diff options
Diffstat (limited to 'lib/Transforms/Utils')
-rw-r--r-- | lib/Transforms/Utils/BypassSlowDivision.cpp | 4 | ||||
-rw-r--r-- | lib/Transforms/Utils/CloneFunction.cpp | 38 | ||||
-rw-r--r-- | lib/Transforms/Utils/InlineFunction.cpp | 67 | ||||
-rw-r--r-- | lib/Transforms/Utils/Local.cpp | 48 | ||||
-rw-r--r-- | lib/Transforms/Utils/SimplifyCFG.cpp | 127 | ||||
-rw-r--r-- | lib/Transforms/Utils/SimplifyLibCalls.cpp | 425 | ||||
-rw-r--r-- | lib/Transforms/Utils/Utils.cpp | 1 | ||||
-rw-r--r-- | lib/Transforms/Utils/ValueMapper.cpp | 2 |
8 files changed, 459 insertions, 253 deletions
diff --git a/lib/Transforms/Utils/BypassSlowDivision.cpp b/lib/Transforms/Utils/BypassSlowDivision.cpp index 00cda8e034..1f517d038d 100644 --- a/lib/Transforms/Utils/BypassSlowDivision.cpp +++ b/lib/Transforms/Utils/BypassSlowDivision.cpp @@ -163,7 +163,7 @@ static bool insertFastDiv(Function &F, Value *AndV = MainBuilder.CreateAnd(OrV, BitMask); // Compare operand values and branch - Value *ZeroV = MainBuilder.getInt32(0); + Value *ZeroV = ConstantInt::getSigned(Dividend->getType(), 0); Value *CmpV = MainBuilder.CreateICmpEQ(AndV, ZeroV); MainBuilder.CreateCondBr(CmpV, FastBB, SlowBB); @@ -244,7 +244,7 @@ bool llvm::bypassSlowDivision(Function &F, // Get bitwidth of div/rem instruction IntegerType *T = cast<IntegerType>(J->getType()); - int bitwidth = T->getBitWidth(); + unsigned int bitwidth = T->getBitWidth(); // Continue if bitwidth is not bypassed DenseMap<unsigned int, unsigned int>::const_iterator BI = BypassWidths.find(bitwidth); diff --git a/lib/Transforms/Utils/CloneFunction.cpp b/lib/Transforms/Utils/CloneFunction.cpp index a309bce544..be8d39e128 100644 --- a/lib/Transforms/Utils/CloneFunction.cpp +++ b/lib/Transforms/Utils/CloneFunction.cpp @@ -87,26 +87,26 @@ void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc, assert(VMap.count(I) && "No mapping from source argument specified!"); #endif - // Clone any attributes. - if (NewFunc->arg_size() == OldFunc->arg_size()) - NewFunc->copyAttributesFrom(OldFunc); - else { - //Some arguments were deleted with the VMap. Copy arguments one by one - for (Function::const_arg_iterator I = OldFunc->arg_begin(), - E = OldFunc->arg_end(); I != E; ++I) - if (Argument* Anew = dyn_cast<Argument>(VMap[I])) - Anew->addAttr(OldFunc->getAttributes() - .getParamAttributes(I->getArgNo() + 1)); - NewFunc->setAttributes(NewFunc->getAttributes() - .addAttributes(NewFunc->getContext(), - AttributeSet::ReturnIndex, - OldFunc->getAttributes())); - NewFunc->setAttributes(NewFunc->getAttributes() - .addAttributes(NewFunc->getContext(), - AttributeSet::FunctionIndex, - OldFunc->getAttributes())); + AttributeSet OldAttrs = OldFunc->getAttributes(); + // Clone any argument attributes that are present in the VMap. + for (Function::const_arg_iterator I = OldFunc->arg_begin(), + E = OldFunc->arg_end(); + I != E; ++I) + if (Argument *Anew = dyn_cast<Argument>(VMap[I])) { + AttributeSet attrs = + OldAttrs.getParamAttributes(I->getArgNo() + 1); + if (attrs.getNumSlots() > 0) + Anew->addAttr(attrs); + } - } + NewFunc->setAttributes(NewFunc->getAttributes() + .addAttributes(NewFunc->getContext(), + AttributeSet::ReturnIndex, + OldAttrs.getRetAttributes())); + NewFunc->setAttributes(NewFunc->getAttributes() + .addAttributes(NewFunc->getContext(), + AttributeSet::FunctionIndex, + OldAttrs.getFnAttributes())); // Loop over all of the basic blocks in the function, cloning them as // appropriate. Note that we save BE this way in order to handle cloning of diff --git a/lib/Transforms/Utils/InlineFunction.cpp b/lib/Transforms/Utils/InlineFunction.cpp index 0d2598a221..dabb67b921 100644 --- a/lib/Transforms/Utils/InlineFunction.cpp +++ b/lib/Transforms/Utils/InlineFunction.cpp @@ -82,7 +82,8 @@ namespace { /// a simple branch. When there is more than one predecessor, we need to /// split the landing pad block after the landingpad instruction and jump /// to there. - void forwardResume(ResumeInst *RI); + void forwardResume(ResumeInst *RI, + SmallPtrSet<LandingPadInst*, 16> &InlinedLPads); /// addIncomingPHIValuesFor - Add incoming-PHI values to the unwind /// destination block for the given basic block, using the values for the @@ -140,8 +141,10 @@ BasicBlock *InvokeInliningInfo::getInnerResumeDest() { /// block. When the landing pad block has only one predecessor, this is a simple /// branch. When there is more than one predecessor, we need to split the /// landing pad block after the landingpad instruction and jump to there. -void InvokeInliningInfo::forwardResume(ResumeInst *RI) { +void InvokeInliningInfo::forwardResume(ResumeInst *RI, + SmallPtrSet<LandingPadInst*, 16> &InlinedLPads) { BasicBlock *Dest = getInnerResumeDest(); + LandingPadInst *OuterLPad = getLandingPadInst(); BasicBlock *Src = RI->getParent(); BranchInst::Create(Dest, Src); @@ -152,6 +155,16 @@ void InvokeInliningInfo::forwardResume(ResumeInst *RI) { InnerEHValuesPHI->addIncoming(RI->getOperand(0), Src); RI->eraseFromParent(); + + // Append the clauses from the outer landing pad instruction into the inlined + // landing pad instructions. + for (SmallPtrSet<LandingPadInst*, 16>::iterator I = InlinedLPads.begin(), + E = InlinedLPads.end(); I != E; ++I) { + LandingPadInst *InlinedLPad = *I; + for (unsigned OuterIdx = 0, OuterNum = OuterLPad->getNumClauses(); + OuterIdx != OuterNum; ++OuterIdx) + InlinedLPad->addClause(OuterLPad->getClause(OuterIdx)); + } } /// HandleCallsInBlockInlinedThroughInvoke - When we inline a basic block into @@ -229,19 +242,15 @@ static void HandleInlinedInvoke(InvokeInst *II, BasicBlock *FirstNewBlock, // The inlined code is currently at the end of the function, scan from the // start of the inlined code to its end, checking for stuff we need to - // rewrite. If the code doesn't have calls or unwinds, we know there is - // nothing to rewrite. - if (!InlinedCodeInfo.ContainsCalls) { - // Now that everything is happy, we have one final detail. The PHI nodes in - // the exception destination block still have entries due to the original - // invoke instruction. Eliminate these entries (which might even delete the - // PHI node) now. - InvokeDest->removePredecessor(II->getParent()); - return; - } - + // rewrite. InvokeInliningInfo Invoke(II); - + + // Get all of the inlined landing pad instructions. + SmallPtrSet<LandingPadInst*, 16> InlinedLPads; + for (Function::iterator I = FirstNewBlock, E = Caller->end(); I != E; ++I) + if (InvokeInst *II = dyn_cast<InvokeInst>(I->getTerminator())) + InlinedLPads.insert(II->getLandingPadInst()); + for (Function::iterator BB = FirstNewBlock, E = Caller->end(); BB != E; ++BB){ if (InlinedCodeInfo.ContainsCalls) if (HandleCallsInBlockInlinedThroughInvoke(BB, Invoke)) { @@ -250,13 +259,14 @@ static void HandleInlinedInvoke(InvokeInst *II, BasicBlock *FirstNewBlock, continue; } + // Forward any resumes that are remaining here. if (ResumeInst *RI = dyn_cast<ResumeInst>(BB->getTerminator())) - Invoke.forwardResume(RI); + Invoke.forwardResume(RI, InlinedLPads); } // Now that everything is happy, we have one final detail. The PHI nodes in // the exception destination block still have entries due to the original - // invoke instruction. Eliminate these entries (which might even delete the + // invoke instruction. Eliminate these entries (which might even delete the // PHI node) now. InvokeDest->removePredecessor(II->getParent()); } @@ -748,8 +758,10 @@ bool llvm::InlineFunction(CallSite CS, InlineFunctionInfo &IFI, // If the call site was an invoke instruction, add a branch to the normal // destination. - if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) - BranchInst::Create(II->getNormalDest(), TheCall); + if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) { + BranchInst *NewBr = BranchInst::Create(II->getNormalDest(), TheCall); + NewBr->setDebugLoc(Returns[0]->getDebugLoc()); + } // If the return instruction returned a value, replace uses of the call with // uses of the returned value. @@ -777,15 +789,16 @@ bool llvm::InlineFunction(CallSite CS, InlineFunctionInfo &IFI, // "starter" and "ender" blocks. How we accomplish this depends on whether // this is an invoke instruction or a call instruction. BasicBlock *AfterCallBB; + BranchInst *CreatedBranchToNormalDest = NULL; if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) { // Add an unconditional branch to make this look like the CallInst case... - BranchInst *NewBr = BranchInst::Create(II->getNormalDest(), TheCall); + CreatedBranchToNormalDest = BranchInst::Create(II->getNormalDest(), TheCall); // Split the basic block. This guarantees that no PHI nodes will have to be // updated due to new incoming edges, and make the invoke case more // symmetric to the call case. - AfterCallBB = OrigBB->splitBasicBlock(NewBr, + AfterCallBB = OrigBB->splitBasicBlock(CreatedBranchToNormalDest, CalledFunc->getName()+".exit"); } else { // It's a call @@ -840,11 +853,20 @@ bool llvm::InlineFunction(CallSite CS, InlineFunctionInfo &IFI, // Add a branch to the merge points and remove return instructions. + DebugLoc Loc; for (unsigned i = 0, e = Returns.size(); i != e; ++i) { ReturnInst *RI = Returns[i]; - BranchInst::Create(AfterCallBB, RI); + BranchInst* BI = BranchInst::Create(AfterCallBB, RI); + Loc = RI->getDebugLoc(); + BI->setDebugLoc(Loc); RI->eraseFromParent(); } + // We need to set the debug location to *somewhere* inside the + // inlined function. The line number may be nonsensical, but the + // instruction will at least be associated with the right + // function. + if (CreatedBranchToNormalDest) + CreatedBranchToNormalDest->setDebugLoc(Loc); } else if (!Returns.empty()) { // Otherwise, if there is exactly one return value, just replace anything // using the return value of the call with the computed value. @@ -864,6 +886,9 @@ bool llvm::InlineFunction(CallSite CS, InlineFunctionInfo &IFI, AfterCallBB->getInstList().splice(AfterCallBB->begin(), ReturnBB->getInstList()); + if (CreatedBranchToNormalDest) + CreatedBranchToNormalDest->setDebugLoc(Returns[0]->getDebugLoc()); + // Delete the return instruction now and empty ReturnBB now. Returns[0]->eraseFromParent(); ReturnBB->eraseFromParent(); diff --git a/lib/Transforms/Utils/Local.cpp b/lib/Transforms/Utils/Local.cpp index a54ee08b67..12e5b3e9d2 100644 --- a/lib/Transforms/Utils/Local.cpp +++ b/lib/Transforms/Utils/Local.cpp @@ -832,7 +832,24 @@ unsigned llvm::getOrEnforceKnownAlignment(Value *V, unsigned PrefAlign, /// Dbg Intrinsic utilities /// -/// Inserts a llvm.dbg.value instrinsic before the stores to an alloca'd value +/// See if there is a dbg.value intrinsic for DIVar before I. +static bool LdStHasDebugValue(DIVariable &DIVar, Instruction *I) { + // Since we can't guarantee that the original dbg.declare instrinsic + // is removed by LowerDbgDeclare(), we need to make sure that we are + // not inserting the same dbg.value intrinsic over and over. + llvm::BasicBlock::InstListType::iterator PrevI(I); + if (PrevI != I->getParent()->getInstList().begin()) { + --PrevI; + if (DbgValueInst *DVI = dyn_cast<DbgValueInst>(PrevI)) + if (DVI->getValue() == I->getOperand(0) && + DVI->getOffset() == 0 && + DVI->getVariable() == DIVar) + return true; + } + return false; +} + +/// Inserts a llvm.dbg.value intrinsic before a store to an alloca'd value /// that has an associated llvm.dbg.decl intrinsic. bool llvm::ConvertDebugDeclareToDebugValue(DbgDeclareInst *DDI, StoreInst *SI, DIBuilder &Builder) { @@ -840,6 +857,9 @@ bool llvm::ConvertDebugDeclareToDebugValue(DbgDeclareInst *DDI, if (!DIVar.Verify()) return false; + if (LdStHasDebugValue(DIVar, SI)) + return true; + Instruction *DbgVal = NULL; // If an argument is zero extended then use argument directly. The ZExt // may be zapped by an optimization pass in future. @@ -863,7 +883,7 @@ bool llvm::ConvertDebugDeclareToDebugValue(DbgDeclareInst *DDI, return true; } -/// Inserts a llvm.dbg.value instrinsic before the stores to an alloca'd value +/// Inserts a llvm.dbg.value intrinsic before a load of an alloca'd value /// that has an associated llvm.dbg.decl intrinsic. bool llvm::ConvertDebugDeclareToDebugValue(DbgDeclareInst *DDI, LoadInst *LI, DIBuilder &Builder) { @@ -871,6 +891,9 @@ bool llvm::ConvertDebugDeclareToDebugValue(DbgDeclareInst *DDI, if (!DIVar.Verify()) return false; + if (LdStHasDebugValue(DIVar, LI)) + return true; + Instruction *DbgVal = Builder.insertDbgValueIntrinsic(LI->getOperand(0), 0, DIVar, LI); @@ -902,6 +925,8 @@ bool llvm::LowerDbgDeclare(Function &F) { E = Dbgs.end(); I != E; ++I) { DbgDeclareInst *DDI = *I; if (AllocaInst *AI = dyn_cast_or_null<AllocaInst>(DDI->getAddress())) { + // We only remove the dbg.declare intrinsic if all uses are + // converted to dbg.value intrinsics. bool RemoveDDI = true; for (Value::use_iterator UI = AI->use_begin(), E = AI->use_end(); UI != E; ++UI) @@ -985,22 +1010,17 @@ bool llvm::removeUnreachableBlocks(Function &F) { if (Reachable.count(I)) continue; - // Remove the block as predecessor of all its reachable successors. - // Unreachable successors don't matter as they'll soon be removed, too. for (succ_iterator SI = succ_begin(I), SE = succ_end(I); SI != SE; ++SI) if (Reachable.count(*SI)) (*SI)->removePredecessor(I); + I->dropAllReferences(); + } - // Zap all instructions in this basic block. - while (!I->empty()) { - Instruction &Inst = I->back(); - if (!Inst.use_empty()) - Inst.replaceAllUsesWith(UndefValue::get(Inst.getType())); - I->getInstList().pop_back(); - } + for (Function::iterator I = llvm::next(F.begin()), E=F.end(); I != E;) + if (!Reachable.count(I)) + I = F.getBasicBlockList().erase(I); + else + ++I; - --I; - llvm::next(I)->eraseFromParent(); - } return true; } diff --git a/lib/Transforms/Utils/SimplifyCFG.cpp b/lib/Transforms/Utils/SimplifyCFG.cpp index a63d31d5af..052ad85551 100644 --- a/lib/Transforms/Utils/SimplifyCFG.cpp +++ b/lib/Transforms/Utils/SimplifyCFG.cpp @@ -59,6 +59,10 @@ static cl::opt<bool> SinkCommon("simplifycfg-sink-common", cl::Hidden, cl::init(true), cl::desc("Sink common instructions down to the end block")); +static cl::opt<bool> +HoistCondStores("simplifycfg-hoist-cond-stores", cl::Hidden, cl::init(true), + cl::desc("Hoist conditional stores if an unconditional store preceeds")); + STATISTIC(NumBitMaps, "Number of switch instructions turned into bitmaps"); STATISTIC(NumLookupTables, "Number of switch instructions turned into lookup tables"); STATISTIC(NumSinkCommons, "Number of common instructions sunk down to the end block"); @@ -1332,6 +1336,66 @@ static bool SinkThenElseCodeToEnd(BranchInst *BI1) { return Changed; } +/// \brief Determine if we can hoist sink a sole store instruction out of a +/// conditional block. +/// +/// We are looking for code like the following: +/// BrBB: +/// store i32 %add, i32* %arrayidx2 +/// ... // No other stores or function calls (we could be calling a memory +/// ... // function). +/// %cmp = icmp ult %x, %y +/// br i1 %cmp, label %EndBB, label %ThenBB +/// ThenBB: +/// store i32 %add5, i32* %arrayidx2 +/// br label EndBB +/// EndBB: +/// ... +/// We are going to transform this into: +/// BrBB: +/// store i32 %add, i32* %arrayidx2 +/// ... // +/// %cmp = icmp ult %x, %y +/// %add.add5 = select i1 %cmp, i32 %add, %add5 +/// store i32 %add.add5, i32* %arrayidx2 +/// ... +/// +/// \return The pointer to the value of the previous store if the store can be +/// hoisted into the predecessor block. 0 otherwise. +Value *isSafeToSpeculateStore(Instruction *I, BasicBlock *BrBB, + BasicBlock *StoreBB, BasicBlock *EndBB) { + StoreInst *StoreToHoist = dyn_cast<StoreInst>(I); + if (!StoreToHoist) + return 0; + + // Volatile or atomic. + if (!StoreToHoist->isSimple()) + return 0; + + Value *StorePtr = StoreToHoist->getPointerOperand(); + + // Look for a store to the same pointer in BrBB. + unsigned MaxNumInstToLookAt = 10; + for (BasicBlock::reverse_iterator RI = BrBB->rbegin(), + RE = BrBB->rend(); RI != RE && (--MaxNumInstToLookAt); ++RI) { + Instruction *CurI = &*RI; + + // Could be calling an instruction that effects memory like free(). + if (CurI->mayHaveSideEffects() && !isa<StoreInst>(CurI)) + return 0; + + StoreInst *SI = dyn_cast<StoreInst>(CurI); + // Found the previous store make sure it stores to the same location. + if (SI && SI->getPointerOperand() == StorePtr) + // Found the previous store, return its value operand. + return SI->getValueOperand(); + else if (SI) + return 0; // Unknown store. + } + + return 0; +} + /// \brief Speculate a conditional basic block flattening the CFG. /// /// Note that this is a very risky transform currently. Speculating @@ -1395,6 +1459,8 @@ static bool SpeculativelyExecuteBB(BranchInst *BI, BasicBlock *ThenBB) { SmallDenseMap<Instruction *, unsigned, 4> SinkCandidateUseCounts; unsigned SpeculationCost = 0; + Value *SpeculatedStoreValue = 0; + StoreInst *SpeculatedStore = 0; for (BasicBlock::iterator BBI = ThenBB->begin(), BBE = llvm::prior(ThenBB->end()); BBI != BBE; ++BBI) { @@ -1410,13 +1476,21 @@ static bool SpeculativelyExecuteBB(BranchInst *BI, BasicBlock *ThenBB) { return false; // Don't hoist the instruction if it's unsafe or expensive. - if (!isSafeToSpeculativelyExecute(I)) + if (!isSafeToSpeculativelyExecute(I) && + !(HoistCondStores && + (SpeculatedStoreValue = isSafeToSpeculateStore(I, BB, ThenBB, + EndBB)))) return false; - if (ComputeSpeculationCost(I) > PHINodeFoldingThreshold) + if (!SpeculatedStoreValue && + ComputeSpeculationCost(I) > PHINodeFoldingThreshold) return false; + // Store the store speculation candidate. + if (SpeculatedStoreValue) + SpeculatedStore = cast<StoreInst>(I); + // Do not hoist the instruction if any of its operands are defined but not - // used in this BB. The transformation will prevent the operand from + // used in BB. The transformation will prevent the operand from // being sunk into the use block. for (User::op_iterator i = I->op_begin(), e = I->op_end(); i != e; ++i) { @@ -1473,12 +1547,24 @@ static bool SpeculativelyExecuteBB(BranchInst *BI, BasicBlock *ThenBB) { // If there are no PHIs to process, bail early. This helps ensure idempotence // as well. - if (!HaveRewritablePHIs) + if (!HaveRewritablePHIs && !(HoistCondStores && SpeculatedStoreValue)) return false; // If we get here, we can hoist the instruction and if-convert. DEBUG(dbgs() << "SPECULATIVELY EXECUTING BB" << *ThenBB << "\n";); + // Insert a select of the value of the speculated store. + if (SpeculatedStoreValue) { + IRBuilder<true, NoFolder> Builder(BI); + Value *TrueV = SpeculatedStore->getValueOperand(); + Value *FalseV = SpeculatedStoreValue; + if (Invert) + std::swap(TrueV, FalseV); + Value *S = Builder.CreateSelect(BrCond, TrueV, FalseV, TrueV->getName() + + "." + FalseV->getName()); + SpeculatedStore->setOperand(0, S); + } + // Hoist the instructions. BB->getInstList().splice(BI, ThenBB->getInstList(), ThenBB->begin(), llvm::prior(ThenBB->end())); @@ -2789,9 +2875,20 @@ bool SimplifyCFGOpt::SimplifyResume(ResumeInst *RI, IRBuilder<> &Builder) { return false; // Turn all invokes that unwind here into calls and delete the basic block. + bool InvokeRequiresTableEntry = false; + bool Changed = false; for (pred_iterator PI = pred_begin(BB), PE = pred_end(BB); PI != PE;) { InvokeInst *II = cast<InvokeInst>((*PI++)->getTerminator()); + + if (II->hasFnAttr(Attribute::UWTable)) { + // Don't remove an `invoke' instruction if the ABI requires an entry into + // the table. + InvokeRequiresTableEntry = true; + continue; + } + SmallVector<Value*, 8> Args(II->op_begin(), II->op_end() - 3); + // Insert a call instruction before the invoke. CallInst *Call = CallInst::Create(II->getCalledValue(), Args, "", II); Call->takeName(II); @@ -2811,11 +2908,14 @@ bool SimplifyCFGOpt::SimplifyResume(ResumeInst *RI, IRBuilder<> &Builder) { // Finally, delete the invoke instruction! II->eraseFromParent(); + Changed = true; } - // The landingpad is now unreachable. Zap it. - BB->eraseFromParent(); - return true; + if (!InvokeRequiresTableEntry) + // The landingpad is now unreachable. Zap it. + BB->eraseFromParent(); + + return Changed; } bool SimplifyCFGOpt::SimplifyReturn(ReturnInst *RI, IRBuilder<> &Builder) { @@ -3059,7 +3159,12 @@ static bool TurnSwitchRangeIntoICmp(SwitchInst *SI, IRBuilder<> &Builder) { Value *Sub = SI->getCondition(); if (!Offset->isNullValue()) Sub = Builder.CreateAdd(Sub, Offset, Sub->getName()+".off"); - Value *Cmp = Builder.CreateICmpULT(Sub, NumCases, "switch"); + Value *Cmp; + // If NumCases overflowed, then all possible values jump to the successor. + if (NumCases->isNullValue() && SI->getNumCases() != 0) + Cmp = ConstantInt::getTrue(SI->getContext()); + else + Cmp = Builder.CreateICmpULT(Sub, NumCases, "switch"); BranchInst *NewBI = Builder.CreateCondBr( Cmp, SI->case_begin().getCaseSuccessor(), SI->getDefaultDest()); @@ -3944,11 +4049,13 @@ static bool passingValueIsAlwaysUndefined(Value *V, Instruction *I) { // Load from null is undefined. if (LoadInst *LI = dyn_cast<LoadInst>(Use)) - return LI->getPointerAddressSpace() == 0; + if (!LI->isVolatile()) + return LI->getPointerAddressSpace() == 0; // Store to null is undefined. if (StoreInst *SI = dyn_cast<StoreInst>(Use)) - return SI->getPointerAddressSpace() == 0 && SI->getPointerOperand() == I; + if (!SI->isVolatile()) + return SI->getPointerAddressSpace() == 0 && SI->getPointerOperand() == I; } return false; } diff --git a/lib/Transforms/Utils/SimplifyLibCalls.cpp b/lib/Transforms/Utils/SimplifyLibCalls.cpp index b25febaa14..cadec21c50 100644 --- a/lib/Transforms/Utils/SimplifyLibCalls.cpp +++ b/lib/Transforms/Utils/SimplifyLibCalls.cpp @@ -15,11 +15,13 @@ //===----------------------------------------------------------------------===// #include "llvm/Transforms/Utils/SimplifyLibCalls.h" +#include "llvm/ADT/SmallString.h" #include "llvm/ADT/StringMap.h" #include "llvm/Analysis/ValueTracking.h" #include "llvm/IR/DataLayout.h" #include "llvm/IR/Function.h" #include "llvm/IR/IRBuilder.h" +#include "llvm/IR/IntrinsicInst.h" #include "llvm/IR/Intrinsics.h" #include "llvm/IR/LLVMContext.h" #include "llvm/IR/Module.h" @@ -798,8 +800,7 @@ struct StrToOpt : public LibCallOptimization { if (isa<ConstantPointerNull>(EndPtr)) { // With a null EndPtr, this function won't capture the main argument. // It would be readonly too, except that it still may write to errno. - CI->addAttribute(1, Attribute::get(Callee->getContext(), - Attribute::NoCapture)); + CI->addAttribute(1, Attribute::NoCapture); } return 0; @@ -1517,6 +1518,12 @@ struct FPrintFOpt : public LibCallOptimization { if (!getConstantStringInfo(CI->getArgOperand(1), FormatStr)) return 0; + // Do not do any of the following transformations if the fprintf return + // value is used, in general the fprintf return value is not compatible + // with fwrite(), fputc() or fputs(). + if (!CI->use_empty()) + return 0; + // fprintf(F, "foo") --> fwrite("foo", 3, 1, F) if (CI->getNumArgOperands() == 2) { for (unsigned i = 0, e = FormatStr.size(); i != e; ++i) @@ -1526,11 +1533,10 @@ struct FPrintFOpt : public LibCallOptimization { // These optimizations require DataLayout. if (!TD) return 0; - Value *NewCI = EmitFWrite(CI->getArgOperand(1), - ConstantInt::get(TD->getIntPtrType(*Context), - FormatStr.size()), - CI->getArgOperand(0), B, TD, TLI); - return NewCI ? ConstantInt::get(CI->getType(), FormatStr.size()) : 0; + return EmitFWrite(CI->getArgOperand(1), + ConstantInt::get(TD->getIntPtrType(*Context), + FormatStr.size()), + CI->getArgOperand(0), B, TD, TLI); } // The remaining optimizations require the format string to be "%s" or "%c" @@ -1543,14 +1549,12 @@ struct FPrintFOpt : public LibCallOptimization { if (FormatStr[1] == 'c') { // fprintf(F, "%c", chr) --> fputc(chr, F) if (!CI->getArgOperand(2)->getType()->isIntegerTy()) return 0; - Value *NewCI = EmitFPutC(CI->getArgOperand(2), CI->getArgOperand(0), B, - TD, TLI); - return NewCI ? ConstantInt::get(CI->getType(), 1) : 0; + return EmitFPutC(CI->getArgOperand(2), CI->getArgOperand(0), B, TD, TLI); } if (FormatStr[1] == 's') { // fprintf(F, "%s", str) --> fputs(str, F) - if (!CI->getArgOperand(2)->getType()->isPointerTy() || !CI->use_empty()) + if (!CI->getArgOperand(2)->getType()->isPointerTy()) return 0; return EmitFPutS(CI->getArgOperand(2), CI->getArgOperand(0), B, TD, TLI); } @@ -1673,67 +1677,17 @@ class LibCallSimplifierImpl { const TargetLibraryInfo *TLI; const LibCallSimplifier *LCS; bool UnsafeFPShrink; - StringMap<LibCallOptimization*, BumpPtrAllocator> Optimizations; - - // Fortified library call optimizations. - MemCpyChkOpt MemCpyChk; - MemMoveChkOpt MemMoveChk; - MemSetChkOpt MemSetChk; - StrCpyChkOpt StrCpyChk; - StpCpyChkOpt StpCpyChk; - StrNCpyChkOpt StrNCpyChk; - - // String library call optimizations. - StrCatOpt StrCat; - StrNCatOpt StrNCat; - StrChrOpt StrChr; - StrRChrOpt StrRChr; - StrCmpOpt StrCmp; - StrNCmpOpt StrNCmp; - StrCpyOpt StrCpy; - StpCpyOpt StpCpy; - StrNCpyOpt StrNCpy; - StrLenOpt StrLen; - StrPBrkOpt StrPBrk; - StrToOpt StrTo; - StrSpnOpt StrSpn; - StrCSpnOpt StrCSpn; - StrStrOpt StrStr; - - // Memory library call optimizations. - MemCmpOpt MemCmp; - MemCpyOpt MemCpy; - MemMoveOpt MemMove; - MemSetOpt MemSet; - // Math library call optimizations. - UnaryDoubleFPOpt UnaryDoubleFP, UnsafeUnaryDoubleFP; - CosOpt Cos; PowOpt Pow; Exp2Opt Exp2; - // Integer library call optimizations. - FFSOpt FFS; - AbsOpt Abs; - IsDigitOpt IsDigit; - IsAsciiOpt IsAscii; - ToAsciiOpt ToAscii; - - // Formatting and IO library call optimizations. - PrintFOpt PrintF; - SPrintFOpt SPrintF; - FPrintFOpt FPrintF; - FWriteOpt FWrite; - FPutsOpt FPuts; - PutsOpt Puts; - - void initOptimizations(); - void addOpt(LibFunc::Func F, LibCallOptimization* Opt); - void addOpt(LibFunc::Func F1, LibFunc::Func F2, LibCallOptimization* Opt); + // Math library call optimizations. + CosOpt Cos; + PowOpt Pow; + Exp2Opt Exp2; public: LibCallSimplifierImpl(const DataLayout *TD, const TargetLibraryInfo *TLI, const LibCallSimplifier *LCS, bool UnsafeFPShrink = false) - : UnaryDoubleFP(false), UnsafeUnaryDoubleFP(true), - Cos(UnsafeFPShrink), Pow(UnsafeFPShrink), Exp2(UnsafeFPShrink) { + : Cos(UnsafeFPShrink), Pow(UnsafeFPShrink), Exp2(UnsafeFPShrink) { this->TD = TD; this->TLI = TLI; this->LCS = LCS; @@ -1741,123 +1695,234 @@ public: } Value *optimizeCall(CallInst *CI); + LibCallOptimization *lookupOptimization(CallInst *CI); + bool hasFloatVersion(StringRef FuncName); }; -void LibCallSimplifierImpl::initOptimizations() { - // Fortified library call optimizations. - Optimizations["__memcpy_chk"] = &MemCpyChk; - Optimizations["__memmove_chk"] = &MemMoveChk; - Optimizations["__memset_chk"] = &MemSetChk; - Optimizations["__strcpy_chk"] = &StrCpyChk; - Optimizations["__stpcpy_chk"] = &StpCpyChk; - Optimizations["__strncpy_chk"] = &StrNCpyChk; - Optimizations["__stpncpy_chk"] = &StrNCpyChk; - - // String library call optimizations. - addOpt(LibFunc::strcat, &StrCat); - addOpt(LibFunc::strncat, &StrNCat); - addOpt(LibFunc::strchr, &StrChr); - addOpt(LibFunc::strrchr, &StrRChr); - addOpt(LibFunc::strcmp, &StrCmp); - addOpt(LibFunc::strncmp, &StrNCmp); - addOpt(LibFunc::strcpy, &StrCpy); - addOpt(LibFunc::stpcpy, &StpCpy); - addOpt(LibFunc::strncpy, &StrNCpy); - addOpt(LibFunc::strlen, &StrLen); - addOpt(LibFunc::strpbrk, &StrPBrk); - addOpt(LibFunc::strtol, &StrTo); - addOpt(LibFunc::strtod, &StrTo); - addOpt(LibFunc::strtof, &StrTo); - addOpt(LibFunc::strtoul, &StrTo); - addOpt(LibFunc::strtoll, &StrTo); - addOpt(LibFunc::strtold, &StrTo); - addOpt(LibFunc::strtoull, &StrTo); - addOpt(LibFunc::strspn, &StrSpn); - addOpt(LibFunc::strcspn, &StrCSpn); - addOpt(LibFunc::strstr, &StrStr); - - // Memory library call optimizations. - addOpt(LibFunc::memcmp, &MemCmp); - addOpt(LibFunc::memcpy, &MemCpy); - addOpt(LibFunc::memmove, &MemMove); - addOpt(LibFunc::memset, &MemSet); +bool LibCallSimplifierImpl::hasFloatVersion(StringRef FuncName) { + LibFunc::Func Func; + SmallString<20> FloatFuncName = FuncName; + FloatFuncName += 'f'; + if (TLI->getLibFunc(FloatFuncName, Func)) + return TLI->has(Func); + return false; +} - // Math library call optimizations. - addOpt(LibFunc::ceil, LibFunc::ceilf, &UnaryDoubleFP); - addOpt(LibFunc::fabs, LibFunc::fabsf, &UnaryDoubleFP); - addOpt(LibFunc::floor, LibFunc::floorf, &UnaryDoubleFP); - addOpt(LibFunc::rint, LibFunc::rintf, &UnaryDoubleFP); - addOpt(LibFunc::round, LibFunc::roundf, &UnaryDoubleFP); - addOpt(LibFunc::nearbyint, LibFunc::nearbyintf, &UnaryDoubleFP); - addOpt(LibFunc::trunc, LibFunc::truncf, &UnaryDoubleFP); - - if(UnsafeFPShrink) { - addOpt(LibFunc::acos, LibFunc::acosf, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::acosh, LibFunc::acoshf, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::asin, LibFunc::asinf, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::asinh, LibFunc::asinhf, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::atan, LibFunc::atanf, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::atanh, LibFunc::atanhf, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::cbrt, LibFunc::cbrtf, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::cosh, LibFunc::coshf, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::exp, LibFunc::expf, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::exp10, LibFunc::exp10f, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::expm1, LibFunc::expm1f, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::log, LibFunc::logf, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::log10, LibFunc::log10f, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::log1p, LibFunc::log1pf, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::log2, LibFunc::log2f, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::logb, LibFunc::logbf, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::sin, LibFunc::sinf, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::sinh, LibFunc::sinhf, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::sqrt, LibFunc::sqrtf, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::tan, LibFunc::tanf, &UnsafeUnaryDoubleFP); - addOpt(LibFunc::tanh, LibFunc::tanhf, &UnsafeUnaryDoubleFP); +// Fortified library call optimizations. +static MemCpyChkOpt MemCpyChk; +static MemMoveChkOpt MemMoveChk; +static MemSetChkOpt MemSetChk; +static StrCpyChkOpt StrCpyChk; +static StpCpyChkOpt StpCpyChk; +static StrNCpyChkOpt StrNCpyChk; + +// String library call optimizations. +static StrCatOpt StrCat; +static StrNCatOpt StrNCat; +static StrChrOpt StrChr; +static StrRChrOpt StrRChr; +static StrCmpOpt StrCmp; +static StrNCmpOpt StrNCmp; +static StrCpyOpt StrCpy; +static StpCpyOpt StpCpy; +static StrNCpyOpt StrNCpy; +static StrLenOpt StrLen; +static StrPBrkOpt StrPBrk; +static StrToOpt StrTo; +static StrSpnOpt StrSpn; +static StrCSpnOpt StrCSpn; +static StrStrOpt StrStr; + +// Memory library call optimizations. +static MemCmpOpt MemCmp; +static MemCpyOpt MemCpy; +static MemMoveOpt MemMove; +static MemSetOpt MemSet; + +// Math library call optimizations. +static UnaryDoubleFPOpt UnaryDoubleFP(false); +static UnaryDoubleFPOpt UnsafeUnaryDoubleFP(true); + + // Integer library call optimizations. +static FFSOpt FFS; +static AbsOpt Abs; +static IsDigitOpt IsDigit; +static IsAsciiOpt IsAscii; +static ToAsciiOpt ToAscii; + +// Formatting and IO library call optimizations. +static PrintFOpt PrintF; +static SPrintFOpt SPrintF; +static FPrintFOpt FPrintF; +static FWriteOpt FWrite; +static FPutsOpt FPuts; +static PutsOpt Puts; + +LibCallOptimization *LibCallSimplifierImpl::lookupOptimization(CallInst *CI) { + LibFunc::Func Func; + Function *Callee = CI->getCalledFunction(); + StringRef FuncName = Callee->getName(); + + // Next check for intrinsics. + if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(CI)) { + switch (II->getIntrinsicID()) { + case Intrinsic::pow: + return &Pow; + case Intrinsic::exp2: + return &Exp2; + default: + return 0; + } } - addOpt(LibFunc::cosf, &Cos); - addOpt(LibFunc::cos, &Cos); - addOpt(LibFunc::cosl, &Cos); - addOpt(LibFunc::powf, &Pow); - addOpt(LibFunc::pow, &Pow); - addOpt(LibFunc::powl, &Pow); - Optimizations["llvm.pow.f32"] = &Pow; - Optimizations["llvm.pow.f64"] = &Pow; - Optimizations["llvm.pow.f80"] = &Pow; - Optimizations["llvm.pow.f128"] = &Pow; - Optimizations["llvm.pow.ppcf128"] = &Pow; - addOpt(LibFunc::exp2l, &Exp2); - addOpt(LibFunc::exp2, &Exp2); - addOpt(LibFunc::exp2f, &Exp2); - Optimizations["llvm.exp2.ppcf128"] = &Exp2; - Optimizations["llvm.exp2.f128"] = &Exp2; - Optimizations["llvm.exp2.f80"] = &Exp2; - Optimizations["llvm.exp2.f64"] = &Exp2; - Optimizations["llvm.exp2.f32"] = &Exp2; + // Then check for known library functions. + if (TLI->getLibFunc(FuncName, Func) && TLI->has(Func)) { + switch (Func) { + case LibFunc::strcat: + return &StrCat; + case LibFunc::strncat: + return &StrNCat; + case LibFunc::strchr: + return &StrChr; + case LibFunc::strrchr: + return &StrRChr; + case LibFunc::strcmp: + return &StrCmp; + ca |