//===- LoopSimplify.cpp - Loop Canonicalization Pass ----------------------===//
-//
+//
// 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 pass performs several transformations to transform natural loops into a
//
// Loop exit-block insertion guarantees that all exit blocks from the loop
// (blocks which are outside of the loop that have predecessors inside of the
-// loop) are dominated by the loop header. This simplifies transformations such
-// as store-sinking that are built into LICM.
+// loop) only have predecessors from inside of the loop (and are thus dominated
+// by the loop header). This simplifies transformations such as store-sinking
+// that are built into LICM.
//
// This pass also guarantees that loops will have exactly one backedge.
//
//
//===----------------------------------------------------------------------===//
+#define DEBUG_TYPE "loopsimplify"
#include "llvm/Transforms/Scalar.h"
+#include "llvm/Constant.h"
+#include "llvm/Instructions.h"
+#include "llvm/Function.h"
+#include "llvm/Type.h"
+#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/Dominators.h"
#include "llvm/Analysis/LoopInfo.h"
-#include "llvm/Function.h"
-#include "llvm/iTerminators.h"
-#include "llvm/iPHINode.h"
-#include "llvm/Constant.h"
#include "llvm/Support/CFG.h"
-#include "Support/SetOperations.h"
-#include "Support/Statistic.h"
-#include "Support/DepthFirstIterator.h"
+#include "llvm/Support/Compiler.h"
+#include "llvm/ADT/SetOperations.h"
+#include "llvm/ADT/SetVector.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/ADT/DepthFirstIterator.h"
+using namespace llvm;
-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 blocks inserted");
-
- struct LoopSimplify : public FunctionPass {
+ struct VISIBILITY_HIDDEN LoopSimplify : public FunctionPass {
+ static char ID; // Pass identification, replacement for typeid
+ LoopSimplify() : FunctionPass((intptr_t)&ID) {}
+
+ // AA - If we have an alias analysis object to update, this is it, otherwise
+ // this is null.
+ AliasAnalysis *AA;
+ LoopInfo *LI;
+ DominatorTree *DT;
virtual bool runOnFunction(Function &F);
-
+
virtual void getAnalysisUsage(AnalysisUsage &AU) const {
// We need loop information to identify the loops...
AU.addRequired<LoopInfo>();
- AU.addRequired<DominatorSet>();
+ AU.addRequired<DominatorTree>();
AU.addPreserved<LoopInfo>();
- AU.addPreserved<DominatorSet>();
- AU.addPreserved<ImmediateDominators>();
AU.addPreserved<DominatorTree>();
AU.addPreserved<DominanceFrontier>();
- AU.addPreservedID(BreakCriticalEdgesID); // No crit edges added....
+ AU.addPreservedID(BreakCriticalEdgesID); // No critical edges added.
}
+
+ /// verifyAnalysis() - Verify loop nest.
+ void verifyAnalysis() const {
+#ifndef NDEBUG
+ LoopInfo *NLI = &getAnalysis<LoopInfo>();
+ for (LoopInfo::iterator I = NLI->begin(), E = NLI->end(); I != E; ++I)
+ (*I)->verifyLoop();
+#endif
+ }
+
private:
bool ProcessLoop(Loop *L);
BasicBlock *SplitBlockPredecessors(BasicBlock *BB, const char *Suffix,
const std::vector<BasicBlock*> &Preds);
- void RewriteLoopExitBlock(Loop *L, BasicBlock *Exit);
+ BasicBlock *RewriteLoopExitBlock(Loop *L, BasicBlock *Exit);
void InsertPreheaderForLoop(Loop *L);
+ Loop *SeparateNestedLoop(Loop *L);
void InsertUniqueBackedgeBlock(Loop *L);
-
- void UpdateDomInfoForRevectoredPreds(BasicBlock *NewBB,
- std::vector<BasicBlock*> &PredBlocks);
+ void PlaceSplitBlockCarefully(BasicBlock *NewBB,
+ std::vector<BasicBlock*> &SplitPreds,
+ Loop *L);
};
- RegisterOpt<LoopSimplify>
+ char LoopSimplify::ID = 0;
+ RegisterPass<LoopSimplify>
X("loopsimplify", "Canonicalize natural loops", true);
}
// Publically exposed interface to pass...
-const PassInfo *LoopSimplifyID = X.getPassInfo();
-Pass *createLoopSimplifyPass() { return new LoopSimplify(); }
+const PassInfo *llvm::LoopSimplifyID = X.getPassInfo();
+FunctionPass *llvm::createLoopSimplifyPass() { return new LoopSimplify(); }
/// runOnFunction - 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 Changed = false;
- LoopInfo &LI = getAnalysis<LoopInfo>();
-
- for (unsigned i = 0, e = LI.getTopLevelLoops().size(); i != e; ++i)
- Changed |= ProcessLoop(LI.getTopLevelLoops()[i]);
+ LI = &getAnalysis<LoopInfo>();
+ AA = getAnalysisToUpdate<AliasAnalysis>();
+ DT = &getAnalysis<DominatorTree>();
+
+ // 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);
return Changed;
}
-
/// ProcessLoop - Walk the loop structure in depth first order, ensuring that
/// all loops have preheaders.
///
bool LoopSimplify::ProcessLoop(Loop *L) {
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?");
- // Does the loop already have a preheader? If so, don't modify the loop...
+ // Does the loop already have a preheader? If so, don't insert one.
if (L->getLoopPreheader() == 0) {
InsertPreheaderForLoop(L);
NumInserted++;
Changed = true;
}
- // Regardless of whether or not we added a preheader to the loop we must
- // guarantee that the preheader dominates all exit nodes. If there are any
- // exit nodes not dominated, split them now.
- DominatorSet &DS = getAnalysis<DominatorSet>();
- BasicBlock *Header = L->getHeader();
- for (unsigned i = 0, e = L->getExitBlocks().size(); i != e; ++i)
- if (!DS.dominates(Header, L->getExitBlocks()[i])) {
- RewriteLoopExitBlock(L, L->getExitBlocks()[i]);
- assert(DS.dominates(Header, L->getExitBlocks()[i]) &&
- "RewriteLoopExitBlock failed?");
- 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;
+ L->getExitBlocks(ExitBlocks);
+
+ SetVector<BasicBlock*> ExitBlockSet(ExitBlocks.begin(), ExitBlocks.end());
+ for (SetVector<BasicBlock*>::iterator I = ExitBlockSet.begin(),
+ E = ExitBlockSet.end(); I != E; ++I) {
+ BasicBlock *ExitBlock = *I;
+ for (pred_iterator PI = pred_begin(ExitBlock), PE = pred_end(ExitBlock);
+ PI != PE; ++PI)
+ // Must be exactly this loop: no subloops, parent loops, or non-loop preds
+ // allowed.
+ if (!L->contains(*PI)) {
+ 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) {
+ // 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)) {
+ ++NumNested;
+ // This is a big restructuring change, reprocess the whole loop.
+ ProcessLoop(NL);
+ Changed = true;
+ // GCC doesn't tail recursion eliminate this.
+ goto ReprocessLoop;
+ }
}
- // The preheader may have more than two predecessors at this point (from the
- // preheader and from the backedges). To simplify the loop more, insert an
- // extra back-edge block in the loop so that there is exactly one backedge.
- if (L->getNumBackEdges() != 1) {
+ // 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;
}
- const std::vector<Loop*> &SubLoops = L->getSubLoops();
- for (unsigned i = 0, e = SubLoops.size(); i != e; ++i)
- Changed |= ProcessLoop(SubLoops[i]);
+ // Scan over the PHI nodes in the loop header. Since they now have only two
+ // incoming values (the loop is canonicalized), we may have simplified the PHI
+ // down to 'X = phi [X, Y]', which should be replaced with 'Y'.
+ 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;
}
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);
+ 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->getInstList().push_back(BI);
-
+ 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
// incoming edges in BB into new PHI nodes in NewBB.
//
if (!Preds.empty()) { // Is the loop not obviously dead?
- for (BasicBlock::iterator I = BB->begin();
- PHINode *PN = dyn_cast<PHINode>(I); ++I) {
-
- // 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);
-
- // 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]);
- NewPHI->addIncoming(V, Preds[i]);
+ // 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;
+ 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(NewPHI, NewBB);
+ // edge.
+ PN->addIncoming(InVal, NewBB);
+
+ // Can we eliminate this phi node now?
+ if (Value *V = PN->hasConstantValue(true)) {
+ Instruction *I = dyn_cast<Instruction>(V);
+ // If I is in NewBB, the Dominator call will fail, because NewBB isn't
+ // registered in DominatorTree yet. Handle this case explicitly.
+ if (!I || (I->getParent() != NewBB &&
+ getAnalysis<DominatorTree>().dominates(I, PN))) {
+ PN->replaceAllUsesWith(V);
+ if (AA) AA->deleteValue(PN);
+ BB->getInstList().erase(PN);
+ }
+ }
}
-
+
// Now that the PHI nodes are updated, actually move the edges from
// Preds to point to NewBB instead of BB.
//
if (TI->getSuccessor(s) == BB)
TI->setSuccessor(s, NewBB);
}
-
+
} else { // Otherwise the loop is dead...
- for (BasicBlock::iterator I = BB->begin();
- PHINode *PN = dyn_cast<PHINode>(I); ++I)
+ 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);
- }
- return NewBB;
-}
-
-// ChangeExitBlock - This recursive function is used to change any exit blocks
-// that use OldExit to use NewExit instead. This is recursive because children
-// may need to be processed as well.
-//
-static void ChangeExitBlock(Loop *L, BasicBlock *OldExit, BasicBlock *NewExit) {
- if (L->hasExitBlock(OldExit)) {
- L->changeExitBlock(OldExit, NewExit);
- const std::vector<Loop*> &SubLoops = L->getSubLoops();
- for (unsigned i = 0, e = SubLoops.size(); i != e; ++i)
- ChangeExitBlock(SubLoops[i], OldExit, NewExit);
+ }
}
-}
+ return NewBB;
+}
/// InsertPreheaderForLoop - Once we discover that a loop doesn't have a
/// preheader, this method is called to insert one. This method has two phases:
std::vector<BasicBlock*> 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...
-
- // Split out the loop pre-header
+ if (!L->contains(*PI)) // Coming in from outside the loop?
+ OutsideBlocks.push_back(*PI); // Keep track of it...
+
+ // Split out the loop pre-header.
BasicBlock *NewBB =
SplitBlockPredecessors(Header, ".preheader", OutsideBlocks);
+
//===--------------------------------------------------------------------===//
// Update analysis results now that we have performed the transformation
//
-
+
// We know that we have loop information to update... update it now.
if (Loop *Parent = L->getParentLoop())
- Parent->addBasicBlockToLoop(NewBB, getAnalysis<LoopInfo>());
+ Parent->addBasicBlockToLoop(NewBB, *LI);
- // If the header for the loop used to be an exit node for another loop, then
- // we need to update this to know that the loop-preheader is now the exit
- // node. Note that the only loop that could have our header as an exit node
- // is a sibling loop, ie, one with the same parent loop, or one if it's
- // children.
- //
- const std::vector<Loop*> *ParentSubLoops;
- if (Loop *Parent = L->getParentLoop())
- ParentSubLoops = &Parent->getSubLoops();
- else // Must check top-level loops...
- ParentSubLoops = &getAnalysis<LoopInfo>().getTopLevelLoops();
-
- // Loop over all sibling loops, performing the substitution (recursively to
- // include child loops)...
- for (unsigned i = 0, e = ParentSubLoops->size(); i != e; ++i)
- ChangeExitBlock((*ParentSubLoops)[i], Header, NewBB);
-
- DominatorSet &DS = getAnalysis<DominatorSet>(); // Update dominator info
- {
- // The blocks that dominate NewBB are the blocks that dominate Header,
- // minus Header, plus NewBB.
- DominatorSet::DomSetType DomSet = DS.getDominators(Header);
- DomSet.insert(NewBB); // We dominate ourself
- DomSet.erase(Header); // Header does not dominate us...
- DS.addBasicBlock(NewBB, DomSet);
-
- // The newly created basic block dominates all nodes dominated by Header.
- for (Function::iterator I = Header->getParent()->begin(),
- E = Header->getParent()->end(); I != E; ++I)
- if (DS.dominates(Header, I))
- DS.addDominator(I, NewBB);
- }
-
- // Update immediate dominator information if we have it...
- if (ImmediateDominators *ID = getAnalysisToUpdate<ImmediateDominators>()) {
- // Whatever i-dominated the header node now immediately dominates NewBB
- ID->addNewBlock(NewBB, ID->get(Header));
-
- // The preheader now is the immediate dominator for the header node...
- ID->setImmediateDominator(Header, NewBB);
- }
-
- // Update DominatorTree information if it is active.
- if (DominatorTree *DT = getAnalysisToUpdate<DominatorTree>()) {
- // The immediate dominator of the preheader is the immediate dominator of
- // the old header.
- //
- DominatorTree::Node *HeaderNode = DT->getNode(Header);
- DominatorTree::Node *PHNode = DT->createNewNode(NewBB,
- HeaderNode->getIDom());
-
- // Change the header node so that PNHode is the new immediate dominator
- DT->changeImmediateDominator(HeaderNode, PHNode);
- }
+ DT->splitBlock(NewBB);
+ if (DominanceFrontier *DF = getAnalysisToUpdate<DominanceFrontier>())
+ DF->splitBlock(NewBB);
- // Update dominance frontier information...
- if (DominanceFrontier *DF = getAnalysisToUpdate<DominanceFrontier>()) {
- // The DF(NewBB) is just (DF(Header)-Header), because NewBB dominates
- // everything that Header does, and it strictly dominates Header in
- // addition.
- assert(DF->find(Header) != DF->end() && "Header node doesn't have DF set?");
- DominanceFrontier::DomSetType NewDFSet = DF->find(Header)->second;
- NewDFSet.erase(Header);
- DF->addBasicBlock(NewBB, NewDFSet);
-
- // Now we must loop over all of the dominance frontiers in the function,
- // replacing occurrences of Header with NewBB in some cases. If a block
- // dominates a (now) predecessor of NewBB, but did not strictly dominate
- // Header, it will have Header in it's DF set, but should now have NewBB in
- // its set.
- for (unsigned i = 0, e = OutsideBlocks.size(); i != e; ++i) {
- // Get all of the dominators of the predecessor...
- const DominatorSet::DomSetType &PredDoms =
- DS.getDominators(OutsideBlocks[i]);
- for (DominatorSet::DomSetType::const_iterator PDI = PredDoms.begin(),
- PDE = PredDoms.end(); PDI != PDE; ++PDI) {
- BasicBlock *PredDom = *PDI;
- // If the loop header is in DF(PredDom), then PredDom didn't dominate
- // the header but did dominate a predecessor outside of the loop. Now
- // we change this entry to include the preheader in the DF instead of
- // the header.
- DominanceFrontier::iterator DFI = DF->find(PredDom);
- assert(DFI != DF->end() && "No dominance frontier for node?");
- if (DFI->second.count(Header)) {
- DF->removeFromFrontier(DFI, Header);
- DF->addToFrontier(DFI, NewBB);
- }
- }
- }
- }
+ // Make sure that NewBB is put someplace intelligent, which doesn't mess up
+ // code layout too horribly.
+ PlaceSplitBlockCarefully(NewBB, OutsideBlocks, L);
}
-void LoopSimplify::RewriteLoopExitBlock(Loop *L, BasicBlock *Exit) {
- DominatorSet &DS = getAnalysis<DominatorSet>();
- assert(!DS.dominates(L->getHeader(), Exit) &&
- "Loop already dominates exit block??");
- assert(std::find(L->getExitBlocks().begin(), L->getExitBlocks().end(), Exit)
- != L->getExitBlocks().end() && "Not a current exit block!");
-
+/// 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;
for (pred_iterator I = pred_begin(Exit), E = pred_end(Exit); I != E; ++I)
if (L->contains(*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 the parent
- // loop of L.
+ // 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
+ DT->splitBlock(NewBB);
+ if (DominanceFrontier *DF = getAnalysisToUpdate<DominanceFrontier>())
+ DF->splitBlock(NewBB);
+
+ 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 *InputBB, BasicBlock *StopBlock,
+ std::set<BasicBlock*> &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, 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) || DT->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;
+ }
+
+ // Scan this PHI node looking for a use of the PHI node by itself.
+ for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
+ if (PN->getIncomingValue(i) == PN &&
+ L->contains(PN->getIncomingBlock(i)))
+ // We found something tasty to remove.
+ return PN;
+ }
+ return 0;
+}
+
+// PlaceSplitBlockCarefully - If the block isn't already, move the new block to
+// 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,
+ Loop *L) {
+ // Check to see if NewBB is already well placed.
+ Function::iterator BBI = NewBB; --BBI;
+ for (unsigned i = 0, e = SplitPreds.size(); i != e; ++i) {
+ if (&*BBI == SplitPreds[i])
+ return;
+ }
+
+ // If it isn't already after an outside block, move it after one. This is
+ // always good as it makes the uncond branch from the outside block into a
+ // fall-through.
+
+ // Figure out *which* outside block to put this after. Prefer an outside
+ // block that neighbors a BB actually in the loop.
+ BasicBlock *FoundBB = 0;
+ for (unsigned i = 0, e = SplitPreds.size(); i != e; ++i) {
+ Function::iterator BBI = SplitPreds[i];
+ if (++BBI != NewBB->getParent()->end() &&
+ L->contains(BBI)) {
+ FoundBB = SplitPreds[i];
+ break;
+ }
+ }
+
+ // If our heuristic for a *good* bb to place this after doesn't find
+ // anything, just pick something. It's likely better than leaving it within
+ // the loop.
+ if (!FoundBB)
+ FoundBB = SplitPreds[0];
+ NewBB->moveAfter(FoundBB);
+}
+
+
+/// SeparateNestedLoop - If this loop has multiple backedges, try to pull one of
+/// them out into a nested loop. This is important for code that looks like
+/// this:
+///
+/// Loop:
+/// ...
+/// br cond, Loop, Next
+/// ...
+/// br cond2, Loop, Out
+///
+/// To identify this common case, we look at the PHI nodes in the header of the
+/// loop. PHI nodes with unchanging values on one backedge correspond to values
+/// that change in the "outer" loop, but not in the "inner" loop.
+///
+/// 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, 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;
+ for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
+ if (PN->getIncomingValue(i) != PN ||
+ !L->contains(PN->getIncomingBlock(i)))
+ OuterLoopPreds.push_back(PN->getIncomingBlock(i));
+
+ BasicBlock *Header = L->getHeader();
+ BasicBlock *NewBB = SplitBlockPredecessors(Header, ".outer", OuterLoopPreds);
+
+ // Update dominator information
+ DT->splitBlock(NewBB);
+ if (DominanceFrontier *DF = getAnalysisToUpdate<DominanceFrontier>())
+ DF->splitBlock(NewBB);
+
+ // Make sure that NewBB is put someplace intelligent, which doesn't mess up
+ // code layout too horribly.
+ PlaceSplitBlockCarefully(NewBB, OuterLoopPreds, L);
+
+ // Create the new outer loop.
+ Loop *NewOuter = new Loop();
+
+ // Change the parent loop to use the outer loop as its child now.
if (Loop *Parent = L->getParentLoop())
- Parent->addBasicBlockToLoop(NewBB, getAnalysis<LoopInfo>());
+ Parent->replaceChildLoopWith(L, NewOuter);
+ 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);
- // Replace any instances of Exit with NewBB in this and any nested loops...
- for (df_iterator<Loop*> I = df_begin(L), E = df_end(L); I != E; ++I)
- if (I->hasExitBlock(Exit))
- I->changeExitBlock(Exit, NewBB); // Update exit block information
+ for (unsigned i = 0, e = L->getBlocks().size(); i != e; ++i)
+ NewOuter->addBlockEntry(L->getBlocks()[i]);
- // Update dominator information (set, immdom, domtree, and domfrontier)
- UpdateDomInfoForRevectoredPreds(NewBB, LoopBlocks);
+ // Determine which blocks should stay in L and which should be moved out to
+ // the Outer loop now.
+ std::set<BasicBlock*> BlocksInL;
+ for (pred_iterator PI = pred_begin(Header), E = pred_end(Header); PI!=E; ++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.
+ 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(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.
+ for (unsigned i = 0; i != L->getBlocks().size(); ++i) {
+ BasicBlock *BB = L->getBlocks()[i];
+ if (!BlocksInL.count(BB)) {
+ // Move this block to the parent, updating the exit blocks sets
+ L->removeBlockFromLoop(BB);
+ if ((*LI)[BB] == L)
+ LI->changeLoopFor(BB, NewOuter);
+ --i;
+ }
+ }
+
+ return NewOuter;
}
+
+
/// InsertUniqueBackedgeBlock - This method is called when the specified loop
/// has more than one backedge in it. If this occurs, revector all of these
/// backedges to target a new basic block and have that block branch to the loop
// Create and insert the new backedge block...
BasicBlock *BEBlock = new BasicBlock(Header->getName()+".backedge", F);
- Instruction *BETerminator = new BranchInst(Header);
- BEBlock->getInstList().push_back(BETerminator);
+ BranchInst *BETerminator = new BranchInst(Header, BEBlock);
// Move the new backedge block to right after the last backedge block.
Function::iterator InsertPos = BackedgeBlocks.back(); ++InsertPos;
F->getBasicBlockList().splice(InsertPos, F->getBasicBlockList(), BEBlock);
-
+
// Now that the block has been inserted into the function, create PHI nodes in
// the backedge block which correspond to any PHI nodes in the header block.
- for (BasicBlock::iterator I = Header->begin();
- PHINode *PN = dyn_cast<PHINode>(I); ++I) {
+ 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);
- NewPN->op_reserve(2*BackedgeBlocks.size());
+ NewPN->reserveOperandSpace(BackedgeBlocks.size());
+ if (AA) AA->copyValue(PN, NewPN);
// Loop over the PHI node, moving all entries except the one for the
// preheader over to the new PHI node.
}
}
}
-
+
// Delete all of the incoming values from the old PN except the preheader's
assert(PreheaderIdx != ~0U && "PHI has no preheader entry??");
if (PreheaderIdx != 0) {
PN->setIncomingValue(0, PN->getIncomingValue(PreheaderIdx));
PN->setIncomingBlock(0, PN->getIncomingBlock(PreheaderIdx));
}
- PN->op_erase(PN->op_begin()+2, PN->op_end());
+ // Nuke all entries except the zero'th.
+ for (unsigned i = 0, e = PN->getNumIncomingValues()-1; i != e; ++i)
+ PN->removeIncomingValue(e-i, false);
// Finally, add the newly constructed PHI node as the entry for the BEBlock.
PN->addIncoming(NewPN, BEBlock);
// eliminate the PHI Node.
if (HasUniqueIncomingValue) {
NewPN->replaceAllUsesWith(UniqueValue);
+ if (AA) AA->deleteValue(NewPN);
BEBlock->getInstList().erase(NewPN);
}
}
// Update Loop Information - we know that this block is now in the current
// loop and all parent loops.
- L->addBasicBlockToLoop(BEBlock, getAnalysis<LoopInfo>());
+ L->addBasicBlockToLoop(BEBlock, *LI);
- // Replace any instances of Exit with NewBB in this and any nested loops...
- for (df_iterator<Loop*> I = df_begin(L), E = df_end(L); I != E; ++I)
- if (I->hasExitBlock(Header))
- I->changeExitBlock(Header, BEBlock); // Update exit block information
-
- // Update dominator information (set, immdom, domtree, and domfrontier)
- UpdateDomInfoForRevectoredPreds(BEBlock, BackedgeBlocks);
+ // Update dominator information
+ DT->splitBlock(BEBlock);
+ if (DominanceFrontier *DF = getAnalysisToUpdate<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 "Exit"), 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 Exit, and moves some predecessors of "Exit" 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(succ_begin(NewBB) != succ_end(NewBB) &&
- ++succ_begin(NewBB) == succ_end(NewBB) &&
- "NewBB should have a single successor!");
- DominatorSet &DS = getAnalysis<DominatorSet>();
-
- // Update dominator information... The blocks that dominate NewBB are the
- // intersection of the dominators of predecessors, plus the block itself.
- // The newly created basic block does not dominate anything except itself.
- //
- DominatorSet::DomSetType NewBBDomSet = DS.getDominators(PredBlocks[0]);
- for (unsigned i = 1, e = PredBlocks.size(); i != e; ++i)
- set_intersect(NewBBDomSet, DS.getDominators(PredBlocks[i]));
- NewBBDomSet.insert(NewBB); // All blocks dominate themselves...
- DS.addBasicBlock(NewBB, NewBBDomSet);
-
- // Update immediate dominator information if we have it...
- BasicBlock *NewBBIDom = 0;
- if (ImmediateDominators *ID = getAnalysisToUpdate<ImmediateDominators>()) {
- // This block does not strictly dominate anything, so it is not an immediate
- // dominator. 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...
- 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...
- }
-
- // Update DominatorTree information if it is active.
- if (DominatorTree *DT = getAnalysisToUpdate<DominatorTree>()) {
- // NewBB doesn't dominate anything, so just create a node and link it into
- // its immediate dominator. If we don't have ImmediateDominator info
- // around, calculate the idom as above.
- DominatorTree::Node *NewBBIDomNode;
- if (NewBBIDom) {
- NewBBIDomNode = DT->getNode(NewBBIDom);
- } else {
- NewBBIDomNode = DT->getNode(PredBlocks[0]); // Random pred
- while (!NewBBDomSet.count(NewBBIDomNode->getBlock())) {
- NewBBIDomNode = NewBBIDomNode->getIDom();
- assert(NewBBIDomNode && "No shared dominator found??");
- }
- }
-
- // Create the new dominator tree node...
- DT->createNewNode(NewBB, NewBBIDomNode);
- }
-
- // Update dominance frontier information...
- if (DominanceFrontier *DF = getAnalysisToUpdate<DominanceFrontier>()) {
- // DF(NewBB) is {Exit} because NewBB does not strictly dominate Exit, but it
- // does dominate itself (and there is an edge (NewBB -> Exit)). Exit is the
- // single successor of NewBB.
- DominanceFrontier::DomSetType NewDFSet;
- BasicBlock *Exit = *succ_begin(NewBB);
- NewDFSet.insert(Exit);
- DF->addBasicBlock(NewBB, NewDFSet);
-
- // Now we must loop over all of the dominance frontiers in the function,
- // replacing occurrences of Exit with NewBB in some cases. All blocks that
- // dominate a block in PredBlocks and contained Exit in their dominance
- // frontier must be updated to contain NewBB instead. This only occurs if
- // there is more than one block in PredBlocks.
- //
- if (PredBlocks.size() > 1) {
- 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 Exit node is in DF(PredDom), then PredDom didn't dominate
- // Exit but did dominate a predecessor of it. Now we change this
- // entry to include NewBB in the DF instead of Exit.
- DominanceFrontier::iterator DFI = DF->find(PredDom);
- assert(DFI != DF->end() && "No dominance frontier for node?");
- if (DFI->second.count(Exit)) {
- DF->removeFromFrontier(DFI, Exit);
- DF->addToFrontier(DFI, NewBB);
- }
- }
- }
- }
- }
-}
-} // End llvm namespace
projects / oota-llvm.git / blobdiff