// pointer. There are no calls in the loop which mod/ref the pointer.
// If these conditions are true, we can promote the loads and stores in the
// loop of the pointer to use a temporary alloca'd variable. We then use
-// the mem2reg functionality to construct the appropriate SSA form for the
-// variable.
+// the SSAUpdater to construct the appropriate SSA form for the value.
//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "licm"
#include "llvm/Transforms/Scalar.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/IntrinsicInst.h"
-#include "llvm/Instructions.h"
-#include "llvm/Target/TargetData.h"
-#include "llvm/Analysis/LoopInfo.h"
-#include "llvm/Analysis/LoopPass.h"
+#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/AliasSetTracker.h"
-#include "llvm/Analysis/Dominators.h"
+#include "llvm/Analysis/ConstantFolding.h"
+#include "llvm/Analysis/LoopInfo.h"
+#include "llvm/Analysis/LoopPass.h"
#include "llvm/Analysis/ScalarEvolution.h"
-#include "llvm/Transforms/Utils/PromoteMemToReg.h"
-#include "llvm/Support/CFG.h"
+#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/IR/CFG.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Dominators.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/Metadata.h"
+#include "llvm/IR/PredIteratorCache.h"
#include "llvm/Support/CommandLine.h"
-#include "llvm/Support/raw_ostream.h"
#include "llvm/Support/Debug.h"
-#include "llvm/ADT/Statistic.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetLibraryInfo.h"
+#include "llvm/Transforms/Utils/Local.h"
+#include "llvm/Transforms/Utils/LoopUtils.h"
+#include "llvm/Transforms/Utils/SSAUpdater.h"
#include <algorithm>
using namespace llvm;
+#define DEBUG_TYPE "licm"
+
STATISTIC(NumSunk , "Number of instructions sunk out of loop");
STATISTIC(NumHoisted , "Number of instructions hoisted out of loop");
STATISTIC(NumMovedLoads, "Number of load insts hoisted or sunk");
namespace {
struct LICM : public LoopPass {
static char ID; // Pass identification, replacement for typeid
- LICM() : LoopPass(&ID) {}
+ LICM() : LoopPass(ID) {
+ initializeLICMPass(*PassRegistry::getPassRegistry());
+ }
- virtual bool runOnLoop(Loop *L, LPPassManager &LPM);
+ bool runOnLoop(Loop *L, LPPassManager &LPM) override;
/// This transformation requires natural loop information & requires that
/// loop preheaders be inserted into the CFG...
///
- virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.setPreservesCFG();
- AU.addRequiredID(LoopSimplifyID);
+ AU.addRequired<DominatorTreeWrapperPass>();
AU.addRequired<LoopInfo>();
- AU.addRequired<DominatorTree>();
- AU.addRequired<DominanceFrontier>(); // For scalar promotion (mem2reg)
+ AU.addRequiredID(LoopSimplifyID);
+ AU.addPreservedID(LoopSimplifyID);
+ AU.addRequiredID(LCSSAID);
+ AU.addPreservedID(LCSSAID);
AU.addRequired<AliasAnalysis>();
+ AU.addPreserved<AliasAnalysis>();
AU.addPreserved<ScalarEvolution>();
- AU.addPreserved<DominanceFrontier>();
- AU.addPreservedID(LoopSimplifyID);
+ AU.addRequired<TargetLibraryInfo>();
}
- bool doFinalization() {
- // Free the values stored in the map
- for (std::map<Loop *, AliasSetTracker *>::iterator
- I = LoopToAliasMap.begin(), E = LoopToAliasMap.end(); I != E; ++I)
- delete I->second;
+ using llvm::Pass::doFinalization;
- LoopToAliasMap.clear();
+ bool doFinalization() override {
+ assert(LoopToAliasSetMap.empty() && "Didn't free loop alias sets");
return false;
}
private:
- // Various analyses that we use...
AliasAnalysis *AA; // Current AliasAnalysis information
LoopInfo *LI; // Current LoopInfo
- DominatorTree *DT; // Dominator Tree for the current Loop...
- DominanceFrontier *DF; // Current Dominance Frontier
+ DominatorTree *DT; // Dominator Tree for the current Loop.
+
+ const DataLayout *DL; // DataLayout for constant folding.
+ TargetLibraryInfo *TLI; // TargetLibraryInfo for constant folding.
- // State that is updated as we process loops
+ // State that is updated as we process loops.
bool Changed; // Set to true when we change anything.
BasicBlock *Preheader; // The preheader block of the current loop...
Loop *CurLoop; // The current loop we are working on...
AliasSetTracker *CurAST; // AliasSet information for the current loop...
- std::map<Loop *, AliasSetTracker *> LoopToAliasMap;
+ bool MayThrow; // The current loop contains an instruction which
+ // may throw, thus preventing code motion of
+ // instructions with side effects.
+ DenseMap<Loop*, AliasSetTracker*> LoopToAliasSetMap;
/// cloneBasicBlockAnalysis - Simple Analysis hook. Clone alias set info.
- void cloneBasicBlockAnalysis(BasicBlock *From, BasicBlock *To, Loop *L);
+ void cloneBasicBlockAnalysis(BasicBlock *From, BasicBlock *To,
+ Loop *L) override;
/// deleteAnalysisValue - Simple Analysis hook. Delete value V from alias
/// set.
- void deleteAnalysisValue(Value *V, Loop *L);
+ void deleteAnalysisValue(Value *V, Loop *L) override;
/// SinkRegion - Walk the specified region of the CFG (defined by all blocks
/// dominated by the specified block, and that are in the current loop) in
///
bool inSubLoop(BasicBlock *BB) {
assert(CurLoop->contains(BB) && "Only valid if BB is IN the loop");
- for (Loop::iterator I = CurLoop->begin(), E = CurLoop->end(); I != E; ++I)
- if ((*I)->contains(BB))
- return true; // A subloop actually contains this block!
- return false;
- }
-
- /// isExitBlockDominatedByBlockInLoop - This method checks to see if the
- /// specified exit block of the loop is dominated by the specified block
- /// that is in the body of the loop. We use these constraints to
- /// dramatically limit the amount of the dominator tree that needs to be
- /// searched.
- bool isExitBlockDominatedByBlockInLoop(BasicBlock *ExitBlock,
- BasicBlock *BlockInLoop) const {
- // If the block in the loop is the loop header, it must be dominated!
- BasicBlock *LoopHeader = CurLoop->getHeader();
- if (BlockInLoop == LoopHeader)
- return true;
-
- DomTreeNode *BlockInLoopNode = DT->getNode(BlockInLoop);
- DomTreeNode *IDom = DT->getNode(ExitBlock);
-
- // Because the exit block is not in the loop, we know we have to get _at
- // least_ its immediate dominator.
- IDom = IDom->getIDom();
-
- while (IDom && IDom != BlockInLoopNode) {
- // If we have got to the header of the loop, then the instructions block
- // did not dominate the exit node, so we can't hoist it.
- if (IDom->getBlock() == LoopHeader)
- return false;
-
- // Get next Immediate Dominator.
- IDom = IDom->getIDom();
- };
-
- return true;
+ return LI->getLoopFor(BB) != CurLoop;
}
/// sink - When an instruction is found to only be used outside of the loop,
///
bool isSafeToExecuteUnconditionally(Instruction &I);
+ /// isGuaranteedToExecute - Check that the instruction is guaranteed to
+ /// execute.
+ ///
+ bool isGuaranteedToExecute(Instruction &I);
+
/// pointerInvalidatedByLoop - Return true if the body of this loop may
/// store into the memory location pointed to by V.
///
- bool pointerInvalidatedByLoop(Value *V, unsigned Size) {
+ bool pointerInvalidatedByLoop(Value *V, uint64_t Size,
+ const AAMDNodes &AAInfo) {
// Check to see if any of the basic blocks in CurLoop invalidate *V.
- return CurAST->getAliasSetForPointer(V, Size).isMod();
+ return CurAST->getAliasSetForPointer(V, Size, AAInfo).isMod();
}
bool canSinkOrHoistInst(Instruction &I);
- bool isLoopInvariantInst(Instruction &I);
bool isNotUsedInLoop(Instruction &I);
- /// PromoteValuesInLoop - Look at the stores in the loop and promote as many
- /// to scalars as we can.
- ///
- void PromoteValuesInLoop();
-
- /// FindPromotableValuesInLoop - Check the current loop for stores to
- /// definite pointers, which are not loaded and stored through may aliases.
- /// If these are found, create an alloca for the value, add it to the
- /// PromotedValues list, and keep track of the mapping from value to
- /// alloca...
- ///
- void FindPromotableValuesInLoop(
- std::vector<std::pair<AllocaInst*, Value*> > &PromotedValues,
- std::map<Value*, AllocaInst*> &Val2AlMap);
+ void PromoteAliasSet(AliasSet &AS,
+ SmallVectorImpl<BasicBlock*> &ExitBlocks,
+ SmallVectorImpl<Instruction*> &InsertPts,
+ PredIteratorCache &PIC);
+
+ /// \brief Create a copy of the instruction in the exit block and patch up
+ /// SSA.
+ /// PN is a user of I in ExitBlock that can be used to get the number and
+ /// list of predecessors fast.
+ Instruction *CloneInstructionInExitBlock(Instruction &I,
+ BasicBlock &ExitBlock,
+ PHINode &PN);
};
}
char LICM::ID = 0;
-static RegisterPass<LICM> X("licm", "Loop Invariant Code Motion");
+INITIALIZE_PASS_BEGIN(LICM, "licm", "Loop Invariant Code Motion", false, false)
+INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
+INITIALIZE_PASS_DEPENDENCY(LoopInfo)
+INITIALIZE_PASS_DEPENDENCY(LoopSimplify)
+INITIALIZE_PASS_DEPENDENCY(LCSSA)
+INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)
+INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfo)
+INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
+INITIALIZE_PASS_END(LICM, "licm", "Loop Invariant Code Motion", false, false)
Pass *llvm::createLICMPass() { return new LICM(); }
/// Hoist expressions out of the specified loop. Note, alias info for inner
-/// loop is not preserved so it is not a good idea to run LICM multiple
+/// loop is not preserved so it is not a good idea to run LICM multiple
/// times on one loop.
///
bool LICM::runOnLoop(Loop *L, LPPassManager &LPM) {
+ if (skipOptnoneFunction(L))
+ return false;
+
Changed = false;
// Get our Loop and Alias Analysis information...
LI = &getAnalysis<LoopInfo>();
AA = &getAnalysis<AliasAnalysis>();
- DF = &getAnalysis<DominanceFrontier>();
- DT = &getAnalysis<DominatorTree>();
+ DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree();
+
+ DataLayoutPass *DLP = getAnalysisIfAvailable<DataLayoutPass>();
+ DL = DLP ? &DLP->getDataLayout() : nullptr;
+ TLI = &getAnalysis<TargetLibraryInfo>();
+
+ assert(L->isLCSSAForm(*DT) && "Loop is not in LCSSA form.");
CurAST = new AliasSetTracker(*AA);
- // Collect Alias info from subloops
+ // Collect Alias info from subloops.
for (Loop::iterator LoopItr = L->begin(), LoopItrE = L->end();
LoopItr != LoopItrE; ++LoopItr) {
Loop *InnerL = *LoopItr;
- AliasSetTracker *InnerAST = LoopToAliasMap[InnerL];
- assert (InnerAST && "Where is my AST?");
+ AliasSetTracker *InnerAST = LoopToAliasSetMap[InnerL];
+ assert(InnerAST && "Where is my AST?");
// What if InnerLoop was modified by other passes ?
CurAST->add(*InnerAST);
+
+ // Once we've incorporated the inner loop's AST into ours, we don't need the
+ // subloop's anymore.
+ delete InnerAST;
+ LoopToAliasSetMap.erase(InnerL);
}
-
+
CurLoop = L;
// Get the preheader block to move instructions into...
for (Loop::block_iterator I = L->block_begin(), E = L->block_end();
I != E; ++I) {
BasicBlock *BB = *I;
- if (LI->getLoopFor(BB) == L) // Ignore blocks in subloops...
+ if (LI->getLoopFor(BB) == L) // Ignore blocks in subloops.
CurAST->add(*BB); // Incorporate the specified basic block
}
+ MayThrow = false;
+ // TODO: We've already searched for instructions which may throw in subloops.
+ // We may want to reuse this information.
+ for (Loop::block_iterator BB = L->block_begin(), BBE = L->block_end();
+ (BB != BBE) && !MayThrow ; ++BB)
+ for (BasicBlock::iterator I = (*BB)->begin(), E = (*BB)->end();
+ (I != E) && !MayThrow; ++I)
+ MayThrow |= I->mayThrow();
+
// We want to visit all of the instructions in this loop... that are not parts
// of our subloops (they have already had their invariants hoisted out of
// their loop, into this loop, so there is no need to process the BODIES of
HoistRegion(DT->getNode(L->getHeader()));
// Now that all loop invariants have been removed from the loop, promote any
- // memory references to scalars that we can...
- if (!DisablePromotion && Preheader && L->hasDedicatedExits())
- PromoteValuesInLoop();
+ // memory references to scalars that we can.
+ if (!DisablePromotion && (Preheader || L->hasDedicatedExits())) {
+ SmallVector<BasicBlock *, 8> ExitBlocks;
+ SmallVector<Instruction *, 8> InsertPts;
+ PredIteratorCache PIC;
+
+ // Loop over all of the alias sets in the tracker object.
+ for (AliasSetTracker::iterator I = CurAST->begin(), E = CurAST->end();
+ I != E; ++I)
+ PromoteAliasSet(*I, ExitBlocks, InsertPts, PIC);
+
+ // Once we have promoted values across the loop body we have to recursively
+ // reform LCSSA as any nested loop may now have values defined within the
+ // loop used in the outer loop.
+ // FIXME: This is really heavy handed. It would be a bit better to use an
+ // SSAUpdater strategy during promotion that was LCSSA aware and reformed
+ // it as it went.
+ if (Changed)
+ formLCSSARecursively(*L, *DT, getAnalysisIfAvailable<ScalarEvolution>());
+ }
- // Clear out loops state information for the next iteration
- CurLoop = 0;
- Preheader = 0;
+ // Check that neither this loop nor its parent have had LCSSA broken. LICM is
+ // specifically moving instructions across the loop boundary and so it is
+ // especially in need of sanity checking here.
+ assert(L->isLCSSAForm(*DT) && "Loop not left in LCSSA form after LICM!");
+ assert((!L->getParentLoop() || L->getParentLoop()->isLCSSAForm(*DT)) &&
+ "Parent loop not left in LCSSA form after LICM!");
- LoopToAliasMap[L] = CurAST;
+ // Clear out loops state information for the next iteration
+ CurLoop = nullptr;
+ Preheader = nullptr;
+
+ // If this loop is nested inside of another one, save the alias information
+ // for when we process the outer loop.
+ if (L->getParentLoop())
+ LoopToAliasSetMap[L] = CurAST;
+ else
+ delete CurAST;
return Changed;
}
/// iteration.
///
void LICM::SinkRegion(DomTreeNode *N) {
- assert(N != 0 && "Null dominator tree node?");
+ assert(N != nullptr && "Null dominator tree node?");
BasicBlock *BB = N->getBlock();
// If this subregion is not in the top level loop at all, exit.
if (!CurLoop->contains(BB)) return;
- // We are processing blocks in reverse dfo, so process children first...
+ // We are processing blocks in reverse dfo, so process children first.
const std::vector<DomTreeNode*> &Children = N->getChildren();
for (unsigned i = 0, e = Children.size(); i != e; ++i)
SinkRegion(Children[i]);
for (BasicBlock::iterator II = BB->end(); II != BB->begin(); ) {
Instruction &I = *--II;
+ // If the instruction is dead, we would try to sink it because it isn't used
+ // in the loop, instead, just delete it.
+ if (isInstructionTriviallyDead(&I, TLI)) {
+ DEBUG(dbgs() << "LICM deleting dead inst: " << I << '\n');
+ ++II;
+ CurAST->deleteValue(&I);
+ I.eraseFromParent();
+ Changed = true;
+ continue;
+ }
+
// Check to see if we can sink this instruction to the exit blocks
// of the loop. We can do this if the all users of the instruction are
// outside of the loop. In this case, it doesn't even matter if the
/// before uses, allowing us to hoist a loop body in one pass without iteration.
///
void LICM::HoistRegion(DomTreeNode *N) {
- assert(N != 0 && "Null dominator tree node?");
+ assert(N != nullptr && "Null dominator tree node?");
BasicBlock *BB = N->getBlock();
// If this subregion is not in the top level loop at all, exit.
for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ) {
Instruction &I = *II++;
+ // Try constant folding this instruction. If all the operands are
+ // constants, it is technically hoistable, but it would be better to just
+ // fold it.
+ if (Constant *C = ConstantFoldInstruction(&I, DL, TLI)) {
+ DEBUG(dbgs() << "LICM folding inst: " << I << " --> " << *C << '\n');
+ CurAST->copyValue(&I, C);
+ CurAST->deleteValue(&I);
+ I.replaceAllUsesWith(C);
+ I.eraseFromParent();
+ continue;
+ }
+
// Try hoisting the instruction out to the preheader. We can only do this
// if all of the operands of the instruction are loop invariant and if it
// is safe to hoist the instruction.
//
- if (isLoopInvariantInst(I) && canSinkOrHoistInst(I) &&
+ if (CurLoop->hasLoopInvariantOperands(&I) && canSinkOrHoistInst(I) &&
isSafeToExecuteUnconditionally(I))
hoist(I);
- }
+ }
const std::vector<DomTreeNode*> &Children = N->getChildren();
for (unsigned i = 0, e = Children.size(); i != e; ++i)
bool LICM::canSinkOrHoistInst(Instruction &I) {
// Loads have extra constraints we have to verify before we can hoist them.
if (LoadInst *LI = dyn_cast<LoadInst>(&I)) {
- if (LI->isVolatile())
- return false; // Don't hoist volatile loads!
+ if (!LI->isUnordered())
+ return false; // Don't hoist volatile/atomic loads!
// Loads from constant memory are always safe to move, even if they end up
// in the same alias set as something that ends up being modified.
if (AA->pointsToConstantMemory(LI->getOperand(0)))
return true;
-
+ if (LI->getMetadata("invariant.load"))
+ return true;
+
// Don't hoist loads which have may-aliased stores in loop.
- unsigned Size = 0;
+ uint64_t Size = 0;
if (LI->getType()->isSized())
Size = AA->getTypeStoreSize(LI->getType());
- return !pointerInvalidatedByLoop(LI->getOperand(0), Size);
+
+ AAMDNodes AAInfo;
+ LI->getAAMetadata(AAInfo);
+
+ return !pointerInvalidatedByLoop(LI->getOperand(0), Size, AAInfo);
} else if (CallInst *CI = dyn_cast<CallInst>(&I)) {
- // Handle obvious cases efficiently.
+ // Don't sink or hoist dbg info; it's legal, but not useful.
+ if (isa<DbgInfoIntrinsic>(I))
+ return false;
+
+ // Handle simple cases by querying alias analysis.
AliasAnalysis::ModRefBehavior Behavior = AA->getModRefBehavior(CI);
if (Behavior == AliasAnalysis::DoesNotAccessMemory)
return true;
- else if (Behavior == AliasAnalysis::OnlyReadsMemory) {
+ if (AliasAnalysis::onlyReadsMemory(Behavior)) {
// If this call only reads from memory and there are no writes to memory
// in the loop, we can hoist or sink the call as appropriate.
bool FoundMod = false;
if (!FoundMod) return true;
}
- // FIXME: This should use mod/ref information to see if we can hoist or sink
- // the call.
+ // FIXME: This should use mod/ref information to see if we can hoist or
+ // sink the call.
return false;
}
- // Otherwise these instructions are hoistable/sinkable
- return isa<BinaryOperator>(I) || isa<CastInst>(I) ||
- isa<SelectInst>(I) || isa<GetElementPtrInst>(I) || isa<CmpInst>(I) ||
- isa<InsertElementInst>(I) || isa<ExtractElementInst>(I) ||
- isa<ShuffleVectorInst>(I);
+ // Only these instructions are hoistable/sinkable.
+ if (!isa<BinaryOperator>(I) && !isa<CastInst>(I) && !isa<SelectInst>(I) &&
+ !isa<GetElementPtrInst>(I) && !isa<CmpInst>(I) &&
+ !isa<InsertElementInst>(I) && !isa<ExtractElementInst>(I) &&
+ !isa<ShuffleVectorInst>(I) && !isa<ExtractValueInst>(I) &&
+ !isa<InsertValueInst>(I))
+ return false;
+
+ return isSafeToExecuteUnconditionally(I);
+}
+
+/// \brief Returns true if a PHINode is a trivially replaceable with an
+/// Instruction.
+///
+/// This is true when all incoming values are that instruction. This pattern
+/// occurs most often with LCSSA PHI nodes.
+static bool isTriviallyReplacablePHI(PHINode &PN, Instruction &I) {
+ for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i)
+ if (PN.getIncomingValue(i) != &I)
+ return false;
+
+ return true;
}
/// isNotUsedInLoop - Return true if the only users of this instruction are
/// exit blocks of the loop.
///
bool LICM::isNotUsedInLoop(Instruction &I) {
- for (Value::use_iterator UI = I.use_begin(), E = I.use_end(); UI != E; ++UI) {
- Instruction *User = cast<Instruction>(*UI);
- if (PHINode *PN = dyn_cast<PHINode>(User)) {
- // PHI node uses occur in predecessor blocks!
+ for (User *U : I.users()) {
+ Instruction *UI = cast<Instruction>(U);
+ if (PHINode *PN = dyn_cast<PHINode>(UI)) {
+ // A PHI node where all of the incoming values are this instruction are
+ // special -- they can just be RAUW'ed with the instruction and thus
+ // don't require a use in the predecessor. This is a particular important
+ // special case because it is the pattern found in LCSSA form.
+ if (isTriviallyReplacablePHI(*PN, I)) {
+ if (CurLoop->contains(PN))
+ return false;
+ else
+ continue;
+ }
+
+ // Otherwise, PHI node uses occur in predecessor blocks if the incoming
+ // values. Check for such a use being inside the loop.
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
if (PN->getIncomingValue(i) == &I)
if (CurLoop->contains(PN->getIncomingBlock(i)))
return false;
- } else if (CurLoop->contains(User)) {
- return false;
+
+ continue;
}
+
+ if (CurLoop->contains(UI))
+ return false;
}
return true;
}
-
-/// isLoopInvariantInst - Return true if all operands of this instruction are
-/// loop invariant. We also filter out non-hoistable instructions here just for
-/// efficiency.
-///
-bool LICM::isLoopInvariantInst(Instruction &I) {
- // The instruction is loop invariant if all of its operands are loop-invariant
- for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i)
- if (!CurLoop->isLoopInvariant(I.getOperand(i)))
- return false;
-
- // If we got this far, the instruction is loop invariant!
- return true;
+Instruction *LICM::CloneInstructionInExitBlock(Instruction &I,
+ BasicBlock &ExitBlock,
+ PHINode &PN) {
+ Instruction *New = I.clone();
+ ExitBlock.getInstList().insert(ExitBlock.getFirstInsertionPt(), New);
+ if (!I.getName().empty()) New->setName(I.getName() + ".le");
+
+ // Build LCSSA PHI nodes for any in-loop operands. Note that this is
+ // particularly cheap because we can rip off the PHI node that we're
+ // replacing for the number and blocks of the predecessors.
+ // OPT: If this shows up in a profile, we can instead finish sinking all
+ // invariant instructions, and then walk their operands to re-establish
+ // LCSSA. That will eliminate creating PHI nodes just to nuke them when
+ // sinking bottom-up.
+ for (User::op_iterator OI = New->op_begin(), OE = New->op_end(); OI != OE;
+ ++OI)
+ if (Instruction *OInst = dyn_cast<Instruction>(*OI))
+ if (Loop *OLoop = LI->getLoopFor(OInst->getParent()))
+ if (!OLoop->contains(&PN)) {
+ PHINode *OpPN =
+ PHINode::Create(OInst->getType(), PN.getNumIncomingValues(),
+ OInst->getName() + ".lcssa", ExitBlock.begin());
+ for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i)
+ OpPN->addIncoming(OInst, PN.getIncomingBlock(i));
+ *OI = OpPN;
+ }
+ return New;
}
/// sink - When an instruction is found to only be used outside of the loop,
/// position, and may either delete it or move it to outside of the loop.
///
void LICM::sink(Instruction &I) {
- DEBUG(dbgs() << "LICM sinking instruction: " << I);
-
- SmallVector<BasicBlock*, 8> ExitBlocks;
- CurLoop->getExitBlocks(ExitBlocks);
+ DEBUG(dbgs() << "LICM sinking instruction: " << I << "\n");
if (isa<LoadInst>(I)) ++NumMovedLoads;
else if (isa<CallInst>(I)) ++NumMovedCalls;
++NumSunk;
Changed = true;
- // The case where there is only a single exit node of this loop is common
- // enough that we handle it as a special (more efficient) case. It is more
- // efficient to handle because there are no PHI nodes that need to be placed.
- if (ExitBlocks.size() == 1) {
- if (!isExitBlockDominatedByBlockInLoop(ExitBlocks[0], I.getParent())) {
- // Instruction is not used, just delete it.
- CurAST->deleteValue(&I);
- // If I has users in unreachable blocks, eliminate.
- // If I is not void type then replaceAllUsesWith undef.
- // This allows ValueHandlers and custom metadata to adjust itself.
- if (!I.getType()->isVoidTy())
- I.replaceAllUsesWith(UndefValue::get(I.getType()));
- I.eraseFromParent();
- } else {
- // Move the instruction to the start of the exit block, after any PHI
- // nodes in it.
- I.removeFromParent();
- BasicBlock::iterator InsertPt = ExitBlocks[0]->getFirstNonPHI();
- ExitBlocks[0]->getInstList().insert(InsertPt, &I);
- }
- } else if (ExitBlocks.empty()) {
- // The instruction is actually dead if there ARE NO exit blocks.
- CurAST->deleteValue(&I);
- // If I has users in unreachable blocks, eliminate.
- // If I is not void type then replaceAllUsesWith undef.
- // This allows ValueHandlers and custom metadata to adjust itself.
- if (!I.getType()->isVoidTy())
- I.replaceAllUsesWith(UndefValue::get(I.getType()));
- I.eraseFromParent();
- } else {
- // Otherwise, if we have multiple exits, use the PromoteMem2Reg function to
- // do all of the hard work of inserting PHI nodes as necessary. We convert
- // the value into a stack object to get it to do this.
-
- // Firstly, we create a stack object to hold the value...
- AllocaInst *AI = 0;
-
- if (!I.getType()->isVoidTy()) {
- AI = new AllocaInst(I.getType(), 0, I.getName(),
- I.getParent()->getParent()->getEntryBlock().begin());
- CurAST->add(AI);
- }
-
- // Secondly, insert load instructions for each use of the instruction
- // outside of the loop.
- while (!I.use_empty()) {
- Instruction *U = cast<Instruction>(I.use_back());
-
- // If the user is a PHI Node, we actually have to insert load instructions
- // in all predecessor blocks, not in the PHI block itself!
- if (PHINode *UPN = dyn_cast<PHINode>(U)) {
- // Only insert into each predecessor once, so that we don't have
- // different incoming values from the same block!
- std::map<BasicBlock*, Value*> InsertedBlocks;
- for (unsigned i = 0, e = UPN->getNumIncomingValues(); i != e; ++i)
- if (UPN->getIncomingValue(i) == &I) {
- BasicBlock *Pred = UPN->getIncomingBlock(i);
- Value *&PredVal = InsertedBlocks[Pred];
- if (!PredVal) {
- // Insert a new load instruction right before the terminator in
- // the predecessor block.
- PredVal = new LoadInst(AI, "", Pred->getTerminator());
- CurAST->add(cast<LoadInst>(PredVal));
- }
-
- UPN->setIncomingValue(i, PredVal);
- }
-
- } else {
- LoadInst *L = new LoadInst(AI, "", U);
- U->replaceUsesOfWith(&I, L);
- CurAST->add(L);
- }
- }
-
- // Thirdly, insert a copy of the instruction in each exit block of the loop
- // that is dominated by the instruction, storing the result into the memory
- // location. Be careful not to insert the instruction into any particular
- // basic block more than once.
- std::set<BasicBlock*> InsertedBlocks;
- BasicBlock *InstOrigBB = I.getParent();
-
- for (unsigned i = 0, e = ExitBlocks.size(); i != e; ++i) {
- BasicBlock *ExitBlock = ExitBlocks[i];
-
- if (isExitBlockDominatedByBlockInLoop(ExitBlock, InstOrigBB)) {
- // If we haven't already processed this exit block, do so now.
- if (InsertedBlocks.insert(ExitBlock).second) {
- // Insert the code after the last PHI node...
- BasicBlock::iterator InsertPt = ExitBlock->getFirstNonPHI();
-
- // If this is the first exit block processed, just move the original
- // instruction, otherwise clone the original instruction and insert
- // the copy.
- Instruction *New;
- if (InsertedBlocks.size() == 1) {
- I.removeFromParent();
- ExitBlock->getInstList().insert(InsertPt, &I);
- New = &I;
- } else {
- New = I.clone();
- CurAST->copyValue(&I, New);
- if (!I.getName().empty())
- New->setName(I.getName()+".le");
- ExitBlock->getInstList().insert(InsertPt, New);
- }
-
- // Now that we have inserted the instruction, store it into the alloca
- if (AI) new StoreInst(New, AI, InsertPt);
- }
- }
- }
-
- // If the instruction doesn't dominate any exit blocks, it must be dead.
- if (InsertedBlocks.empty()) {
- CurAST->deleteValue(&I);
- I.eraseFromParent();
- }
-
- // Finally, promote the fine value to SSA form.
- if (AI) {
- std::vector<AllocaInst*> Allocas;
- Allocas.push_back(AI);
- PromoteMemToReg(Allocas, *DT, *DF, CurAST);
- }
+#ifndef NDEBUG
+ SmallVector<BasicBlock *, 32> ExitBlocks;
+ CurLoop->getUniqueExitBlocks(ExitBlocks);
+ SmallPtrSet<BasicBlock *, 32> ExitBlockSet(ExitBlocks.begin(), ExitBlocks.end());
+#endif
+
+ // Clones of this instruction. Don't create more than one per exit block!
+ SmallDenseMap<BasicBlock *, Instruction *, 32> SunkCopies;
+
+ // If this instruction is only used outside of the loop, then all users are
+ // PHI nodes in exit blocks due to LCSSA form. Just RAUW them with clones of
+ // the instruction.
+ while (!I.use_empty()) {
+ // The user must be a PHI node.
+ PHINode *PN = cast<PHINode>(I.user_back());
+
+ BasicBlock *ExitBlock = PN->getParent();
+ assert(ExitBlockSet.count(ExitBlock) &&
+ "The LCSSA PHI is not in an exit block!");
+
+ Instruction *New;
+ auto It = SunkCopies.find(ExitBlock);
+ if (It != SunkCopies.end())
+ New = It->second;
+ else
+ New = SunkCopies[ExitBlock] =
+ CloneInstructionInExitBlock(I, *ExitBlock, *PN);
+
+ PN->replaceAllUsesWith(New);
+ PN->eraseFromParent();
}
+
+ CurAST->deleteValue(&I);
+ I.eraseFromParent();
}
/// hoist - When an instruction is found to only use loop invariant operands
DEBUG(dbgs() << "LICM hoisting to " << Preheader->getName() << ": "
<< I << "\n");
- // Remove the instruction from its current basic block... but don't delete the
- // instruction.
- I.removeFromParent();
-
- // Insert the new node in Preheader, before the terminator.
- Preheader->getInstList().insert(Preheader->getTerminator(), &I);
+ // Move the new node to the Preheader, before its terminator.
+ I.moveBefore(Preheader->getTerminator());
if (isa<LoadInst>(I)) ++NumMovedLoads;
else if (isa<CallInst>(I)) ++NumMovedCalls;
///
bool LICM::isSafeToExecuteUnconditionally(Instruction &Inst) {
// If it is not a trapping instruction, it is always safe to hoist.
- if (Inst.isSafeToSpeculativelyExecute())
+ if (isSafeToSpeculativelyExecute(&Inst, DL))
return true;
+ return isGuaranteedToExecute(Inst);
+}
+
+bool LICM::isGuaranteedToExecute(Instruction &Inst) {
+
+ // Somewhere in this loop there is an instruction which may throw and make us
+ // exit the loop.
+ if (MayThrow)
+ return false;
+
// Otherwise we have to check to make sure that the instruction dominates all
// of the exit blocks. If it doesn't, then there is a path out of the loop
// which does not execute this instruction, so we can't hoist it.
SmallVector<BasicBlock*, 8> ExitBlocks;
CurLoop->getExitBlocks(ExitBlocks);
- // For each exit block, get the DT node and walk up the DT until the
- // instruction's basic block is found or we exit the loop.
+ // Verify that the block dominates each of the exit blocks of the loop.
for (unsigned i = 0, e = ExitBlocks.size(); i != e; ++i)
- if (!isExitBlockDominatedByBlockInLoop(ExitBlocks[i], Inst.getParent()))
+ if (!DT->dominates(Inst.getParent(), ExitBlocks[i]))
return false;
+ // As a degenerate case, if the loop is statically infinite then we haven't
+ // proven anything since there are no exit blocks.
+ if (ExitBlocks.empty())
+ return false;
+
return true;
}
+namespace {
+ class LoopPromoter : public LoadAndStorePromoter {
+ Value *SomePtr; // Designated pointer to store to.
+ SmallPtrSet<Value*, 4> &PointerMustAliases;
+ SmallVectorImpl<BasicBlock*> &LoopExitBlocks;
+ SmallVectorImpl<Instruction*> &LoopInsertPts;
+ PredIteratorCache &PredCache;
+ AliasSetTracker &AST;
+ LoopInfo &LI;
+ DebugLoc DL;
+ int Alignment;
+ AAMDNodes AATags;
+
+ Value *maybeInsertLCSSAPHI(Value *V, BasicBlock *BB) const {
+ if (Instruction *I = dyn_cast<Instruction>(V))
+ if (Loop *L = LI.getLoopFor(I->getParent()))
+ if (!L->contains(BB)) {
+ // We need to create an LCSSA PHI node for the incoming value and
+ // store that.
+ PHINode *PN = PHINode::Create(
+ I->getType(), PredCache.GetNumPreds(BB),
+ I->getName() + ".lcssa", BB->begin());
+ for (BasicBlock **PI = PredCache.GetPreds(BB); *PI; ++PI)
+ PN->addIncoming(I, *PI);
+ return PN;
+ }
+ return V;
+ }
-/// PromoteValuesInLoop - Try to promote memory values to scalars by sinking
-/// stores out of the loop and moving loads to before the loop. We do this by
-/// looping over the stores in the loop, looking for stores to Must pointers
-/// which are loop invariant. We promote these memory locations to use allocas
-/// instead. These allocas can easily be raised to register values by the
-/// PromoteMem2Reg functionality.
-///
-void LICM::PromoteValuesInLoop() {
- // PromotedValues - List of values that are promoted out of the loop. Each
- // value has an alloca instruction for it, and a canonical version of the
- // pointer.
- std::vector<std::pair<AllocaInst*, Value*> > PromotedValues;
- std::map<Value*, AllocaInst*> ValueToAllocaMap; // Map of ptr to alloca
-
- FindPromotableValuesInLoop(PromotedValues, ValueToAllocaMap);
- if (ValueToAllocaMap.empty()) return; // If there are values to promote.
+ public:
+ LoopPromoter(Value *SP, const SmallVectorImpl<Instruction *> &Insts,
+ SSAUpdater &S, SmallPtrSet<Value *, 4> &PMA,
+ SmallVectorImpl<BasicBlock *> &LEB,
+ SmallVectorImpl<Instruction *> &LIP, PredIteratorCache &PIC,
+ AliasSetTracker &ast, LoopInfo &li, DebugLoc dl, int alignment,
+ const AAMDNodes &AATags)
+ : LoadAndStorePromoter(Insts, S), SomePtr(SP), PointerMustAliases(PMA),
+ LoopExitBlocks(LEB), LoopInsertPts(LIP), PredCache(PIC), AST(ast),
+ LI(li), DL(dl), Alignment(alignment), AATags(AATags) {}
+
+ bool isInstInList(Instruction *I,
+ const SmallVectorImpl<Instruction*> &) const override {
+ Value *Ptr;
+ if (LoadInst *LI = dyn_cast<LoadInst>(I))
+ Ptr = LI->getOperand(0);
+ else
+ Ptr = cast<StoreInst>(I)->getPointerOperand();
+ return PointerMustAliases.count(Ptr);
+ }
- Changed = true;
- NumPromoted += PromotedValues.size();
-
- std::vector<Value*> PointerValueNumbers;
-
- // Emit a copy from the value into the alloca'd value in the loop preheader
- TerminatorInst *LoopPredInst = Preheader->getTerminator();
- for (unsigned i = 0, e = PromotedValues.size(); i != e; ++i) {
- Value *Ptr = PromotedValues[i].second;
-
- // If we are promoting a pointer value, update alias information for the
- // inserted load.
- Value *LoadValue = 0;
- if (cast<PointerType>(Ptr->getType())->getElementType()->isPointerTy()) {
- // Locate a load or store through the pointer, and assign the same value
- // to LI as we are loading or storing. Since we know that the value is
- // stored in this loop, this will always succeed.
- for (Value::use_iterator UI = Ptr->use_begin(), E = Ptr->use_end();
- UI != E; ++UI)
- if (LoadInst *LI = dyn_cast<LoadInst>(*UI)) {
- LoadValue = LI;
- break;
- } else if (StoreInst *SI = dyn_cast<StoreInst>(*UI)) {
- if (SI->getOperand(1) == Ptr) {
- LoadValue = SI->getOperand(0);
- break;
- }
- }
- assert(LoadValue && "No store through the pointer found!");
- PointerValueNumbers.push_back(LoadValue); // Remember this for later.
+ void doExtraRewritesBeforeFinalDeletion() const override {
+ // Insert stores after in the loop exit blocks. Each exit block gets a
+ // store of the live-out values that feed them. Since we've already told
+ // the SSA updater about the defs in the loop and the preheader
+ // definition, it is all set and we can start using it.
+ for (unsigned i = 0, e = LoopExitBlocks.size(); i != e; ++i) {
+ BasicBlock *ExitBlock = LoopExitBlocks[i];
+ Value *LiveInValue = SSA.GetValueInMiddleOfBlock(ExitBlock);
+ LiveInValue = maybeInsertLCSSAPHI(LiveInValue, ExitBlock);
+ Value *Ptr = maybeInsertLCSSAPHI(SomePtr, ExitBlock);
+ Instruction *InsertPos = LoopInsertPts[i];
+ StoreInst *NewSI = new StoreInst(LiveInValue, Ptr, InsertPos);
+ NewSI->setAlignment(Alignment);
+ NewSI->setDebugLoc(DL);
+ if (AATags) NewSI->setAAMetadata(AATags);
+ }
}
- // Load from the memory we are promoting.
- LoadInst *LI = new LoadInst(Ptr, Ptr->getName()+".promoted", LoopPredInst);
+ void replaceLoadWithValue(LoadInst *LI, Value *V) const override {
+ // Update alias analysis.
+ AST.copyValue(LI, V);
+ }
+ void instructionDeleted(Instruction *I) const override {
+ AST.deleteValue(I);
+ }
+ };
+} // end anon namespace
- if (LoadValue) CurAST->copyValue(LoadValue, LI);
+/// PromoteAliasSet - Try to promote memory values to scalars by sinking
+/// stores out of the loop and moving loads to before the loop. We do this by
+/// looping over the stores in the loop, looking for stores to Must pointers
+/// which are loop invariant.
+///
+void LICM::PromoteAliasSet(AliasSet &AS,
+ SmallVectorImpl<BasicBlock*> &ExitBlocks,
+ SmallVectorImpl<Instruction*> &InsertPts,
+ PredIteratorCache &PIC) {
+ // We can promote this alias set if it has a store, if it is a "Must" alias
+ // set, if the pointer is loop invariant, and if we are not eliminating any
+ // volatile loads or stores.
+ if (AS.isForwardingAliasSet() || !AS.isMod() || !AS.isMustAlias() ||
+ AS.isVolatile() || !CurLoop->isLoopInvariant(AS.begin()->getValue()))
+ return;
- // Store into the temporary alloca.
- new StoreInst(LI, PromotedValues[i].first, LoopPredInst);
- }
+ assert(!AS.empty() &&
+ "Must alias set should have at least one pointer element in it!");
+ Value *SomePtr = AS.begin()->getValue();
- // Scan the basic blocks in the loop, replacing uses of our pointers with
- // uses of the allocas in question.
+ // It isn't safe to promote a load/store from the loop if the load/store is
+ // conditional. For example, turning:
//
- for (Loop::block_iterator I = CurLoop->block_begin(),
- E = CurLoop->block_end(); I != E; ++I) {
- BasicBlock *BB = *I;
- // Rewrite all loads and stores in the block of the pointer...
- for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
- if (LoadInst *L = dyn_cast<LoadInst>(II)) {
- std::map<Value*, AllocaInst*>::iterator
- I = ValueToAllocaMap.find(L->getOperand(0));
- if (I != ValueToAllocaMap.end())
- L->setOperand(0, I->second); // Rewrite load instruction...
- } else if (StoreInst *S = dyn_cast<StoreInst>(II)) {
- std::map<Value*, AllocaInst*>::iterator
- I = ValueToAllocaMap.find(S->getOperand(1));
- if (I != ValueToAllocaMap.end())
- S->setOperand(1, I->second); // Rewrite store instruction...
- }
- }
- }
-
- // Now that the body of the loop uses the allocas instead of the original
- // memory locations, insert code to copy the alloca value back into the
- // original memory location on all exits from the loop. Note that we only
- // want to insert one copy of the code in each exit block, though the loop may
- // exit to the same block more than once.
+ // for () { if (c) *P += 1; }
+ //
+ // into:
+ //
+ // tmp = *P; for () { if (c) tmp +=1; } *P = tmp;
//
- SmallPtrSet<BasicBlock*, 16> ProcessedBlocks;
+ // is not safe, because *P may only be valid to access if 'c' is true.
+ //
+ // It is safe to promote P if all uses are direct load/stores and if at
+ // least one is guaranteed to be executed.
+ bool GuaranteedToExecute = false;
- SmallVector<BasicBlock*, 8> ExitBlocks;
- CurLoop->getExitBlocks(ExitBlocks);
- for (unsigned i = 0, e = ExitBlocks.size(); i != e; ++i) {
- if (!ProcessedBlocks.insert(ExitBlocks[i]))
- continue;
-
- // Copy all of the allocas into their memory locations.
- BasicBlock::iterator BI = ExitBlocks[i]->getFirstNonPHI();
- Instruction *InsertPos = BI;
- unsigned PVN = 0;
- for (unsigned i = 0, e = PromotedValues.size(); i != e; ++i) {
- // Load from the alloca.
- LoadInst *LI = new LoadInst(PromotedValues[i].first, "", InsertPos);
-
- // If this is a pointer type, update alias info appropriately.
- if (LI->getType()->isPointerTy())
- CurAST->copyValue(PointerValueNumbers[PVN++], LI);
-
- // Store into the memory we promoted.
- new StoreInst(LI, PromotedValues[i].second, InsertPos);
- }
- }
+ SmallVector<Instruction*, 64> LoopUses;
+ SmallPtrSet<Value*, 4> PointerMustAliases;
- // Now that we have done the deed, use the mem2reg functionality to promote
- // all of the new allocas we just created into real SSA registers.
- //
- std::vector<AllocaInst*> PromotedAllocas;
- PromotedAllocas.reserve(PromotedValues.size());
- for (unsigned i = 0, e = PromotedValues.size(); i != e; ++i)
- PromotedAllocas.push_back(PromotedValues[i].first);
- PromoteMemToReg(PromotedAllocas, *DT, *DF, CurAST);
-}
+ // We start with an alignment of one and try to find instructions that allow
+ // us to prove better alignment.
+ unsigned Alignment = 1;
+ AAMDNodes AATags;
-/// FindPromotableValuesInLoop - Check the current loop for stores to definite
-/// pointers, which are not loaded and stored through may aliases and are safe
-/// for promotion. If these are found, create an alloca for the value, add it
-/// to the PromotedValues list, and keep track of the mapping from value to
-/// alloca.
-void LICM::FindPromotableValuesInLoop(
- std::vector<std::pair<AllocaInst*, Value*> > &PromotedValues,
- std::map<Value*, AllocaInst*> &ValueToAllocaMap) {
- Instruction *FnStart = CurLoop->getHeader()->getParent()->begin()->begin();
-
- // Loop over all of the alias sets in the tracker object.
- for (AliasSetTracker::iterator I = CurAST->begin(), E = CurAST->end();
- I != E; ++I) {
- AliasSet &AS = *I;
- // We can promote this alias set if it has a store, if it is a "Must" alias
- // set, if the pointer is loop invariant, and if we are not eliminating any
- // volatile loads or stores.
- if (AS.isForwardingAliasSet() || !AS.isMod() || !AS.isMustAlias() ||
- AS.isVolatile() || !CurLoop->isLoopInvariant(AS.begin()->getValue()))
- continue;
-
- assert(!AS.empty() &&
- "Must alias set should have at least one pointer element in it!");
- Value *V = AS.begin()->getValue();
+ // Check that all of the pointers in the alias set have the same type. We
+ // cannot (yet) promote a memory location that is loaded and stored in
+ // different sizes. While we are at it, collect alignment and AA info.
+ for (AliasSet::iterator ASI = AS.begin(), E = AS.end(); ASI != E; ++ASI) {
+ Value *ASIV = ASI->getValue();
+ PointerMustAliases.insert(ASIV);
// Check that all of the pointers in the alias set have the same type. We
// cannot (yet) promote a memory location that is loaded and stored in
// different sizes.
- {
- bool PointerOk = true;
- for (AliasSet::iterator I = AS.begin(), E = AS.end(); I != E; ++I)
- if (V->getType() != I->getValue()->getType()) {
- PointerOk = false;
- break;
- }
- if (!PointerOk)
- continue;
- }
+ if (SomePtr->getType() != ASIV->getType())
+ return;
- // It isn't safe to promote a load/store from the loop if the load/store is
- // conditional. For example, turning:
- //
- // for () { if (c) *P += 1; }
- //
- // into:
- //
- // tmp = *P; for () { if (c) tmp +=1; } *P = tmp;
- //
- // is not safe, because *P may only be valid to access if 'c' is true.
- //
- // It is safe to promote P if all uses are direct load/stores and if at
- // least one is guaranteed to be executed.
- bool GuaranteedToExecute = false;
- bool InvalidInst = false;
- for (Value::use_iterator UI = V->use_begin(), UE = V->use_end();
- UI != UE; ++UI) {
- // Ignore instructions not in this loop.
- Instruction *Use = dyn_cast<Instruction>(*UI);
- if (!Use || !CurLoop->contains(Use))
+ for (User *U : ASIV->users()) {
+ // Ignore instructions that are outside the loop.
+ Instruction *UI = dyn_cast<Instruction>(U);
+ if (!UI || !CurLoop->contains(UI))
continue;
- if (!isa<LoadInst>(Use) && !isa<StoreInst>(Use)) {
- InvalidInst = true;
- break;
- }
-
- if (!GuaranteedToExecute)
- GuaranteedToExecute = isSafeToExecuteUnconditionally(*Use);
- }
+ // If there is an non-load/store instruction in the loop, we can't promote
+ // it.
+ if (LoadInst *load = dyn_cast<LoadInst>(UI)) {
+ assert(!load->isVolatile() && "AST broken");
+ if (!load->isSimple())
+ return;
+ } else if (StoreInst *store = dyn_cast<StoreInst>(UI)) {
+ // Stores *of* the pointer are not interesting, only stores *to* the
+ // pointer.
+ if (UI->getOperand(1) != ASIV)
+ continue;
+ assert(!store->isVolatile() && "AST broken");
+ if (!store->isSimple())
+ return;
+
+ // Note that we only check GuaranteedToExecute inside the store case
+ // so that we do not introduce stores where they did not exist before
+ // (which would break the LLVM concurrency model).
+
+ // If the alignment of this instruction allows us to specify a more
+ // restrictive (and performant) alignment and if we are sure this
+ // instruction will be executed, update the alignment.
+ // Larger is better, with the exception of 0 being the best alignment.
+ unsigned InstAlignment = store->getAlignment();
+ if ((InstAlignment > Alignment || InstAlignment == 0) && Alignment != 0)
+ if (isGuaranteedToExecute(*UI)) {
+ GuaranteedToExecute = true;
+ Alignment = InstAlignment;
+ }
- // If there is an non-load/store instruction in the loop, we can't promote
- // it. If there isn't a guaranteed-to-execute instruction, we can't
- // promote.
- if (InvalidInst || !GuaranteedToExecute)
- continue;
-
- const Type *Ty = cast<PointerType>(V->getType())->getElementType();
- AllocaInst *AI = new AllocaInst(Ty, 0, V->getName()+".tmp", FnStart);
- PromotedValues.push_back(std::make_pair(AI, V));
+ if (!GuaranteedToExecute)
+ GuaranteedToExecute = isGuaranteedToExecute(*UI);
+
+ } else
+ return; // Not a load or store.
- // Update the AST and alias analysis.
- CurAST->copyValue(V, AI);
+ // Merge the AA tags.
+ if (LoopUses.empty()) {
+ // On the first load/store, just take its AA tags.
+ UI->getAAMetadata(AATags);
+ } else if (AATags) {
+ UI->getAAMetadata(AATags, /* Merge = */ true);
+ }
+
+ LoopUses.push_back(UI);
+ }
+ }
- for (AliasSet::iterator I = AS.begin(), E = AS.end(); I != E; ++I)
- ValueToAllocaMap.insert(std::make_pair(I->getValue(), AI));
+ // If there isn't a guaranteed-to-execute instruction, we can't promote.
+ if (!GuaranteedToExecute)
+ return;
- DEBUG(dbgs() << "LICM: Promoting value: " << *V << "\n");
+ // Otherwise, this is safe to promote, lets do it!
+ DEBUG(dbgs() << "LICM: Promoting value stored to in loop: " <<*SomePtr<<'\n');
+ Changed = true;
+ ++NumPromoted;
+
+ // Grab a debug location for the inserted loads/stores; given that the
+ // inserted loads/stores have little relation to the original loads/stores,
+ // this code just arbitrarily picks a location from one, since any debug
+ // location is better than none.
+ DebugLoc DL = LoopUses[0]->getDebugLoc();
+
+ // Figure out the loop exits and their insertion points, if this is the
+ // first promotion.
+ if (ExitBlocks.empty()) {
+ CurLoop->getUniqueExitBlocks(ExitBlocks);
+ InsertPts.resize(ExitBlocks.size());
+ for (unsigned i = 0, e = ExitBlocks.size(); i != e; ++i)
+ InsertPts[i] = ExitBlocks[i]->getFirstInsertionPt();
}
+
+ // We use the SSAUpdater interface to insert phi nodes as required.
+ SmallVector<PHINode*, 16> NewPHIs;
+ SSAUpdater SSA(&NewPHIs);
+ LoopPromoter Promoter(SomePtr, LoopUses, SSA, PointerMustAliases, ExitBlocks,
+ InsertPts, PIC, *CurAST, *LI, DL, Alignment, AATags);
+
+ // Set up the preheader to have a definition of the value. It is the live-out
+ // value from the preheader that uses in the loop will use.
+ LoadInst *PreheaderLoad =
+ new LoadInst(SomePtr, SomePtr->getName()+".promoted",
+ Preheader->getTerminator());
+ PreheaderLoad->setAlignment(Alignment);
+ PreheaderLoad->setDebugLoc(DL);
+ if (AATags) PreheaderLoad->setAAMetadata(AATags);
+ SSA.AddAvailableValue(Preheader, PreheaderLoad);
+
+ // Rewrite all the loads in the loop and remember all the definitions from
+ // stores in the loop.
+ Promoter.run(LoopUses);
+
+ // If the SSAUpdater didn't use the load in the preheader, just zap it now.
+ if (PreheaderLoad->use_empty())
+ PreheaderLoad->eraseFromParent();
}
+
/// cloneBasicBlockAnalysis - Simple Analysis hook. Clone alias set info.
void LICM::cloneBasicBlockAnalysis(BasicBlock *From, BasicBlock *To, Loop *L) {
- AliasSetTracker *AST = LoopToAliasMap[L];
+ AliasSetTracker *AST = LoopToAliasSetMap.lookup(L);
if (!AST)
return;
/// deleteAnalysisValue - Simple Analysis hook. Delete value V from alias
/// set.
void LICM::deleteAnalysisValue(Value *V, Loop *L) {
- AliasSetTracker *AST = LoopToAliasMap[L];
+ AliasSetTracker *AST = LoopToAliasSetMap.lookup(L);
if (!AST)
return;
projects / oota-llvm.git / blobdiff