X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTransforms%2FScalar%2FJumpThreading.cpp;h=e845a7f586bc3160143efb663980efaf7527a0f6;hb=4a45f0871a26e65a48a74dbb277c70f662344235;hp=84dbb1a4750d6520e434fbd8b73c8a8214f3ebd9;hpb=8e8cdc8fe03db3eb4bed463281d30a053b8770cc;p=oota-llvm.git diff --git a/lib/Transforms/Scalar/JumpThreading.cpp b/lib/Transforms/Scalar/JumpThreading.cpp index 84dbb1a4750..e845a7f586b 100644 --- a/lib/Transforms/Scalar/JumpThreading.cpp +++ b/lib/Transforms/Scalar/JumpThreading.cpp @@ -11,36 +11,41 @@ // //===----------------------------------------------------------------------===// -#define DEBUG_TYPE "jump-threading" #include "llvm/Transforms/Scalar.h" -#include "llvm/IntrinsicInst.h" -#include "llvm/LLVMContext.h" -#include "llvm/Pass.h" -#include "llvm/Analysis/InstructionSimplify.h" -#include "llvm/Analysis/LazyValueInfo.h" -#include "llvm/Analysis/Loads.h" -#include "llvm/Transforms/Utils/BasicBlockUtils.h" -#include "llvm/Transforms/Utils/Local.h" -#include "llvm/Transforms/Utils/SSAUpdater.h" -#include "llvm/Target/TargetData.h" #include "llvm/ADT/DenseMap.h" #include "llvm/ADT/DenseSet.h" -#include "llvm/ADT/Statistic.h" #include "llvm/ADT/STLExtras.h" #include "llvm/ADT/SmallPtrSet.h" #include "llvm/ADT/SmallSet.h" +#include "llvm/ADT/Statistic.h" +#include "llvm/Analysis/CFG.h" +#include "llvm/Analysis/ConstantFolding.h" +#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/ValueHandle.h" #include "llvm/Support/raw_ostream.h" +#include "llvm/Transforms/Utils/BasicBlockUtils.h" +#include "llvm/Transforms/Utils/Local.h" +#include "llvm/Transforms/Utils/SSAUpdater.h" using namespace llvm; +#define DEBUG_TYPE "jump-threading" + STATISTIC(NumThreads, "Number of jumps threaded"); STATISTIC(NumFolds, "Number of terminators folded"); STATISTIC(NumDupes, "Number of branch blocks duplicated to eliminate phi"); static cl::opt -Threshold("jump-threading-threshold", +BBDuplicateThreshold("jump-threading-threshold", cl::desc("Max block size to duplicate for jump threading"), cl::init(6), cl::Hidden); @@ -73,7 +78,7 @@ namespace { /// revectored to the false side of the second if. /// class JumpThreading : public FunctionPass { - TargetData *TD; + TargetLibraryInfo *TLI; LazyValueInfo *LVI; #ifdef NDEBUG SmallPtrSet LoopHeaders; @@ -82,6 +87,8 @@ namespace { #endif DenseSet > RecursionSet; + unsigned BBDupThreshold; + // RAII helper for updating the recursion stack. struct RecursionSetRemover { DenseSet > &TheSet; @@ -97,15 +104,17 @@ namespace { }; public: static char ID; // Pass identification - JumpThreading() : FunctionPass(ID) { + JumpThreading(int T = -1) : FunctionPass(ID) { + BBDupThreshold = (T == -1) ? BBDuplicateThreshold : unsigned(T); initializeJumpThreadingPass(*PassRegistry::getPassRegistry()); } - bool runOnFunction(Function &F); + bool runOnFunction(Function &F) override; - virtual void getAnalysisUsage(AnalysisUsage &AU) const { + void getAnalysisUsage(AnalysisUsage &AU) const override { AU.addRequired(); AU.addPreserved(); + AU.addRequired(); } void FindLoopHeaders(Function &F); @@ -117,14 +126,17 @@ namespace { 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); bool SimplifyPartiallyRedundantLoad(LoadInst *LI); + bool TryToUnfoldSelect(CmpInst *CondCmp, BasicBlock *BB); }; } @@ -132,19 +144,32 @@ char JumpThreading::ID = 0; INITIALIZE_PASS_BEGIN(JumpThreading, "jump-threading", "Jump Threading", false, false) INITIALIZE_PASS_DEPENDENCY(LazyValueInfo) +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. /// bool JumpThreading::runOnFunction(Function &F) { + if (skipOptnoneFunction(F)) + return false; + DEBUG(dbgs() << "Jump threading on function '" << F.getName() << "'\n"); - TD = getAnalysisIfAvailable(); + TLI = &getAnalysis().getTLI(); LVI = &getAnalysis(); + // Remove unreachable blocks from function as they may result in infinite + // loop. We do threading if we found something profitable. Jump threading a + // branch can create other opportunities. If these opportunities form a cycle + // i.e. if any jump treading is undoing previous threading in the path, then + // we will loop forever. We take care of this issue by not jump threading for + // back edges. This works for normal cases but not for unreachable blocks as + // they may have cycle with no back edge. + removeUnreachableBlocks(F); + FindLoopHeaders(F); bool Changed, EverChanged = false; @@ -160,7 +185,7 @@ bool JumpThreading::runOnFunction(Function &F) { // 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'); @@ -168,40 +193,38 @@ bool JumpThreading::runOnFunction(Function &F) { LVI->eraseBlock(BB); DeleteDeadBlock(BB); Changed = true; - } else if (BranchInst *BI = dyn_cast(BB->getTerminator())) { - // Can't thread an unconditional jump, but if the block is "almost - // empty", we can replace uses of it with uses of the successor and make - // this dead. - if (BI->isUnconditional() && - BB != &BB->getParent()->getEntryBlock()) { - BasicBlock::iterator BBI = BB->getFirstNonPHI(); - // Ignore dbg intrinsics. - while (isa(BBI)) - ++BBI; + continue; + } + + BranchInst *BI = dyn_cast(BB->getTerminator()); + + // Can't thread an unconditional jump, but if the block is "almost + // empty", we can replace uses of it with uses of the successor and make + // this dead. + if (BI && BI->isUnconditional() && + BB != &BB->getParent()->getEntryBlock() && // If the terminator is the only non-phi instruction, try to nuke it. - if (BBI->isTerminator()) { - // Since TryToSimplifyUncondBranchFromEmptyBlock may delete the - // block, we have to make sure it isn't in the LoopHeaders set. We - // reinsert afterward if needed. - bool ErasedFromLoopHeaders = LoopHeaders.erase(BB); - BasicBlock *Succ = BI->getSuccessor(0); - - // FIXME: It is always conservatively correct to drop the info - // for a block even if it doesn't get erased. This isn't totally - // awesome, but it allows us to use AssertingVH to prevent nasty - // dangling pointer issues within LazyValueInfo. - LVI->eraseBlock(BB); - if (TryToSimplifyUncondBranchFromEmptyBlock(BB)) { - Changed = true; - // If we deleted BB and BB was the header of a loop, then the - // successor is now the header of the loop. - BB = Succ; - } - - if (ErasedFromLoopHeaders) - LoopHeaders.insert(BB); - } + BB->getFirstNonPHIOrDbg()->isTerminator()) { + // Since TryToSimplifyUncondBranchFromEmptyBlock may delete the + // block, we have to make sure it isn't in the LoopHeaders set. We + // reinsert afterward if needed. + bool ErasedFromLoopHeaders = LoopHeaders.erase(BB); + BasicBlock *Succ = BI->getSuccessor(0); + + // FIXME: It is always conservatively correct to drop the info + // for a block even if it doesn't get erased. This isn't totally + // awesome, but it allows us to use AssertingVH to prevent nasty + // dangling pointer issues within LazyValueInfo. + LVI->eraseBlock(BB); + if (TryToSimplifyUncondBranchFromEmptyBlock(BB)) { + Changed = true; + // If we deleted BB and BB was the header of a loop, then the + // successor is now the header of the loop. + BB = Succ; } + + if (ErasedFromLoopHeaders) + LoopHeaders.insert(BB); } } EverChanged |= Changed; @@ -212,19 +235,24 @@ bool JumpThreading::runOnFunction(Function &F) { } /// getJumpThreadDuplicationCost - Return the cost of duplicating this block to -/// thread across it. -static unsigned getJumpThreadDuplicationCost(const BasicBlock *BB) { +/// thread across it. Stop scanning the block when passing the threshold. +static unsigned getJumpThreadDuplicationCost(const BasicBlock *BB, + unsigned Threshold) { /// Ignore PHI nodes, these will be flattened when duplication happens. BasicBlock::const_iterator I = BB->getFirstNonPHI(); // FIXME: THREADING will delete values that are just used to compute the // branch, so they shouldn't count against the duplication cost. - // Sum up the cost of each instruction until we get to the terminator. Don't // include the terminator because the copy won't include it. unsigned Size = 0; for (; !isa(I); ++I) { + + // Stop scanning the block if we've reached the threshold. + if (Size > Threshold) + return Size; + // Debugger intrinsics don't incur code size. if (isa(I)) continue; @@ -240,7 +268,11 @@ static unsigned getJumpThreadDuplicationCost(const BasicBlock *BB) { // as having cost of 2 total, and if they are a vector intrinsic, we model // them as having cost 1. if (const CallInst *CI = dyn_cast(I)) { - if (!isa(CI)) + if (CI->cannotDuplicate()) + // Blocks with NoDuplicate are modelled as having infinite cost, so they + // are never duplicated. + return ~0U; + else if (!isa(CI)) Size += 3; else if (!CI->getType()->isVectorTy()) Size += 1; @@ -289,7 +321,7 @@ void JumpThreading::FindLoopHeaders(Function &F) { /// Returns null if Val is null or not an appropriate constant. static Constant *getKnownConstant(Value *Val, ConstantPreference Preference) { if (!Val) - return 0; + return nullptr; // Undef is "known" enough. if (UndefValue *U = dyn_cast(Val)) @@ -310,7 +342,8 @@ static Constant *getKnownConstant(Value *Val, ConstantPreference Preference) { /// 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 @@ -333,7 +366,7 @@ ComputeValueKnownInPredecessors(Value *V, BasicBlock *BB, PredValueInfo &Result, // If V is a non-instruction value, or an instruction in a different block, // then it can't be derived from a PHI. Instruction *I = dyn_cast(V); - if (I == 0 || I->getParent() != BB) { + if (!I || I->getParent() != BB) { // Okay, if this is a live-in value, see if it has a known value at the end // of any of our predecessors. @@ -352,7 +385,7 @@ ComputeValueKnownInPredecessors(Value *V, BasicBlock *BB, PredValueInfo &Result, 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)); } @@ -368,7 +401,8 @@ ComputeValueKnownInPredecessors(Value *V, BasicBlock *BB, PredValueInfo &Result, 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))); } @@ -387,9 +421,9 @@ ComputeValueKnownInPredecessors(Value *V, BasicBlock *BB, PredValueInfo &Result, 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; @@ -430,7 +464,7 @@ ComputeValueKnownInPredecessors(Value *V, BasicBlock *BB, PredValueInfo &Result, isa(I->getOperand(1)) && cast(I->getOperand(1))->isOne()) { ComputeValueKnownInPredecessors(I->getOperand(0), BB, Result, - WantInteger); + WantInteger, CxtI); if (Result.empty()) return false; @@ -448,7 +482,7 @@ ComputeValueKnownInPredecessors(Value *V, BasicBlock *BB, PredValueInfo &Result, if (ConstantInt *CI = dyn_cast(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) { @@ -468,6 +502,7 @@ ComputeValueKnownInPredecessors(Value *V, BasicBlock *BB, PredValueInfo &Result, assert(Preference == WantInteger && "Compares only produce integers"); PHINode *PN = dyn_cast(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) { @@ -475,14 +510,15 @@ ComputeValueKnownInPredecessors(Value *V, BasicBlock *BB, PredValueInfo &Result, Value *LHS = PN->getIncomingValue(i); Value *RHS = Cmp->getOperand(1)->DoPHITranslation(BB, PredBB); - Value *Res = SimplifyCmpInst(Cmp->getPredicate(), LHS, RHS, TD); - if (Res == 0) { + Value *Res = SimplifyCmpInst(Cmp->getPredicate(), LHS, RHS, DL); + if (!Res) { if (!isa(RHS)) continue; LazyValueInfo::Tristate ResT = LVI->getPredicateOnEdge(Cmp->getPredicate(), LHS, - cast(RHS), PredBB, BB); + cast(RHS), PredBB, BB, + CxtI ? CxtI : Cmp); if (ResT == LazyValueInfo::Unknown) continue; Res = ConstantInt::get(Type::getInt1Ty(LHS->getContext()), ResT); @@ -495,7 +531,6 @@ ComputeValueKnownInPredecessors(Value *V, BasicBlock *BB, PredValueInfo &Result, 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(Cmp->getOperand(1)) && Cmp->getType()->isIntegerTy()) { @@ -509,7 +544,7 @@ ComputeValueKnownInPredecessors(Value *V, BasicBlock *BB, PredValueInfo &Result, // 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; @@ -525,7 +560,7 @@ ComputeValueKnownInPredecessors(Value *V, BasicBlock *BB, PredValueInfo &Result, if (Constant *CmpConst = dyn_cast(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; @@ -540,8 +575,42 @@ ComputeValueKnownInPredecessors(Value *V, BasicBlock *BB, PredValueInfo &Result, } } + if (SelectInst *SI = dyn_cast(I)) { + // Handle select instructions where at least one operand is a known constant + // and we can figure out the condition value for any predecessor block. + Constant *TrueVal = getKnownConstant(SI->getTrueValue(), Preference); + Constant *FalseVal = getKnownConstant(SI->getFalseValue(), Preference); + PredValueInfoTy Conds; + if ((TrueVal || FalseVal) && + ComputeValueKnownInPredecessors(SI->getCondition(), BB, Conds, + WantInteger, CxtI)) { + for (unsigned i = 0, e = Conds.size(); i != e; ++i) { + Constant *Cond = Conds[i].first; + + // Figure out what value to use for the condition. + bool KnownCond; + if (ConstantInt *CI = dyn_cast(Cond)) { + // A known boolean. + KnownCond = CI->isOne(); + } else { + assert(isa(Cond) && "Unexpected condition value"); + // Either operand will do, so be sure to pick the one that's a known + // constant. + // FIXME: Do this more cleverly if both values are known constants? + KnownCond = (TrueVal != nullptr); + } + + // See if the select has a known constant value for this predecessor. + if (Constant *Val = KnownCond ? TrueVal : FalseVal) + Result.push_back(std::make_pair(Val, Conds[i].second)); + } + + return !Result.empty(); + } + } + // 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)); @@ -567,19 +636,31 @@ static unsigned GetBestDestForJumpOnUndef(BasicBlock *BB) { for (unsigned i = 1, e = BBTerm->getNumSuccessors(); i != e; ++i) { TestBB = BBTerm->getSuccessor(i); unsigned NumPreds = std::distance(pred_begin(TestBB), pred_end(TestBB)); - if (NumPreds < MinNumPreds) + if (NumPreds < MinNumPreds) { MinSucc = i; + MinNumPreds = NumPreds; + } } return MinSucc; } +static bool hasAddressTakenAndUsed(BasicBlock *BB) { + if (!BB->hasAddressTaken()) return false; + + // If the block has its address taken, it may be a tree of dead constants + // hanging off of it. These shouldn't keep the block alive. + BlockAddress *BA = BlockAddress::get(BB); + BA->removeDeadConstantUsers(); + return !BA->use_empty(); +} + /// ProcessBlock - If there are any predecessors whose control can be threaded /// through to a successor, transform them now. 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; @@ -588,20 +669,16 @@ bool JumpThreading::ProcessBlock(BasicBlock *BB) { // because now the condition in this block can be threaded through // predecessors of our predecessor block. if (BasicBlock *SinglePred = BB->getSinglePredecessor()) { - if (SinglePred->getTerminator()->getNumSuccessors() == 1 && - SinglePred != BB) { + const TerminatorInst *TI = SinglePred->getTerminator(); + if (!TI->isExceptional() && TI->getNumSuccessors() == 1 && + SinglePred != BB && !hasAddressTakenAndUsed(BB)) { // If SinglePred was a loop header, BB becomes one. 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; } } @@ -620,12 +697,26 @@ bool JumpThreading::ProcessBlock(BasicBlock *BB) { } else if (SwitchInst *SI = dyn_cast(Terminator)) { Condition = SI->getCondition(); } else if (IndirectBrInst *IB = dyn_cast(Terminator)) { + // Can't thread indirect branch with no successors. + if (IB->getNumSuccessors() == 0) return false; Condition = IB->getAddress()->stripPointerCasts(); Preference = WantBlockAddress; } else { return false; // Must be an invoke. } + // Run constant folding to see if we can reduce the condition to a simple + // constant. + if (Instruction *I = dyn_cast(Condition)) { + Value *SimpleVal = + ConstantFoldInstruction(I, BB->getModule()->getDataLayout(), TLI); + if (SimpleVal) { + I->replaceAllUsesWith(SimpleVal); + I->eraseFromParent(); + Condition = SimpleVal; + } + } + // If the terminator is branching on an undef, we can pick any of the // successors to branch to. Let GetBestDestForJumpOnUndef decide. if (isa(Condition)) { @@ -652,58 +743,55 @@ bool JumpThreading::ProcessBlock(BasicBlock *BB) { DEBUG(dbgs() << " In block '" << BB->getName() << "' folding terminator: " << *BB->getTerminator() << '\n'); ++NumFolds; - ConstantFoldTerminator(BB); + ConstantFoldTerminator(BB, true); return true; } Instruction *CondInst = dyn_cast(Condition); // All the rest of our checks depend on the condition being an instruction. - if (CondInst == 0) { + 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(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 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(BB->getTerminator()); Constant *CondConst = dyn_cast(CondCmp->getOperand(1)); - pred_iterator PI = pred_begin(BB), PE = pred_end(BB); - if (CondBr && CondConst && CondBr->isConditional() && PI != PE && - (!isa(CondCmp->getOperand(0)) || - cast(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)) + return true; } // Check for some cases that are worth simplifying. Right now we want to look @@ -727,7 +815,7 @@ bool JumpThreading::ProcessBlock(BasicBlock *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 @@ -749,14 +837,13 @@ bool JumpThreading::ProcessBlock(BasicBlock *BB) { return false; } - /// SimplifyPartiallyRedundantLoad - If LI is an obviously partially redundant /// load instruction, eliminate it by replacing it with a PHI node. This is an /// important optimization that encourages jump threading, and needs to be run /// interlaced with other jump threading tasks. bool JumpThreading::SimplifyPartiallyRedundantLoad(LoadInst *LI) { - // Don't hack volatile loads. - if (LI->isVolatile()) return false; + // Don't hack volatile/atomic loads. + if (!LI->isSimple()) return false; // If the load is defined in a block with exactly one predecessor, it can't be // partially redundant. @@ -764,6 +851,12 @@ bool JumpThreading::SimplifyPartiallyRedundantLoad(LoadInst *LI) { if (LoadBB->getSinglePredecessor()) return false; + // If the load is defined in a landing pad, it can't be partially redundant, + // because the edges between the invoke and the landing pad cannot have other + // instructions between them. + if (LoadBB->isLandingPad()) + return false; + Value *LoadedPtr = LI->getOperand(0); // If the loaded operand is defined in the LoadBB, it can't be available. @@ -785,6 +878,9 @@ bool JumpThreading::SimplifyPartiallyRedundantLoad(LoadInst *LI) { // 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; @@ -796,11 +892,15 @@ bool JumpThreading::SimplifyPartiallyRedundantLoad(LoadInst *LI) { if (BBIt != LoadBB->begin()) return false; + // 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 PredsScanned; typedef SmallVector, 8> AvailablePredsTy; AvailablePredsTy AvailablePreds; - BasicBlock *OneUnavailablePred = 0; + BasicBlock *OneUnavailablePred = nullptr; // If we got here, the loaded value is transparent through to the start of the // block. Check to see if it is available in any of the predecessor blocks. @@ -809,17 +909,22 @@ bool JumpThreading::SimplifyPartiallyRedundantLoad(LoadInst *LI) { 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(); - Value *PredAvailable = FindAvailableLoadedValue(LoadedPtr, PredBB, BBIt, 6); + AAMDNodes ThisAATags; + Value *PredAvailable = FindAvailableLoadedValue(LoadedPtr, PredBB, BBIt, 6, + nullptr, &ThisAATags); if (!PredAvailable) { OneUnavailablePred = PredBB; continue; } + // 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. AvailablePreds.push_back(std::make_pair(PredBB, PredAvailable)); @@ -834,7 +939,7 @@ bool JumpThreading::SimplifyPartiallyRedundantLoad(LoadInst *LI) { // predecessor, we want to insert a merge block for those common predecessors. // This ensures that we only have to insert one reload, thus not increasing // code size. - BasicBlock *UnavailablePred = 0; + BasicBlock *UnavailablePred = nullptr; // If there is exactly one predecessor where the value is unavailable, the // already computed 'OneUnavailablePred' block is it. If it ends in an @@ -865,8 +970,7 @@ bool JumpThreading::SimplifyPartiallyRedundantLoad(LoadInst *LI) { // Split them out to their own block. UnavailablePred = - SplitBlockPredecessors(LoadBB, &PredsToSplit[0], PredsToSplit.size(), - "thread-pre-split", this); + SplitBlockPredecessors(LoadBB, PredsToSplit, "thread-pre-split"); } // If the value isn't available in all predecessors, then there will be @@ -875,9 +979,13 @@ bool JumpThreading::SimplifyPartiallyRedundantLoad(LoadInst *LI) { if (UnavailablePred) { assert(UnavailablePred->getTerminator()->getNumSuccessors() == 1 && "Can't handle critical edge here!"); - Value *NewVal = new LoadInst(LoadedPtr, LI->getName()+".pr", false, + LoadInst *NewVal = new LoadInst(LoadedPtr, LI->getName()+".pr", false, LI->getAlignment(), UnavailablePred->getTerminator()); + NewVal->setDebugLoc(LI->getDebugLoc()); + if (AATags) + NewVal->setAAMetadata(AATags); + AvailablePreds.push_back(std::make_pair(UnavailablePred, NewVal)); } @@ -886,22 +994,33 @@ bool JumpThreading::SimplifyPartiallyRedundantLoad(LoadInst *LI) { array_pod_sort(AvailablePreds.begin(), AvailablePreds.end()); // Create a PHI node at the start of the block for the PRE'd load value. - PHINode *PN = PHINode::Create(LI->getType(), "", LoadBB->begin()); + pred_iterator PB = pred_begin(LoadBB), PE = pred_end(LoadBB); + PHINode *PN = PHINode::Create(LI->getType(), std::distance(PB, PE), "", + LoadBB->begin()); PN->takeName(LI); + PN->setDebugLoc(LI->getDebugLoc()); // Insert new entries into the PHI for each predecessor. A single block may // have multiple entries here. - for (pred_iterator PI = pred_begin(LoadBB), E = pred_end(LoadBB); PI != E; - ++PI) { + for (pred_iterator PI = PB; PI != PE; ++PI) { BasicBlock *P = *PI; AvailablePredsTy::iterator I = std::lower_bound(AvailablePreds.begin(), AvailablePreds.end(), - std::make_pair(P, (Value*)0)); + std::make_pair(P, (Value*)nullptr)); 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"; @@ -952,7 +1071,7 @@ FindMostPopularDest(BasicBlock *BB, } } - // Okay, now we know the most popular destination. If there is more than + // Okay, now we know the most popular destination. If there is more than one // destination, we need to determine one. This is arbitrary, but we need // to make a deterministic decision. Pick the first one that appears in the // successor list. @@ -976,14 +1095,15 @@ FindMostPopularDest(BasicBlock *BB, } 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() && @@ -1003,12 +1123,12 @@ bool JumpThreading::ProcessThreadableEdges(Value *Cond, BasicBlock *BB, SmallPtrSet SeenPreds; SmallVector, 16> PredToDestList; - BasicBlock *OnlyDest = 0; + BasicBlock *OnlyDest = nullptr; BasicBlock *MultipleDestSentinel = (BasicBlock*)(intptr_t)~0ULL; 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 @@ -1020,19 +1140,19 @@ bool JumpThreading::ProcessThreadableEdges(Value *Cond, BasicBlock *BB, BasicBlock *DestBB; if (isa(Val)) - DestBB = 0; + DestBB = nullptr; else if (BranchInst *BI = dyn_cast(BB->getTerminator())) DestBB = BI->getSuccessor(cast(Val)->isZero()); - else if (SwitchInst *SI = dyn_cast(BB->getTerminator())) - DestBB = SI->getSuccessor(SI->findCaseValue(cast(Val))); - else { + else if (SwitchInst *SI = dyn_cast(BB->getTerminator())) { + DestBB = SI->findCaseValue(cast(Val)).getCaseSuccessor(); + } else { assert(isa(BB->getTerminator()) && "Unexpected terminator"); DestBB = cast(Val)->getBasicBlock(); } // If we have exactly one destination, remember it for efficiency below. - if (i == 0) + if (PredToDestList.empty()) OnlyDest = DestBB; else if (OnlyDest != DestBB) OnlyDest = MultipleDestSentinel; @@ -1071,7 +1191,7 @@ bool JumpThreading::ProcessThreadableEdges(Value *Cond, BasicBlock *BB, // If the threadable edges are branching on an undefined value, we get to pick // the destination that these predecessors should get to. - if (MostPopularDest == 0) + if (!MostPopularDest) MostPopularDest = BB->getTerminator()-> getSuccessor(GetBestDestForJumpOnUndef(BB)); @@ -1148,10 +1268,10 @@ bool JumpThreading::ProcessBranchOnXOR(BinaryOperator *BO) { 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; } @@ -1173,7 +1293,7 @@ bool JumpThreading::ProcessBranchOnXOR(BinaryOperator *BO) { } // Determine which value to split on, true, false, or undef if neither. - ConstantInt *SplitVal = 0; + ConstantInt *SplitVal = nullptr; if (NumTrue > NumFalse) SplitVal = ConstantInt::getTrue(BB->getContext()); else if (NumTrue != 0 || NumFalse != 0) @@ -1194,7 +1314,7 @@ bool JumpThreading::ProcessBranchOnXOR(BinaryOperator *BO) { // help us. However, we can just replace the LHS or RHS with the constant. if (BlocksToFoldInto.size() == cast(BB->front()).getNumIncomingValues()) { - if (SplitVal == 0) { + if (!SplitVal) { // If all preds provide undef, just nuke the xor, because it is undef too. BO->replaceAllUsesWith(UndefValue::get(BO->getType())); BO->eraseFromParent(); @@ -1261,8 +1381,8 @@ bool JumpThreading::ThreadEdge(BasicBlock *BB, return false; } - unsigned JumpThreadCost = getJumpThreadDuplicationCost(BB); - 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; @@ -1275,8 +1395,7 @@ bool JumpThreading::ThreadEdge(BasicBlock *BB, else { DEBUG(dbgs() << " Factoring out " << PredBBs.size() << " common predecessors.\n"); - PredBB = SplitBlockPredecessors(BB, &PredBBs[0], PredBBs.size(), - ".thr_comm", this); + PredBB = SplitBlockPredecessors(BB, PredBBs, ".thr_comm"); } // And finally, do it! @@ -1320,7 +1439,8 @@ bool JumpThreading::ThreadEdge(BasicBlock *BB, // We didn't copy the terminator from BB over to NewBB, because there is now // an unconditional jump to SuccBB. Insert the unconditional jump. - BranchInst::Create(SuccBB, NewBB); + BranchInst *NewBI =BranchInst::Create(SuccBB, NewBB); + NewBI->setDebugLoc(BB->getTerminator()->getDebugLoc()); // Check to see if SuccBB has PHI nodes. If so, we need to add entries to the // PHI nodes for NewBB now. @@ -1335,16 +1455,15 @@ bool JumpThreading::ThreadEdge(BasicBlock *BB, for (BasicBlock::iterator I = BB->begin(); I != BB->end(); ++I) { // Scan all uses of this instruction to see if it is used outside of its // block, and if so, record them in UsesToRename. - for (Value::use_iterator UI = I->use_begin(), E = I->use_end(); UI != E; - ++UI) { - Instruction *User = cast(*UI); + for (Use &U : I->uses()) { + Instruction *User = cast(U.getUser()); if (PHINode *UserPN = dyn_cast(User)) { - if (UserPN->getIncomingBlock(UI) == BB) + if (UserPN->getIncomingBlock(U) == BB) continue; } else if (User->getParent() == BB) continue; - UsesToRename.push_back(&UI.getUse()); + UsesToRename.push_back(&U); } // If there are no uses outside the block, we're done with this instruction. @@ -1379,7 +1498,7 @@ bool JumpThreading::ThreadEdge(BasicBlock *BB, // 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, TD); + SimplifyInstructionsInBlock(NewBB, TLI); // Threaded an edge! ++NumThreads; @@ -1405,8 +1524,8 @@ bool JumpThreading::DuplicateCondBranchOnPHIIntoPred(BasicBlock *BB, return false; } - unsigned DuplicationCost = getJumpThreadDuplicationCost(BB); - 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; @@ -1419,8 +1538,7 @@ bool JumpThreading::DuplicateCondBranchOnPHIIntoPred(BasicBlock *BB, else { DEBUG(dbgs() << " Factoring out " << PredBBs.size() << " common predecessors.\n"); - PredBB = SplitBlockPredecessors(BB, &PredBBs[0], PredBBs.size(), - ".thr_comm", this); + PredBB = SplitBlockPredecessors(BB, PredBBs, ".thr_comm"); } // Okay, we decided to do this! Clone all the instructions in BB onto the end @@ -1433,8 +1551,8 @@ bool JumpThreading::DuplicateCondBranchOnPHIIntoPred(BasicBlock *BB, // can just clone the bits from BB into the end of the new PredBB. BranchInst *OldPredBranch = dyn_cast(PredBB->getTerminator()); - if (OldPredBranch == 0 || !OldPredBranch->isUnconditional()) { - PredBB = SplitEdge(PredBB, BB, this); + if (!OldPredBranch || !OldPredBranch->isUnconditional()) { + PredBB = SplitEdge(PredBB, BB); OldPredBranch = cast(PredBB->getTerminator()); } @@ -1445,7 +1563,6 @@ bool JumpThreading::DuplicateCondBranchOnPHIIntoPred(BasicBlock *BB, BasicBlock::iterator BI = BB->begin(); for (; PHINode *PN = dyn_cast(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) { @@ -1462,7 +1579,8 @@ bool JumpThreading::DuplicateCondBranchOnPHIIntoPred(BasicBlock *BB, // 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, TD)) { + if (Value *IV = + SimplifyInstruction(New, BB->getModule()->getDataLayout())) { delete New; ValueMapping[BI] = IV; } else { @@ -1490,16 +1608,15 @@ bool JumpThreading::DuplicateCondBranchOnPHIIntoPred(BasicBlock *BB, for (BasicBlock::iterator I = BB->begin(); I != BB->end(); ++I) { // Scan all uses of this instruction to see if it is used outside of its // block, and if so, record them in UsesToRename. - for (Value::use_iterator UI = I->use_begin(), E = I->use_end(); UI != E; - ++UI) { - Instruction *User = cast(*UI); + for (Use &U : I->uses()) { + Instruction *User = cast(U.getUser()); if (PHINode *UserPN = dyn_cast(User)) { - if (UserPN->getIncomingBlock(UI) == BB) + if (UserPN->getIncomingBlock(U) == BB) continue; } else if (User->getParent() == BB) continue; - UsesToRename.push_back(&UI.getUse()); + UsesToRename.push_back(&U); } // If there are no uses outside the block, we're done with this instruction. @@ -1531,4 +1648,80 @@ bool JumpThreading::DuplicateCondBranchOnPHIIntoPred(BasicBlock *BB, return true; } +/// TryToUnfoldSelect - Look for blocks of the form +/// bb1: +/// %a = select +/// br bb +/// +/// bb2: +/// %p = phi [%a, %bb] ... +/// %c = icmp %p +/// br i1 %c +/// +/// And expand the select into a branch structure if one of its arms allows %c +/// to be folded. This later enables threading from bb1 over bb2. +bool JumpThreading::TryToUnfoldSelect(CmpInst *CondCmp, BasicBlock *BB) { + BranchInst *CondBr = dyn_cast(BB->getTerminator()); + PHINode *CondLHS = dyn_cast(CondCmp->getOperand(0)); + Constant *CondRHS = cast(CondCmp->getOperand(1)); + + if (!CondBr || !CondBr->isConditional() || !CondLHS || + CondLHS->getParent() != BB) + return false; + + for (unsigned I = 0, E = CondLHS->getNumIncomingValues(); I != E; ++I) { + BasicBlock *Pred = CondLHS->getIncomingBlock(I); + SelectInst *SI = dyn_cast(CondLHS->getIncomingValue(I)); + // Look if one of the incoming values is a select in the corresponding + // predecessor. + if (!SI || SI->getParent() != Pred || !SI->hasOneUse()) + continue; + + BranchInst *PredTerm = dyn_cast(Pred->getTerminator()); + if (!PredTerm || !PredTerm->isUnconditional()) + continue; + + // Now check if one of the select values would allow us to constant fold the + // terminator in BB. We don't do the transform if both sides fold, those + // cases will be threaded in any case. + LazyValueInfo::Tristate LHSFolds = + LVI->getPredicateOnEdge(CondCmp->getPredicate(), SI->getOperand(1), + CondRHS, Pred, BB, CondCmp); + LazyValueInfo::Tristate RHSFolds = + LVI->getPredicateOnEdge(CondCmp->getPredicate(), SI->getOperand(2), + CondRHS, Pred, BB, CondCmp); + if ((LHSFolds != LazyValueInfo::Unknown || + RHSFolds != LazyValueInfo::Unknown) && + LHSFolds != RHSFolds) { + // Expand the select. + // + // Pred -- + // | v + // | NewBB + // | | + // |----- + // v + // BB + BasicBlock *NewBB = BasicBlock::Create(BB->getContext(), "select.unfold", + BB->getParent(), BB); + // Move the unconditional branch to NewBB. + PredTerm->removeFromParent(); + NewBB->getInstList().insert(NewBB->end(), PredTerm); + // Create a conditional branch and update PHI nodes. + BranchInst::Create(NewBB, BB, SI->getCondition(), Pred); + CondLHS->setIncomingValue(I, SI->getFalseValue()); + CondLHS->addIncoming(SI->getTrueValue(), NewBB); + // The select is now dead. + SI->eraseFromParent(); + + // Update any other PHI nodes in BB. + for (BasicBlock::iterator BI = BB->begin(); + PHINode *Phi = dyn_cast(BI); ++BI) + if (Phi != CondLHS) + Phi->addIncoming(Phi->getIncomingValueForBlock(Pred), NewBB); + return true; + } + } + return false; +}