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 //===----------------------------------------------------------------------===//
10 // This is the LLVM loop vectorizer. This pass modifies 'vectorizable' loops
11 // and generates target-independent LLVM-IR. Legalization of the IR is done
12 // in the codegen. However, the vectorizes uses (will use) the codegen
13 // interfaces to generate IR that is likely to result in an optimal binary.
15 // The loop vectorizer combines consecutive loop iteration into a single
16 // 'wide' iteration. After this transformation the index is incremented
17 // by the SIMD vector width, and not by one.
19 // This pass has three parts:
20 // 1. The main loop pass that drives the different parts.
21 // 2. LoopVectorizationLegality - A unit that checks for the legality
22 // of the vectorization.
23 // 3. SingleBlockLoopVectorizer - A unit that performs the actual
24 // widening of instructions.
25 // 4. LoopVectorizationCostModel - A unit that checks for the profitability
26 // of vectorization. It decides on the optimal vector width, which
27 // can be one, if vectorization is not profitable.
28 //===----------------------------------------------------------------------===//
30 // The reduction-variable vectorization is based on the paper:
31 // D. Nuzman and R. Henderson. Multi-platform Auto-vectorization.
33 // Variable uniformity checks are inspired by:
34 // Karrenberg, R. and Hack, S. Whole Function Vectorization.
36 // Other ideas/concepts are from:
37 // A. Zaks and D. Nuzman. Autovectorization in GCC-two years later.
39 //===----------------------------------------------------------------------===//
40 #define LV_NAME "loop-vectorize"
41 #define DEBUG_TYPE LV_NAME
42 #include "llvm/Constants.h"
43 #include "llvm/DerivedTypes.h"
44 #include "llvm/Instructions.h"
45 #include "llvm/LLVMContext.h"
46 #include "llvm/Pass.h"
47 #include "llvm/Analysis/LoopPass.h"
48 #include "llvm/Value.h"
49 #include "llvm/Function.h"
50 #include "llvm/Analysis/Verifier.h"
51 #include "llvm/Module.h"
52 #include "llvm/Type.h"
53 #include "llvm/ADT/SmallVector.h"
54 #include "llvm/ADT/StringExtras.h"
55 #include "llvm/Analysis/AliasAnalysis.h"
56 #include "llvm/Analysis/AliasSetTracker.h"
57 #include "llvm/Analysis/ScalarEvolution.h"
58 #include "llvm/Analysis/Dominators.h"
59 #include "llvm/Analysis/ScalarEvolutionExpressions.h"
60 #include "llvm/Analysis/ScalarEvolutionExpander.h"
61 #include "llvm/Analysis/LoopInfo.h"
62 #include "llvm/Analysis/ValueTracking.h"
63 #include "llvm/Transforms/Scalar.h"
64 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
65 #include "llvm/TargetTransformInfo.h"
66 #include "llvm/Support/CommandLine.h"
67 #include "llvm/Support/Debug.h"
68 #include "llvm/Support/raw_ostream.h"
69 #include "llvm/DataLayout.h"
70 #include "llvm/Transforms/Utils/Local.h"
74 static cl::opt<unsigned>
75 VectorizationFactor("force-vector-width", cl::init(0), cl::Hidden,
76 cl::desc("Set the default vectorization width. Zero is autoselect."));
78 /// We don't vectorize loops with a known constant trip count below this number.
79 const unsigned TinyTripCountThreshold = 16;
81 /// When performing a runtime memory check, do not check more than this
82 /// number of pointers. Notice that the check is quadratic!
83 const unsigned RuntimeMemoryCheckThreshold = 2;
87 // Forward declarations.
88 class LoopVectorizationLegality;
89 class LoopVectorizationCostModel;
91 /// SingleBlockLoopVectorizer vectorizes loops which contain only one basic
92 /// block to a specified vectorization factor (VF).
93 /// This class performs the widening of scalars into vectors, or multiple
94 /// scalars. This class also implements the following features:
95 /// * It inserts an epilogue loop for handling loops that don't have iteration
96 /// counts that are known to be a multiple of the vectorization factor.
97 /// * It handles the code generation for reduction variables.
98 /// * Scalarization (implementation using scalars) of un-vectorizable
100 /// SingleBlockLoopVectorizer does not perform any vectorization-legality
101 /// checks, and relies on the caller to check for the different legality
102 /// aspects. The SingleBlockLoopVectorizer relies on the
103 /// LoopVectorizationLegality class to provide information about the induction
104 /// and reduction variables that were found to a given vectorization factor.
105 class SingleBlockLoopVectorizer {
108 SingleBlockLoopVectorizer(Loop *Orig, ScalarEvolution *Se, LoopInfo *Li,
109 DominatorTree *dt, LPPassManager *Lpm,
111 OrigLoop(Orig), SE(Se), LI(Li), DT(dt), LPM(Lpm), VF(VecWidth),
112 Builder(Se->getContext()), Induction(0), OldInduction(0) { }
114 // Perform the actual loop widening (vectorization).
115 void vectorize(LoopVectorizationLegality *Legal) {
116 ///Create a new empty loop. Unlink the old loop and connect the new one.
117 createEmptyLoop(Legal);
118 /// Widen each instruction in the old loop to a new one in the new loop.
119 /// Use the Legality module to find the induction and reduction variables.
120 vectorizeLoop(Legal);
121 // Register the new loop and update the analysis passes.
126 /// Create an empty loop, based on the loop ranges of the old loop.
127 void createEmptyLoop(LoopVectorizationLegality *Legal);
128 /// Copy and widen the instructions from the old loop.
129 void vectorizeLoop(LoopVectorizationLegality *Legal);
130 /// Insert the new loop to the loop hierarchy and pass manager
131 /// and update the analysis passes.
132 void updateAnalysis();
134 /// This instruction is un-vectorizable. Implement it as a sequence
136 void scalarizeInstruction(Instruction *Instr);
138 /// Create a broadcast instruction. This method generates a broadcast
139 /// instruction (shuffle) for loop invariant values and for the induction
140 /// value. If this is the induction variable then we extend it to N, N+1, ...
141 /// this is needed because each iteration in the loop corresponds to a SIMD
143 Value *getBroadcastInstrs(Value *V);
145 /// This is a helper function used by getBroadcastInstrs. It adds 0, 1, 2 ..
146 /// for each element in the vector. Starting from zero.
147 Value *getConsecutiveVector(Value* Val);
149 /// When we go over instructions in the basic block we rely on previous
150 /// values within the current basic block or on loop invariant values.
151 /// When we widen (vectorize) values we place them in the map. If the values
152 /// are not within the map, they have to be loop invariant, so we simply
153 /// broadcast them into a vector.
154 Value *getVectorValue(Value *V);
156 /// Get a uniform vector of constant integers. We use this to get
157 /// vectors of ones and zeros for the reduction code.
158 Constant* getUniformVector(unsigned Val, Type* ScalarTy);
160 typedef DenseMap<Value*, Value*> ValueMap;
162 /// The original loop.
164 // Scev analysis to use.
170 // Loop Pass Manager;
172 // The vectorization factor to use.
175 // The builder that we use
178 // --- Vectorization state ---
180 /// The vector-loop preheader.
181 BasicBlock *LoopVectorPreHeader;
182 /// The scalar-loop preheader.
183 BasicBlock *LoopScalarPreHeader;
184 /// Middle Block between the vector and the scalar.
185 BasicBlock *LoopMiddleBlock;
186 ///The ExitBlock of the scalar loop.
187 BasicBlock *LoopExitBlock;
188 ///The vector loop body.
189 BasicBlock *LoopVectorBody;
190 ///The scalar loop body.
191 BasicBlock *LoopScalarBody;
192 ///The first bypass block.
193 BasicBlock *LoopBypassBlock;
195 /// The new Induction variable which was added to the new block.
197 /// The induction variable of the old basic block.
198 PHINode *OldInduction;
199 // Maps scalars to widened vectors.
203 /// LoopVectorizationLegality checks if it is legal to vectorize a loop, and
204 /// to what vectorization factor.
205 /// This class does not look at the profitability of vectorization, only the
206 /// legality. This class has two main kinds of checks:
207 /// * Memory checks - The code in canVectorizeMemory checks if vectorization
208 /// will change the order of memory accesses in a way that will change the
209 /// correctness of the program.
210 /// * Scalars checks - The code in canVectorizeBlock checks for a number
211 /// of different conditions, such as the availability of a single induction
212 /// variable, that all types are supported and vectorize-able, etc.
213 /// This code reflects the capabilities of SingleBlockLoopVectorizer.
214 /// This class is also used by SingleBlockLoopVectorizer for identifying
215 /// induction variable and the different reduction variables.
216 class LoopVectorizationLegality {
218 LoopVectorizationLegality(Loop *Lp, ScalarEvolution *Se, DataLayout *Dl):
219 TheLoop(Lp), SE(Se), DL(Dl), Induction(0) { }
221 /// This represents the kinds of reductions that we support.
223 NoReduction, /// Not a reduction.
224 IntegerAdd, /// Sum of numbers.
225 IntegerMult, /// Product of numbers.
226 IntegerOr, /// Bitwise or logical OR of numbers.
227 IntegerAnd, /// Bitwise or logical AND of numbers.
228 IntegerXor /// Bitwise or logical XOR of numbers.
231 /// This POD struct holds information about reduction variables.
232 struct ReductionDescriptor {
234 ReductionDescriptor():
235 StartValue(0), LoopExitInstr(0), Kind(NoReduction) {}
238 ReductionDescriptor(Value *Start, Instruction *Exit, ReductionKind K):
239 StartValue(Start), LoopExitInstr(Exit), Kind(K) {}
241 // The starting value of the reduction.
242 // It does not have to be zero!
244 // The instruction who's value is used outside the loop.
245 Instruction *LoopExitInstr;
246 // The kind of the reduction.
250 // This POD struct holds information about the memory runtime legality
251 // check that a group of pointers do not overlap.
252 struct RuntimePointerCheck {
253 /// This flag indicates if we need to add the runtime check.
255 /// Holds the pointers that we need to check.
256 SmallVector<Value*, 2> Pointers;
259 /// ReductionList contains the reduction descriptors for all
260 /// of the reductions that were found in the loop.
261 typedef DenseMap<PHINode*, ReductionDescriptor> ReductionList;
263 /// Returns true if it is legal to vectorize this loop.
264 /// This does not mean that it is profitable to vectorize this
265 /// loop, only that it is legal to do so.
268 /// Returns the Induction variable.
269 PHINode *getInduction() {return Induction;}
271 /// Returns the reduction variables found in the loop.
272 ReductionList *getReductionVars() { return &Reductions; }
274 /// Check if the pointer returned by this GEP is consecutive
275 /// when the index is vectorized. This happens when the last
276 /// index of the GEP is consecutive, like the induction variable.
277 /// This check allows us to vectorize A[idx] into a wide load/store.
278 bool isConsecutiveGep(Value *Ptr);
280 /// Returns true if the value V is uniform within the loop.
281 bool isUniform(Value *V);
283 /// Returns true if this instruction will remain scalar after vectorization.
284 bool isUniformAfterVectorization(Instruction* I) {return Uniforms.count(I);}
286 /// Returns the information that we collected about runtime memory check.
287 RuntimePointerCheck *getRuntimePointerCheck() {return &PtrRtCheck; }
289 /// Check if a single basic block loop is vectorizable.
290 /// At this point we know that this is a loop with a constant trip count
291 /// and we only need to check individual instructions.
292 bool canVectorizeBlock(BasicBlock &BB);
294 /// When we vectorize loops we may change the order in which
295 /// we read and write from memory. This method checks if it is
296 /// legal to vectorize the code, considering only memory constrains.
297 /// Returns true if BB is vectorizable
298 bool canVectorizeMemory(BasicBlock &BB);
300 /// Returns True, if 'Phi' is the kind of reduction variable for type
301 /// 'Kind'. If this is a reduction variable, it adds it to ReductionList.
302 bool AddReductionVar(PHINode *Phi, ReductionKind Kind);
303 /// Returns true if the instruction I can be a reduction variable of type
305 bool isReductionInstr(Instruction *I, ReductionKind Kind);
306 /// Returns True, if 'Phi' is an induction variable.
307 bool isInductionVariable(PHINode *Phi);
308 /// Return true if can compute the address bounds of Ptr within the loop.
309 bool hasComputableBounds(Value *Ptr);
311 /// The loop that we evaluate.
315 /// DataLayout analysis.
318 // --- vectorization state --- //
320 /// Holds the induction variable.
322 /// Holds the reduction variables.
323 ReductionList Reductions;
324 /// Allowed outside users. This holds the reduction
325 /// vars which can be accessed from outside the loop.
326 SmallPtrSet<Value*, 4> AllowedExit;
327 /// This set holds the variables which are known to be uniform after
329 SmallPtrSet<Instruction*, 4> Uniforms;
330 /// We need to check that all of the pointers in this list are disjoint
332 RuntimePointerCheck PtrRtCheck;
335 /// LoopVectorizationCostModel - estimates the expected speedups due to
337 /// In many cases vectorization is not profitable. This can happen because
338 /// of a number of reasons. In this class we mainly attempt to predict
339 /// the expected speedup/slowdowns due to the supported instruction set.
340 /// We use the VectorTargetTransformInfo to query the different backends
341 /// for the cost of different operations.
342 class LoopVectorizationCostModel {
345 LoopVectorizationCostModel(Loop *Lp, ScalarEvolution *Se,
346 LoopVectorizationLegality *Leg,
347 const VectorTargetTransformInfo *Vtti):
348 TheLoop(Lp), SE(Se), Legal(Leg), VTTI(Vtti) { }
350 /// Returns the most profitable vectorization factor for the loop that is
351 /// smaller or equal to the VF argument. This method checks every power
353 unsigned findBestVectorizationFactor(unsigned VF = 8);
356 /// Returns the expected execution cost. The unit of the cost does
357 /// not matter because we use the 'cost' units to compare different
358 /// vector widths. The cost that is returned is *not* normalized by
359 /// the factor width.
360 unsigned expectedCost(unsigned VF);
362 /// Returns the execution time cost of an instruction for a given vector
363 /// width. Vector width of one means scalar.
364 unsigned getInstructionCost(Instruction *I, unsigned VF);
366 /// A helper function for converting Scalar types to vector types.
367 /// If the incoming type is void, we return void. If the VF is 1, we return
369 static Type* ToVectorTy(Type *Scalar, unsigned VF);
371 /// The loop that we evaluate.
376 /// Vectorization legality.
377 LoopVectorizationLegality *Legal;
378 /// Vector target information.
379 const VectorTargetTransformInfo *VTTI;
382 struct LoopVectorize : public LoopPass {
383 static char ID; // Pass identification, replacement for typeid
385 LoopVectorize() : LoopPass(ID) {
386 initializeLoopVectorizePass(*PassRegistry::getPassRegistry());
392 TargetTransformInfo *TTI;
395 virtual bool runOnLoop(Loop *L, LPPassManager &LPM) {
396 // We only vectorize innermost loops.
400 SE = &getAnalysis<ScalarEvolution>();
401 DL = getAnalysisIfAvailable<DataLayout>();
402 LI = &getAnalysis<LoopInfo>();
403 TTI = getAnalysisIfAvailable<TargetTransformInfo>();
404 DT = &getAnalysis<DominatorTree>();
406 DEBUG(dbgs() << "LV: Checking a loop in \"" <<
407 L->getHeader()->getParent()->getName() << "\"\n");
409 // Check if it is legal to vectorize the loop.
410 LoopVectorizationLegality LVL(L, SE, DL);
411 if (!LVL.canVectorize()) {
412 DEBUG(dbgs() << "LV: Not vectorizing.\n");
416 // Select the preffered vectorization factor.
418 if (VectorizationFactor == 0) {
419 const VectorTargetTransformInfo *VTTI = 0;
421 VTTI = TTI->getVectorTargetTransformInfo();
422 // Use the cost model.
423 LoopVectorizationCostModel CM(L, SE, &LVL, VTTI);
424 VF = CM.findBestVectorizationFactor();
427 DEBUG(dbgs() << "LV: Vectorization is possible but not beneficial.\n");
432 // Use the user command flag.
433 VF = VectorizationFactor;
436 DEBUG(dbgs() << "LV: Found a vectorizable loop ("<< VF << ") in "<<
437 L->getHeader()->getParent()->getParent()->getModuleIdentifier()<<
440 // If we decided that it is *legal* to vectorizer the loop then do it.
441 SingleBlockLoopVectorizer LB(L, SE, LI, DT, &LPM, VF);
444 DEBUG(verifyFunction(*L->getHeader()->getParent()));
448 virtual void getAnalysisUsage(AnalysisUsage &AU) const {
449 LoopPass::getAnalysisUsage(AU);
450 AU.addRequiredID(LoopSimplifyID);
451 AU.addRequiredID(LCSSAID);
452 AU.addRequired<LoopInfo>();
453 AU.addRequired<ScalarEvolution>();
454 AU.addRequired<DominatorTree>();
455 AU.addPreserved<LoopInfo>();
456 AU.addPreserved<DominatorTree>();
461 Value *SingleBlockLoopVectorizer::getBroadcastInstrs(Value *V) {
462 // Instructions that access the old induction variable
463 // actually want to get the new one.
464 if (V == OldInduction)
467 LLVMContext &C = V->getContext();
468 Type *VTy = VectorType::get(V->getType(), VF);
469 Type *I32 = IntegerType::getInt32Ty(C);
470 Constant *Zero = ConstantInt::get(I32, 0);
471 Value *Zeros = ConstantAggregateZero::get(VectorType::get(I32, VF));
472 Value *UndefVal = UndefValue::get(VTy);
473 // Insert the value into a new vector.
474 Value *SingleElem = Builder.CreateInsertElement(UndefVal, V, Zero);
475 // Broadcast the scalar into all locations in the vector.
476 Value *Shuf = Builder.CreateShuffleVector(SingleElem, UndefVal, Zeros,
478 // We are accessing the induction variable. Make sure to promote the
479 // index for each consecutive SIMD lane. This adds 0,1,2 ... to all lanes.
481 return getConsecutiveVector(Shuf);
485 Value *SingleBlockLoopVectorizer::getConsecutiveVector(Value* Val) {
486 assert(Val->getType()->isVectorTy() && "Must be a vector");
487 assert(Val->getType()->getScalarType()->isIntegerTy() &&
488 "Elem must be an integer");
490 Type *ITy = Val->getType()->getScalarType();
491 VectorType *Ty = cast<VectorType>(Val->getType());
492 unsigned VLen = Ty->getNumElements();
493 SmallVector<Constant*, 8> Indices;
495 // Create a vector of consecutive numbers from zero to VF.
496 for (unsigned i = 0; i < VLen; ++i)
497 Indices.push_back(ConstantInt::get(ITy, i));
499 // Add the consecutive indices to the vector value.
500 Constant *Cv = ConstantVector::get(Indices);
501 assert(Cv->getType() == Val->getType() && "Invalid consecutive vec");
502 return Builder.CreateAdd(Val, Cv, "induction");
505 bool LoopVectorizationLegality::isConsecutiveGep(Value *Ptr) {
506 GetElementPtrInst *Gep = dyn_cast_or_null<GetElementPtrInst>(Ptr);
510 unsigned NumOperands = Gep->getNumOperands();
511 Value *LastIndex = Gep->getOperand(NumOperands - 1);
513 // Check that all of the gep indices are uniform except for the last.
514 for (unsigned i = 0; i < NumOperands - 1; ++i)
515 if (!SE->isLoopInvariant(SE->getSCEV(Gep->getOperand(i)), TheLoop))
518 // We can emit wide load/stores only of the last index is the induction
520 const SCEV *Last = SE->getSCEV(LastIndex);
521 if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(Last)) {
522 const SCEV *Step = AR->getStepRecurrence(*SE);
524 // The memory is consecutive because the last index is consecutive
525 // and all other indices are loop invariant.
533 bool LoopVectorizationLegality::isUniform(Value *V) {
534 return (SE->isLoopInvariant(SE->getSCEV(V), TheLoop));
537 Value *SingleBlockLoopVectorizer::getVectorValue(Value *V) {
538 assert(!V->getType()->isVectorTy() && "Can't widen a vector");
539 // If we saved a vectorized copy of V, use it.
540 Value *&MapEntry = WidenMap[V];
544 // Broadcast V and save the value for future uses.
545 Value *B = getBroadcastInstrs(V);
551 SingleBlockLoopVectorizer::getUniformVector(unsigned Val, Type* ScalarTy) {
552 SmallVector<Constant*, 8> Indices;
553 // Create a vector of consecutive numbers from zero to VF.
554 for (unsigned i = 0; i < VF; ++i)
555 Indices.push_back(ConstantInt::get(ScalarTy, Val, true));
557 // Add the consecutive indices to the vector value.
558 return ConstantVector::get(Indices);
561 void SingleBlockLoopVectorizer::scalarizeInstruction(Instruction *Instr) {
562 assert(!Instr->getType()->isAggregateType() && "Can't handle vectors");
563 // Holds vector parameters or scalars, in case of uniform vals.
564 SmallVector<Value*, 8> Params;
566 // Find all of the vectorized parameters.
567 for (unsigned op = 0, e = Instr->getNumOperands(); op != e; ++op) {
568 Value *SrcOp = Instr->getOperand(op);
570 // If we are accessing the old induction variable, use the new one.
571 if (SrcOp == OldInduction) {
572 Params.push_back(getBroadcastInstrs(Induction));
576 // Try using previously calculated values.
577 Instruction *SrcInst = dyn_cast<Instruction>(SrcOp);
579 // If the src is an instruction that appeared earlier in the basic block
580 // then it should already be vectorized.
581 if (SrcInst && SrcInst->getParent() == Instr->getParent()) {
582 assert(WidenMap.count(SrcInst) && "Source operand is unavailable");
583 // The parameter is a vector value from earlier.
584 Params.push_back(WidenMap[SrcInst]);
586 // The parameter is a scalar from outside the loop. Maybe even a constant.
587 Params.push_back(SrcOp);
591 assert(Params.size() == Instr->getNumOperands() &&
592 "Invalid number of operands");
594 // Does this instruction return a value ?
595 bool IsVoidRetTy = Instr->getType()->isVoidTy();
596 Value *VecResults = 0;
598 // If we have a return value, create an empty vector. We place the scalarized
599 // instructions in this vector.
601 VecResults = UndefValue::get(VectorType::get(Instr->getType(), VF));
603 // For each scalar that we create:
604 for (unsigned i = 0; i < VF; ++i) {
605 Instruction *Cloned = Instr->clone();
607 Cloned->setName(Instr->getName() + ".cloned");
608 // Replace the operands of the cloned instrucions with extracted scalars.
609 for (unsigned op = 0, e = Instr->getNumOperands(); op != e; ++op) {
610 Value *Op = Params[op];
611 // Param is a vector. Need to extract the right lane.
612 if (Op->getType()->isVectorTy())
613 Op = Builder.CreateExtractElement(Op, Builder.getInt32(i));
614 Cloned->setOperand(op, Op);
617 // Place the cloned scalar in the new loop.
618 Builder.Insert(Cloned);
620 // If the original scalar returns a value we need to place it in a vector
621 // so that future users will be able to use it.
623 VecResults = Builder.CreateInsertElement(VecResults, Cloned,
624 Builder.getInt32(i));
628 WidenMap[Instr] = VecResults;
632 SingleBlockLoopVectorizer::createEmptyLoop(LoopVectorizationLegality *Legal) {
634 In this function we generate a new loop. The new loop will contain
635 the vectorized instructions while the old loop will continue to run the
638 [ ] <-- vector loop bypass.
641 | [ ] <-- vector pre header.
645 | [ ]_| <-- vector loop.
648 >[ ] <--- middle-block.
651 | [ ] <--- new preheader.
655 | [ ]_| <-- old scalar loop to handle remainder.
662 OldInduction = Legal->getInduction();
663 assert(OldInduction && "We must have a single phi node.");
664 Type *IdxTy = OldInduction->getType();
666 // Find the loop boundaries.
667 const SCEV *ExitCount = SE->getExitCount(OrigLoop, OrigLoop->getHeader());
668 assert(ExitCount != SE->getCouldNotCompute() && "Invalid loop count");
670 // Get the total trip count from the count by adding 1.
671 ExitCount = SE->getAddExpr(ExitCount,
672 SE->getConstant(ExitCount->getType(), 1));
673 // We may need to extend the index in case there is a type mismatch.
674 // We know that the count starts at zero and does not overflow.
675 // We are using Zext because it should be less expensive.
676 if (ExitCount->getType() != IdxTy)
677 ExitCount = SE->getZeroExtendExpr(ExitCount, IdxTy);
679 // This is the original scalar-loop preheader.
680 BasicBlock *BypassBlock = OrigLoop->getLoopPreheader();
681 BasicBlock *ExitBlock = OrigLoop->getExitBlock();
682 assert(ExitBlock && "Must have an exit block");
684 // The loop index does not have to start at Zero. It starts with this value.
685 Value *StartIdx = OldInduction->getIncomingValueForBlock(BypassBlock);
687 assert(OrigLoop->getNumBlocks() == 1 && "Invalid loop");
688 assert(BypassBlock && "Invalid loop structure");
690 BasicBlock *VectorPH =
691 BypassBlock->splitBasicBlock(BypassBlock->getTerminator(), "vector.ph");
692 BasicBlock *VecBody = VectorPH->splitBasicBlock(VectorPH->getTerminator(),
695 BasicBlock *MiddleBlock = VecBody->splitBasicBlock(VecBody->getTerminator(),
697 BasicBlock *ScalarPH =
698 MiddleBlock->splitBasicBlock(MiddleBlock->getTerminator(),
700 // Find the induction variable.
701 BasicBlock *OldBasicBlock = OrigLoop->getHeader();
703 // Use this IR builder to create the loop instructions (Phi, Br, Cmp)
705 Builder.SetInsertPoint(VecBody->getFirstInsertionPt());
707 // Generate the induction variable.
708 Induction = Builder.CreatePHI(IdxTy, 2, "index");
709 Constant *Step = ConstantInt::get(IdxTy, VF);
711 // Expand the trip count and place the new instructions in the preheader.
712 // Notice that the pre-header does not change, only the loop body.
713 SCEVExpander Exp(*SE, "induction");
714 Instruction *Loc = BypassBlock->getTerminator();
716 // Count holds the overall loop count (N).
717 Value *Count = Exp.expandCodeFor(ExitCount, Induction->getType(), Loc);
719 // Add the start index to the loop count to get the new end index.
720 Value *IdxEnd = BinaryOperator::CreateAdd(Count, StartIdx, "end.idx", Loc);
722 // Now we need to generate the expression for N - (N % VF), which is
723 // the part that the vectorized body will execute.
724 Constant *CIVF = ConstantInt::get(IdxTy, VF);
725 Value *R = BinaryOperator::CreateURem(Count, CIVF, "n.mod.vf", Loc);
726 Value *CountRoundDown = BinaryOperator::CreateSub(Count, R, "n.vec", Loc);
727 Value *IdxEndRoundDown = BinaryOperator::CreateAdd(CountRoundDown, StartIdx,
728 "end.idx.rnd.down", Loc);
730 // Now, compare the new count to zero. If it is zero, jump to the scalar part.
731 Value *Cmp = CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_EQ,
736 LoopVectorizationLegality::RuntimePointerCheck *PtrRtCheck =
737 Legal->getRuntimePointerCheck();
738 Value *MemoryRuntimeCheck = 0;
739 if (PtrRtCheck->Need) {
740 unsigned NumPointers = PtrRtCheck->Pointers.size();
741 SmallVector<Value* , 2> Starts;
742 SmallVector<Value* , 2> Ends;
744 // Use this type for pointer arithmetic.
745 Type* PtrArithTy = PtrRtCheck->Pointers[0]->getType();
747 for (unsigned i=0; i < NumPointers; ++i) {
748 Value *Ptr = PtrRtCheck->Pointers[i];
749 const SCEV *Sc = SE->getSCEV(Ptr);
751 if (SE->isLoopInvariant(Sc, OrigLoop)) {
752 DEBUG(dbgs() << "LV1: Adding RT check for a loop invariant ptr:" <<
754 Starts.push_back(Ptr);
757 DEBUG(dbgs() << "LV: Adding RT check for range:" << *Ptr <<"\n");
758 const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(Sc);
759 Value *Start = Exp.expandCodeFor(AR->getStart(), PtrArithTy, Loc);
760 const SCEV *Ex = SE->getExitCount(OrigLoop, OrigLoop->getHeader());
761 const SCEV *ScEnd = AR->evaluateAtIteration(Ex, *SE);
762 assert(!isa<SCEVCouldNotCompute>(ScEnd) && "Invalid scev range.");
763 Value *End = Exp.expandCodeFor(ScEnd, PtrArithTy, Loc);
764 Starts.push_back(Start);
769 for (unsigned i=0; i < NumPointers; ++i) {
770 for (unsigned j=i+1; j < NumPointers; ++j) {
771 Value *Cmp0 = CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_ULE,
772 Starts[0], Ends[1], "bound0", Loc);
773 Value *Cmp1 = CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_ULE,
774 Starts[1], Ends[0], "bound1", Loc);
775 Value *IsConflict = BinaryOperator::Create(Instruction::And, Cmp0, Cmp1,
776 "found.conflict", Loc);
777 if (MemoryRuntimeCheck) {
778 MemoryRuntimeCheck = BinaryOperator::Create(Instruction::Or,
781 "conflict.rdx", Loc);
783 MemoryRuntimeCheck = IsConflict;
787 }// end of need-runtime-check code.
789 // If we are using memory runtime checks, include them in.
790 if (MemoryRuntimeCheck) {
791 Cmp = BinaryOperator::Create(Instruction::Or, Cmp, MemoryRuntimeCheck,
795 BranchInst::Create(MiddleBlock, VectorPH, Cmp, Loc);
796 // Remove the old terminator.
797 Loc->eraseFromParent();
799 // We are going to resume the execution of the scalar loop.
800 // This PHI decides on what number to start. If we come from the
801 // vector loop then we need to start with the end index minus the
802 // index modulo VF. If we come from a bypass edge then we need to start
803 // from the real start.
804 PHINode* ResumeIndex = PHINode::Create(IdxTy, 2, "resume.idx",
805 MiddleBlock->getTerminator());
806 ResumeIndex->addIncoming(StartIdx, BypassBlock);
807 ResumeIndex->addIncoming(IdxEndRoundDown, VecBody);
809 // Add a check in the middle block to see if we have completed
810 // all of the iterations in the first vector loop.
811 // If (N - N%VF) == N, then we *don't* need to run the remainder.
812 Value *CmpN = CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_EQ, IdxEnd,
813 ResumeIndex, "cmp.n",
814 MiddleBlock->getTerminator());
816 BranchInst::Create(ExitBlock, ScalarPH, CmpN, MiddleBlock->getTerminator());
817 // Remove the old terminator.
818 MiddleBlock->getTerminator()->eraseFromParent();
820 // Create i+1 and fill the PHINode.
821 Value *NextIdx = Builder.CreateAdd(Induction, Step, "index.next");
822 Induction->addIncoming(StartIdx, VectorPH);
823 Induction->addIncoming(NextIdx, VecBody);
824 // Create the compare.
825 Value *ICmp = Builder.CreateICmpEQ(NextIdx, IdxEndRoundDown);
826 Builder.CreateCondBr(ICmp, MiddleBlock, VecBody);
828 // Now we have two terminators. Remove the old one from the block.
829 VecBody->getTerminator()->eraseFromParent();
831 // Fix the scalar body iteration count.
832 unsigned BlockIdx = OldInduction->getBasicBlockIndex(ScalarPH);
833 OldInduction->setIncomingValue(BlockIdx, ResumeIndex);
835 // Get ready to start creating new instructions into the vectorized body.
836 Builder.SetInsertPoint(VecBody->getFirstInsertionPt());
838 // Register the new loop.
839 Loop* Lp = new Loop();
840 LPM->insertLoop(Lp, OrigLoop->getParentLoop());
842 Lp->addBasicBlockToLoop(VecBody, LI->getBase());
844 Loop *ParentLoop = OrigLoop->getParentLoop();
846 ParentLoop->addBasicBlockToLoop(ScalarPH, LI->getBase());
847 ParentLoop->addBasicBlockToLoop(VectorPH, LI->getBase());
848 ParentLoop->addBasicBlockToLoop(MiddleBlock, LI->getBase());
852 LoopVectorPreHeader = VectorPH;
853 LoopScalarPreHeader = ScalarPH;
854 LoopMiddleBlock = MiddleBlock;
855 LoopExitBlock = ExitBlock;
856 LoopVectorBody = VecBody;
857 LoopScalarBody = OldBasicBlock;
858 LoopBypassBlock = BypassBlock;
861 /// This function returns the identity element (or neutral element) for
864 getReductionIdentity(LoopVectorizationLegality::ReductionKind K) {
866 case LoopVectorizationLegality::IntegerXor:
867 case LoopVectorizationLegality::IntegerAdd:
868 case LoopVectorizationLegality::IntegerOr:
869 // Adding, Xoring, Oring zero to a number does not change it.
871 case LoopVectorizationLegality::IntegerMult:
872 // Multiplying a number by 1 does not change it.
874 case LoopVectorizationLegality::IntegerAnd:
875 // AND-ing a number with an all-1 value does not change it.
878 llvm_unreachable("Unknown reduction kind");
883 SingleBlockLoopVectorizer::vectorizeLoop(LoopVectorizationLegality *Legal) {
884 //===------------------------------------------------===//
886 // Notice: any optimization or new instruction that go
887 // into the code below should be also be implemented in
890 //===------------------------------------------------===//
891 typedef SmallVector<PHINode*, 4> PhiVector;
892 BasicBlock &BB = *OrigLoop->getHeader();
893 Constant *Zero = ConstantInt::get(
894 IntegerType::getInt32Ty(BB.getContext()), 0);
896 // In order to support reduction variables we need to be able to vectorize
897 // Phi nodes. Phi nodes have cycles, so we need to vectorize them in two
898 // steages. First, we create a new vector PHI node with no incoming edges.
899 // We use this value when we vectorize all of the instructions that use the
900 // PHI. Next, after all of the instructions in the block are complete we
901 // add the new incoming edges to the PHI. At this point all of the
902 // instructions in the basic block are vectorized, so we can use them to
903 // construct the PHI.
906 // For each instruction in the old loop.
907 for (BasicBlock::iterator it = BB.begin(), e = BB.end(); it != e; ++it) {
908 Instruction *Inst = it;
910 switch (Inst->getOpcode()) {
911 case Instruction::Br:
912 // Nothing to do for PHIs and BR, since we already took care of the
913 // loop control flow instructions.
915 case Instruction::PHI:{
916 PHINode* P = cast<PHINode>(Inst);
917 // Special handling for the induction var.
918 if (OldInduction == Inst)
920 // This is phase one of vectorizing PHIs.
921 // This has to be a reduction variable.
922 assert(Legal->getReductionVars()->count(P) && "Not a Reduction");
923 Type *VecTy = VectorType::get(Inst->getType(), VF);
924 WidenMap[Inst] = Builder.CreatePHI(VecTy, 2, "vec.phi");
925 PHIsToFix.push_back(P);
928 case Instruction::Add:
929 case Instruction::FAdd:
930 case Instruction::Sub:
931 case Instruction::FSub:
932 case Instruction::Mul:
933 case Instruction::FMul:
934 case Instruction::UDiv:
935 case Instruction::SDiv:
936 case Instruction::FDiv:
937 case Instruction::URem:
938 case Instruction::SRem:
939 case Instruction::FRem:
940 case Instruction::Shl:
941 case Instruction::LShr:
942 case Instruction::AShr:
943 case Instruction::And:
944 case Instruction::Or:
945 case Instruction::Xor: {
946 // Just widen binops.
947 BinaryOperator *BinOp = dyn_cast<BinaryOperator>(Inst);
948 Value *A = getVectorValue(Inst->getOperand(0));
949 Value *B = getVectorValue(Inst->getOperand(1));
951 // Use this vector value for all users of the original instruction.
952 Value *V = Builder.CreateBinOp(BinOp->getOpcode(), A, B);
955 // Update the NSW, NUW and Exact flags.
956 BinaryOperator *VecOp = cast<BinaryOperator>(V);
957 if (isa<OverflowingBinaryOperator>(BinOp)) {
958 VecOp->setHasNoSignedWrap(BinOp->hasNoSignedWrap());
959 VecOp->setHasNoUnsignedWrap(BinOp->hasNoUnsignedWrap());
961 if (isa<PossiblyExactOperator>(VecOp))
962 VecOp->setIsExact(BinOp->isExact());
965 case Instruction::Select: {
967 // If the selector is loop invariant we can create a select
968 // instruction with a scalar condition. Otherwise, use vector-select.
969 Value *Cond = Inst->getOperand(0);
970 bool InvariantCond = SE->isLoopInvariant(SE->getSCEV(Cond), OrigLoop);
972 // The condition can be loop invariant but still defined inside the
973 // loop. This means that we can't just use the original 'cond' value.
974 // We have to take the 'vectorized' value and pick the first lane.
975 // Instcombine will make this a no-op.
976 Cond = getVectorValue(Cond);
978 Cond = Builder.CreateExtractElement(Cond, Builder.getInt32(0));
980 Value *Op0 = getVectorValue(Inst->getOperand(1));
981 Value *Op1 = getVectorValue(Inst->getOperand(2));
982 WidenMap[Inst] = Builder.CreateSelect(Cond, Op0, Op1);
986 case Instruction::ICmp:
987 case Instruction::FCmp: {
988 // Widen compares. Generate vector compares.
989 bool FCmp = (Inst->getOpcode() == Instruction::FCmp);
990 CmpInst *Cmp = dyn_cast<CmpInst>(Inst);
991 Value *A = getVectorValue(Inst->getOperand(0));
992 Value *B = getVectorValue(Inst->getOperand(1));
994 WidenMap[Inst] = Builder.CreateFCmp(Cmp->getPredicate(), A, B);
996 WidenMap[Inst] = Builder.CreateICmp(Cmp->getPredicate(), A, B);
1000 case Instruction::Store: {
1001 // Attempt to issue a wide store.
1002 StoreInst *SI = dyn_cast<StoreInst>(Inst);
1003 Type *StTy = VectorType::get(SI->getValueOperand()->getType(), VF);
1004 Value *Ptr = SI->getPointerOperand();
1005 unsigned Alignment = SI->getAlignment();
1007 assert(!Legal->isUniform(Ptr) &&
1008 "We do not allow storing to uniform addresses");
1010 GetElementPtrInst *Gep = dyn_cast<GetElementPtrInst>(Ptr);
1012 // This store does not use GEPs.
1013 if (!Legal->isConsecutiveGep(Gep)) {
1014 scalarizeInstruction(Inst);
1018 // The last index does not have to be the induction. It can be
1019 // consecutive and be a function of the index. For example A[I+1];
1020 unsigned NumOperands = Gep->getNumOperands();
1021 Value *LastIndex = getVectorValue(Gep->getOperand(NumOperands - 1));
1022 LastIndex = Builder.CreateExtractElement(LastIndex, Zero);
1024 // Create the new GEP with the new induction variable.
1025 GetElementPtrInst *Gep2 = cast<GetElementPtrInst>(Gep->clone());
1026 Gep2->setOperand(NumOperands - 1, LastIndex);
1027 Ptr = Builder.Insert(Gep2);
1028 Ptr = Builder.CreateBitCast(Ptr, StTy->getPointerTo());
1029 Value *Val = getVectorValue(SI->getValueOperand());
1030 Builder.CreateStore(Val, Ptr)->setAlignment(Alignment);
1033 case Instruction::Load: {
1034 // Attempt to issue a wide load.
1035 LoadInst *LI = dyn_cast<LoadInst>(Inst);
1036 Type *RetTy = VectorType::get(LI->getType(), VF);
1037 Value *Ptr = LI->getPointerOperand();
1038 unsigned Alignment = LI->getAlignment();
1039 GetElementPtrInst *Gep = dyn_cast<GetElementPtrInst>(Ptr);
1041 // If we don't have a gep, or that the pointer is loop invariant,
1042 // scalarize the load.
1043 if (!Gep || Legal->isUniform(Gep) || !Legal->isConsecutiveGep(Gep)) {
1044 scalarizeInstruction(Inst);
1048 // The last index does not have to be the induction. It can be
1049 // consecutive and be a function of the index. For example A[I+1];
1050 unsigned NumOperands = Gep->getNumOperands();
1051 Value *LastIndex = getVectorValue(Gep->getOperand(NumOperands -1));
1052 LastIndex = Builder.CreateExtractElement(LastIndex, Zero);
1054 // Create the new GEP with the new induction variable.
1055 GetElementPtrInst *Gep2 = cast<GetElementPtrInst>(Gep->clone());
1056 Gep2->setOperand(NumOperands - 1, LastIndex);
1057 Ptr = Builder.Insert(Gep2);
1058 Ptr = Builder.CreateBitCast(Ptr, RetTy->getPointerTo());
1059 LI = Builder.CreateLoad(Ptr);
1060 LI->setAlignment(Alignment);
1061 // Use this vector value for all users of the load.
1062 WidenMap[Inst] = LI;
1065 case Instruction::ZExt:
1066 case Instruction::SExt:
1067 case Instruction::FPToUI:
1068 case Instruction::FPToSI:
1069 case Instruction::FPExt:
1070 case Instruction::PtrToInt:
1071 case Instruction::IntToPtr:
1072 case Instruction::SIToFP:
1073 case Instruction::UIToFP:
1074 case Instruction::Trunc:
1075 case Instruction::FPTrunc:
1076 case Instruction::BitCast: {
1077 /// Vectorize bitcasts.
1078 CastInst *CI = dyn_cast<CastInst>(Inst);
1079 Value *A = getVectorValue(Inst->getOperand(0));
1080 Type *DestTy = VectorType::get(CI->getType()->getScalarType(), VF);
1081 WidenMap[Inst] = Builder.CreateCast(CI->getOpcode(), A, DestTy);
1086 /// All other instructions are unsupported. Scalarize them.
1087 scalarizeInstruction(Inst);
1090 }// end of for_each instr.
1092 // At this point every instruction in the original loop is widended to
1093 // a vector form. We are almost done. Now, we need to fix the PHI nodes
1094 // that we vectorized. The PHI nodes are currently empty because we did
1095 // not want to introduce cycles. Notice that the remaining PHI nodes
1096 // that we need to fix are reduction variables.
1098 // Create the 'reduced' values for each of the induction vars.
1099 // The reduced values are the vector values that we scalarize and combine
1100 // after the loop is finished.
1101 for (PhiVector::iterator it = PHIsToFix.begin(), e = PHIsToFix.end();
1103 PHINode *RdxPhi = *it;
1104 PHINode *VecRdxPhi = dyn_cast<PHINode>(WidenMap[RdxPhi]);
1105 assert(RdxPhi && "Unable to recover vectorized PHI");
1107 // Find the reduction variable descriptor.
1108 assert(Legal->getReductionVars()->count(RdxPhi) &&
1109 "Unable to find the reduction variable");
1110 LoopVectorizationLegality::ReductionDescriptor RdxDesc =
1111 (*Legal->getReductionVars())[RdxPhi];
1113 // We need to generate a reduction vector from the incoming scalar.
1114 // To do so, we need to generate the 'identity' vector and overide
1115 // one of the elements with the incoming scalar reduction. We need
1116 // to do it in the vector-loop preheader.
1117 Builder.SetInsertPoint(LoopBypassBlock->getTerminator());
1119 // This is the vector-clone of the value that leaves the loop.
1120 Value *VectorExit = getVectorValue(RdxDesc.LoopExitInstr);
1121 Type *VecTy = VectorExit->getType();
1123 // Find the reduction identity variable. Zero for addition, or, xor,
1124 // one for multiplication, -1 for And.
1125 Constant *Identity = getUniformVector(getReductionIdentity(RdxDesc.Kind),
1126 VecTy->getScalarType());
1128 // This vector is the Identity vector where the first element is the
1129 // incoming scalar reduction.
1130 Value *VectorStart = Builder.CreateInsertElement(Identity,
1131 RdxDesc.StartValue, Zero);
1134 // Fix the vector-loop phi.
1135 // We created the induction variable so we know that the
1136 // preheader is the first entry.
1137 BasicBlock *VecPreheader = Induction->getIncomingBlock(0);
1139 // Reductions do not have to start at zero. They can start with
1140 // any loop invariant values.
1141 VecRdxPhi->addIncoming(VectorStart, VecPreheader);
1142 unsigned SelfEdgeIdx = (RdxPhi)->getBasicBlockIndex(LoopScalarBody);
1143 Value *Val = getVectorValue(RdxPhi->getIncomingValue(SelfEdgeIdx));
1144 VecRdxPhi->addIncoming(Val, LoopVectorBody);
1146 // Before each round, move the insertion point right between
1147 // the PHIs and the values we are going to write.
1148 // This allows us to write both PHINodes and the extractelement
1150 Builder.SetInsertPoint(LoopMiddleBlock->getFirstInsertionPt());
1152 // This PHINode contains the vectorized reduction variable, or
1153 // the initial value vector, if we bypass the vector loop.
1154 PHINode *NewPhi = Builder.CreatePHI(VecTy, 2, "rdx.vec.exit.phi");
1155 NewPhi->addIncoming(VectorStart, LoopBypassBlock);
1156 NewPhi->addIncoming(getVectorValue(RdxDesc.LoopExitInstr), LoopVectorBody);
1158 // Extract the first scalar.
1160 Builder.CreateExtractElement(NewPhi, Builder.getInt32(0));
1161 // Extract and reduce the remaining vector elements.
1162 for (unsigned i=1; i < VF; ++i) {
1164 Builder.CreateExtractElement(NewPhi, Builder.getInt32(i));
1165 switch (RdxDesc.Kind) {
1166 case LoopVectorizationLegality::IntegerAdd:
1167 Scalar0 = Builder.CreateAdd(Scalar0, Scalar1);
1169 case LoopVectorizationLegality::IntegerMult:
1170 Scalar0 = Builder.CreateMul(Scalar0, Scalar1);
1172 case LoopVectorizationLegality::IntegerOr:
1173 Scalar0 = Builder.CreateOr(Scalar0, Scalar1);
1175 case LoopVectorizationLegality::IntegerAnd:
1176 Scalar0 = Builder.CreateAnd(Scalar0, Scalar1);
1178 case LoopVectorizationLegality::IntegerXor:
1179 Scalar0 = Builder.CreateXor(Scalar0, Scalar1);
1182 llvm_unreachable("Unknown reduction operation");
1186 // Now, we need to fix the users of the reduction variable
1187 // inside and outside of the scalar remainder loop.
1188 // We know that the loop is in LCSSA form. We need to update the
1189 // PHI nodes in the exit blocks.
1190 for (BasicBlock::iterator LEI = LoopExitBlock->begin(),
1191 LEE = LoopExitBlock->end(); LEI != LEE; ++LEI) {
1192 PHINode *LCSSAPhi = dyn_cast<PHINode>(LEI);
1193 if (!LCSSAPhi) continue;
1195 // All PHINodes need to have a single entry edge, or two if
1196 // we already fixed them.
1197 assert(LCSSAPhi->getNumIncomingValues() < 3 && "Invalid LCSSA PHI");
1199 // We found our reduction value exit-PHI. Update it with the
1200 // incoming bypass edge.
1201 if (LCSSAPhi->getIncomingValue(0) == RdxDesc.LoopExitInstr) {
1202 // Add an edge coming from the bypass.
1203 LCSSAPhi->addIncoming(Scalar0, LoopMiddleBlock);
1206 }// end of the LCSSA phi scan.
1208 // Fix the scalar loop reduction variable with the incoming reduction sum
1209 // from the vector body and from the backedge value.
1210 int IncomingEdgeBlockIdx = (RdxPhi)->getBasicBlockIndex(LoopScalarBody);
1211 int SelfEdgeBlockIdx = (IncomingEdgeBlockIdx ? 0 : 1); // The other block.
1212 (RdxPhi)->setIncomingValue(SelfEdgeBlockIdx, Scalar0);
1213 (RdxPhi)->setIncomingValue(IncomingEdgeBlockIdx, RdxDesc.LoopExitInstr);
1214 }// end of for each redux variable.
1217 void SingleBlockLoopVectorizer::updateAnalysis() {
1218 // The original basic block.
1219 SE->forgetLoop(OrigLoop);
1221 // Update the dominator tree information.
1222 assert(DT->properlyDominates(LoopBypassBlock, LoopExitBlock) &&
1223 "Entry does not dominate exit.");
1225 DT->addNewBlock(LoopVectorPreHeader, LoopBypassBlock);
1226 DT->addNewBlock(LoopVectorBody, LoopVectorPreHeader);
1227 DT->addNewBlock(LoopMiddleBlock, LoopBypassBlock);
1228 DT->addNewBlock(LoopScalarPreHeader, LoopMiddleBlock);
1229 DT->changeImmediateDominator(LoopScalarBody, LoopScalarPreHeader);
1230 DT->changeImmediateDominator(LoopExitBlock, LoopMiddleBlock);
1232 DEBUG(DT->verifyAnalysis());
1235 bool LoopVectorizationLegality::canVectorize() {
1236 if (!TheLoop->getLoopPreheader()) {
1237 assert(false && "No preheader!!");
1238 DEBUG(dbgs() << "LV: Loop not normalized." << "\n");
1242 // We can only vectorize single basic block loops.
1243 unsigned NumBlocks = TheLoop->getNumBlocks();
1244 if (NumBlocks != 1) {
1245 DEBUG(dbgs() << "LV: Too many blocks:" << NumBlocks << "\n");
1249 // We need to have a loop header.
1250 BasicBlock *BB = TheLoop->getHeader();
1251 DEBUG(dbgs() << "LV: Found a loop: " << BB->getName() << "\n");
1253 // ScalarEvolution needs to be able to find the exit count.
1254 const SCEV *ExitCount = SE->getExitCount(TheLoop, BB);
1255 if (ExitCount == SE->getCouldNotCompute()) {
1256 DEBUG(dbgs() << "LV: SCEV could not compute the loop exit count.\n");
1260 // Do not loop-vectorize loops with a tiny trip count.
1261 unsigned TC = SE->getSmallConstantTripCount(TheLoop, BB);
1262 if (TC > 0u && TC < TinyTripCountThreshold) {
1263 DEBUG(dbgs() << "LV: Found a loop with a very small trip count. " <<
1264 "This loop is not worth vectorizing.\n");
1268 // Go over each instruction and look at memory deps.
1269 if (!canVectorizeBlock(*BB)) {
1270 DEBUG(dbgs() << "LV: Can't vectorize this loop header\n");
1274 DEBUG(dbgs() << "LV: We can vectorize this loop" <<
1275 (PtrRtCheck.Need ? " (with a runtime bound check)" : "")
1278 // Okay! We can vectorize. At this point we don't have any other mem analysis
1279 // which may limit our maximum vectorization factor, so just return true with
1284 bool LoopVectorizationLegality::canVectorizeBlock(BasicBlock &BB) {
1285 // Scan the instructions in the block and look for hazards.
1286 for (BasicBlock::iterator it = BB.begin(), e = BB.end(); it != e; ++it) {
1287 Instruction *I = it;
1289 PHINode *Phi = dyn_cast<PHINode>(I);
1291 // This should not happen because the loop should be normalized.
1292 if (Phi->getNumIncomingValues() != 2) {
1293 DEBUG(dbgs() << "LV: Found an invalid PHI.\n");
1296 // We only look at integer phi nodes.
1297 if (!Phi->getType()->isIntegerTy()) {
1298 DEBUG(dbgs() << "LV: Found an non-int PHI.\n");
1302 if (isInductionVariable(Phi)) {
1304 DEBUG(dbgs() << "LV: Found too many inductions."<< *Phi <<"\n");
1307 DEBUG(dbgs() << "LV: Found the induction PHI."<< *Phi <<"\n");
1311 if (AddReductionVar(Phi, IntegerAdd)) {
1312 DEBUG(dbgs() << "LV: Found an ADD reduction PHI."<< *Phi <<"\n");
1315 if (AddReductionVar(Phi, IntegerMult)) {
1316 DEBUG(dbgs() << "LV: Found a MUL reduction PHI."<< *Phi <<"\n");
1319 if (AddReductionVar(Phi, IntegerOr)) {
1320 DEBUG(dbgs() << "LV: Found an OR reduction PHI."<< *Phi <<"\n");
1323 if (AddReductionVar(Phi, IntegerAnd)) {
1324 DEBUG(dbgs() << "LV: Found an AND reduction PHI."<< *Phi <<"\n");
1327 if (AddReductionVar(Phi, IntegerXor)) {
1328 DEBUG(dbgs() << "LV: Found a XOR reduction PHI."<< *Phi <<"\n");
1332 DEBUG(dbgs() << "LV: Found an unidentified PHI."<< *Phi <<"\n");
1334 }// end of PHI handling
1336 // We still don't handle functions.
1337 CallInst *CI = dyn_cast<CallInst>(I);
1339 DEBUG(dbgs() << "LV: Found a call site.\n");
1343 // We do not re-vectorize vectors.
1344 if (!VectorType::isValidElementType(I->getType()) &&
1345 !I->getType()->isVoidTy()) {
1346 DEBUG(dbgs() << "LV: Found unvectorizable type." << "\n");
1350 // Reduction instructions are allowed to have exit users.
1351 // All other instructions must not have external users.
1352 if (!AllowedExit.count(I))
1353 //Check that all of the users of the loop are inside the BB.
1354 for (Value::use_iterator it = I->use_begin(), e = I->use_end();
1356 Instruction *U = cast<Instruction>(*it);
1357 // This user may be a reduction exit value.
1358 BasicBlock *Parent = U->getParent();
1359 if (Parent != &BB) {
1360 DEBUG(dbgs() << "LV: Found an outside user for : "<< *U << "\n");
1367 DEBUG(dbgs() << "LV: Did not find an induction var.\n");
1371 // Don't vectorize if the memory dependencies do not allow vectorization.
1372 if (!canVectorizeMemory(BB))
1375 // We now know that the loop is vectorizable!
1376 // Collect variables that will remain uniform after vectorization.
1377 std::vector<Value*> Worklist;
1379 // Start with the conditional branch and walk up the block.
1380 Worklist.push_back(BB.getTerminator()->getOperand(0));
1382 while (Worklist.size()) {
1383 Instruction *I = dyn_cast<Instruction>(Worklist.back());
1384 Worklist.pop_back();
1385 // Look at instructions inside this block.
1387 if (I->getParent() != &BB) continue;
1389 // Stop when reaching PHI nodes.
1390 if (isa<PHINode>(I)) {
1391 assert(I == Induction && "Found a uniform PHI that is not the induction");
1395 // This is a known uniform.
1398 // Insert all operands.
1399 for (int i=0, Op = I->getNumOperands(); i < Op; ++i) {
1400 Worklist.push_back(I->getOperand(i));
1407 bool LoopVectorizationLegality::canVectorizeMemory(BasicBlock &BB) {
1408 typedef SmallVector<Value*, 16> ValueVector;
1409 typedef SmallPtrSet<Value*, 16> ValueSet;
1410 // Holds the Load and Store *instructions*.
1413 PtrRtCheck.Pointers.clear();
1414 PtrRtCheck.Need = false;
1416 // Scan the BB and collect legal loads and stores.
1417 for (BasicBlock::iterator it = BB.begin(), e = BB.end(); it != e; ++it) {
1418 Instruction *I = it;
1420 // If this is a load, save it. If this instruction can read from memory
1421 // but is not a load, then we quit. Notice that we don't handle function
1422 // calls that read or write.
1423 if (I->mayReadFromMemory()) {
1424 LoadInst *Ld = dyn_cast<LoadInst>(I);
1425 if (!Ld) return false;
1426 if (!Ld->isSimple()) {
1427 DEBUG(dbgs() << "LV: Found a non-simple load.\n");
1430 Loads.push_back(Ld);
1434 // Save store instructions. Abort if other instructions write to memory.
1435 if (I->mayWriteToMemory()) {
1436 StoreInst *St = dyn_cast<StoreInst>(I);
1437 if (!St) return false;
1438 if (!St->isSimple()) {
1439 DEBUG(dbgs() << "LV: Found a non-simple store.\n");
1442 Stores.push_back(St);
1446 // Now we have two lists that hold the loads and the stores.
1447 // Next, we find the pointers that they use.
1449 // Check if we see any stores. If there are no stores, then we don't
1450 // care if the pointers are *restrict*.
1451 if (!Stores.size()) {
1452 DEBUG(dbgs() << "LV: Found a read-only loop!\n");
1456 // Holds the read and read-write *pointers* that we find.
1458 ValueVector ReadWrites;
1460 // Holds the analyzed pointers. We don't want to call GetUnderlyingObjects
1461 // multiple times on the same object. If the ptr is accessed twice, once
1462 // for read and once for write, it will only appear once (on the write
1463 // list). This is okay, since we are going to check for conflicts between
1464 // writes and between reads and writes, but not between reads and reads.
1467 ValueVector::iterator I, IE;
1468 for (I = Stores.begin(), IE = Stores.end(); I != IE; ++I) {
1469 StoreInst *ST = dyn_cast<StoreInst>(*I);
1470 assert(ST && "Bad StoreInst");
1471 Value* Ptr = ST->getPointerOperand();
1473 if (isUniform(Ptr)) {
1474 DEBUG(dbgs() << "LV: We don't allow storing to uniform addresses\n");
1478 // If we did *not* see this pointer before, insert it to
1479 // the read-write list. At this phase it is only a 'write' list.
1480 if (Seen.insert(Ptr))
1481 ReadWrites.push_back(Ptr);
1484 for (I = Loads.begin(), IE = Loads.end(); I != IE; ++I) {
1485 LoadInst *LD = dyn_cast<LoadInst>(*I);
1486 assert(LD && "Bad LoadInst");
1487 Value* Ptr = LD->getPointerOperand();
1488 // If we did *not* see this pointer before, insert it to the
1489 // read list. If we *did* see it before, then it is already in
1490 // the read-write list. This allows us to vectorize expressions
1491 // such as A[i] += x; Because the address of A[i] is a read-write
1492 // pointer. This only works if the index of A[i] is consecutive.
1493 // If the address of i is unknown (for example A[B[i]]) then we may
1494 // read a few words, modify, and write a few words, and some of the
1495 // words may be written to the same address.
1496 if (Seen.insert(Ptr) || !isConsecutiveGep(Ptr))
1497 Reads.push_back(Ptr);
1500 // If we write (or read-write) to a single destination and there are no
1501 // other reads in this loop then is it safe to vectorize.
1502 if (ReadWrites.size() == 1 && Reads.size() == 0) {
1503 DEBUG(dbgs() << "LV: Found a write-only loop!\n");
1507 // Find pointers with computable bounds. We are going to use this information
1508 // to place a runtime bound check.
1510 for (I = ReadWrites.begin(), IE = ReadWrites.end(); I != IE; ++I)
1511 if (hasComputableBounds(*I)) {
1512 PtrRtCheck.Pointers.push_back(*I);
1513 DEBUG(dbgs() << "LV: Found a runtime check ptr:" << **I <<"\n");
1518 for (I = Reads.begin(), IE = Reads.end(); I != IE; ++I)
1519 if (hasComputableBounds(*I)) {
1520 PtrRtCheck.Pointers.push_back(*I);
1521 DEBUG(dbgs() << "LV: Found a runtime check ptr:" << **I <<"\n");
1527 // Check that we did not collect too many pointers or found a
1528 // unsizeable pointer.
1529 if (!RT || PtrRtCheck.Pointers.size() > RuntimeMemoryCheckThreshold) {
1530 PtrRtCheck.Pointers.clear();
1534 PtrRtCheck.Need = RT;
1537 DEBUG(dbgs() << "LV: We can perform a memory runtime check if needed.\n");
1540 // Now that the pointers are in two lists (Reads and ReadWrites), we
1541 // can check that there are no conflicts between each of the writes and
1542 // between the writes to the reads.
1543 ValueSet WriteObjects;
1544 ValueVector TempObjects;
1546 // Check that the read-writes do not conflict with other read-write
1548 for (I = ReadWrites.begin(), IE = ReadWrites.end(); I != IE; ++I) {
1549 GetUnderlyingObjects(*I, TempObjects, DL);
1550 for (ValueVector::iterator it=TempObjects.begin(), e=TempObjects.end();
1552 if (!isIdentifiedObject(*it)) {
1553 DEBUG(dbgs() << "LV: Found an unidentified write ptr:"<< **it <<"\n");
1556 if (!WriteObjects.insert(*it)) {
1557 DEBUG(dbgs() << "LV: Found a possible write-write reorder:"
1562 TempObjects.clear();
1565 /// Check that the reads don't conflict with the read-writes.
1566 for (I = Reads.begin(), IE = Reads.end(); I != IE; ++I) {
1567 GetUnderlyingObjects(*I, TempObjects, DL);
1568 for (ValueVector::iterator it=TempObjects.begin(), e=TempObjects.end();
1570 if (!isIdentifiedObject(*it)) {
1571 DEBUG(dbgs() << "LV: Found an unidentified read ptr:"<< **it <<"\n");
1574 if (WriteObjects.count(*it)) {
1575 DEBUG(dbgs() << "LV: Found a possible read/write reorder:"
1580 TempObjects.clear();
1583 // It is safe to vectorize and we don't need any runtime checks.
1584 DEBUG(dbgs() << "LV: We don't need a runtime memory check.\n");
1585 PtrRtCheck.Pointers.clear();
1586 PtrRtCheck.Need = false;
1590 bool LoopVectorizationLegality::AddReductionVar(PHINode *Phi,
1591 ReductionKind Kind) {
1592 if (Phi->getNumIncomingValues() != 2)
1595 // Find the possible incoming reduction variable.
1596 BasicBlock *BB = Phi->getParent();
1597 int SelfEdgeIdx = Phi->getBasicBlockIndex(BB);
1598 int InEdgeBlockIdx = (SelfEdgeIdx ? 0 : 1); // The other entry.
1599 Value *RdxStart = Phi->getIncomingValue(InEdgeBlockIdx);
1601 // ExitInstruction is the single value which is used outside the loop.
1602 // We only allow for a single reduction value to be used outside the loop.
1603 // This includes users of the reduction, variables (which form a cycle
1604 // which ends in the phi node).
1605 Instruction *ExitInstruction = 0;
1607 // Iter is our iterator. We start with the PHI node and scan for all of the
1608 // users of this instruction. All users must be instructions which can be
1609 // used as reduction variables (such as ADD). We may have a single
1610 // out-of-block user. They cycle must end with the original PHI.
1611 // Also, we can't have multiple block-local users.
1612 Instruction *Iter = Phi;
1614 // Any reduction instr must be of one of the allowed kinds.
1615 if (!isReductionInstr(Iter, Kind))
1618 // Did we found a user inside this block ?
1619 bool FoundInBlockUser = false;
1620 // Did we reach the initial PHI node ?
1621 bool FoundStartPHI = false;
1623 // If the instruction has no users then this is a broken
1624 // chain and can't be a reduction variable.
1625 if (Iter->use_empty())
1628 // For each of the *users* of iter.
1629 for (Value::use_iterator it = Iter->use_begin(), e = Iter->use_end();
1631 Instruction *U = cast<Instruction>(*it);
1632 // We already know that the PHI is a user.
1634 FoundStartPHI = true;
1637 // Check if we found the exit user.
1638 BasicBlock *Parent = U->getParent();
1640 // We must have a single exit instruction.
1641 if (ExitInstruction != 0)
1643 ExitInstruction = Iter;
1645 // We can't have multiple inside users.
1646 if (FoundInBlockUser)
1648 FoundInBlockUser = true;
1652 // We found a reduction var if we have reached the original
1653 // phi node and we only have a single instruction with out-of-loop
1655 if (FoundStartPHI && ExitInstruction) {
1656 // This instruction is allowed to have out-of-loop users.
1657 AllowedExit.insert(ExitInstruction);
1659 // Save the description of this reduction variable.
1660 ReductionDescriptor RD(RdxStart, ExitInstruction, Kind);
1661 Reductions[Phi] = RD;
1668 LoopVectorizationLegality::isReductionInstr(Instruction *I,
1669 ReductionKind Kind) {
1670 switch (I->getOpcode()) {
1673 case Instruction::PHI:
1676 case Instruction::Add:
1677 case Instruction::Sub:
1678 return Kind == IntegerAdd;
1679 case Instruction::Mul:
1680 case Instruction::UDiv:
1681 case Instruction::SDiv:
1682 return Kind == IntegerMult;
1683 case Instruction::And:
1684 return Kind == IntegerAnd;
1685 case Instruction::Or:
1686 return Kind == IntegerOr;
1687 case Instruction::Xor:
1688 return Kind == IntegerXor;
1692 bool LoopVectorizationLegality::isInductionVariable(PHINode *Phi) {
1693 // Check that the PHI is consecutive and starts at zero.
1694 const SCEV *PhiScev = SE->getSCEV(Phi);
1695 const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(PhiScev);
1697 DEBUG(dbgs() << "LV: PHI is not a poly recurrence.\n");
1700 const SCEV *Step = AR->getStepRecurrence(*SE);
1702 if (!Step->isOne()) {
1703 DEBUG(dbgs() << "LV: PHI stride does not equal one.\n");
1709 bool LoopVectorizationLegality::hasComputableBounds(Value *Ptr) {
1710 const SCEV *PhiScev = SE->getSCEV(Ptr);
1711 const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(PhiScev);
1715 return AR->isAffine();
1719 LoopVectorizationCostModel::findBestVectorizationFactor(unsigned VF) {
1721 DEBUG(dbgs() << "LV: No vector target information. Not vectorizing. \n");
1725 float Cost = expectedCost(1);
1727 DEBUG(dbgs() << "LV: Scalar loop costs: "<< (int)Cost << ".\n");
1728 for (unsigned i=2; i <= VF; i*=2) {
1729 // Notice that the vector loop needs to be executed less times, so
1730 // we need to divide the cost of the vector loops by the width of
1731 // the vector elements.
1732 float VectorCost = expectedCost(i) / (float)i;
1733 DEBUG(dbgs() << "LV: Vector loop of width "<< i << " costs: " <<
1734 (int)VectorCost << ".\n");
1735 if (VectorCost < Cost) {
1741 DEBUG(dbgs() << "LV: Selecting VF = : "<< Width << ".\n");
1745 unsigned LoopVectorizationCostModel::expectedCost(unsigned VF) {
1746 // We can only estimate the cost of single basic block loops.
1747 assert(1 == TheLoop->getNumBlocks() && "Too many blocks in loop");
1749 BasicBlock *BB = TheLoop->getHeader();
1752 // For each instruction in the old loop.
1753 for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; ++it) {
1754 Instruction *Inst = it;
1755 unsigned C = getInstructionCost(Inst, VF);
1757 DEBUG(dbgs() << "LV: Found an estimated cost of "<< C <<" for VF "<< VF <<
1758 " For instruction: "<< *Inst << "\n");
1765 LoopVectorizationCostModel::getInstructionCost(Instruction *I, unsigned VF) {
1766 assert(VTTI && "Invalid vector target transformation info");
1768 // If we know that this instruction will remain uniform, check the cost of
1769 // the scalar version.
1770 if (Legal->isUniformAfterVectorization(I))
1773 Type *RetTy = I->getType();
1774 Type *VectorTy = ToVectorTy(RetTy, VF);
1777 // TODO: We need to estimate the cost of intrinsic calls.
1778 switch (I->getOpcode()) {
1779 case Instruction::GetElementPtr:
1780 // We mark this instruction as zero-cost because scalar GEPs are usually
1781 // lowered to the intruction addressing mode. At the moment we don't
1782 // generate vector geps.
1784 case Instruction::Br: {
1785 return VTTI->getCFInstrCost(I->getOpcode());
1787 case Instruction::PHI:
1789 case Instruction::Add:
1790 case Instruction::FAdd:
1791 case Instruction::Sub:
1792 case Instruction::FSub:
1793 case Instruction::Mul:
1794 case Instruction::FMul:
1795 case Instruction::UDiv:
1796 case Instruction::SDiv:
1797 case Instruction::FDiv:
1798 case Instruction::URem:
1799 case Instruction::SRem:
1800 case Instruction::FRem:
1801 case Instruction::Shl:
1802 case Instruction::LShr:
1803 case Instruction::AShr:
1804 case Instruction::And:
1805 case Instruction::Or:
1806 case Instruction::Xor: {
1807 return VTTI->getArithmeticInstrCost(I->getOpcode(), VectorTy);
1809 case Instruction::Select: {
1810 SelectInst *SI = cast<SelectInst>(I);
1811 const SCEV *CondSCEV = SE->getSCEV(SI->getCondition());
1812 bool ScalarCond = (SE->isLoopInvariant(CondSCEV, TheLoop));
1813 Type *CondTy = SI->getCondition()->getType();
1815 CondTy = VectorType::get(CondTy, VF);
1817 return VTTI->getCmpSelInstrCost(I->getOpcode(), VectorTy, CondTy);
1819 case Instruction::ICmp:
1820 case Instruction::FCmp: {
1821 Type *ValTy = I->getOperand(0)->getType();
1822 VectorTy = ToVectorTy(ValTy, VF);
1823 return VTTI->getCmpSelInstrCost(I->getOpcode(), VectorTy);
1825 case Instruction::Store: {
1826 StoreInst *SI = cast<StoreInst>(I);
1827 Type *ValTy = SI->getValueOperand()->getType();
1828 VectorTy = ToVectorTy(ValTy, VF);
1831 return VTTI->getMemoryOpCost(I->getOpcode(), ValTy,
1832 SI->getAlignment(), SI->getPointerAddressSpace());
1834 // Scalarized stores.
1835 if (!Legal->isConsecutiveGep(SI->getPointerOperand())) {
1837 unsigned ExtCost = VTTI->getInstrCost(Instruction::ExtractElement,
1839 // The cost of extracting from the value vector.
1840 Cost += VF * (ExtCost);
1841 // The cost of the scalar stores.
1842 Cost += VF * VTTI->getMemoryOpCost(I->getOpcode(),
1843 ValTy->getScalarType(),
1845 SI->getPointerAddressSpace());
1850 return VTTI->getMemoryOpCost(I->getOpcode(), VectorTy, SI->getAlignment(),
1851 SI->getPointerAddressSpace());
1853 case Instruction::Load: {
1854 LoadInst *LI = cast<LoadInst>(I);
1857 return VTTI->getMemoryOpCost(I->getOpcode(), RetTy,
1859 LI->getPointerAddressSpace());
1861 // Scalarized loads.
1862 if (!Legal->isConsecutiveGep(LI->getPointerOperand())) {
1864 unsigned InCost = VTTI->getInstrCost(Instruction::InsertElement, RetTy);
1865 // The cost of inserting the loaded value into the result vector.
1866 Cost += VF * (InCost);
1867 // The cost of the scalar stores.
1868 Cost += VF * VTTI->getMemoryOpCost(I->getOpcode(),
1869 RetTy->getScalarType(),
1871 LI->getPointerAddressSpace());
1876 return VTTI->getMemoryOpCost(I->getOpcode(), VectorTy, LI->getAlignment(),
1877 LI->getPointerAddressSpace());
1879 case Instruction::ZExt:
1880 case Instruction::SExt:
1881 case Instruction::FPToUI:
1882 case Instruction::FPToSI:
1883 case Instruction::FPExt:
1884 case Instruction::PtrToInt:
1885 case Instruction::IntToPtr:
1886 case Instruction::SIToFP:
1887 case Instruction::UIToFP:
1888 case Instruction::Trunc:
1889 case Instruction::FPTrunc:
1890 case Instruction::BitCast: {
1891 Type *SrcVecTy = ToVectorTy(I->getOperand(0)->getType(), VF);
1892 return VTTI->getCastInstrCost(I->getOpcode(), VectorTy, SrcVecTy);
1895 // We are scalarizing the instruction. Return the cost of the scalar
1896 // instruction, plus the cost of insert and extract into vector
1897 // elements, times the vector width.
1900 bool IsVoid = RetTy->isVoidTy();
1902 unsigned InsCost = (IsVoid ? 0 :
1903 VTTI->getInstrCost(Instruction::InsertElement,
1906 unsigned ExtCost = VTTI->getInstrCost(Instruction::ExtractElement,
1909 // The cost of inserting the results plus extracting each one of the
1911 Cost += VF * (InsCost + ExtCost * I->getNumOperands());
1913 // The cost of executing VF copies of the scalar instruction.
1914 Cost += VF * VTTI->getInstrCost(I->getOpcode(), RetTy);
1920 Type* LoopVectorizationCostModel::ToVectorTy(Type *Scalar, unsigned VF) {
1921 if (Scalar->isVoidTy() || VF == 1)
1923 return VectorType::get(Scalar, VF);
1928 char LoopVectorize::ID = 0;
1929 static const char lv_name[] = "Loop Vectorization";
1930 INITIALIZE_PASS_BEGIN(LoopVectorize, LV_NAME, lv_name, false, false)
1931 INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
1932 INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)
1933 INITIALIZE_PASS_DEPENDENCY(LoopSimplify)
1934 INITIALIZE_PASS_END(LoopVectorize, LV_NAME, lv_name, false, false)
1937 Pass *createLoopVectorizePass() {
1938 return new LoopVectorize();