diff options
| -rw-r--r-- | llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp | 14 | ||||
| -rw-r--r-- | llvm/test/Transforms/InstCombine/insertelement-bitcast.ll | 20 |
2 files changed, 26 insertions, 8 deletions
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp b/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp index 28d5f187c3d..e76f41b4794 100644 --- a/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp @@ -867,6 +867,20 @@ Instruction *InstCombiner::visitInsertElementInst(InsertElementInst &IE) { if (isa<UndefValue>(ScalarOp) || isa<UndefValue>(IdxOp)) replaceInstUsesWith(IE, VecOp); + // If the vector and scalar are both bitcast from the same element type, do + // the insert in that source type followed by bitcast. + Value *VecSrc, *ScalarSrc; + if (match(VecOp, m_BitCast(m_Value(VecSrc))) && + match(ScalarOp, m_BitCast(m_Value(ScalarSrc))) && + (VecOp->hasOneUse() || ScalarOp->hasOneUse()) && + VecSrc->getType()->isVectorTy() && !ScalarSrc->getType()->isVectorTy() && + VecSrc->getType()->getVectorElementType() == ScalarSrc->getType()) { + // inselt (bitcast VecSrc), (bitcast ScalarSrc), IdxOp --> + // bitcast (inselt VecSrc, ScalarSrc, IdxOp) + Value *NewInsElt = Builder.CreateInsertElement(VecSrc, ScalarSrc, IdxOp); + return new BitCastInst(NewInsElt, IE.getType()); + } + // If the inserted element was extracted from some other vector and both // indexes are constant, try to turn this into a shuffle. uint64_t InsertedIdx, ExtractedIdx; diff --git a/llvm/test/Transforms/InstCombine/insertelement-bitcast.ll b/llvm/test/Transforms/InstCombine/insertelement-bitcast.ll index 92d2aeeff18..5f698d015d7 100644 --- a/llvm/test/Transforms/InstCombine/insertelement-bitcast.ll +++ b/llvm/test/Transforms/InstCombine/insertelement-bitcast.ll @@ -6,9 +6,8 @@ declare void @use_v4f32(<4 x float>) define <4 x float> @bitcast_inselt(i32 %x, <4 x i32> %v) { ; CHECK-LABEL: @bitcast_inselt( -; CHECK-NEXT: [[XF:%.*]] = bitcast i32 [[X:%.*]] to float -; CHECK-NEXT: [[VF:%.*]] = bitcast <4 x i32> [[V:%.*]] to <4 x float> -; CHECK-NEXT: [[R:%.*]] = insertelement <4 x float> [[VF]], float [[XF]], i32 0 +; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x i32> [[V:%.*]], i32 [[X:%.*]], i32 0 +; CHECK-NEXT: [[R:%.*]] = bitcast <4 x i32> [[TMP1]] to <4 x float> ; CHECK-NEXT: ret <4 x float> [[R]] ; %xf = bitcast i32 %x to float @@ -21,8 +20,8 @@ define <4 x float> @bitcast_inselt_use1(i32 %x, <4 x i32> %v) { ; CHECK-LABEL: @bitcast_inselt_use1( ; CHECK-NEXT: [[XF:%.*]] = bitcast i32 [[X:%.*]] to float ; CHECK-NEXT: call void @use_f32(float [[XF]]) -; CHECK-NEXT: [[VF:%.*]] = bitcast <4 x i32> [[V:%.*]] to <4 x float> -; CHECK-NEXT: [[R:%.*]] = insertelement <4 x float> [[VF]], float [[XF]], i32 1 +; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x i32> [[V:%.*]], i32 [[X]], i32 1 +; CHECK-NEXT: [[R:%.*]] = bitcast <4 x i32> [[TMP1]] to <4 x float> ; CHECK-NEXT: ret <4 x float> [[R]] ; %xf = bitcast i32 %x to float @@ -34,10 +33,10 @@ define <4 x float> @bitcast_inselt_use1(i32 %x, <4 x i32> %v) { define <4 x float> @bitcast_inselt_use2(i32 %x, <4 x i32> %v, i32 %index) { ; CHECK-LABEL: @bitcast_inselt_use2( -; CHECK-NEXT: [[XF:%.*]] = bitcast i32 [[X:%.*]] to float ; CHECK-NEXT: [[VF:%.*]] = bitcast <4 x i32> [[V:%.*]] to <4 x float> ; CHECK-NEXT: call void @use_v4f32(<4 x float> [[VF]]) -; CHECK-NEXT: [[R:%.*]] = insertelement <4 x float> [[VF]], float [[XF]], i32 [[INDEX:%.*]] +; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x i32> [[V]], i32 [[X:%.*]], i32 [[INDEX:%.*]] +; CHECK-NEXT: [[R:%.*]] = bitcast <4 x i32> [[TMP1]] to <4 x float> ; CHECK-NEXT: ret <4 x float> [[R]] ; %xf = bitcast i32 %x to float @@ -47,6 +46,8 @@ define <4 x float> @bitcast_inselt_use2(i32 %x, <4 x i32> %v, i32 %index) { ret <4 x float> %r } +; Negative test - 2 extra uses would require an extra instruction. + define <4 x float> @bitcast_inselt_use3(i32 %x, <4 x i32> %v) { ; CHECK-LABEL: @bitcast_inselt_use3( ; CHECK-NEXT: [[XF:%.*]] = bitcast i32 [[X:%.*]] to float @@ -64,6 +65,8 @@ define <4 x float> @bitcast_inselt_use3(i32 %x, <4 x i32> %v) { ret <4 x float> %r } +; Negative test - wrong source type for vector. + define <2 x float> @bitcast_inselt_wrong_bitcast1(i32 %x, i64 %notv) { ; CHECK-LABEL: @bitcast_inselt_wrong_bitcast1( ; CHECK-NEXT: [[XF:%.*]] = bitcast i32 [[X:%.*]] to float @@ -77,6 +80,8 @@ define <2 x float> @bitcast_inselt_wrong_bitcast1(i32 %x, i64 %notv) { ret <2 x float> %r } +; Negative test - wrong source type for scalar. + define <2 x float> @bitcast_inselt_wrong_bitcast2(<2 x i16> %notscalar, <2 x i32> %v) { ; CHECK-LABEL: @bitcast_inselt_wrong_bitcast2( ; CHECK-NEXT: [[XF:%.*]] = bitcast <2 x i16> [[NOTSCALAR:%.*]] to float @@ -89,4 +94,3 @@ define <2 x float> @bitcast_inselt_wrong_bitcast2(<2 x i16> %notscalar, <2 x i32 %r = insertelement <2 x float> %vf, float %xf, i32 0 ret <2 x float> %r } - |
