/// store into the memory location pointed to by V.
///
bool pointerInvalidatedByLoop(Value *V, uint64_t Size,
- const MDNode *TBAAInfo) {
+ const AAMDNodes &AAInfo) {
// Check to see if any of the basic blocks in CurLoop invalidate *V.
- return CurAST->getAliasSetForPointer(V, Size, TBAAInfo).isMod();
+ return CurAST->getAliasSetForPointer(V, Size, AAInfo).isMod();
}
bool canSinkOrHoistInst(Instruction &I);
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);
};
}
uint64_t Size = 0;
if (LI->getType()->isSized())
Size = AA->getTypeStoreSize(LI->getType());
- return !pointerInvalidatedByLoop(LI->getOperand(0), Size,
- LI->getMetadata(LLVMContext::MD_tbaa));
+
+ AAMDNodes AAInfo;
+ LI->getAAMetadata(AAInfo);
+
+ return !pointerInvalidatedByLoop(LI->getOperand(0), Size, AAInfo);
} else if (CallInst *CI = dyn_cast<CallInst>(&I)) {
// Don't sink or hoist dbg info; it's legal, but not useful.
if (isa<DbgInfoIntrinsic>(I))
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,
/// this function moves it to the exit blocks and patches up SSA form as needed.
/// This method is guaranteed to remove the original instruction from its
Instruction *New;
auto It = SunkCopies.find(ExitBlock);
- if (It != SunkCopies.end()) {
+ if (It != SunkCopies.end())
New = It->second;
- } else {
- New = I.clone();
- SunkCopies[ExitBlock] = New;
- 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;
- }
- }
+ else
+ New = SunkCopies[ExitBlock] =
+ CloneInstructionInExitBlock(I, *ExitBlock, *PN);
PN->replaceAllUsesWith(New);
PN->eraseFromParent();
///
bool LICM::isSafeToExecuteUnconditionally(Instruction &Inst) {
// If it is not a trapping instruction, it is always safe to hoist.
- if (isSafeToSpeculativelyExecute(&Inst))
+ if (isSafeToSpeculativelyExecute(&Inst, DL))
return true;
return isGuaranteedToExecute(Inst);
LoopInfo &LI;
DebugLoc DL;
int Alignment;
- MDNode *TBAATag;
+ AAMDNodes AATags;
Value *maybeInsertLCSSAPHI(Value *V, BasicBlock *BB) const {
if (Instruction *I = dyn_cast<Instruction>(V))
SmallVectorImpl<BasicBlock *> &LEB,
SmallVectorImpl<Instruction *> &LIP, PredIteratorCache &PIC,
AliasSetTracker &ast, LoopInfo &li, DebugLoc dl, int alignment,
- MDNode *TBAATag)
+ const AAMDNodes &AATags)
: LoadAndStorePromoter(Insts, S), SomePtr(SP), PointerMustAliases(PMA),
LoopExitBlocks(LEB), LoopInsertPts(LIP), PredCache(PIC), AST(ast),
- LI(li), DL(dl), Alignment(alignment), TBAATag(TBAATag) {}
+ LI(li), DL(dl), Alignment(alignment), AATags(AATags) {}
bool isInstInList(Instruction *I,
const SmallVectorImpl<Instruction*> &) const override {
StoreInst *NewSI = new StoreInst(LiveInValue, Ptr, InsertPos);
NewSI->setAlignment(Alignment);
NewSI->setDebugLoc(DL);
- if (TBAATag) NewSI->setMetadata(LLVMContext::MD_tbaa, TBAATag);
+ if (AATags) NewSI->setAAMetadata(AATags);
}
}
// We start with an alignment of one and try to find instructions that allow
// us to prove better alignment.
unsigned Alignment = 1;
- MDNode *TBAATag = nullptr;
+ AAMDNodes AATags;
// 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 TBAA info.
+ // 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);
} else
return; // Not a load or store.
- // Merge the TBAA tags.
+ // Merge the AA tags.
if (LoopUses.empty()) {
- // On the first load/store, just take its TBAA tag.
- TBAATag = UI->getMetadata(LLVMContext::MD_tbaa);
- } else if (TBAATag) {
- TBAATag = MDNode::getMostGenericTBAA(TBAATag,
- UI->getMetadata(LLVMContext::MD_tbaa));
+ // 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);
SmallVector<PHINode*, 16> NewPHIs;
SSAUpdater SSA(&NewPHIs);
LoopPromoter Promoter(SomePtr, LoopUses, SSA, PointerMustAliases, ExitBlocks,
- InsertPts, PIC, *CurAST, *LI, DL, Alignment, TBAATag);
+ 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.
Preheader->getTerminator());
PreheaderLoad->setAlignment(Alignment);
PreheaderLoad->setDebugLoc(DL);
- if (TBAATag) PreheaderLoad->setMetadata(LLVMContext::MD_tbaa, TBAATag);
+ if (AATags) PreheaderLoad->setAAMetadata(AATags);
SSA.AddAvailableValue(Preheader, PreheaderLoad);
// Rewrite all the loads in the loop and remember all the definitions from