#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/LLVMContext.h"
+#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/AliasSetTracker.h"
#include "llvm/Analysis/ConstantFolding.h"
+#include "llvm/Analysis/Dominators.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/LoopPass.h"
-#include "llvm/Analysis/Dominators.h"
-#include "llvm/Transforms/Utils/Local.h"
-#include "llvm/Transforms/Utils/SSAUpdater.h"
+#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/LLVMContext.h"
#include "llvm/Support/CFG.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/SSAUpdater.h"
#include <algorithm>
using namespace llvm;
AU.addPreserved<AliasAnalysis>();
AU.addPreserved("scalar-evolution");
AU.addPreservedID(LoopSimplifyID);
+ AU.addRequired<TargetLibraryInfo>();
}
+ using llvm::Pass::doFinalization;
+
bool doFinalization() {
assert(LoopToAliasSetMap.empty() && "Didn't free loop alias sets");
return false;
LoopInfo *LI; // Current LoopInfo
DominatorTree *DT; // Dominator Tree for the current Loop.
+ DataLayout *TD; // DataLayout for constant folding.
+ TargetLibraryInfo *TLI; // TargetLibraryInfo for constant folding.
+
// 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...
+ 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.
bool canSinkOrHoistInst(Instruction &I);
bool isNotUsedInLoop(Instruction &I);
- void PromoteAliasSet(AliasSet &AS);
+ void PromoteAliasSet(AliasSet &AS,
+ SmallVectorImpl<BasicBlock*> &ExitBlocks,
+ SmallVectorImpl<Instruction*> &InsertPts);
};
}
INITIALIZE_PASS_DEPENDENCY(DominatorTree)
INITIALIZE_PASS_DEPENDENCY(LoopInfo)
INITIALIZE_PASS_DEPENDENCY(LoopSimplify)
+INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfo)
INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
INITIALIZE_PASS_END(LICM, "licm", "Loop Invariant Code Motion", false, false)
AA = &getAnalysis<AliasAnalysis>();
DT = &getAnalysis<DominatorTree>();
+ TD = getAnalysisIfAvailable<DataLayout>();
+ TLI = &getAnalysis<TargetLibraryInfo>();
+
CurAST = new AliasSetTracker(*AA);
// Collect Alias info from subloops.
for (Loop::iterator LoopItr = L->begin(), LoopItrE = L->end();
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
// 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()) {
+ SmallVector<BasicBlock *, 8> ExitBlocks;
+ SmallVector<Instruction *, 8> InsertPts;
+
// 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);
+ PromoteAliasSet(*I, ExitBlocks, InsertPts);
}
// Clear out loops state information for the next iteration
// 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)) {
+ if (isInstructionTriviallyDead(&I, TLI)) {
DEBUG(dbgs() << "LICM deleting dead inst: " << I << '\n');
++II;
CurAST->deleteValue(&I);
// 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)) {
+ if (Constant *C = ConstantFoldInstruction(&I, TD, TLI)) {
DEBUG(dbgs() << "LICM folding inst: " << I << " --> " << *C << '\n');
CurAST->copyValue(&I, C);
CurAST->deleteValue(&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.
uint64_t Size = 0;
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.
+ bool HoistableKind = (isa<BinaryOperator>(I) || isa<CastInst>(I) ||
+ isa<SelectInst>(I) || isa<GetElementPtrInst>(I) ||
+ isa<CmpInst>(I) || isa<InsertElementInst>(I) ||
+ isa<ExtractElementInst>(I) ||
+ isa<ShuffleVectorInst>(I));
+ if (!HoistableKind)
+ return false;
+
+ return isSafeToExecuteUnconditionally(I);
}
/// isNotUsedInLoop - Return true if the only users of this instruction are
} else {
// Move the instruction to the start of the exit block, after any PHI
// nodes in it.
- I.moveBefore(ExitBlocks[0]->getFirstNonPHI());
+ I.moveBefore(ExitBlocks[0]->getFirstInsertionPt());
// This instruction is no longer in the AST for the current loop, because
// we just sunk it out of the loop. If we just sunk it into an outer
continue;
// Insert the code after the last PHI node.
- BasicBlock::iterator InsertPt = ExitBlock->getFirstNonPHI();
+ BasicBlock::iterator InsertPt = ExitBlock->getFirstInsertionPt();
// If this is the first exit block processed, just move the original
// instruction, otherwise clone the original instruction and insert
///
bool LICM::isSafeToExecuteUnconditionally(Instruction &Inst) {
// If it is not a trapping instruction, it is always safe to hoist.
- if (Inst.isSafeToSpeculativelyExecute())
+ if (isSafeToSpeculativelyExecute(&Inst))
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.
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;
}
Value *SomePtr; // Designated pointer to store to.
SmallPtrSet<Value*, 4> &PointerMustAliases;
SmallVectorImpl<BasicBlock*> &LoopExitBlocks;
+ SmallVectorImpl<Instruction*> &LoopInsertPts;
AliasSetTracker &AST;
DebugLoc DL;
int Alignment;
+ MDNode *TBAATag;
public:
LoopPromoter(Value *SP,
const SmallVectorImpl<Instruction*> &Insts, SSAUpdater &S,
SmallPtrSet<Value*, 4> &PMA,
- SmallVectorImpl<BasicBlock*> &LEB, AliasSetTracker &ast,
- DebugLoc dl, int alignment)
+ SmallVectorImpl<BasicBlock*> &LEB,
+ SmallVectorImpl<Instruction*> &LIP,
+ AliasSetTracker &ast, DebugLoc dl, int alignment,
+ MDNode *TBAATag)
: LoadAndStorePromoter(Insts, S), SomePtr(SP),
- PointerMustAliases(PMA), LoopExitBlocks(LEB), AST(ast), DL(dl),
- Alignment(alignment) {}
+ PointerMustAliases(PMA), LoopExitBlocks(LEB), LoopInsertPts(LIP),
+ AST(ast), DL(dl), Alignment(alignment), TBAATag(TBAATag) {}
virtual bool isInstInList(Instruction *I,
const SmallVectorImpl<Instruction*> &) const {
for (unsigned i = 0, e = LoopExitBlocks.size(); i != e; ++i) {
BasicBlock *ExitBlock = LoopExitBlocks[i];
Value *LiveInValue = SSA.GetValueInMiddleOfBlock(ExitBlock);
- Instruction *InsertPos = ExitBlock->getFirstNonPHI();
+ Instruction *InsertPos = LoopInsertPts[i];
StoreInst *NewSI = new StoreInst(LiveInValue, SomePtr, InsertPos);
NewSI->setAlignment(Alignment);
NewSI->setDebugLoc(DL);
+ if (TBAATag) NewSI->setMetadata(LLVMContext::MD_tbaa, TBAATag);
}
}
/// looping over the stores in the loop, looking for stores to Must pointers
/// which are loop invariant.
///
-void LICM::PromoteAliasSet(AliasSet &AS) {
+void LICM::PromoteAliasSet(AliasSet &AS,
+ SmallVectorImpl<BasicBlock*> &ExitBlocks,
+ SmallVectorImpl<Instruction*> &InsertPts) {
// 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.
// We start with an alignment of one and try to find instructions that allow
// us to prove better alignment.
unsigned Alignment = 1;
+ MDNode *TBAATag = 0;
// 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.
+ // different sizes. While we are at it, collect alignment and TBAA info.
for (AliasSet::iterator ASI = AS.begin(), E = AS.end(); ASI != E; ++ASI) {
Value *ASIV = ASI->getValue();
PointerMustAliases.insert(ASIV);
// If there is an non-load/store instruction in the loop, we can't promote
// it.
- if (isa<LoadInst>(Use)) {
- assert(!cast<LoadInst>(Use)->isVolatile() && "AST broken");
+ if (LoadInst *load = dyn_cast<LoadInst>(Use)) {
+ assert(!load->isVolatile() && "AST broken");
+ if (!load->isSimple())
+ return;
} else if (StoreInst *store = dyn_cast<StoreInst>(Use)) {
// Stores *of* the pointer are not interesting, only stores *to* the
// pointer.
if (Use->getOperand(1) != ASIV)
continue;
- unsigned InstAlignment = store->getAlignment();
- assert(!cast<StoreInst>(Use)->isVolatile() && "AST broken");
+ 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
// 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.
- if ((InstAlignment > Alignment || InstAlignment == 0)
- && (Alignment != 0))
+ unsigned InstAlignment = store->getAlignment();
+ if ((InstAlignment > Alignment || InstAlignment == 0) && Alignment != 0)
if (isGuaranteedToExecute(*Use)) {
GuaranteedToExecute = true;
Alignment = InstAlignment;
} else
return; // Not a load or store.
+ // Merge the TBAA tags.
+ if (LoopUses.empty()) {
+ // On the first load/store, just take its TBAA tag.
+ TBAATag = Use->getMetadata(LLVMContext::MD_tbaa);
+ } else if (TBAATag && TBAATag != Use->getMetadata(LLVMContext::MD_tbaa)) {
+ TBAATag = 0;
+ }
+
LoopUses.push_back(Use);
}
}
// location is better than none.
DebugLoc DL = LoopUses[0]->getDebugLoc();
- SmallVector<BasicBlock*, 8> ExitBlocks;
- CurLoop->getUniqueExitBlocks(ExitBlocks);
+ // 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,
- *CurAST, DL, Alignment);
+ InsertPts, *CurAST, DL, Alignment, TBAATag);
// 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.
Preheader->getTerminator());
PreheaderLoad->setAlignment(Alignment);
PreheaderLoad->setDebugLoc(DL);
+ if (TBAATag) PreheaderLoad->setMetadata(LLVMContext::MD_tbaa, TBAATag);
SSA.AddAvailableValue(Preheader, PreheaderLoad);
// Rewrite all the loads in the loop and remember all the definitions from