diff options
| -rw-r--r-- | llvm/lib/Target/X86/X86ISelLowering.cpp | 10 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/widened-broadcast.ll | 20 |
2 files changed, 11 insertions, 19 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp index 3cd1704c9c6..01636128df0 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.cpp +++ b/llvm/lib/Target/X86/X86ISelLowering.cpp @@ -10483,9 +10483,13 @@ static SDValue lowerVectorShuffleAsBroadcast(const SDLoc &DL, MVT VT, // We only support broadcasting from 128-bit vectors to minimize the // number of patterns we need to deal with in isel. So extract down to - // 128-bits. - if (SrcVT.getSizeInBits() > 128) - V = extract128BitVector(V, 0, DAG, DL); + // 128-bits, removing as many bitcasts as possible. + if (SrcVT.getSizeInBits() > 128) { + MVT ExtVT = MVT::getVectorVT(SrcVT.getScalarType(), + 128 / SrcVT.getScalarSizeInBits()); + V = extract128BitVector(peekThroughBitcasts(V), 0, DAG, DL); + V = DAG.getBitcast(ExtVT, V); + } return DAG.getBitcast(VT, DAG.getNode(Opcode, DL, BroadcastVT, V)); } diff --git a/llvm/test/CodeGen/X86/widened-broadcast.ll b/llvm/test/CodeGen/X86/widened-broadcast.ll index 564371c7321..3de39f72022 100644 --- a/llvm/test/CodeGen/X86/widened-broadcast.ll +++ b/llvm/test/CodeGen/X86/widened-broadcast.ll @@ -526,22 +526,10 @@ define <32 x i8> @load_splat_32i8_32i8_01230123012301230123012301230123(<32 x i8 ; SSE-NEXT: movdqa %xmm0, %xmm1 ; SSE-NEXT: retq ; -; AVX1-LABEL: load_splat_32i8_32i8_01230123012301230123012301230123: -; AVX1: # %bb.0: # %entry -; AVX1-NEXT: vbroadcastss (%rdi), %ymm0 -; AVX1-NEXT: retq -; -; AVX2-LABEL: load_splat_32i8_32i8_01230123012301230123012301230123: -; AVX2: # %bb.0: # %entry -; AVX2-NEXT: vmovaps (%rdi), %ymm0 -; AVX2-NEXT: vbroadcastss %xmm0, %ymm0 -; AVX2-NEXT: retq -; -; AVX512-LABEL: load_splat_32i8_32i8_01230123012301230123012301230123: -; AVX512: # %bb.0: # %entry -; AVX512-NEXT: vmovaps (%rdi), %ymm0 -; AVX512-NEXT: vbroadcastss %xmm0, %ymm0 -; AVX512-NEXT: retq +; AVX-LABEL: load_splat_32i8_32i8_01230123012301230123012301230123: +; AVX: # %bb.0: # %entry +; AVX-NEXT: vbroadcastss (%rdi), %ymm0 +; AVX-NEXT: retq entry: %ld = load <32 x i8>, <32 x i8>* %ptr %ret = shufflevector <32 x i8> %ld, <32 x i8> undef, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3> |
