1 //===- LoopVectorize.cpp - A Loop Vectorizer ------------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
9 #include "LoopVectorize.h"
10 #include "llvm/ADT/StringExtras.h"
11 #include "llvm/Analysis/AliasAnalysis.h"
12 #include "llvm/Analysis/AliasSetTracker.h"
13 #include "llvm/Analysis/Dominators.h"
14 #include "llvm/Analysis/LoopInfo.h"
15 #include "llvm/Analysis/LoopIterator.h"
16 #include "llvm/Analysis/LoopPass.h"
17 #include "llvm/Analysis/ScalarEvolutionExpander.h"
18 #include "llvm/Analysis/ScalarEvolutionExpander.h"
19 #include "llvm/Analysis/ScalarEvolutionExpressions.h"
20 #include "llvm/Analysis/ValueTracking.h"
21 #include "llvm/Analysis/Verifier.h"
22 #include "llvm/Constants.h"
23 #include "llvm/DataLayout.h"
24 #include "llvm/DerivedTypes.h"
25 #include "llvm/Function.h"
26 #include "llvm/Instructions.h"
27 #include "llvm/IntrinsicInst.h"
28 #include "llvm/LLVMContext.h"
29 #include "llvm/Module.h"
30 #include "llvm/Pass.h"
31 #include "llvm/Support/CommandLine.h"
32 #include "llvm/Support/Debug.h"
33 #include "llvm/Support/raw_ostream.h"
34 #include "llvm/TargetTransformInfo.h"
35 #include "llvm/Transforms/Scalar.h"
36 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
37 #include "llvm/Transforms/Utils/Local.h"
38 #include "llvm/Transforms/Vectorize.h"
39 #include "llvm/Type.h"
40 #include "llvm/Value.h"
42 static cl::opt<unsigned>
43 VectorizationFactor("force-vector-width", cl::init(0), cl::Hidden,
44 cl::desc("Sets the SIMD width. Zero is autoselect."));
47 EnableIfConversion("enable-if-conversion", cl::init(true), cl::Hidden,
48 cl::desc("Enable if-conversion during vectorization."));
52 /// The LoopVectorize Pass.
53 struct LoopVectorize : public LoopPass {
54 static char ID; // Pass identification, replacement for typeid
56 LoopVectorize() : LoopPass(ID) {
57 initializeLoopVectorizePass(*PassRegistry::getPassRegistry());
63 TargetTransformInfo *TTI;
66 virtual bool runOnLoop(Loop *L, LPPassManager &LPM) {
67 // We only vectorize innermost loops.
71 SE = &getAnalysis<ScalarEvolution>();
72 DL = getAnalysisIfAvailable<DataLayout>();
73 LI = &getAnalysis<LoopInfo>();
74 TTI = getAnalysisIfAvailable<TargetTransformInfo>();
75 DT = &getAnalysis<DominatorTree>();
77 DEBUG(dbgs() << "LV: Checking a loop in \"" <<
78 L->getHeader()->getParent()->getName() << "\"\n");
80 // Check if it is legal to vectorize the loop.
81 LoopVectorizationLegality LVL(L, SE, DL, DT);
82 if (!LVL.canVectorize()) {
83 DEBUG(dbgs() << "LV: Not vectorizing.\n");
87 // Select the preffered vectorization factor.
89 if (VectorizationFactor == 0) {
90 const VectorTargetTransformInfo *VTTI = 0;
92 VTTI = TTI->getVectorTargetTransformInfo();
93 // Use the cost model.
94 LoopVectorizationCostModel CM(L, SE, &LVL, VTTI);
95 VF = CM.findBestVectorizationFactor();
98 DEBUG(dbgs() << "LV: Vectorization is possible but not beneficial.\n");
103 // Use the user command flag.
104 VF = VectorizationFactor;
107 DEBUG(dbgs() << "LV: Found a vectorizable loop ("<< VF << ") in "<<
108 L->getHeader()->getParent()->getParent()->getModuleIdentifier()<<
111 // If we decided that it is *legal* to vectorizer the loop then do it.
112 InnerLoopVectorizer LB(L, SE, LI, DT, DL, VF);
115 DEBUG(verifyFunction(*L->getHeader()->getParent()));
119 virtual void getAnalysisUsage(AnalysisUsage &AU) const {
120 LoopPass::getAnalysisUsage(AU);
121 AU.addRequiredID(LoopSimplifyID);
122 AU.addRequiredID(LCSSAID);
123 AU.addRequired<LoopInfo>();
124 AU.addRequired<ScalarEvolution>();
125 AU.addRequired<DominatorTree>();
126 AU.addPreserved<LoopInfo>();
127 AU.addPreserved<DominatorTree>();
134 //===----------------------------------------------------------------------===//
135 // Implementation of LoopVectorizationLegality, InnerLoopVectorizer and
136 // LoopVectorizationCostModel.
137 //===----------------------------------------------------------------------===//
140 LoopVectorizationLegality::RuntimePointerCheck::insert(ScalarEvolution *SE,
141 Loop *Lp, Value *Ptr) {
142 const SCEV *Sc = SE->getSCEV(Ptr);
143 const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(Sc);
144 assert(AR && "Invalid addrec expression");
145 const SCEV *Ex = SE->getExitCount(Lp, Lp->getLoopLatch());
146 const SCEV *ScEnd = AR->evaluateAtIteration(Ex, *SE);
147 Pointers.push_back(Ptr);
148 Starts.push_back(AR->getStart());
149 Ends.push_back(ScEnd);
152 Value *InnerLoopVectorizer::getBroadcastInstrs(Value *V) {
154 LLVMContext &C = V->getContext();
155 Type *VTy = VectorType::get(V->getType(), VF);
156 Type *I32 = IntegerType::getInt32Ty(C);
158 // Save the current insertion location.
159 Instruction *Loc = Builder.GetInsertPoint();
161 // We need to place the broadcast of invariant variables outside the loop.
162 Instruction *Instr = dyn_cast<Instruction>(V);
163 bool NewInstr = (Instr && Instr->getParent() == LoopVectorBody);
164 bool Invariant = OrigLoop->isLoopInvariant(V) && !NewInstr;
166 // Place the code for broadcasting invariant variables in the new preheader.
168 Builder.SetInsertPoint(LoopVectorPreHeader->getTerminator());
170 Constant *Zero = ConstantInt::get(I32, 0);
171 Value *Zeros = ConstantAggregateZero::get(VectorType::get(I32, VF));
172 Value *UndefVal = UndefValue::get(VTy);
173 // Insert the value into a new vector.
174 Value *SingleElem = Builder.CreateInsertElement(UndefVal, V, Zero);
175 // Broadcast the scalar into all locations in the vector.
176 Value *Shuf = Builder.CreateShuffleVector(SingleElem, UndefVal, Zeros,
179 // Restore the builder insertion point.
181 Builder.SetInsertPoint(Loc);
186 Value *InnerLoopVectorizer::getConsecutiveVector(Value* Val, bool Negate) {
187 assert(Val->getType()->isVectorTy() && "Must be a vector");
188 assert(Val->getType()->getScalarType()->isIntegerTy() &&
189 "Elem must be an integer");
191 Type *ITy = Val->getType()->getScalarType();
192 VectorType *Ty = cast<VectorType>(Val->getType());
193 int VLen = Ty->getNumElements();
194 SmallVector<Constant*, 8> Indices;
196 // Create a vector of consecutive numbers from zero to VF.
197 for (int i = 0; i < VLen; ++i)
198 Indices.push_back(ConstantInt::get(ITy, Negate ? (-i): i ));
200 // Add the consecutive indices to the vector value.
201 Constant *Cv = ConstantVector::get(Indices);
202 assert(Cv->getType() == Val->getType() && "Invalid consecutive vec");
203 return Builder.CreateAdd(Val, Cv, "induction");
206 bool LoopVectorizationLegality::isConsecutivePtr(Value *Ptr) {
207 assert(Ptr->getType()->isPointerTy() && "Unexpected non ptr");
209 // If this value is a pointer induction variable we know it is consecutive.
210 PHINode *Phi = dyn_cast_or_null<PHINode>(Ptr);
211 if (Phi && Inductions.count(Phi)) {
212 InductionInfo II = Inductions[Phi];
213 if (PtrInduction == II.IK)
217 GetElementPtrInst *Gep = dyn_cast_or_null<GetElementPtrInst>(Ptr);
221 unsigned NumOperands = Gep->getNumOperands();
222 Value *LastIndex = Gep->getOperand(NumOperands - 1);
224 // Check that all of the gep indices are uniform except for the last.
225 for (unsigned i = 0; i < NumOperands - 1; ++i)
226 if (!SE->isLoopInvariant(SE->getSCEV(Gep->getOperand(i)), TheLoop))
229 // We can emit wide load/stores only if the last index is the induction
231 const SCEV *Last = SE->getSCEV(LastIndex);
232 if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(Last)) {
233 const SCEV *Step = AR->getStepRecurrence(*SE);
235 // The memory is consecutive because the last index is consecutive
236 // and all other indices are loop invariant.
244 bool LoopVectorizationLegality::isUniform(Value *V) {
245 return (SE->isLoopInvariant(SE->getSCEV(V), TheLoop));
248 Value *InnerLoopVectorizer::getVectorValue(Value *V) {
249 assert(V != Induction && "The new induction variable should not be used.");
250 assert(!V->getType()->isVectorTy() && "Can't widen a vector");
251 // If we saved a vectorized copy of V, use it.
252 Value *&MapEntry = WidenMap[V];
256 // Broadcast V and save the value for future uses.
257 Value *B = getBroadcastInstrs(V);
263 InnerLoopVectorizer::getUniformVector(unsigned Val, Type* ScalarTy) {
264 return ConstantVector::getSplat(VF, ConstantInt::get(ScalarTy, Val, true));
267 void InnerLoopVectorizer::scalarizeInstruction(Instruction *Instr) {
268 assert(!Instr->getType()->isAggregateType() && "Can't handle vectors");
269 // Holds vector parameters or scalars, in case of uniform vals.
270 SmallVector<Value*, 8> Params;
272 // Find all of the vectorized parameters.
273 for (unsigned op = 0, e = Instr->getNumOperands(); op != e; ++op) {
274 Value *SrcOp = Instr->getOperand(op);
276 // If we are accessing the old induction variable, use the new one.
277 if (SrcOp == OldInduction) {
278 Params.push_back(getVectorValue(SrcOp));
282 // Try using previously calculated values.
283 Instruction *SrcInst = dyn_cast<Instruction>(SrcOp);
285 // If the src is an instruction that appeared earlier in the basic block
286 // then it should already be vectorized.
287 if (SrcInst && SrcInst->getParent() == Instr->getParent()) {
288 assert(WidenMap.count(SrcInst) && "Source operand is unavailable");
289 // The parameter is a vector value from earlier.
290 Params.push_back(WidenMap[SrcInst]);
292 // The parameter is a scalar from outside the loop. Maybe even a constant.
293 Params.push_back(SrcOp);
297 assert(Params.size() == Instr->getNumOperands() &&
298 "Invalid number of operands");
300 // Does this instruction return a value ?
301 bool IsVoidRetTy = Instr->getType()->isVoidTy();
302 Value *VecResults = 0;
304 // If we have a return value, create an empty vector. We place the scalarized
305 // instructions in this vector.
307 VecResults = UndefValue::get(VectorType::get(Instr->getType(), VF));
309 // For each scalar that we create:
310 for (unsigned i = 0; i < VF; ++i) {
311 Instruction *Cloned = Instr->clone();
313 Cloned->setName(Instr->getName() + ".cloned");
314 // Replace the operands of the cloned instrucions with extracted scalars.
315 for (unsigned op = 0, e = Instr->getNumOperands(); op != e; ++op) {
316 Value *Op = Params[op];
317 // Param is a vector. Need to extract the right lane.
318 if (Op->getType()->isVectorTy())
319 Op = Builder.CreateExtractElement(Op, Builder.getInt32(i));
320 Cloned->setOperand(op, Op);
323 // Place the cloned scalar in the new loop.
324 Builder.Insert(Cloned);
326 // If the original scalar returns a value we need to place it in a vector
327 // so that future users will be able to use it.
329 VecResults = Builder.CreateInsertElement(VecResults, Cloned,
330 Builder.getInt32(i));
334 WidenMap[Instr] = VecResults;
338 InnerLoopVectorizer::addRuntimeCheck(LoopVectorizationLegality *Legal,
340 LoopVectorizationLegality::RuntimePointerCheck *PtrRtCheck =
341 Legal->getRuntimePointerCheck();
343 if (!PtrRtCheck->Need)
346 Value *MemoryRuntimeCheck = 0;
347 unsigned NumPointers = PtrRtCheck->Pointers.size();
348 SmallVector<Value* , 2> Starts;
349 SmallVector<Value* , 2> Ends;
351 SCEVExpander Exp(*SE, "induction");
353 // Use this type for pointer arithmetic.
354 Type* PtrArithTy = Type::getInt8PtrTy(Loc->getContext(), 0);
356 for (unsigned i = 0; i < NumPointers; ++i) {
357 Value *Ptr = PtrRtCheck->Pointers[i];
358 const SCEV *Sc = SE->getSCEV(Ptr);
360 if (SE->isLoopInvariant(Sc, OrigLoop)) {
361 DEBUG(dbgs() << "LV: Adding RT check for a loop invariant ptr:" <<
363 Starts.push_back(Ptr);
366 DEBUG(dbgs() << "LV: Adding RT check for range:" << *Ptr <<"\n");
368 Value *Start = Exp.expandCodeFor(PtrRtCheck->Starts[i], PtrArithTy, Loc);
369 Value *End = Exp.expandCodeFor(PtrRtCheck->Ends[i], PtrArithTy, Loc);
370 Starts.push_back(Start);
375 for (unsigned i = 0; i < NumPointers; ++i) {
376 for (unsigned j = i+1; j < NumPointers; ++j) {
377 Instruction::CastOps Op = Instruction::BitCast;
378 Value *Start0 = CastInst::Create(Op, Starts[i], PtrArithTy, "bc", Loc);
379 Value *Start1 = CastInst::Create(Op, Starts[j], PtrArithTy, "bc", Loc);
380 Value *End0 = CastInst::Create(Op, Ends[i], PtrArithTy, "bc", Loc);
381 Value *End1 = CastInst::Create(Op, Ends[j], PtrArithTy, "bc", Loc);
383 Value *Cmp0 = CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_ULE,
384 Start0, End1, "bound0", Loc);
385 Value *Cmp1 = CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_ULE,
386 Start1, End0, "bound1", Loc);
387 Value *IsConflict = BinaryOperator::Create(Instruction::And, Cmp0, Cmp1,
388 "found.conflict", Loc);
389 if (MemoryRuntimeCheck)
390 MemoryRuntimeCheck = BinaryOperator::Create(Instruction::Or,
393 "conflict.rdx", Loc);
395 MemoryRuntimeCheck = IsConflict;
400 return MemoryRuntimeCheck;
404 InnerLoopVectorizer::createEmptyLoop(LoopVectorizationLegality *Legal) {
406 In this function we generate a new loop. The new loop will contain
407 the vectorized instructions while the old loop will continue to run the
410 [ ] <-- vector loop bypass.
413 | [ ] <-- vector pre header.
417 | [ ]_| <-- vector loop.
420 >[ ] <--- middle-block.
423 | [ ] <--- new preheader.
427 | [ ]_| <-- old scalar loop to handle remainder.
434 BasicBlock *OldBasicBlock = OrigLoop->getHeader();
435 BasicBlock *BypassBlock = OrigLoop->getLoopPreheader();
436 BasicBlock *ExitBlock = OrigLoop->getExitBlock();
437 assert(ExitBlock && "Must have an exit block");
439 // Some loops have a single integer induction variable, while other loops
440 // don't. One example is c++ iterators that often have multiple pointer
441 // induction variables. In the code below we also support a case where we
442 // don't have a single induction variable.
443 OldInduction = Legal->getInduction();
444 Type *IdxTy = OldInduction ? OldInduction->getType() :
445 DL->getIntPtrType(SE->getContext());
447 // Find the loop boundaries.
448 const SCEV *ExitCount = SE->getExitCount(OrigLoop, OrigLoop->getLoopLatch());
449 assert(ExitCount != SE->getCouldNotCompute() && "Invalid loop count");
451 // Get the total trip count from the count by adding 1.
452 ExitCount = SE->getAddExpr(ExitCount,
453 SE->getConstant(ExitCount->getType(), 1));
455 // Expand the trip count and place the new instructions in the preheader.
456 // Notice that the pre-header does not change, only the loop body.
457 SCEVExpander Exp(*SE, "induction");
459 // Count holds the overall loop count (N).
460 Value *Count = Exp.expandCodeFor(ExitCount, ExitCount->getType(),
461 BypassBlock->getTerminator());
463 // The loop index does not have to start at Zero. Find the original start
464 // value from the induction PHI node. If we don't have an induction variable
465 // then we know that it starts at zero.
466 Value *StartIdx = OldInduction ?
467 OldInduction->getIncomingValueForBlock(BypassBlock):
468 ConstantInt::get(IdxTy, 0);
470 assert(BypassBlock && "Invalid loop structure");
472 // Generate the code that checks in runtime if arrays overlap.
473 Value *MemoryRuntimeCheck = addRuntimeCheck(Legal,
474 BypassBlock->getTerminator());
476 // Split the single block loop into the two loop structure described above.
477 BasicBlock *VectorPH =
478 BypassBlock->splitBasicBlock(BypassBlock->getTerminator(), "vector.ph");
479 BasicBlock *VecBody =
480 VectorPH->splitBasicBlock(VectorPH->getTerminator(), "vector.body");
481 BasicBlock *MiddleBlock =
482 VecBody->splitBasicBlock(VecBody->getTerminator(), "middle.block");
483 BasicBlock *ScalarPH =
484 MiddleBlock->splitBasicBlock(MiddleBlock->getTerminator(), "scalar.ph");
486 // This is the location in which we add all of the logic for bypassing
487 // the new vector loop.
488 Instruction *Loc = BypassBlock->getTerminator();
490 // Use this IR builder to create the loop instructions (Phi, Br, Cmp)
492 Builder.SetInsertPoint(VecBody->getFirstInsertionPt());
494 // Generate the induction variable.
495 Induction = Builder.CreatePHI(IdxTy, 2, "index");
496 Constant *Step = ConstantInt::get(IdxTy, VF);
498 // We may need to extend the index in case there is a type mismatch.
499 // We know that the count starts at zero and does not overflow.
500 if (Count->getType() != IdxTy) {
501 // The exit count can be of pointer type. Convert it to the correct
503 if (ExitCount->getType()->isPointerTy())
504 Count = CastInst::CreatePointerCast(Count, IdxTy, "ptrcnt.to.int", Loc);
506 Count = CastInst::CreateZExtOrBitCast(Count, IdxTy, "zext.cnt", Loc);
509 // Add the start index to the loop count to get the new end index.
510 Value *IdxEnd = BinaryOperator::CreateAdd(Count, StartIdx, "end.idx", Loc);
512 // Now we need to generate the expression for N - (N % VF), which is
513 // the part that the vectorized body will execute.
514 Constant *CIVF = ConstantInt::get(IdxTy, VF);
515 Value *R = BinaryOperator::CreateURem(Count, CIVF, "n.mod.vf", Loc);
516 Value *CountRoundDown = BinaryOperator::CreateSub(Count, R, "n.vec", Loc);
517 Value *IdxEndRoundDown = BinaryOperator::CreateAdd(CountRoundDown, StartIdx,
518 "end.idx.rnd.down", Loc);
520 // Now, compare the new count to zero. If it is zero skip the vector loop and
521 // jump to the scalar loop.
522 Value *Cmp = CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_EQ,
527 // If we are using memory runtime checks, include them in.
528 if (MemoryRuntimeCheck)
529 Cmp = BinaryOperator::Create(Instruction::Or, Cmp, MemoryRuntimeCheck,
532 BranchInst::Create(MiddleBlock, VectorPH, Cmp, Loc);
533 // Remove the old terminator.
534 Loc->eraseFromParent();
536 // We are going to resume the execution of the scalar loop.
537 // Go over all of the induction variables that we found and fix the
538 // PHIs that are left in the scalar version of the loop.
539 // The starting values of PHI nodes depend on the counter of the last
540 // iteration in the vectorized loop.
541 // If we come from a bypass edge then we need to start from the original
544 // This variable saves the new starting index for the scalar loop.
545 PHINode *ResumeIndex = 0;
546 LoopVectorizationLegality::InductionList::iterator I, E;
547 LoopVectorizationLegality::InductionList *List = Legal->getInductionVars();
548 for (I = List->begin(), E = List->end(); I != E; ++I) {
549 PHINode *OrigPhi = I->first;
550 LoopVectorizationLegality::InductionInfo II = I->second;
551 PHINode *ResumeVal = PHINode::Create(OrigPhi->getType(), 2, "resume.val",
552 MiddleBlock->getTerminator());
555 case LoopVectorizationLegality::NoInduction:
556 llvm_unreachable("Unknown induction");
557 case LoopVectorizationLegality::IntInduction: {
558 // Handle the integer induction counter:
559 assert(OrigPhi->getType()->isIntegerTy() && "Invalid type");
560 assert(OrigPhi == OldInduction && "Unknown integer PHI");
561 // We know what the end value is.
562 EndValue = IdxEndRoundDown;
563 // We also know which PHI node holds it.
564 ResumeIndex = ResumeVal;
567 case LoopVectorizationLegality::ReverseIntInduction: {
568 // Convert the CountRoundDown variable to the PHI size.
569 unsigned CRDSize = CountRoundDown->getType()->getScalarSizeInBits();
570 unsigned IISize = II.StartValue->getType()->getScalarSizeInBits();
571 Value *CRD = CountRoundDown;
572 if (CRDSize > IISize)
573 CRD = CastInst::Create(Instruction::Trunc, CountRoundDown,
574 II.StartValue->getType(),
575 "tr.crd", BypassBlock->getTerminator());
576 else if (CRDSize < IISize)
577 CRD = CastInst::Create(Instruction::SExt, CountRoundDown,
578 II.StartValue->getType(),
579 "sext.crd", BypassBlock->getTerminator());
580 // Handle reverse integer induction counter:
581 EndValue = BinaryOperator::CreateSub(II.StartValue, CRD, "rev.ind.end",
582 BypassBlock->getTerminator());
585 case LoopVectorizationLegality::PtrInduction: {
586 // For pointer induction variables, calculate the offset using
588 EndValue = GetElementPtrInst::Create(II.StartValue, CountRoundDown,
590 BypassBlock->getTerminator());
595 // The new PHI merges the original incoming value, in case of a bypass,
596 // or the value at the end of the vectorized loop.
597 ResumeVal->addIncoming(II.StartValue, BypassBlock);
598 ResumeVal->addIncoming(EndValue, VecBody);
600 // Fix the scalar body counter (PHI node).
601 unsigned BlockIdx = OrigPhi->getBasicBlockIndex(ScalarPH);
602 OrigPhi->setIncomingValue(BlockIdx, ResumeVal);
605 // If we are generating a new induction variable then we also need to
606 // generate the code that calculates the exit value. This value is not
607 // simply the end of the counter because we may skip the vectorized body
608 // in case of a runtime check.
610 assert(!ResumeIndex && "Unexpected resume value found");
611 ResumeIndex = PHINode::Create(IdxTy, 2, "new.indc.resume.val",
612 MiddleBlock->getTerminator());
613 ResumeIndex->addIncoming(StartIdx, BypassBlock);
614 ResumeIndex->addIncoming(IdxEndRoundDown, VecBody);
617 // Make sure that we found the index where scalar loop needs to continue.
618 assert(ResumeIndex && ResumeIndex->getType()->isIntegerTy() &&
619 "Invalid resume Index");
621 // Add a check in the middle block to see if we have completed
622 // all of the iterations in the first vector loop.
623 // If (N - N%VF) == N, then we *don't* need to run the remainder.
624 Value *CmpN = CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_EQ, IdxEnd,
625 ResumeIndex, "cmp.n",
626 MiddleBlock->getTerminator());
628 BranchInst::Create(ExitBlock, ScalarPH, CmpN, MiddleBlock->getTerminator());
629 // Remove the old terminator.
630 MiddleBlock->getTerminator()->eraseFromParent();
632 // Create i+1 and fill the PHINode.
633 Value *NextIdx = Builder.CreateAdd(Induction, Step, "index.next");
634 Induction->addIncoming(StartIdx, VectorPH);
635 Induction->addIncoming(NextIdx, VecBody);
636 // Create the compare.
637 Value *ICmp = Builder.CreateICmpEQ(NextIdx, IdxEndRoundDown);
638 Builder.CreateCondBr(ICmp, MiddleBlock, VecBody);
640 // Now we have two terminators. Remove the old one from the block.
641 VecBody->getTerminator()->eraseFromParent();
643 // Get ready to start creating new instructions into the vectorized body.
644 Builder.SetInsertPoint(VecBody->getFirstInsertionPt());
646 // Create and register the new vector loop.
647 Loop* Lp = new Loop();
648 Loop *ParentLoop = OrigLoop->getParentLoop();
650 // Insert the new loop into the loop nest and register the new basic blocks.
652 ParentLoop->addChildLoop(Lp);
653 ParentLoop->addBasicBlockToLoop(ScalarPH, LI->getBase());
654 ParentLoop->addBasicBlockToLoop(VectorPH, LI->getBase());
655 ParentLoop->addBasicBlockToLoop(MiddleBlock, LI->getBase());
657 LI->addTopLevelLoop(Lp);
660 Lp->addBasicBlockToLoop(VecBody, LI->getBase());
663 LoopVectorPreHeader = VectorPH;
664 LoopScalarPreHeader = ScalarPH;
665 LoopMiddleBlock = MiddleBlock;
666 LoopExitBlock = ExitBlock;
667 LoopVectorBody = VecBody;
668 LoopScalarBody = OldBasicBlock;
669 LoopBypassBlock = BypassBlock;
672 /// This function returns the identity element (or neutral element) for
675 getReductionIdentity(LoopVectorizationLegality::ReductionKind K) {
677 case LoopVectorizationLegality::IntegerXor:
678 case LoopVectorizationLegality::IntegerAdd:
679 case LoopVectorizationLegality::IntegerOr:
680 // Adding, Xoring, Oring zero to a number does not change it.
682 case LoopVectorizationLegality::IntegerMult:
683 // Multiplying a number by 1 does not change it.
685 case LoopVectorizationLegality::IntegerAnd:
686 // AND-ing a number with an all-1 value does not change it.
689 llvm_unreachable("Unknown reduction kind");
694 isTriviallyVectorizableIntrinsic(Instruction *Inst) {
695 IntrinsicInst *II = dyn_cast<IntrinsicInst>(Inst);
698 switch (II->getIntrinsicID()) {
699 case Intrinsic::sqrt:
703 case Intrinsic::exp2:
705 case Intrinsic::log10:
706 case Intrinsic::log2:
707 case Intrinsic::fabs:
708 case Intrinsic::floor:
709 case Intrinsic::ceil:
710 case Intrinsic::trunc:
711 case Intrinsic::rint:
712 case Intrinsic::nearbyint:
723 InnerLoopVectorizer::vectorizeLoop(LoopVectorizationLegality *Legal) {
724 //===------------------------------------------------===//
726 // Notice: any optimization or new instruction that go
727 // into the code below should be also be implemented in
730 //===------------------------------------------------===//
731 BasicBlock &BB = *OrigLoop->getHeader();
733 ConstantInt::get(IntegerType::getInt32Ty(BB.getContext()), 0);
735 // In order to support reduction variables we need to be able to vectorize
736 // Phi nodes. Phi nodes have cycles, so we need to vectorize them in two
737 // stages. First, we create a new vector PHI node with no incoming edges.
738 // We use this value when we vectorize all of the instructions that use the
739 // PHI. Next, after all of the instructions in the block are complete we
740 // add the new incoming edges to the PHI. At this point all of the
741 // instructions in the basic block are vectorized, so we can use them to
742 // construct the PHI.
743 PhiVector RdxPHIsToFix;
745 // Scan the loop in a topological order to ensure that defs are vectorized
747 LoopBlocksDFS DFS(OrigLoop);
750 // Vectorize all of the blocks in the original loop.
751 for (LoopBlocksDFS::RPOIterator bb = DFS.beginRPO(),
752 be = DFS.endRPO(); bb != be; ++bb)
753 vectorizeBlockInLoop(Legal, *bb, &RdxPHIsToFix);
755 // At this point every instruction in the original loop is widened to
756 // a vector form. We are almost done. Now, we need to fix the PHI nodes
757 // that we vectorized. The PHI nodes are currently empty because we did
758 // not want to introduce cycles. Notice that the remaining PHI nodes
759 // that we need to fix are reduction variables.
761 // Create the 'reduced' values for each of the induction vars.
762 // The reduced values are the vector values that we scalarize and combine
763 // after the loop is finished.
764 for (PhiVector::iterator it = RdxPHIsToFix.begin(), e = RdxPHIsToFix.end();
766 PHINode *RdxPhi = *it;
767 PHINode *VecRdxPhi = dyn_cast<PHINode>(WidenMap[RdxPhi]);
768 assert(RdxPhi && "Unable to recover vectorized PHI");
770 // Find the reduction variable descriptor.
771 assert(Legal->getReductionVars()->count(RdxPhi) &&
772 "Unable to find the reduction variable");
773 LoopVectorizationLegality::ReductionDescriptor RdxDesc =
774 (*Legal->getReductionVars())[RdxPhi];
776 // We need to generate a reduction vector from the incoming scalar.
777 // To do so, we need to generate the 'identity' vector and overide
778 // one of the elements with the incoming scalar reduction. We need
779 // to do it in the vector-loop preheader.
780 Builder.SetInsertPoint(LoopBypassBlock->getTerminator());
782 // This is the vector-clone of the value that leaves the loop.
783 Value *VectorExit = getVectorValue(RdxDesc.LoopExitInstr);
784 Type *VecTy = VectorExit->getType();
786 // Find the reduction identity variable. Zero for addition, or, xor,
787 // one for multiplication, -1 for And.
788 Constant *Identity = getUniformVector(getReductionIdentity(RdxDesc.Kind),
789 VecTy->getScalarType());
791 // This vector is the Identity vector where the first element is the
792 // incoming scalar reduction.
793 Value *VectorStart = Builder.CreateInsertElement(Identity,
794 RdxDesc.StartValue, Zero);
796 // Fix the vector-loop phi.
797 // We created the induction variable so we know that the
798 // preheader is the first entry.
799 BasicBlock *VecPreheader = Induction->getIncomingBlock(0);
801 // Reductions do not have to start at zero. They can start with
802 // any loop invariant values.
803 VecRdxPhi->addIncoming(VectorStart, VecPreheader);
805 getVectorValue(RdxPhi->getIncomingValueForBlock(OrigLoop->getLoopLatch()));
806 VecRdxPhi->addIncoming(Val, LoopVectorBody);
808 // Before each round, move the insertion point right between
809 // the PHIs and the values we are going to write.
810 // This allows us to write both PHINodes and the extractelement
812 Builder.SetInsertPoint(LoopMiddleBlock->getFirstInsertionPt());
814 // This PHINode contains the vectorized reduction variable, or
815 // the initial value vector, if we bypass the vector loop.
816 PHINode *NewPhi = Builder.CreatePHI(VecTy, 2, "rdx.vec.exit.phi");
817 NewPhi->addIncoming(VectorStart, LoopBypassBlock);
818 NewPhi->addIncoming(getVectorValue(RdxDesc.LoopExitInstr), LoopVectorBody);
820 // Extract the first scalar.
822 Builder.CreateExtractElement(NewPhi, Builder.getInt32(0));
823 // Extract and reduce the remaining vector elements.
824 for (unsigned i=1; i < VF; ++i) {
826 Builder.CreateExtractElement(NewPhi, Builder.getInt32(i));
827 switch (RdxDesc.Kind) {
828 case LoopVectorizationLegality::IntegerAdd:
829 Scalar0 = Builder.CreateAdd(Scalar0, Scalar1, "add.rdx");
831 case LoopVectorizationLegality::IntegerMult:
832 Scalar0 = Builder.CreateMul(Scalar0, Scalar1, "mul.rdx");
834 case LoopVectorizationLegality::IntegerOr:
835 Scalar0 = Builder.CreateOr(Scalar0, Scalar1, "or.rdx");
837 case LoopVectorizationLegality::IntegerAnd:
838 Scalar0 = Builder.CreateAnd(Scalar0, Scalar1, "and.rdx");
840 case LoopVectorizationLegality::IntegerXor:
841 Scalar0 = Builder.CreateXor(Scalar0, Scalar1, "xor.rdx");
844 llvm_unreachable("Unknown reduction operation");
848 // Now, we need to fix the users of the reduction variable
849 // inside and outside of the scalar remainder loop.
850 // We know that the loop is in LCSSA form. We need to update the
851 // PHI nodes in the exit blocks.
852 for (BasicBlock::iterator LEI = LoopExitBlock->begin(),
853 LEE = LoopExitBlock->end(); LEI != LEE; ++LEI) {
854 PHINode *LCSSAPhi = dyn_cast<PHINode>(LEI);
855 if (!LCSSAPhi) continue;
857 // All PHINodes need to have a single entry edge, or two if
858 // we already fixed them.
859 assert(LCSSAPhi->getNumIncomingValues() < 3 && "Invalid LCSSA PHI");
861 // We found our reduction value exit-PHI. Update it with the
862 // incoming bypass edge.
863 if (LCSSAPhi->getIncomingValue(0) == RdxDesc.LoopExitInstr) {
864 // Add an edge coming from the bypass.
865 LCSSAPhi->addIncoming(Scalar0, LoopMiddleBlock);
868 }// end of the LCSSA phi scan.
870 // Fix the scalar loop reduction variable with the incoming reduction sum
871 // from the vector body and from the backedge value.
872 int IncomingEdgeBlockIdx =
873 (RdxPhi)->getBasicBlockIndex(OrigLoop->getLoopLatch());
874 assert(IncomingEdgeBlockIdx >= 0 && "Invalid block index");
875 // Pick the other block.
876 int SelfEdgeBlockIdx = (IncomingEdgeBlockIdx ? 0 : 1);
877 (RdxPhi)->setIncomingValue(SelfEdgeBlockIdx, Scalar0);
878 (RdxPhi)->setIncomingValue(IncomingEdgeBlockIdx, RdxDesc.LoopExitInstr);
879 }// end of for each redux variable.
882 Value *InnerLoopVectorizer::createEdgeMask(BasicBlock *Src, BasicBlock *Dst) {
883 assert(std::find(pred_begin(Dst), pred_end(Dst), Src) != pred_end(Dst) &&
886 Value *SrcMask = createBlockInMask(Src);
888 // The terminator has to be a branch inst!
889 BranchInst *BI = dyn_cast<BranchInst>(Src->getTerminator());
890 assert(BI && "Unexpected terminator found");
892 Value *EdgeMask = SrcMask;
893 if (BI->isConditional()) {
894 EdgeMask = getVectorValue(BI->getCondition());
895 if (BI->getSuccessor(0) != Dst)
896 EdgeMask = Builder.CreateNot(EdgeMask);
899 return Builder.CreateAnd(EdgeMask, SrcMask);
902 Value *InnerLoopVectorizer::createBlockInMask(BasicBlock *BB) {
903 assert(OrigLoop->contains(BB) && "Block is not a part of a loop");
905 // Loop incoming mask is all-one.
906 if (OrigLoop->getHeader() == BB) {
907 Value *C = ConstantInt::get(IntegerType::getInt1Ty(BB->getContext()), 1);
908 return getVectorValue(C);
911 // This is the block mask. We OR all incoming edges, and with zero.
912 Value *Zero = ConstantInt::get(IntegerType::getInt1Ty(BB->getContext()), 0);
913 Value *BlockMask = getVectorValue(Zero);
916 for (pred_iterator it = pred_begin(BB), e = pred_end(BB); it != e; ++it)
917 BlockMask = Builder.CreateOr(BlockMask, createEdgeMask(*it, BB));
923 InnerLoopVectorizer::vectorizeBlockInLoop(LoopVectorizationLegality *Legal,
924 BasicBlock *BB, PhiVector *PV) {
926 ConstantInt::get(IntegerType::getInt32Ty(BB->getContext()), 0);
928 // For each instruction in the old loop.
929 for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; ++it) {
930 switch (it->getOpcode()) {
931 case Instruction::Br:
932 // Nothing to do for PHIs and BR, since we already took care of the
933 // loop control flow instructions.
935 case Instruction::PHI:{
936 PHINode* P = cast<PHINode>(it);
937 // Handle reduction variables:
938 if (Legal->getReductionVars()->count(P)) {
939 // This is phase one of vectorizing PHIs.
940 Type *VecTy = VectorType::get(it->getType(), VF);
942 PHINode::Create(VecTy, 2, "vec.phi",
943 LoopVectorBody->getFirstInsertionPt());
948 // Check for PHI nodes that are lowered to vector selects.
949 if (P->getParent() != OrigLoop->getHeader()) {
950 // We know that all PHIs in non header blocks are converted into
951 // selects, so we don't have to worry about the insertion order and we
952 // can just use the builder.
954 // At this point we generate the predication tree. There may be
955 // duplications since this is a simple recursive scan, but future
956 // optimizations will clean it up.
957 Value *Cond = createBlockInMask(P->getIncomingBlock(0));
959 Builder.CreateSelect(Cond,
960 getVectorValue(P->getIncomingValue(0)),
961 getVectorValue(P->getIncomingValue(1)),
966 // This PHINode must be an induction variable.
967 // Make sure that we know about it.
968 assert(Legal->getInductionVars()->count(P) &&
969 "Not an induction variable");
971 LoopVectorizationLegality::InductionInfo II =
972 Legal->getInductionVars()->lookup(P);
975 case LoopVectorizationLegality::NoInduction:
976 llvm_unreachable("Unknown induction");
977 case LoopVectorizationLegality::IntInduction: {
978 assert(P == OldInduction && "Unexpected PHI");
979 Value *Broadcasted = getBroadcastInstrs(Induction);
980 // After broadcasting the induction variable we need to make the
981 // vector consecutive by adding 0, 1, 2 ...
982 Value *ConsecutiveInduction = getConsecutiveVector(Broadcasted);
983 WidenMap[OldInduction] = ConsecutiveInduction;
986 case LoopVectorizationLegality::ReverseIntInduction:
987 case LoopVectorizationLegality::PtrInduction:
988 // Handle reverse integer and pointer inductions.
990 // If we have a single integer induction variable then use it.
991 // Otherwise, start counting at zero.
993 LoopVectorizationLegality::InductionInfo OldII =
994 Legal->getInductionVars()->lookup(OldInduction);
995 StartIdx = OldII.StartValue;
997 StartIdx = ConstantInt::get(Induction->getType(), 0);
999 // This is the normalized GEP that starts counting at zero.
1000 Value *NormalizedIdx = Builder.CreateSub(Induction, StartIdx,
1003 // Handle the reverse integer induction variable case.
1004 if (LoopVectorizationLegality::ReverseIntInduction == II.IK) {
1005 IntegerType *DstTy = cast<IntegerType>(II.StartValue->getType());
1006 Value *CNI = Builder.CreateSExtOrTrunc(NormalizedIdx, DstTy,
1008 Value *ReverseInd = Builder.CreateSub(II.StartValue, CNI,
1011 // This is a new value so do not hoist it out.
1012 Value *Broadcasted = getBroadcastInstrs(ReverseInd);
1013 // After broadcasting the induction variable we need to make the
1014 // vector consecutive by adding ... -3, -2, -1, 0.
1015 Value *ConsecutiveInduction = getConsecutiveVector(Broadcasted,
1017 WidenMap[it] = ConsecutiveInduction;
1021 // Handle the pointer induction variable case.
1022 assert(P->getType()->isPointerTy() && "Unexpected type.");
1024 // This is the vector of results. Notice that we don't generate
1025 // vector geps because scalar geps result in better code.
1026 Value *VecVal = UndefValue::get(VectorType::get(P->getType(), VF));
1027 for (unsigned int i = 0; i < VF; ++i) {
1028 Constant *Idx = ConstantInt::get(Induction->getType(), i);
1029 Value *GlobalIdx = Builder.CreateAdd(NormalizedIdx, Idx,
1031 Value *SclrGep = Builder.CreateGEP(II.StartValue, GlobalIdx,
1033 VecVal = Builder.CreateInsertElement(VecVal, SclrGep,
1034 Builder.getInt32(i),
1038 WidenMap[it] = VecVal;
1044 case Instruction::Add:
1045 case Instruction::FAdd:
1046 case Instruction::Sub:
1047 case Instruction::FSub:
1048 case Instruction::Mul:
1049 case Instruction::FMul:
1050 case Instruction::UDiv:
1051 case Instruction::SDiv:
1052 case Instruction::FDiv:
1053 case Instruction::URem:
1054 case Instruction::SRem:
1055 case Instruction::FRem:
1056 case Instruction::Shl:
1057 case Instruction::LShr:
1058 case Instruction::AShr:
1059 case Instruction::And:
1060 case Instruction::Or:
1061 case Instruction::Xor: {
1062 // Just widen binops.
1063 BinaryOperator *BinOp = dyn_cast<BinaryOperator>(it);
1064 Value *A = getVectorValue(it->getOperand(0));
1065 Value *B = getVectorValue(it->getOperand(1));
1067 // Use this vector value for all users of the original instruction.
1068 Value *V = Builder.CreateBinOp(BinOp->getOpcode(), A, B);
1071 // Update the NSW, NUW and Exact flags.
1072 BinaryOperator *VecOp = cast<BinaryOperator>(V);
1073 if (isa<OverflowingBinaryOperator>(BinOp)) {
1074 VecOp->setHasNoSignedWrap(BinOp->hasNoSignedWrap());
1075 VecOp->setHasNoUnsignedWrap(BinOp->hasNoUnsignedWrap());
1077 if (isa<PossiblyExactOperator>(VecOp))
1078 VecOp->setIsExact(BinOp->isExact());
1081 case Instruction::Select: {
1083 // If the selector is loop invariant we can create a select
1084 // instruction with a scalar condition. Otherwise, use vector-select.
1085 Value *Cond = it->getOperand(0);
1086 bool InvariantCond = SE->isLoopInvariant(SE->getSCEV(Cond), OrigLoop);
1088 // The condition can be loop invariant but still defined inside the
1089 // loop. This means that we can't just use the original 'cond' value.
1090 // We have to take the 'vectorized' value and pick the first lane.
1091 // Instcombine will make this a no-op.
1092 Cond = getVectorValue(Cond);
1094 Cond = Builder.CreateExtractElement(Cond, Builder.getInt32(0));
1096 Value *Op0 = getVectorValue(it->getOperand(1));
1097 Value *Op1 = getVectorValue(it->getOperand(2));
1098 WidenMap[it] = Builder.CreateSelect(Cond, Op0, Op1);
1102 case Instruction::ICmp:
1103 case Instruction::FCmp: {
1104 // Widen compares. Generate vector compares.
1105 bool FCmp = (it->getOpcode() == Instruction::FCmp);
1106 CmpInst *Cmp = dyn_cast<CmpInst>(it);
1107 Value *A = getVectorValue(it->getOperand(0));
1108 Value *B = getVectorValue(it->getOperand(1));
1110 WidenMap[it] = Builder.CreateFCmp(Cmp->getPredicate(), A, B);
1112 WidenMap[it] = Builder.CreateICmp(Cmp->getPredicate(), A, B);
1116 case Instruction::Store: {
1117 // Attempt to issue a wide store.
1118 StoreInst *SI = dyn_cast<StoreInst>(it);
1119 Type *StTy = VectorType::get(SI->getValueOperand()->getType(), VF);
1120 Value *Ptr = SI->getPointerOperand();
1121 unsigned Alignment = SI->getAlignment();
1123 assert(!Legal->isUniform(Ptr) &&
1124 "We do not allow storing to uniform addresses");
1126 GetElementPtrInst *Gep = dyn_cast<GetElementPtrInst>(Ptr);
1128 // This store does not use GEPs.
1129 if (!Legal->isConsecutivePtr(Ptr)) {
1130 scalarizeInstruction(it);
1135 // The last index does not have to be the induction. It can be
1136 // consecutive and be a function of the index. For example A[I+1];
1137 unsigned NumOperands = Gep->getNumOperands();
1138 Value *LastIndex = getVectorValue(Gep->getOperand(NumOperands - 1));
1139 LastIndex = Builder.CreateExtractElement(LastIndex, Zero);
1141 // Create the new GEP with the new induction variable.
1142 GetElementPtrInst *Gep2 = cast<GetElementPtrInst>(Gep->clone());
1143 Gep2->setOperand(NumOperands - 1, LastIndex);
1144 Ptr = Builder.Insert(Gep2);
1146 // Use the induction element ptr.
1147 assert(isa<PHINode>(Ptr) && "Invalid induction ptr");
1148 Ptr = Builder.CreateExtractElement(getVectorValue(Ptr), Zero);
1150 Ptr = Builder.CreateBitCast(Ptr, StTy->getPointerTo());
1151 Value *Val = getVectorValue(SI->getValueOperand());
1152 Builder.CreateStore(Val, Ptr)->setAlignment(Alignment);
1155 case Instruction::Load: {
1156 // Attempt to issue a wide load.
1157 LoadInst *LI = dyn_cast<LoadInst>(it);
1158 Type *RetTy = VectorType::get(LI->getType(), VF);
1159 Value *Ptr = LI->getPointerOperand();
1160 unsigned Alignment = LI->getAlignment();
1161 GetElementPtrInst *Gep = dyn_cast<GetElementPtrInst>(Ptr);
1163 // If the pointer is loop invariant or if it is non consecutive,
1164 // scalarize the load.
1165 bool Con = Legal->isConsecutivePtr(Ptr);
1166 if (Legal->isUniform(Ptr) || !Con) {
1167 scalarizeInstruction(it);
1172 // The last index does not have to be the induction. It can be
1173 // consecutive and be a function of the index. For example A[I+1];
1174 unsigned NumOperands = Gep->getNumOperands();
1175 Value *LastIndex = getVectorValue(Gep->getOperand(NumOperands -1));
1176 LastIndex = Builder.CreateExtractElement(LastIndex, Zero);
1178 // Create the new GEP with the new induction variable.
1179 GetElementPtrInst *Gep2 = cast<GetElementPtrInst>(Gep->clone());
1180 Gep2->setOperand(NumOperands - 1, LastIndex);
1181 Ptr = Builder.Insert(Gep2);
1183 // Use the induction element ptr.
1184 assert(isa<PHINode>(Ptr) && "Invalid induction ptr");
1185 Ptr = Builder.CreateExtractElement(getVectorValue(Ptr), Zero);
1188 Ptr = Builder.CreateBitCast(Ptr, RetTy->getPointerTo());
1189 LI = Builder.CreateLoad(Ptr);
1190 LI->setAlignment(Alignment);
1191 // Use this vector value for all users of the load.
1195 case Instruction::ZExt:
1196 case Instruction::SExt:
1197 case Instruction::FPToUI:
1198 case Instruction::FPToSI:
1199 case Instruction::FPExt:
1200 case Instruction::PtrToInt:
1201 case Instruction::IntToPtr:
1202 case Instruction::SIToFP:
1203 case Instruction::UIToFP:
1204 case Instruction::Trunc:
1205 case Instruction::FPTrunc:
1206 case Instruction::BitCast: {
1207 CastInst *CI = dyn_cast<CastInst>(it);
1208 /// Optimize the special case where the source is the induction
1209 /// variable. Notice that we can only optimize the 'trunc' case
1210 /// because: a. FP conversions lose precision, b. sext/zext may wrap,
1211 /// c. other casts depend on pointer size.
1212 if (CI->getOperand(0) == OldInduction &&
1213 it->getOpcode() == Instruction::Trunc) {
1214 Value *ScalarCast = Builder.CreateCast(CI->getOpcode(), Induction,
1216 Value *Broadcasted = getBroadcastInstrs(ScalarCast);
1217 WidenMap[it] = getConsecutiveVector(Broadcasted);
1220 /// Vectorize casts.
1221 Value *A = getVectorValue(it->getOperand(0));
1222 Type *DestTy = VectorType::get(CI->getType()->getScalarType(), VF);
1223 WidenMap[it] = Builder.CreateCast(CI->getOpcode(), A, DestTy);
1227 case Instruction::Call: {
1228 assert(isTriviallyVectorizableIntrinsic(it));
1229 Module *M = BB->getParent()->getParent();
1230 IntrinsicInst *II = cast<IntrinsicInst>(it);
1231 Intrinsic::ID ID = II->getIntrinsicID();
1232 SmallVector<Value*, 4> Args;
1233 for (unsigned i = 0, ie = II->getNumArgOperands(); i != ie; ++i)
1234 Args.push_back(getVectorValue(II->getArgOperand(i)));
1235 Type *Tys[] = { VectorType::get(II->getType()->getScalarType(), VF) };
1236 Function *F = Intrinsic::getDeclaration(M, ID, Tys);
1237 WidenMap[it] = Builder.CreateCall(F, Args);
1242 // All other instructions are unsupported. Scalarize them.
1243 scalarizeInstruction(it);
1246 }// end of for_each instr.
1249 void InnerLoopVectorizer::updateAnalysis() {
1250 // Forget the original basic block.
1251 SE->forgetLoop(OrigLoop);
1253 // Update the dominator tree information.
1254 assert(DT->properlyDominates(LoopBypassBlock, LoopExitBlock) &&
1255 "Entry does not dominate exit.");
1257 DT->addNewBlock(LoopVectorPreHeader, LoopBypassBlock);
1258 DT->addNewBlock(LoopVectorBody, LoopVectorPreHeader);
1259 DT->addNewBlock(LoopMiddleBlock, LoopBypassBlock);
1260 DT->addNewBlock(LoopScalarPreHeader, LoopMiddleBlock);
1261 DT->changeImmediateDominator(LoopScalarBody, LoopScalarPreHeader);
1262 DT->changeImmediateDominator(LoopExitBlock, LoopMiddleBlock);
1264 DEBUG(DT->verifyAnalysis());
1267 bool LoopVectorizationLegality::canVectorizeWithIfConvert() {
1268 if (!EnableIfConversion)
1271 assert(TheLoop->getNumBlocks() > 1 && "Single block loops are vectorizable");
1272 std::vector<BasicBlock*> &LoopBlocks = TheLoop->getBlocksVector();
1274 // Collect the blocks that need predication.
1275 for (unsigned i = 0, e = LoopBlocks.size(); i < e; ++i) {
1276 BasicBlock *BB = LoopBlocks[i];
1278 // We don't support switch statements inside loops.
1279 if (!isa<BranchInst>(BB->getTerminator()))
1282 // We must have at most two predecessors because we need to convert
1283 // all PHIs to selects.
1284 unsigned Preds = std::distance(pred_begin(BB), pred_end(BB));
1288 // We must be able to predicate all blocks that need to be predicated.
1289 if (blockNeedsPredication(BB) && !blockCanBePredicated(BB))
1293 // We can if-convert this loop.
1297 bool LoopVectorizationLegality::canVectorize() {
1298 assert(TheLoop->getLoopPreheader() && "No preheader!!");
1300 // We can only vectorize innermost loops.
1301 if (TheLoop->getSubLoopsVector().size())
1304 // We must have a single backedge.
1305 if (TheLoop->getNumBackEdges() != 1)
1308 // We must have a single exiting block.
1309 if (!TheLoop->getExitingBlock())
1312 unsigned NumBlocks = TheLoop->getNumBlocks();
1314 // Check if we can if-convert non single-bb loops.
1315 if (NumBlocks != 1 && !canVectorizeWithIfConvert()) {
1316 DEBUG(dbgs() << "LV: Can't if-convert the loop.\n");
1320 // We need to have a loop header.
1321 BasicBlock *Latch = TheLoop->getLoopLatch();
1322 DEBUG(dbgs() << "LV: Found a loop: " <<
1323 TheLoop->getHeader()->getName() << "\n");
1325 // ScalarEvolution needs to be able to find the exit count.
1326 const SCEV *ExitCount = SE->getExitCount(TheLoop, Latch);
1327 if (ExitCount == SE->getCouldNotCompute()) {
1328 DEBUG(dbgs() << "LV: SCEV could not compute the loop exit count.\n");
1332 // Do not loop-vectorize loops with a tiny trip count.
1333 unsigned TC = SE->getSmallConstantTripCount(TheLoop, Latch);
1334 if (TC > 0u && TC < TinyTripCountThreshold) {
1335 DEBUG(dbgs() << "LV: Found a loop with a very small trip count. " <<
1336 "This loop is not worth vectorizing.\n");
1340 // Check if we can vectorize the instructions and CFG in this loop.
1341 if (!canVectorizeInstrs()) {
1342 DEBUG(dbgs() << "LV: Can't vectorize the instructions or CFG\n");
1346 // Go over each instruction and look at memory deps.
1347 if (!canVectorizeMemory()) {
1348 DEBUG(dbgs() << "LV: Can't vectorize due to memory conflicts\n");
1352 // Collect all of the variables that remain uniform after vectorization.
1353 collectLoopUniforms();
1355 DEBUG(dbgs() << "LV: We can vectorize this loop" <<
1356 (PtrRtCheck.Need ? " (with a runtime bound check)" : "")
1359 // Okay! We can vectorize. At this point we don't have any other mem analysis
1360 // which may limit our maximum vectorization factor, so just return true with
1365 bool LoopVectorizationLegality::canVectorizeInstrs() {
1366 BasicBlock *PreHeader = TheLoop->getLoopPreheader();
1367 BasicBlock *Header = TheLoop->getHeader();
1369 // For each block in the loop.
1370 for (Loop::block_iterator bb = TheLoop->block_begin(),
1371 be = TheLoop->block_end(); bb != be; ++bb) {
1373 // Scan the instructions in the block and look for hazards.
1374 for (BasicBlock::iterator it = (*bb)->begin(), e = (*bb)->end(); it != e;
1377 if (PHINode *Phi = dyn_cast<PHINode>(it)) {
1378 // This should not happen because the loop should be normalized.
1379 if (Phi->getNumIncomingValues() != 2) {
1380 DEBUG(dbgs() << "LV: Found an invalid PHI.\n");
1384 // Check that this PHI type is allowed.
1385 if (!Phi->getType()->isIntegerTy() &&
1386 !Phi->getType()->isPointerTy()) {
1387 DEBUG(dbgs() << "LV: Found an non-int non-pointer PHI.\n");
1391 // If this PHINode is not in the header block, then we know that we
1392 // can convert it to select during if-conversion. No need to check if
1393 // the PHIs in this block are induction or reduction variables.
1397 // This is the value coming from the preheader.
1398 Value *StartValue = Phi->getIncomingValueForBlock(PreHeader);
1399 // Check if this is an induction variable.
1400 InductionKind IK = isInductionVariable(Phi);
1402 if (NoInduction != IK) {
1403 // Int inductions are special because we only allow one IV.
1404 if (IK == IntInduction) {
1406 DEBUG(dbgs() << "LV: Found too many inductions."<< *Phi <<"\n");
1412 DEBUG(dbgs() << "LV: Found an induction variable.\n");
1413 Inductions[Phi] = InductionInfo(StartValue, IK);
1417 if (AddReductionVar(Phi, IntegerAdd)) {
1418 DEBUG(dbgs() << "LV: Found an ADD reduction PHI."<< *Phi <<"\n");
1421 if (AddReductionVar(Phi, IntegerMult)) {
1422 DEBUG(dbgs() << "LV: Found a MUL reduction PHI."<< *Phi <<"\n");
1425 if (AddReductionVar(Phi, IntegerOr)) {
1426 DEBUG(dbgs() << "LV: Found an OR reduction PHI."<< *Phi <<"\n");
1429 if (AddReductionVar(Phi, IntegerAnd)) {
1430 DEBUG(dbgs() << "LV: Found an AND reduction PHI."<< *Phi <<"\n");
1433 if (AddReductionVar(Phi, IntegerXor)) {
1434 DEBUG(dbgs() << "LV: Found a XOR reduction PHI."<< *Phi <<"\n");
1438 DEBUG(dbgs() << "LV: Found an unidentified PHI."<< *Phi <<"\n");
1440 }// end of PHI handling
1442 // We still don't handle functions.
1443 CallInst *CI = dyn_cast<CallInst>(it);
1444 if (CI && !isTriviallyVectorizableIntrinsic(it)) {
1445 DEBUG(dbgs() << "LV: Found a call site.\n");
1449 // We do not re-vectorize vectors.
1450 if (!VectorType::isValidElementType(it->getType()) &&
1451 !it->getType()->isVoidTy()) {
1452 DEBUG(dbgs() << "LV: Found unvectorizable type." << "\n");
1456 // Reduction instructions are allowed to have exit users.
1457 // All other instructions must not have external users.
1458 if (!AllowedExit.count(it))
1459 //Check that all of the users of the loop are inside the BB.
1460 for (Value::use_iterator I = it->use_begin(), E = it->use_end();
1462 Instruction *U = cast<Instruction>(*I);
1463 // This user may be a reduction exit value.
1464 if (!TheLoop->contains(U)) {
1465 DEBUG(dbgs() << "LV: Found an outside user for : "<< *U << "\n");
1474 DEBUG(dbgs() << "LV: Did not find one integer induction var.\n");
1475 assert(getInductionVars()->size() && "No induction variables");
1481 void LoopVectorizationLegality::collectLoopUniforms() {
1482 // We now know that the loop is vectorizable!
1483 // Collect variables that will remain uniform after vectorization.
1484 std::vector<Value*> Worklist;
1485 BasicBlock *Latch = TheLoop->getLoopLatch();
1487 // Start with the conditional branch and walk up the block.
1488 Worklist.push_back(Latch->getTerminator()->getOperand(0));
1490 while (Worklist.size()) {
1491 Instruction *I = dyn_cast<Instruction>(Worklist.back());
1492 Worklist.pop_back();
1494 // Look at instructions inside this loop.
1495 // Stop when reaching PHI nodes.
1496 // TODO: we need to follow values all over the loop, not only in this block.
1497 if (!I || !TheLoop->contains(I) || isa<PHINode>(I))
1500 // This is a known uniform.
1503 // Insert all operands.
1504 for (int i = 0, Op = I->getNumOperands(); i < Op; ++i) {
1505 Worklist.push_back(I->getOperand(i));
1510 bool LoopVectorizationLegality::canVectorizeMemory() {
1511 typedef SmallVector<Value*, 16> ValueVector;
1512 typedef SmallPtrSet<Value*, 16> ValueSet;
1513 // Holds the Load and Store *instructions*.
1516 PtrRtCheck.Pointers.clear();
1517 PtrRtCheck.Need = false;
1520 for (Loop::block_iterator bb = TheLoop->block_begin(),
1521 be = TheLoop->block_end(); bb != be; ++bb) {
1523 // Scan the BB and collect legal loads and stores.
1524 for (BasicBlock::iterator it = (*bb)->begin(), e = (*bb)->end(); it != e;
1527 // If this is a load, save it. If this instruction can read from memory
1528 // but is not a load, then we quit. Notice that we don't handle function
1529 // calls that read or write.
1530 if (it->mayReadFromMemory()) {
1531 LoadInst *Ld = dyn_cast<LoadInst>(it);
1532 if (!Ld) return false;
1533 if (!Ld->isSimple()) {
1534 DEBUG(dbgs() << "LV: Found a non-simple load.\n");
1537 Loads.push_back(Ld);
1541 // Save 'store' instructions. Abort if other instructions write to memory.
1542 if (it->mayWriteToMemory()) {
1543 StoreInst *St = dyn_cast<StoreInst>(it);
1544 if (!St) return false;
1545 if (!St->isSimple()) {
1546 DEBUG(dbgs() << "LV: Found a non-simple store.\n");
1549 Stores.push_back(St);
1554 // Now we have two lists that hold the loads and the stores.
1555 // Next, we find the pointers that they use.
1557 // Check if we see any stores. If there are no stores, then we don't
1558 // care if the pointers are *restrict*.
1559 if (!Stores.size()) {
1560 DEBUG(dbgs() << "LV: Found a read-only loop!\n");
1564 // Holds the read and read-write *pointers* that we find.
1566 ValueVector ReadWrites;
1568 // Holds the analyzed pointers. We don't want to call GetUnderlyingObjects
1569 // multiple times on the same object. If the ptr is accessed twice, once
1570 // for read and once for write, it will only appear once (on the write
1571 // list). This is okay, since we are going to check for conflicts between
1572 // writes and between reads and writes, but not between reads and reads.
1575 ValueVector::iterator I, IE;
1576 for (I = Stores.begin(), IE = Stores.end(); I != IE; ++I) {
1577 StoreInst *ST = dyn_cast<StoreInst>(*I);
1578 assert(ST && "Bad StoreInst");
1579 Value* Ptr = ST->getPointerOperand();
1581 if (isUniform(Ptr)) {
1582 DEBUG(dbgs() << "LV: We don't allow storing to uniform addresses\n");
1586 // If we did *not* see this pointer before, insert it to
1587 // the read-write list. At this phase it is only a 'write' list.
1588 if (Seen.insert(Ptr))
1589 ReadWrites.push_back(Ptr);
1592 for (I = Loads.begin(), IE = Loads.end(); I != IE; ++I) {
1593 LoadInst *LD = dyn_cast<LoadInst>(*I);
1594 assert(LD && "Bad LoadInst");
1595 Value* Ptr = LD->getPointerOperand();
1596 // If we did *not* see this pointer before, insert it to the
1597 // read list. If we *did* see it before, then it is already in
1598 // the read-write list. This allows us to vectorize expressions
1599 // such as A[i] += x; Because the address of A[i] is a read-write
1600 // pointer. This only works if the index of A[i] is consecutive.
1601 // If the address of i is unknown (for example A[B[i]]) then we may
1602 // read a few words, modify, and write a few words, and some of the
1603 // words may be written to the same address.
1604 if (Seen.insert(Ptr) || !isConsecutivePtr(Ptr))
1605 Reads.push_back(Ptr);
1608 // If we write (or read-write) to a single destination and there are no
1609 // other reads in this loop then is it safe to vectorize.
1610 if (ReadWrites.size() == 1 && Reads.size() == 0) {
1611 DEBUG(dbgs() << "LV: Found a write-only loop!\n");
1615 // Find pointers with computable bounds. We are going to use this information
1616 // to place a runtime bound check.
1618 for (I = ReadWrites.begin(), IE = ReadWrites.end(); I != IE; ++I)
1619 if (hasComputableBounds(*I)) {
1620 PtrRtCheck.insert(SE, TheLoop, *I);
1621 DEBUG(dbgs() << "LV: Found a runtime check ptr:" << **I <<"\n");
1626 for (I = Reads.begin(), IE = Reads.end(); I != IE; ++I)
1627 if (hasComputableBounds(*I)) {
1628 PtrRtCheck.insert(SE, TheLoop, *I);
1629 DEBUG(dbgs() << "LV: Found a runtime check ptr:" << **I <<"\n");
1635 // Check that we did not collect too many pointers or found a
1636 // unsizeable pointer.
1637 if (!RT || PtrRtCheck.Pointers.size() > RuntimeMemoryCheckThreshold) {
1642 PtrRtCheck.Need = RT;
1645 DEBUG(dbgs() << "LV: We can perform a memory runtime check if needed.\n");
1648 // Now that the pointers are in two lists (Reads and ReadWrites), we
1649 // can check that there are no conflicts between each of the writes and
1650 // between the writes to the reads.
1651 ValueSet WriteObjects;
1652 ValueVector TempObjects;
1654 // Check that the read-writes do not conflict with other read-write
1656 for (I = ReadWrites.begin(), IE = ReadWrites.end(); I != IE; ++I) {
1657 GetUnderlyingObjects(*I, TempObjects, DL);
1658 for (ValueVector::iterator it=TempObjects.begin(), e=TempObjects.end();
1660 if (!isIdentifiedObject(*it)) {
1661 DEBUG(dbgs() << "LV: Found an unidentified write ptr:"<< **it <<"\n");
1664 if (!WriteObjects.insert(*it)) {
1665 DEBUG(dbgs() << "LV: Found a possible write-write reorder:"
1670 TempObjects.clear();
1673 /// Check that the reads don't conflict with the read-writes.
1674 for (I = Reads.begin(), IE = Reads.end(); I != IE; ++I) {
1675 GetUnderlyingObjects(*I, TempObjects, DL);
1676 for (ValueVector::iterator it=TempObjects.begin(), e=TempObjects.end();
1678 if (!isIdentifiedObject(*it)) {
1679 DEBUG(dbgs() << "LV: Found an unidentified read ptr:"<< **it <<"\n");
1682 if (WriteObjects.count(*it)) {
1683 DEBUG(dbgs() << "LV: Found a possible read/write reorder:"
1688 TempObjects.clear();
1691 // It is safe to vectorize and we don't need any runtime checks.
1692 DEBUG(dbgs() << "LV: We don't need a runtime memory check.\n");
1697 bool LoopVectorizationLegality::AddReductionVar(PHINode *Phi,
1698 ReductionKind Kind) {
1699 if (Phi->getNumIncomingValues() != 2)
1702 // Reduction variables are only found in the loop header block.
1703 if (Phi->getParent() != TheLoop->getHeader())
1706 // Obtain the reduction start value from the value that comes from the loop
1708 Value *RdxStart = Phi->getIncomingValueForBlock(TheLoop->getLoopPreheader());
1710 // ExitInstruction is the single value which is used outside the loop.
1711 // We only allow for a single reduction value to be used outside the loop.
1712 // This includes users of the reduction, variables (which form a cycle
1713 // which ends in the phi node).
1714 Instruction *ExitInstruction = 0;
1716 // Iter is our iterator. We start with the PHI node and scan for all of the
1717 // users of this instruction. All users must be instructions which can be
1718 // used as reduction variables (such as ADD). We may have a single
1719 // out-of-block user. They cycle must end with the original PHI.
1720 // Also, we can't have multiple block-local users.
1721 Instruction *Iter = Phi;
1723 // If the instruction has no users then this is a broken
1724 // chain and can't be a reduction variable.
1725 if (Iter->use_empty())
1728 // Any reduction instr must be of one of the allowed kinds.
1729 if (!isReductionInstr(Iter, Kind))
1732 // Did we find a user inside this block ?
1733 bool FoundInBlockUser = false;
1734 // Did we reach the initial PHI node ?
1735 bool FoundStartPHI = false;
1737 // For each of the *users* of iter.
1738 for (Value::use_iterator it = Iter->use_begin(), e = Iter->use_end();
1740 Instruction *U = cast<Instruction>(*it);
1741 // We already know that the PHI is a user.
1743 FoundStartPHI = true;
1747 // Check if we found the exit user.
1748 BasicBlock *Parent = U->getParent();
1749 if (!TheLoop->contains(Parent)) {
1750 // Exit if you find multiple outside users.
1751 if (ExitInstruction != 0)
1753 ExitInstruction = Iter;
1756 // We allow in-loop PHINodes which are not the original reduction PHI
1757 // node. If this PHI is the only user of Iter (happens in IF w/ no ELSE
1758 // structure) then don't skip this PHI.
1759 if (isa<PHINode>(U) && U->getParent() != TheLoop->getHeader() &&
1760 TheLoop->contains(U) && Iter->getNumUses() > 1)
1763 // We can't have multiple inside users.
1764 if (FoundInBlockUser)
1766 FoundInBlockUser = true;
1770 // We found a reduction var if we have reached the original
1771 // phi node and we only have a single instruction with out-of-loop
1773 if (FoundStartPHI && ExitInstruction) {
1774 // This instruction is allowed to have out-of-loop users.
1775 AllowedExit.insert(ExitInstruction);
1777 // Save the description of this reduction variable.
1778 ReductionDescriptor RD(RdxStart, ExitInstruction, Kind);
1779 Reductions[Phi] = RD;
1783 // If we've reached the start PHI but did not find an outside user then
1784 // this is dead code. Abort.
1791 LoopVectorizationLegality::isReductionInstr(Instruction *I,
1792 ReductionKind Kind) {
1793 switch (I->getOpcode()) {
1796 case Instruction::PHI:
1799 case Instruction::Add:
1800 case Instruction::Sub:
1801 return Kind == IntegerAdd;
1802 case Instruction::Mul:
1803 return Kind == IntegerMult;
1804 case Instruction::And:
1805 return Kind == IntegerAnd;
1806 case Instruction::Or:
1807 return Kind == IntegerOr;
1808 case Instruction::Xor:
1809 return Kind == IntegerXor;
1813 LoopVectorizationLegality::InductionKind
1814 LoopVectorizationLegality::isInductionVariable(PHINode *Phi) {
1815 Type *PhiTy = Phi->getType();
1816 // We only handle integer and pointer inductions variables.
1817 if (!PhiTy->isIntegerTy() && !PhiTy->isPointerTy())
1820 // Check that the PHI is consecutive and starts at zero.
1821 const SCEV *PhiScev = SE->getSCEV(Phi);
1822 const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(PhiScev);
1824 DEBUG(dbgs() << "LV: PHI is not a poly recurrence.\n");
1827 const SCEV *Step = AR->getStepRecurrence(*SE);
1829 // Integer inductions need to have a stride of one.
1830 if (PhiTy->isIntegerTy()) {
1832 return IntInduction;
1833 if (Step->isAllOnesValue())
1834 return ReverseIntInduction;
1838 // Calculate the pointer stride and check if it is consecutive.
1839 const SCEVConstant *C = dyn_cast<SCEVConstant>(Step);
1843 assert(PhiTy->isPointerTy() && "The PHI must be a pointer");
1844 uint64_t Size = DL->getTypeAllocSize(PhiTy->getPointerElementType());
1845 if (C->getValue()->equalsInt(Size))
1846 return PtrInduction;
1851 bool LoopVectorizationLegality::blockNeedsPredication(BasicBlock *BB) {
1852 assert(TheLoop->contains(BB) && "Unknown block used");
1854 // Blocks that do not dominate the latch need predication.
1855 BasicBlock* Latch = TheLoop->getLoopLatch();
1856 return !DT->dominates(BB, Latch);
1859 bool LoopVectorizationLegality::blockCanBePredicated(BasicBlock *BB) {
1860 for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; ++it) {
1861 // We don't predicate loads/stores at the moment.
1862 if (it->mayReadFromMemory() || it->mayWriteToMemory() || it->mayThrow())
1865 // The isntructions below can trap.
1866 switch (it->getOpcode()) {
1868 case Instruction::UDiv:
1869 case Instruction::SDiv:
1870 case Instruction::URem:
1871 case Instruction::SRem:
1879 bool LoopVectorizationLegality::hasComputableBounds(Value *Ptr) {
1880 const SCEV *PhiScev = SE->getSCEV(Ptr);
1881 const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(PhiScev);
1885 return AR->isAffine();
1889 LoopVectorizationCostModel::findBestVectorizationFactor(unsigned VF) {
1891 DEBUG(dbgs() << "LV: No vector target information. Not vectorizing. \n");
1895 float Cost = expectedCost(1);
1897 DEBUG(dbgs() << "LV: Scalar loop costs: "<< (int)Cost << ".\n");
1898 for (unsigned i=2; i <= VF; i*=2) {
1899 // Notice that the vector loop needs to be executed less times, so
1900 // we need to divide the cost of the vector loops by the width of
1901 // the vector elements.
1902 float VectorCost = expectedCost(i) / (float)i;
1903 DEBUG(dbgs() << "LV: Vector loop of width "<< i << " costs: " <<
1904 (int)VectorCost << ".\n");
1905 if (VectorCost < Cost) {
1911 DEBUG(dbgs() << "LV: Selecting VF = : "<< Width << ".\n");
1915 unsigned LoopVectorizationCostModel::expectedCost(unsigned VF) {
1919 for (Loop::block_iterator bb = TheLoop->block_begin(),
1920 be = TheLoop->block_end(); bb != be; ++bb) {
1921 unsigned BlockCost = 0;
1922 BasicBlock *BB = *bb;
1924 // For each instruction in the old loop.
1925 for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; ++it) {
1926 unsigned C = getInstructionCost(it, VF);
1928 DEBUG(dbgs() << "LV: Found an estimated cost of "<< C <<" for VF " <<
1929 VF << " For instruction: "<< *it << "\n");
1932 // We assume that if-converted blocks have a 50% chance of being executed.
1933 // When the code is scalar then some of the blocks are avoided due to CF.
1934 // When the code is vectorized we execute all code paths.
1935 if (Legal->blockNeedsPredication(*bb) && VF == 1)
1945 LoopVectorizationCostModel::getInstructionCost(Instruction *I, unsigned VF) {
1946 assert(VTTI && "Invalid vector target transformation info");
1948 // If we know that this instruction will remain uniform, check the cost of
1949 // the scalar version.
1950 if (Legal->isUniformAfterVectorization(I))
1953 Type *RetTy = I->getType();
1954 Type *VectorTy = ToVectorTy(RetTy, VF);
1956 // TODO: We need to estimate the cost of intrinsic calls.
1957 switch (I->getOpcode()) {
1958 case Instruction::GetElementPtr:
1959 // We mark this instruction as zero-cost because scalar GEPs are usually
1960 // lowered to the intruction addressing mode. At the moment we don't
1961 // generate vector geps.
1963 case Instruction::Br: {
1964 return VTTI->getCFInstrCost(I->getOpcode());
1966 case Instruction::PHI:
1967 //TODO: IF-converted IFs become selects.
1969 case Instruction::Add:
1970 case Instruction::FAdd:
1971 case Instruction::Sub:
1972 case Instruction::FSub:
1973 case Instruction::Mul:
1974 case Instruction::FMul:
1975 case Instruction::UDiv:
1976 case Instruction::SDiv:
1977 case Instruction::FDiv:
1978 case Instruction::URem:
1979 case Instruction::SRem:
1980 case Instruction::FRem:
1981 case Instruction::Shl:
1982 case Instruction::LShr:
1983 case Instruction::AShr:
1984 case Instruction::And:
1985 case Instruction::Or:
1986 case Instruction::Xor:
1987 return VTTI->getArithmeticInstrCost(I->getOpcode(), VectorTy);
1988 case Instruction::Select: {
1989 SelectInst *SI = cast<SelectInst>(I);
1990 const SCEV *CondSCEV = SE->getSCEV(SI->getCondition());
1991 bool ScalarCond = (SE->isLoopInvariant(CondSCEV, TheLoop));
1992 Type *CondTy = SI->getCondition()->getType();
1994 CondTy = VectorType::get(CondTy, VF);
1996 return VTTI->getCmpSelInstrCost(I->getOpcode(), VectorTy, CondTy);
1998 case Instruction::ICmp:
1999 case Instruction::FCmp: {
2000 Type *ValTy = I->getOperand(0)->getType();
2001 VectorTy = ToVectorTy(ValTy, VF);
2002 return VTTI->getCmpSelInstrCost(I->getOpcode(), VectorTy);
2004 case Instruction::Store: {
2005 StoreInst *SI = cast<StoreInst>(I);
2006 Type *ValTy = SI->getValueOperand()->getType();
2007 VectorTy = ToVectorTy(ValTy, VF);
2010 return VTTI->getMemoryOpCost(I->getOpcode(), ValTy,
2012 SI->getPointerAddressSpace());
2014 // Scalarized stores.
2015 if (!Legal->isConsecutivePtr(SI->getPointerOperand())) {
2017 unsigned ExtCost = VTTI->getInstrCost(Instruction::ExtractElement,
2019 // The cost of extracting from the value vector.
2020 Cost += VF * (ExtCost);
2021 // The cost of the scalar stores.
2022 Cost += VF * VTTI->getMemoryOpCost(I->getOpcode(),
2023 ValTy->getScalarType(),
2025 SI->getPointerAddressSpace());
2030 return VTTI->getMemoryOpCost(I->getOpcode(), VectorTy, SI->getAlignment(),
2031 SI->getPointerAddressSpace());
2033 case Instruction::Load: {
2034 LoadInst *LI = cast<LoadInst>(I);
2037 return VTTI->getMemoryOpCost(I->getOpcode(), RetTy,
2039 LI->getPointerAddressSpace());
2041 // Scalarized loads.
2042 if (!Legal->isConsecutivePtr(LI->getPointerOperand())) {
2044 unsigned InCost = VTTI->getInstrCost(Instruction::InsertElement, RetTy);
2045 // The cost of inserting the loaded value into the result vector.
2046 Cost += VF * (InCost);
2047 // The cost of the scalar stores.
2048 Cost += VF * VTTI->getMemoryOpCost(I->getOpcode(),
2049 RetTy->getScalarType(),
2051 LI->getPointerAddressSpace());
2056 return VTTI->getMemoryOpCost(I->getOpcode(), VectorTy, LI->getAlignment(),
2057 LI->getPointerAddressSpace());
2059 case Instruction::ZExt:
2060 case Instruction::SExt:
2061 case Instruction::FPToUI:
2062 case Instruction::FPToSI:
2063 case Instruction::FPExt:
2064 case Instruction::PtrToInt:
2065 case Instruction::IntToPtr:
2066 case Instruction::SIToFP:
2067 case Instruction::UIToFP:
2068 case Instruction::Trunc:
2069 case Instruction::FPTrunc:
2070 case Instruction::BitCast: {
2071 Type *SrcVecTy = ToVectorTy(I->getOperand(0)->getType(), VF);
2072 return VTTI->getCastInstrCost(I->getOpcode(), VectorTy, SrcVecTy);
2074 case Instruction::Call: {
2075 assert(isTriviallyVectorizableIntrinsic(I));
2076 IntrinsicInst *II = cast<IntrinsicInst>(I);
2077 Type *RetTy = ToVectorTy(II->getType(), VF);
2078 SmallVector<Type*, 4> Tys;
2079 for (unsigned i = 0, ie = II->getNumArgOperands(); i != ie; ++i)
2080 Tys.push_back(ToVectorTy(II->getArgOperand(i)->getType(), VF));
2081 return VTTI->getIntrinsicInstrCost(II->getIntrinsicID(), RetTy, Tys);
2084 // We are scalarizing the instruction. Return the cost of the scalar
2085 // instruction, plus the cost of insert and extract into vector
2086 // elements, times the vector width.
2089 bool IsVoid = RetTy->isVoidTy();
2091 unsigned InsCost = (IsVoid ? 0 :
2092 VTTI->getInstrCost(Instruction::InsertElement,
2095 unsigned ExtCost = VTTI->getInstrCost(Instruction::ExtractElement,
2098 // The cost of inserting the results plus extracting each one of the
2100 Cost += VF * (InsCost + ExtCost * I->getNumOperands());
2102 // The cost of executing VF copies of the scalar instruction.
2103 Cost += VF * VTTI->getInstrCost(I->getOpcode(), RetTy);
2109 Type* LoopVectorizationCostModel::ToVectorTy(Type *Scalar, unsigned VF) {
2110 if (Scalar->isVoidTy() || VF == 1)
2112 return VectorType::get(Scalar, VF);
2115 char LoopVectorize::ID = 0;
2116 static const char lv_name[] = "Loop Vectorization";
2117 INITIALIZE_PASS_BEGIN(LoopVectorize, LV_NAME, lv_name, false, false)
2118 INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
2119 INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)
2120 INITIALIZE_PASS_DEPENDENCY(LoopSimplify)
2121 INITIALIZE_PASS_END(LoopVectorize, LV_NAME, lv_name, false, false)
2124 Pass *createLoopVectorizePass() {
2125 return new LoopVectorize();