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 iterated dominator frontiers to place PHI nodes, then
13 // traversing the function in depth-first order to rewrite loads and stores as
16 // The algorithm used here is based on:
18 // Sreedhar and Gao. A linear time algorithm for placing phi-nodes.
19 // In Proceedings of the 22nd ACM SIGPLAN-SIGACT Symposium on Principles of
20 // Programming Languages
21 // POPL '95. ACM, New York, NY, 62-73.
23 // It has been modified to not explicitly use the DJ graph data structure and to
24 // directly compute pruned SSA using per-variable liveness information.
26 //===----------------------------------------------------------------------===//
28 #define DEBUG_TYPE "mem2reg"
29 #include "llvm/Transforms/Utils/PromoteMemToReg.h"
30 #include "llvm/Constants.h"
31 #include "llvm/DerivedTypes.h"
32 #include "llvm/Function.h"
33 #include "llvm/Instructions.h"
34 #include "llvm/IntrinsicInst.h"
35 #include "llvm/Metadata.h"
36 #include "llvm/Analysis/AliasSetTracker.h"
37 #include "llvm/Analysis/DebugInfo.h"
38 #include "llvm/Analysis/DIBuilder.h"
39 #include "llvm/Analysis/Dominators.h"
40 #include "llvm/Analysis/InstructionSimplify.h"
41 #include "llvm/Transforms/Utils/Local.h"
42 #include "llvm/ADT/DenseMap.h"
43 #include "llvm/ADT/SmallPtrSet.h"
44 #include "llvm/ADT/SmallVector.h"
45 #include "llvm/ADT/Statistic.h"
46 #include "llvm/ADT/STLExtras.h"
47 #include "llvm/Support/CFG.h"
52 STATISTIC(NumLocalPromoted, "Number of alloca's promoted within one block");
53 STATISTIC(NumSingleStore, "Number of alloca's promoted with a single store");
54 STATISTIC(NumDeadAlloca, "Number of dead alloca's removed");
55 STATISTIC(NumPHIInsert, "Number of PHI nodes inserted");
59 struct DenseMapInfo<std::pair<BasicBlock*, unsigned> > {
60 typedef std::pair<BasicBlock*, unsigned> EltTy;
61 static inline EltTy getEmptyKey() {
62 return EltTy(reinterpret_cast<BasicBlock*>(-1), ~0U);
64 static inline EltTy getTombstoneKey() {
65 return EltTy(reinterpret_cast<BasicBlock*>(-2), 0U);
67 static unsigned getHashValue(const std::pair<BasicBlock*, unsigned> &Val) {
68 return DenseMapInfo<void*>::getHashValue(Val.first) + Val.second*2;
70 static bool isEqual(const EltTy &LHS, const EltTy &RHS) {
76 /// isAllocaPromotable - Return true if this alloca is legal for promotion.
77 /// This is true if there are only loads and stores to the alloca.
79 bool llvm::isAllocaPromotable(const AllocaInst *AI) {
80 // FIXME: If the memory unit is of pointer or integer type, we can permit
81 // assignments to subsections of the memory unit.
83 // Only allow direct and non-volatile loads and stores...
84 for (Value::const_use_iterator UI = AI->use_begin(), UE = AI->use_end();
85 UI != UE; ++UI) { // Loop over all of the uses of the alloca
87 if (const LoadInst *LI = dyn_cast<LoadInst>(U)) {
90 } else if (const StoreInst *SI = dyn_cast<StoreInst>(U)) {
91 if (SI->getOperand(0) == AI)
92 return false; // Don't allow a store OF the AI, only INTO the AI.
106 // Data package used by RenamePass()
107 class RenamePassData {
109 typedef std::vector<Value *> ValVector;
111 RenamePassData() : BB(NULL), Pred(NULL), Values() {}
112 RenamePassData(BasicBlock *B, BasicBlock *P,
113 const ValVector &V) : BB(B), Pred(P), Values(V) {}
118 void swap(RenamePassData &RHS) {
119 std::swap(BB, RHS.BB);
120 std::swap(Pred, RHS.Pred);
121 Values.swap(RHS.Values);
125 /// LargeBlockInfo - This assigns and keeps a per-bb relative ordering of
126 /// load/store instructions in the block that directly load or store an alloca.
128 /// This functionality is important because it avoids scanning large basic
129 /// blocks multiple times when promoting many allocas in the same block.
130 class LargeBlockInfo {
131 /// InstNumbers - For each instruction that we track, keep the index of the
132 /// instruction. The index starts out as the number of the instruction from
133 /// the start of the block.
134 DenseMap<const Instruction *, unsigned> InstNumbers;
137 /// isInterestingInstruction - This code only looks at accesses to allocas.
138 static bool isInterestingInstruction(const Instruction *I) {
139 return (isa<LoadInst>(I) && isa<AllocaInst>(I->getOperand(0))) ||
140 (isa<StoreInst>(I) && isa<AllocaInst>(I->getOperand(1)));
143 /// getInstructionIndex - Get or calculate the index of the specified
145 unsigned getInstructionIndex(const Instruction *I) {
146 assert(isInterestingInstruction(I) &&
147 "Not a load/store to/from an alloca?");
149 // If we already have this instruction number, return it.
150 DenseMap<const Instruction *, unsigned>::iterator It = InstNumbers.find(I);
151 if (It != InstNumbers.end()) return It->second;
153 // Scan the whole block to get the instruction. This accumulates
154 // information for every interesting instruction in the block, in order to
155 // avoid gratuitus rescans.
156 const BasicBlock *BB = I->getParent();
158 for (BasicBlock::const_iterator BBI = BB->begin(), E = BB->end();
160 if (isInterestingInstruction(BBI))
161 InstNumbers[BBI] = InstNo++;
162 It = InstNumbers.find(I);
164 assert(It != InstNumbers.end() && "Didn't insert instruction?");
168 void deleteValue(const Instruction *I) {
169 InstNumbers.erase(I);
177 struct PromoteMem2Reg {
178 /// Allocas - The alloca instructions being promoted.
180 std::vector<AllocaInst*> Allocas;
184 /// AST - An AliasSetTracker object to update. If null, don't update it.
186 AliasSetTracker *AST;
188 /// AllocaLookup - Reverse mapping of Allocas.
190 DenseMap<AllocaInst*, unsigned> AllocaLookup;
192 /// NewPhiNodes - The PhiNodes we're adding.
194 DenseMap<std::pair<BasicBlock*, unsigned>, PHINode*> NewPhiNodes;
196 /// PhiToAllocaMap - For each PHI node, keep track of which entry in Allocas
197 /// it corresponds to.
198 DenseMap<PHINode*, unsigned> PhiToAllocaMap;
200 /// PointerAllocaValues - If we are updating an AliasSetTracker, then for
201 /// each alloca that is of pointer type, we keep track of what to copyValue
202 /// to the inserted PHI nodes here.
204 std::vector<Value*> PointerAllocaValues;
206 /// AllocaDbgDeclares - For each alloca, we keep track of the dbg.declare
207 /// intrinsic that describes it, if any, so that we can convert it to a
208 /// dbg.value intrinsic if the alloca gets promoted.
209 SmallVector<DbgDeclareInst*, 8> AllocaDbgDeclares;
211 /// Visited - The set of basic blocks the renamer has already visited.
213 SmallPtrSet<BasicBlock*, 16> Visited;
215 /// BBNumbers - Contains a stable numbering of basic blocks to avoid
216 /// non-determinstic behavior.
217 DenseMap<BasicBlock*, unsigned> BBNumbers;
219 /// DomLevels - Maps DomTreeNodes to their level in the dominator tree.
220 DenseMap<DomTreeNode*, unsigned> DomLevels;
222 /// BBNumPreds - Lazily compute the number of predecessors a block has.
223 DenseMap<const BasicBlock*, unsigned> BBNumPreds;
225 PromoteMem2Reg(const std::vector<AllocaInst*> &A, DominatorTree &dt,
226 AliasSetTracker *ast)
227 : Allocas(A), DT(dt), DIB(0), AST(ast) {}
234 /// dominates - Return true if BB1 dominates BB2 using the DominatorTree.
236 bool dominates(BasicBlock *BB1, BasicBlock *BB2) const {
237 return DT.dominates(BB1, BB2);
241 void RemoveFromAllocasList(unsigned &AllocaIdx) {
242 Allocas[AllocaIdx] = Allocas.back();
247 unsigned getNumPreds(const BasicBlock *BB) {
248 unsigned &NP = BBNumPreds[BB];
250 NP = std::distance(pred_begin(BB), pred_end(BB))+1;
254 void DetermineInsertionPoint(AllocaInst *AI, unsigned AllocaNum,
256 void ComputeLiveInBlocks(AllocaInst *AI, AllocaInfo &Info,
257 const SmallPtrSet<BasicBlock*, 32> &DefBlocks,
258 SmallPtrSet<BasicBlock*, 32> &LiveInBlocks);
260 void RewriteSingleStoreAlloca(AllocaInst *AI, AllocaInfo &Info,
261 LargeBlockInfo &LBI);
262 void PromoteSingleBlockAlloca(AllocaInst *AI, AllocaInfo &Info,
263 LargeBlockInfo &LBI);
265 void RenamePass(BasicBlock *BB, BasicBlock *Pred,
266 RenamePassData::ValVector &IncVals,
267 std::vector<RenamePassData> &Worklist);
268 bool QueuePhiNode(BasicBlock *BB, unsigned AllocaIdx, unsigned &Version);
272 SmallVector<BasicBlock*, 32> DefiningBlocks;
273 SmallVector<BasicBlock*, 32> UsingBlocks;
275 StoreInst *OnlyStore;
276 BasicBlock *OnlyBlock;
277 bool OnlyUsedInOneBlock;
279 Value *AllocaPointerVal;
280 DbgDeclareInst *DbgDeclare;
283 DefiningBlocks.clear();
287 OnlyUsedInOneBlock = true;
288 AllocaPointerVal = 0;
292 /// AnalyzeAlloca - Scan the uses of the specified alloca, filling in our
294 void AnalyzeAlloca(AllocaInst *AI) {
297 // As we scan the uses of the alloca instruction, keep track of stores,
298 // and decide whether all of the loads and stores to the alloca are within
299 // the same basic block.
300 for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end();
302 Instruction *User = cast<Instruction>(*UI++);
304 if (StoreInst *SI = dyn_cast<StoreInst>(User)) {
305 // Remember the basic blocks which define new values for the alloca
306 DefiningBlocks.push_back(SI->getParent());
307 AllocaPointerVal = SI->getOperand(0);
310 LoadInst *LI = cast<LoadInst>(User);
311 // Otherwise it must be a load instruction, keep track of variable
313 UsingBlocks.push_back(LI->getParent());
314 AllocaPointerVal = LI;
317 if (OnlyUsedInOneBlock) {
319 OnlyBlock = User->getParent();
320 else if (OnlyBlock != User->getParent())
321 OnlyUsedInOneBlock = false;
325 DbgDeclare = FindAllocaDbgDeclare(AI);
329 typedef std::pair<DomTreeNode*, unsigned> DomTreeNodePair;
331 struct DomTreeNodeCompare {
332 bool operator()(const DomTreeNodePair &LHS, const DomTreeNodePair &RHS) {
333 return LHS.second < RHS.second;
336 } // end of anonymous namespace
339 void PromoteMem2Reg::run() {
340 Function &F = *DT.getRoot()->getParent();
342 if (AST) PointerAllocaValues.resize(Allocas.size());
343 AllocaDbgDeclares.resize(Allocas.size());
348 for (unsigned AllocaNum = 0; AllocaNum != Allocas.size(); ++AllocaNum) {
349 AllocaInst *AI = Allocas[AllocaNum];
351 assert(isAllocaPromotable(AI) &&
352 "Cannot promote non-promotable alloca!");
353 assert(AI->getParent()->getParent() == &F &&
354 "All allocas should be in the same function, which is same as DF!");
356 if (AI->use_empty()) {
357 // If there are no uses of the alloca, just delete it now.
358 if (AST) AST->deleteValue(AI);
359 AI->eraseFromParent();
361 // Remove the alloca from the Allocas list, since it has been processed
362 RemoveFromAllocasList(AllocaNum);
367 // Calculate the set of read and write-locations for each alloca. This is
368 // analogous to finding the 'uses' and 'definitions' of each variable.
369 Info.AnalyzeAlloca(AI);
371 // If there is only a single store to this value, replace any loads of
372 // it that are directly dominated by the definition with the value stored.
373 if (Info.DefiningBlocks.size() == 1) {
374 RewriteSingleStoreAlloca(AI, Info, LBI);
376 // Finally, after the scan, check to see if the store is all that is left.
377 if (Info.UsingBlocks.empty()) {
378 // Record debuginfo for the store and remove the declaration's debuginfo.
379 if (DbgDeclareInst *DDI = Info.DbgDeclare) {
381 DIB = new DIBuilder(*DDI->getParent()->getParent()->getParent());
382 ConvertDebugDeclareToDebugValue(DDI, Info.OnlyStore, *DIB);
383 DDI->eraseFromParent();
385 // Remove the (now dead) store and alloca.
386 Info.OnlyStore->eraseFromParent();
387 LBI.deleteValue(Info.OnlyStore);
389 if (AST) AST->deleteValue(AI);
390 AI->eraseFromParent();
393 // The alloca has been processed, move on.
394 RemoveFromAllocasList(AllocaNum);
401 // If the alloca is only read and written in one basic block, just perform a
402 // linear sweep over the block to eliminate it.
403 if (Info.OnlyUsedInOneBlock) {
404 PromoteSingleBlockAlloca(AI, Info, LBI);
406 // Finally, after the scan, check to see if the stores are all that is
408 if (Info.UsingBlocks.empty()) {
410 // Remove the (now dead) stores and alloca.
411 while (!AI->use_empty()) {
412 StoreInst *SI = cast<StoreInst>(AI->use_back());
413 // Record debuginfo for the store before removing it.
414 if (DbgDeclareInst *DDI = Info.DbgDeclare) {
416 DIB = new DIBuilder(*SI->getParent()->getParent()->getParent());
417 ConvertDebugDeclareToDebugValue(DDI, SI, *DIB);
419 SI->eraseFromParent();
423 if (AST) AST->deleteValue(AI);
424 AI->eraseFromParent();
427 // The alloca has been processed, move on.
428 RemoveFromAllocasList(AllocaNum);
430 // The alloca's debuginfo can be removed as well.
431 if (DbgDeclareInst *DDI = Info.DbgDeclare)
432 DDI->eraseFromParent();
439 // If we haven't computed dominator tree levels, do so now.
440 if (DomLevels.empty()) {
441 SmallVector<DomTreeNode*, 32> Worklist;
443 DomTreeNode *Root = DT.getRootNode();
445 Worklist.push_back(Root);
447 while (!Worklist.empty()) {
448 DomTreeNode *Node = Worklist.pop_back_val();
449 unsigned ChildLevel = DomLevels[Node] + 1;
450 for (DomTreeNode::iterator CI = Node->begin(), CE = Node->end();
452 DomLevels[*CI] = ChildLevel;
453 Worklist.push_back(*CI);
458 // If we haven't computed a numbering for the BB's in the function, do so
460 if (BBNumbers.empty()) {
462 for (Function::iterator I = F.begin(), E = F.end(); I != E; ++I)
466 // If we have an AST to keep updated, remember some pointer value that is
467 // stored into the alloca.
469 PointerAllocaValues[AllocaNum] = Info.AllocaPointerVal;
471 // Remember the dbg.declare intrinsic describing this alloca, if any.
472 if (Info.DbgDeclare) AllocaDbgDeclares[AllocaNum] = Info.DbgDeclare;
474 // Keep the reverse mapping of the 'Allocas' array for the rename pass.
475 AllocaLookup[Allocas[AllocaNum]] = AllocaNum;
477 // At this point, we're committed to promoting the alloca using IDF's, and
478 // the standard SSA construction algorithm. Determine which blocks need PHI
479 // nodes and see if we can optimize out some work by avoiding insertion of
481 DetermineInsertionPoint(AI, AllocaNum, Info);
485 return; // All of the allocas must have been trivial!
490 // Set the incoming values for the basic block to be null values for all of
491 // the alloca's. We do this in case there is a load of a value that has not
492 // been stored yet. In this case, it will get this null value.
494 RenamePassData::ValVector Values(Allocas.size());
495 for (unsigned i = 0, e = Allocas.size(); i != e; ++i)
496 Values[i] = UndefValue::get(Allocas[i]->getAllocatedType());
498 // Walks all basic blocks in the function performing the SSA rename algorithm
499 // and inserting the phi nodes we marked as necessary
501 std::vector<RenamePassData> RenamePassWorkList;
502 RenamePassWorkList.push_back(RenamePassData(F.begin(), 0, Values));
505 RPD.swap(RenamePassWorkList.back());
506 RenamePassWorkList.pop_back();
507 // RenamePass may add new worklist entries.
508 RenamePass(RPD.BB, RPD.Pred, RPD.Values, RenamePassWorkList);
509 } while (!RenamePassWorkList.empty());
511 // The renamer uses the Visited set to avoid infinite loops. Clear it now.
514 // Remove the allocas themselves from the function.
515 for (unsigned i = 0, e = Allocas.size(); i != e; ++i) {
516 Instruction *A = Allocas[i];
518 // If there are any uses of the alloca instructions left, they must be in
519 // unreachable basic blocks that were not processed by walking the dominator
520 // tree. Just delete the users now.
522 A->replaceAllUsesWith(UndefValue::get(A->getType()));
523 if (AST) AST->deleteValue(A);
524 A->eraseFromParent();
527 // Remove alloca's dbg.declare instrinsics from the function.
528 for (unsigned i = 0, e = AllocaDbgDeclares.size(); i != e; ++i)
529 if (DbgDeclareInst *DDI = AllocaDbgDeclares[i])
530 DDI->eraseFromParent();
532 // Loop over all of the PHI nodes and see if there are any that we can get
533 // rid of because they merge all of the same incoming values. This can
534 // happen due to undef values coming into the PHI nodes. This process is
535 // iterative, because eliminating one PHI node can cause others to be removed.
536 bool EliminatedAPHI = true;
537 while (EliminatedAPHI) {
538 EliminatedAPHI = false;
540 for (DenseMap<std::pair<BasicBlock*, unsigned>, PHINode*>::iterator I =
541 NewPhiNodes.begin(), E = NewPhiNodes.end(); I != E;) {
542 PHINode *PN = I->second;
544 // If this PHI node merges one value and/or undefs, get the value.
545 if (Value *V = SimplifyInstruction(PN, 0, &DT)) {
546 if (AST && PN->getType()->isPointerTy())
547 AST->deleteValue(PN);
548 PN->replaceAllUsesWith(V);
549 PN->eraseFromParent();
550 NewPhiNodes.erase(I++);
551 EliminatedAPHI = true;
558 // At this point, the renamer has added entries to PHI nodes for all reachable
559 // code. Unfortunately, there may be unreachable blocks which the renamer
560 // hasn't traversed. If this is the case, the PHI nodes may not
561 // have incoming values for all predecessors. Loop over all PHI nodes we have
562 // created, inserting undef values if they are missing any incoming values.
564 for (DenseMap<std::pair<BasicBlock*, unsigned>, PHINode*>::iterator I =
565 NewPhiNodes.begin(), E = NewPhiNodes.end(); I != E; ++I) {
566 // We want to do this once per basic block. As such, only process a block
567 // when we find the PHI that is the first entry in the block.
568 PHINode *SomePHI = I->second;
569 BasicBlock *BB = SomePHI->getParent();
570 if (&BB->front() != SomePHI)
573 // Only do work here if there the PHI nodes are missing incoming values. We
574 // know that all PHI nodes that were inserted in a block will have the same
575 // number of incoming values, so we can just check any of them.
576 if (SomePHI->getNumIncomingValues() == getNumPreds(BB))
579 // Get the preds for BB.
580 SmallVector<BasicBlock*, 16> Preds(pred_begin(BB), pred_end(BB));
582 // Ok, now we know that all of the PHI nodes are missing entries for some
583 // basic blocks. Start by sorting the incoming predecessors for efficient
585 std::sort(Preds.begin(), Preds.end());
587 // Now we loop through all BB's which have entries in SomePHI and remove
588 // them from the Preds list.
589 for (unsigned i = 0, e = SomePHI->getNumIncomingValues(); i != e; ++i) {
590 // Do a log(n) search of the Preds list for the entry we want.
591 SmallVector<BasicBlock*, 16>::iterator EntIt =
592 std::lower_bound(Preds.begin(), Preds.end(),
593 SomePHI->getIncomingBlock(i));
594 assert(EntIt != Preds.end() && *EntIt == SomePHI->getIncomingBlock(i)&&
595 "PHI node has entry for a block which is not a predecessor!");
601 // At this point, the blocks left in the preds list must have dummy
602 // entries inserted into every PHI nodes for the block. Update all the phi
603 // nodes in this block that we are inserting (there could be phis before
605 unsigned NumBadPreds = SomePHI->getNumIncomingValues();
606 BasicBlock::iterator BBI = BB->begin();
607 while ((SomePHI = dyn_cast<PHINode>(BBI++)) &&
608 SomePHI->getNumIncomingValues() == NumBadPreds) {
609 Value *UndefVal = UndefValue::get(SomePHI->getType());
610 for (unsigned pred = 0, e = Preds.size(); pred != e; ++pred)
611 SomePHI->addIncoming(UndefVal, Preds[pred]);
619 /// ComputeLiveInBlocks - Determine which blocks the value is live in. These
620 /// are blocks which lead to uses. Knowing this allows us to avoid inserting
621 /// PHI nodes into blocks which don't lead to uses (thus, the inserted phi nodes
623 void PromoteMem2Reg::
624 ComputeLiveInBlocks(AllocaInst *AI, AllocaInfo &Info,
625 const SmallPtrSet<BasicBlock*, 32> &DefBlocks,
626 SmallPtrSet<BasicBlock*, 32> &LiveInBlocks) {
628 // To determine liveness, we must iterate through the predecessors of blocks
629 // where the def is live. Blocks are added to the worklist if we need to
630 // check their predecessors. Start with all the using blocks.
631 SmallVector<BasicBlock*, 64> LiveInBlockWorklist(Info.UsingBlocks.begin(),
632 Info.UsingBlocks.end());
634 // If any of the using blocks is also a definition block, check to see if the
635 // definition occurs before or after the use. If it happens before the use,
636 // the value isn't really live-in.
637 for (unsigned i = 0, e = LiveInBlockWorklist.size(); i != e; ++i) {
638 BasicBlock *BB = LiveInBlockWorklist[i];
639 if (!DefBlocks.count(BB)) continue;
641 // Okay, this is a block that both uses and defines the value. If the first
642 // reference to the alloca is a def (store), then we know it isn't live-in.
643 for (BasicBlock::iterator I = BB->begin(); ; ++I) {
644 if (StoreInst *SI = dyn_cast<StoreInst>(I)) {
645 if (SI->getOperand(1) != AI) continue;
647 // We found a store to the alloca before a load. The alloca is not
648 // actually live-in here.
649 LiveInBlockWorklist[i] = LiveInBlockWorklist.back();
650 LiveInBlockWorklist.pop_back();
655 if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
656 if (LI->getOperand(0) != AI) continue;
658 // Okay, we found a load before a store to the alloca. It is actually
659 // live into this block.
665 // Now that we have a set of blocks where the phi is live-in, recursively add
666 // their predecessors until we find the full region the value is live.
667 while (!LiveInBlockWorklist.empty()) {
668 BasicBlock *BB = LiveInBlockWorklist.pop_back_val();
670 // The block really is live in here, insert it into the set. If already in
671 // the set, then it has already been processed.
672 if (!LiveInBlocks.insert(BB))
675 // Since the value is live into BB, it is either defined in a predecessor or
676 // live into it to. Add the preds to the worklist unless they are a
678 for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
681 // The value is not live into a predecessor if it defines the value.
682 if (DefBlocks.count(P))
685 // Otherwise it is, add to the worklist.
686 LiveInBlockWorklist.push_back(P);
691 /// DetermineInsertionPoint - At this point, we're committed to promoting the
692 /// alloca using IDF's, and the standard SSA construction algorithm. Determine
693 /// which blocks need phi nodes and see if we can optimize out some work by
694 /// avoiding insertion of dead phi nodes.
695 void PromoteMem2Reg::DetermineInsertionPoint(AllocaInst *AI, unsigned AllocaNum,
697 // Unique the set of defining blocks for efficient lookup.
698 SmallPtrSet<BasicBlock*, 32> DefBlocks;
699 DefBlocks.insert(Info.DefiningBlocks.begin(), Info.DefiningBlocks.end());
701 // Determine which blocks the value is live in. These are blocks which lead
703 SmallPtrSet<BasicBlock*, 32> LiveInBlocks;
704 ComputeLiveInBlocks(AI, Info, DefBlocks, LiveInBlocks);
706 // Use a priority queue keyed on dominator tree level so that inserted nodes
707 // are handled from the bottom of the dominator tree upwards.
708 typedef std::priority_queue<DomTreeNodePair, SmallVector<DomTreeNodePair, 32>,
709 DomTreeNodeCompare> IDFPriorityQueue;
712 for (SmallPtrSet<BasicBlock*, 32>::const_iterator I = DefBlocks.begin(),
713 E = DefBlocks.end(); I != E; ++I) {
714 if (DomTreeNode *Node = DT.getNode(*I))
715 PQ.push(std::make_pair(Node, DomLevels[Node]));
718 SmallVector<std::pair<unsigned, BasicBlock*>, 32> DFBlocks;
719 SmallPtrSet<DomTreeNode*, 32> Visited;
720 SmallVector<DomTreeNode*, 32> Worklist;
721 while (!PQ.empty()) {
722 DomTreeNodePair RootPair = PQ.top();
724 DomTreeNode *Root = RootPair.first;
725 unsigned RootLevel = RootPair.second;
727 // Walk all dominator tree children of Root, inspecting their CFG edges with
728 // targets elsewhere on the dominator tree. Only targets whose level is at
729 // most Root's level are added to the iterated dominance frontier of the
733 Worklist.push_back(Root);
735 while (!Worklist.empty()) {
736 DomTreeNode *Node = Worklist.pop_back_val();
737 BasicBlock *BB = Node->getBlock();
739 for (succ_iterator SI = succ_begin(BB), SE = succ_end(BB); SI != SE;
741 DomTreeNode *SuccNode = DT.getNode(*SI);
743 // Quickly skip all CFG edges that are also dominator tree edges instead
744 // of catching them below.
745 if (SuccNode->getIDom() == Node)
748 unsigned SuccLevel = DomLevels[SuccNode];
749 if (SuccLevel > RootLevel)
752 if (!Visited.insert(SuccNode))
755 BasicBlock *SuccBB = SuccNode->getBlock();
756 if (!LiveInBlocks.count(SuccBB))
759 DFBlocks.push_back(std::make_pair(BBNumbers[SuccBB], SuccBB));
760 if (!DefBlocks.count(SuccBB))
761 PQ.push(std::make_pair(SuccNode, SuccLevel));
764 for (DomTreeNode::iterator CI = Node->begin(), CE = Node->end(); CI != CE;
766 if (!Visited.count(*CI))
767 Worklist.push_back(*CI);
772 if (DFBlocks.size() > 1)
773 std::sort(DFBlocks.begin(), DFBlocks.end());
775 unsigned CurrentVersion = 0;
776 for (unsigned i = 0, e = DFBlocks.size(); i != e; ++i)
777 QueuePhiNode(DFBlocks[i].second, AllocaNum, CurrentVersion);
780 /// RewriteSingleStoreAlloca - If there is only a single store to this value,
781 /// replace any loads of it that are directly dominated by the definition with
782 /// the value stored.
783 void PromoteMem2Reg::RewriteSingleStoreAlloca(AllocaInst *AI,
785 LargeBlockInfo &LBI) {
786 StoreInst *OnlyStore = Info.OnlyStore;
787 bool StoringGlobalVal = !isa<Instruction>(OnlyStore->getOperand(0));
788 BasicBlock *StoreBB = OnlyStore->getParent();
791 // Clear out UsingBlocks. We will reconstruct it here if needed.
792 Info.UsingBlocks.clear();
794 for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end(); UI != E; ) {
795 Instruction *UserInst = cast<Instruction>(*UI++);
796 if (!isa<LoadInst>(UserInst)) {
797 assert(UserInst == OnlyStore && "Should only have load/stores");
800 LoadInst *LI = cast<LoadInst>(UserInst);
802 // Okay, if we have a load from the alloca, we want to replace it with the
803 // only value stored to the alloca. We can do this if the value is
804 // dominated by the store. If not, we use the rest of the mem2reg machinery
805 // to insert the phi nodes as needed.
806 if (!StoringGlobalVal) { // Non-instructions are always dominated.
807 if (LI->getParent() == StoreBB) {
808 // If we have a use that is in the same block as the store, compare the
809 // indices of the two instructions to see which one came first. If the
810 // load came before the store, we can't handle it.
811 if (StoreIndex == -1)
812 StoreIndex = LBI.getInstructionIndex(OnlyStore);
814 if (unsigned(StoreIndex) > LBI.getInstructionIndex(LI)) {
815 // Can't handle this load, bail out.
816 Info.UsingBlocks.push_back(StoreBB);
820 } else if (LI->getParent() != StoreBB &&
821 !dominates(StoreBB, LI->getParent())) {
822 // If the load and store are in different blocks, use BB dominance to
823 // check their relationships. If the store doesn't dom the use, bail
825 Info.UsingBlocks.push_back(LI->getParent());
830 // Otherwise, we *can* safely rewrite this load.
831 Value *ReplVal = OnlyStore->getOperand(0);
832 // If the replacement value is the load, this must occur in unreachable
835 ReplVal = UndefValue::get(LI->getType());
836 LI->replaceAllUsesWith(ReplVal);
837 if (AST && LI->getType()->isPointerTy())
838 AST->deleteValue(LI);
839 LI->eraseFromParent();
846 /// StoreIndexSearchPredicate - This is a helper predicate used to search by the
847 /// first element of a pair.
848 struct StoreIndexSearchPredicate {
849 bool operator()(const std::pair<unsigned, StoreInst*> &LHS,
850 const std::pair<unsigned, StoreInst*> &RHS) {
851 return LHS.first < RHS.first;
857 /// PromoteSingleBlockAlloca - Many allocas are only used within a single basic
858 /// block. If this is the case, avoid traversing the CFG and inserting a lot of
859 /// potentially useless PHI nodes by just performing a single linear pass over
860 /// the basic block using the Alloca.
862 /// If we cannot promote this alloca (because it is read before it is written),
863 /// return true. This is necessary in cases where, due to control flow, the
864 /// alloca is potentially undefined on some control flow paths. e.g. code like
865 /// this is potentially correct:
867 /// for (...) { if (c) { A = undef; undef = B; } }
869 /// ... so long as A is not used before undef is set.
871 void PromoteMem2Reg::PromoteSingleBlockAlloca(AllocaInst *AI, AllocaInfo &Info,
872 LargeBlockInfo &LBI) {
873 // The trickiest case to handle is when we have large blocks. Because of this,
874 // this code is optimized assuming that large blocks happen. This does not
875 // significantly pessimize the small block case. This uses LargeBlockInfo to
876 // make it efficient to get the index of various operations in the block.
878 // Clear out UsingBlocks. We will reconstruct it here if needed.
879 Info.UsingBlocks.clear();
881 // Walk the use-def list of the alloca, getting the locations of all stores.
882 typedef SmallVector<std::pair<unsigned, StoreInst*>, 64> StoresByIndexTy;
883 StoresByIndexTy StoresByIndex;
885 for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end();
887 if (StoreInst *SI = dyn_cast<StoreInst>(*UI))
888 StoresByIndex.push_back(std::make_pair(LBI.getInstructionIndex(SI), SI));
890 // If there are no stores to the alloca, just replace any loads with undef.
891 if (StoresByIndex.empty()) {
892 for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end(); UI != E;)
893 if (LoadInst *LI = dyn_cast<LoadInst>(*UI++)) {
894 LI->replaceAllUsesWith(UndefValue::get(LI->getType()));
895 if (AST && LI->getType()->isPointerTy())
896 AST->deleteValue(LI);
898 LI->eraseFromParent();
903 // Sort the stores by their index, making it efficient to do a lookup with a
905 std::sort(StoresByIndex.begin(), StoresByIndex.end());
907 // Walk all of the loads from this alloca, replacing them with the nearest
908 // store above them, if any.
909 for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end(); UI != E;) {
910 LoadInst *LI = dyn_cast<LoadInst>(*UI++);
913 unsigned LoadIdx = LBI.getInstructionIndex(LI);
915 // Find the nearest store that has a lower than this load.
916 StoresByIndexTy::iterator I =
917 std::lower_bound(StoresByIndex.begin(), StoresByIndex.end(),
918 std::pair<unsigned, StoreInst*>(LoadIdx, static_cast<StoreInst*>(0)),
919 StoreIndexSearchPredicate());
921 // If there is no store before this load, then we can't promote this load.
922 if (I == StoresByIndex.begin()) {
923 // Can't handle this load, bail out.
924 Info.UsingBlocks.push_back(LI->getParent());
928 // Otherwise, there was a store before this load, the load takes its value.
930 LI->replaceAllUsesWith(I->second->getOperand(0));
931 if (AST && LI->getType()->isPointerTy())
932 AST->deleteValue(LI);
933 LI->eraseFromParent();
938 // QueuePhiNode - queues a phi-node to be added to a basic-block for a specific
939 // Alloca returns true if there wasn't already a phi-node for that variable
941 bool PromoteMem2Reg::QueuePhiNode(BasicBlock *BB, unsigned AllocaNo,
943 // Look up the basic-block in question.
944 PHINode *&PN = NewPhiNodes[std::make_pair(BB, AllocaNo)];
946 // If the BB already has a phi node added for the i'th alloca then we're done!
947 if (PN) return false;
949 // Create a PhiNode using the dereferenced type... and add the phi-node to the
951 PN = PHINode::Create(Allocas[AllocaNo]->getAllocatedType(), getNumPreds(BB),
952 Allocas[AllocaNo]->getName() + "." + Twine(Version++),
955 PhiToAllocaMap[PN] = AllocaNo;
957 if (AST && PN->getType()->isPointerTy())
958 AST->copyValue(PointerAllocaValues[AllocaNo], PN);
963 // RenamePass - Recursively traverse the CFG of the function, renaming loads and
964 // stores to the allocas which we are promoting. IncomingVals indicates what
965 // value each Alloca contains on exit from the predecessor block Pred.
967 void PromoteMem2Reg::RenamePass(BasicBlock *BB, BasicBlock *Pred,
968 RenamePassData::ValVector &IncomingVals,
969 std::vector<RenamePassData> &Worklist) {
971 // If we are inserting any phi nodes into this BB, they will already be in the
973 if (PHINode *APN = dyn_cast<PHINode>(BB->begin())) {
974 // If we have PHI nodes to update, compute the number of edges from Pred to
976 if (PhiToAllocaMap.count(APN)) {
977 // We want to be able to distinguish between PHI nodes being inserted by
978 // this invocation of mem2reg from those phi nodes that already existed in
979 // the IR before mem2reg was run. We determine that APN is being inserted
980 // because it is missing incoming edges. All other PHI nodes being
981 // inserted by this pass of mem2reg will have the same number of incoming
982 // operands so far. Remember this count.
983 unsigned NewPHINumOperands = APN->getNumOperands();
985 unsigned NumEdges = 0;
986 for (succ_iterator I = succ_begin(Pred), E = succ_end(Pred); I != E; ++I)
989 assert(NumEdges && "Must be at least one edge from Pred to BB!");
991 // Add entries for all the phis.
992 BasicBlock::iterator PNI = BB->begin();
994 unsigned AllocaNo = PhiToAllocaMap[APN];
996 // Add N incoming values to the PHI node.
997 for (unsigned i = 0; i != NumEdges; ++i)
998 APN->addIncoming(IncomingVals[AllocaNo], Pred);
1000 // The currently active variable for this block is now the PHI.
1001 IncomingVals[AllocaNo] = APN;
1003 // Get the next phi node.
1005 APN = dyn_cast<PHINode>(PNI);
1006 if (APN == 0) break;
1008 // Verify that it is missing entries. If not, it is not being inserted
1009 // by this mem2reg invocation so we want to ignore it.
1010 } while (APN->getNumOperands() == NewPHINumOperands);
1014 // Don't revisit blocks.
1015 if (!Visited.insert(BB)) return;
1017 for (BasicBlock::iterator II = BB->begin(); !isa<TerminatorInst>(II); ) {
1018 Instruction *I = II++; // get the instruction, increment iterator
1020 if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
1021 AllocaInst *Src = dyn_cast<AllocaInst>(LI->getPointerOperand());
1024 DenseMap<AllocaInst*, unsigned>::iterator AI = AllocaLookup.find(Src);
1025 if (AI == AllocaLookup.end()) continue;
1027 Value *V = IncomingVals[AI->second];
1029 // Anything using the load now uses the current value.
1030 LI->replaceAllUsesWith(V);
1031 if (AST && LI->getType()->isPointerTy())
1032 AST->deleteValue(LI);
1033 BB->getInstList().erase(LI);
1034 } else if (StoreInst *SI = dyn_cast<StoreInst>(I)) {
1035 // Delete this instruction and mark the name as the current holder of the
1037 AllocaInst *Dest = dyn_cast<AllocaInst>(SI->getPointerOperand());
1038 if (!Dest) continue;
1040 DenseMap<AllocaInst *, unsigned>::iterator ai = AllocaLookup.find(Dest);
1041 if (ai == AllocaLookup.end())
1044 // what value were we writing?
1045 IncomingVals[ai->second] = SI->getOperand(0);
1046 // Record debuginfo for the store before removing it.
1047 if (DbgDeclareInst *DDI = AllocaDbgDeclares[ai->second]) {
1049 DIB = new DIBuilder(*SI->getParent()->getParent()->getParent());
1050 ConvertDebugDeclareToDebugValue(DDI, SI, *DIB);
1052 BB->getInstList().erase(SI);
1056 // 'Recurse' to our successors.
1057 succ_iterator I = succ_begin(BB), E = succ_end(BB);
1060 // Keep track of the successors so we don't visit the same successor twice
1061 SmallPtrSet<BasicBlock*, 8> VisitedSuccs;
1063 // Handle the first successor without using the worklist.
1064 VisitedSuccs.insert(*I);
1070 if (VisitedSuccs.insert(*I))
1071 Worklist.push_back(RenamePassData(*I, Pred, IncomingVals));
1076 /// PromoteMemToReg - Promote the specified list of alloca instructions into
1077 /// scalar registers, inserting PHI nodes as appropriate. This function does
1078 /// not modify the CFG of the function at all. All allocas must be from the
1081 /// If AST is specified, the specified tracker is updated to reflect changes
1084 void llvm::PromoteMemToReg(const std::vector<AllocaInst*> &Allocas,
1085 DominatorTree &DT, AliasSetTracker *AST) {
1086 // If there is nothing to do, bail out...
1087 if (Allocas.empty()) return;
1089 PromoteMem2Reg(Allocas, DT, AST).run();