//===- LoopRotation.cpp - Loop Rotation Pass ------------------------------===// // // The LLVM Compiler Infrastructure // // This file is distributed under the University of Illinois Open Source // License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// // // This file implements Loop Rotation Pass. // //===----------------------------------------------------------------------===// #include "llvm/Transforms/Scalar.h" #include "llvm/ADT/Statistic.h" #include "llvm/Analysis/CodeMetrics.h" #include "llvm/Analysis/InstructionSimplify.h" #include "llvm/Analysis/LoopPass.h" #include "llvm/Analysis/ScalarEvolution.h" #include "llvm/Analysis/TargetTransformInfo.h" #include "llvm/Analysis/ValueTracking.h" #include "llvm/IR/CFG.h" #include "llvm/IR/Dominators.h" #include "llvm/IR/Function.h" #include "llvm/IR/IntrinsicInst.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Debug.h" #include "llvm/Transforms/Utils/BasicBlockUtils.h" #include "llvm/Transforms/Utils/Local.h" #include "llvm/Transforms/Utils/SSAUpdater.h" #include "llvm/Transforms/Utils/ValueMapper.h" using namespace llvm; #define DEBUG_TYPE "loop-rotate" static cl::opt<unsigned> DefaultRotationThreshold("rotation-max-header-size", cl::init(16), cl::Hidden, cl::desc("The default maximum header size for automatic loop rotation")); STATISTIC(NumRotated, "Number of loops rotated"); namespace { class LoopRotate : public LoopPass { public: static char ID; // Pass ID, replacement for typeid LoopRotate(int SpecifiedMaxHeaderSize = -1) : LoopPass(ID) { initializeLoopRotatePass(*PassRegistry::getPassRegistry()); if (SpecifiedMaxHeaderSize == -1) MaxHeaderSize = DefaultRotationThreshold; else MaxHeaderSize = unsigned(SpecifiedMaxHeaderSize); } // LCSSA form makes instruction renaming easier. void getAnalysisUsage(AnalysisUsage &AU) const override { AU.addPreserved<DominatorTreeWrapperPass>(); AU.addRequired<LoopInfo>(); AU.addPreserved<LoopInfo>(); AU.addRequiredID(LoopSimplifyID); AU.addPreservedID(LoopSimplifyID); AU.addRequiredID(LCSSAID); AU.addPreservedID(LCSSAID); AU.addPreserved<ScalarEvolution>(); AU.addRequired<TargetTransformInfo>(); } bool runOnLoop(Loop *L, LPPassManager &LPM) override; bool simplifyLoopLatch(Loop *L); bool rotateLoop(Loop *L, bool SimplifiedLatch); private: unsigned MaxHeaderSize; LoopInfo *LI; const TargetTransformInfo *TTI; }; } char LoopRotate::ID = 0; INITIALIZE_PASS_BEGIN(LoopRotate, "loop-rotate", "Rotate Loops", false, false) INITIALIZE_AG_DEPENDENCY(TargetTransformInfo) INITIALIZE_PASS_DEPENDENCY(LoopInfo) INITIALIZE_PASS_DEPENDENCY(LoopSimplify) INITIALIZE_PASS_DEPENDENCY(LCSSA) INITIALIZE_PASS_END(LoopRotate, "loop-rotate", "Rotate Loops", false, false) Pass *llvm::createLoopRotatePass(int MaxHeaderSize) { return new LoopRotate(MaxHeaderSize); } /// Rotate Loop L as many times as possible. Return true if /// the loop is rotated at least once. bool LoopRotate::runOnLoop(Loop *L, LPPassManager &LPM) { if (skipOptnoneFunction(L)) return false; // Save the loop metadata. MDNode *LoopMD = L->getLoopID(); LI = &getAnalysis<LoopInfo>(); TTI = &getAnalysis<TargetTransformInfo>(); // Simplify the loop latch before attempting to rotate the header // upward. Rotation may not be needed if the loop tail can be folded into the // loop exit. bool SimplifiedLatch = simplifyLoopLatch(L); // One loop can be rotated multiple times. bool MadeChange = false; while (rotateLoop(L, SimplifiedLatch)) { MadeChange = true; SimplifiedLatch = false; } // Restore the loop metadata. // NB! We presume LoopRotation DOESN'T ADD its own metadata. if ((MadeChange || SimplifiedLatch) && LoopMD) L->setLoopID(LoopMD); return MadeChange; } /// RewriteUsesOfClonedInstructions - We just cloned the instructions from the /// old header into the preheader. If there were uses of the values produced by /// these instruction that were outside of the loop, we have to insert PHI nodes /// to merge the two values. Do this now. static void RewriteUsesOfClonedInstructions(BasicBlock *OrigHeader, BasicBlock *OrigPreheader, ValueToValueMapTy &ValueMap) { // Remove PHI node entries that are no longer live. BasicBlock::iterator I, E = OrigHeader->end(); for (I = OrigHeader->begin(); PHINode *PN = dyn_cast<PHINode>(I); ++I) PN->removeIncomingValue(PN->getBasicBlockIndex(OrigPreheader)); // Now fix up users of the instructions in OrigHeader, inserting PHI nodes // as necessary. SSAUpdater SSA; for (I = OrigHeader->begin(); I != E; ++I) { Value *OrigHeaderVal = I; // If there are no uses of the value (e.g. because it returns void), there // is nothing to rewrite. if (OrigHeaderVal->use_empty()) continue; Value *OrigPreHeaderVal = ValueMap[OrigHeaderVal]; // The value now exits in two versions: the initial value in the preheader // and the loop "next" value in the original header. SSA.Initialize(OrigHeaderVal->getType(), OrigHeaderVal->getName()); SSA.AddAvailableValue(OrigHeader, OrigHeaderVal); SSA.AddAvailableValue(OrigPreheader, OrigPreHeaderVal); // Visit each use of the OrigHeader instruction. for (Value::use_iterator UI = OrigHeaderVal->use_begin(), UE = OrigHeaderVal->use_end(); UI != UE; ) { // Grab the use before incrementing the iterator. Use &U = *UI; // Increment the iterator before removing the use from the list. ++UI; // SSAUpdater can't handle a non-PHI use in the same block as an // earlier def. We can easily handle those cases manually. Instruction *UserInst = cast<Instruction>(U.getUser()); if (!isa<PHINode>(UserInst)) { BasicBlock *UserBB = UserInst->getParent(); // The original users in the OrigHeader are already using the // original definitions. if (UserBB == OrigHeader) continue; // Users in the OrigPreHeader need to use the value to which the // original definitions are mapped. if (UserBB == OrigPreheader) { U = OrigPreHeaderVal; continue; } } // Anything else can be handled by SSAUpdater. SSA.RewriteUse(U); } } } /// Determine whether the instructions in this range my be safely and cheaply /// speculated. This is not an important enough situation to develop complex /// heuristics. We handle a single arithmetic instruction along with any type /// conversions. static bool shouldSpeculateInstrs(BasicBlock::iterator Begin, BasicBlock::iterator End) { bool seenIncrement = false; for (BasicBlock::iterator I = Begin; I != End; ++I) { if (!isSafeToSpeculativelyExecute(I)) return false; if (isa<DbgInfoIntrinsic>(I)) continue; switch (I->getOpcode()) { default: return false; case Instruction::GetElementPtr: // GEPs are cheap if all indices are constant. if (!cast<GEPOperator>(I)->hasAllConstantIndices()) return false; // fall-thru to increment case case Instruction::Add: case Instruction::Sub: case Instruction::And: case Instruction::Or: case Instruction::Xor: case Instruction::Shl: case Instruction::LShr: case Instruction::AShr: if (seenIncrement) return false; seenIncrement = true; break; case Instruction::Trunc: case Instruction::ZExt: case Instruction::SExt: // ignore type conversions break; } } return true; } /// Fold the loop tail into the loop exit by speculating the loop tail /// instructions. Typically, this is a single post-increment. In the case of a /// simple 2-block loop, hoisting the increment can be much better than /// duplicating the entire loop header. In the cast of loops with early exits, /// rotation will not work anyway, but simplifyLoopLatch will put the loop in /// canonical form so downstream passes can handle it. /// /// I don't believe this invalidates SCEV. bool LoopRotate::simplifyLoopLatch(Loop *L) { BasicBlock *Latch = L->getLoopLatch(); if (!Latch || Latch->hasAddressTaken()) return false; BranchInst *Jmp = dyn_cast<BranchInst>(Latch->getTerminator()); if (!Jmp || !Jmp->isUnconditional()) return false; BasicBlock *LastExit = Latch->getSinglePredecessor(); if (!LastExit || !L->isLoopExiting(LastExit)) return false; BranchInst *BI = dyn_cast<BranchInst>(LastExit->getTerminator()); if (!BI) return false; if (!shouldSpeculateInstrs(Latch->begin(), Jmp)) return false; DEBUG(dbgs() << "Folding loop latch " << Latch->getName() << " into " << LastExit->getName() << "\n"); // Hoist the instructions from Latch into LastExit. LastExit->getInstList().splice(BI, Latch->getInstList(), Latch->begin(), Jmp); unsigned FallThruPath = BI->getSuccessor(0) == Latch ? 0 : 1; BasicBlock *Header = Jmp->getSuccessor(0); assert(Header == L->getHeader() && "expected a backward branch"); // Remove Latch from the CFG so that LastExit becomes the new Latch. BI->setSuccessor(FallThruPath, Header); Latch->replaceSuccessorsPhiUsesWith(LastExit); Jmp->eraseFromParent(); // Nuke the Latch block. assert(Latch->empty() && "unable to evacuate Latch"); LI->removeBlock(Latch); if (DominatorTreeWrapperPass *DTWP = getAnalysisIfAvailable<DominatorTreeWrapperPass>()) DTWP->getDomTree().eraseNode(Latch); Latch->eraseFromParent(); return true; } /// Rotate loop LP. Return true if the loop is rotated. /// /// \param SimplifiedLatch is true if the latch was just folded into the final /// loop exit. In this case we may want to rotate even though the new latch is /// now an exiting branch. This rotation would have happened had the latch not /// been simplified. However, if SimplifiedLatch is false, then we avoid /// rotating loops in which the latch exits to avoid excessive or endless /// rotation. LoopRotate should be repeatable and converge to a canonical /// form. This property is satisfied because simplifying the loop latch can only /// happen once across multiple invocations of the LoopRotate pass. bool LoopRotate::rotateLoop(Loop *L, bool SimplifiedLatch) { // If the loop has only one block then there is not much to rotate. if (L->getBlocks().size() == 1) return false; BasicBlock *OrigHeader = L->getHeader(); BasicBlock *OrigLatch = L->getLoopLatch(); BranchInst *BI = dyn_cast<BranchInst>(OrigHeader->getTerminator()); if (!BI || BI->isUnconditional()) return false; // If the loop header is not one of the loop exiting blocks then // either this loop is already rotated or it is not // suitable for loop rotation transformations. if (!L->isLoopExiting(OrigHeader)) return false; // If the loop latch already contains a branch that leaves the loop then the // loop is already rotated. if (!OrigLatch) return false; // Rotate if either the loop latch does *not* exit the loop, or if the loop // latch was just simplified. if (L->isLoopExiting(OrigLatch) && !SimplifiedLatch) return false; // Check size of original header and reject loop if it is very big or we can't // duplicate blocks inside it. { CodeMetrics Metrics; Metrics.analyzeBasicBlock(OrigHeader, *TTI); if (Metrics.notDuplicatable) { DEBUG(dbgs() << "LoopRotation: NOT rotating - contains non-duplicatable" << " instructions: "; L->dump()); return false; } if (Metrics.NumInsts > MaxHeaderSize) return false; } // Now, this loop is suitable for rotation. BasicBlock *OrigPreheader = L->getLoopPreheader(); // If the loop could not be converted to canonical form, it must have an // indirectbr in it, just give up. if (!OrigPreheader) return false; // Anything ScalarEvolution may know about this loop or the PHI nodes // in its header will soon be invalidated. if (ScalarEvolution *SE = getAnalysisIfAvailable<ScalarEvolution>()) SE->forgetLoop(L); DEBUG(dbgs() << "LoopRotation: rotating "; L->dump()); // Find new Loop header. NewHeader is a Header's one and only successor // that is inside loop. Header's other successor is outside the // loop. Otherwise loop is not suitable for rotation. BasicBlock *Exit = BI->getSuccessor(0); BasicBlock *NewHeader = BI->getSuccessor(1); if (L->contains(Exit)) std::swap(Exit, NewHeader); assert(NewHeader && "Unable to determine new loop header"); assert(L->contains(NewHeader) && !L->contains(Exit) && "Unable to determine loop header and exit blocks"); // This code assumes that the new header has exactly one predecessor. // Remove any single-entry PHI nodes in it. assert(NewHeader->getSinglePredecessor() && "New header doesn't have one pred!"); FoldSingleEntryPHINodes(NewHeader); // Begin by walking OrigHeader and populating ValueMap with an entry for // each Instruction. BasicBlock::iterator I = OrigHeader->begin(), E = OrigHeader->end(); ValueToValueMapTy ValueMap; // For PHI nodes, the value available in OldPreHeader is just the // incoming value from OldPreHeader. for (; PHINode *PN = dyn_cast<PHINode>(I); ++I) ValueMap[PN] = PN->getIncomingValueForBlock(OrigPreheader); // For the rest of the instructions, either hoist to the OrigPreheader if // possible or create a clone in the OldPreHeader if not. TerminatorInst *LoopEntryBranch = OrigPreheader->getTerminator(); while (I != E) { Instruction *Inst = I++; // If the instruction's operands are invariant and it doesn't read or write // memory, then it is safe to hoist. Doing this doesn't change the order of // execution in the preheader, but does prevent the instruction from // executing in each iteration of the loop. This means it is safe to hoist // something that might trap, but isn't safe to hoist something that reads // memory (without proving that the loop doesn't write). if (L->hasLoopInvariantOperands(Inst) && !Inst->mayReadFromMemory() && !Inst->mayWriteToMemory() && !isa<TerminatorInst>(Inst) && !isa<DbgInfoIntrinsic>(Inst) && !isa<AllocaInst>(Inst)) { Inst->moveBefore(LoopEntryBranch); continue; } // Otherwise, create a duplicate of the instruction. Instruction *C = Inst->clone(); // Eagerly remap the operands of the instruction. RemapInstruction(C, ValueMap, RF_NoModuleLevelChanges|RF_IgnoreMissingEntries); // With the operands remapped, see if the instruction constant folds or is // otherwise simplifyable. This commonly occurs because the entry from PHI // nodes allows icmps and other instructions to fold. Value *V = SimplifyInstruction(C); if (V && LI->replacementPreservesLCSSAForm(C, V)) { // If so, then delete the temporary instruction and stick the folded value // in the map. delete C; ValueMap[Inst] = V; } else { // Otherwise, stick the new instruction into the new block! C->setName(Inst->getName()); C->insertBefore(LoopEntryBranch); ValueMap[Inst] = C; } } // Along with all the other instructions, we just cloned OrigHeader's // terminator into OrigPreHeader. Fix up the PHI nodes in each of OrigHeader's // successors by duplicating their incoming values for OrigHeader. TerminatorInst *TI = OrigHeader->getTerminator(); for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i) for (BasicBlock::iterator BI = TI->getSuccessor(i)->begin(); PHINode *PN = dyn_cast<PHINode>(BI); ++BI) PN->addIncoming(PN->getIncomingValueForBlock(OrigHeader), OrigPreheader); // Now that OrigPreHeader has a clone of OrigHeader's terminator, remove // OrigPreHeader's old terminator (the original branch into the loop), and // remove the corresponding incoming values from the PHI nodes in OrigHeader. LoopEntryBranch->eraseFromParent(); // If there were any uses of instructions in the duplicated block outside the // loop, update them, inserting PHI nodes as required RewriteUsesOfClonedInstructions(OrigHeader, OrigPreheader, ValueMap); // NewHeader is now the header of the loop. L->moveToHeader(NewHeader); assert(L->getHeader() == NewHeader && "Latch block is our new header"); // At this point, we've finished our major CFG changes. As part of cloning // the loop into the preheader we've simplified instructions and the // duplicated conditional branch may now be branching on a constant. If it is // branching on a constant and if that constant means that we enter the loop, // then we fold away the cond branch to an uncond branch. This simplifies the // loop in cases important for nested loops, and it also means we don't have // to split as many edges. BranchInst *PHBI = cast<BranchInst>(OrigPreheader->getTerminator()); assert(PHBI->isConditional() && "Should be clone of BI condbr!"); if (!isa<ConstantInt>(PHBI->getCondition()) || PHBI->getSuccessor(cast<ConstantInt>(PHBI->getCondition())->isZero()) != NewHeader) { // The conditional branch can't be folded, handle the general case. // Update DominatorTree to reflect the CFG change we just made. Then split // edges as necessary to preserve LoopSimplify form. if (DominatorTreeWrapperPass *DTWP = getAnalysisIfAvailable<DominatorTreeWrapperPass>()) { DominatorTree &DT = DTWP->getDomTree(); // Everything that was dominated by the old loop header is now dominated // by the original loop preheader. Conceptually the header was merged // into the preheader, even though we reuse the actual block as a new // loop latch. DomTreeNode *OrigHeaderNode = DT.getNode(OrigHeader); SmallVector<DomTreeNode *, 8> HeaderChildren(OrigHeaderNode->begin(), OrigHeaderNode->end()); DomTreeNode *OrigPreheaderNode = DT.getNode(OrigPreheader); for (unsigned I = 0, E = HeaderChildren.size(); I != E; ++I) DT.changeImmediateDominator(HeaderChildren[I], OrigPreheaderNode); assert(DT.getNode(Exit)->getIDom() == OrigPreheaderNode); assert(DT.getNode(NewHeader)->getIDom() == OrigPreheaderNode); // Update OrigHeader to be dominated by the new header block. DT.changeImmediateDominator(OrigHeader, OrigLatch); } // Right now OrigPreHeader has two successors, NewHeader and ExitBlock, and // thus is not a preheader anymore. // Split the edge to form a real preheader. BasicBlock *NewPH = SplitCriticalEdge(OrigPreheader, NewHeader, this); NewPH->setName(NewHeader->getName() + ".lr.ph"); // Preserve canonical loop form, which means that 'Exit' should have only // one predecessor. Note that Exit could be an exit block for multiple // nested loops, causing both of the edges to now be critical and need to // be split. SmallVector<BasicBlock *, 4> ExitPreds(pred_begin(Exit), pred_end(Exit)); bool SplitLatchEdge = false; for (SmallVectorImpl<BasicBlock *>::iterator PI = ExitPreds.begin(), PE = ExitPreds.end(); PI != PE; ++PI) { // We only need to split loop exit edges. Loop *PredLoop = LI->getLoopFor(*PI); if (!PredLoop || PredLoop->contains(Exit)) continue; SplitLatchEdge |= L->getLoopLatch() == *PI; BasicBlock *ExitSplit = SplitCriticalEdge(*PI, Exit, this); ExitSplit->moveBefore(Exit); } assert(SplitLatchEdge && "Despite splitting all preds, failed to split latch exit?"); } else { // We can fold the conditional branch in the preheader, this makes things // simpler. The first step is to remove the extra edge to the Exit block. Exit->removePredecessor(OrigPreheader, true /*preserve LCSSA*/); BranchInst *NewBI = BranchInst::Create(NewHeader, PHBI); NewBI->setDebugLoc(PHBI->getDebugLoc()); PHBI->eraseFromParent(); // With our CFG finalized, update DomTree if it is available. if (DominatorTreeWrapperPass *DTWP = getAnalysisIfAvailable<DominatorTreeWrapperPass>()) { DominatorTree &DT = DTWP->getDomTree(); // Update OrigHeader to be dominated by the new header block. DT.changeImmediateDominator(NewHeader, OrigPreheader); DT.changeImmediateDominator(OrigHeader, OrigLatch); // Brute force incremental dominator tree update. Call // findNearestCommonDominator on all CFG predecessors of each child of the // original header. DomTreeNode *OrigHeaderNode = DT.getNode(OrigHeader); SmallVector<DomTreeNode *, 8> HeaderChildren(OrigHeaderNode->begin(), OrigHeaderNode->end()); bool Changed; do { Changed = false; for (unsigned I = 0, E = HeaderChildren.size(); I != E; ++I) { DomTreeNode *Node = HeaderChildren[I]; BasicBlock *BB = Node->getBlock(); pred_iterator PI = pred_begin(BB); BasicBlock *NearestDom = *PI; for (pred_iterator PE = pred_end(BB); PI != PE; ++PI) NearestDom = DT.findNearestCommonDominator(NearestDom, *PI); // Remember if this changes the DomTree. if (Node->getIDom()->getBlock() != NearestDom) { DT.changeImmediateDominator(BB, NearestDom); Changed = true; } } // If the dominator changed, this may have an effect on other // predecessors, continue until we reach a fixpoint. } while (Changed); } } assert(L->getLoopPreheader() && "Invalid loop preheader after loop rotation"); assert(L->getLoopLatch() && "Invalid loop latch after loop rotation"); // Now that the CFG and DomTree are in a consistent state again, try to merge // the OrigHeader block into OrigLatch. This will succeed if they are // connected by an unconditional branch. This is just a cleanup so the // emitted code isn't too gross in this common case. MergeBlockIntoPredecessor(OrigHeader, this); DEBUG(dbgs() << "LoopRotation: into "; L->dump()); ++NumRotated; return true; }