#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Analysis/LazyValueInfo.h"
#include "llvm/Analysis/Loads.h"
+#include "llvm/Analysis/TargetLibraryInfo.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/Metadata.h"
#include "llvm/IR/ValueHandle.h"
#include "llvm/Pass.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
-#include "llvm/Target/TargetLibraryInfo.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Transforms/Utils/Local.h"
#include "llvm/Transforms/Utils/SSAUpdater.h"
STATISTIC(NumDupes, "Number of branch blocks duplicated to eliminate phi");
static cl::opt<unsigned>
-Threshold("jump-threading-threshold",
+BBDuplicateThreshold("jump-threading-threshold",
cl::desc("Max block size to duplicate for jump threading"),
cl::init(6), cl::Hidden);
/// revectored to the false side of the second if.
///
class JumpThreading : public FunctionPass {
- const DataLayout *DL;
TargetLibraryInfo *TLI;
LazyValueInfo *LVI;
#ifdef NDEBUG
#endif
DenseSet<std::pair<Value*, BasicBlock*> > RecursionSet;
+ unsigned BBDupThreshold;
+
// RAII helper for updating the recursion stack.
struct RecursionSetRemover {
DenseSet<std::pair<Value*, BasicBlock*> > &TheSet;
};
public:
static char ID; // Pass identification
- JumpThreading() : FunctionPass(ID) {
+ JumpThreading(int T = -1) : FunctionPass(ID) {
+ BBDupThreshold = (T == -1) ? BBDuplicateThreshold : unsigned(T);
initializeJumpThreadingPass(*PassRegistry::getPassRegistry());
}
void getAnalysisUsage(AnalysisUsage &AU) const override {
AU.addRequired<LazyValueInfo>();
AU.addPreserved<LazyValueInfo>();
- AU.addRequired<TargetLibraryInfo>();
+ AU.addRequired<TargetLibraryInfoWrapperPass>();
}
void FindLoopHeaders(Function &F);
bool ComputeValueKnownInPredecessors(Value *V, BasicBlock *BB,
PredValueInfo &Result,
- ConstantPreference Preference);
+ ConstantPreference Preference,
+ Instruction *CxtI = nullptr);
bool ProcessThreadableEdges(Value *Cond, BasicBlock *BB,
- ConstantPreference Preference);
+ ConstantPreference Preference,
+ Instruction *CxtI = nullptr);
bool ProcessBranchOnPHI(PHINode *PN);
bool ProcessBranchOnXOR(BinaryOperator *BO);
INITIALIZE_PASS_BEGIN(JumpThreading, "jump-threading",
"Jump Threading", false, false)
INITIALIZE_PASS_DEPENDENCY(LazyValueInfo)
-INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfo)
+INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass)
INITIALIZE_PASS_END(JumpThreading, "jump-threading",
"Jump Threading", false, false)
// Public interface to the Jump Threading pass
-FunctionPass *llvm::createJumpThreadingPass() { return new JumpThreading(); }
+FunctionPass *llvm::createJumpThreadingPass(int Threshold) { return new JumpThreading(Threshold); }
/// runOnFunction - Top level algorithm.
///
return false;
DEBUG(dbgs() << "Jump threading on function '" << F.getName() << "'\n");
- DataLayoutPass *DLP = getAnalysisIfAvailable<DataLayoutPass>();
- DL = DLP ? &DLP->getDataLayout() : nullptr;
- TLI = &getAnalysis<TargetLibraryInfo>();
+ TLI = &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI();
LVI = &getAnalysis<LazyValueInfo>();
// Remove unreachable blocks from function as they may result in infinite
// If the block is trivially dead, zap it. This eliminates the successor
// edges which simplifies the CFG.
- if (pred_begin(BB) == pred_end(BB) &&
+ if (pred_empty(BB) &&
BB != &BB->getParent()->getEntryBlock()) {
DEBUG(dbgs() << " JT: Deleting dead block '" << BB->getName()
<< "' with terminator: " << *BB->getTerminator() << '\n');
///
bool JumpThreading::
ComputeValueKnownInPredecessors(Value *V, BasicBlock *BB, PredValueInfo &Result,
- ConstantPreference Preference) {
+ ConstantPreference Preference,
+ Instruction *CxtI) {
// This method walks up use-def chains recursively. Because of this, we could
// get into an infinite loop going around loops in the use-def chain. To
// prevent this, keep track of what (value, block) pairs we've already visited
BasicBlock *P = *PI;
// If the value is known by LazyValueInfo to be a constant in a
// predecessor, use that information to try to thread this block.
- Constant *PredCst = LVI->getConstantOnEdge(V, P, BB);
+ Constant *PredCst = LVI->getConstantOnEdge(V, P, BB, CxtI);
if (Constant *KC = getKnownConstant(PredCst, Preference))
Result.push_back(std::make_pair(KC, P));
}
Result.push_back(std::make_pair(KC, PN->getIncomingBlock(i)));
} else {
Constant *CI = LVI->getConstantOnEdge(InVal,
- PN->getIncomingBlock(i), BB);
+ PN->getIncomingBlock(i),
+ BB, CxtI);
if (Constant *KC = getKnownConstant(CI, Preference))
Result.push_back(std::make_pair(KC, PN->getIncomingBlock(i)));
}
if (I->getOpcode() == Instruction::Or ||
I->getOpcode() == Instruction::And) {
ComputeValueKnownInPredecessors(I->getOperand(0), BB, LHSVals,
- WantInteger);
+ WantInteger, CxtI);
ComputeValueKnownInPredecessors(I->getOperand(1), BB, RHSVals,
- WantInteger);
+ WantInteger, CxtI);
if (LHSVals.empty() && RHSVals.empty())
return false;
isa<ConstantInt>(I->getOperand(1)) &&
cast<ConstantInt>(I->getOperand(1))->isOne()) {
ComputeValueKnownInPredecessors(I->getOperand(0), BB, Result,
- WantInteger);
+ WantInteger, CxtI);
if (Result.empty())
return false;
if (ConstantInt *CI = dyn_cast<ConstantInt>(BO->getOperand(1))) {
PredValueInfoTy LHSVals;
ComputeValueKnownInPredecessors(BO->getOperand(0), BB, LHSVals,
- WantInteger);
+ WantInteger, CxtI);
// Try to use constant folding to simplify the binary operator.
for (unsigned i = 0, e = LHSVals.size(); i != e; ++i) {
assert(Preference == WantInteger && "Compares only produce integers");
PHINode *PN = dyn_cast<PHINode>(Cmp->getOperand(0));
if (PN && PN->getParent() == BB) {
+ const DataLayout &DL = PN->getModule()->getDataLayout();
// We can do this simplification if any comparisons fold to true or false.
// See if any do.
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) {
LazyValueInfo::Tristate
ResT = LVI->getPredicateOnEdge(Cmp->getPredicate(), LHS,
- cast<Constant>(RHS), PredBB, BB);
+ cast<Constant>(RHS), PredBB, BB,
+ CxtI ? CxtI : Cmp);
if (ResT == LazyValueInfo::Unknown)
continue;
Res = ConstantInt::get(Type::getInt1Ty(LHS->getContext()), ResT);
return !Result.empty();
}
-
// If comparing a live-in value against a constant, see if we know the
// live-in value on any predecessors.
if (isa<Constant>(Cmp->getOperand(1)) && Cmp->getType()->isIntegerTy()) {
// predecessor, use that information to try to thread this block.
LazyValueInfo::Tristate Res =
LVI->getPredicateOnEdge(Cmp->getPredicate(), Cmp->getOperand(0),
- RHSCst, P, BB);
+ RHSCst, P, BB, CxtI ? CxtI : Cmp);
if (Res == LazyValueInfo::Unknown)
continue;
if (Constant *CmpConst = dyn_cast<Constant>(Cmp->getOperand(1))) {
PredValueInfoTy LHSVals;
ComputeValueKnownInPredecessors(I->getOperand(0), BB, LHSVals,
- WantInteger);
+ WantInteger, CxtI);
for (unsigned i = 0, e = LHSVals.size(); i != e; ++i) {
Constant *V = LHSVals[i].first;
PredValueInfoTy Conds;
if ((TrueVal || FalseVal) &&
ComputeValueKnownInPredecessors(SI->getCondition(), BB, Conds,
- WantInteger)) {
+ WantInteger, CxtI)) {
for (unsigned i = 0, e = Conds.size(); i != e; ++i) {
Constant *Cond = Conds[i].first;
}
// If all else fails, see if LVI can figure out a constant value for us.
- Constant *CI = LVI->getConstant(V, BB);
+ Constant *CI = LVI->getConstant(V, BB, CxtI);
if (Constant *KC = getKnownConstant(CI, Preference)) {
for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI)
Result.push_back(std::make_pair(KC, *PI));
bool JumpThreading::ProcessBlock(BasicBlock *BB) {
// If the block is trivially dead, just return and let the caller nuke it.
// This simplifies other transformations.
- if (pred_begin(BB) == pred_end(BB) &&
+ if (pred_empty(BB) &&
BB != &BB->getParent()->getEntryBlock())
return false;
if (LoopHeaders.erase(SinglePred))
LoopHeaders.insert(BB);
- // Remember if SinglePred was the entry block of the function. If so, we
- // will need to move BB back to the entry position.
- bool isEntry = SinglePred == &SinglePred->getParent()->getEntryBlock();
LVI->eraseBlock(SinglePred);
MergeBasicBlockIntoOnlyPred(BB);
- if (isEntry && BB != &BB->getParent()->getEntryBlock())
- BB->moveBefore(&BB->getParent()->getEntryBlock());
return true;
}
}
// Run constant folding to see if we can reduce the condition to a simple
// constant.
if (Instruction *I = dyn_cast<Instruction>(Condition)) {
- Value *SimpleVal = ConstantFoldInstruction(I, DL, TLI);
+ Value *SimpleVal =
+ ConstantFoldInstruction(I, BB->getModule()->getDataLayout(), TLI);
if (SimpleVal) {
I->replaceAllUsesWith(SimpleVal);
I->eraseFromParent();
// All the rest of our checks depend on the condition being an instruction.
if (!CondInst) {
// FIXME: Unify this with code below.
- if (ProcessThreadableEdges(Condition, BB, Preference))
+ if (ProcessThreadableEdges(Condition, BB, Preference, Terminator))
return true;
return false;
}
if (CmpInst *CondCmp = dyn_cast<CmpInst>(CondInst)) {
- // For a comparison where the LHS is outside this block, it's possible
- // that we've branched on it before. Used LVI to see if we can simplify
- // the branch based on that.
+ // If we're branching on a conditional, LVI might be able to determine
+ // it's value at the the branch instruction. We only handle comparisons
+ // against a constant at this time.
+ // TODO: This should be extended to handle switches as well.
BranchInst *CondBr = dyn_cast<BranchInst>(BB->getTerminator());
Constant *CondConst = dyn_cast<Constant>(CondCmp->getOperand(1));
- pred_iterator PI = pred_begin(BB), PE = pred_end(BB);
- if (CondBr && CondConst && CondBr->isConditional() && PI != PE &&
- (!isa<Instruction>(CondCmp->getOperand(0)) ||
- cast<Instruction>(CondCmp->getOperand(0))->getParent() != BB)) {
- // For predecessor edge, determine if the comparison is true or false
- // on that edge. If they're all true or all false, we can simplify the
- // branch.
- // FIXME: We could handle mixed true/false by duplicating code.
- LazyValueInfo::Tristate Baseline =
- LVI->getPredicateOnEdge(CondCmp->getPredicate(), CondCmp->getOperand(0),
- CondConst, *PI, BB);
- if (Baseline != LazyValueInfo::Unknown) {
- // Check that all remaining incoming values match the first one.
- while (++PI != PE) {
- LazyValueInfo::Tristate Ret =
- LVI->getPredicateOnEdge(CondCmp->getPredicate(),
- CondCmp->getOperand(0), CondConst, *PI, BB);
- if (Ret != Baseline) break;
- }
-
- // If we terminated early, then one of the values didn't match.
- if (PI == PE) {
- unsigned ToRemove = Baseline == LazyValueInfo::True ? 1 : 0;
- unsigned ToKeep = Baseline == LazyValueInfo::True ? 0 : 1;
- CondBr->getSuccessor(ToRemove)->removePredecessor(BB, true);
- BranchInst::Create(CondBr->getSuccessor(ToKeep), CondBr);
- CondBr->eraseFromParent();
- return true;
+ if (CondBr && CondConst && CondBr->isConditional()) {
+ LazyValueInfo::Tristate Ret =
+ LVI->getPredicateAt(CondCmp->getPredicate(), CondCmp->getOperand(0),
+ CondConst, CondBr);
+ if (Ret != LazyValueInfo::Unknown) {
+ unsigned ToRemove = Ret == LazyValueInfo::True ? 1 : 0;
+ unsigned ToKeep = Ret == LazyValueInfo::True ? 0 : 1;
+ CondBr->getSuccessor(ToRemove)->removePredecessor(BB, true);
+ BranchInst::Create(CondBr->getSuccessor(ToKeep), CondBr);
+ CondBr->eraseFromParent();
+ if (CondCmp->use_empty())
+ CondCmp->eraseFromParent();
+ else if (CondCmp->getParent() == BB) {
+ // If the fact we just learned is true for all uses of the
+ // condition, replace it with a constant value
+ auto *CI = Ret == LazyValueInfo::True ?
+ ConstantInt::getTrue(CondCmp->getType()) :
+ ConstantInt::getFalse(CondCmp->getType());
+ CondCmp->replaceAllUsesWith(CI);
+ CondCmp->eraseFromParent();
}
+ return true;
}
-
}
if (CondBr && CondConst && TryToUnfoldSelect(CondCmp, BB))
// a PHI node in the current block. If we can prove that any predecessors
// compute a predictable value based on a PHI node, thread those predecessors.
//
- if (ProcessThreadableEdges(CondInst, BB, Preference))
+ if (ProcessThreadableEdges(CondInst, BB, Preference, Terminator))
return true;
// If this is an otherwise-unfoldable branch on a phi node in the current
// If the returned value is the load itself, replace with an undef. This can
// only happen in dead loops.
if (AvailableVal == LI) AvailableVal = UndefValue::get(LI->getType());
+ if (AvailableVal->getType() != LI->getType())
+ AvailableVal =
+ CastInst::CreateBitOrPointerCast(AvailableVal, LI->getType(), "", LI);
LI->replaceAllUsesWith(AvailableVal);
LI->eraseFromParent();
return true;
if (BBIt != LoadBB->begin())
return false;
- // If all of the loads and stores that feed the value have the same TBAA tag,
- // then we can propagate it onto any newly inserted loads.
- MDNode *TBAATag = LI->getMetadata(LLVMContext::MD_tbaa);
+ // If all of the loads and stores that feed the value have the same AA tags,
+ // then we can propagate them onto any newly inserted loads.
+ AAMDNodes AATags;
+ LI->getAAMetadata(AATags);
SmallPtrSet<BasicBlock*, 8> PredsScanned;
typedef SmallVector<std::pair<BasicBlock*, Value*>, 8> AvailablePredsTy;
BasicBlock *PredBB = *PI;
// If we already scanned this predecessor, skip it.
- if (!PredsScanned.insert(PredBB))
+ if (!PredsScanned.insert(PredBB).second)
continue;
// Scan the predecessor to see if the value is available in the pred.
BBIt = PredBB->end();
- MDNode *ThisTBAATag = nullptr;
+ AAMDNodes ThisAATags;
Value *PredAvailable = FindAvailableLoadedValue(LoadedPtr, PredBB, BBIt, 6,
- nullptr, &ThisTBAATag);
+ nullptr, &ThisAATags);
if (!PredAvailable) {
OneUnavailablePred = PredBB;
continue;
}
- // If tbaa tags disagree or are not present, forget about them.
- if (TBAATag != ThisTBAATag) TBAATag = nullptr;
+ // If AA tags disagree or are not present, forget about them.
+ if (AATags != ThisAATags) AATags = AAMDNodes();
// If so, this load is partially redundant. Remember this info so that we
// can create a PHI node.
// Split them out to their own block.
UnavailablePred =
- SplitBlockPredecessors(LoadBB, PredsToSplit, "thread-pre-split", this);
+ SplitBlockPredecessors(LoadBB, PredsToSplit, "thread-pre-split");
}
// If the value isn't available in all predecessors, then there will be
LI->getAlignment(),
UnavailablePred->getTerminator());
NewVal->setDebugLoc(LI->getDebugLoc());
- if (TBAATag)
- NewVal->setMetadata(LLVMContext::MD_tbaa, TBAATag);
+ if (AATags)
+ NewVal->setAAMetadata(AATags);
AvailablePreds.push_back(std::make_pair(UnavailablePred, NewVal));
}
assert(I != AvailablePreds.end() && I->first == P &&
"Didn't find entry for predecessor!");
- PN->addIncoming(I->second, I->first);
+ // If we have an available predecessor but it requires casting, insert the
+ // cast in the predecessor and use the cast. Note that we have to update the
+ // AvailablePreds vector as we go so that all of the PHI entries for this
+ // predecessor use the same bitcast.
+ Value *&PredV = I->second;
+ if (PredV->getType() != LI->getType())
+ PredV = CastInst::CreateBitOrPointerCast(PredV, LI->getType(), "",
+ P->getTerminator());
+
+ PN->addIncoming(PredV, I->first);
}
//cerr << "PRE: " << *LI << *PN << "\n";
}
bool JumpThreading::ProcessThreadableEdges(Value *Cond, BasicBlock *BB,
- ConstantPreference Preference) {
+ ConstantPreference Preference,
+ Instruction *CxtI) {
// If threading this would thread across a loop header, don't even try to
// thread the edge.
if (LoopHeaders.count(BB))
return false;
PredValueInfoTy PredValues;
- if (!ComputeValueKnownInPredecessors(Cond, BB, PredValues, Preference))
+ if (!ComputeValueKnownInPredecessors(Cond, BB, PredValues, Preference, CxtI))
return false;
assert(!PredValues.empty() &&
for (unsigned i = 0, e = PredValues.size(); i != e; ++i) {
BasicBlock *Pred = PredValues[i].second;
- if (!SeenPreds.insert(Pred))
+ if (!SeenPreds.insert(Pred).second)
continue; // Duplicate predecessor entry.
// If the predecessor ends with an indirect goto, we can't change its
PredValueInfoTy XorOpValues;
bool isLHS = true;
if (!ComputeValueKnownInPredecessors(BO->getOperand(0), BB, XorOpValues,
- WantInteger)) {
+ WantInteger, BO)) {
assert(XorOpValues.empty());
if (!ComputeValueKnownInPredecessors(BO->getOperand(1), BB, XorOpValues,
- WantInteger))
+ WantInteger, BO))
return false;
isLHS = false;
}
return false;
}
- unsigned JumpThreadCost = getJumpThreadDuplicationCost(BB, Threshold);
- if (JumpThreadCost > Threshold) {
+ unsigned JumpThreadCost = getJumpThreadDuplicationCost(BB, BBDupThreshold);
+ if (JumpThreadCost > BBDupThreshold) {
DEBUG(dbgs() << " Not threading BB '" << BB->getName()
<< "' - Cost is too high: " << JumpThreadCost << "\n");
return false;
else {
DEBUG(dbgs() << " Factoring out " << PredBBs.size()
<< " common predecessors.\n");
- PredBB = SplitBlockPredecessors(BB, PredBBs, ".thr_comm", this);
+ PredBB = SplitBlockPredecessors(BB, PredBBs, ".thr_comm");
}
// And finally, do it!
// At this point, the IR is fully up to date and consistent. Do a quick scan
// over the new instructions and zap any that are constants or dead. This
// frequently happens because of phi translation.
- SimplifyInstructionsInBlock(NewBB, DL, TLI);
+ SimplifyInstructionsInBlock(NewBB, TLI);
// Threaded an edge!
++NumThreads;
return false;
}
- unsigned DuplicationCost = getJumpThreadDuplicationCost(BB, Threshold);
- if (DuplicationCost > Threshold) {
+ unsigned DuplicationCost = getJumpThreadDuplicationCost(BB, BBDupThreshold);
+ if (DuplicationCost > BBDupThreshold) {
DEBUG(dbgs() << " Not duplicating BB '" << BB->getName()
<< "' - Cost is too high: " << DuplicationCost << "\n");
return false;
else {
DEBUG(dbgs() << " Factoring out " << PredBBs.size()
<< " common predecessors.\n");
- PredBB = SplitBlockPredecessors(BB, PredBBs, ".thr_comm", this);
+ PredBB = SplitBlockPredecessors(BB, PredBBs, ".thr_comm");
}
// Okay, we decided to do this! Clone all the instructions in BB onto the end
BranchInst *OldPredBranch = dyn_cast<BranchInst>(PredBB->getTerminator());
if (!OldPredBranch || !OldPredBranch->isUnconditional()) {
- PredBB = SplitEdge(PredBB, BB, this);
+ PredBB = SplitEdge(PredBB, BB);
OldPredBranch = cast<BranchInst>(PredBB->getTerminator());
}
BasicBlock::iterator BI = BB->begin();
for (; PHINode *PN = dyn_cast<PHINode>(BI); ++BI)
ValueMapping[PN] = PN->getIncomingValueForBlock(PredBB);
-
// Clone the non-phi instructions of BB into PredBB, keeping track of the
// mapping and using it to remap operands in the cloned instructions.
for (; BI != BB->end(); ++BI) {
// If this instruction can be simplified after the operands are updated,
// just use the simplified value instead. This frequently happens due to
// phi translation.
- if (Value *IV = SimplifyInstruction(New, DL)) {
+ if (Value *IV =
+ SimplifyInstruction(New, BB->getModule()->getDataLayout())) {
delete New;
ValueMapping[BI] = IV;
} else {
// cases will be threaded in any case.
LazyValueInfo::Tristate LHSFolds =
LVI->getPredicateOnEdge(CondCmp->getPredicate(), SI->getOperand(1),
- CondRHS, Pred, BB);
+ CondRHS, Pred, BB, CondCmp);
LazyValueInfo::Tristate RHSFolds =
LVI->getPredicateOnEdge(CondCmp->getPredicate(), SI->getOperand(2),
- CondRHS, Pred, BB);
+ CondRHS, Pred, BB, CondCmp);
if ((LHSFolds != LazyValueInfo::Unknown ||
RHSFolds != LazyValueInfo::Unknown) &&
LHSFolds != RHSFolds) {