diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp --- a/llvm/lib/Target/X86/X86ISelLowering.cpp +++ b/llvm/lib/Target/X86/X86ISelLowering.cpp @@ -31748,6 +31748,19 @@ if (supportedVectorVarShift(VT, Subtarget, ShiftOpc) || Subtarget.hasXOP()) return SDValue(); + // Attempt to fold per-element (ExtVT) shift as unpack(y,x) << zext(z) + if (((IsCst || !Subtarget.hasAVX512()) && !IsFSHR && EltSizeInBits <= 16) || + supportedVectorVarShift(ExtVT, Subtarget, ShiftOpc)) { + SDValue Z = DAG.getConstant(0, DL, VT); + SDValue RLo = DAG.getBitcast(ExtVT, getUnpackl(DAG, DL, VT, Op1, Op0)); + SDValue RHi = DAG.getBitcast(ExtVT, getUnpackh(DAG, DL, VT, Op1, Op0)); + SDValue ALo = DAG.getBitcast(ExtVT, getUnpackl(DAG, DL, VT, AmtMod, Z)); + SDValue AHi = DAG.getBitcast(ExtVT, getUnpackh(DAG, DL, VT, AmtMod, Z)); + SDValue Lo = DAG.getNode(ShiftOpc, DL, ExtVT, RLo, ALo); + SDValue Hi = DAG.getNode(ShiftOpc, DL, ExtVT, RHi, AHi); + return getPack(DAG, Subtarget, DL, VT, Lo, Hi, !IsFSHR); + } + // Attempt to fold as: // fshl(x,y,z) -> (((aext(x) << bw) | zext(y)) << (z & (bw-1))) >> bw. // fshr(x,y,z) -> (((aext(x) << bw) | zext(y)) >> (z & (bw-1))). @@ -31766,19 +31779,6 @@ return DAG.getNode(ISD::TRUNCATE, DL, VT, Res); } - // Attempt to fold per-element (ExtVT) shift as unpack(y,x) << zext(z) - if (((IsCst || !Subtarget.hasAVX512()) && !IsFSHR && EltSizeInBits <= 16) || - supportedVectorVarShift(ExtVT, Subtarget, ShiftOpc)) { - SDValue Z = DAG.getConstant(0, DL, VT); - SDValue RLo = DAG.getBitcast(ExtVT, getUnpackl(DAG, DL, VT, Op1, Op0)); - SDValue RHi = DAG.getBitcast(ExtVT, getUnpackh(DAG, DL, VT, Op1, Op0)); - SDValue ALo = DAG.getBitcast(ExtVT, getUnpackl(DAG, DL, VT, AmtMod, Z)); - SDValue AHi = DAG.getBitcast(ExtVT, getUnpackh(DAG, DL, VT, AmtMod, Z)); - SDValue Lo = DAG.getNode(ShiftOpc, DL, ExtVT, RLo, ALo); - SDValue Hi = DAG.getNode(ShiftOpc, DL, ExtVT, RHi, AHi); - return getPack(DAG, Subtarget, DL, VT, Lo, Hi, !IsFSHR); - } - // Fallback to generic expansion. return SDValue(); } diff --git a/llvm/test/CodeGen/X86/vector-fshl-128.ll b/llvm/test/CodeGen/X86/vector-fshl-128.ll --- a/llvm/test/CodeGen/X86/vector-fshl-128.ll +++ b/llvm/test/CodeGen/X86/vector-fshl-128.ll @@ -502,46 +502,47 @@ ; ; AVX2-LABEL: var_funnnel_v8i16: ; AVX2: # %bb.0: -; AVX2-NEXT: vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero -; AVX2-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero -; AVX2-NEXT: vpslld $16, %ymm0, %ymm0 -; AVX2-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4],ymm0[5],ymm1[6],ymm0[7],ymm1[8],ymm0[9],ymm1[10],ymm0[11],ymm1[12],ymm0[13],ymm1[14],ymm0[15] -; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm1 -; AVX2-NEXT: vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero -; AVX2-NEXT: vpsllvd %ymm1, %ymm0, %ymm0 -; AVX2-NEXT: vpsrld $16, %ymm0, %ymm0 -; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1 -; AVX2-NEXT: vpackusdw %xmm1, %xmm0, %xmm0 -; AVX2-NEXT: vzeroupper +; AVX2-NEXT: vpunpckhwd {{.*#+}} xmm3 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2 +; AVX2-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX2-NEXT: vpunpckhwd {{.*#+}} xmm4 = xmm2[4],xmm4[4],xmm2[5],xmm4[5],xmm2[6],xmm4[6],xmm2[7],xmm4[7] +; AVX2-NEXT: vpsllvd %xmm4, %xmm3, %xmm3 +; AVX2-NEXT: vpsrld $16, %xmm3, %xmm3 +; AVX2-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3] +; AVX2-NEXT: vpmovzxwd {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero +; AVX2-NEXT: vpsllvd %xmm1, %xmm0, %xmm0 +; AVX2-NEXT: vpsrld $16, %xmm0, %xmm0 +; AVX2-NEXT: vpackusdw %xmm3, %xmm0, %xmm0 ; AVX2-NEXT: retq ; ; AVX512F-LABEL: var_funnnel_v8i16: ; AVX512F: # %bb.0: -; AVX512F-NEXT: vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero -; AVX512F-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero -; AVX512F-NEXT: vpslld $16, %ymm0, %ymm0 -; AVX512F-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4],ymm0[5],ymm1[6],ymm0[7],ymm1[8],ymm0[9],ymm1[10],ymm0[11],ymm1[12],ymm0[13],ymm1[14],ymm0[15] -; AVX512F-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm1 -; AVX512F-NEXT: vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero -; AVX512F-NEXT: vpsllvd %ymm1, %ymm0, %ymm0 -; AVX512F-NEXT: vpsrld $16, %ymm0, %ymm0 -; AVX512F-NEXT: vpmovdw %zmm0, %ymm0 -; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 -; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: vpunpckhwd {{.*#+}} xmm3 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512F-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2 +; AVX512F-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512F-NEXT: vpunpckhwd {{.*#+}} xmm4 = xmm2[4],xmm4[4],xmm2[5],xmm4[5],xmm2[6],xmm4[6],xmm2[7],xmm4[7] +; AVX512F-NEXT: vpsllvd %xmm4, %xmm3, %xmm3 +; AVX512F-NEXT: vpsrld $16, %xmm3, %xmm3 +; AVX512F-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3] +; AVX512F-NEXT: vpmovzxwd {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero +; AVX512F-NEXT: vpsllvd %xmm1, %xmm0, %xmm0 +; AVX512F-NEXT: vpsrld $16, %xmm0, %xmm0 +; AVX512F-NEXT: vpackusdw %xmm3, %xmm0, %xmm0 ; AVX512F-NEXT: retq ; ; AVX512VL-LABEL: var_funnnel_v8i16: ; AVX512VL: # %bb.0: -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero -; AVX512VL-NEXT: vpslld $16, %ymm0, %ymm0 -; AVX512VL-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4],ymm0[5],ymm1[6],ymm0[7],ymm1[8],ymm0[9],ymm1[10],ymm0[11],ymm1[12],ymm0[13],ymm1[14],ymm0[15] -; AVX512VL-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm1 -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero -; AVX512VL-NEXT: vpsllvd %ymm1, %ymm0, %ymm0 -; AVX512VL-NEXT: vpsrld $16, %ymm0, %ymm0 -; AVX512VL-NEXT: vpmovdw %ymm0, %xmm0 -; AVX512VL-NEXT: vzeroupper +; AVX512VL-NEXT: vpunpckhwd {{.*#+}} xmm3 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512VL-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2 +; AVX512VL-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512VL-NEXT: vpunpckhwd {{.*#+}} xmm4 = xmm2[4],xmm4[4],xmm2[5],xmm4[5],xmm2[6],xmm4[6],xmm2[7],xmm4[7] +; AVX512VL-NEXT: vpsllvd %xmm4, %xmm3, %xmm3 +; AVX512VL-NEXT: vpsrld $16, %xmm3, %xmm3 +; AVX512VL-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3] +; AVX512VL-NEXT: vpmovzxwd {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero +; AVX512VL-NEXT: vpsllvd %xmm1, %xmm0, %xmm0 +; AVX512VL-NEXT: vpsrld $16, %xmm0, %xmm0 +; AVX512VL-NEXT: vpackusdw %xmm3, %xmm0, %xmm0 ; AVX512VL-NEXT: retq ; ; AVX512BW-LABEL: var_funnnel_v8i16: @@ -815,60 +816,64 @@ ; ; AVX512BW-LABEL: var_funnnel_v16i8: ; AVX512BW: # %bb.0: +; AVX512BW-NEXT: vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15] ; AVX512BW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2 -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BW-NEXT: vpsllw $8, %ymm0, %ymm0 -; AVX512BW-NEXT: vpor %ymm1, %ymm0, %ymm0 -; AVX512BW-NEXT: vpsllvw %zmm2, %zmm0, %zmm0 -; AVX512BW-NEXT: vpsrlw $8, %ymm0, %ymm0 -; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 +; AVX512BW-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512BW-NEXT: vpunpckhbw {{.*#+}} xmm4 = xmm2[8],xmm4[8],xmm2[9],xmm4[9],xmm2[10],xmm4[10],xmm2[11],xmm4[11],xmm2[12],xmm4[12],xmm2[13],xmm4[13],xmm2[14],xmm4[14],xmm2[15],xmm4[15] +; AVX512BW-NEXT: vpsllvw %zmm4, %zmm3, %zmm3 +; AVX512BW-NEXT: vpsrlw $8, %xmm3, %xmm3 +; AVX512BW-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512BW-NEXT: vpmovzxbw {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero +; AVX512BW-NEXT: vpsllvw %zmm1, %zmm0, %zmm0 +; AVX512BW-NEXT: vpsrlw $8, %xmm0, %xmm0 +; AVX512BW-NEXT: vpackuswb %xmm3, %xmm0, %xmm0 ; AVX512BW-NEXT: vzeroupper ; AVX512BW-NEXT: retq ; ; AVX512VBMI2-LABEL: var_funnnel_v16i8: ; AVX512VBMI2: # %bb.0: -; AVX512VBMI2-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1 -; AVX512VBMI2-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 -; AVX512VBMI2-NEXT: vmovdqa {{.*#+}} ymm3 = [0,64,1,65,2,66,3,67,4,68,5,69,6,70,7,71,8,72,9,73,10,74,11,75,12,76,13,77,14,78,15,79] -; AVX512VBMI2-NEXT: vpermt2b %zmm0, %zmm3, %zmm1 -; AVX512VBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm0 -; AVX512VBMI2-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512VBMI2-NEXT: vpsllvw %zmm0, %zmm1, %zmm0 -; AVX512VBMI2-NEXT: vpsrlw $8, %ymm0, %ymm0 -; AVX512VBMI2-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512VBMI2-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 +; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15] +; AVX512VBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2 +; AVX512VBMI2-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} xmm4 = xmm2[8],xmm4[8],xmm2[9],xmm4[9],xmm2[10],xmm4[10],xmm2[11],xmm4[11],xmm2[12],xmm4[12],xmm2[13],xmm4[13],xmm2[14],xmm4[14],xmm2[15],xmm4[15] +; AVX512VBMI2-NEXT: vpsllvw %zmm4, %zmm3, %zmm3 +; AVX512VBMI2-NEXT: vpsrlw $8, %xmm3, %xmm3 +; AVX512VBMI2-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512VBMI2-NEXT: vpmovzxbw {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero +; AVX512VBMI2-NEXT: vpsllvw %zmm1, %zmm0, %zmm0 +; AVX512VBMI2-NEXT: vpsrlw $8, %xmm0, %xmm0 +; AVX512VBMI2-NEXT: vpackuswb %xmm3, %xmm0, %xmm0 ; AVX512VBMI2-NEXT: vzeroupper ; AVX512VBMI2-NEXT: retq ; ; AVX512VLBW-LABEL: var_funnnel_v16i8: ; AVX512VLBW: # %bb.0: -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512VLBW-NEXT: vpsllw $8, %ymm0, %ymm0 -; AVX512VLBW-NEXT: vpor %ymm1, %ymm0, %ymm0 -; AVX512VLBW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm1 -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero -; AVX512VLBW-NEXT: vpsllvw %ymm1, %ymm0, %ymm0 -; AVX512VLBW-NEXT: vpsrlw $8, %ymm0, %ymm0 -; AVX512VLBW-NEXT: vpmovwb %ymm0, %xmm0 -; AVX512VLBW-NEXT: vzeroupper +; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15] +; AVX512VLBW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2 +; AVX512VLBW-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} xmm4 = xmm2[8],xmm4[8],xmm2[9],xmm4[9],xmm2[10],xmm4[10],xmm2[11],xmm4[11],xmm2[12],xmm4[12],xmm2[13],xmm4[13],xmm2[14],xmm4[14],xmm2[15],xmm4[15] +; AVX512VLBW-NEXT: vpsllvw %xmm4, %xmm3, %xmm3 +; AVX512VLBW-NEXT: vpsrlw $8, %xmm3, %xmm3 +; AVX512VLBW-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero +; AVX512VLBW-NEXT: vpsllvw %xmm1, %xmm0, %xmm0 +; AVX512VLBW-NEXT: vpsrlw $8, %xmm0, %xmm0 +; AVX512VLBW-NEXT: vpackuswb %xmm3, %xmm0, %xmm0 ; AVX512VLBW-NEXT: retq ; ; AVX512VLVBMI2-LABEL: var_funnnel_v16i8: ; AVX512VLVBMI2: # %bb.0: -; AVX512VLVBMI2-NEXT: # kill: def $xmm1 killed $xmm1 def $ymm1 -; AVX512VLVBMI2-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0 -; AVX512VLVBMI2-NEXT: vmovdqa {{.*#+}} ymm3 = [0,32,1,33,2,34,3,35,4,36,5,37,6,38,7,39,8,40,9,41,10,42,11,43,12,44,13,45,14,46,15,47] -; AVX512VLVBMI2-NEXT: vpermi2b %ymm0, %ymm1, %ymm3 -; AVX512VLVBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm0 -; AVX512VLVBMI2-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512VLVBMI2-NEXT: vpsllvw %ymm0, %ymm3, %ymm0 -; AVX512VLVBMI2-NEXT: vpsrlw $8, %ymm0, %ymm0 -; AVX512VLVBMI2-NEXT: vpmovwb %ymm0, %xmm0 -; AVX512VLVBMI2-NEXT: vzeroupper +; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15] +; AVX512VLVBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2 +; AVX512VLVBMI2-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} xmm4 = xmm2[8],xmm4[8],xmm2[9],xmm4[9],xmm2[10],xmm4[10],xmm2[11],xmm4[11],xmm2[12],xmm4[12],xmm2[13],xmm4[13],xmm2[14],xmm4[14],xmm2[15],xmm4[15] +; AVX512VLVBMI2-NEXT: vpsllvw %xmm4, %xmm3, %xmm3 +; AVX512VLVBMI2-NEXT: vpsrlw $8, %xmm3, %xmm3 +; AVX512VLVBMI2-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512VLVBMI2-NEXT: vpmovzxbw {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero +; AVX512VLVBMI2-NEXT: vpsllvw %xmm1, %xmm0, %xmm0 +; AVX512VLVBMI2-NEXT: vpsrlw $8, %xmm0, %xmm0 +; AVX512VLVBMI2-NEXT: vpackuswb %xmm3, %xmm0, %xmm0 ; AVX512VLVBMI2-NEXT: retq ; ; XOP-LABEL: var_funnnel_v16i8: @@ -2031,54 +2036,52 @@ ; ; AVX512BW-LABEL: constant_funnnel_v16i8: ; AVX512BW: # %bb.0: -; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm2 = [0,1,2,3,4,5,6,7,0,7,6,5,4,3,2,1] -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BW-NEXT: vpsllw $8, %ymm0, %ymm0 -; AVX512BW-NEXT: vpor %ymm1, %ymm0, %ymm0 -; AVX512BW-NEXT: vpsllvw %zmm2, %zmm0, %zmm0 -; AVX512BW-NEXT: vpsrlw $8, %ymm0, %ymm0 -; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 +; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm2 = [0,0,7,0,6,0,5,0,4,0,3,0,2,0,1,0] +; AVX512BW-NEXT: vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15] +; AVX512BW-NEXT: vpsllvw %zmm2, %zmm3, %zmm2 +; AVX512BW-NEXT: vpsrlw $8, %xmm2, %xmm2 +; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm3 = [0,0,1,0,2,0,3,0,4,0,5,0,6,0,7,0] +; AVX512BW-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512BW-NEXT: vpsllvw %zmm3, %zmm0, %zmm0 +; AVX512BW-NEXT: vpsrlw $8, %xmm0, %xmm0 +; AVX512BW-NEXT: vpackuswb %xmm2, %xmm0, %xmm0 ; AVX512BW-NEXT: vzeroupper ; AVX512BW-NEXT: retq ; ; AVX512VBMI2-LABEL: constant_funnnel_v16i8: ; AVX512VBMI2: # %bb.0: -; AVX512VBMI2-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1 -; AVX512VBMI2-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 -; AVX512VBMI2-NEXT: vmovdqa {{.*#+}} ymm2 = [0,64,1,65,2,66,3,67,4,68,5,69,6,70,7,71,8,72,9,73,10,74,11,75,12,76,13,77,14,78,15,79] -; AVX512VBMI2-NEXT: vpermt2b %zmm0, %zmm2, %zmm1 -; AVX512VBMI2-NEXT: vmovdqa {{.*#+}} ymm0 = [0,1,2,3,4,5,6,7,0,7,6,5,4,3,2,1] -; AVX512VBMI2-NEXT: vpsllvw %zmm0, %zmm1, %zmm0 -; AVX512VBMI2-NEXT: vpsrlw $8, %ymm0, %ymm0 -; AVX512VBMI2-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512VBMI2-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 +; AVX512VBMI2-NEXT: vmovdqa {{.*#+}} xmm2 = [0,0,7,0,6,0,5,0,4,0,3,0,2,0,1,0] +; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15] +; AVX512VBMI2-NEXT: vpsllvw %zmm2, %zmm3, %zmm2 +; AVX512VBMI2-NEXT: vpsrlw $8, %xmm2, %xmm2 +; AVX512VBMI2-NEXT: vmovdqa {{.*#+}} xmm3 = [0,0,1,0,2,0,3,0,4,0,5,0,6,0,7,0] +; AVX512VBMI2-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512VBMI2-NEXT: vpsllvw %zmm3, %zmm0, %zmm0 +; AVX512VBMI2-NEXT: vpsrlw $8, %xmm0, %xmm0 +; AVX512VBMI2-NEXT: vpackuswb %xmm2, %xmm0, %xmm0 ; AVX512VBMI2-NEXT: vzeroupper ; AVX512VBMI2-NEXT: retq ; ; AVX512VLBW-LABEL: constant_funnnel_v16i8: ; AVX512VLBW: # %bb.0: -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512VLBW-NEXT: vpsllw $8, %ymm0, %ymm0 -; AVX512VLBW-NEXT: vpor %ymm1, %ymm0, %ymm0 -; AVX512VLBW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0 -; AVX512VLBW-NEXT: vpsrlw $8, %ymm0, %ymm0 -; AVX512VLBW-NEXT: vpmovwb %ymm0, %xmm0 -; AVX512VLBW-NEXT: vzeroupper +; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} xmm2 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15] +; AVX512VLBW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2 +; AVX512VLBW-NEXT: vpsrlw $8, %xmm2, %xmm2 +; AVX512VLBW-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512VLBW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 +; AVX512VLBW-NEXT: vpsrlw $8, %xmm0, %xmm0 +; AVX512VLBW-NEXT: vpackuswb %xmm2, %xmm0, %xmm0 ; AVX512VLBW-NEXT: retq ; ; AVX512VLVBMI2-LABEL: constant_funnnel_v16i8: ; AVX512VLVBMI2: # %bb.0: -; AVX512VLVBMI2-NEXT: # kill: def $xmm1 killed $xmm1 def $ymm1 -; AVX512VLVBMI2-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0 -; AVX512VLVBMI2-NEXT: vmovdqa {{.*#+}} ymm2 = [0,32,1,33,2,34,3,35,4,36,5,37,6,38,7,39,8,40,9,41,10,42,11,43,12,44,13,45,14,46,15,47] -; AVX512VLVBMI2-NEXT: vpermi2b %ymm0, %ymm1, %ymm2 -; AVX512VLVBMI2-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm0 -; AVX512VLVBMI2-NEXT: vpsrlw $8, %ymm0, %ymm0 -; AVX512VLVBMI2-NEXT: vpmovwb %ymm0, %xmm0 -; AVX512VLVBMI2-NEXT: vzeroupper +; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} xmm2 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15] +; AVX512VLVBMI2-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2 +; AVX512VLVBMI2-NEXT: vpsrlw $8, %xmm2, %xmm2 +; AVX512VLVBMI2-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512VLVBMI2-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 +; AVX512VLVBMI2-NEXT: vpsrlw $8, %xmm0, %xmm0 +; AVX512VLVBMI2-NEXT: vpackuswb %xmm2, %xmm0, %xmm0 ; AVX512VLVBMI2-NEXT: retq ; ; XOP-LABEL: constant_funnnel_v16i8: diff --git a/llvm/test/CodeGen/X86/vector-fshl-256.ll b/llvm/test/CodeGen/X86/vector-fshl-256.ll --- a/llvm/test/CodeGen/X86/vector-fshl-256.ll +++ b/llvm/test/CodeGen/X86/vector-fshl-256.ll @@ -368,28 +368,32 @@ ; ; AVX512F-LABEL: var_funnnel_v16i16: ; AVX512F: # %bb.0: -; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero -; AVX512F-NEXT: vpslld $16, %zmm0, %zmm0 -; AVX512F-NEXT: vpord %zmm1, %zmm0, %zmm0 -; AVX512F-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm1 -; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512F-NEXT: vpsllvd %zmm1, %zmm0, %zmm0 -; AVX512F-NEXT: vpsrld $16, %zmm0, %zmm0 -; AVX512F-NEXT: vpmovdw %zmm0, %ymm0 +; AVX512F-NEXT: vpunpckhwd {{.*#+}} ymm3 = ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15] +; AVX512F-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2 +; AVX512F-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512F-NEXT: vpunpckhwd {{.*#+}} ymm5 = ymm2[4],ymm4[4],ymm2[5],ymm4[5],ymm2[6],ymm4[6],ymm2[7],ymm4[7],ymm2[12],ymm4[12],ymm2[13],ymm4[13],ymm2[14],ymm4[14],ymm2[15],ymm4[15] +; AVX512F-NEXT: vpsllvd %ymm5, %ymm3, %ymm3 +; AVX512F-NEXT: vpsrld $16, %ymm3, %ymm3 +; AVX512F-NEXT: vpunpcklwd {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11] +; AVX512F-NEXT: vpunpcklwd {{.*#+}} ymm1 = ymm2[0],ymm4[0],ymm2[1],ymm4[1],ymm2[2],ymm4[2],ymm2[3],ymm4[3],ymm2[8],ymm4[8],ymm2[9],ymm4[9],ymm2[10],ymm4[10],ymm2[11],ymm4[11] +; AVX512F-NEXT: vpsllvd %ymm1, %ymm0, %ymm0 +; AVX512F-NEXT: vpsrld $16, %ymm0, %ymm0 +; AVX512F-NEXT: vpackusdw %ymm3, %ymm0, %ymm0 ; AVX512F-NEXT: retq ; ; AVX512VL-LABEL: var_funnnel_v16i16: ; AVX512VL: # %bb.0: -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero -; AVX512VL-NEXT: vpslld $16, %zmm0, %zmm0 -; AVX512VL-NEXT: vpord %zmm1, %zmm0, %zmm0 -; AVX512VL-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm1 -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512VL-NEXT: vpsllvd %zmm1, %zmm0, %zmm0 -; AVX512VL-NEXT: vpsrld $16, %zmm0, %zmm0 -; AVX512VL-NEXT: vpmovdw %zmm0, %ymm0 +; AVX512VL-NEXT: vpunpckhwd {{.*#+}} ymm3 = ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15] +; AVX512VL-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2 +; AVX512VL-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512VL-NEXT: vpunpckhwd {{.*#+}} ymm5 = ymm2[4],ymm4[4],ymm2[5],ymm4[5],ymm2[6],ymm4[6],ymm2[7],ymm4[7],ymm2[12],ymm4[12],ymm2[13],ymm4[13],ymm2[14],ymm4[14],ymm2[15],ymm4[15] +; AVX512VL-NEXT: vpsllvd %ymm5, %ymm3, %ymm3 +; AVX512VL-NEXT: vpsrld $16, %ymm3, %ymm3 +; AVX512VL-NEXT: vpunpcklwd {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11] +; AVX512VL-NEXT: vpunpcklwd {{.*#+}} ymm1 = ymm2[0],ymm4[0],ymm2[1],ymm4[1],ymm2[2],ymm4[2],ymm2[3],ymm4[3],ymm2[8],ymm4[8],ymm2[9],ymm4[9],ymm2[10],ymm4[10],ymm2[11],ymm4[11] +; AVX512VL-NEXT: vpsllvd %ymm1, %ymm0, %ymm0 +; AVX512VL-NEXT: vpsrld $16, %ymm0, %ymm0 +; AVX512VL-NEXT: vpackusdw %ymm3, %ymm0, %ymm0 ; AVX512VL-NEXT: retq ; ; AVX512BW-LABEL: var_funnnel_v16i16: @@ -641,54 +645,62 @@ ; ; AVX512BW-LABEL: var_funnnel_v32i8: ; AVX512BW: # %bb.0: -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero -; AVX512BW-NEXT: vpsllw $8, %zmm0, %zmm0 -; AVX512BW-NEXT: vporq %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm1 -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero +; AVX512BW-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31] +; AVX512BW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2 +; AVX512BW-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512BW-NEXT: vpunpckhbw {{.*#+}} ymm5 = ymm2[8],ymm4[8],ymm2[9],ymm4[9],ymm2[10],ymm4[10],ymm2[11],ymm4[11],ymm2[12],ymm4[12],ymm2[13],ymm4[13],ymm2[14],ymm4[14],ymm2[15],ymm4[15],ymm2[24],ymm4[24],ymm2[25],ymm4[25],ymm2[26],ymm4[26],ymm2[27],ymm4[27],ymm2[28],ymm4[28],ymm2[29],ymm4[29],ymm2[30],ymm4[30],ymm2[31],ymm4[31] +; AVX512BW-NEXT: vpsllvw %zmm5, %zmm3, %zmm3 +; AVX512BW-NEXT: vpsrlw $8, %ymm3, %ymm3 +; AVX512BW-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23] +; AVX512BW-NEXT: vpunpcklbw {{.*#+}} ymm1 = ymm2[0],ymm4[0],ymm2[1],ymm4[1],ymm2[2],ymm4[2],ymm2[3],ymm4[3],ymm2[4],ymm4[4],ymm2[5],ymm4[5],ymm2[6],ymm4[6],ymm2[7],ymm4[7],ymm2[16],ymm4[16],ymm2[17],ymm4[17],ymm2[18],ymm4[18],ymm2[19],ymm4[19],ymm2[20],ymm4[20],ymm2[21],ymm4[21],ymm2[22],ymm4[22],ymm2[23],ymm4[23] ; AVX512BW-NEXT: vpsllvw %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpsrlw $8, %zmm0, %zmm0 -; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 +; AVX512BW-NEXT: vpsrlw $8, %ymm0, %ymm0 +; AVX512BW-NEXT: vpackuswb %ymm3, %ymm0, %ymm0 ; AVX512BW-NEXT: retq ; ; AVX512VBMI2-LABEL: var_funnnel_v32i8: ; AVX512VBMI2: # %bb.0: -; AVX512VBMI2-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1 -; AVX512VBMI2-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 -; AVX512VBMI2-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,64,1,65,2,66,3,67,4,68,5,69,6,70,7,71,8,72,9,73,10,74,11,75,12,76,13,77,14,78,15,79,16,80,17,81,18,82,19,83,20,84,21,85,22,86,23,87,24,88,25,89,26,90,27,91,28,92,29,93,30,94,31,95] -; AVX512VBMI2-NEXT: vpermi2b %zmm0, %zmm1, %zmm3 -; AVX512VBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm0 -; AVX512VBMI2-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero -; AVX512VBMI2-NEXT: vpsllvw %zmm0, %zmm3, %zmm0 -; AVX512VBMI2-NEXT: vpsrlw $8, %zmm0, %zmm0 -; AVX512VBMI2-NEXT: vpmovwb %zmm0, %ymm0 +; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31] +; AVX512VBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2 +; AVX512VBMI2-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} ymm5 = ymm2[8],ymm4[8],ymm2[9],ymm4[9],ymm2[10],ymm4[10],ymm2[11],ymm4[11],ymm2[12],ymm4[12],ymm2[13],ymm4[13],ymm2[14],ymm4[14],ymm2[15],ymm4[15],ymm2[24],ymm4[24],ymm2[25],ymm4[25],ymm2[26],ymm4[26],ymm2[27],ymm4[27],ymm2[28],ymm4[28],ymm2[29],ymm4[29],ymm2[30],ymm4[30],ymm2[31],ymm4[31] +; AVX512VBMI2-NEXT: vpsllvw %zmm5, %zmm3, %zmm3 +; AVX512VBMI2-NEXT: vpsrlw $8, %ymm3, %ymm3 +; AVX512VBMI2-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23] +; AVX512VBMI2-NEXT: vpunpcklbw {{.*#+}} ymm1 = ymm2[0],ymm4[0],ymm2[1],ymm4[1],ymm2[2],ymm4[2],ymm2[3],ymm4[3],ymm2[4],ymm4[4],ymm2[5],ymm4[5],ymm2[6],ymm4[6],ymm2[7],ymm4[7],ymm2[16],ymm4[16],ymm2[17],ymm4[17],ymm2[18],ymm4[18],ymm2[19],ymm4[19],ymm2[20],ymm4[20],ymm2[21],ymm4[21],ymm2[22],ymm4[22],ymm2[23],ymm4[23] +; AVX512VBMI2-NEXT: vpsllvw %zmm1, %zmm0, %zmm0 +; AVX512VBMI2-NEXT: vpsrlw $8, %ymm0, %ymm0 +; AVX512VBMI2-NEXT: vpackuswb %ymm3, %ymm0, %ymm0 ; AVX512VBMI2-NEXT: retq ; ; AVX512VLBW-LABEL: var_funnnel_v32i8: ; AVX512VLBW: # %bb.0: -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero -; AVX512VLBW-NEXT: vpsllw $8, %zmm0, %zmm0 -; AVX512VLBW-NEXT: vporq %zmm1, %zmm0, %zmm0 -; AVX512VLBW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm1 -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero -; AVX512VLBW-NEXT: vpsllvw %zmm1, %zmm0, %zmm0 -; AVX512VLBW-NEXT: vpsrlw $8, %zmm0, %zmm0 -; AVX512VLBW-NEXT: vpmovwb %zmm0, %ymm0 +; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31] +; AVX512VLBW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2 +; AVX512VLBW-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} ymm5 = ymm2[8],ymm4[8],ymm2[9],ymm4[9],ymm2[10],ymm4[10],ymm2[11],ymm4[11],ymm2[12],ymm4[12],ymm2[13],ymm4[13],ymm2[14],ymm4[14],ymm2[15],ymm4[15],ymm2[24],ymm4[24],ymm2[25],ymm4[25],ymm2[26],ymm4[26],ymm2[27],ymm4[27],ymm2[28],ymm4[28],ymm2[29],ymm4[29],ymm2[30],ymm4[30],ymm2[31],ymm4[31] +; AVX512VLBW-NEXT: vpsllvw %ymm5, %ymm3, %ymm3 +; AVX512VLBW-NEXT: vpsrlw $8, %ymm3, %ymm3 +; AVX512VLBW-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23] +; AVX512VLBW-NEXT: vpunpcklbw {{.*#+}} ymm1 = ymm2[0],ymm4[0],ymm2[1],ymm4[1],ymm2[2],ymm4[2],ymm2[3],ymm4[3],ymm2[4],ymm4[4],ymm2[5],ymm4[5],ymm2[6],ymm4[6],ymm2[7],ymm4[7],ymm2[16],ymm4[16],ymm2[17],ymm4[17],ymm2[18],ymm4[18],ymm2[19],ymm4[19],ymm2[20],ymm4[20],ymm2[21],ymm4[21],ymm2[22],ymm4[22],ymm2[23],ymm4[23] +; AVX512VLBW-NEXT: vpsllvw %ymm1, %ymm0, %ymm0 +; AVX512VLBW-NEXT: vpsrlw $8, %ymm0, %ymm0 +; AVX512VLBW-NEXT: vpackuswb %ymm3, %ymm0, %ymm0 ; AVX512VLBW-NEXT: retq ; ; AVX512VLVBMI2-LABEL: var_funnnel_v32i8: ; AVX512VLVBMI2: # %bb.0: -; AVX512VLVBMI2-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1 -; AVX512VLVBMI2-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 -; AVX512VLVBMI2-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,64,1,65,2,66,3,67,4,68,5,69,6,70,7,71,8,72,9,73,10,74,11,75,12,76,13,77,14,78,15,79,16,80,17,81,18,82,19,83,20,84,21,85,22,86,23,87,24,88,25,89,26,90,27,91,28,92,29,93,30,94,31,95] -; AVX512VLVBMI2-NEXT: vpermi2b %zmm0, %zmm1, %zmm3 -; AVX512VLVBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm0 -; AVX512VLVBMI2-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero -; AVX512VLVBMI2-NEXT: vpsllvw %zmm0, %zmm3, %zmm0 -; AVX512VLVBMI2-NEXT: vpsrlw $8, %zmm0, %zmm0 -; AVX512VLVBMI2-NEXT: vpmovwb %zmm0, %ymm0 +; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31] +; AVX512VLVBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2 +; AVX512VLVBMI2-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} ymm5 = ymm2[8],ymm4[8],ymm2[9],ymm4[9],ymm2[10],ymm4[10],ymm2[11],ymm4[11],ymm2[12],ymm4[12],ymm2[13],ymm4[13],ymm2[14],ymm4[14],ymm2[15],ymm4[15],ymm2[24],ymm4[24],ymm2[25],ymm4[25],ymm2[26],ymm4[26],ymm2[27],ymm4[27],ymm2[28],ymm4[28],ymm2[29],ymm4[29],ymm2[30],ymm4[30],ymm2[31],ymm4[31] +; AVX512VLVBMI2-NEXT: vpsllvw %ymm5, %ymm3, %ymm3 +; AVX512VLVBMI2-NEXT: vpsrlw $8, %ymm3, %ymm3 +; AVX512VLVBMI2-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23] +; AVX512VLVBMI2-NEXT: vpunpcklbw {{.*#+}} ymm1 = ymm2[0],ymm4[0],ymm2[1],ymm4[1],ymm2[2],ymm4[2],ymm2[3],ymm4[3],ymm2[4],ymm4[4],ymm2[5],ymm4[5],ymm2[6],ymm4[6],ymm2[7],ymm4[7],ymm2[16],ymm4[16],ymm2[17],ymm4[17],ymm2[18],ymm4[18],ymm2[19],ymm4[19],ymm2[20],ymm4[20],ymm2[21],ymm4[21],ymm2[22],ymm4[22],ymm2[23],ymm4[23] +; AVX512VLVBMI2-NEXT: vpsllvw %ymm1, %ymm0, %ymm0 +; AVX512VLVBMI2-NEXT: vpsrlw $8, %ymm0, %ymm0 +; AVX512VLVBMI2-NEXT: vpackuswb %ymm3, %ymm0, %ymm0 ; AVX512VLVBMI2-NEXT: retq ; ; XOPAVX1-LABEL: var_funnnel_v32i8: @@ -1900,46 +1912,50 @@ ; ; AVX512BW-LABEL: constant_funnnel_v32i8: ; AVX512BW: # %bb.0: -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero -; AVX512BW-NEXT: vpsllw $8, %zmm0, %zmm0 -; AVX512BW-NEXT: vporq %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0 -; AVX512BW-NEXT: vpsrlw $8, %zmm0, %zmm0 -; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 +; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm2 = [0,0,7,0,6,0,5,0,4,0,3,0,2,0,1,0,0,0,7,0,6,0,5,0,4,0,3,0,2,0,1,0] +; AVX512BW-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31] +; AVX512BW-NEXT: vpsllvw %zmm2, %zmm3, %zmm2 +; AVX512BW-NEXT: vpsrlw $8, %ymm2, %ymm2 +; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm3 = [0,0,1,0,2,0,3,0,4,0,5,0,6,0,7,0,0,0,1,0,2,0,3,0,4,0,5,0,6,0,7,0] +; AVX512BW-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23] +; AVX512BW-NEXT: vpsllvw %zmm3, %zmm0, %zmm0 +; AVX512BW-NEXT: vpsrlw $8, %ymm0, %ymm0 +; AVX512BW-NEXT: vpackuswb %ymm2, %ymm0, %ymm0 ; AVX512BW-NEXT: retq ; ; AVX512VBMI2-LABEL: constant_funnnel_v32i8: ; AVX512VBMI2: # %bb.0: -; AVX512VBMI2-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1 -; AVX512VBMI2-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 -; AVX512VBMI2-NEXT: vmovdqa64 {{.*#+}} zmm2 = [0,64,1,65,2,66,3,67,4,68,5,69,6,70,7,71,8,72,9,73,10,74,11,75,12,76,13,77,14,78,15,79,16,80,17,81,18,82,19,83,20,84,21,85,22,86,23,87,24,88,25,89,26,90,27,91,28,92,29,93,30,94,31,95] -; AVX512VBMI2-NEXT: vpermi2b %zmm0, %zmm1, %zmm2 -; AVX512VBMI2-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm0 -; AVX512VBMI2-NEXT: vpsrlw $8, %zmm0, %zmm0 -; AVX512VBMI2-NEXT: vpmovwb %zmm0, %ymm0 +; AVX512VBMI2-NEXT: vmovdqa {{.*#+}} ymm2 = [0,0,7,0,6,0,5,0,4,0,3,0,2,0,1,0,0,0,7,0,6,0,5,0,4,0,3,0,2,0,1,0] +; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31] +; AVX512VBMI2-NEXT: vpsllvw %zmm2, %zmm3, %zmm2 +; AVX512VBMI2-NEXT: vpsrlw $8, %ymm2, %ymm2 +; AVX512VBMI2-NEXT: vmovdqa {{.*#+}} ymm3 = [0,0,1,0,2,0,3,0,4,0,5,0,6,0,7,0,0,0,1,0,2,0,3,0,4,0,5,0,6,0,7,0] +; AVX512VBMI2-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23] +; AVX512VBMI2-NEXT: vpsllvw %zmm3, %zmm0, %zmm0 +; AVX512VBMI2-NEXT: vpsrlw $8, %ymm0, %ymm0 +; AVX512VBMI2-NEXT: vpackuswb %ymm2, %ymm0, %ymm0 ; AVX512VBMI2-NEXT: retq ; ; AVX512VLBW-LABEL: constant_funnnel_v32i8: ; AVX512VLBW: # %bb.0: -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero -; AVX512VLBW-NEXT: vpsllw $8, %zmm0, %zmm0 -; AVX512VLBW-NEXT: vporq %zmm1, %zmm0, %zmm0 -; AVX512VLBW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0 -; AVX512VLBW-NEXT: vpsrlw $8, %zmm0, %zmm0 -; AVX512VLBW-NEXT: vpmovwb %zmm0, %ymm0 +; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} ymm2 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31] +; AVX512VLBW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2 +; AVX512VLBW-NEXT: vpsrlw $8, %ymm2, %ymm2 +; AVX512VLBW-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23] +; AVX512VLBW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0 +; AVX512VLBW-NEXT: vpsrlw $8, %ymm0, %ymm0 +; AVX512VLBW-NEXT: vpackuswb %ymm2, %ymm0, %ymm0 ; AVX512VLBW-NEXT: retq ; ; AVX512VLVBMI2-LABEL: constant_funnnel_v32i8: ; AVX512VLVBMI2: # %bb.0: -; AVX512VLVBMI2-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1 -; AVX512VLVBMI2-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 -; AVX512VLVBMI2-NEXT: vmovdqa64 {{.*#+}} zmm2 = [0,64,1,65,2,66,3,67,4,68,5,69,6,70,7,71,8,72,9,73,10,74,11,75,12,76,13,77,14,78,15,79,16,80,17,81,18,82,19,83,20,84,21,85,22,86,23,87,24,88,25,89,26,90,27,91,28,92,29,93,30,94,31,95] -; AVX512VLVBMI2-NEXT: vpermi2b %zmm0, %zmm1, %zmm2 -; AVX512VLVBMI2-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm0 -; AVX512VLVBMI2-NEXT: vpsrlw $8, %zmm0, %zmm0 -; AVX512VLVBMI2-NEXT: vpmovwb %zmm0, %ymm0 +; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} ymm2 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31] +; AVX512VLVBMI2-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2 +; AVX512VLVBMI2-NEXT: vpsrlw $8, %ymm2, %ymm2 +; AVX512VLVBMI2-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23] +; AVX512VLVBMI2-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0 +; AVX512VLVBMI2-NEXT: vpsrlw $8, %ymm0, %ymm0 +; AVX512VLVBMI2-NEXT: vpackuswb %ymm2, %ymm0, %ymm0 ; AVX512VLVBMI2-NEXT: retq ; ; XOPAVX1-LABEL: constant_funnnel_v32i8: diff --git a/llvm/test/CodeGen/X86/vector-fshl-512.ll b/llvm/test/CodeGen/X86/vector-fshl-512.ll --- a/llvm/test/CodeGen/X86/vector-fshl-512.ll +++ b/llvm/test/CodeGen/X86/vector-fshl-512.ll @@ -134,52 +134,58 @@ define <32 x i16> @var_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %amt) nounwind { ; AVX512F-LABEL: var_funnnel_v32i16: ; AVX512F: # %bb.0: -; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm4 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero -; AVX512F-NEXT: vpslld $16, %zmm4, %zmm4 -; AVX512F-NEXT: vpord %zmm3, %zmm4, %zmm3 +; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm3 +; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm4 +; AVX512F-NEXT: vpunpckhwd {{.*#+}} ymm5 = ymm4[4],ymm3[4],ymm4[5],ymm3[5],ymm4[6],ymm3[6],ymm4[7],ymm3[7],ymm4[12],ymm3[12],ymm4[13],ymm3[13],ymm4[14],ymm3[14],ymm4[15],ymm3[15] ; AVX512F-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2 -; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero -; AVX512F-NEXT: vpsllvd %zmm4, %zmm3, %zmm3 -; AVX512F-NEXT: vpsrld $16, %zmm3, %zmm3 -; AVX512F-NEXT: vpmovdw %zmm3, %ymm3 -; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm1 -; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm0 -; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero -; AVX512F-NEXT: vpslld $16, %zmm0, %zmm0 -; AVX512F-NEXT: vpord %zmm1, %zmm0, %zmm0 -; AVX512F-NEXT: vextracti64x4 $1, %zmm2, %ymm1 -; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512F-NEXT: vpsllvd %zmm1, %zmm0, %zmm0 -; AVX512F-NEXT: vpsrld $16, %zmm0, %zmm0 -; AVX512F-NEXT: vpmovdw %zmm0, %ymm0 -; AVX512F-NEXT: vinserti64x4 $1, %ymm0, %zmm3, %zmm0 +; AVX512F-NEXT: vextracti64x4 $1, %zmm2, %ymm6 +; AVX512F-NEXT: vpxor %xmm7, %xmm7, %xmm7 +; AVX512F-NEXT: vpunpckhwd {{.*#+}} ymm8 = ymm6[4],ymm7[4],ymm6[5],ymm7[5],ymm6[6],ymm7[6],ymm6[7],ymm7[7],ymm6[12],ymm7[12],ymm6[13],ymm7[13],ymm6[14],ymm7[14],ymm6[15],ymm7[15] +; AVX512F-NEXT: vpsllvd %ymm8, %ymm5, %ymm5 +; AVX512F-NEXT: vpsrld $16, %ymm5, %ymm5 +; AVX512F-NEXT: vpunpcklwd {{.*#+}} ymm3 = ymm4[0],ymm3[0],ymm4[1],ymm3[1],ymm4[2],ymm3[2],ymm4[3],ymm3[3],ymm4[8],ymm3[8],ymm4[9],ymm3[9],ymm4[10],ymm3[10],ymm4[11],ymm3[11] +; AVX512F-NEXT: vpunpcklwd {{.*#+}} ymm4 = ymm6[0],ymm7[0],ymm6[1],ymm7[1],ymm6[2],ymm7[2],ymm6[3],ymm7[3],ymm6[8],ymm7[8],ymm6[9],ymm7[9],ymm6[10],ymm7[10],ymm6[11],ymm7[11] +; AVX512F-NEXT: vpsllvd %ymm4, %ymm3, %ymm3 +; AVX512F-NEXT: vpsrld $16, %ymm3, %ymm3 +; AVX512F-NEXT: vpackusdw %ymm5, %ymm3, %ymm3 +; AVX512F-NEXT: vpunpckhwd {{.*#+}} ymm4 = ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15] +; AVX512F-NEXT: vpunpckhwd {{.*#+}} ymm5 = ymm2[4],ymm7[4],ymm2[5],ymm7[5],ymm2[6],ymm7[6],ymm2[7],ymm7[7],ymm2[12],ymm7[12],ymm2[13],ymm7[13],ymm2[14],ymm7[14],ymm2[15],ymm7[15] +; AVX512F-NEXT: vpsllvd %ymm5, %ymm4, %ymm4 +; AVX512F-NEXT: vpsrld $16, %ymm4, %ymm4 +; AVX512F-NEXT: vpunpcklwd {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11] +; AVX512F-NEXT: vpunpcklwd {{.*#+}} ymm1 = ymm2[0],ymm7[0],ymm2[1],ymm7[1],ymm2[2],ymm7[2],ymm2[3],ymm7[3],ymm2[8],ymm7[8],ymm2[9],ymm7[9],ymm2[10],ymm7[10],ymm2[11],ymm7[11] +; AVX512F-NEXT: vpsllvd %ymm1, %ymm0, %ymm0 +; AVX512F-NEXT: vpsrld $16, %ymm0, %ymm0 +; AVX512F-NEXT: vpackusdw %ymm4, %ymm0, %ymm0 +; AVX512F-NEXT: vinserti64x4 $1, %ymm3, %zmm0, %zmm0 ; AVX512F-NEXT: retq ; ; AVX512VL-LABEL: var_funnnel_v32i16: ; AVX512VL: # %bb.0: -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm4 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero -; AVX512VL-NEXT: vpslld $16, %zmm4, %zmm4 -; AVX512VL-NEXT: vpord %zmm3, %zmm4, %zmm3 +; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm3 +; AVX512VL-NEXT: vextracti64x4 $1, %zmm1, %ymm4 +; AVX512VL-NEXT: vpunpckhwd {{.*#+}} ymm5 = ymm4[4],ymm3[4],ymm4[5],ymm3[5],ymm4[6],ymm3[6],ymm4[7],ymm3[7],ymm4[12],ymm3[12],ymm4[13],ymm3[13],ymm4[14],ymm3[14],ymm4[15],ymm3[15] ; AVX512VL-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2 -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero -; AVX512VL-NEXT: vpsllvd %zmm4, %zmm3, %zmm3 -; AVX512VL-NEXT: vpsrld $16, %zmm3, %zmm3 -; AVX512VL-NEXT: vpmovdw %zmm3, %ymm3 -; AVX512VL-NEXT: vextracti64x4 $1, %zmm1, %ymm1 -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero -; AVX512VL-NEXT: vpslld $16, %zmm0, %zmm0 -; AVX512VL-NEXT: vpord %zmm1, %zmm0, %zmm0 -; AVX512VL-NEXT: vextracti64x4 $1, %zmm2, %ymm1 -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512VL-NEXT: vpsllvd %zmm1, %zmm0, %zmm0 -; AVX512VL-NEXT: vpsrld $16, %zmm0, %zmm0 -; AVX512VL-NEXT: vpmovdw %zmm0, %ymm0 -; AVX512VL-NEXT: vinserti64x4 $1, %ymm0, %zmm3, %zmm0 +; AVX512VL-NEXT: vextracti64x4 $1, %zmm2, %ymm6 +; AVX512VL-NEXT: vpxor %xmm7, %xmm7, %xmm7 +; AVX512VL-NEXT: vpunpckhwd {{.*#+}} ymm8 = ymm6[4],ymm7[4],ymm6[5],ymm7[5],ymm6[6],ymm7[6],ymm6[7],ymm7[7],ymm6[12],ymm7[12],ymm6[13],ymm7[13],ymm6[14],ymm7[14],ymm6[15],ymm7[15] +; AVX512VL-NEXT: vpsllvd %ymm8, %ymm5, %ymm5 +; AVX512VL-NEXT: vpsrld $16, %ymm5, %ymm5 +; AVX512VL-NEXT: vpunpcklwd {{.*#+}} ymm3 = ymm4[0],ymm3[0],ymm4[1],ymm3[1],ymm4[2],ymm3[2],ymm4[3],ymm3[3],ymm4[8],ymm3[8],ymm4[9],ymm3[9],ymm4[10],ymm3[10],ymm4[11],ymm3[11] +; AVX512VL-NEXT: vpunpcklwd {{.*#+}} ymm4 = ymm6[0],ymm7[0],ymm6[1],ymm7[1],ymm6[2],ymm7[2],ymm6[3],ymm7[3],ymm6[8],ymm7[8],ymm6[9],ymm7[9],ymm6[10],ymm7[10],ymm6[11],ymm7[11] +; AVX512VL-NEXT: vpsllvd %ymm4, %ymm3, %ymm3 +; AVX512VL-NEXT: vpsrld $16, %ymm3, %ymm3 +; AVX512VL-NEXT: vpackusdw %ymm5, %ymm3, %ymm3 +; AVX512VL-NEXT: vpunpckhwd {{.*#+}} ymm4 = ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15] +; AVX512VL-NEXT: vpunpckhwd {{.*#+}} ymm5 = ymm2[4],ymm7[4],ymm2[5],ymm7[5],ymm2[6],ymm7[6],ymm2[7],ymm7[7],ymm2[12],ymm7[12],ymm2[13],ymm7[13],ymm2[14],ymm7[14],ymm2[15],ymm7[15] +; AVX512VL-NEXT: vpsllvd %ymm5, %ymm4, %ymm4 +; AVX512VL-NEXT: vpsrld $16, %ymm4, %ymm4 +; AVX512VL-NEXT: vpunpcklwd {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11] +; AVX512VL-NEXT: vpunpcklwd {{.*#+}} ymm1 = ymm2[0],ymm7[0],ymm2[1],ymm7[1],ymm2[2],ymm7[2],ymm2[3],ymm7[3],ymm2[8],ymm7[8],ymm2[9],ymm7[9],ymm2[10],ymm7[10],ymm2[11],ymm7[11] +; AVX512VL-NEXT: vpsllvd %ymm1, %ymm0, %ymm0 +; AVX512VL-NEXT: vpsrld $16, %ymm0, %ymm0 +; AVX512VL-NEXT: vpackusdw %ymm4, %ymm0, %ymm0 +; AVX512VL-NEXT: vinserti64x4 $1, %ymm3, %zmm0, %zmm0 ; AVX512VL-NEXT: retq ; ; AVX512BW-LABEL: var_funnnel_v32i16: diff --git a/llvm/test/CodeGen/X86/vector-fshr-128.ll b/llvm/test/CodeGen/X86/vector-fshr-128.ll --- a/llvm/test/CodeGen/X86/vector-fshr-128.ll +++ b/llvm/test/CodeGen/X86/vector-fshr-128.ll @@ -561,42 +561,45 @@ ; ; AVX2-LABEL: var_funnnel_v8i16: ; AVX2: # %bb.0: -; AVX2-NEXT: vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero -; AVX2-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero -; AVX2-NEXT: vpslld $16, %ymm0, %ymm0 -; AVX2-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4],ymm0[5],ymm1[6],ymm0[7],ymm1[8],ymm0[9],ymm1[10],ymm0[11],ymm1[12],ymm0[13],ymm1[14],ymm0[15] -; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm1 -; AVX2-NEXT: vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero -; AVX2-NEXT: vpsrlvd %ymm1, %ymm0, %ymm0 -; AVX2-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13,u,u,u,u,u,u,u,u,16,17,20,21,24,25,28,29,u,u,u,u,u,u,u,u] -; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3] -; AVX2-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 -; AVX2-NEXT: vzeroupper +; AVX2-NEXT: vpunpckhwd {{.*#+}} xmm3 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2 +; AVX2-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX2-NEXT: vpunpckhwd {{.*#+}} xmm5 = xmm2[4],xmm4[4],xmm2[5],xmm4[5],xmm2[6],xmm4[6],xmm2[7],xmm4[7] +; AVX2-NEXT: vpsrlvd %xmm5, %xmm3, %xmm3 +; AVX2-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1],xmm3[2],xmm4[3],xmm3[4],xmm4[5],xmm3[6],xmm4[7] +; AVX2-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3] +; AVX2-NEXT: vpmovzxwd {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero +; AVX2-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 +; AVX2-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm4[1],xmm0[2],xmm4[3],xmm0[4],xmm4[5],xmm0[6],xmm4[7] +; AVX2-NEXT: vpackusdw %xmm3, %xmm0, %xmm0 ; AVX2-NEXT: retq ; ; AVX512F-LABEL: var_funnnel_v8i16: ; AVX512F: # %bb.0: -; AVX512F-NEXT: vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero -; AVX512F-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero -; AVX512F-NEXT: vpslld $16, %ymm0, %ymm0 -; AVX512F-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4],ymm0[5],ymm1[6],ymm0[7],ymm1[8],ymm0[9],ymm1[10],ymm0[11],ymm1[12],ymm0[13],ymm1[14],ymm0[15] -; AVX512F-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm1 -; AVX512F-NEXT: vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero -; AVX512F-NEXT: vpsrlvd %ymm1, %ymm0, %ymm0 -; AVX512F-NEXT: vpmovdw %zmm0, %ymm0 -; AVX512F-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 -; AVX512F-NEXT: vzeroupper +; AVX512F-NEXT: vpunpckhwd {{.*#+}} xmm3 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512F-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2 +; AVX512F-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512F-NEXT: vpunpckhwd {{.*#+}} xmm5 = xmm2[4],xmm4[4],xmm2[5],xmm4[5],xmm2[6],xmm4[6],xmm2[7],xmm4[7] +; AVX512F-NEXT: vpsrlvd %xmm5, %xmm3, %xmm3 +; AVX512F-NEXT: vpblendw {{.*#+}} xmm3 = xmm3[0],xmm4[1],xmm3[2],xmm4[3],xmm3[4],xmm4[5],xmm3[6],xmm4[7] +; AVX512F-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3] +; AVX512F-NEXT: vpmovzxwd {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero +; AVX512F-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 +; AVX512F-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm4[1],xmm0[2],xmm4[3],xmm0[4],xmm4[5],xmm0[6],xmm4[7] +; AVX512F-NEXT: vpackusdw %xmm3, %xmm0, %xmm0 ; AVX512F-NEXT: retq ; ; AVX512VL-LABEL: var_funnnel_v8i16: ; AVX512VL: # %bb.0: -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero -; AVX512VL-NEXT: vpslld $16, %ymm0, %ymm0 -; AVX512VL-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3],ymm1[4],ymm0[5],ymm1[6],ymm0[7],ymm1[8],ymm0[9],ymm1[10],ymm0[11],ymm1[12],ymm0[13],ymm1[14],ymm0[15] -; AVX512VL-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm1 -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero -; AVX512VL-NEXT: vpsrlvd %ymm1, %ymm0, %ymm0 +; AVX512VL-NEXT: vpunpckhwd {{.*#+}} xmm3 = xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512VL-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2 +; AVX512VL-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512VL-NEXT: vpunpckhwd {{.*#+}} xmm4 = xmm2[4],xmm4[4],xmm2[5],xmm4[5],xmm2[6],xmm4[6],xmm2[7],xmm4[7] +; AVX512VL-NEXT: vpsrlvd %xmm4, %xmm3, %xmm3 +; AVX512VL-NEXT: vpunpcklwd {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3] +; AVX512VL-NEXT: vpmovzxwd {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero +; AVX512VL-NEXT: vpsrlvd %xmm1, %xmm0, %xmm0 +; AVX512VL-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm0 ; AVX512VL-NEXT: vpmovdw %ymm0, %xmm0 ; AVX512VL-NEXT: vzeroupper ; AVX512VL-NEXT: retq @@ -893,54 +896,63 @@ ; ; AVX512BW-LABEL: var_funnnel_v16i8: ; AVX512BW: # %bb.0: +; AVX512BW-NEXT: vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15] ; AVX512BW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2 -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm2 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero,xmm2[8],zero,xmm2[9],zero,xmm2[10],zero,xmm2[11],zero,xmm2[12],zero,xmm2[13],zero,xmm2[14],zero,xmm2[15],zero -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BW-NEXT: vpsllw $8, %ymm0, %ymm0 -; AVX512BW-NEXT: vpor %ymm1, %ymm0, %ymm0 -; AVX512BW-NEXT: vpsrlvw %zmm2, %zmm0, %zmm0 -; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 +; AVX512BW-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512BW-NEXT: vpunpckhbw {{.*#+}} xmm4 = xmm2[8],xmm4[8],xmm2[9],xmm4[9],xmm2[10],xmm4[10],xmm2[11],xmm4[11],xmm2[12],xmm4[12],xmm2[13],xmm4[13],xmm2[14],xmm4[14],xmm2[15],xmm4[15] +; AVX512BW-NEXT: vpsrlvw %zmm4, %zmm3, %zmm3 +; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm4 = [255,255,255,255,255,255,255,255] +; AVX512BW-NEXT: vpand %xmm4, %xmm3, %xmm3 +; AVX512BW-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512BW-NEXT: vpmovzxbw {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero +; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0 +; AVX512BW-NEXT: vpand %xmm4, %xmm0, %xmm0 +; AVX512BW-NEXT: vpackuswb %xmm3, %xmm0, %xmm0 ; AVX512BW-NEXT: vzeroupper ; AVX512BW-NEXT: retq ; ; AVX512VBMI2-LABEL: var_funnnel_v16i8: ; AVX512VBMI2: # %bb.0: -; AVX512VBMI2-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1 -; AVX512VBMI2-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 -; AVX512VBMI2-NEXT: vmovdqa {{.*#+}} ymm3 = [0,64,1,65,2,66,3,67,4,68,5,69,6,70,7,71,8,72,9,73,10,74,11,75,12,76,13,77,14,78,15,79] -; AVX512VBMI2-NEXT: vpermt2b %zmm0, %zmm3, %zmm1 -; AVX512VBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm0 -; AVX512VBMI2-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512VBMI2-NEXT: vpsrlvw %zmm0, %zmm1, %zmm0 -; AVX512VBMI2-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512VBMI2-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 +; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15] +; AVX512VBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2 +; AVX512VBMI2-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} xmm4 = xmm2[8],xmm4[8],xmm2[9],xmm4[9],xmm2[10],xmm4[10],xmm2[11],xmm4[11],xmm2[12],xmm4[12],xmm2[13],xmm4[13],xmm2[14],xmm4[14],xmm2[15],xmm4[15] +; AVX512VBMI2-NEXT: vpsrlvw %zmm4, %zmm3, %zmm3 +; AVX512VBMI2-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512VBMI2-NEXT: vpmovzxbw {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero +; AVX512VBMI2-NEXT: vpsrlvw %zmm1, %zmm0, %zmm1 +; AVX512VBMI2-NEXT: vmovdqa {{.*#+}} xmm0 = [0,2,4,6,8,10,12,14,64,66,68,70,72,74,76,78] +; AVX512VBMI2-NEXT: vpermi2b %zmm3, %zmm1, %zmm0 +; AVX512VBMI2-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0 ; AVX512VBMI2-NEXT: vzeroupper ; AVX512VBMI2-NEXT: retq ; ; AVX512VLBW-LABEL: var_funnnel_v16i8: ; AVX512VLBW: # %bb.0: -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512VLBW-NEXT: vpsllw $8, %ymm0, %ymm0 -; AVX512VLBW-NEXT: vpor %ymm1, %ymm0, %ymm0 -; AVX512VLBW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm1 -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero -; AVX512VLBW-NEXT: vpsrlvw %ymm1, %ymm0, %ymm0 +; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15] +; AVX512VLBW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2 +; AVX512VLBW-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} xmm4 = xmm2[8],xmm4[8],xmm2[9],xmm4[9],xmm2[10],xmm4[10],xmm2[11],xmm4[11],xmm2[12],xmm4[12],xmm2[13],xmm4[13],xmm2[14],xmm4[14],xmm2[15],xmm4[15] +; AVX512VLBW-NEXT: vpsrlvw %xmm4, %xmm3, %xmm3 +; AVX512VLBW-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero +; AVX512VLBW-NEXT: vpsrlvw %xmm1, %xmm0, %xmm0 +; AVX512VLBW-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm0 ; AVX512VLBW-NEXT: vpmovwb %ymm0, %xmm0 ; AVX512VLBW-NEXT: vzeroupper ; AVX512VLBW-NEXT: retq ; ; AVX512VLVBMI2-LABEL: var_funnnel_v16i8: ; AVX512VLVBMI2: # %bb.0: -; AVX512VLVBMI2-NEXT: # kill: def $xmm1 killed $xmm1 def $ymm1 -; AVX512VLVBMI2-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0 -; AVX512VLVBMI2-NEXT: vmovdqa {{.*#+}} ymm3 = [0,32,1,33,2,34,3,35,4,36,5,37,6,38,7,39,8,40,9,41,10,42,11,43,12,44,13,45,14,46,15,47] -; AVX512VLVBMI2-NEXT: vpermi2b %ymm0, %ymm1, %ymm3 -; AVX512VLVBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm0 -; AVX512VLVBMI2-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512VLVBMI2-NEXT: vpsrlvw %ymm0, %ymm3, %ymm0 +; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15] +; AVX512VLVBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2 +; AVX512VLVBMI2-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} xmm4 = xmm2[8],xmm4[8],xmm2[9],xmm4[9],xmm2[10],xmm4[10],xmm2[11],xmm4[11],xmm2[12],xmm4[12],xmm2[13],xmm4[13],xmm2[14],xmm4[14],xmm2[15],xmm4[15] +; AVX512VLVBMI2-NEXT: vpsrlvw %xmm4, %xmm3, %xmm3 +; AVX512VLVBMI2-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512VLVBMI2-NEXT: vpmovzxbw {{.*#+}} xmm1 = xmm2[0],zero,xmm2[1],zero,xmm2[2],zero,xmm2[3],zero,xmm2[4],zero,xmm2[5],zero,xmm2[6],zero,xmm2[7],zero +; AVX512VLVBMI2-NEXT: vpsrlvw %xmm1, %xmm0, %xmm0 +; AVX512VLVBMI2-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm0 ; AVX512VLVBMI2-NEXT: vpmovwb %ymm0, %xmm0 ; AVX512VLVBMI2-NEXT: vzeroupper ; AVX512VLVBMI2-NEXT: retq @@ -1974,48 +1986,51 @@ ; ; AVX512BW-LABEL: constant_funnnel_v16i8: ; AVX512BW: # %bb.0: -; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm2 = [0,1,2,3,4,5,6,7,0,7,6,5,4,3,2,1] -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512BW-NEXT: vpsllw $8, %ymm0, %ymm0 -; AVX512BW-NEXT: vpor %ymm1, %ymm0, %ymm0 -; AVX512BW-NEXT: vpsrlvw %zmm2, %zmm0, %zmm0 -; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512BW-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 +; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm2 = [0,0,7,0,6,0,5,0,4,0,3,0,2,0,1,0] +; AVX512BW-NEXT: vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15] +; AVX512BW-NEXT: vpsrlvw %zmm2, %zmm3, %zmm2 +; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm3 = [255,255,255,255,255,255,255,255] +; AVX512BW-NEXT: vpand %xmm3, %xmm2, %xmm2 +; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm4 = [0,0,1,0,2,0,3,0,4,0,5,0,6,0,7,0] +; AVX512BW-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512BW-NEXT: vpsrlvw %zmm4, %zmm0, %zmm0 +; AVX512BW-NEXT: vpand %xmm3, %xmm0, %xmm0 +; AVX512BW-NEXT: vpackuswb %xmm2, %xmm0, %xmm0 ; AVX512BW-NEXT: vzeroupper ; AVX512BW-NEXT: retq ; ; AVX512VBMI2-LABEL: constant_funnnel_v16i8: ; AVX512VBMI2: # %bb.0: -; AVX512VBMI2-NEXT: # kill: def $xmm1 killed $xmm1 def $zmm1 -; AVX512VBMI2-NEXT: # kill: def $xmm0 killed $xmm0 def $zmm0 -; AVX512VBMI2-NEXT: vmovdqa {{.*#+}} ymm2 = [0,64,1,65,2,66,3,67,4,68,5,69,6,70,7,71,8,72,9,73,10,74,11,75,12,76,13,77,14,78,15,79] -; AVX512VBMI2-NEXT: vpermt2b %zmm0, %zmm2, %zmm1 -; AVX512VBMI2-NEXT: vmovdqa {{.*#+}} ymm0 = [0,1,2,3,4,5,6,7,0,7,6,5,4,3,2,1] -; AVX512VBMI2-NEXT: vpsrlvw %zmm0, %zmm1, %zmm0 -; AVX512VBMI2-NEXT: vpmovwb %zmm0, %ymm0 -; AVX512VBMI2-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0 +; AVX512VBMI2-NEXT: vmovdqa {{.*#+}} xmm2 = [0,0,7,0,6,0,5,0,4,0,3,0,2,0,1,0] +; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} xmm3 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15] +; AVX512VBMI2-NEXT: vpsrlvw %zmm2, %zmm3, %zmm2 +; AVX512VBMI2-NEXT: vmovdqa {{.*#+}} xmm3 = [0,0,1,0,2,0,3,0,4,0,5,0,6,0,7,0] +; AVX512VBMI2-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512VBMI2-NEXT: vpsrlvw %zmm3, %zmm0, %zmm1 +; AVX512VBMI2-NEXT: vmovdqa {{.*#+}} xmm0 = [0,2,4,6,8,10,12,14,64,66,68,70,72,74,76,78] +; AVX512VBMI2-NEXT: vpermi2b %zmm2, %zmm1, %zmm0 +; AVX512VBMI2-NEXT: # kill: def $xmm0 killed $xmm0 killed $zmm0 ; AVX512VBMI2-NEXT: vzeroupper ; AVX512VBMI2-NEXT: retq ; ; AVX512VLBW-LABEL: constant_funnnel_v16i8: ; AVX512VLBW: # %bb.0: -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero -; AVX512VLBW-NEXT: vpsllw $8, %ymm0, %ymm0 -; AVX512VLBW-NEXT: vpor %ymm1, %ymm0, %ymm0 -; AVX512VLBW-NEXT: vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0 +; AVX512VLBW-NEXT: vpunpcklbw {{.*#+}} xmm2 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512VLBW-NEXT: vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2 +; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} xmm0 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15] +; AVX512VLBW-NEXT: vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 +; AVX512VLBW-NEXT: vinserti128 $1, %xmm0, %ymm2, %ymm0 ; AVX512VLBW-NEXT: vpmovwb %ymm0, %xmm0 ; AVX512VLBW-NEXT: vzeroupper ; AVX512VLBW-NEXT: retq ; ; AVX512VLVBMI2-LABEL: constant_funnnel_v16i8: ; AVX512VLVBMI2: # %bb.0: -; AVX512VLVBMI2-NEXT: # kill: def $xmm1 killed $xmm1 def $ymm1 -; AVX512VLVBMI2-NEXT: # kill: def $xmm0 killed $xmm0 def $ymm0 -; AVX512VLVBMI2-NEXT: vmovdqa {{.*#+}} ymm2 = [0,32,1,33,2,34,3,35,4,36,5,37,6,38,7,39,8,40,9,41,10,42,11,43,12,44,13,45,14,46,15,47] -; AVX512VLVBMI2-NEXT: vpermi2b %ymm0, %ymm1, %ymm2 -; AVX512VLVBMI2-NEXT: vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm0 +; AVX512VLVBMI2-NEXT: vpunpcklbw {{.*#+}} xmm2 = xmm1[0],xmm0[0],xmm1[1],xmm0[1],xmm1[2],xmm0[2],xmm1[3],xmm0[3],xmm1[4],xmm0[4],xmm1[5],xmm0[5],xmm1[6],xmm0[6],xmm1[7],xmm0[7] +; AVX512VLVBMI2-NEXT: vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2 +; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} xmm0 = xmm1[8],xmm0[8],xmm1[9],xmm0[9],xmm1[10],xmm0[10],xmm1[11],xmm0[11],xmm1[12],xmm0[12],xmm1[13],xmm0[13],xmm1[14],xmm0[14],xmm1[15],xmm0[15] +; AVX512VLVBMI2-NEXT: vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 +; AVX512VLVBMI2-NEXT: vinserti128 $1, %xmm0, %ymm2, %ymm0 ; AVX512VLVBMI2-NEXT: vpmovwb %ymm0, %xmm0 ; AVX512VLVBMI2-NEXT: vzeroupper ; AVX512VLVBMI2-NEXT: retq diff --git a/llvm/test/CodeGen/X86/vector-fshr-256.ll b/llvm/test/CodeGen/X86/vector-fshr-256.ll --- a/llvm/test/CodeGen/X86/vector-fshr-256.ll +++ b/llvm/test/CodeGen/X86/vector-fshr-256.ll @@ -398,26 +398,32 @@ ; ; AVX512F-LABEL: var_funnnel_v16i16: ; AVX512F: # %bb.0: -; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero -; AVX512F-NEXT: vpslld $16, %zmm0, %zmm0 -; AVX512F-NEXT: vpord %zmm1, %zmm0, %zmm0 -; AVX512F-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm1 -; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512F-NEXT: vpsrlvd %zmm1, %zmm0, %zmm0 -; AVX512F-NEXT: vpmovdw %zmm0, %ymm0 +; AVX512F-NEXT: vpunpckhwd {{.*#+}} ymm3 = ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15] +; AVX512F-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2 +; AVX512F-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512F-NEXT: vpunpckhwd {{.*#+}} ymm5 = ymm2[4],ymm4[4],ymm2[5],ymm4[5],ymm2[6],ymm4[6],ymm2[7],ymm4[7],ymm2[12],ymm4[12],ymm2[13],ymm4[13],ymm2[14],ymm4[14],ymm2[15],ymm4[15] +; AVX512F-NEXT: vpsrlvd %ymm5, %ymm3, %ymm3 +; AVX512F-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0],ymm4[1],ymm3[2],ymm4[3],ymm3[4],ymm4[5],ymm3[6],ymm4[7],ymm3[8],ymm4[9],ymm3[10],ymm4[11],ymm3[12],ymm4[13],ymm3[14],ymm4[15] +; AVX512F-NEXT: vpunpcklwd {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11] +; AVX512F-NEXT: vpunpcklwd {{.*#+}} ymm1 = ymm2[0],ymm4[0],ymm2[1],ymm4[1],ymm2[2],ymm4[2],ymm2[3],ymm4[3],ymm2[8],ymm4[8],ymm2[9],ymm4[9],ymm2[10],ymm4[10],ymm2[11],ymm4[11] +; AVX512F-NEXT: vpsrlvd %ymm1, %ymm0, %ymm0 +; AVX512F-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0],ymm4[1],ymm0[2],ymm4[3],ymm0[4],ymm4[5],ymm0[6],ymm4[7],ymm0[8],ymm4[9],ymm0[10],ymm4[11],ymm0[12],ymm4[13],ymm0[14],ymm4[15] +; AVX512F-NEXT: vpackusdw %ymm3, %ymm0, %ymm0 ; AVX512F-NEXT: retq ; ; AVX512VL-LABEL: var_funnnel_v16i16: ; AVX512VL: # %bb.0: -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero -; AVX512VL-NEXT: vpslld $16, %zmm0, %zmm0 -; AVX512VL-NEXT: vpord %zmm1, %zmm0, %zmm0 -; AVX512VL-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm1 -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512VL-NEXT: vpsrlvd %zmm1, %zmm0, %zmm0 -; AVX512VL-NEXT: vpmovdw %zmm0, %ymm0 +; AVX512VL-NEXT: vpunpckhwd {{.*#+}} ymm3 = ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15] +; AVX512VL-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2 +; AVX512VL-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512VL-NEXT: vpunpckhwd {{.*#+}} ymm5 = ymm2[4],ymm4[4],ymm2[5],ymm4[5],ymm2[6],ymm4[6],ymm2[7],ymm4[7],ymm2[12],ymm4[12],ymm2[13],ymm4[13],ymm2[14],ymm4[14],ymm2[15],ymm4[15] +; AVX512VL-NEXT: vpsrlvd %ymm5, %ymm3, %ymm3 +; AVX512VL-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0],ymm4[1],ymm3[2],ymm4[3],ymm3[4],ymm4[5],ymm3[6],ymm4[7],ymm3[8],ymm4[9],ymm3[10],ymm4[11],ymm3[12],ymm4[13],ymm3[14],ymm4[15] +; AVX512VL-NEXT: vpunpcklwd {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11] +; AVX512VL-NEXT: vpunpcklwd {{.*#+}} ymm1 = ymm2[0],ymm4[0],ymm2[1],ymm4[1],ymm2[2],ymm4[2],ymm2[3],ymm4[3],ymm2[8],ymm4[8],ymm2[9],ymm4[9],ymm2[10],ymm4[10],ymm2[11],ymm4[11] +; AVX512VL-NEXT: vpsrlvd %ymm1, %ymm0, %ymm0 +; AVX512VL-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0],ymm4[1],ymm0[2],ymm4[3],ymm0[4],ymm4[5],ymm0[6],ymm4[7],ymm0[8],ymm4[9],ymm0[10],ymm4[11],ymm0[12],ymm4[13],ymm0[14],ymm4[15] +; AVX512VL-NEXT: vpackusdw %ymm3, %ymm0, %ymm0 ; AVX512VL-NEXT: retq ; ; AVX512BW-LABEL: var_funnnel_v16i16: @@ -672,50 +678,63 @@ ; ; AVX512BW-LABEL: var_funnnel_v32i8: ; AVX512BW: # %bb.0: -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero -; AVX512BW-NEXT: vpsllw $8, %zmm0, %zmm0 -; AVX512BW-NEXT: vporq %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm1 -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero +; AVX512BW-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31] +; AVX512BW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2 +; AVX512BW-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512BW-NEXT: vpunpckhbw {{.*#+}} ymm5 = ymm2[8],ymm4[8],ymm2[9],ymm4[9],ymm2[10],ymm4[10],ymm2[11],ymm4[11],ymm2[12],ymm4[12],ymm2[13],ymm4[13],ymm2[14],ymm4[14],ymm2[15],ymm4[15],ymm2[24],ymm4[24],ymm2[25],ymm4[25],ymm2[26],ymm4[26],ymm2[27],ymm4[27],ymm2[28],ymm4[28],ymm2[29],ymm4[29],ymm2[30],ymm4[30],ymm2[31],ymm4[31] +; AVX512BW-NEXT: vpsrlvw %zmm5, %zmm3, %zmm3 +; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm5 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255] +; AVX512BW-NEXT: vpand %ymm5, %ymm3, %ymm3 +; AVX512BW-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23] +; AVX512BW-NEXT: vpunpcklbw {{.*#+}} ymm1 = ymm2[0],ymm4[0],ymm2[1],ymm4[1],ymm2[2],ymm4[2],ymm2[3],ymm4[3],ymm2[4],ymm4[4],ymm2[5],ymm4[5],ymm2[6],ymm4[6],ymm2[7],ymm4[7],ymm2[16],ymm4[16],ymm2[17],ymm4[17],ymm2[18],ymm4[18],ymm2[19],ymm4[19],ymm2[20],ymm4[20],ymm2[21],ymm4[21],ymm2[22],ymm4[22],ymm2[23],ymm4[23] ; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 +; AVX512BW-NEXT: vpand %ymm5, %ymm0, %ymm0 +; AVX512BW-NEXT: vpackuswb %ymm3, %ymm0, %ymm0 ; AVX512BW-NEXT: retq ; ; AVX512VBMI2-LABEL: var_funnnel_v32i8: ; AVX512VBMI2: # %bb.0: -; AVX512VBMI2-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1 -; AVX512VBMI2-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 -; AVX512VBMI2-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,64,1,65,2,66,3,67,4,68,5,69,6,70,7,71,8,72,9,73,10,74,11,75,12,76,13,77,14,78,15,79,16,80,17,81,18,82,19,83,20,84,21,85,22,86,23,87,24,88,25,89,26,90,27,91,28,92,29,93,30,94,31,95] -; AVX512VBMI2-NEXT: vpermi2b %zmm0, %zmm1, %zmm3 -; AVX512VBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm0 -; AVX512VBMI2-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero -; AVX512VBMI2-NEXT: vpsrlvw %zmm0, %zmm3, %zmm0 -; AVX512VBMI2-NEXT: vpmovwb %zmm0, %ymm0 +; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31] +; AVX512VBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2 +; AVX512VBMI2-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} ymm5 = ymm2[8],ymm4[8],ymm2[9],ymm4[9],ymm2[10],ymm4[10],ymm2[11],ymm4[11],ymm2[12],ymm4[12],ymm2[13],ymm4[13],ymm2[14],ymm4[14],ymm2[15],ymm4[15],ymm2[24],ymm4[24],ymm2[25],ymm4[25],ymm2[26],ymm4[26],ymm2[27],ymm4[27],ymm2[28],ymm4[28],ymm2[29],ymm4[29],ymm2[30],ymm4[30],ymm2[31],ymm4[31] +; AVX512VBMI2-NEXT: vpsrlvw %zmm5, %zmm3, %zmm3 +; AVX512VBMI2-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23] +; AVX512VBMI2-NEXT: vpunpcklbw {{.*#+}} ymm1 = ymm2[0],ymm4[0],ymm2[1],ymm4[1],ymm2[2],ymm4[2],ymm2[3],ymm4[3],ymm2[4],ymm4[4],ymm2[5],ymm4[5],ymm2[6],ymm4[6],ymm2[7],ymm4[7],ymm2[16],ymm4[16],ymm2[17],ymm4[17],ymm2[18],ymm4[18],ymm2[19],ymm4[19],ymm2[20],ymm4[20],ymm2[21],ymm4[21],ymm2[22],ymm4[22],ymm2[23],ymm4[23] +; AVX512VBMI2-NEXT: vpsrlvw %zmm1, %zmm0, %zmm1 +; AVX512VBMI2-NEXT: vmovdqa {{.*#+}} ymm0 = [0,2,4,6,8,10,12,14,64,66,68,70,72,74,76,78,16,18,20,22,24,26,28,30,80,82,84,86,88,90,92,94] +; AVX512VBMI2-NEXT: vpermi2b %zmm3, %zmm1, %zmm0 +; AVX512VBMI2-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0 ; AVX512VBMI2-NEXT: retq ; ; AVX512VLBW-LABEL: var_funnnel_v32i8: ; AVX512VLBW: # %bb.0: -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero -; AVX512VLBW-NEXT: vpsllw $8, %zmm0, %zmm0 -; AVX512VLBW-NEXT: vporq %zmm1, %zmm0, %zmm0 -; AVX512VLBW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm1 -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero -; AVX512VLBW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0 -; AVX512VLBW-NEXT: vpmovwb %zmm0, %ymm0 +; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31] +; AVX512VLBW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2 +; AVX512VLBW-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} ymm5 = ymm2[8],ymm4[8],ymm2[9],ymm4[9],ymm2[10],ymm4[10],ymm2[11],ymm4[11],ymm2[12],ymm4[12],ymm2[13],ymm4[13],ymm2[14],ymm4[14],ymm2[15],ymm4[15],ymm2[24],ymm4[24],ymm2[25],ymm4[25],ymm2[26],ymm4[26],ymm2[27],ymm4[27],ymm2[28],ymm4[28],ymm2[29],ymm4[29],ymm2[30],ymm4[30],ymm2[31],ymm4[31] +; AVX512VLBW-NEXT: vpsrlvw %ymm5, %ymm3, %ymm3 +; AVX512VLBW-NEXT: vmovdqa {{.*#+}} ymm5 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255] +; AVX512VLBW-NEXT: vpand %ymm5, %ymm3, %ymm3 +; AVX512VLBW-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23] +; AVX512VLBW-NEXT: vpunpcklbw {{.*#+}} ymm1 = ymm2[0],ymm4[0],ymm2[1],ymm4[1],ymm2[2],ymm4[2],ymm2[3],ymm4[3],ymm2[4],ymm4[4],ymm2[5],ymm4[5],ymm2[6],ymm4[6],ymm2[7],ymm4[7],ymm2[16],ymm4[16],ymm2[17],ymm4[17],ymm2[18],ymm4[18],ymm2[19],ymm4[19],ymm2[20],ymm4[20],ymm2[21],ymm4[21],ymm2[22],ymm4[22],ymm2[23],ymm4[23] +; AVX512VLBW-NEXT: vpsrlvw %ymm1, %ymm0, %ymm0 +; AVX512VLBW-NEXT: vpand %ymm5, %ymm0, %ymm0 +; AVX512VLBW-NEXT: vpackuswb %ymm3, %ymm0, %ymm0 ; AVX512VLBW-NEXT: retq ; ; AVX512VLVBMI2-LABEL: var_funnnel_v32i8: ; AVX512VLVBMI2: # %bb.0: -; AVX512VLVBMI2-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1 -; AVX512VLVBMI2-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 -; AVX512VLVBMI2-NEXT: vmovdqa64 {{.*#+}} zmm3 = [0,64,1,65,2,66,3,67,4,68,5,69,6,70,7,71,8,72,9,73,10,74,11,75,12,76,13,77,14,78,15,79,16,80,17,81,18,82,19,83,20,84,21,85,22,86,23,87,24,88,25,89,26,90,27,91,28,92,29,93,30,94,31,95] -; AVX512VLVBMI2-NEXT: vpermi2b %zmm0, %zmm1, %zmm3 -; AVX512VLVBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm0 -; AVX512VLVBMI2-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero -; AVX512VLVBMI2-NEXT: vpsrlvw %zmm0, %zmm3, %zmm0 -; AVX512VLVBMI2-NEXT: vpmovwb %zmm0, %ymm0 +; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31] +; AVX512VLVBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2 +; AVX512VLVBMI2-NEXT: vpxor %xmm4, %xmm4, %xmm4 +; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} ymm5 = ymm2[8],ymm4[8],ymm2[9],ymm4[9],ymm2[10],ymm4[10],ymm2[11],ymm4[11],ymm2[12],ymm4[12],ymm2[13],ymm4[13],ymm2[14],ymm4[14],ymm2[15],ymm4[15],ymm2[24],ymm4[24],ymm2[25],ymm4[25],ymm2[26],ymm4[26],ymm2[27],ymm4[27],ymm2[28],ymm4[28],ymm2[29],ymm4[29],ymm2[30],ymm4[30],ymm2[31],ymm4[31] +; AVX512VLVBMI2-NEXT: vpsrlvw %ymm5, %ymm3, %ymm3 +; AVX512VLVBMI2-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23] +; AVX512VLVBMI2-NEXT: vpunpcklbw {{.*#+}} ymm1 = ymm2[0],ymm4[0],ymm2[1],ymm4[1],ymm2[2],ymm4[2],ymm2[3],ymm4[3],ymm2[4],ymm4[4],ymm2[5],ymm4[5],ymm2[6],ymm4[6],ymm2[7],ymm4[7],ymm2[16],ymm4[16],ymm2[17],ymm4[17],ymm2[18],ymm4[18],ymm2[19],ymm4[19],ymm2[20],ymm4[20],ymm2[21],ymm4[21],ymm2[22],ymm4[22],ymm2[23],ymm4[23] +; AVX512VLVBMI2-NEXT: vpsrlvw %ymm1, %ymm0, %ymm1 +; AVX512VLVBMI2-NEXT: vmovdqa {{.*#+}} ymm0 = [0,2,4,6,8,10,12,14,32,34,36,38,40,42,44,46,16,18,20,22,24,26,28,30,48,50,52,54,56,58,60,62] +; AVX512VLVBMI2-NEXT: vpermi2b %ymm3, %ymm1, %ymm0 ; AVX512VLVBMI2-NEXT: retq ; ; XOPAVX1-LABEL: var_funnnel_v32i8: @@ -1709,42 +1728,51 @@ ; ; AVX512BW-LABEL: constant_funnnel_v32i8: ; AVX512BW: # %bb.0: -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero -; AVX512BW-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero -; AVX512BW-NEXT: vpsllw $8, %zmm0, %zmm0 -; AVX512BW-NEXT: vporq %zmm1, %zmm0, %zmm0 -; AVX512BW-NEXT: vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0 -; AVX512BW-NEXT: vpmovwb %zmm0, %ymm0 +; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm2 = [0,0,7,0,6,0,5,0,4,0,3,0,2,0,1,0,0,0,7,0,6,0,5,0,4,0,3,0,2,0,1,0] +; AVX512BW-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31] +; AVX512BW-NEXT: vpsrlvw %zmm2, %zmm3, %zmm2 +; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm3 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255] +; AVX512BW-NEXT: vpand %ymm3, %ymm2, %ymm2 +; AVX512BW-NEXT: vmovdqa {{.*#+}} ymm4 = [0,0,1,0,2,0,3,0,4,0,5,0,6,0,7,0,0,0,1,0,2,0,3,0,4,0,5,0,6,0,7,0] +; AVX512BW-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23] +; AVX512BW-NEXT: vpsrlvw %zmm4, %zmm0, %zmm0 +; AVX512BW-NEXT: vpand %ymm3, %ymm0, %ymm0 +; AVX512BW-NEXT: vpackuswb %ymm2, %ymm0, %ymm0 ; AVX512BW-NEXT: retq ; ; AVX512VBMI2-LABEL: constant_funnnel_v32i8: ; AVX512VBMI2: # %bb.0: -; AVX512VBMI2-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1 -; AVX512VBMI2-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 -; AVX512VBMI2-NEXT: vmovdqa64 {{.*#+}} zmm2 = [0,64,1,65,2,66,3,67,4,68,5,69,6,70,7,71,8,72,9,73,10,74,11,75,12,76,13,77,14,78,15,79,16,80,17,81,18,82,19,83,20,84,21,85,22,86,23,87,24,88,25,89,26,90,27,91,28,92,29,93,30,94,31,95] -; AVX512VBMI2-NEXT: vpermi2b %zmm0, %zmm1, %zmm2 -; AVX512VBMI2-NEXT: vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm0 -; AVX512VBMI2-NEXT: vpmovwb %zmm0, %ymm0 +; AVX512VBMI2-NEXT: vmovdqa {{.*#+}} ymm2 = [0,0,7,0,6,0,5,0,4,0,3,0,2,0,1,0,0,0,7,0,6,0,5,0,4,0,3,0,2,0,1,0] +; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31] +; AVX512VBMI2-NEXT: vpsrlvw %zmm2, %zmm3, %zmm2 +; AVX512VBMI2-NEXT: vmovdqa {{.*#+}} ymm3 = [0,0,1,0,2,0,3,0,4,0,5,0,6,0,7,0,0,0,1,0,2,0,3,0,4,0,5,0,6,0,7,0] +; AVX512VBMI2-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23] +; AVX512VBMI2-NEXT: vpsrlvw %zmm3, %zmm0, %zmm1 +; AVX512VBMI2-NEXT: vmovdqa {{.*#+}} ymm0 = [0,2,4,6,8,10,12,14,64,66,68,70,72,74,76,78,16,18,20,22,24,26,28,30,80,82,84,86,88,90,92,94] +; AVX512VBMI2-NEXT: vpermi2b %zmm2, %zmm1, %zmm0 +; AVX512VBMI2-NEXT: # kill: def $ymm0 killed $ymm0 killed $zmm0 ; AVX512VBMI2-NEXT: retq ; ; AVX512VLBW-LABEL: constant_funnnel_v32i8: ; AVX512VLBW: # %bb.0: -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero,ymm1[16],zero,ymm1[17],zero,ymm1[18],zero,ymm1[19],zero,ymm1[20],zero,ymm1[21],zero,ymm1[22],zero,ymm1[23],zero,ymm1[24],zero,ymm1[25],zero,ymm1[26],zero,ymm1[27],zero,ymm1[28],zero,ymm1[29],zero,ymm1[30],zero,ymm1[31],zero -; AVX512VLBW-NEXT: vpmovzxbw {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero,ymm0[16],zero,ymm0[17],zero,ymm0[18],zero,ymm0[19],zero,ymm0[20],zero,ymm0[21],zero,ymm0[22],zero,ymm0[23],zero,ymm0[24],zero,ymm0[25],zero,ymm0[26],zero,ymm0[27],zero,ymm0[28],zero,ymm0[29],zero,ymm0[30],zero,ymm0[31],zero -; AVX512VLBW-NEXT: vpsllw $8, %zmm0, %zmm0 -; AVX512VLBW-NEXT: vporq %zmm1, %zmm0, %zmm0 -; AVX512VLBW-NEXT: vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0 -; AVX512VLBW-NEXT: vpmovwb %zmm0, %ymm0 +; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} ymm2 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31] +; AVX512VLBW-NEXT: vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2 +; AVX512VLBW-NEXT: vmovdqa {{.*#+}} ymm3 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255] +; AVX512VLBW-NEXT: vpand %ymm3, %ymm2, %ymm2 +; AVX512VLBW-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23] +; AVX512VLBW-NEXT: vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0 +; AVX512VLBW-NEXT: vpand %ymm3, %ymm0, %ymm0 +; AVX512VLBW-NEXT: vpackuswb %ymm2, %ymm0, %ymm0 ; AVX512VLBW-NEXT: retq ; ; AVX512VLVBMI2-LABEL: constant_funnnel_v32i8: ; AVX512VLVBMI2: # %bb.0: -; AVX512VLVBMI2-NEXT: # kill: def $ymm1 killed $ymm1 def $zmm1 -; AVX512VLVBMI2-NEXT: # kill: def $ymm0 killed $ymm0 def $zmm0 -; AVX512VLVBMI2-NEXT: vmovdqa64 {{.*#+}} zmm2 = [0,64,1,65,2,66,3,67,4,68,5,69,6,70,7,71,8,72,9,73,10,74,11,75,12,76,13,77,14,78,15,79,16,80,17,81,18,82,19,83,20,84,21,85,22,86,23,87,24,88,25,89,26,90,27,91,28,92,29,93,30,94,31,95] -; AVX512VLVBMI2-NEXT: vpermi2b %zmm0, %zmm1, %zmm2 -; AVX512VLVBMI2-NEXT: vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm0 -; AVX512VLVBMI2-NEXT: vpmovwb %zmm0, %ymm0 +; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} ymm2 = ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15],ymm1[24],ymm0[24],ymm1[25],ymm0[25],ymm1[26],ymm0[26],ymm1[27],ymm0[27],ymm1[28],ymm0[28],ymm1[29],ymm0[29],ymm1[30],ymm0[30],ymm1[31],ymm0[31] +; AVX512VLVBMI2-NEXT: vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm2, %ymm2 +; AVX512VLVBMI2-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[16],ymm0[16],ymm1[17],ymm0[17],ymm1[18],ymm0[18],ymm1[19],ymm0[19],ymm1[20],ymm0[20],ymm1[21],ymm0[21],ymm1[22],ymm0[22],ymm1[23],ymm0[23] +; AVX512VLVBMI2-NEXT: vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm1 +; AVX512VLVBMI2-NEXT: vmovdqa {{.*#+}} ymm0 = [0,2,4,6,8,10,12,14,32,34,36,38,40,42,44,46,16,18,20,22,24,26,28,30,48,50,52,54,56,58,60,62] +; AVX512VLVBMI2-NEXT: vpermi2b %ymm2, %ymm1, %ymm0 ; AVX512VLVBMI2-NEXT: retq ; ; XOPAVX1-LABEL: constant_funnnel_v32i8: diff --git a/llvm/test/CodeGen/X86/vector-fshr-512.ll b/llvm/test/CodeGen/X86/vector-fshr-512.ll --- a/llvm/test/CodeGen/X86/vector-fshr-512.ll +++ b/llvm/test/CodeGen/X86/vector-fshr-512.ll @@ -138,48 +138,58 @@ define <32 x i16> @var_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %amt) nounwind { ; AVX512F-LABEL: var_funnnel_v32i16: ; AVX512F: # %bb.0: -; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm4 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero -; AVX512F-NEXT: vpslld $16, %zmm4, %zmm4 -; AVX512F-NEXT: vpord %zmm3, %zmm4, %zmm3 +; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm3 +; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm4 +; AVX512F-NEXT: vpunpckhwd {{.*#+}} ymm5 = ymm4[4],ymm3[4],ymm4[5],ymm3[5],ymm4[6],ymm3[6],ymm4[7],ymm3[7],ymm4[12],ymm3[12],ymm4[13],ymm3[13],ymm4[14],ymm3[14],ymm4[15],ymm3[15] ; AVX512F-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2 -; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero -; AVX512F-NEXT: vpsrlvd %zmm4, %zmm3, %zmm3 -; AVX512F-NEXT: vpmovdw %zmm3, %ymm3 -; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm1 -; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm0 -; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero -; AVX512F-NEXT: vpslld $16, %zmm0, %zmm0 -; AVX512F-NEXT: vpord %zmm1, %zmm0, %zmm0 -; AVX512F-NEXT: vextracti64x4 $1, %zmm2, %ymm1 -; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512F-NEXT: vpsrlvd %zmm1, %zmm0, %zmm0 -; AVX512F-NEXT: vpmovdw %zmm0, %ymm0 -; AVX512F-NEXT: vinserti64x4 $1, %ymm0, %zmm3, %zmm0 +; AVX512F-NEXT: vextracti64x4 $1, %zmm2, %ymm6 +; AVX512F-NEXT: vpxor %xmm7, %xmm7, %xmm7 +; AVX512F-NEXT: vpunpckhwd {{.*#+}} ymm8 = ymm6[4],ymm7[4],ymm6[5],ymm7[5],ymm6[6],ymm7[6],ymm6[7],ymm7[7],ymm6[12],ymm7[12],ymm6[13],ymm7[13],ymm6[14],ymm7[14],ymm6[15],ymm7[15] +; AVX512F-NEXT: vpsrlvd %ymm8, %ymm5, %ymm5 +; AVX512F-NEXT: vpblendw {{.*#+}} ymm5 = ymm5[0],ymm7[1],ymm5[2],ymm7[3],ymm5[4],ymm7[5],ymm5[6],ymm7[7],ymm5[8],ymm7[9],ymm5[10],ymm7[11],ymm5[12],ymm7[13],ymm5[14],ymm7[15] +; AVX512F-NEXT: vpunpcklwd {{.*#+}} ymm3 = ymm4[0],ymm3[0],ymm4[1],ymm3[1],ymm4[2],ymm3[2],ymm4[3],ymm3[3],ymm4[8],ymm3[8],ymm4[9],ymm3[9],ymm4[10],ymm3[10],ymm4[11],ymm3[11] +; AVX512F-NEXT: vpunpcklwd {{.*#+}} ymm4 = ymm6[0],ymm7[0],ymm6[1],ymm7[1],ymm6[2],ymm7[2],ymm6[3],ymm7[3],ymm6[8],ymm7[8],ymm6[9],ymm7[9],ymm6[10],ymm7[10],ymm6[11],ymm7[11] +; AVX512F-NEXT: vpsrlvd %ymm4, %ymm3, %ymm3 +; AVX512F-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0],ymm7[1],ymm3[2],ymm7[3],ymm3[4],ymm7[5],ymm3[6],ymm7[7],ymm3[8],ymm7[9],ymm3[10],ymm7[11],ymm3[12],ymm7[13],ymm3[14],ymm7[15] +; AVX512F-NEXT: vpackusdw %ymm5, %ymm3, %ymm3 +; AVX512F-NEXT: vpunpckhwd {{.*#+}} ymm4 = ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15] +; AVX512F-NEXT: vpunpckhwd {{.*#+}} ymm5 = ymm2[4],ymm7[4],ymm2[5],ymm7[5],ymm2[6],ymm7[6],ymm2[7],ymm7[7],ymm2[12],ymm7[12],ymm2[13],ymm7[13],ymm2[14],ymm7[14],ymm2[15],ymm7[15] +; AVX512F-NEXT: vpsrlvd %ymm5, %ymm4, %ymm4 +; AVX512F-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0],ymm7[1],ymm4[2],ymm7[3],ymm4[4],ymm7[5],ymm4[6],ymm7[7],ymm4[8],ymm7[9],ymm4[10],ymm7[11],ymm4[12],ymm7[13],ymm4[14],ymm7[15] +; AVX512F-NEXT: vpunpcklwd {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11] +; AVX512F-NEXT: vpunpcklwd {{.*#+}} ymm1 = ymm2[0],ymm7[0],ymm2[1],ymm7[1],ymm2[2],ymm7[2],ymm2[3],ymm7[3],ymm2[8],ymm7[8],ymm2[9],ymm7[9],ymm2[10],ymm7[10],ymm2[11],ymm7[11] +; AVX512F-NEXT: vpsrlvd %ymm1, %ymm0, %ymm0 +; AVX512F-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0],ymm7[1],ymm0[2],ymm7[3],ymm0[4],ymm7[5],ymm0[6],ymm7[7],ymm0[8],ymm7[9],ymm0[10],ymm7[11],ymm0[12],ymm7[13],ymm0[14],ymm7[15] +; AVX512F-NEXT: vpackusdw %ymm4, %ymm0, %ymm0 +; AVX512F-NEXT: vinserti64x4 $1, %ymm3, %zmm0, %zmm0 ; AVX512F-NEXT: retq ; ; AVX512VL-LABEL: var_funnnel_v32i16: ; AVX512VL: # %bb.0: -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm3 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm4 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero -; AVX512VL-NEXT: vpslld $16, %zmm4, %zmm4 -; AVX512VL-NEXT: vpord %zmm3, %zmm4, %zmm3 +; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm3 +; AVX512VL-NEXT: vextracti64x4 $1, %zmm1, %ymm4 +; AVX512VL-NEXT: vpunpckhwd {{.*#+}} ymm5 = ymm4[4],ymm3[4],ymm4[5],ymm3[5],ymm4[6],ymm3[6],ymm4[7],ymm3[7],ymm4[12],ymm3[12],ymm4[13],ymm3[13],ymm4[14],ymm3[14],ymm4[15],ymm3[15] ; AVX512VL-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2 -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm4 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero -; AVX512VL-NEXT: vpsrlvd %zmm4, %zmm3, %zmm3 -; AVX512VL-NEXT: vpmovdw %zmm3, %ymm3 -; AVX512VL-NEXT: vextracti64x4 $1, %zmm1, %ymm1 -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm0 = ymm0[0],zero,ymm0[1],zero,ymm0[2],zero,ymm0[3],zero,ymm0[4],zero,ymm0[5],zero,ymm0[6],zero,ymm0[7],zero,ymm0[8],zero,ymm0[9],zero,ymm0[10],zero,ymm0[11],zero,ymm0[12],zero,ymm0[13],zero,ymm0[14],zero,ymm0[15],zero -; AVX512VL-NEXT: vpslld $16, %zmm0, %zmm0 -; AVX512VL-NEXT: vpord %zmm1, %zmm0, %zmm0 -; AVX512VL-NEXT: vextracti64x4 $1, %zmm2, %ymm1 -; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm1 = ymm1[0],zero,ymm1[1],zero,ymm1[2],zero,ymm1[3],zero,ymm1[4],zero,ymm1[5],zero,ymm1[6],zero,ymm1[7],zero,ymm1[8],zero,ymm1[9],zero,ymm1[10],zero,ymm1[11],zero,ymm1[12],zero,ymm1[13],zero,ymm1[14],zero,ymm1[15],zero -; AVX512VL-NEXT: vpsrlvd %zmm1, %zmm0, %zmm0 -; AVX512VL-NEXT: vpmovdw %zmm0, %ymm0 -; AVX512VL-NEXT: vinserti64x4 $1, %ymm0, %zmm3, %zmm0 +; AVX512VL-NEXT: vextracti64x4 $1, %zmm2, %ymm6 +; AVX512VL-NEXT: vpxor %xmm7, %xmm7, %xmm7 +; AVX512VL-NEXT: vpunpckhwd {{.*#+}} ymm8 = ymm6[4],ymm7[4],ymm6[5],ymm7[5],ymm6[6],ymm7[6],ymm6[7],ymm7[7],ymm6[12],ymm7[12],ymm6[13],ymm7[13],ymm6[14],ymm7[14],ymm6[15],ymm7[15] +; AVX512VL-NEXT: vpsrlvd %ymm8, %ymm5, %ymm5 +; AVX512VL-NEXT: vpblendw {{.*#+}} ymm5 = ymm5[0],ymm7[1],ymm5[2],ymm7[3],ymm5[4],ymm7[5],ymm5[6],ymm7[7],ymm5[8],ymm7[9],ymm5[10],ymm7[11],ymm5[12],ymm7[13],ymm5[14],ymm7[15] +; AVX512VL-NEXT: vpunpcklwd {{.*#+}} ymm3 = ymm4[0],ymm3[0],ymm4[1],ymm3[1],ymm4[2],ymm3[2],ymm4[3],ymm3[3],ymm4[8],ymm3[8],ymm4[9],ymm3[9],ymm4[10],ymm3[10],ymm4[11],ymm3[11] +; AVX512VL-NEXT: vpunpcklwd {{.*#+}} ymm4 = ymm6[0],ymm7[0],ymm6[1],ymm7[1],ymm6[2],ymm7[2],ymm6[3],ymm7[3],ymm6[8],ymm7[8],ymm6[9],ymm7[9],ymm6[10],ymm7[10],ymm6[11],ymm7[11] +; AVX512VL-NEXT: vpsrlvd %ymm4, %ymm3, %ymm3 +; AVX512VL-NEXT: vpblendw {{.*#+}} ymm3 = ymm3[0],ymm7[1],ymm3[2],ymm7[3],ymm3[4],ymm7[5],ymm3[6],ymm7[7],ymm3[8],ymm7[9],ymm3[10],ymm7[11],ymm3[12],ymm7[13],ymm3[14],ymm7[15] +; AVX512VL-NEXT: vpackusdw %ymm5, %ymm3, %ymm3 +; AVX512VL-NEXT: vpunpckhwd {{.*#+}} ymm4 = ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15] +; AVX512VL-NEXT: vpunpckhwd {{.*#+}} ymm5 = ymm2[4],ymm7[4],ymm2[5],ymm7[5],ymm2[6],ymm7[6],ymm2[7],ymm7[7],ymm2[12],ymm7[12],ymm2[13],ymm7[13],ymm2[14],ymm7[14],ymm2[15],ymm7[15] +; AVX512VL-NEXT: vpsrlvd %ymm5, %ymm4, %ymm4 +; AVX512VL-NEXT: vpblendw {{.*#+}} ymm4 = ymm4[0],ymm7[1],ymm4[2],ymm7[3],ymm4[4],ymm7[5],ymm4[6],ymm7[7],ymm4[8],ymm7[9],ymm4[10],ymm7[11],ymm4[12],ymm7[13],ymm4[14],ymm7[15] +; AVX512VL-NEXT: vpunpcklwd {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11] +; AVX512VL-NEXT: vpunpcklwd {{.*#+}} ymm1 = ymm2[0],ymm7[0],ymm2[1],ymm7[1],ymm2[2],ymm7[2],ymm2[3],ymm7[3],ymm2[8],ymm7[8],ymm2[9],ymm7[9],ymm2[10],ymm7[10],ymm2[11],ymm7[11] +; AVX512VL-NEXT: vpsrlvd %ymm1, %ymm0, %ymm0 +; AVX512VL-NEXT: vpblendw {{.*#+}} ymm0 = ymm0[0],ymm7[1],ymm0[2],ymm7[3],ymm0[4],ymm7[5],ymm0[6],ymm7[7],ymm0[8],ymm7[9],ymm0[10],ymm7[11],ymm0[12],ymm7[13],ymm0[14],ymm7[15] +; AVX512VL-NEXT: vpackusdw %ymm4, %ymm0, %ymm0 +; AVX512VL-NEXT: vinserti64x4 $1, %ymm3, %zmm0, %zmm0 ; AVX512VL-NEXT: retq ; ; AVX512BW-LABEL: var_funnnel_v32i16: