diff options
| -rw-r--r-- | llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp | 10 | ||||
| -rw-r--r-- | llvm/test/Transforms/InstCombine/extractelement.ll | 18 |
2 files changed, 17 insertions, 11 deletions
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp b/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp index 24e278d..09c04e2 100644 --- a/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp @@ -192,11 +192,9 @@ Instruction *InstCombinerImpl::foldBitcastExtElt(ExtractElementInst &Ext) { // If we are casting an integer to vector and extracting a portion, that is // a shift-right and truncate. - // TODO: If no shift is needed, allow extra use? // TODO: Allow FP dest type by casting the trunc to FP? if (X->getType()->isIntegerTy() && DestTy->isIntegerTy() && - isDesirableIntType(X->getType()->getPrimitiveSizeInBits()) && - Ext.getVectorOperand()->hasOneUse()) { + isDesirableIntType(X->getType()->getPrimitiveSizeInBits())) { assert(isa<FixedVectorType>(Ext.getVectorOperand()->getType()) && "Expected fixed vector type for bitcast from scalar integer"); @@ -206,8 +204,10 @@ Instruction *InstCombinerImpl::foldBitcastExtElt(ExtractElementInst &Ext) { if (IsBigEndian) ExtIndexC = NumElts.getKnownMinValue() - 1 - ExtIndexC; unsigned ShiftAmountC = ExtIndexC * DestTy->getPrimitiveSizeInBits(); - Value *Lshr = Builder.CreateLShr(X, ShiftAmountC, "extelt.offset"); - return new TruncInst(Lshr, DestTy); + if (!ShiftAmountC || Ext.getVectorOperand()->hasOneUse()) { + Value *Lshr = Builder.CreateLShr(X, ShiftAmountC, "extelt.offset"); + return new TruncInst(Lshr, DestTy); + } } if (!X->getType()->isVectorTy()) diff --git a/llvm/test/Transforms/InstCombine/extractelement.ll b/llvm/test/Transforms/InstCombine/extractelement.ll index 49d2965..d2ac127 100644 --- a/llvm/test/Transforms/InstCombine/extractelement.ll +++ b/llvm/test/Transforms/InstCombine/extractelement.ll @@ -433,14 +433,20 @@ define i8 @bitcast_scalar_index_variable(i32 %x, i64 %y) { ret i8 %r } -; negative test - no extra uses +; extra use is ok if we don't need a shift define i8 @bitcast_scalar_index0_use(i64 %x) { -; ANY-LABEL: @bitcast_scalar_index0_use( -; ANY-NEXT: [[V:%.*]] = bitcast i64 [[X:%.*]] to <8 x i8> -; ANY-NEXT: call void @use(<8 x i8> [[V]]) -; ANY-NEXT: [[R:%.*]] = extractelement <8 x i8> [[V]], i64 0 -; ANY-NEXT: ret i8 [[R]] +; LE-LABEL: @bitcast_scalar_index0_use( +; LE-NEXT: [[V:%.*]] = bitcast i64 [[X:%.*]] to <8 x i8> +; LE-NEXT: call void @use(<8 x i8> [[V]]) +; LE-NEXT: [[R:%.*]] = trunc i64 [[X]] to i8 +; LE-NEXT: ret i8 [[R]] +; +; BE-LABEL: @bitcast_scalar_index0_use( +; BE-NEXT: [[V:%.*]] = bitcast i64 [[X:%.*]] to <8 x i8> +; BE-NEXT: call void @use(<8 x i8> [[V]]) +; BE-NEXT: [[R:%.*]] = extractelement <8 x i8> [[V]], i64 0 +; BE-NEXT: ret i8 [[R]] ; %v = bitcast i64 %x to <8 x i8> call void @use(<8 x i8> %v) |