1 //===- PromoteMemoryToRegister.cpp - Convert allocas to registers ---------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file promotes memory references to be register references. It promotes
11 // alloca instructions which only have loads and stores as uses. An alloca is
12 // transformed by using dominator frontiers to place PHI nodes, then traversing
13 // the function in depth-first order to rewrite loads and stores as appropriate.
14 // This is just the standard SSA construction algorithm to construct "pruned"
17 //===----------------------------------------------------------------------===//
19 #define DEBUG_TYPE "mem2reg"
20 #include "llvm/Transforms/Utils/PromoteMemToReg.h"
21 #include "llvm/Constants.h"
22 #include "llvm/DerivedTypes.h"
23 #include "llvm/Function.h"
24 #include "llvm/Instructions.h"
25 #include "llvm/IntrinsicInst.h"
26 #include "llvm/Metadata.h"
27 #include "llvm/Analysis/DebugInfo.h"
28 #include "llvm/Analysis/Dominators.h"
29 #include "llvm/Analysis/AliasSetTracker.h"
30 #include "llvm/ADT/DenseMap.h"
31 #include "llvm/ADT/SmallPtrSet.h"
32 #include "llvm/ADT/SmallVector.h"
33 #include "llvm/ADT/Statistic.h"
34 #include "llvm/ADT/STLExtras.h"
35 #include "llvm/Support/CFG.h"
39 STATISTIC(NumLocalPromoted, "Number of alloca's promoted within one block");
40 STATISTIC(NumSingleStore, "Number of alloca's promoted with a single store");
41 STATISTIC(NumDeadAlloca, "Number of dead alloca's removed");
42 STATISTIC(NumPHIInsert, "Number of PHI nodes inserted");
46 struct DenseMapInfo<std::pair<BasicBlock*, unsigned> > {
47 typedef std::pair<BasicBlock*, unsigned> EltTy;
48 static inline EltTy getEmptyKey() {
49 return EltTy(reinterpret_cast<BasicBlock*>(-1), ~0U);
51 static inline EltTy getTombstoneKey() {
52 return EltTy(reinterpret_cast<BasicBlock*>(-2), 0U);
54 static unsigned getHashValue(const std::pair<BasicBlock*, unsigned> &Val) {
55 return DenseMapInfo<void*>::getHashValue(Val.first) + Val.second*2;
57 static bool isEqual(const EltTy &LHS, const EltTy &RHS) {
63 /// isAllocaPromotable - Return true if this alloca is legal for promotion.
64 /// This is true if there are only loads and stores to the alloca.
66 bool llvm::isAllocaPromotable(const AllocaInst *AI) {
67 // FIXME: If the memory unit is of pointer or integer type, we can permit
68 // assignments to subsections of the memory unit.
70 // Only allow direct and non-volatile loads and stores...
71 for (Value::use_const_iterator UI = AI->use_begin(), UE = AI->use_end();
72 UI != UE; ++UI) // Loop over all of the uses of the alloca
73 if (const LoadInst *LI = dyn_cast<LoadInst>(*UI)) {
76 } else if (const StoreInst *SI = dyn_cast<StoreInst>(*UI)) {
77 if (SI->getOperand(0) == AI)
78 return false; // Don't allow a store OF the AI, only INTO the AI.
88 /// FindAllocaDbgDeclare - Finds the llvm.dbg.declare intrinsic describing the
89 /// alloca 'V', if any.
90 static DbgDeclareInst *FindAllocaDbgDeclare(Value *V) {
91 if (MDNode *DebugNode = MDNode::getIfExists(V->getContext(), &V, 1))
92 for (Value::use_iterator UI = DebugNode->use_begin(),
93 E = DebugNode->use_end(); UI != E; ++UI)
94 if (DbgDeclareInst *DDI = dyn_cast<DbgDeclareInst>(*UI))
103 // Data package used by RenamePass()
104 class RenamePassData {
106 typedef std::vector<Value *> ValVector;
108 RenamePassData() : BB(NULL), Pred(NULL), Values() {}
109 RenamePassData(BasicBlock *B, BasicBlock *P,
110 const ValVector &V) : BB(B), Pred(P), Values(V) {}
115 void swap(RenamePassData &RHS) {
116 std::swap(BB, RHS.BB);
117 std::swap(Pred, RHS.Pred);
118 Values.swap(RHS.Values);
122 /// LargeBlockInfo - This assigns and keeps a per-bb relative ordering of
123 /// load/store instructions in the block that directly load or store an alloca.
125 /// This functionality is important because it avoids scanning large basic
126 /// blocks multiple times when promoting many allocas in the same block.
127 class LargeBlockInfo {
128 /// InstNumbers - For each instruction that we track, keep the index of the
129 /// instruction. The index starts out as the number of the instruction from
130 /// the start of the block.
131 DenseMap<const Instruction *, unsigned> InstNumbers;
134 /// isInterestingInstruction - This code only looks at accesses to allocas.
135 static bool isInterestingInstruction(const Instruction *I) {
136 return (isa<LoadInst>(I) && isa<AllocaInst>(I->getOperand(0))) ||
137 (isa<StoreInst>(I) && isa<AllocaInst>(I->getOperand(1)));
140 /// getInstructionIndex - Get or calculate the index of the specified
142 unsigned getInstructionIndex(const Instruction *I) {
143 assert(isInterestingInstruction(I) &&
144 "Not a load/store to/from an alloca?");
146 // If we already have this instruction number, return it.
147 DenseMap<const Instruction *, unsigned>::iterator It = InstNumbers.find(I);
148 if (It != InstNumbers.end()) return It->second;
150 // Scan the whole block to get the instruction. This accumulates
151 // information for every interesting instruction in the block, in order to
152 // avoid gratuitus rescans.
153 const BasicBlock *BB = I->getParent();
155 for (BasicBlock::const_iterator BBI = BB->begin(), E = BB->end();
157 if (isInterestingInstruction(BBI))
158 InstNumbers[BBI] = InstNo++;
159 It = InstNumbers.find(I);
161 assert(It != InstNumbers.end() && "Didn't insert instruction?");
165 void deleteValue(const Instruction *I) {
166 InstNumbers.erase(I);
174 struct PromoteMem2Reg {
175 /// Allocas - The alloca instructions being promoted.
177 std::vector<AllocaInst*> Allocas;
179 DominanceFrontier &DF;
182 /// AST - An AliasSetTracker object to update. If null, don't update it.
184 AliasSetTracker *AST;
186 /// AllocaLookup - Reverse mapping of Allocas.
188 std::map<AllocaInst*, unsigned> AllocaLookup;
190 /// NewPhiNodes - The PhiNodes we're adding.
192 DenseMap<std::pair<BasicBlock*, unsigned>, PHINode*> NewPhiNodes;
194 /// PhiToAllocaMap - For each PHI node, keep track of which entry in Allocas
195 /// it corresponds to.
196 DenseMap<PHINode*, unsigned> PhiToAllocaMap;
198 /// PointerAllocaValues - If we are updating an AliasSetTracker, then for
199 /// each alloca that is of pointer type, we keep track of what to copyValue
200 /// to the inserted PHI nodes here.
202 std::vector<Value*> PointerAllocaValues;
204 /// AllocaDbgDeclares - For each alloca, we keep track of the dbg.declare
205 /// intrinsic that describes it, if any, so that we can convert it to a
206 /// dbg.value intrinsic if the alloca gets promoted.
207 SmallVector<DbgDeclareInst*, 8> AllocaDbgDeclares;
209 /// Visited - The set of basic blocks the renamer has already visited.
211 SmallPtrSet<BasicBlock*, 16> Visited;
213 /// BBNumbers - Contains a stable numbering of basic blocks to avoid
214 /// non-determinstic behavior.
215 DenseMap<BasicBlock*, unsigned> BBNumbers;
217 /// BBNumPreds - Lazily compute the number of predecessors a block has.
218 DenseMap<const BasicBlock*, unsigned> BBNumPreds;
220 PromoteMem2Reg(const std::vector<AllocaInst*> &A, DominatorTree &dt,
221 DominanceFrontier &df, AliasSetTracker *ast)
222 : Allocas(A), DT(dt), DF(df), DIF(0), AST(ast) {}
229 /// properlyDominates - Return true if I1 properly dominates I2.
231 bool properlyDominates(Instruction *I1, Instruction *I2) const {
232 if (InvokeInst *II = dyn_cast<InvokeInst>(I1))
233 I1 = II->getNormalDest()->begin();
234 return DT.properlyDominates(I1->getParent(), I2->getParent());
237 /// dominates - Return true if BB1 dominates BB2 using the DominatorTree.
239 bool dominates(BasicBlock *BB1, BasicBlock *BB2) const {
240 return DT.dominates(BB1, BB2);
244 void RemoveFromAllocasList(unsigned &AllocaIdx) {
245 Allocas[AllocaIdx] = Allocas.back();
250 unsigned getNumPreds(const BasicBlock *BB) {
251 unsigned &NP = BBNumPreds[BB];
253 NP = std::distance(pred_begin(BB), pred_end(BB))+1;
257 void DetermineInsertionPoint(AllocaInst *AI, unsigned AllocaNum,
259 void ComputeLiveInBlocks(AllocaInst *AI, AllocaInfo &Info,
260 const SmallPtrSet<BasicBlock*, 32> &DefBlocks,
261 SmallPtrSet<BasicBlock*, 32> &LiveInBlocks);
263 void RewriteSingleStoreAlloca(AllocaInst *AI, AllocaInfo &Info,
264 LargeBlockInfo &LBI);
265 void PromoteSingleBlockAlloca(AllocaInst *AI, AllocaInfo &Info,
266 LargeBlockInfo &LBI);
267 void ConvertDebugDeclareToDebugValue(DbgDeclareInst *DDI, StoreInst *SI);
270 void RenamePass(BasicBlock *BB, BasicBlock *Pred,
271 RenamePassData::ValVector &IncVals,
272 std::vector<RenamePassData> &Worklist);
273 bool QueuePhiNode(BasicBlock *BB, unsigned AllocaIdx, unsigned &Version,
274 SmallPtrSet<PHINode*, 16> &InsertedPHINodes);
278 std::vector<BasicBlock*> DefiningBlocks;
279 std::vector<BasicBlock*> UsingBlocks;
281 StoreInst *OnlyStore;
282 BasicBlock *OnlyBlock;
283 bool OnlyUsedInOneBlock;
285 Value *AllocaPointerVal;
286 DbgDeclareInst *DbgDeclare;
289 DefiningBlocks.clear();
293 OnlyUsedInOneBlock = true;
294 AllocaPointerVal = 0;
298 /// AnalyzeAlloca - Scan the uses of the specified alloca, filling in our
300 void AnalyzeAlloca(AllocaInst *AI) {
303 // As we scan the uses of the alloca instruction, keep track of stores,
304 // and decide whether all of the loads and stores to the alloca are within
305 // the same basic block.
306 for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end();
308 Instruction *User = cast<Instruction>(*UI++);
310 if (StoreInst *SI = dyn_cast<StoreInst>(User)) {
311 // Remember the basic blocks which define new values for the alloca
312 DefiningBlocks.push_back(SI->getParent());
313 AllocaPointerVal = SI->getOperand(0);
316 LoadInst *LI = cast<LoadInst>(User);
317 // Otherwise it must be a load instruction, keep track of variable
319 UsingBlocks.push_back(LI->getParent());
320 AllocaPointerVal = LI;
323 if (OnlyUsedInOneBlock) {
325 OnlyBlock = User->getParent();
326 else if (OnlyBlock != User->getParent())
327 OnlyUsedInOneBlock = false;
331 DbgDeclare = FindAllocaDbgDeclare(AI);
334 } // end of anonymous namespace
337 void PromoteMem2Reg::run() {
338 Function &F = *DF.getRoot()->getParent();
340 if (AST) PointerAllocaValues.resize(Allocas.size());
341 AllocaDbgDeclares.resize(Allocas.size());
346 for (unsigned AllocaNum = 0; AllocaNum != Allocas.size(); ++AllocaNum) {
347 AllocaInst *AI = Allocas[AllocaNum];
349 assert(isAllocaPromotable(AI) &&
350 "Cannot promote non-promotable alloca!");
351 assert(AI->getParent()->getParent() == &F &&
352 "All allocas should be in the same function, which is same as DF!");
354 if (AI->use_empty()) {
355 // If there are no uses of the alloca, just delete it now.
356 if (AST) AST->deleteValue(AI);
357 AI->eraseFromParent();
359 // Remove the alloca from the Allocas list, since it has been processed
360 RemoveFromAllocasList(AllocaNum);
365 // Calculate the set of read and write-locations for each alloca. This is
366 // analogous to finding the 'uses' and 'definitions' of each variable.
367 Info.AnalyzeAlloca(AI);
369 // If there is only a single store to this value, replace any loads of
370 // it that are directly dominated by the definition with the value stored.
371 if (Info.DefiningBlocks.size() == 1) {
372 RewriteSingleStoreAlloca(AI, Info, LBI);
374 // Finally, after the scan, check to see if the store is all that is left.
375 if (Info.UsingBlocks.empty()) {
376 // Record debuginfo for the store and remove the declaration's debuginfo.
377 if (DbgDeclareInst *DDI = Info.DbgDeclare) {
378 ConvertDebugDeclareToDebugValue(DDI, Info.OnlyStore);
379 DDI->eraseFromParent();
381 // Remove the (now dead) store and alloca.
382 Info.OnlyStore->eraseFromParent();
383 LBI.deleteValue(Info.OnlyStore);
385 if (AST) AST->deleteValue(AI);
386 AI->eraseFromParent();
389 // The alloca has been processed, move on.
390 RemoveFromAllocasList(AllocaNum);
397 // If the alloca is only read and written in one basic block, just perform a
398 // linear sweep over the block to eliminate it.
399 if (Info.OnlyUsedInOneBlock) {
400 PromoteSingleBlockAlloca(AI, Info, LBI);
402 // Finally, after the scan, check to see if the stores are all that is
404 if (Info.UsingBlocks.empty()) {
406 // Remove the (now dead) stores and alloca.
407 while (!AI->use_empty()) {
408 StoreInst *SI = cast<StoreInst>(AI->use_back());
409 // Record debuginfo for the store before removing it.
410 if (DbgDeclareInst *DDI = Info.DbgDeclare)
411 ConvertDebugDeclareToDebugValue(DDI, SI);
412 SI->eraseFromParent();
416 if (AST) AST->deleteValue(AI);
417 AI->eraseFromParent();
420 // The alloca has been processed, move on.
421 RemoveFromAllocasList(AllocaNum);
423 // The alloca's debuginfo can be removed as well.
424 if (DbgDeclareInst *DDI = Info.DbgDeclare)
425 DDI->eraseFromParent();
432 // If we haven't computed a numbering for the BB's in the function, do so
434 if (BBNumbers.empty()) {
436 for (Function::iterator I = F.begin(), E = F.end(); I != E; ++I)
440 // If we have an AST to keep updated, remember some pointer value that is
441 // stored into the alloca.
443 PointerAllocaValues[AllocaNum] = Info.AllocaPointerVal;
445 // Remember the dbg.declare intrinsic describing this alloca, if any.
446 if (Info.DbgDeclare) AllocaDbgDeclares[AllocaNum] = Info.DbgDeclare;
448 // Keep the reverse mapping of the 'Allocas' array for the rename pass.
449 AllocaLookup[Allocas[AllocaNum]] = AllocaNum;
451 // At this point, we're committed to promoting the alloca using IDF's, and
452 // the standard SSA construction algorithm. Determine which blocks need PHI
453 // nodes and see if we can optimize out some work by avoiding insertion of
455 DetermineInsertionPoint(AI, AllocaNum, Info);
459 return; // All of the allocas must have been trivial!
464 // Set the incoming values for the basic block to be null values for all of
465 // the alloca's. We do this in case there is a load of a value that has not
466 // been stored yet. In this case, it will get this null value.
468 RenamePassData::ValVector Values(Allocas.size());
469 for (unsigned i = 0, e = Allocas.size(); i != e; ++i)
470 Values[i] = UndefValue::get(Allocas[i]->getAllocatedType());
472 // Walks all basic blocks in the function performing the SSA rename algorithm
473 // and inserting the phi nodes we marked as necessary
475 std::vector<RenamePassData> RenamePassWorkList;
476 RenamePassWorkList.push_back(RenamePassData(F.begin(), 0, Values));
479 RPD.swap(RenamePassWorkList.back());
480 RenamePassWorkList.pop_back();
481 // RenamePass may add new worklist entries.
482 RenamePass(RPD.BB, RPD.Pred, RPD.Values, RenamePassWorkList);
483 } while (!RenamePassWorkList.empty());
485 // The renamer uses the Visited set to avoid infinite loops. Clear it now.
488 // Remove the allocas themselves from the function.
489 for (unsigned i = 0, e = Allocas.size(); i != e; ++i) {
490 Instruction *A = Allocas[i];
492 // If there are any uses of the alloca instructions left, they must be in
493 // sections of dead code that were not processed on the dominance frontier.
494 // Just delete the users now.
497 A->replaceAllUsesWith(UndefValue::get(A->getType()));
498 if (AST) AST->deleteValue(A);
499 A->eraseFromParent();
502 // Remove alloca's dbg.declare instrinsics from the function.
503 for (unsigned i = 0, e = AllocaDbgDeclares.size(); i != e; ++i)
504 if (DbgDeclareInst *DDI = AllocaDbgDeclares[i])
505 DDI->eraseFromParent();
507 // Loop over all of the PHI nodes and see if there are any that we can get
508 // rid of because they merge all of the same incoming values. This can
509 // happen due to undef values coming into the PHI nodes. This process is
510 // iterative, because eliminating one PHI node can cause others to be removed.
511 bool EliminatedAPHI = true;
512 while (EliminatedAPHI) {
513 EliminatedAPHI = false;
515 for (DenseMap<std::pair<BasicBlock*, unsigned>, PHINode*>::iterator I =
516 NewPhiNodes.begin(), E = NewPhiNodes.end(); I != E;) {
517 PHINode *PN = I->second;
519 // If this PHI node merges one value and/or undefs, get the value.
520 if (Value *V = PN->hasConstantValue(&DT)) {
521 if (AST && PN->getType()->isPointerTy())
522 AST->deleteValue(PN);
523 PN->replaceAllUsesWith(V);
524 PN->eraseFromParent();
525 NewPhiNodes.erase(I++);
526 EliminatedAPHI = true;
533 // At this point, the renamer has added entries to PHI nodes for all reachable
534 // code. Unfortunately, there may be unreachable blocks which the renamer
535 // hasn't traversed. If this is the case, the PHI nodes may not
536 // have incoming values for all predecessors. Loop over all PHI nodes we have
537 // created, inserting undef values if they are missing any incoming values.
539 for (DenseMap<std::pair<BasicBlock*, unsigned>, PHINode*>::iterator I =
540 NewPhiNodes.begin(), E = NewPhiNodes.end(); I != E; ++I) {
541 // We want to do this once per basic block. As such, only process a block
542 // when we find the PHI that is the first entry in the block.
543 PHINode *SomePHI = I->second;
544 BasicBlock *BB = SomePHI->getParent();
545 if (&BB->front() != SomePHI)
548 // Only do work here if there the PHI nodes are missing incoming values. We
549 // know that all PHI nodes that were inserted in a block will have the same
550 // number of incoming values, so we can just check any of them.
551 if (SomePHI->getNumIncomingValues() == getNumPreds(BB))
554 // Get the preds for BB.
555 SmallVector<BasicBlock*, 16> Preds(pred_begin(BB), pred_end(BB));
557 // Ok, now we know that all of the PHI nodes are missing entries for some
558 // basic blocks. Start by sorting the incoming predecessors for efficient
560 std::sort(Preds.begin(), Preds.end());
562 // Now we loop through all BB's which have entries in SomePHI and remove
563 // them from the Preds list.
564 for (unsigned i = 0, e = SomePHI->getNumIncomingValues(); i != e; ++i) {
565 // Do a log(n) search of the Preds list for the entry we want.
566 SmallVector<BasicBlock*, 16>::iterator EntIt =
567 std::lower_bound(Preds.begin(), Preds.end(),
568 SomePHI->getIncomingBlock(i));
569 assert(EntIt != Preds.end() && *EntIt == SomePHI->getIncomingBlock(i)&&
570 "PHI node has entry for a block which is not a predecessor!");
576 // At this point, the blocks left in the preds list must have dummy
577 // entries inserted into every PHI nodes for the block. Update all the phi
578 // nodes in this block that we are inserting (there could be phis before
580 unsigned NumBadPreds = SomePHI->getNumIncomingValues();
581 BasicBlock::iterator BBI = BB->begin();
582 while ((SomePHI = dyn_cast<PHINode>(BBI++)) &&
583 SomePHI->getNumIncomingValues() == NumBadPreds) {
584 Value *UndefVal = UndefValue::get(SomePHI->getType());
585 for (unsigned pred = 0, e = Preds.size(); pred != e; ++pred)
586 SomePHI->addIncoming(UndefVal, Preds[pred]);
594 /// ComputeLiveInBlocks - Determine which blocks the value is live in. These
595 /// are blocks which lead to uses. Knowing this allows us to avoid inserting
596 /// PHI nodes into blocks which don't lead to uses (thus, the inserted phi nodes
598 void PromoteMem2Reg::
599 ComputeLiveInBlocks(AllocaInst *AI, AllocaInfo &Info,
600 const SmallPtrSet<BasicBlock*, 32> &DefBlocks,
601 SmallPtrSet<BasicBlock*, 32> &LiveInBlocks) {
603 // To determine liveness, we must iterate through the predecessors of blocks
604 // where the def is live. Blocks are added to the worklist if we need to
605 // check their predecessors. Start with all the using blocks.
606 SmallVector<BasicBlock*, 64> LiveInBlockWorklist;
607 LiveInBlockWorklist.insert(LiveInBlockWorklist.end(),
608 Info.UsingBlocks.begin(), Info.UsingBlocks.end());
610 // If any of the using blocks is also a definition block, check to see if the
611 // definition occurs before or after the use. If it happens before the use,
612 // the value isn't really live-in.
613 for (unsigned i = 0, e = LiveInBlockWorklist.size(); i != e; ++i) {
614 BasicBlock *BB = LiveInBlockWorklist[i];
615 if (!DefBlocks.count(BB)) continue;
617 // Okay, this is a block that both uses and defines the value. If the first
618 // reference to the alloca is a def (store), then we know it isn't live-in.
619 for (BasicBlock::iterator I = BB->begin(); ; ++I) {
620 if (StoreInst *SI = dyn_cast<StoreInst>(I)) {
621 if (SI->getOperand(1) != AI) continue;
623 // We found a store to the alloca before a load. The alloca is not
624 // actually live-in here.
625 LiveInBlockWorklist[i] = LiveInBlockWorklist.back();
626 LiveInBlockWorklist.pop_back();
631 if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
632 if (LI->getOperand(0) != AI) continue;
634 // Okay, we found a load before a store to the alloca. It is actually
635 // live into this block.
641 // Now that we have a set of blocks where the phi is live-in, recursively add
642 // their predecessors until we find the full region the value is live.
643 while (!LiveInBlockWorklist.empty()) {
644 BasicBlock *BB = LiveInBlockWorklist.pop_back_val();
646 // The block really is live in here, insert it into the set. If already in
647 // the set, then it has already been processed.
648 if (!LiveInBlocks.insert(BB))
651 // Since the value is live into BB, it is either defined in a predecessor or
652 // live into it to. Add the preds to the worklist unless they are a
654 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
657 // The value is not live into a predecessor if it defines the value.
658 if (DefBlocks.count(P))
661 // Otherwise it is, add to the worklist.
662 LiveInBlockWorklist.push_back(P);
667 /// DetermineInsertionPoint - At this point, we're committed to promoting the
668 /// alloca using IDF's, and the standard SSA construction algorithm. Determine
669 /// which blocks need phi nodes and see if we can optimize out some work by
670 /// avoiding insertion of dead phi nodes.
671 void PromoteMem2Reg::DetermineInsertionPoint(AllocaInst *AI, unsigned AllocaNum,
674 // Unique the set of defining blocks for efficient lookup.
675 SmallPtrSet<BasicBlock*, 32> DefBlocks;
676 DefBlocks.insert(Info.DefiningBlocks.begin(), Info.DefiningBlocks.end());
678 // Determine which blocks the value is live in. These are blocks which lead
680 SmallPtrSet<BasicBlock*, 32> LiveInBlocks;
681 ComputeLiveInBlocks(AI, Info, DefBlocks, LiveInBlocks);
683 // Compute the locations where PhiNodes need to be inserted. Look at the
684 // dominance frontier of EACH basic-block we have a write in.
685 unsigned CurrentVersion = 0;
686 SmallPtrSet<PHINode*, 16> InsertedPHINodes;
687 std::vector<std::pair<unsigned, BasicBlock*> > DFBlocks;
688 while (!Info.DefiningBlocks.empty()) {
689 BasicBlock *BB = Info.DefiningBlocks.back();
690 Info.DefiningBlocks.pop_back();
692 // Look up the DF for this write, add it to defining blocks.
693 DominanceFrontier::const_iterator it = DF.find(BB);
694 if (it == DF.end()) continue;
696 const DominanceFrontier::DomSetType &S = it->second;
698 // In theory we don't need the indirection through the DFBlocks vector.
699 // In practice, the order of calling QueuePhiNode would depend on the
700 // (unspecified) ordering of basic blocks in the dominance frontier,
701 // which would give PHI nodes non-determinstic subscripts. Fix this by
702 // processing blocks in order of the occurance in the function.
703 for (DominanceFrontier::DomSetType::const_iterator P = S.begin(),
704 PE = S.end(); P != PE; ++P) {
705 // If the frontier block is not in the live-in set for the alloca, don't
706 // bother processing it.
707 if (!LiveInBlocks.count(*P))
710 DFBlocks.push_back(std::make_pair(BBNumbers[*P], *P));
713 // Sort by which the block ordering in the function.
714 if (DFBlocks.size() > 1)
715 std::sort(DFBlocks.begin(), DFBlocks.end());
717 for (unsigned i = 0, e = DFBlocks.size(); i != e; ++i) {
718 BasicBlock *BB = DFBlocks[i].second;
719 if (QueuePhiNode(BB, AllocaNum, CurrentVersion, InsertedPHINodes))
720 Info.DefiningBlocks.push_back(BB);
726 /// RewriteSingleStoreAlloca - If there is only a single store to this value,
727 /// replace any loads of it that are directly dominated by the definition with
728 /// the value stored.
729 void PromoteMem2Reg::RewriteSingleStoreAlloca(AllocaInst *AI,
731 LargeBlockInfo &LBI) {
732 StoreInst *OnlyStore = Info.OnlyStore;
733 bool StoringGlobalVal = !isa<Instruction>(OnlyStore->getOperand(0));
734 BasicBlock *StoreBB = OnlyStore->getParent();
737 // Clear out UsingBlocks. We will reconstruct it here if needed.
738 Info.UsingBlocks.clear();
740 for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end(); UI != E; ) {
741 Instruction *UserInst = cast<Instruction>(*UI++);
742 if (!isa<LoadInst>(UserInst)) {
743 assert(UserInst == OnlyStore && "Should only have load/stores");
746 LoadInst *LI = cast<LoadInst>(UserInst);
748 // Okay, if we have a load from the alloca, we want to replace it with the
749 // only value stored to the alloca. We can do this if the value is
750 // dominated by the store. If not, we use the rest of the mem2reg machinery
751 // to insert the phi nodes as needed.
752 if (!StoringGlobalVal) { // Non-instructions are always dominated.
753 if (LI->getParent() == StoreBB) {
754 // If we have a use that is in the same block as the store, compare the
755 // indices of the two instructions to see which one came first. If the
756 // load came before the store, we can't handle it.
757 if (StoreIndex == -1)
758 StoreIndex = LBI.getInstructionIndex(OnlyStore);
760 if (unsigned(StoreIndex) > LBI.getInstructionIndex(LI)) {
761 // Can't handle this load, bail out.
762 Info.UsingBlocks.push_back(StoreBB);
766 } else if (LI->getParent() != StoreBB &&
767 !dominates(StoreBB, LI->getParent())) {
768 // If the load and store are in different blocks, use BB dominance to
769 // check their relationships. If the store doesn't dom the use, bail
771 Info.UsingBlocks.push_back(LI->getParent());
776 // Otherwise, we *can* safely rewrite this load.
777 Value *ReplVal = OnlyStore->getOperand(0);
778 // If the replacement value is the load, this must occur in unreachable
781 ReplVal = UndefValue::get(LI->getType());
782 LI->replaceAllUsesWith(ReplVal);
783 if (AST && LI->getType()->isPointerTy())
784 AST->deleteValue(LI);
785 LI->eraseFromParent();
792 /// StoreIndexSearchPredicate - This is a helper predicate used to search by the
793 /// first element of a pair.
794 struct StoreIndexSearchPredicate {
795 bool operator()(const std::pair<unsigned, StoreInst*> &LHS,
796 const std::pair<unsigned, StoreInst*> &RHS) {
797 return LHS.first < RHS.first;
803 /// PromoteSingleBlockAlloca - Many allocas are only used within a single basic
804 /// block. If this is the case, avoid traversing the CFG and inserting a lot of
805 /// potentially useless PHI nodes by just performing a single linear pass over
806 /// the basic block using the Alloca.
808 /// If we cannot promote this alloca (because it is read before it is written),
809 /// return true. This is necessary in cases where, due to control flow, the
810 /// alloca is potentially undefined on some control flow paths. e.g. code like
811 /// this is potentially correct:
813 /// for (...) { if (c) { A = undef; undef = B; } }
815 /// ... so long as A is not used before undef is set.
817 void PromoteMem2Reg::PromoteSingleBlockAlloca(AllocaInst *AI, AllocaInfo &Info,
818 LargeBlockInfo &LBI) {
819 // The trickiest case to handle is when we have large blocks. Because of this,
820 // this code is optimized assuming that large blocks happen. This does not
821 // significantly pessimize the small block case. This uses LargeBlockInfo to
822 // make it efficient to get the index of various operations in the block.
824 // Clear out UsingBlocks. We will reconstruct it here if needed.
825 Info.UsingBlocks.clear();
827 // Walk the use-def list of the alloca, getting the locations of all stores.
828 typedef SmallVector<std::pair<unsigned, StoreInst*>, 64> StoresByIndexTy;
829 StoresByIndexTy StoresByIndex;
831 for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end();
833 if (StoreInst *SI = dyn_cast<StoreInst>(*UI))
834 StoresByIndex.push_back(std::make_pair(LBI.getInstructionIndex(SI), SI));
836 // If there are no stores to the alloca, just replace any loads with undef.
837 if (StoresByIndex.empty()) {
838 for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end(); UI != E;)
839 if (LoadInst *LI = dyn_cast<LoadInst>(*UI++)) {
840 LI->replaceAllUsesWith(UndefValue::get(LI->getType()));
841 if (AST && LI->getType()->isPointerTy())
842 AST->deleteValue(LI);
844 LI->eraseFromParent();
849 // Sort the stores by their index, making it efficient to do a lookup with a
851 std::sort(StoresByIndex.begin(), StoresByIndex.end());
853 // Walk all of the loads from this alloca, replacing them with the nearest
854 // store above them, if any.
855 for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end(); UI != E;) {
856 LoadInst *LI = dyn_cast<LoadInst>(*UI++);
859 unsigned LoadIdx = LBI.getInstructionIndex(LI);
861 // Find the nearest store that has a lower than this load.
862 StoresByIndexTy::iterator I =
863 std::lower_bound(StoresByIndex.begin(), StoresByIndex.end(),
864 std::pair<unsigned, StoreInst*>(LoadIdx, 0),
865 StoreIndexSearchPredicate());
867 // If there is no store before this load, then we can't promote this load.
868 if (I == StoresByIndex.begin()) {
869 // Can't handle this load, bail out.
870 Info.UsingBlocks.push_back(LI->getParent());
874 // Otherwise, there was a store before this load, the load takes its value.
876 LI->replaceAllUsesWith(I->second->getOperand(0));
877 if (AST && LI->getType()->isPointerTy())
878 AST->deleteValue(LI);
879 LI->eraseFromParent();
884 // Inserts a llvm.dbg.value instrinsic before the stores to an alloca'd value
885 // that has an associated llvm.dbg.decl intrinsic.
886 void PromoteMem2Reg::ConvertDebugDeclareToDebugValue(DbgDeclareInst *DDI,
888 DIVariable DIVar(DDI->getVariable());
889 if (!DIVar.getNode())
893 DIF = new DIFactory(*SI->getParent()->getParent()->getParent());
894 Instruction *DbgVal = DIF->InsertDbgValueIntrinsic(SI->getOperand(0), 0,
897 // Propagate any debug metadata from the store onto the dbg.value.
898 if (MDNode *SIMD = SI->getMetadata("dbg"))
899 DbgVal->setMetadata("dbg", SIMD);
902 // QueuePhiNode - queues a phi-node to be added to a basic-block for a specific
903 // Alloca returns true if there wasn't already a phi-node for that variable
905 bool PromoteMem2Reg::QueuePhiNode(BasicBlock *BB, unsigned AllocaNo,
907 SmallPtrSet<PHINode*, 16> &InsertedPHINodes) {
908 // Look up the basic-block in question.
909 PHINode *&PN = NewPhiNodes[std::make_pair(BB, AllocaNo)];
911 // If the BB already has a phi node added for the i'th alloca then we're done!
912 if (PN) return false;
914 // Create a PhiNode using the dereferenced type... and add the phi-node to the
916 PN = PHINode::Create(Allocas[AllocaNo]->getAllocatedType(),
917 Allocas[AllocaNo]->getName() + "." + Twine(Version++),
920 PhiToAllocaMap[PN] = AllocaNo;
921 PN->reserveOperandSpace(getNumPreds(BB));
923 InsertedPHINodes.insert(PN);
925 if (AST && PN->getType()->isPointerTy())
926 AST->copyValue(PointerAllocaValues[AllocaNo], PN);
931 // RenamePass - Recursively traverse the CFG of the function, renaming loads and
932 // stores to the allocas which we are promoting. IncomingVals indicates what
933 // value each Alloca contains on exit from the predecessor block Pred.
935 void PromoteMem2Reg::RenamePass(BasicBlock *BB, BasicBlock *Pred,
936 RenamePassData::ValVector &IncomingVals,
937 std::vector<RenamePassData> &Worklist) {
939 // If we are inserting any phi nodes into this BB, they will already be in the
941 if (PHINode *APN = dyn_cast<PHINode>(BB->begin())) {
942 // If we have PHI nodes to update, compute the number of edges from Pred to
944 if (PhiToAllocaMap.count(APN)) {
945 // We want to be able to distinguish between PHI nodes being inserted by
946 // this invocation of mem2reg from those phi nodes that already existed in
947 // the IR before mem2reg was run. We determine that APN is being inserted
948 // because it is missing incoming edges. All other PHI nodes being
949 // inserted by this pass of mem2reg will have the same number of incoming
950 // operands so far. Remember this count.
951 unsigned NewPHINumOperands = APN->getNumOperands();
953 unsigned NumEdges = 0;
954 for (succ_iterator I = succ_begin(Pred), E = succ_end(Pred); I != E; ++I)
957 assert(NumEdges && "Must be at least one edge from Pred to BB!");
959 // Add entries for all the phis.
960 BasicBlock::iterator PNI = BB->begin();
962 unsigned AllocaNo = PhiToAllocaMap[APN];
964 // Add N incoming values to the PHI node.
965 for (unsigned i = 0; i != NumEdges; ++i)
966 APN->addIncoming(IncomingVals[AllocaNo], Pred);
968 // The currently active variable for this block is now the PHI.
969 IncomingVals[AllocaNo] = APN;
971 // Get the next phi node.
973 APN = dyn_cast<PHINode>(PNI);
976 // Verify that it is missing entries. If not, it is not being inserted
977 // by this mem2reg invocation so we want to ignore it.
978 } while (APN->getNumOperands() == NewPHINumOperands);
982 // Don't revisit blocks.
983 if (!Visited.insert(BB)) return;
985 for (BasicBlock::iterator II = BB->begin(); !isa<TerminatorInst>(II); ) {
986 Instruction *I = II++; // get the instruction, increment iterator
988 if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
989 AllocaInst *Src = dyn_cast<AllocaInst>(LI->getPointerOperand());
992 std::map<AllocaInst*, unsigned>::iterator AI = AllocaLookup.find(Src);
993 if (AI == AllocaLookup.end()) continue;
995 Value *V = IncomingVals[AI->second];
997 // Anything using the load now uses the current value.
998 LI->replaceAllUsesWith(V);
999 if (AST && LI->getType()->isPointerTy())
1000 AST->deleteValue(LI);
1001 BB->getInstList().erase(LI);
1002 } else if (StoreInst *SI = dyn_cast<StoreInst>(I)) {
1003 // Delete this instruction and mark the name as the current holder of the
1005 AllocaInst *Dest = dyn_cast<AllocaInst>(SI->getPointerOperand());
1006 if (!Dest) continue;
1008 std::map<AllocaInst *, unsigned>::iterator ai = AllocaLookup.find(Dest);
1009 if (ai == AllocaLookup.end())
1012 // what value were we writing?
1013 IncomingVals[ai->second] = SI->getOperand(0);
1014 // Record debuginfo for the store before removing it.
1015 if (DbgDeclareInst *DDI = AllocaDbgDeclares[ai->second])
1016 ConvertDebugDeclareToDebugValue(DDI, SI);
1017 BB->getInstList().erase(SI);
1021 // 'Recurse' to our successors.
1022 succ_iterator I = succ_begin(BB), E = succ_end(BB);
1025 // Keep track of the successors so we don't visit the same successor twice
1026 SmallPtrSet<BasicBlock*, 8> VisitedSuccs;
1028 // Handle the first successor without using the worklist.
1029 VisitedSuccs.insert(*I);
1035 if (VisitedSuccs.insert(*I))
1036 Worklist.push_back(RenamePassData(*I, Pred, IncomingVals));
1041 /// PromoteMemToReg - Promote the specified list of alloca instructions into
1042 /// scalar registers, inserting PHI nodes as appropriate. This function makes
1043 /// use of DominanceFrontier information. This function does not modify the CFG
1044 /// of the function at all. All allocas must be from the same function.
1046 /// If AST is specified, the specified tracker is updated to reflect changes
1049 void llvm::PromoteMemToReg(const std::vector<AllocaInst*> &Allocas,
1050 DominatorTree &DT, DominanceFrontier &DF,
1051 AliasSetTracker *AST) {
1052 // If there is nothing to do, bail out...
1053 if (Allocas.empty()) return;
1055 PromoteMem2Reg(Allocas, DT, DF, AST).run();