diff options
| -rw-r--r-- | llvm/lib/Target/X86/X86ISelLowering.cpp | 8 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll | 19 |
2 files changed, 12 insertions, 15 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp index 6319c9cc223..60085493913 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.cpp +++ b/llvm/lib/Target/X86/X86ISelLowering.cpp @@ -33225,6 +33225,14 @@ bool X86TargetLowering::SimplifyDemandedVectorEltsForTargetNode( } break; } + case X86ISD::VPERMV: { + SDValue Mask = Op.getOperand(0); + APInt MaskUndef, MaskZero; + if (SimplifyDemandedVectorElts(Mask, DemandedElts, MaskUndef, MaskZero, TLO, + Depth + 1)) + return true; + break; + } case X86ISD::PSHUFB: case X86ISD::VPERMILPV: { // TODO - simplify other variable shuffle masks. diff --git a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll index f2f45fc6700..1d416edbfda 100644 --- a/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll +++ b/llvm/test/CodeGen/X86/vector-shuffle-combining-avx2.ll @@ -343,21 +343,10 @@ define <8 x float> @combine_permps_as_permpd(<8 x float> %a) { } define <8 x float> @combine_permps_as_vpermilps(<8 x float> %a, i32 %a1) { -; X86-LABEL: combine_permps_as_vpermilps: -; X86: # %bb.0: -; X86-NEXT: vmovss {{.*#+}} xmm1 = mem[0],zero,zero,zero -; X86-NEXT: vblendps {{.*#+}} ymm1 = ymm1[0],mem[1,2,3,4,5,6,7] -; X86-NEXT: vpermps %ymm0, %ymm1, %ymm0 -; X86-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,1,2,3,4,5,6,7] -; X86-NEXT: retl -; -; X64-LABEL: combine_permps_as_vpermilps: -; X64: # %bb.0: -; X64-NEXT: vmovd %edi, %xmm1 -; X64-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0],mem[1,2,3,4,5,6,7] -; X64-NEXT: vpermd %ymm0, %ymm1, %ymm0 -; X64-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[1,1,2,3,4,5,6,7] -; X64-NEXT: retq +; CHECK-LABEL: combine_permps_as_vpermilps: +; CHECK: # %bb.0: +; CHECK-NEXT: vpermilps {{.*#+}} ymm0 = ymm0[2,2,1,0,7,6,5,4] +; CHECK-NEXT: ret{{[l|q]}} %1 = insertelement <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4>, i32 %a1, i32 0 %2 = call <8 x float> @llvm.x86.avx2.permps(<8 x float> %a, <8 x i32> %1) %3 = shufflevector <8 x float> %2, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> |
