ChainInstr) << '\n'
; }; } } while (false)
586 << " Aliased instruction and pointer:\n"do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Found alias:\n"
" Aliasing instruction and pointer:\n" << " " <<
*MemInstr << '\n' << " " << *getLoadStorePointerOperand
(MemInstr) << '\n' << " Aliased instruction and pointer:\n"
<< " " << *ChainInstr << '\n' << " "
<< *getLoadStorePointerOperand(ChainInstr) << '\n'
; }; } } while (false)
587 << " " << *ChainInstr << '\n'do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Found alias:\n"
" Aliasing instruction and pointer:\n" << " " <<
*MemInstr << '\n' << " " << *getLoadStorePointerOperand
(MemInstr) << '\n' << " Aliased instruction and pointer:\n"
<< " " << *ChainInstr << '\n' << " "
<< *getLoadStorePointerOperand(ChainInstr) << '\n'
; }; } } while (false)
588 << " " << *getLoadStorePointerOperand(ChainInstr) << '\n';do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Found alias:\n"
" Aliasing instruction and pointer:\n" << " " <<
*MemInstr << '\n' << " " << *getLoadStorePointerOperand
(MemInstr) << '\n' << " Aliased instruction and pointer:\n"
<< " " << *ChainInstr << '\n' << " "
<< *getLoadStorePointerOperand(ChainInstr) << '\n'
; }; } } while (false)
589 })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Found alias:\n"
" Aliasing instruction and pointer:\n" << " " <<
*MemInstr << '\n' << " " << *getLoadStorePointerOperand
(MemInstr) << '\n' << " Aliased instruction and pointer:\n"
<< " " << *ChainInstr << '\n' << " "
<< *getLoadStorePointerOperand(ChainInstr) << '\n'
; }; } } while (false)
;
590 // Save this aliasing memory instruction as a barrier, but allow other
591 // instructions that precede the barrier to be vectorized with this one.
592 BarrierMemoryInstr = MemInstr;
593 break;
594 }
595 }
596 // Continue the search only for store chains, since vectorizing stores that
597 // precede an aliasing load is valid. Conversely, vectorizing loads is valid
598 // up to an aliasing store, but should not pull loads from further down in
599 // the basic block.
600 if (IsLoadChain && BarrierMemoryInstr) {
601 // The BarrierMemoryInstr is a store that precedes ChainInstr.
602 assert(OBB.dominates(BarrierMemoryInstr, ChainInstr))(static_cast <bool> (OBB.dominates(BarrierMemoryInstr, ChainInstr
)) ? void (0) : __assert_fail ("OBB.dominates(BarrierMemoryInstr, ChainInstr)"
, "/build/llvm-toolchain-snapshot-7~svn329677/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp"
, 602, __extension__ __PRETTY_FUNCTION__))
;
603 break;
604 }
605 }
606
607 // Find the largest prefix of Chain whose elements are all in
608 // ChainInstrs[0, ChainInstrIdx). This is the largest vectorizable prefix of
609 // Chain. (Recall that Chain is in address order, but ChainInstrs is in BB
610 // order.)
611 SmallPtrSet<Instruction *, 8> VectorizableChainInstrs(
612 ChainInstrs.begin(), ChainInstrs.begin() + ChainInstrIdx);
613 unsigned ChainIdx = 0;
614 for (unsigned ChainLen = Chain.size(); ChainIdx < ChainLen; ++ChainIdx) {
615 if (!VectorizableChainInstrs.count(Chain[ChainIdx]))
616 break;
617 }
618 return Chain.slice(0, ChainIdx);
619}
620
621std::pair<InstrListMap, InstrListMap>
622Vectorizer::collectInstructions(BasicBlock *BB) {
623 InstrListMap LoadRefs;
624 InstrListMap StoreRefs;
625
626 for (Instruction &I : *BB) {
627 if (!I.mayReadOrWriteMemory())
628 continue;
629
630 if (LoadInst *LI = dyn_cast<LoadInst>(&I)) {
631 if (!LI->isSimple())
632 continue;
633
634 // Skip if it's not legal.
635 if (!TTI.isLegalToVectorizeLoad(LI))
636 continue;
637
638 Type *Ty = LI->getType();
639 if (!VectorType::isValidElementType(Ty->getScalarType()))
640 continue;
641
642 // Skip weird non-byte sizes. They probably aren't worth the effort of
643 // handling correctly.
644 unsigned TySize = DL.getTypeSizeInBits(Ty);
645 if ((TySize % 8) != 0)
646 continue;
647
648 // Skip vectors of pointers. The vectorizeLoadChain/vectorizeStoreChain
649 // functions are currently using an integer type for the vectorized
650 // load/store, and does not support casting between the integer type and a
651 // vector of pointers (e.g. i64 to <2 x i16*>)
652 if (Ty->isVectorTy() && Ty->isPtrOrPtrVectorTy())
653 continue;
654
655 Value *Ptr = LI->getPointerOperand();
656 unsigned AS = Ptr->getType()->getPointerAddressSpace();
657 unsigned VecRegSize = TTI.getLoadStoreVecRegBitWidth(AS);
658
659 unsigned VF = VecRegSize / TySize;
660 VectorType *VecTy = dyn_cast<VectorType>(Ty);
661
662 // No point in looking at these if they're too big to vectorize.
663 if (TySize > VecRegSize / 2 ||
664 (VecTy && TTI.getLoadVectorFactor(VF, TySize, TySize / 8, VecTy) == 0))
665 continue;
666
667 // Make sure all the users of a vector are constant-index extracts.
668 if (isa<VectorType>(Ty) && !llvm::all_of(LI->users(), [](const User *U) {
669 const ExtractElementInst *EEI = dyn_cast<ExtractElementInst>(U);
670 return EEI && isa<ConstantInt>(EEI->getOperand(1));
671 }))
672 continue;
673
674 // Save the load locations.
675 Value *ObjPtr = GetUnderlyingObject(Ptr, DL);
676 LoadRefs[ObjPtr].push_back(LI);
677 } else if (StoreInst *SI = dyn_cast<StoreInst>(&I)) {
678 if (!SI->isSimple())
679 continue;
680
681 // Skip if it's not legal.
682 if (!TTI.isLegalToVectorizeStore(SI))
683 continue;
684
685 Type *Ty = SI->getValueOperand()->getType();
686 if (!VectorType::isValidElementType(Ty->getScalarType()))
687 continue;
688
689 // Skip vectors of pointers. The vectorizeLoadChain/vectorizeStoreChain
690 // functions are currently using an integer type for the vectorized
691 // load/store, and does not support casting between the integer type and a
692 // vector of pointers (e.g. i64 to <2 x i16*>)
693 if (Ty->isVectorTy() && Ty->isPtrOrPtrVectorTy())
694 continue;
695
696 // Skip weird non-byte sizes. They probably aren't worth the effort of
697 // handling correctly.
698 unsigned TySize = DL.getTypeSizeInBits(Ty);
699 if ((TySize % 8) != 0)
700 continue;
701
702 Value *Ptr = SI->getPointerOperand();
703 unsigned AS = Ptr->getType()->getPointerAddressSpace();
704 unsigned VecRegSize = TTI.getLoadStoreVecRegBitWidth(AS);
705
706 unsigned VF = VecRegSize / TySize;
707 VectorType *VecTy = dyn_cast<VectorType>(Ty);
708
709 // No point in looking at these if they're too big to vectorize.
710 if (TySize > VecRegSize / 2 ||
711 (VecTy && TTI.getStoreVectorFactor(VF, TySize, TySize / 8, VecTy) == 0))
712 continue;
713
714 if (isa<VectorType>(Ty) && !llvm::all_of(SI->users(), [](const User *U) {
715 const ExtractElementInst *EEI = dyn_cast<ExtractElementInst>(U);
716 return EEI && isa<ConstantInt>(EEI->getOperand(1));
717 }))
718 continue;
719
720 // Save store location.
721 Value *ObjPtr = GetUnderlyingObject(Ptr, DL);
722 StoreRefs[ObjPtr].push_back(SI);
723 }
724 }
725
726 return {LoadRefs, StoreRefs};
727}
728
729bool Vectorizer::vectorizeChains(InstrListMap &Map) {
730 bool Changed = false;
731
732 for (const std::pair<Value *, InstrList> &Chain : Map) {
733 unsigned Size = Chain.second.size();
734 if (Size < 2)
1
Assuming 'Size' is >= 2
2
Taking false branch
735 continue;
736
737 DEBUG(dbgs() << "LSV: Analyzing a chain of length " << Size << ".\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { dbgs() << "LSV: Analyzing a chain of length "
<< Size << ".\n"; } } while (false)
;
738
739 // Process the stores in chunks of 64.
740 for (unsigned CI = 0, CE = Size; CI < CE; CI += 64) {
3
Loop condition is true. Entering loop body
741 unsigned Len = std::min<unsigned>(CE - CI, 64);
742 ArrayRef<Instruction *> Chunk(&Chain.second[CI], Len);
743 Changed |= vectorizeInstructions(Chunk);
4
Calling 'Vectorizer::vectorizeInstructions'
744 }
745 }
746
747 return Changed;
748}
749
750bool Vectorizer::vectorizeInstructions(ArrayRef<Instruction *> Instrs) {
751 DEBUG(dbgs() << "LSV: Vectorizing " << Instrs.size() << " instructions.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { dbgs() << "LSV: Vectorizing "
<< Instrs.size() << " instructions.\n"; } } while
(false)
;
752 SmallVector<int, 16> Heads, Tails;
753 int ConsecutiveChain[64];
754
755 // Do a quadratic search on all of the given loads/stores and find all of the
756 // pairs of loads/stores that follow each other.
757 for (int i = 0, e = Instrs.size(); i < e; ++i) {
5
Assuming 'i' is >= 'e'
6
Loop condition is false. Execution continues on line 778
758 ConsecutiveChain[i] = -1;
759 for (int j = e - 1; j >= 0; --j) {
760 if (i == j)
761 continue;
762
763 if (isConsecutiveAccess(Instrs[i], Instrs[j])) {
764 if (ConsecutiveChain[i] != -1) {
765 int CurDistance = std::abs(ConsecutiveChain[i] - i);
766 int NewDistance = std::abs(ConsecutiveChain[i] - j);
767 if (j < i || NewDistance > CurDistance)
768 continue; // Should not insert.
769 }
770
771 Tails.push_back(j);
772 Heads.push_back(i);
773 ConsecutiveChain[i] = j;
774 }
775 }
776 }
777
778 bool Changed = false;
779 SmallPtrSet<Instruction *, 16> InstructionsProcessed;
780
781 for (int Head : Heads) {
7
Assuming '__begin1' is not equal to '__end1'
782 if (InstructionsProcessed.count(Instrs[Head]))
8
Assuming the condition is false
9
Taking false branch
783 continue;
784 bool LongerChainExists = false;
785 for (unsigned TIt = 0; TIt < Tails.size(); TIt++)
10
Assuming the condition is false
11
Loop condition is false. Execution continues on line 791
786 if (Head == Tails[TIt] &&
787 !InstructionsProcessed.count(Instrs[Heads[TIt]])) {
788 LongerChainExists = true;
789 break;
790 }
791 if (LongerChainExists)
12
Taking false branch
792 continue;
793
794 // We found an instr that starts a chain. Now follow the chain and try to
795 // vectorize it.
796 SmallVector<Instruction *, 16> Operands;
797 int I = Head;
798 while (I != -1 && (is_contained(Tails, I) || is_contained(Heads, I))) {
13
Assuming the condition is false
799 if (InstructionsProcessed.count(Instrs[I]))
800 break;
801
802 Operands.push_back(Instrs[I]);
803 I = ConsecutiveChain[I];
804 }
805
806 bool Vectorized = false;
807 if (isa<LoadInst>(*Operands.begin()))
14
Taking true branch
808 Vectorized = vectorizeLoadChain(Operands, &InstructionsProcessed);
15
Calling 'Vectorizer::vectorizeLoadChain'
809 else
810 Vectorized = vectorizeStoreChain(Operands, &InstructionsProcessed);
811
812 Changed |= Vectorized;
813 }
814
815 return Changed;
816}
817
818bool Vectorizer::vectorizeStoreChain(
819 ArrayRef<Instruction *> Chain,
820 SmallPtrSet<Instruction *, 16> *InstructionsProcessed) {
821 StoreInst *S0 = cast<StoreInst>(Chain[0]);
822
823 // If the vector has an int element, default to int for the whole store.
824 Type *StoreTy;
825 for (Instruction *I : Chain) {
826 StoreTy = cast<StoreInst>(I)->getValueOperand()->getType();
827 if (StoreTy->isIntOrIntVectorTy())
828 break;
829
830 if (StoreTy->isPtrOrPtrVectorTy()) {
831 StoreTy = Type::getIntNTy(F.getParent()->getContext(),
832 DL.getTypeSizeInBits(StoreTy));
833 break;
834 }
835 }
836
837 unsigned Sz = DL.getTypeSizeInBits(StoreTy);
838 unsigned AS = S0->getPointerAddressSpace();
839 unsigned VecRegSize = TTI.getLoadStoreVecRegBitWidth(AS);
840 unsigned VF = VecRegSize / Sz;
841 unsigned ChainSize = Chain.size();
842 unsigned Alignment = getAlignment(S0);
843
844 if (!isPowerOf2_32(Sz) || VF < 2 || ChainSize < 2) {
845 InstructionsProcessed->insert(Chain.begin(), Chain.end());
846 return false;
847 }
848
849 ArrayRef<Instruction *> NewChain = getVectorizablePrefix(Chain);
850 if (NewChain.empty()) {
851 // No vectorization possible.
852 InstructionsProcessed->insert(Chain.begin(), Chain.end());
853 return false;
854 }
855 if (NewChain.size() == 1) {
856 // Failed after the first instruction. Discard it and try the smaller chain.
857 InstructionsProcessed->insert(NewChain.front());
858 return false;
859 }
860
861 // Update Chain to the valid vectorizable subchain.
862 Chain = NewChain;
863 ChainSize = Chain.size();
864
865 // Check if it's legal to vectorize this chain. If not, split the chain and
866 // try again.
867 unsigned EltSzInBytes = Sz / 8;
868 unsigned SzInBytes = EltSzInBytes * ChainSize;
869 if (!TTI.isLegalToVectorizeStoreChain(SzInBytes, Alignment, AS)) {
870 auto Chains = splitOddVectorElts(Chain, Sz);
871 return vectorizeStoreChain(Chains.first, InstructionsProcessed) |
872 vectorizeStoreChain(Chains.second, InstructionsProcessed);
873 }
874
875 VectorType *VecTy;
876 VectorType *VecStoreTy = dyn_cast<VectorType>(StoreTy);
877 if (VecStoreTy)
878 VecTy = VectorType::get(StoreTy->getScalarType(),
879 Chain.size() * VecStoreTy->getNumElements());
880 else
881 VecTy = VectorType::get(StoreTy, Chain.size());
882
883 // If it's more than the max vector size or the target has a better
884 // vector factor, break it into two pieces.
885 unsigned TargetVF = TTI.getStoreVectorFactor(VF, Sz, SzInBytes, VecTy);
886 if (ChainSize > VF || (VF != TargetVF && TargetVF < ChainSize)) {
887 DEBUG(dbgs() << "LSV: Chain doesn't match with the vector factor."do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { dbgs() << "LSV: Chain doesn't match with the vector factor."
" Creating two separate arrays.\n"; } } while (false)
888 " Creating two separate arrays.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { dbgs() << "LSV: Chain doesn't match with the vector factor."
" Creating two separate arrays.\n"; } } while (false)
;
889 return vectorizeStoreChain(Chain.slice(0, TargetVF),
890 InstructionsProcessed) |
891 vectorizeStoreChain(Chain.slice(TargetVF), InstructionsProcessed);
892 }
893
894 DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Stores to vectorize:\n"
; for (Instruction *I : Chain) dbgs() << " " << *
I << "\n"; }; } } while (false)
895 dbgs() << "LSV: Stores to vectorize:\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Stores to vectorize:\n"
; for (Instruction *I : Chain) dbgs() << " " << *
I << "\n"; }; } } while (false)
896 for (Instruction *I : Chain)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Stores to vectorize:\n"
; for (Instruction *I : Chain) dbgs() << " " << *
I << "\n"; }; } } while (false)
897 dbgs() << " " << *I << "\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Stores to vectorize:\n"
; for (Instruction *I : Chain) dbgs() << " " << *
I << "\n"; }; } } while (false)
898 })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Stores to vectorize:\n"
; for (Instruction *I : Chain) dbgs() << " " << *
I << "\n"; }; } } while (false)
;
899
900 // We won't try again to vectorize the elements of the chain, regardless of
901 // whether we succeed below.
902 InstructionsProcessed->insert(Chain.begin(), Chain.end());
903
904 // If the store is going to be misaligned, don't vectorize it.
905 if (accessIsMisaligned(SzInBytes, AS, Alignment)) {
906 if (S0->getPointerAddressSpace() != 0)
907 return false;
908
909 unsigned NewAlign = getOrEnforceKnownAlignment(S0->getPointerOperand(),
910 StackAdjustedAlignment,
911 DL, S0, nullptr, &DT);
912 if (NewAlign < StackAdjustedAlignment)
913 return false;
914 }
915
916 BasicBlock::iterator First, Last;
917 std::tie(First, Last) = getBoundaryInstrs(Chain);
918 Builder.SetInsertPoint(&*Last);
919
920 Value *Vec = UndefValue::get(VecTy);
921
922 if (VecStoreTy) {
923 unsigned VecWidth = VecStoreTy->getNumElements();
924 for (unsigned I = 0, E = Chain.size(); I != E; ++I) {
925 StoreInst *Store = cast<StoreInst>(Chain[I]);
926 for (unsigned J = 0, NE = VecStoreTy->getNumElements(); J != NE; ++J) {
927 unsigned NewIdx = J + I * VecWidth;
928 Value *Extract = Builder.CreateExtractElement(Store->getValueOperand(),
929 Builder.getInt32(J));
930 if (Extract->getType() != StoreTy->getScalarType())
931 Extract = Builder.CreateBitCast(Extract, StoreTy->getScalarType());
932
933 Value *Insert =
934 Builder.CreateInsertElement(Vec, Extract, Builder.getInt32(NewIdx));
935 Vec = Insert;
936 }
937 }
938 } else {
939 for (unsigned I = 0, E = Chain.size(); I != E; ++I) {
940 StoreInst *Store = cast<StoreInst>(Chain[I]);
941 Value *Extract = Store->getValueOperand();
942 if (Extract->getType() != StoreTy->getScalarType())
943 Extract =
944 Builder.CreateBitOrPointerCast(Extract, StoreTy->getScalarType());
945
946 Value *Insert =
947 Builder.CreateInsertElement(Vec, Extract, Builder.getInt32(I));
948 Vec = Insert;
949 }
950 }
951
952 // This cast is safe because Builder.CreateStore() always creates a bona fide
953 // StoreInst.
954 StoreInst *SI = cast<StoreInst>(
955 Builder.CreateStore(Vec, Builder.CreateBitCast(S0->getPointerOperand(),
956 VecTy->getPointerTo(AS))));
957 propagateMetadata(SI, Chain);
958 SI->setAlignment(Alignment);
959
960 eraseInstructions(Chain);
961 ++NumVectorInstructions;
962 NumScalarsVectorized += Chain.size();
963 return true;
964}
965
966bool Vectorizer::vectorizeLoadChain(
967 ArrayRef<Instruction *> Chain,
968 SmallPtrSet<Instruction *, 16> *InstructionsProcessed) {
969 LoadInst *L0 = cast<LoadInst>(Chain[0]);
970
971 // If the vector has an int element, default to int for the whole load.
972 Type *LoadTy;
16
'LoadTy' declared without an initial value
973 for (const auto &V : Chain) {
17
Assuming '__begin1' is equal to '__end1'
974 LoadTy = cast<LoadInst>(V)->getType();
975 if (LoadTy->isIntOrIntVectorTy())
976 break;
977
978 if (LoadTy->isPtrOrPtrVectorTy()) {
979 LoadTy = Type::getIntNTy(F.getParent()->getContext(),
980 DL.getTypeSizeInBits(LoadTy));
981 break;
982 }
983 }
984
985 unsigned Sz = DL.getTypeSizeInBits(LoadTy);
18
1st function call argument is an uninitialized value
986 unsigned AS = L0->getPointerAddressSpace();
987 unsigned VecRegSize = TTI.getLoadStoreVecRegBitWidth(AS);
988 unsigned VF = VecRegSize / Sz;
989 unsigned ChainSize = Chain.size();
990 unsigned Alignment = getAlignment(L0);
991
992 if (!isPowerOf2_32(Sz) || VF < 2 || ChainSize < 2) {
993 InstructionsProcessed->insert(Chain.begin(), Chain.end());
994 return false;
995 }
996
997 ArrayRef<Instruction *> NewChain = getVectorizablePrefix(Chain);
998 if (NewChain.empty()) {
999 // No vectorization possible.
1000 InstructionsProcessed->insert(Chain.begin(), Chain.end());
1001 return false;
1002 }
1003 if (NewChain.size() == 1) {
1004 // Failed after the first instruction. Discard it and try the smaller chain.
1005 InstructionsProcessed->insert(NewChain.front());
1006 return false;
1007 }
1008
1009 // Update Chain to the valid vectorizable subchain.
1010 Chain = NewChain;
1011 ChainSize = Chain.size();
1012
1013 // Check if it's legal to vectorize this chain. If not, split the chain and
1014 // try again.
1015 unsigned EltSzInBytes = Sz / 8;
1016 unsigned SzInBytes = EltSzInBytes * ChainSize;
1017 if (!TTI.isLegalToVectorizeLoadChain(SzInBytes, Alignment, AS)) {
1018 auto Chains = splitOddVectorElts(Chain, Sz);
1019 return vectorizeLoadChain(Chains.first, InstructionsProcessed) |
1020 vectorizeLoadChain(Chains.second, InstructionsProcessed);
1021 }
1022
1023 VectorType *VecTy;
1024 VectorType *VecLoadTy = dyn_cast<VectorType>(LoadTy);
1025 if (VecLoadTy)
1026 VecTy = VectorType::get(LoadTy->getScalarType(),
1027 Chain.size() * VecLoadTy->getNumElements());
1028 else
1029 VecTy = VectorType::get(LoadTy, Chain.size());
1030
1031 // If it's more than the max vector size or the target has a better
1032 // vector factor, break it into two pieces.
1033 unsigned TargetVF = TTI.getLoadVectorFactor(VF, Sz, SzInBytes, VecTy);
1034 if (ChainSize > VF || (VF != TargetVF && TargetVF < ChainSize)) {
1035 DEBUG(dbgs() << "LSV: Chain doesn't match with the vector factor."do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { dbgs() << "LSV: Chain doesn't match with the vector factor."
" Creating two separate arrays.\n"; } } while (false)
1036 " Creating two separate arrays.\n")do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { dbgs() << "LSV: Chain doesn't match with the vector factor."
" Creating two separate arrays.\n"; } } while (false)
;
1037 return vectorizeLoadChain(Chain.slice(0, TargetVF), InstructionsProcessed) |
1038 vectorizeLoadChain(Chain.slice(TargetVF), InstructionsProcessed);
1039 }
1040
1041 // We won't try again to vectorize the elements of the chain, regardless of
1042 // whether we succeed below.
1043 InstructionsProcessed->insert(Chain.begin(), Chain.end());
1044
1045 // If the load is going to be misaligned, don't vectorize it.
1046 if (accessIsMisaligned(SzInBytes, AS, Alignment)) {
1047 if (L0->getPointerAddressSpace() != 0)
1048 return false;
1049
1050 unsigned NewAlign = getOrEnforceKnownAlignment(L0->getPointerOperand(),
1051 StackAdjustedAlignment,
1052 DL, L0, nullptr, &DT);
1053 if (NewAlign < StackAdjustedAlignment)
1054 return false;
1055
1056 Alignment = NewAlign;
1057 }
1058
1059 DEBUG({do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Loads to vectorize:\n"
; for (Instruction *I : Chain) I->dump(); }; } } while (false
)
1060 dbgs() << "LSV: Loads to vectorize:\n";do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Loads to vectorize:\n"
; for (Instruction *I : Chain) I->dump(); }; } } while (false
)
1061 for (Instruction *I : Chain)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Loads to vectorize:\n"
; for (Instruction *I : Chain) I->dump(); }; } } while (false
)
1062 I->dump();do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Loads to vectorize:\n"
; for (Instruction *I : Chain) I->dump(); }; } } while (false
)
1063 })do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { { dbgs() << "LSV: Loads to vectorize:\n"
; for (Instruction *I : Chain) I->dump(); }; } } while (false
)
;
1064
1065 // getVectorizablePrefix already computed getBoundaryInstrs. The value of
1066 // Last may have changed since then, but the value of First won't have. If it
1067 // matters, we could compute getBoundaryInstrs only once and reuse it here.
1068 BasicBlock::iterator First, Last;
1069 std::tie(First, Last) = getBoundaryInstrs(Chain);
1070 Builder.SetInsertPoint(&*First);
1071
1072 Value *Bitcast =
1073 Builder.CreateBitCast(L0->getPointerOperand(), VecTy->getPointerTo(AS));
1074 // This cast is safe because Builder.CreateLoad always creates a bona fide
1075 // LoadInst.
1076 LoadInst *LI = cast<LoadInst>(Builder.CreateLoad(Bitcast));
1077 propagateMetadata(LI, Chain);
1078 LI->setAlignment(Alignment);
1079
1080 if (VecLoadTy) {
1081 SmallVector<Instruction *, 16> InstrsToErase;
1082
1083 unsigned VecWidth = VecLoadTy->getNumElements();
1084 for (unsigned I = 0, E = Chain.size(); I != E; ++I) {
1085 for (auto Use : Chain[I]->users()) {
1086 // All users of vector loads are ExtractElement instructions with
1087 // constant indices, otherwise we would have bailed before now.
1088 Instruction *UI = cast<Instruction>(Use);
1089 unsigned Idx = cast<ConstantInt>(UI->getOperand(1))->getZExtValue();
1090 unsigned NewIdx = Idx + I * VecWidth;
1091 Value *V = Builder.CreateExtractElement(LI, Builder.getInt32(NewIdx),
1092 UI->getName());
1093 if (V->getType() != UI->getType())
1094 V = Builder.CreateBitCast(V, UI->getType());
1095
1096 // Replace the old instruction.
1097 UI->replaceAllUsesWith(V);
1098 InstrsToErase.push_back(UI);
1099 }
1100 }
1101
1102 // Bitcast might not be an Instruction, if the value being loaded is a
1103 // constant. In that case, no need to reorder anything.
1104 if (Instruction *BitcastInst = dyn_cast<Instruction>(Bitcast))
1105 reorder(BitcastInst);
1106
1107 for (auto I : InstrsToErase)
1108 I->eraseFromParent();
1109 } else {
1110 for (unsigned I = 0, E = Chain.size(); I != E; ++I) {
1111 Value *CV = Chain[I];
1112 Value *V =
1113 Builder.CreateExtractElement(LI, Builder.getInt32(I), CV->getName());
1114 if (V->getType() != CV->getType()) {
1115 V = Builder.CreateBitOrPointerCast(V, CV->getType());
1116 }
1117
1118 // Replace the old instruction.
1119 CV->replaceAllUsesWith(V);
1120 }
1121
1122 if (Instruction *BitcastInst = dyn_cast<Instruction>(Bitcast))
1123 reorder(BitcastInst);
1124 }
1125
1126 eraseInstructions(Chain);
1127
1128 ++NumVectorInstructions;
1129 NumScalarsVectorized += Chain.size();
1130 return true;
1131}
1132
1133bool Vectorizer::accessIsMisaligned(unsigned SzInBytes, unsigned AddressSpace,
1134 unsigned Alignment) {
1135 if (Alignment % SzInBytes == 0)
1136 return false;
1137
1138 bool Fast = false;
1139 bool Allows = TTI.allowsMisalignedMemoryAccesses(F.getParent()->getContext(),
1140 SzInBytes * 8, AddressSpace,
1141 Alignment, &Fast);
1142 DEBUG(dbgs() << "LSV: Target said misaligned is allowed? " << Allowsdo { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { dbgs() << "LSV: Target said misaligned is allowed? "
<< Allows << " and fast? " << Fast <<
"\n";; } } while (false)
1143 << " and fast? " << Fast << "\n";)do { if (::llvm::DebugFlag && ::llvm::isCurrentDebugType
("load-store-vectorizer")) { dbgs() << "LSV: Target said misaligned is allowed? "
<< Allows << " and fast? " << Fast <<
"\n";; } } while (false)
;
1144 return !Allows || !Fast;
1145}