//
// The LLVM Compiler Infrastructure
//
-// This file was developed by Devang Patel 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.
//
//===----------------------------------------------------------------------===//
//
//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "loop-rotation"
-
+#define DEBUG_TYPE "loop-rotate"
#include "llvm/Transforms/Scalar.h"
#include "llvm/Function.h"
-#include "llvm/Instructions.h"
-#include "llvm/Analysis/LoopInfo.h"
+#include "llvm/IntrinsicInst.h"
+#include "llvm/Analysis/CodeMetrics.h"
#include "llvm/Analysis/LoopPass.h"
+#include "llvm/Analysis/InstructionSimplify.h"
+#include "llvm/Analysis/ScalarEvolution.h"
+#include "llvm/Analysis/ValueTracking.h"
#include "llvm/Transforms/Utils/Local.h"
-#include "llvm/Support/CommandLine.h"
+#include "llvm/Transforms/Utils/BasicBlockUtils.h"
+#include "llvm/Transforms/Utils/SSAUpdater.h"
+#include "llvm/Transforms/Utils/ValueMapper.h"
#include "llvm/Support/Debug.h"
#include "llvm/ADT/Statistic.h"
-#include "llvm/ADT/SmallVector.h"
-#include <map>
-
using namespace llvm;
#define MAX_HEADER_SIZE 16
STATISTIC(NumRotated, "Number of loops rotated");
namespace {
- cl::opt<unsigned>
- RotateThreshold("rotate-threshold", cl::init(200), cl::Hidden,
- cl::desc("The cut-off point for loop rotating"));
-
- class VISIBILITY_HIDDEN InsnReplacementData {
- public:
- InsnReplacementData(Instruction *O, Instruction *P, Instruction *H)
- : Original(O), PreHeader(P), Header(H) {}
+ class LoopRotate : public LoopPass {
public:
- Instruction *Original; // Original instruction
- Instruction *PreHeader; // New pre-header replacement
- Instruction *Header; // New header replacement
- };
-
- class VISIBILITY_HIDDEN LoopRotate : public LoopPass {
+ static char ID; // Pass ID, replacement for typeid
+ LoopRotate() : LoopPass(ID) {
+ initializeLoopRotatePass(*PassRegistry::getPassRegistry());
+ }
- public:
- bool runOnLoop(Loop *L, LPPassManager &LPM);
+ // LCSSA form makes instruction renaming easier.
virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.addPreserved<DominatorTree>();
+ AU.addRequired<LoopInfo>();
+ AU.addPreserved<LoopInfo>();
+ AU.addRequiredID(LoopSimplifyID);
+ AU.addPreservedID(LoopSimplifyID);
AU.addRequiredID(LCSSAID);
AU.addPreservedID(LCSSAID);
- //AU.addRequired<LoopInfo>();
- //AU.addPreserved<LoopInfo>();
+ AU.addPreserved<ScalarEvolution>();
}
- // Helper functions
-
- /// Do actual work
- bool rotateLoop(Loop *L, LPPassManager &LPM);
-
- /// Initialize local data
- void initialize();
-
- /// Make sure all Exit block PHINodes have required incoming values.
- /// If incoming value is constant or defined outside the loop then
- /// PHINode may not have an entry for new pre-header.
- void updateExitBlock();
-
- /// Return true if this instruction is used outside original header.
- bool usedOutsideOriginalHeader(Instruction *In);
-
- /// Find Replacement information for instruction. Return NULL if it is
- /// not available.
- InsnReplacementData *findReplacementData(Instruction *I);
+ bool runOnLoop(Loop *L, LPPassManager &LPM);
+ void simplifyLoopLatch(Loop *L);
+ bool rotateLoop(Loop *L);
private:
-
- Loop *L;
- BasicBlock *OrigHeader;
- BasicBlock *OrigPreHeader;
- BasicBlock *OrigLatch;
- BasicBlock *NewHeader;
- BasicBlock *NewPreHeader;
- BasicBlock *Exit;
-
- SmallVector<InsnReplacementData, MAX_HEADER_SIZE> RD;
+ LoopInfo *LI;
};
-
- RegisterPass<LoopRotate> X ("loop-rotate", "Rotate Loops");
}
-LoopPass *llvm::createLoopRotatePass() { return new LoopRotate(); }
+char LoopRotate::ID = 0;
+INITIALIZE_PASS_BEGIN(LoopRotate, "loop-rotate", "Rotate Loops", false, false)
+INITIALIZE_PASS_DEPENDENCY(LoopInfo)
+INITIALIZE_PASS_DEPENDENCY(LoopSimplify)
+INITIALIZE_PASS_DEPENDENCY(LCSSA)
+INITIALIZE_PASS_END(LoopRotate, "loop-rotate", "Rotate Loops", false, false)
-bool LoopRotate::runOnLoop(Loop *Lp, LPPassManager &LPM) {
-
- bool RotatedOneLoop = false;
- initialize();
+Pass *llvm::createLoopRotatePass() { return new LoopRotate(); }
+
+/// 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) {
+ LI = &getAnalysis<LoopInfo>();
+
+ // 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.
+ simplifyLoopLatch(L);
// One loop can be rotated multiple times.
- while (rotateLoop(Lp,LPM)) {
- RotatedOneLoop = true;
- initialize();
- }
+ bool MadeChange = false;
+ while (rotateLoop(L))
+ MadeChange = true;
- return RotatedOneLoop;
+ return MadeChange;
}
-bool LoopRotate::rotateLoop(Loop *Lp, LPPassManager &LPM) {
+/// 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;
- L = Lp;
- if ( NumRotated >= RotateThreshold)
- return false;
+ Value *OrigPreHeaderVal = ValueMap[OrigHeaderVal];
- OrigHeader = L->getHeader();
- OrigPreHeader = L->getLoopPreheader();
- OrigLatch = L->getLoopLatch();
+ // 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);
- // If loop has only one block then there is not much to rotate.
- if (L->getBlocks().size() <= 1)
- return false;
+ // 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.getUse();
- if (!OrigHeader || !OrigLatch || !OrigPreHeader)
- return false;
+ // Increment the iterator before removing the use from the list.
+ ++UI;
- // If loop header is not one of the loop exit block then
- // either this loop is already rotated or it is not
- // suitable for loop rotation transformations.
- if (!L->isLoopExit(OrigHeader))
- return false;
+ // 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();
- BranchInst *BI = dyn_cast<BranchInst>(OrigHeader->getTerminator());
- if (!BI)
- return false;
+ // The original users in the OrigHeader are already using the
+ // original definitions.
+ if (UserBB == OrigHeader)
+ continue;
- std::vector<BasicBlock *> ExitBlocks;
- L->getExitBlocks(ExitBlocks);
- if (ExitBlocks.size() > 1)
- return false;
+ // Users in the OrigPreHeader need to use the value to which the
+ // original definitions are mapped.
+ if (UserBB == OrigPreheader) {
+ U = OrigPreHeaderVal;
+ continue;
+ }
+ }
- // Find new Loop header. NewHeader is a Header's one and only successor
- // that is inside loop. Header's all other successors are out side the
- // loop. Otherwise loop is not suitable for rotation.
- for (unsigned index = 0; index < BI->getNumSuccessors(); ++index) {
- BasicBlock *S = BI->getSuccessor(index);
- if (L->contains(S)) {
- if (!NewHeader)
- NewHeader = S;
- else
- // Loop Header has two successors inside loop. This loop is
- // not suitable for rotation.
- return false;
- } else {
- if (!Exit)
- Exit = S;
- else
- // Loop has multiple exits.
- return false;
+ // Anything else can be handled by SSAUpdater.
+ SSA.RewriteUse(U);
}
}
- assert (NewHeader && "Unable to determine new loop header");
-
- // Check size of original header and reject
- // loop if it is very big.
- if (OrigHeader->getInstList().size() > MAX_HEADER_SIZE)
- return false;
-
- // Now, this loop is suitable for rotation.
+}
- // Copy Prepare PHI nodes and other instructions from original header
- // into new pre-header. Unlike original header, new pre-header is
- // not a member of loop. New pre-header has only one predecessor,
- // that is original loop pre-header.
- //
- // New loop header is one and only successor of original header that
- // is inside the loop. All other original header successors are outside
- // the loop. Copy PHI Nodes from original header into new loop header.
- // Add second incoming value, from new loop pre-header into these phi
- // nodes. If a value defined in original header is used outside original
- // header then new loop header will need new phi nodes with two incoming
- // values, one definition from original header and second definition is
- // from new loop pre-header (which is a clone of original header definition).
-
- NewPreHeader = new BasicBlock("bb.nph", OrigHeader->getParent(), OrigHeader);
- for (BasicBlock::iterator I = OrigHeader->begin(), E = OrigHeader->end();
- I != E; ++I) {
- Instruction *In = I;
-
- if (PHINode *PN = dyn_cast<PHINode>(I)) {
-
- // Create new PHI node with one value incoming from OrigPreHeader.
- // NewPreHeader has only one predecessor, OrigPreHeader.
- PHINode *NPH = new PHINode(In->getType(), In->getName());
- NPH->addIncoming(PN->getIncomingValueForBlock(OrigPreHeader),
- OrigPreHeader);
- NewPreHeader->getInstList().push_back(NPH);
-
- // Create new PHI node with two incoming values for NewHeader.
- // One incoming value is from OrigLatch (through OrigHeader) and
- // second incoming value is from NewPreHeader.
- PHINode *NH = new PHINode(In->getType(), In->getName());
- NH->addIncoming(PN->getIncomingValueForBlock(OrigLatch), OrigHeader);
- NH->addIncoming(NPH, NewPreHeader);
- NewHeader->getInstList().push_front(NH);
-
- RD.push_back(InsnReplacementData(In, NPH, NH));
- } else {
- // This is not a PHI instruction. Insert its clone into NewPreHeader.
- // If this instruction is using a value from same basic block then
- // update it to use value from cloned instruction.
- Instruction *C = In->clone();
- C->setName(In->getName());
- NewPreHeader->getInstList().push_back(C);
-
- // If this instruction is used outside this basic block then
- // create new PHINode for this instruction.
- Instruction *NewHeaderReplacement = NULL;
- if (usedOutsideOriginalHeader(In)) {
- PHINode *PN = new PHINode(In->getType(), In->getName());
- PN->addIncoming(In, OrigHeader);
- PN->addIncoming(C, NewPreHeader);
- NewHeader->getInstList().push_front(PN);
- NewHeaderReplacement = PN;
- }
- RD.push_back(InsnReplacementData(In, C, NewHeaderReplacement));
- }
- }
+/// 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) {
- // Update new pre-header.
- // Rename values that are defined in original header to reflects values
- // defined in new pre-header.
- for (SmallVector<InsnReplacementData, MAX_HEADER_SIZE>::iterator
- I = RD.begin(), E = RD.end(); I != E; ++I) {
-
- InsnReplacementData IRD = (*I);
- Instruction *In = IRD.Original;
- Instruction *C = IRD.PreHeader;
-
- if (C->getParent() != NewPreHeader)
- continue;
+ if (!isSafeToSpeculativelyExecute(I))
+ return false;
- // PHINodes uses value from pre-header predecessors.
- if (isa<PHINode>(In))
+ if (isa<DbgInfoIntrinsic>(I))
continue;
- for (unsigned opi = 0; opi < In->getNumOperands(); ++opi) {
- if (Instruction *OpPhi = dyn_cast<PHINode>(In->getOperand(opi))) {
- if (InsnReplacementData *D = findReplacementData(OpPhi))
- C->setOperand(opi, D->PreHeader);
- }
- else if (Instruction *OpInsn =
- dyn_cast<Instruction>(In->getOperand(opi))) {
- if (InsnReplacementData *D = findReplacementData(OpInsn))
- C->setOperand(opi, D->PreHeader);
- }
+ 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;
+}
- // Rename uses of original header instructions to reflect their new
- // definitions (either from new pre-header node or from newly created
- // new header PHINodes.
- //
- // Original header instructions are used in
- // 1) Original header:
- //
- // If instruction is used in non-phi instructions then it is using
- // defintion from original heder iteself. Do not replace this use
- // with definition from new header or new pre-header.
- //
- // If instruction is used in phi node then it is an incoming
- // value. Rename its use to reflect new definition from new-preheader
- // or new header.
- //
- // 2) Inside loop but not in original header
- //
- // Replace this use to reflect definition from new header.
- for (SmallVector<InsnReplacementData, MAX_HEADER_SIZE>::iterator
- I = RD.begin(), E = RD.end(); I != E; ++I) {
-
- InsnReplacementData IRD = (*I);
- if (!IRD.Header)
- continue;
-
- Instruction *OldPhi = IRD.Original;
- Instruction *NewPhi = IRD.Header;
-
- // Before replacing uses, collect them first, so that iterator is
- // not invalidated.
- SmallVector<Instruction *, 16> AllUses;
- for (Value::use_iterator UI = OldPhi->use_begin(), UE = OldPhi->use_end();
- UI != UE; ++UI ) {
- Instruction *U = cast<Instruction>(UI);
- AllUses.push_back(U);
- }
-
- for (SmallVector<Instruction *, 16>::iterator UI = AllUses.begin(),
- UE = AllUses.end(); UI != UE; ++UI) {
- Instruction *U = *UI;
- BasicBlock *Parent = U->getParent();
-
- // Used inside original header
- if (Parent == OrigHeader) {
- // Do not rename uses inside original header non-phi instructions.
- if (!isa<PHINode>(U))
- continue;
- PHINode *PU = dyn_cast<PHINode>(U);
- // Do not rename uses inside original header phi nodes, if the
- // incoming value is for new header.
- if (PU->getBasicBlockIndex(NewHeader) != -1
- && PU->getIncomingValueForBlock(NewHeader) == U)
- continue;
+/// 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.
+void LoopRotate::simplifyLoopLatch(Loop *L) {
+ BasicBlock *Latch = L->getLoopLatch();
+ if (!Latch || Latch->hasAddressTaken())
+ return;
+
+ BranchInst *Jmp = dyn_cast<BranchInst>(Latch->getTerminator());
+ if (!Jmp || !Jmp->isUnconditional())
+ return;
+
+ BasicBlock *LastExit = Latch->getSinglePredecessor();
+ if (!LastExit || !L->isLoopExiting(LastExit))
+ return;
+
+ BranchInst *BI = dyn_cast<BranchInst>(LastExit->getTerminator());
+ if (!BI)
+ return;
- U->replaceUsesOfWith(OldPhi, NewPhi);
- continue;
- }
+ if (!shouldSpeculateInstrs(Latch->begin(), Jmp))
+ return;
- // Used inside loop, but not in original header.
- if (L->contains(U->getParent())) {
- if (U != NewPhi )
- U->replaceUsesOfWith(OldPhi, NewPhi);
- continue;
- }
+ DEBUG(dbgs() << "Folding loop latch " << Latch->getName() << " into "
+ << LastExit->getName() << "\n");
- // Used inside Exit Block. Since we are in LCSSA form, U must be PHINode.
- assert ( U->getParent() == Exit && "Need to propagate new PHI into Exit blocks");
- assert (isa<PHINode>(U) && "Use in Exit Block that is not PHINode");
+ // Hoist the instructions from Latch into LastExit.
+ LastExit->getInstList().splice(BI, Latch->getInstList(), Latch->begin(), Jmp);
- PHINode *UPhi = cast<PHINode>(U);
+ unsigned FallThruPath = BI->getSuccessor(0) == Latch ? 0 : 1;
+ BasicBlock *Header = Jmp->getSuccessor(0);
+ assert(Header == L->getHeader() && "expected a backward branch");
- // UPhi already has one incoming argument from original header.
- // Add second incoming argument from new Pre header.
-
- UPhi->addIncoming(IRD.PreHeader, NewPreHeader);
- }
- }
-
- /// Make sure all Exit block PHINodes have required incoming values.
- updateExitBlock();
-
- // Update CFG
-
- // Removing incoming branch from loop preheader to original header.
- // Now original header is inside the loop.
- OrigHeader->removePredecessor(OrigPreHeader);
-
- // Establish NewPreHeader as loop preheader. Add unconditional branch
- // from original loop pre-header to new loop pre-header. Add NewPreHEader
- // in loop nest.
- BranchInst *PH_BI = cast<BranchInst>(OrigPreHeader->getTerminator());
- PH_BI->setSuccessor(0, NewPreHeader);
- LoopInfo &LI = LPM.getAnalysis<LoopInfo>();
- if (Loop *PL = LI.getLoopFor(OrigPreHeader))
- PL->addBasicBlockToLoop(NewPreHeader, LI);
-
- // Make NewHeader as the new header for the loop.
- L->moveToHeader(NewHeader);
+ // Remove Latch from the CFG so that LastExit becomes the new Latch.
+ BI->setSuccessor(FallThruPath, Header);
+ Latch->replaceSuccessorsPhiUsesWith(LastExit);
+ Jmp->eraseFromParent();
- NumRotated++;
- return true;
+ // Nuke the Latch block.
+ assert(Latch->empty() && "unable to evacuate Latch");
+ LI->removeBlock(Latch);
+ if (DominatorTree *DT = getAnalysisIfAvailable<DominatorTree>())
+ DT->eraseNode(Latch);
+ Latch->eraseFromParent();
}
+/// Rotate loop LP. Return true if the loop is rotated.
+bool LoopRotate::rotateLoop(Loop *L) {
+ // If the loop has only one block then there is not much to rotate.
+ if (L->getBlocks().size() == 1)
+ return false;
-/// Make sure all Exit block PHINodes have required incoming values.
-/// If incoming value is constant or defined outside the loop then
-/// PHINode may not have an entry for new pre-header.
-void LoopRotate::updateExitBlock() {
+ BasicBlock *OrigHeader = L->getHeader();
- for (BasicBlock::iterator I = Exit->begin(), E = Exit->end();
- I != E; ++I) {
+ BranchInst *BI = dyn_cast<BranchInst>(OrigHeader->getTerminator());
+ if (BI == 0 || BI->isUnconditional())
+ return false;
- if (!isa<PHINode>(I))
- break;
+ // 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;
- PHINode *PN = dyn_cast<PHINode>(I);
+ // Updating PHInodes in loops with multiple exits adds complexity.
+ // Keep it simple, and restrict loop rotation to loops with one exit only.
+ // In future, lift this restriction and support for multiple exits if
+ // required.
+ SmallVector<BasicBlock*, 8> ExitBlocks;
+ L->getExitBlocks(ExitBlocks);
+ if (ExitBlocks.size() > 1)
+ return false;
- if (PN->getBasicBlockIndex(NewPreHeader) == -1) {
- Value *V = PN->getIncomingValueForBlock(OrigHeader);
- if (isa<Constant>(V))
- PN->addIncoming(V, NewPreHeader);
- else {
- InsnReplacementData *IRD = findReplacementData(cast<Instruction>(V));
- assert (IRD && IRD->PreHeader && "Missing New Preheader Instruction");
- PN->addIncoming(IRD->PreHeader, NewPreHeader);
- }
- }
+ // Check size of original header and reject loop if it is very big.
+ {
+ CodeMetrics Metrics;
+ Metrics.analyzeBasicBlock(OrigHeader);
+ if (Metrics.NumInsts > MAX_HEADER_SIZE)
+ return false;
}
-}
+ // Now, this loop is suitable for rotation.
+ BasicBlock *OrigPreheader = L->getLoopPreheader();
+ BasicBlock *OrigLatch = L->getLoopLatch();
-/// Initialize local data
-void LoopRotate::initialize() {
- L = NULL;
- OrigHeader = NULL;
- OrigPreHeader = NULL;
- NewHeader = NULL;
- NewPreHeader = NULL;
- Exit = NULL;
+ // If the loop could not be converted to canonical form, it must have an
+ // indirectbr in it, just give up.
+ if (OrigPreheader == 0 || OrigLatch == 0)
+ return false;
- RD.clear();
-}
+ // 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);
-/// Return true if this instruction is used outside original header.
-bool LoopRotate::usedOutsideOriginalHeader(Instruction *In) {
+ // 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;
+ }
- for (Value::use_iterator UI = In->use_begin(), UE = In->use_end();
- UI != UE; ++UI) {
- Instruction *U = cast<Instruction>(UI);
- if (U->getParent() != OrigHeader) {
- if (L->contains(U->getParent()))
- return true;
+ // 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;
}
}
- return false;
-}
+ // 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 (DominatorTree *DT = getAnalysisIfAvailable<DominatorTree>()) {
+ // Since OrigPreheader now has the conditional branch to Exit block, it is
+ // the dominator of Exit.
+ DT->changeImmediateDominator(Exit, OrigPreheader);
+ DT->changeImmediateDominator(NewHeader, OrigPreheader);
+
+ // 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.
+ BasicBlock *ExitSplit = SplitCriticalEdge(L->getLoopLatch(), Exit, this);
+ ExitSplit->moveBefore(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 (DominatorTree *DT = getAnalysisIfAvailable<DominatorTree>()) {
+ // Update OrigHeader to be dominated by the new header block.
+ DT->changeImmediateDominator(NewHeader, OrigPreheader);
+ DT->changeImmediateDominator(OrigHeader, OrigLatch);
+ }
+ }
-/// Find Replacement information for instruction. Return NULL if it is
-/// not available.
-InsnReplacementData *LoopRotate::findReplacementData(Instruction *In) {
+ assert(L->getLoopPreheader() && "Invalid loop preheader after loop rotation");
+ assert(L->getLoopLatch() && "Invalid loop latch after loop rotation");
- // Since RD is small, linear walk is OK.
- for (SmallVector<InsnReplacementData, MAX_HEADER_SIZE>::iterator
- I = RD.begin(), E = RD.end(); I != E; ++I)
- if ((*I).Original == In)
- return &(*I);
+ // 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);
- return NULL;
+ ++NumRotated;
+ return true;
}
+