From 5ccb0825ed1bdf6271ef451b8239e86d4ff635b1 Mon Sep 17 00:00:00 2001 From: Pete Cooper Date: Wed, 9 Nov 2011 23:07:35 +0000 Subject: DeadStoreElimination can now trim the size of a store if the end of the store is dead. Currently checks alignment and killing stores on a power of 2 boundary as this is likely to trim the size of the earlier store without breaking large vector stores into scalar ones. Fixes git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@144239 91177308-0d34-0410-b5e6-96231b3b80d8 --- lib/Transforms/Scalar/DeadStoreElimination.cpp | 143 +++++++++++++++++++------ 1 file changed, 110 insertions(+), 33 deletions(-) (limited to 'lib/Transforms/Scalar/DeadStoreElimination.cpp') diff --git a/lib/Transforms/Scalar/DeadStoreElimination.cpp b/lib/Transforms/Scalar/DeadStoreElimination.cpp index f11441895c..03a557e060 100644 --- a/lib/Transforms/Scalar/DeadStoreElimination.cpp +++ b/lib/Transforms/Scalar/DeadStoreElimination.cpp @@ -241,6 +241,24 @@ static bool isRemovable(Instruction *I) { } } + +/// isShortenable - Returns true if this instruction can be safely shortened in +/// length. +static bool isShortenable(Instruction *I) { + // Don't shorten stores for now + if (isa(I)) + return false; + + IntrinsicInst *II = cast(I); + switch (II->getIntrinsicID()) { + default: return false; + case Intrinsic::memset: + case Intrinsic::memcpy: + // Do shorten memory intrinsics. + return true; + } +} + /// getStoredPointerOperand - Return the pointer that is being written to. static Value *getStoredPointerOperand(Instruction *I) { if (StoreInst *SI = dyn_cast(I)) @@ -295,11 +313,24 @@ static bool isObjectPointerWithTrustworthySize(const Value *V) { return false; } -/// isCompleteOverwrite - Return true if a store to the 'Later' location +namespace { + enum OverwriteResult + { + OverwriteComplete, + OverwriteEnd, + OverwriteUnknown + }; +} + +/// isOverwrite - Return 'OverwriteComplete' if a store to the 'Later' location /// completely overwrites a store to the 'Earlier' location. -static bool isCompleteOverwrite(const AliasAnalysis::Location &Later, - const AliasAnalysis::Location &Earlier, - AliasAnalysis &AA) { +/// 'OverwriteEnd' if the end of the 'Earlier' location is completely +/// overwritten by 'Later', or 'OverwriteUnknown' if nothing can be determined +static OverwriteResult isOverwrite(const AliasAnalysis::Location &Later, + const AliasAnalysis::Location &Earlier, + AliasAnalysis &AA, + int64_t& EarlierOff, + int64_t& LaterOff) { const Value *P1 = Earlier.Ptr->stripPointerCasts(); const Value *P2 = Later.Ptr->stripPointerCasts(); @@ -313,23 +344,24 @@ static bool isCompleteOverwrite(const AliasAnalysis::Location &Later, // If we have no TargetData information around, then the size of the store // is inferrable from the pointee type. If they are the same type, then // we know that the store is safe. - if (AA.getTargetData() == 0) - return Later.Ptr->getType() == Earlier.Ptr->getType(); - return false; + if (AA.getTargetData() == 0 && + Later.Ptr->getType() == Earlier.Ptr->getType()) + return OverwriteComplete; + + return OverwriteUnknown; } // Make sure that the Later size is >= the Earlier size. - if (Later.Size < Earlier.Size) - return false; - return true; + if (Later.Size >= Earlier.Size) + return OverwriteComplete; } // Otherwise, we have to have size information, and the later store has to be // larger than the earlier one. if (Later.Size == AliasAnalysis::UnknownSize || Earlier.Size == AliasAnalysis::UnknownSize || - Later.Size <= Earlier.Size || AA.getTargetData() == 0) - return false; + AA.getTargetData() == 0) + return OverwriteUnknown; // Check to see if the later store is to the entire object (either a global, // an alloca, or a byval argument). If so, then it clearly overwrites any @@ -342,26 +374,27 @@ static bool isCompleteOverwrite(const AliasAnalysis::Location &Later, // If we can't resolve the same pointers to the same object, then we can't // analyze them at all. if (UO1 != UO2) - return false; + return OverwriteUnknown; // If the "Later" store is to a recognizable object, get its size. if (isObjectPointerWithTrustworthySize(UO2)) { uint64_t ObjectSize = TD.getTypeAllocSize(cast(UO2->getType())->getElementType()); if (ObjectSize == Later.Size) - return true; + return OverwriteComplete; } // Okay, we have stores to two completely different pointers. Try to // decompose the pointer into a "base + constant_offset" form. If the base // pointers are equal, then we can reason about the two stores. - int64_t EarlierOff = 0, LaterOff = 0; + EarlierOff = 0; + LaterOff = 0; const Value *BP1 = GetPointerBaseWithConstantOffset(P1, EarlierOff, TD); const Value *BP2 = GetPointerBaseWithConstantOffset(P2, LaterOff, TD); // If the base pointers still differ, we have two completely different stores. if (BP1 != BP2) - return false; + return OverwriteUnknown; // The later store completely overlaps the earlier store if: // @@ -379,11 +412,25 @@ static bool isCompleteOverwrite(const AliasAnalysis::Location &Later, // // We have to be careful here as *Off is signed while *.Size is unsigned. if (EarlierOff >= LaterOff && + Later.Size > Earlier.Size && uint64_t(EarlierOff - LaterOff) + Earlier.Size <= Later.Size) - return true; + return OverwriteComplete; + + // The other interesting case is if the later store overwrites the end of + // the earlier store + // + // |--earlier--| + // |-- later --| + // + // In this case we may want to trim the size of earlier to avoid generating + // writes to addresses which will definitely be overwritten later + if (LaterOff > EarlierOff && + LaterOff < int64_t(EarlierOff + Earlier.Size) && + LaterOff + Later.Size >= EarlierOff + Earlier.Size) + return OverwriteEnd; // Otherwise, they don't completely overlap. - return false; + return OverwriteUnknown; } /// isPossibleSelfRead - If 'Inst' might be a self read (i.e. a noop copy of a @@ -507,22 +554,52 @@ bool DSE::runOnBasicBlock(BasicBlock &BB) { // If we find a write that is a) removable (i.e., non-volatile), b) is // completely obliterated by the store to 'Loc', and c) which we know that // 'Inst' doesn't load from, then we can remove it. - if (isRemovable(DepWrite) && isCompleteOverwrite(Loc, DepLoc, *AA) && + if (isRemovable(DepWrite) && !isPossibleSelfRead(Inst, Loc, DepWrite, *AA)) { - DEBUG(dbgs() << "DSE: Remove Dead Store:\n DEAD: " - << *DepWrite << "\n KILLER: " << *Inst << '\n'); - - // Delete the store and now-dead instructions that feed it. - DeleteDeadInstruction(DepWrite, *MD); - ++NumFastStores; - MadeChange = true; - - // DeleteDeadInstruction can delete the current instruction in loop - // cases, reset BBI. - BBI = Inst; - if (BBI != BB.begin()) - --BBI; - break; + int64_t InstWriteOffset, DepWriteOffset; + OverwriteResult OR = isOverwrite(Loc, DepLoc, *AA, + DepWriteOffset, InstWriteOffset); + if (OR == OverwriteComplete) { + DEBUG(dbgs() << "DSE: Remove Dead Store:\n DEAD: " + << *DepWrite << "\n KILLER: " << *Inst << '\n'); + + // Delete the store and now-dead instructions that feed it. + DeleteDeadInstruction(DepWrite, *MD); + ++NumFastStores; + MadeChange = true; + + // DeleteDeadInstruction can delete the current instruction in loop + // cases, reset BBI. + BBI = Inst; + if (BBI != BB.begin()) + --BBI; + break; + } else if (OR == OverwriteEnd && isShortenable(DepWrite)) { + // TODO: base this on the target vector size so that if the earlier + // store was too small to get vector writes anyway then its likely + // a good idea to shorten it + // Power of 2 vector writes are probably always a bad idea to optimize + // as any store/memset/memcpy is likely using vector instructions so + // shortening it to not vector size is likely to be slower + MemIntrinsic* DepIntrinsic = cast(DepWrite); + unsigned DepWriteAlign = DepIntrinsic->getAlignment(); + if (llvm::isPowerOf2_64(InstWriteOffset) || + ((DepWriteAlign != 0) && InstWriteOffset % DepWriteAlign == 0)) { + + DEBUG(dbgs() << "DSE: Remove Dead Store:\n OW END: " + << *DepWrite << "\n KILLER (offset " + << InstWriteOffset << ", " + << DepLoc.Size << ")" + << *Inst << '\n'); + + Value* DepWriteLength = DepIntrinsic->getLength(); + Value* TrimmedLength = ConstantInt::get(DepWriteLength->getType(), + InstWriteOffset - + DepWriteOffset); + DepIntrinsic->setLength(TrimmedLength); + MadeChange = true; + } + } } // If this is a may-aliased store that is clobbering the store value, we -- cgit v1.2.3