diff options
| -rw-r--r-- | llvm/lib/Target/X86/X86ISelLowering.cpp | 7 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/vector-shuffle-128-unpck.ll | 12 |
2 files changed, 18 insertions, 1 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp index 49d30c7cbae..b7446002922 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.cpp +++ b/llvm/lib/Target/X86/X86ISelLowering.cpp @@ -9850,11 +9850,16 @@ static bool is128BitUnpackShuffleMask(ArrayRef<int> Mask) { MVT EltVT = MVT::getIntegerVT(128 / Mask.size()); MVT VT = MVT::getVectorVT(EltVT, Mask.size()); + // We can't assume a canonical shuffle mask, so try the commuted version too. + SmallVector<int, 4> CommutedMask(Mask.begin(), Mask.end()); + ShuffleVectorSDNode::commuteMask(CommutedMask); + // Match any of unary/binary or low/high. for (unsigned i = 0; i != 4; ++i) { SmallVector<int, 16> UnpackMask; createUnpackShuffleMask(VT, UnpackMask, (i >> 1) % 2, i % 2); - if (isTargetShuffleEquivalent(Mask, UnpackMask)) + if (isTargetShuffleEquivalent(Mask, UnpackMask) || + isTargetShuffleEquivalent(CommutedMask, UnpackMask)) return true; } return false; diff --git a/llvm/test/CodeGen/X86/vector-shuffle-128-unpck.ll b/llvm/test/CodeGen/X86/vector-shuffle-128-unpck.ll index 47d9c41e019..58e4c10232f 100644 --- a/llvm/test/CodeGen/X86/vector-shuffle-128-unpck.ll +++ b/llvm/test/CodeGen/X86/vector-shuffle-128-unpck.ll @@ -221,3 +221,15 @@ define <16 x i8> @unpckl_unary_extracted_v32i8(<32 x i8> %x) { ret <16 x i8> %r } +; This would infinite loop because we did not recognize the unpack shuffle mask in commuted form. + +define <8 x i32> @extract_unpckl_v8i32(<8 x i32> %a) { +; ALL-LABEL: extract_unpckl_v8i32: +; ALL: # %bb.0: +; ALL-NEXT: vextractf128 $1, %ymm0, %xmm1 +; ALL-NEXT: vunpcklps {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1] +; ALL-NEXT: retq + %shuffle = shufflevector <8 x i32> %a, <8 x i32> undef, <8 x i32> <i32 4, i32 undef, i32 5, i32 1, i32 undef, i32 undef, i32 undef, i32 undef> + ret <8 x i32> %shuffle +} + |
