//
// The LLVM Compiler Infrastructure
//
-// This file was developed by the LLVM research group and is distributed under
-// the University of Illinois Open Source License. See LICENSE.TXT for details.
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
//
// This pass also guarantees that loops will have exactly one backedge.
//
+// Indirectbr instructions introduce several complications. If the loop
+// contains or is entered by an indirectbr instruction, it may not be possible
+// to transform the loop and make these guarantees. Client code should check
+// that these conditions are true before relying on them.
+//
// Note that the simplifycfg pass will clean up blocks which are split out but
// end up being unnecessary, so usage of this pass should not pessimize
// generated code.
//
//===----------------------------------------------------------------------===//
+#define DEBUG_TYPE "loopsimplify"
#include "llvm/Transforms/Scalar.h"
-#include "llvm/Constant.h"
+#include "llvm/Constants.h"
#include "llvm/Instructions.h"
#include "llvm/Function.h"
+#include "llvm/LLVMContext.h"
#include "llvm/Type.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/Dominators.h"
-#include "llvm/Analysis/LoopInfo.h"
+#include "llvm/Analysis/LoopPass.h"
+#include "llvm/Analysis/ScalarEvolution.h"
+#include "llvm/Transforms/Utils/BasicBlockUtils.h"
+#include "llvm/Transforms/Utils/Local.h"
#include "llvm/Support/CFG.h"
-#include "llvm/Support/Compiler.h"
+#include "llvm/Support/Debug.h"
#include "llvm/ADT/SetOperations.h"
#include "llvm/ADT/SetVector.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/DepthFirstIterator.h"
using namespace llvm;
+STATISTIC(NumInserted, "Number of pre-header or exit blocks inserted");
+STATISTIC(NumNested , "Number of nested loops split out");
+
namespace {
- Statistic<>
- NumInserted("loopsimplify", "Number of pre-header or exit blocks inserted");
- Statistic<>
- NumNested("loopsimplify", "Number of nested loops split out");
+ struct LoopSimplify : public LoopPass {
+ static char ID; // Pass identification, replacement for typeid
+ LoopSimplify() : LoopPass(&ID) {}
- struct VISIBILITY_HIDDEN LoopSimplify : public FunctionPass {
// AA - If we have an alias analysis object to update, this is it, otherwise
// this is null.
AliasAnalysis *AA;
LoopInfo *LI;
-
- virtual bool runOnFunction(Function &F);
+ DominatorTree *DT;
+ Loop *L;
+ virtual bool runOnLoop(Loop *L, LPPassManager &LPM);
virtual void getAnalysisUsage(AnalysisUsage &AU) const {
// We need loop information to identify the loops...
- AU.addRequired<LoopInfo>();
- AU.addRequired<DominatorSet>();
- AU.addRequired<DominatorTree>();
+ AU.addRequiredTransitive<LoopInfo>();
+ AU.addRequiredTransitive<DominatorTree>();
AU.addPreserved<LoopInfo>();
- AU.addPreserved<DominatorSet>();
- AU.addPreserved<ImmediateDominators>();
- AU.addPreserved<ETForest>();
AU.addPreserved<DominatorTree>();
AU.addPreserved<DominanceFrontier>();
+ AU.addPreserved<AliasAnalysis>();
+ AU.addPreserved<ScalarEvolution>();
AU.addPreservedID(BreakCriticalEdgesID); // No critical edges added.
}
+
+ /// verifyAnalysis() - Verify LoopSimplifyForm's guarantees.
+ void verifyAnalysis() const;
+
private:
- bool ProcessLoop(Loop *L);
- BasicBlock *SplitBlockPredecessors(BasicBlock *BB, const char *Suffix,
- const std::vector<BasicBlock*> &Preds);
+ bool ProcessLoop(Loop *L, LPPassManager &LPM);
BasicBlock *RewriteLoopExitBlock(Loop *L, BasicBlock *Exit);
- void InsertPreheaderForLoop(Loop *L);
- Loop *SeparateNestedLoop(Loop *L);
- void InsertUniqueBackedgeBlock(Loop *L);
+ BasicBlock *InsertPreheaderForLoop(Loop *L);
+ Loop *SeparateNestedLoop(Loop *L, LPPassManager &LPM);
+ BasicBlock *InsertUniqueBackedgeBlock(Loop *L, BasicBlock *Preheader);
void PlaceSplitBlockCarefully(BasicBlock *NewBB,
- std::vector<BasicBlock*> &SplitPreds,
+ SmallVectorImpl<BasicBlock*> &SplitPreds,
Loop *L);
-
- void UpdateDomInfoForRevectoredPreds(BasicBlock *NewBB,
- std::vector<BasicBlock*> &PredBlocks);
};
-
- RegisterPass<LoopSimplify>
- X("loopsimplify", "Canonicalize natural loops", true);
}
+char LoopSimplify::ID = 0;
+static RegisterPass<LoopSimplify>
+X("loopsimplify", "Canonicalize natural loops", true);
+
// Publically exposed interface to pass...
-const PassInfo *llvm::LoopSimplifyID = X.getPassInfo();
-FunctionPass *llvm::createLoopSimplifyPass() { return new LoopSimplify(); }
+const PassInfo *const llvm::LoopSimplifyID = &X;
+Pass *llvm::createLoopSimplifyPass() { return new LoopSimplify(); }
-/// runOnFunction - Run down all loops in the CFG (recursively, but we could do
+/// runOnLoop - Run down all loops in the CFG (recursively, but we could do
/// it in any convenient order) inserting preheaders...
///
-bool LoopSimplify::runOnFunction(Function &F) {
+bool LoopSimplify::runOnLoop(Loop *l, LPPassManager &LPM) {
+ L = l;
bool Changed = false;
LI = &getAnalysis<LoopInfo>();
- AA = getAnalysisToUpdate<AliasAnalysis>();
-
- // Check to see that no blocks (other than the header) in loops have
- // predecessors that are not in loops. This is not valid for natural loops,
- // but can occur if the blocks are unreachable. Since they are unreachable we
- // can just shamelessly destroy their terminators to make them not branch into
- // the loop!
- for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
- // This case can only occur for unreachable blocks. Blocks that are
- // unreachable can't be in loops, so filter those blocks out.
- if (LI->getLoopFor(BB)) continue;
-
- bool BlockUnreachable = false;
- TerminatorInst *TI = BB->getTerminator();
-
- // Check to see if any successors of this block are non-loop-header loops
- // that are not the header.
- for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i) {
- // If this successor is not in a loop, BB is clearly ok.
- Loop *L = LI->getLoopFor(TI->getSuccessor(i));
- if (!L) continue;
-
- // If the succ is the loop header, and if L is a top-level loop, then this
- // is an entrance into a loop through the header, which is also ok.
- if (L->getHeader() == TI->getSuccessor(i) && L->getParentLoop() == 0)
- continue;
-
- // Otherwise, this is an entrance into a loop from some place invalid.
- // Either the loop structure is invalid and this is not a natural loop (in
- // which case the compiler is buggy somewhere else) or BB is unreachable.
- BlockUnreachable = true;
- break;
- }
-
- // If this block is ok, check the next one.
- if (!BlockUnreachable) continue;
-
- // Otherwise, this block is dead. To clean up the CFG and to allow later
- // loop transformations to ignore this case, we delete the edges into the
- // loop by replacing the terminator.
-
- // Remove PHI entries from the successors.
- for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
- TI->getSuccessor(i)->removePredecessor(BB);
-
- // Add a new unreachable instruction.
- new UnreachableInst(TI);
-
- // Delete the dead terminator.
- if (AA) AA->deleteValue(&BB->back());
- BB->getInstList().pop_back();
- Changed |= true;
- }
-
- for (LoopInfo::iterator I = LI->begin(), E = LI->end(); I != E; ++I)
- Changed |= ProcessLoop(*I);
+ AA = getAnalysisIfAvailable<AliasAnalysis>();
+ DT = &getAnalysis<DominatorTree>();
+
+ Changed |= ProcessLoop(L, LPM);
return Changed;
}
/// ProcessLoop - Walk the loop structure in depth first order, ensuring that
/// all loops have preheaders.
///
-bool LoopSimplify::ProcessLoop(Loop *L) {
+bool LoopSimplify::ProcessLoop(Loop *L, LPPassManager &LPM) {
bool Changed = false;
ReprocessLoop:
-
- // Canonicalize inner loops before outer loops. Inner loop canonicalization
- // can provide work for the outer loop to canonicalize.
- for (Loop::iterator I = L->begin(), E = L->end(); I != E; ++I)
- Changed |= ProcessLoop(*I);
-
- assert(L->getBlocks()[0] == L->getHeader() &&
- "Header isn't first block in loop?");
+
+ // Check to see that no blocks (other than the header) in this loop that has
+ // predecessors that are not in the loop. This is not valid for natural
+ // loops, but can occur if the blocks are unreachable. Since they are
+ // unreachable we can just shamelessly delete those CFG edges!
+ for (Loop::block_iterator BB = L->block_begin(), E = L->block_end();
+ BB != E; ++BB) {
+ if (*BB == L->getHeader()) continue;
+
+ SmallPtrSet<BasicBlock *, 4> BadPreds;
+ for (pred_iterator PI = pred_begin(*BB), PE = pred_end(*BB); PI != PE; ++PI)
+ if (!L->contains(*PI))
+ BadPreds.insert(*PI);
+
+ // Delete each unique out-of-loop (and thus dead) predecessor.
+ for (SmallPtrSet<BasicBlock *, 4>::iterator I = BadPreds.begin(),
+ E = BadPreds.end(); I != E; ++I) {
+
+ DEBUG(dbgs() << "LoopSimplify: Deleting edge from dead predecessor ";
+ WriteAsOperand(dbgs(), *I, false);
+ dbgs() << "\n");
+
+ // Inform each successor of each dead pred.
+ for (succ_iterator SI = succ_begin(*I), SE = succ_end(*I); SI != SE; ++SI)
+ (*SI)->removePredecessor(*I);
+ // Zap the dead pred's terminator and replace it with unreachable.
+ TerminatorInst *TI = (*I)->getTerminator();
+ TI->replaceAllUsesWith(UndefValue::get(TI->getType()));
+ (*I)->getTerminator()->eraseFromParent();
+ new UnreachableInst((*I)->getContext(), *I);
+ Changed = true;
+ }
+ }
+
+ // If there are exiting blocks with branches on undef, resolve the undef in
+ // the direction which will exit the loop. This will help simplify loop
+ // trip count computations.
+ SmallVector<BasicBlock*, 8> ExitingBlocks;
+ L->getExitingBlocks(ExitingBlocks);
+ for (SmallVectorImpl<BasicBlock *>::iterator I = ExitingBlocks.begin(),
+ E = ExitingBlocks.end(); I != E; ++I)
+ if (BranchInst *BI = dyn_cast<BranchInst>((*I)->getTerminator()))
+ if (BI->isConditional()) {
+ if (UndefValue *Cond = dyn_cast<UndefValue>(BI->getCondition())) {
+
+ DEBUG(dbgs() << "LoopSimplify: Resolving \"br i1 undef\" to exit in ";
+ WriteAsOperand(dbgs(), *I, false);
+ dbgs() << "\n");
+
+ BI->setCondition(ConstantInt::get(Cond->getType(),
+ !L->contains(BI->getSuccessor(0))));
+ Changed = true;
+ }
+ }
// Does the loop already have a preheader? If so, don't insert one.
- if (L->getLoopPreheader() == 0) {
- InsertPreheaderForLoop(L);
- NumInserted++;
- Changed = true;
+ BasicBlock *Preheader = L->getLoopPreheader();
+ if (!Preheader) {
+ Preheader = InsertPreheaderForLoop(L);
+ if (Preheader) {
+ NumInserted++;
+ Changed = true;
+ }
}
// Next, check to make sure that all exit nodes of the loop only have
// predecessors that are inside of the loop. This check guarantees that the
// loop preheader/header will dominate the exit blocks. If the exit block has
// predecessors from outside of the loop, split the edge now.
- std::vector<BasicBlock*> ExitBlocks;
+ SmallVector<BasicBlock*, 8> ExitBlocks;
L->getExitBlocks(ExitBlocks);
- SetVector<BasicBlock*> ExitBlockSet(ExitBlocks.begin(), ExitBlocks.end());
- for (SetVector<BasicBlock*>::iterator I = ExitBlockSet.begin(),
+ SmallSetVector<BasicBlock *, 8> ExitBlockSet(ExitBlocks.begin(),
+ ExitBlocks.end());
+ for (SmallSetVector<BasicBlock *, 8>::iterator I = ExitBlockSet.begin(),
E = ExitBlockSet.end(); I != E; ++I) {
BasicBlock *ExitBlock = *I;
for (pred_iterator PI = pred_begin(ExitBlock), PE = pred_end(ExitBlock);
// Must be exactly this loop: no subloops, parent loops, or non-loop preds
// allowed.
if (!L->contains(*PI)) {
- RewriteLoopExitBlock(L, ExitBlock);
- NumInserted++;
- Changed = true;
+ if (RewriteLoopExitBlock(L, ExitBlock)) {
+ NumInserted++;
+ Changed = true;
+ }
break;
}
}
// If the header has more than two predecessors at this point (from the
// preheader and from multiple backedges), we must adjust the loop.
- unsigned NumBackedges = L->getNumBackEdges();
- if (NumBackedges != 1) {
+ BasicBlock *LoopLatch = L->getLoopLatch();
+ if (!LoopLatch) {
// If this is really a nested loop, rip it out into a child loop. Don't do
// this for loops with a giant number of backedges, just factor them into a
// common backedge instead.
- if (NumBackedges < 8) {
- if (Loop *NL = SeparateNestedLoop(L)) {
+ if (L->getNumBackEdges() < 8) {
+ if (SeparateNestedLoop(L, LPM)) {
++NumNested;
// This is a big restructuring change, reprocess the whole loop.
- ProcessLoop(NL);
Changed = true;
// GCC doesn't tail recursion eliminate this.
goto ReprocessLoop;
// If we either couldn't, or didn't want to, identify nesting of the loops,
// insert a new block that all backedges target, then make it jump to the
// loop header.
- InsertUniqueBackedgeBlock(L);
- NumInserted++;
- Changed = true;
+ LoopLatch = InsertUniqueBackedgeBlock(L, Preheader);
+ if (LoopLatch) {
+ NumInserted++;
+ Changed = true;
+ }
}
// Scan over the PHI nodes in the loop header. Since they now have only two
PHINode *PN;
for (BasicBlock::iterator I = L->getHeader()->begin();
(PN = dyn_cast<PHINode>(I++)); )
- if (Value *V = PN->hasConstantValue()) {
- PN->replaceAllUsesWith(V);
- PN->eraseFromParent();
- }
-
- return Changed;
-}
+ if (Value *V = PN->hasConstantValue(DT)) {
+ if (AA) AA->deleteValue(PN);
+ PN->replaceAllUsesWith(V);
+ PN->eraseFromParent();
+ }
-/// SplitBlockPredecessors - Split the specified block into two blocks. We want
-/// to move the predecessors specified in the Preds list to point to the new
-/// block, leaving the remaining predecessors pointing to BB. This method
-/// updates the SSA PHINode's, but no other analyses.
-///
-BasicBlock *LoopSimplify::SplitBlockPredecessors(BasicBlock *BB,
- const char *Suffix,
- const std::vector<BasicBlock*> &Preds) {
-
- // Create new basic block, insert right before the original block...
- BasicBlock *NewBB = new BasicBlock(BB->getName()+Suffix, BB->getParent(), BB);
-
- // The preheader first gets an unconditional branch to the loop header...
- BranchInst *BI = new BranchInst(BB, NewBB);
-
- // For every PHI node in the block, insert a PHI node into NewBB where the
- // incoming values from the out of loop edges are moved to NewBB. We have two
- // possible cases here. If the loop is dead, we just insert dummy entries
- // into the PHI nodes for the new edge. If the loop is not dead, we move the
- // incoming edges in BB into new PHI nodes in NewBB.
- //
- if (!Preds.empty()) { // Is the loop not obviously dead?
- // Check to see if the values being merged into the new block need PHI
- // nodes. If so, insert them.
- for (BasicBlock::iterator I = BB->begin(); isa<PHINode>(I); ) {
- PHINode *PN = cast<PHINode>(I);
- ++I;
-
- // Check to see if all of the values coming in are the same. If so, we
- // don't need to create a new PHI node.
- Value *InVal = PN->getIncomingValueForBlock(Preds[0]);
- for (unsigned i = 1, e = Preds.size(); i != e; ++i)
- if (InVal != PN->getIncomingValueForBlock(Preds[i])) {
- InVal = 0;
+ // If this loop has multiple exits and the exits all go to the same
+ // block, attempt to merge the exits. This helps several passes, such
+ // as LoopRotation, which do not support loops with multiple exits.
+ // SimplifyCFG also does this (and this code uses the same utility
+ // function), however this code is loop-aware, where SimplifyCFG is
+ // not. That gives it the advantage of being able to hoist
+ // loop-invariant instructions out of the way to open up more
+ // opportunities, and the disadvantage of having the responsibility
+ // to preserve dominator information.
+ bool UniqueExit = true;
+ if (!ExitBlocks.empty())
+ for (unsigned i = 1, e = ExitBlocks.size(); i != e; ++i)
+ if (ExitBlocks[i] != ExitBlocks[0]) {
+ UniqueExit = false;
+ break;
+ }
+ if (UniqueExit) {
+ for (unsigned i = 0, e = ExitingBlocks.size(); i != e; ++i) {
+ BasicBlock *ExitingBlock = ExitingBlocks[i];
+ if (!ExitingBlock->getSinglePredecessor()) continue;
+ BranchInst *BI = dyn_cast<BranchInst>(ExitingBlock->getTerminator());
+ if (!BI || !BI->isConditional()) continue;
+ CmpInst *CI = dyn_cast<CmpInst>(BI->getCondition());
+ if (!CI || CI->getParent() != ExitingBlock) continue;
+
+ // Attempt to hoist out all instructions except for the
+ // comparison and the branch.
+ bool AllInvariant = true;
+ for (BasicBlock::iterator I = ExitingBlock->begin(); &*I != BI; ) {
+ Instruction *Inst = I++;
+ if (Inst == CI)
+ continue;
+ if (!L->makeLoopInvariant(Inst, Changed,
+ Preheader ? Preheader->getTerminator() : 0)) {
+ AllInvariant = false;
break;
}
-
- // If the values coming into the block are not the same, we need a PHI.
- if (InVal == 0) {
- // Create the new PHI node, insert it into NewBB at the end of the block
- PHINode *NewPHI = new PHINode(PN->getType(), PN->getName()+".ph", BI);
- if (AA) AA->copyValue(PN, NewPHI);
-
- // Move all of the edges from blocks outside the loop to the new PHI
- for (unsigned i = 0, e = Preds.size(); i != e; ++i) {
- Value *V = PN->removeIncomingValue(Preds[i], false);
- NewPHI->addIncoming(V, Preds[i]);
- }
- InVal = NewPHI;
- } else {
- // Remove all of the edges coming into the PHI nodes from outside of the
- // block.
- for (unsigned i = 0, e = Preds.size(); i != e; ++i)
- PN->removeIncomingValue(Preds[i], false);
}
-
- // Add an incoming value to the PHI node in the loop for the preheader
- // edge.
- PN->addIncoming(InVal, NewBB);
-
- // Can we eliminate this phi node now?
- if (Value *V = PN->hasConstantValue(true)) {
- if (!isa<Instruction>(V) ||
- getAnalysis<DominatorSet>().dominates(cast<Instruction>(V), PN)) {
- PN->replaceAllUsesWith(V);
- if (AA) AA->deleteValue(PN);
- BB->getInstList().erase(PN);
- }
+ if (!AllInvariant) continue;
+
+ // The block has now been cleared of all instructions except for
+ // a comparison and a conditional branch. SimplifyCFG may be able
+ // to fold it now.
+ if (!FoldBranchToCommonDest(BI)) continue;
+
+ // Success. The block is now dead, so remove it from the loop,
+ // update the dominator tree and dominance frontier, and delete it.
+
+ DEBUG(dbgs() << "LoopSimplify: Eliminating exiting block ";
+ WriteAsOperand(dbgs(), ExitingBlock, false);
+ dbgs() << "\n");
+
+ assert(pred_begin(ExitingBlock) == pred_end(ExitingBlock));
+ Changed = true;
+ LI->removeBlock(ExitingBlock);
+
+ DominanceFrontier *DF = getAnalysisIfAvailable<DominanceFrontier>();
+ DomTreeNode *Node = DT->getNode(ExitingBlock);
+ const std::vector<DomTreeNodeBase<BasicBlock> *> &Children =
+ Node->getChildren();
+ while (!Children.empty()) {
+ DomTreeNode *Child = Children.front();
+ DT->changeImmediateDominator(Child, Node->getIDom());
+ if (DF) DF->changeImmediateDominator(Child->getBlock(),
+ Node->getIDom()->getBlock(),
+ DT);
}
- }
+ DT->eraseNode(ExitingBlock);
+ if (DF) DF->removeBlock(ExitingBlock);
- // Now that the PHI nodes are updated, actually move the edges from
- // Preds to point to NewBB instead of BB.
- //
- for (unsigned i = 0, e = Preds.size(); i != e; ++i) {
- TerminatorInst *TI = Preds[i]->getTerminator();
- for (unsigned s = 0, e = TI->getNumSuccessors(); s != e; ++s)
- if (TI->getSuccessor(s) == BB)
- TI->setSuccessor(s, NewBB);
- }
-
- } else { // Otherwise the loop is dead...
- for (BasicBlock::iterator I = BB->begin(); isa<PHINode>(I); ++I) {
- PHINode *PN = cast<PHINode>(I);
- // Insert dummy values as the incoming value...
- PN->addIncoming(Constant::getNullValue(PN->getType()), NewBB);
+ BI->getSuccessor(0)->removePredecessor(ExitingBlock);
+ BI->getSuccessor(1)->removePredecessor(ExitingBlock);
+ ExitingBlock->eraseFromParent();
}
}
- return NewBB;
+
+ return Changed;
}
/// InsertPreheaderForLoop - Once we discover that a loop doesn't have a
/// preheader, this method is called to insert one. This method has two phases:
/// preheader insertion and analysis updating.
///
-void LoopSimplify::InsertPreheaderForLoop(Loop *L) {
+BasicBlock *LoopSimplify::InsertPreheaderForLoop(Loop *L) {
BasicBlock *Header = L->getHeader();
// Compute the set of predecessors of the loop that are not in the loop.
- std::vector<BasicBlock*> OutsideBlocks;
+ SmallVector<BasicBlock*, 8> OutsideBlocks;
for (pred_iterator PI = pred_begin(Header), PE = pred_end(Header);
PI != PE; ++PI)
- if (!L->contains(*PI)) // Coming in from outside the loop?
- OutsideBlocks.push_back(*PI); // Keep track of it...
+ if (!L->contains(*PI)) { // Coming in from outside the loop?
+ // If the loop is branched to from an indirect branch, we won't
+ // be able to fully transform the loop, because it prohibits
+ // edge splitting.
+ if (isa<IndirectBrInst>((*PI)->getTerminator())) return 0;
+
+ // Keep track of it.
+ OutsideBlocks.push_back(*PI);
+ }
// Split out the loop pre-header.
BasicBlock *NewBB =
- SplitBlockPredecessors(Header, ".preheader", OutsideBlocks);
-
+ SplitBlockPredecessors(Header, &OutsideBlocks[0], OutsideBlocks.size(),
+ ".preheader", this);
- //===--------------------------------------------------------------------===//
- // Update analysis results now that we have performed the transformation
- //
+ DEBUG(dbgs() << "LoopSimplify: Creating pre-header ";
+ WriteAsOperand(dbgs(), NewBB, false);
+ dbgs() << "\n");
- // We know that we have loop information to update... update it now.
- if (Loop *Parent = L->getParentLoop())
- Parent->addBasicBlockToLoop(NewBB, *LI);
-
- UpdateDomInfoForRevectoredPreds(NewBB, OutsideBlocks);
-
// Make sure that NewBB is put someplace intelligent, which doesn't mess up
// code layout too horribly.
PlaceSplitBlockCarefully(NewBB, OutsideBlocks, L);
+
+ return NewBB;
}
/// RewriteLoopExitBlock - Ensure that the loop preheader dominates all exit
/// blocks. This method is used to split exit blocks that have predecessors
/// outside of the loop.
BasicBlock *LoopSimplify::RewriteLoopExitBlock(Loop *L, BasicBlock *Exit) {
- std::vector<BasicBlock*> LoopBlocks;
+ SmallVector<BasicBlock*, 8> LoopBlocks;
for (pred_iterator I = pred_begin(Exit), E = pred_end(Exit); I != E; ++I)
- if (L->contains(*I))
+ if (L->contains(*I)) {
+ // Don't do this if the loop is exited via an indirect branch.
+ if (isa<IndirectBrInst>((*I)->getTerminator())) return 0;
+
LoopBlocks.push_back(*I);
+ }
assert(!LoopBlocks.empty() && "No edges coming in from outside the loop?");
- BasicBlock *NewBB = SplitBlockPredecessors(Exit, ".loopexit", LoopBlocks);
-
- // Update Loop Information - we know that the new block will be in whichever
- // loop the Exit block is in. Note that it may not be in that immediate loop,
- // if the successor is some other loop header. In that case, we continue
- // walking up the loop tree to find a loop that contains both the successor
- // block and the predecessor block.
- Loop *SuccLoop = LI->getLoopFor(Exit);
- while (SuccLoop && !SuccLoop->contains(L->getHeader()))
- SuccLoop = SuccLoop->getParentLoop();
- if (SuccLoop)
- SuccLoop->addBasicBlockToLoop(NewBB, *LI);
-
- // Update dominator information (set, immdom, domtree, and domfrontier)
- UpdateDomInfoForRevectoredPreds(NewBB, LoopBlocks);
+ BasicBlock *NewBB = SplitBlockPredecessors(Exit, &LoopBlocks[0],
+ LoopBlocks.size(), ".loopexit",
+ this);
+
+ DEBUG(dbgs() << "LoopSimplify: Creating dedicated exit block ";
+ WriteAsOperand(dbgs(), NewBB, false);
+ dbgs() << "\n");
+
return NewBB;
}
/// AddBlockAndPredsToSet - Add the specified block, and all of its
/// predecessors, to the specified set, if it's not already in there. Stop
/// predecessor traversal when we reach StopBlock.
-static void AddBlockAndPredsToSet(BasicBlock *BB, BasicBlock *StopBlock,
+static void AddBlockAndPredsToSet(BasicBlock *InputBB, BasicBlock *StopBlock,
std::set<BasicBlock*> &Blocks) {
- if (!Blocks.insert(BB).second) return; // already processed.
- if (BB == StopBlock) return; // Stop here!
-
- for (pred_iterator I = pred_begin(BB), E = pred_end(BB); I != E; ++I)
- AddBlockAndPredsToSet(*I, StopBlock, Blocks);
+ std::vector<BasicBlock *> WorkList;
+ WorkList.push_back(InputBB);
+ do {
+ BasicBlock *BB = WorkList.back(); WorkList.pop_back();
+ if (Blocks.insert(BB).second && BB != StopBlock)
+ // If BB is not already processed and it is not a stop block then
+ // insert its predecessor in the work list
+ for (pred_iterator I = pred_begin(BB), E = pred_end(BB); I != E; ++I) {
+ BasicBlock *WBB = *I;
+ WorkList.push_back(WBB);
+ }
+ } while(!WorkList.empty());
}
/// FindPHIToPartitionLoops - The first part of loop-nestification is to find a
/// PHI node that tells us how to partition the loops.
-static PHINode *FindPHIToPartitionLoops(Loop *L, DominatorSet &DS,
+static PHINode *FindPHIToPartitionLoops(Loop *L, DominatorTree *DT,
AliasAnalysis *AA) {
for (BasicBlock::iterator I = L->getHeader()->begin(); isa<PHINode>(I); ) {
PHINode *PN = cast<PHINode>(I);
++I;
- if (Value *V = PN->hasConstantValue())
- if (!isa<Instruction>(V) || DS.dominates(cast<Instruction>(V), PN)) {
- // This is a degenerate PHI already, don't modify it!
- PN->replaceAllUsesWith(V);
- if (AA) AA->deleteValue(PN);
- PN->eraseFromParent();
- continue;
- }
+ if (Value *V = PN->hasConstantValue(DT)) {
+ // This is a degenerate PHI already, don't modify it!
+ PN->replaceAllUsesWith(V);
+ if (AA) AA->deleteValue(PN);
+ PN->eraseFromParent();
+ continue;
+ }
// Scan this PHI node looking for a use of the PHI node by itself.
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
// right after some 'outside block' block. This prevents the preheader from
// being placed inside the loop body, e.g. when the loop hasn't been rotated.
void LoopSimplify::PlaceSplitBlockCarefully(BasicBlock *NewBB,
- std::vector<BasicBlock*>&SplitPreds,
+ SmallVectorImpl<BasicBlock*> &SplitPreds,
Loop *L) {
// Check to see if NewBB is already well placed.
Function::iterator BBI = NewBB; --BBI;
/// If we are able to separate out a loop, return the new outer loop that was
/// created.
///
-Loop *LoopSimplify::SeparateNestedLoop(Loop *L) {
- PHINode *PN = FindPHIToPartitionLoops(L, getAnalysis<DominatorSet>(), AA);
+Loop *LoopSimplify::SeparateNestedLoop(Loop *L, LPPassManager &LPM) {
+ PHINode *PN = FindPHIToPartitionLoops(L, DT, AA);
if (PN == 0) return 0; // No known way to partition.
// Pull out all predecessors that have varying values in the loop. This
// handles the case when a PHI node has multiple instances of itself as
// arguments.
- std::vector<BasicBlock*> OuterLoopPreds;
+ SmallVector<BasicBlock*, 8> OuterLoopPreds;
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
if (PN->getIncomingValue(i) != PN ||
- !L->contains(PN->getIncomingBlock(i)))
+ !L->contains(PN->getIncomingBlock(i))) {
+ // We can't split indirectbr edges.
+ if (isa<IndirectBrInst>(PN->getIncomingBlock(i)->getTerminator()))
+ return 0;
+
OuterLoopPreds.push_back(PN->getIncomingBlock(i));
+ }
- BasicBlock *Header = L->getHeader();
- BasicBlock *NewBB = SplitBlockPredecessors(Header, ".outer", OuterLoopPreds);
+ DEBUG(dbgs() << "LoopSimplify: Splitting out a new outer loop\n");
- // Update dominator information (set, immdom, domtree, and domfrontier)
- UpdateDomInfoForRevectoredPreds(NewBB, OuterLoopPreds);
+ BasicBlock *Header = L->getHeader();
+ BasicBlock *NewBB = SplitBlockPredecessors(Header, &OuterLoopPreds[0],
+ OuterLoopPreds.size(),
+ ".outer", this);
// Make sure that NewBB is put someplace intelligent, which doesn't mess up
// code layout too horribly.
else
LI->changeTopLevelLoop(L, NewOuter);
- // This block is going to be our new header block: add it to this loop and all
- // parent loops.
- NewOuter->addBasicBlockToLoop(NewBB, *LI);
-
// L is now a subloop of our outer loop.
NewOuter->addChildLoop(L);
- for (unsigned i = 0, e = L->getBlocks().size(); i != e; ++i)
- NewOuter->addBlockEntry(L->getBlocks()[i]);
+ // Add the new loop to the pass manager queue.
+ LPM.insertLoopIntoQueue(NewOuter);
+
+ for (Loop::block_iterator I = L->block_begin(), E = L->block_end();
+ I != E; ++I)
+ NewOuter->addBlockEntry(*I);
+
+ // Now reset the header in L, which had been moved by
+ // SplitBlockPredecessors for the outer loop.
+ L->moveToHeader(Header);
// Determine which blocks should stay in L and which should be moved out to
// the Outer loop now.
- DominatorSet &DS = getAnalysis<DominatorSet>();
std::set<BasicBlock*> BlocksInL;
for (pred_iterator PI = pred_begin(Header), E = pred_end(Header); PI!=E; ++PI)
- if (DS.dominates(Header, *PI))
+ if (DT->dominates(Header, *PI))
AddBlockAndPredsToSet(*PI, Header, BlocksInL);
// Scan all of the loop children of L, moving them to OuterLoop if they are
// not part of the inner loop.
- for (Loop::iterator I = L->begin(); I != L->end(); )
- if (BlocksInL.count((*I)->getHeader()))
+ const std::vector<Loop*> &SubLoops = L->getSubLoops();
+ for (size_t I = 0; I != SubLoops.size(); )
+ if (BlocksInL.count(SubLoops[I]->getHeader()))
++I; // Loop remains in L
else
- NewOuter->addChildLoop(L->removeChildLoop(I));
+ NewOuter->addChildLoop(L->removeChildLoop(SubLoops.begin() + I));
// Now that we know which blocks are in L and which need to be moved to
// OuterLoop, move any blocks that need it.
/// backedges to target a new basic block and have that block branch to the loop
/// header. This ensures that loops have exactly one backedge.
///
-void LoopSimplify::InsertUniqueBackedgeBlock(Loop *L) {
+BasicBlock *
+LoopSimplify::InsertUniqueBackedgeBlock(Loop *L, BasicBlock *Preheader) {
assert(L->getNumBackEdges() > 1 && "Must have > 1 backedge!");
// Get information about the loop
- BasicBlock *Preheader = L->getLoopPreheader();
BasicBlock *Header = L->getHeader();
Function *F = Header->getParent();
+ // Unique backedge insertion currently depends on having a preheader.
+ if (!Preheader)
+ return 0;
+
// Figure out which basic blocks contain back-edges to the loop header.
std::vector<BasicBlock*> BackedgeBlocks;
for (pred_iterator I = pred_begin(Header), E = pred_end(Header); I != E; ++I)
if (*I != Preheader) BackedgeBlocks.push_back(*I);
// Create and insert the new backedge block...
- BasicBlock *BEBlock = new BasicBlock(Header->getName()+".backedge", F);
- BranchInst *BETerminator = new BranchInst(Header, BEBlock);
+ BasicBlock *BEBlock = BasicBlock::Create(Header->getContext(),
+ Header->getName()+".backedge", F);
+ BranchInst *BETerminator = BranchInst::Create(Header, BEBlock);
+
+ DEBUG(dbgs() << "LoopSimplify: Inserting unique backedge block ";
+ WriteAsOperand(dbgs(), BEBlock, false);
+ dbgs() << "\n");
// Move the new backedge block to right after the last backedge block.
Function::iterator InsertPos = BackedgeBlocks.back(); ++InsertPos;
// the backedge block which correspond to any PHI nodes in the header block.
for (BasicBlock::iterator I = Header->begin(); isa<PHINode>(I); ++I) {
PHINode *PN = cast<PHINode>(I);
- PHINode *NewPN = new PHINode(PN->getType(), PN->getName()+".be",
- BETerminator);
+ PHINode *NewPN = PHINode::Create(PN->getType(), PN->getName()+".be",
+ BETerminator);
NewPN->reserveOperandSpace(BackedgeBlocks.size());
if (AA) AA->copyValue(PN, NewPN);
// Update Loop Information - we know that this block is now in the current
// loop and all parent loops.
- L->addBasicBlockToLoop(BEBlock, *LI);
+ L->addBasicBlockToLoop(BEBlock, LI->getBase());
- // Update dominator information (set, immdom, domtree, and domfrontier)
- UpdateDomInfoForRevectoredPreds(BEBlock, BackedgeBlocks);
-}
+ // Update dominator information
+ DT->splitBlock(BEBlock);
+ if (DominanceFrontier *DF = getAnalysisIfAvailable<DominanceFrontier>())
+ DF->splitBlock(BEBlock);
-/// UpdateDomInfoForRevectoredPreds - This method is used to update the four
-/// different kinds of dominator information (dominator sets, immediate
-/// dominators, dominator trees, and dominance frontiers) after a new block has
-/// been added to the CFG.
-///
-/// This only supports the case when an existing block (known as "NewBBSucc"),
-/// had some of its predecessors factored into a new basic block. This
-/// transformation inserts a new basic block ("NewBB"), with a single
-/// unconditional branch to NewBBSucc, and moves some predecessors of
-/// "NewBBSucc" to now branch to NewBB. These predecessors are listed in
-/// PredBlocks, even though they are the same as
-/// pred_begin(NewBB)/pred_end(NewBB).
-///
-void LoopSimplify::UpdateDomInfoForRevectoredPreds(BasicBlock *NewBB,
- std::vector<BasicBlock*> &PredBlocks) {
- assert(!PredBlocks.empty() && "No predblocks??");
- assert(succ_begin(NewBB) != succ_end(NewBB) &&
- ++succ_begin(NewBB) == succ_end(NewBB) &&
- "NewBB should have a single successor!");
- BasicBlock *NewBBSucc = *succ_begin(NewBB);
- DominatorSet &DS = getAnalysis<DominatorSet>();
-
- // Update dominator information... The blocks that dominate NewBB are the
- // intersection of the dominators of predecessors, plus the block itself.
- //
- DominatorSet::DomSetType NewBBDomSet = DS.getDominators(PredBlocks[0]);
- {
- unsigned i, e = PredBlocks.size();
- // It is possible for some preds to not be reachable, and thus have empty
- // dominator sets (all blocks must dom themselves, so no domset would
- // otherwise be empty). If we see any of these, don't intersect with them,
- // as that would certainly leave the resultant set empty.
- for (i = 1; NewBBDomSet.empty(); ++i) {
- assert(i != e && "Didn't find reachable pred?");
- NewBBDomSet = DS.getDominators(PredBlocks[i]);
- }
-
- // Intersect the rest of the non-empty sets.
- for (; i != e; ++i) {
- const DominatorSet::DomSetType &PredDS = DS.getDominators(PredBlocks[i]);
- if (!PredDS.empty())
- set_intersect(NewBBDomSet, PredDS);
- }
- NewBBDomSet.insert(NewBB); // All blocks dominate themselves.
- DS.addBasicBlock(NewBB, NewBBDomSet);
- }
+ return BEBlock;
+}
- // The newly inserted basic block will dominate existing basic blocks iff the
- // PredBlocks dominate all of the non-pred blocks. If all predblocks dominate
- // the non-pred blocks, then they all must be the same block!
- //
- bool NewBBDominatesNewBBSucc = true;
- {
- BasicBlock *OnePred = PredBlocks[0];
- unsigned i, e = PredBlocks.size();
- for (i = 1; !DS.isReachable(OnePred); ++i) {
- assert(i != e && "Didn't find reachable pred?");
- OnePred = PredBlocks[i];
- }
-
- for (; i != e; ++i)
- if (PredBlocks[i] != OnePred && DS.isReachable(PredBlocks[i])) {
- NewBBDominatesNewBBSucc = false;
+void LoopSimplify::verifyAnalysis() const {
+ // It used to be possible to just assert L->isLoopSimplifyForm(), however
+ // with the introduction of indirectbr, there are now cases where it's
+ // not possible to transform a loop as necessary. We can at least check
+ // that there is an indirectbr near any time there's trouble.
+
+ // Indirectbr can interfere with preheader and unique backedge insertion.
+ if (!L->getLoopPreheader() || !L->getLoopLatch()) {
+ bool HasIndBrPred = false;
+ for (pred_iterator PI = pred_begin(L->getHeader()),
+ PE = pred_end(L->getHeader()); PI != PE; ++PI)
+ if (isa<IndirectBrInst>((*PI)->getTerminator())) {
+ HasIndBrPred = true;
break;
}
-
- if (NewBBDominatesNewBBSucc)
- for (pred_iterator PI = pred_begin(NewBBSucc), E = pred_end(NewBBSucc);
- PI != E; ++PI)
- if (*PI != NewBB && !DS.dominates(NewBBSucc, *PI)) {
- NewBBDominatesNewBBSucc = false;
- break;
- }
+ assert(HasIndBrPred &&
+ "LoopSimplify has no excuse for missing loop header info!");
}
- // The other scenario where the new block can dominate its successors are when
- // all predecessors of NewBBSucc that are not NewBB are dominated by NewBBSucc
- // already.
- if (!NewBBDominatesNewBBSucc) {
- NewBBDominatesNewBBSucc = true;
- for (pred_iterator PI = pred_begin(NewBBSucc), E = pred_end(NewBBSucc);
- PI != E; ++PI)
- if (*PI != NewBB && !DS.dominates(NewBBSucc, *PI)) {
- NewBBDominatesNewBBSucc = false;
+ // Indirectbr can interfere with exit block canonicalization.
+ if (!L->hasDedicatedExits()) {
+ bool HasIndBrExiting = false;
+ SmallVector<BasicBlock*, 8> ExitingBlocks;
+ L->getExitingBlocks(ExitingBlocks);
+ for (unsigned i = 0, e = ExitingBlocks.size(); i != e; ++i)
+ if (isa<IndirectBrInst>((ExitingBlocks[i])->getTerminator())) {
+ HasIndBrExiting = true;
break;
}
- }
-
- // If NewBB dominates some blocks, then it will dominate all blocks that
- // NewBBSucc does.
- if (NewBBDominatesNewBBSucc) {
- BasicBlock *PredBlock = PredBlocks[0];
- Function *F = NewBB->getParent();
- for (Function::iterator I = F->begin(), E = F->end(); I != E; ++I)
- if (DS.dominates(NewBBSucc, I))
- DS.addDominator(I, NewBB);
- }
-
- // Update immediate dominator information if we have it.
- BasicBlock *NewBBIDom = 0;
- if (ImmediateDominators *ID = getAnalysisToUpdate<ImmediateDominators>()) {
- // To find the immediate dominator of the new exit node, we trace up the
- // immediate dominators of a predecessor until we find a basic block that
- // dominates the exit block.
- //
- BasicBlock *Dom = PredBlocks[0]; // Some random predecessor.
-
- // Find a reachable pred.
- for (unsigned i = 1; !DS.isReachable(Dom); ++i) {
- assert(i != PredBlocks.size() && "Didn't find reachable pred!");
- Dom = PredBlocks[i];
- }
-
- while (!NewBBDomSet.count(Dom)) { // Loop until we find a dominator.
- assert(Dom != 0 && "No shared dominator found???");
- Dom = ID->get(Dom);
- }
-
- // Set the immediate dominator now...
- ID->addNewBlock(NewBB, Dom);
- NewBBIDom = Dom; // Reuse this if calculating DominatorTree info...
-
- // If NewBB strictly dominates other blocks, we need to update their idom's
- // now. The only block that need adjustment is the NewBBSucc block, whose
- // idom should currently be set to PredBlocks[0].
- if (NewBBDominatesNewBBSucc)
- ID->setImmediateDominator(NewBBSucc, NewBB);
- }
-
- // Update DominatorTree information if it is active.
- if (DominatorTree *DT = getAnalysisToUpdate<DominatorTree>()) {
- // If we don't have ImmediateDominator info around, calculate the idom as
- // above.
- DominatorTree::Node *NewBBIDomNode;
- if (NewBBIDom) {
- NewBBIDomNode = DT->getNode(NewBBIDom);
- } else {
- // Scan all the pred blocks that were pulled out. Any individual one may
- // actually be unreachable, which would mean it doesn't have dom info.
- NewBBIDomNode = 0;
- for (unsigned i = 0; !NewBBIDomNode; ++i) {
- assert(i != PredBlocks.size() && "No reachable preds?");
- NewBBIDomNode = DT->getNode(PredBlocks[i]);
- }
-
- while (!NewBBDomSet.count(NewBBIDomNode->getBlock())) {
- NewBBIDomNode = NewBBIDomNode->getIDom();
- assert(NewBBIDomNode && "No shared dominator found??");
- }
- NewBBIDom = NewBBIDomNode->getBlock();
- }
-
- // Create the new dominator tree node... and set the idom of NewBB.
- DominatorTree::Node *NewBBNode = DT->createNewNode(NewBB, NewBBIDomNode);
-
- // If NewBB strictly dominates other blocks, then it is now the immediate
- // dominator of NewBBSucc. Update the dominator tree as appropriate.
- if (NewBBDominatesNewBBSucc) {
- DominatorTree::Node *NewBBSuccNode = DT->getNode(NewBBSucc);
- DT->changeImmediateDominator(NewBBSuccNode, NewBBNode);
- }
- }
-
- // Update ET-Forest information if it is active.
- if (ETForest *EF = getAnalysisToUpdate<ETForest>()) {
- EF->addNewBlock(NewBB, NewBBIDom);
- if (NewBBDominatesNewBBSucc)
- EF->setImmediateDominator(NewBBSucc, NewBB);
- }
-
- // Update dominance frontier information...
- if (DominanceFrontier *DF = getAnalysisToUpdate<DominanceFrontier>()) {
- // If NewBB dominates NewBBSucc, then DF(NewBB) is now going to be the
- // DF(PredBlocks[0]) without the stuff that the new block does not dominate
- // a predecessor of.
- if (NewBBDominatesNewBBSucc) {
- DominanceFrontier::iterator DFI = DF->find(PredBlocks[0]);
- if (DFI != DF->end()) {
- DominanceFrontier::DomSetType Set = DFI->second;
- // Filter out stuff in Set that we do not dominate a predecessor of.
- for (DominanceFrontier::DomSetType::iterator SetI = Set.begin(),
- E = Set.end(); SetI != E;) {
- bool DominatesPred = false;
- for (pred_iterator PI = pred_begin(*SetI), E = pred_end(*SetI);
- PI != E; ++PI)
- if (DS.dominates(NewBB, *PI))
- DominatesPred = true;
- if (!DominatesPred)
- Set.erase(SetI++);
- else
- ++SetI;
- }
-
- DF->addBasicBlock(NewBB, Set);
- }
-
- } else {
- // DF(NewBB) is {NewBBSucc} because NewBB does not strictly dominate
- // NewBBSucc, but it does dominate itself (and there is an edge (NewBB ->
- // NewBBSucc)). NewBBSucc is the single successor of NewBB.
- DominanceFrontier::DomSetType NewDFSet;
- NewDFSet.insert(NewBBSucc);
- DF->addBasicBlock(NewBB, NewDFSet);
- }
-
- // Now we must loop over all of the dominance frontiers in the function,
- // replacing occurrences of NewBBSucc with NewBB in some cases. All
- // blocks that dominate a block in PredBlocks and contained NewBBSucc in
- // their dominance frontier must be updated to contain NewBB instead.
- //
- for (unsigned i = 0, e = PredBlocks.size(); i != e; ++i) {
- BasicBlock *Pred = PredBlocks[i];
- // Get all of the dominators of the predecessor...
- const DominatorSet::DomSetType &PredDoms = DS.getDominators(Pred);
- for (DominatorSet::DomSetType::const_iterator PDI = PredDoms.begin(),
- PDE = PredDoms.end(); PDI != PDE; ++PDI) {
- BasicBlock *PredDom = *PDI;
-
- // If the NewBBSucc node is in DF(PredDom), then PredDom didn't
- // dominate NewBBSucc but did dominate a predecessor of it. Now we
- // change this entry to include NewBB in the DF instead of NewBBSucc.
- DominanceFrontier::iterator DFI = DF->find(PredDom);
- assert(DFI != DF->end() && "No dominance frontier for node?");
- if (DFI->second.count(NewBBSucc)) {
- // If NewBBSucc should not stay in our dominator frontier, remove it.
- // We remove it unless there is a predecessor of NewBBSucc that we
- // dominate, but we don't strictly dominate NewBBSucc.
- bool ShouldRemove = true;
- if (PredDom == NewBBSucc || !DS.dominates(PredDom, NewBBSucc)) {
- // Okay, we know that PredDom does not strictly dominate NewBBSucc.
- // Check to see if it dominates any predecessors of NewBBSucc.
- for (pred_iterator PI = pred_begin(NewBBSucc),
- E = pred_end(NewBBSucc); PI != E; ++PI)
- if (DS.dominates(PredDom, *PI)) {
- ShouldRemove = false;
- break;
- }
- }
-
- if (ShouldRemove)
- DF->removeFromFrontier(DFI, NewBBSucc);
- DF->addToFrontier(DFI, NewBB);
- }
- }
- }
+ assert(HasIndBrExiting &&
+ "LoopSimplify has no excuse for missing exit block info!");
}
}
-