//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "loop-unroll"
#include "llvm/Transforms/Utils/UnrollLoop.h"
-#include "llvm/BasicBlock.h"
+#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/Statistic.h"
+#include "llvm/Analysis/AssumptionCache.h"
#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Analysis/LoopIterator.h"
#include "llvm/Analysis/LoopPass.h"
#include "llvm/Analysis/ScalarEvolution.h"
+#include "llvm/IR/BasicBlock.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DiagnosticInfo.h"
+#include "llvm/IR/Dominators.h"
+#include "llvm/IR/LLVMContext.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Transforms/Utils/Cloning.h"
#include "llvm/Transforms/Utils/Local.h"
+#include "llvm/Transforms/Utils/LoopUtils.h"
#include "llvm/Transforms/Utils/SimplifyIndVar.h"
using namespace llvm;
+#define DEBUG_TYPE "loop-unroll"
+
// TODO: Should these be here or in LoopUnroll?
STATISTIC(NumCompletelyUnrolled, "Number of loops completely unrolled");
STATISTIC(NumUnrolled, "Number of loops unrolled (completely or otherwise)");
/// FoldBlockIntoPredecessor - Folds a basic block into its predecessor if it
/// only has one predecessor, and that predecessor only has one successor.
-/// The LoopInfo Analysis that is passed will be kept consistent.
-/// Returns the new combined block.
-static BasicBlock *FoldBlockIntoPredecessor(BasicBlock *BB, LoopInfo* LI,
- LPPassManager *LPM) {
+/// The LoopInfo Analysis that is passed will be kept consistent. If folding is
+/// successful references to the containing loop must be removed from
+/// ScalarEvolution by calling ScalarEvolution::forgetLoop because SE may have
+/// references to the eliminated BB. The argument ForgottenLoops contains a set
+/// of loops that have already been forgotten to prevent redundant, expensive
+/// calls to ScalarEvolution::forgetLoop. Returns the new combined block.
+static BasicBlock *
+FoldBlockIntoPredecessor(BasicBlock *BB, LoopInfo* LI, ScalarEvolution *SE,
+ SmallPtrSetImpl<Loop *> &ForgottenLoops) {
// Merge basic blocks into their predecessor if there is only one distinct
// pred, and if there is only one distinct successor of the predecessor, and
// if there are no PHI nodes.
BasicBlock *OnlyPred = BB->getSinglePredecessor();
- if (!OnlyPred) return 0;
+ if (!OnlyPred) return nullptr;
if (OnlyPred->getTerminator()->getNumSuccessors() != 1)
- return 0;
+ return nullptr;
DEBUG(dbgs() << "Merging: " << *BB << "into: " << *OnlyPred);
// Move all definitions in the successor to the predecessor...
OnlyPred->getInstList().splice(OnlyPred->end(), BB->getInstList());
- std::string OldName = BB->getName();
+ // OldName will be valid until erased.
+ StringRef OldName = BB->getName();
// Erase basic block from the function...
// ScalarEvolution holds references to loop exit blocks.
- if (ScalarEvolution *SE = LPM->getAnalysisIfAvailable<ScalarEvolution>()) {
- if (Loop *L = LI->getLoopFor(BB))
- SE->forgetLoop(L);
+ if (SE) {
+ if (Loop *L = LI->getLoopFor(BB)) {
+ if (ForgottenLoops.insert(L).second)
+ SE->forgetLoop(L);
+ }
}
LI->removeBlock(BB);
- BB->eraseFromParent();
// Inherit predecessor's name if it exists...
if (!OldName.empty() && !OnlyPred->hasName())
OnlyPred->setName(OldName);
+ BB->eraseFromParent();
+
return OnlyPred;
}
/// Similarly, TripMultiple divides the number of times that the LatchBlock may
/// execute without exiting the loop.
///
-/// The LoopInfo Analysis that is passed will be kept consistent.
+/// If AllowRuntime is true then UnrollLoop will consider unrolling loops that
+/// have a runtime (i.e. not compile time constant) trip count. Unrolling these
+/// loops require a unroll "prologue" that runs "RuntimeTripCount % Count"
+/// iterations before branching into the unrolled loop. UnrollLoop will not
+/// runtime-unroll the loop if computing RuntimeTripCount will be expensive and
+/// AllowExpensiveTripCount is false.
///
-/// If a LoopPassManager is passed in, and the loop is fully removed, it will be
-/// removed from the LoopPassManager as well. LPM can also be NULL.
+/// The LoopInfo Analysis that is passed will be kept consistent.
///
-/// This utility preserves LoopInfo. If DominatorTree or ScalarEvolution are
-/// available it must also preserve those analyses.
+/// This utility preserves LoopInfo. It will also preserve ScalarEvolution and
+/// DominatorTree if they are non-null.
bool llvm::UnrollLoop(Loop *L, unsigned Count, unsigned TripCount,
- bool AllowRuntime, unsigned TripMultiple,
- LoopInfo *LI, LPPassManager *LPM) {
+ bool AllowRuntime, bool AllowExpensiveTripCount,
+ unsigned TripMultiple, LoopInfo *LI, ScalarEvolution *SE,
+ DominatorTree *DT, AssumptionCache *AC,
+ bool PreserveLCSSA) {
BasicBlock *Preheader = L->getLoopPreheader();
if (!Preheader) {
DEBUG(dbgs() << " Can't unroll; loop preheader-insertion failed.\n");
return false;
}
+ // Loops with indirectbr cannot be cloned.
+ if (!L->isSafeToClone()) {
+ DEBUG(dbgs() << " Can't unroll; Loop body cannot be cloned.\n");
+ return false;
+ }
+
BasicBlock *Header = L->getHeader();
BranchInst *BI = dyn_cast<BranchInst>(LatchBlock->getTerminator());
if (TripCount != 0 && Count > TripCount)
Count = TripCount;
+ // Don't enter the unroll code if there is nothing to do. This way we don't
+ // need to support "partial unrolling by 1".
+ if (TripCount == 0 && Count < 2)
+ return false;
+
assert(Count > 0);
assert(TripMultiple > 0);
assert(TripCount == 0 || TripCount % TripMultiple == 0);
// Are we eliminating the loop control altogether?
bool CompletelyUnroll = Count == TripCount;
+ SmallVector<BasicBlock *, 4> ExitBlocks;
+ L->getExitBlocks(ExitBlocks);
+ Loop *ParentL = L->getParentLoop();
+ bool AllExitsAreInsideParentLoop = !ParentL ||
+ std::all_of(ExitBlocks.begin(), ExitBlocks.end(),
+ [&](BasicBlock *BB) { return ParentL->contains(BB); });
// We assume a run-time trip count if the compiler cannot
// figure out the loop trip count and the unroll-runtime
// flag is specified.
bool RuntimeTripCount = (TripCount == 0 && Count > 0 && AllowRuntime);
- if (RuntimeTripCount && !UnrollRuntimeLoopProlog(L, Count, LI, LPM))
+ if (RuntimeTripCount &&
+ !UnrollRuntimeLoopProlog(L, Count, AllowExpensiveTripCount, LI, SE, DT,
+ PreserveLCSSA))
return false;
// Notify ScalarEvolution that the loop will be substantially changed,
// if not outright eliminated.
- ScalarEvolution *SE = LPM->getAnalysisIfAvailable<ScalarEvolution>();
if (SE)
SE->forgetLoop(L);
(unsigned)GreatestCommonDivisor64(Count, TripMultiple);
}
+ // Report the unrolling decision.
+ DebugLoc LoopLoc = L->getStartLoc();
+ Function *F = Header->getParent();
+ LLVMContext &Ctx = F->getContext();
+
if (CompletelyUnroll) {
DEBUG(dbgs() << "COMPLETELY UNROLLING loop %" << Header->getName()
<< " with trip count " << TripCount << "!\n");
+ emitOptimizationRemark(Ctx, DEBUG_TYPE, *F, LoopLoc,
+ Twine("completely unrolled loop with ") +
+ Twine(TripCount) + " iterations");
} else {
+ auto EmitDiag = [&](const Twine &T) {
+ emitOptimizationRemark(Ctx, DEBUG_TYPE, *F, LoopLoc,
+ "unrolled loop by a factor of " + Twine(Count) +
+ T);
+ };
+
DEBUG(dbgs() << "UNROLLING loop %" << Header->getName()
<< " by " << Count);
if (TripMultiple == 0 || BreakoutTrip != TripMultiple) {
DEBUG(dbgs() << " with a breakout at trip " << BreakoutTrip);
+ EmitDiag(" with a breakout at trip " + Twine(BreakoutTrip));
} else if (TripMultiple != 1) {
DEBUG(dbgs() << " with " << TripMultiple << " trips per branch");
+ EmitDiag(" with " + Twine(TripMultiple) + " trips per branch");
} else if (RuntimeTripCount) {
DEBUG(dbgs() << " with run-time trip count");
+ EmitDiag(" with run-time trip count");
}
DEBUG(dbgs() << "!\n");
}
- std::vector<BasicBlock*> LoopBlocks = L->getBlocks();
-
bool ContinueOnTrue = L->contains(BI->getSuccessor(0));
BasicBlock *LoopExit = BI->getSuccessor(ContinueOnTrue);
for (unsigned It = 1; It != Count; ++It) {
std::vector<BasicBlock*> NewBlocks;
+ SmallDenseMap<const Loop *, Loop *, 4> NewLoops;
+ NewLoops[L] = L;
for (LoopBlocksDFS::RPOIterator BB = BlockBegin; BB != BlockEnd; ++BB) {
ValueToValueMapTy VMap;
BasicBlock *New = CloneBasicBlock(*BB, VMap, "." + Twine(It));
Header->getParent()->getBasicBlockList().push_back(New);
- // Loop over all of the PHI nodes in the block, changing them to use the
- // incoming values from the previous block.
+ // Tell LI about New.
+ if (*BB == Header) {
+ assert(LI->getLoopFor(*BB) == L && "Header should not be in a sub-loop");
+ L->addBasicBlockToLoop(New, *LI);
+ } else {
+ // Figure out which loop New is in.
+ const Loop *OldLoop = LI->getLoopFor(*BB);
+ assert(OldLoop && "Should (at least) be in the loop being unrolled!");
+
+ Loop *&NewLoop = NewLoops[OldLoop];
+ if (!NewLoop) {
+ // Found a new sub-loop.
+ assert(*BB == OldLoop->getHeader() &&
+ "Header should be first in RPO");
+
+ Loop *NewLoopParent = NewLoops.lookup(OldLoop->getParentLoop());
+ assert(NewLoopParent &&
+ "Expected parent loop before sub-loop in RPO");
+ NewLoop = new Loop;
+ NewLoopParent->addChildLoop(NewLoop);
+
+ // Forget the old loop, since its inputs may have changed.
+ if (SE)
+ SE->forgetLoop(OldLoop);
+ }
+ NewLoop->addBasicBlockToLoop(New, *LI);
+ }
+
if (*BB == Header)
+ // Loop over all of the PHI nodes in the block, changing them to use
+ // the incoming values from the previous block.
for (unsigned i = 0, e = OrigPHINode.size(); i != e; ++i) {
PHINode *NewPHI = cast<PHINode>(VMap[OrigPHINode[i]]);
Value *InVal = NewPHI->getIncomingValueForBlock(LatchBlock);
VI != VE; ++VI)
LastValueMap[VI->first] = VI->second;
- L->addBasicBlockToLoop(New, LI->getBase());
-
// Add phi entries for newly created values to all exit blocks.
for (succ_iterator SI = succ_begin(*BB), SE = succ_end(*BB);
SI != SE; ++SI) {
for (unsigned i = 0; i < NewBlocks.size(); ++i)
for (BasicBlock::iterator I = NewBlocks[i]->begin(),
E = NewBlocks[i]->end(); I != E; ++I)
- ::RemapInstruction(I, LastValueMap);
+ ::RemapInstruction(&*I, LastValueMap);
}
// Loop over the PHI nodes in the original block, setting incoming values.
// For a complete unroll, make the last iteration end with a branch
// to the exit block.
- if (CompletelyUnroll && j == 0) {
- Dest = LoopExit;
+ if (CompletelyUnroll) {
+ if (j == 0)
+ Dest = LoopExit;
NeedConditional = false;
}
}
// Merge adjacent basic blocks, if possible.
+ SmallPtrSet<Loop *, 4> ForgottenLoops;
for (unsigned i = 0, e = Latches.size(); i != e; ++i) {
BranchInst *Term = cast<BranchInst>(Latches[i]->getTerminator());
if (Term->isUnconditional()) {
BasicBlock *Dest = Term->getSuccessor(0);
- if (BasicBlock *Fold = FoldBlockIntoPredecessor(Dest, LI, LPM))
+ if (BasicBlock *Fold = FoldBlockIntoPredecessor(Dest, LI, SE,
+ ForgottenLoops))
std::replace(Latches.begin(), Latches.end(), Dest, Fold);
}
}
+ // FIXME: We could register any cloned assumptions instead of clearing the
+ // whole function's cache.
+ AC->clear();
+
// FIXME: Reconstruct dom info, because it is not preserved properly.
// Incrementally updating domtree after loop unrolling would be easy.
- if (DominatorTree *DT = LPM->getAnalysisIfAvailable<DominatorTree>())
- DT->runOnFunction(*L->getHeader()->getParent());
+ if (DT)
+ DT->recalculate(*L->getHeader()->getParent());
// Simplify any new induction variables in the partially unrolled loop.
if (SE && !CompletelyUnroll) {
SmallVector<WeakVH, 16> DeadInsts;
- simplifyLoopIVs(L, SE, LPM, DeadInsts);
+ simplifyLoopIVs(L, SE, DT, LI, DeadInsts);
// Aggressively clean up dead instructions that simplifyLoopIVs already
// identified. Any remaining should be cleaned up below.
while (!DeadInsts.empty())
if (Instruction *Inst =
- dyn_cast_or_null<Instruction>(&*DeadInsts.pop_back_val()))
+ dyn_cast_or_null<Instruction>(&*DeadInsts.pop_back_val()))
RecursivelyDeleteTriviallyDeadInstructions(Inst);
}
// At this point, the code is well formed. We now do a quick sweep over the
// inserted code, doing constant propagation and dead code elimination as we
// go.
+ const DataLayout &DL = Header->getModule()->getDataLayout();
const std::vector<BasicBlock*> &NewLoopBlocks = L->getBlocks();
for (std::vector<BasicBlock*>::const_iterator BB = NewLoopBlocks.begin(),
BBE = NewLoopBlocks.end(); BB != BBE; ++BB)
for (BasicBlock::iterator I = (*BB)->begin(), E = (*BB)->end(); I != E; ) {
- Instruction *Inst = I++;
+ Instruction *Inst = &*I++;
if (isInstructionTriviallyDead(Inst))
(*BB)->getInstList().erase(Inst);
- else if (Value *V = SimplifyInstruction(Inst))
+ else if (Value *V = SimplifyInstruction(Inst, DL))
if (LI->replacementPreservesLCSSAForm(Inst, V)) {
Inst->replaceAllUsesWith(V);
(*BB)->getInstList().erase(Inst);
NumCompletelyUnrolled += CompletelyUnroll;
++NumUnrolled;
- // Remove the loop from the LoopPassManager if it's completely removed.
- if (CompletelyUnroll && LPM != NULL)
- LPM->deleteLoopFromQueue(L);
+
+ Loop *OuterL = L->getParentLoop();
+ // Update LoopInfo if the loop is completely removed.
+ if (CompletelyUnroll)
+ LI->updateUnloop(L);;
+
+ // If we have a pass and a DominatorTree we should re-simplify impacted loops
+ // to ensure subsequent analyses can rely on this form. We want to simplify
+ // at least one layer outside of the loop that was unrolled so that any
+ // changes to the parent loop exposed by the unrolling are considered.
+ if (DT) {
+ if (!OuterL && !CompletelyUnroll)
+ OuterL = L;
+ if (OuterL) {
+ bool Simplified = simplifyLoop(OuterL, DT, LI, SE, AC, PreserveLCSSA);
+
+ // LCSSA must be performed on the outermost affected loop. The unrolled
+ // loop's last loop latch is guaranteed to be in the outermost loop after
+ // LoopInfo's been updated by updateUnloop.
+ Loop *LatchLoop = LI->getLoopFor(Latches.back());
+ if (!OuterL->contains(LatchLoop))
+ while (OuterL->getParentLoop() != LatchLoop)
+ OuterL = OuterL->getParentLoop();
+
+ if (CompletelyUnroll && (!AllExitsAreInsideParentLoop || Simplified))
+ formLCSSARecursively(*OuterL, *DT, LI, SE);
+ else
+ assert(OuterL->isLCSSAForm(*DT) &&
+ "Loops should be in LCSSA form after loop-unroll.");
+ }
+ }
return true;
}
+
+/// Given an llvm.loop loop id metadata node, returns the loop hint metadata
+/// node with the given name (for example, "llvm.loop.unroll.count"). If no
+/// such metadata node exists, then nullptr is returned.
+MDNode *llvm::GetUnrollMetadata(MDNode *LoopID, StringRef Name) {
+ // First operand should refer to the loop id itself.
+ assert(LoopID->getNumOperands() > 0 && "requires at least one operand");
+ assert(LoopID->getOperand(0) == LoopID && "invalid loop id");
+
+ for (unsigned i = 1, e = LoopID->getNumOperands(); i < e; ++i) {
+ MDNode *MD = dyn_cast<MDNode>(LoopID->getOperand(i));
+ if (!MD)
+ continue;
+
+ MDString *S = dyn_cast<MDString>(MD->getOperand(0));
+ if (!S)
+ continue;
+
+ if (Name.equals(S->getString()))
+ return MD;
+ }
+ return nullptr;
+}
projects / oota-llvm.git / blobdiff