diff options
Diffstat (limited to 'llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp')
| -rw-r--r-- | llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp | 40 |
1 files changed, 28 insertions, 12 deletions
diff --git a/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp b/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp index 2e36c50b75fc..9afbe0e9a2a5 100644 --- a/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp +++ b/llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp @@ -178,9 +178,9 @@ public: } void addStore(int64_t OffsetFromFirst, StoreInst *SI) { - int64_t StoreSize = DL.getTypeStoreSize(SI->getOperand(0)->getType()); - - addRange(OffsetFromFirst, StoreSize, SI->getPointerOperand(), + TypeSize StoreSize = DL.getTypeStoreSize(SI->getOperand(0)->getType()); + assert(!StoreSize.isScalable() && "Can't track scalable-typed stores"); + addRange(OffsetFromFirst, StoreSize.getFixedSize(), SI->getPointerOperand(), SI->getAlign().value(), SI); } @@ -371,6 +371,11 @@ Instruction *MemCpyOptPass::tryMergingIntoMemset(Instruction *StartInst, Value *ByteVal) { const DataLayout &DL = StartInst->getModule()->getDataLayout(); + // We can't track scalable types + if (StoreInst *SI = dyn_cast<StoreInst>(StartInst)) + if (DL.getTypeStoreSize(SI->getOperand(0)->getType()).isScalable()) + return nullptr; + // Okay, so we now have a single store that can be splatable. Scan to find // all subsequent stores of the same value to offset from the same pointer. // Join these together into ranges, so we can decide whether contiguous blocks @@ -426,6 +431,10 @@ Instruction *MemCpyOptPass::tryMergingIntoMemset(Instruction *StartInst, if (DL.isNonIntegralPointerType(StoredVal->getType()->getScalarType())) break; + // We can't track ranges involving scalable types. + if (DL.getTypeStoreSize(StoredVal->getType()).isScalable()) + break; + // Check to see if this stored value is of the same byte-splattable value. Value *StoredByte = isBytewiseValue(StoredVal, DL); if (isa<UndefValue>(ByteVal) && StoredByte) @@ -859,7 +868,7 @@ bool MemCpyOptPass::processMemSet(MemSetInst *MSI, BasicBlock::iterator &BBI) { /// the call write its result directly into the destination of the memcpy. bool MemCpyOptPass::performCallSlotOptzn(Instruction *cpyLoad, Instruction *cpyStore, Value *cpyDest, - Value *cpySrc, uint64_t cpyLen, + Value *cpySrc, TypeSize cpySize, Align cpyAlign, CallInst *C) { // The general transformation to keep in mind is // @@ -875,6 +884,10 @@ bool MemCpyOptPass::performCallSlotOptzn(Instruction *cpyLoad, // src only holds uninitialized values at the moment of the call, meaning that // the memcpy can be discarded rather than moved. + // We can't optimize scalable types. + if (cpySize.isScalable()) + return false; + // Lifetime marks shouldn't be operated on. if (Function *F = C->getCalledFunction()) if (F->isIntrinsic() && F->getIntrinsicID() == Intrinsic::lifetime_start) @@ -893,13 +906,13 @@ bool MemCpyOptPass::performCallSlotOptzn(Instruction *cpyLoad, uint64_t srcSize = DL.getTypeAllocSize(srcAlloca->getAllocatedType()) * srcArraySize->getZExtValue(); - if (cpyLen < srcSize) + if (cpySize < srcSize) return false; // Check that accessing the first srcSize bytes of dest will not cause a // trap. Otherwise the transform is invalid since it might cause a trap // to occur earlier than it otherwise would. - if (!isDereferenceableAndAlignedPointer(cpyDest, Align(1), APInt(64, cpyLen), + if (!isDereferenceableAndAlignedPointer(cpyDest, Align(1), APInt(64, cpySize), DL, C, DT)) return false; @@ -1452,9 +1465,10 @@ bool MemCpyOptPass::processMemCpy(MemCpyInst *M, BasicBlock::iterator &BBI) { // of conservatively taking the minimum? Align Alignment = std::min(M->getDestAlign().valueOrOne(), M->getSourceAlign().valueOrOne()); - if (performCallSlotOptzn(M, M, M->getDest(), M->getSource(), - CopySize->getZExtValue(), Alignment, - C)) { + if (performCallSlotOptzn( + M, M, M->getDest(), M->getSource(), + TypeSize::getFixed(CopySize->getZExtValue()), Alignment, + C)) { LLVM_DEBUG(dbgs() << "Performed call slot optimization:\n" << " call: " << *C << "\n" << " memcpy: " << *M << "\n"); @@ -1509,7 +1523,8 @@ bool MemCpyOptPass::processMemCpy(MemCpyInst *M, BasicBlock::iterator &BBI) { Align Alignment = std::min(M->getDestAlign().valueOrOne(), M->getSourceAlign().valueOrOne()); if (performCallSlotOptzn(M, M, M->getDest(), M->getSource(), - CopySize->getZExtValue(), Alignment, C)) { + TypeSize::getFixed(CopySize->getZExtValue()), + Alignment, C)) { eraseInstruction(M); ++NumMemCpyInstr; return true; @@ -1584,7 +1599,7 @@ bool MemCpyOptPass::processByValArgument(CallBase &CB, unsigned ArgNo) { // Find out what feeds this byval argument. Value *ByValArg = CB.getArgOperand(ArgNo); Type *ByValTy = CB.getParamByValType(ArgNo); - uint64_t ByValSize = DL.getTypeAllocSize(ByValTy); + TypeSize ByValSize = DL.getTypeAllocSize(ByValTy); MemoryLocation Loc(ByValArg, LocationSize::precise(ByValSize)); MemCpyInst *MDep = nullptr; if (EnableMemorySSA) { @@ -1612,7 +1627,8 @@ bool MemCpyOptPass::processByValArgument(CallBase &CB, unsigned ArgNo) { // The length of the memcpy must be larger or equal to the size of the byval. ConstantInt *C1 = dyn_cast<ConstantInt>(MDep->getLength()); - if (!C1 || C1->getValue().getZExtValue() < ByValSize) + if (!C1 || !TypeSize::isKnownGE( + TypeSize::getFixed(C1->getValue().getZExtValue()), ByValSize)) return false; // Get the alignment of the byval. If the call doesn't specify the alignment, |