; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512F
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefixes=AVX512VL
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX512BW
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vbmi,+avx512vbmi2 | FileCheck %s --check-prefixes=AVX512VBMI2
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefixes=AVX512VLBW
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vbmi,+avx512vbmi2,+avx512vl | FileCheck %s --check-prefixes=AVX512VLVBMI2
declare <8 x i64> @llvm.fshl.v8i64(<8 x i64>, <8 x i64>, <8 x i64>)
declare <16 x i32> @llvm.fshl.v16i32(<16 x i32>, <16 x i32>, <16 x i32>)
declare <32 x i16> @llvm.fshl.v32i16(<32 x i16>, <32 x i16>, <32 x i16>)
declare <64 x i8> @llvm.fshl.v64i8(<64 x i8>, <64 x i8>, <64 x i8>)
;
; Variable Shifts
;
define <8 x i64> @var_funnnel_v8i64(<8 x i64> %x, <8 x i64> %y, <8 x i64> %amt) nounwind {
; AVX512F-LABEL: var_funnnel_v8i64:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
; AVX512F-NEXT: vpandnq %zmm3, %zmm2, %zmm4
; AVX512F-NEXT: vpsrlq $1, %zmm1, %zmm1
; AVX512F-NEXT: vpsrlvq %zmm4, %zmm1, %zmm1
; AVX512F-NEXT: vpandq %zmm3, %zmm2, %zmm2
; AVX512F-NEXT: vpsllvq %zmm2, %zmm0, %zmm0
; AVX512F-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: var_funnnel_v8i64:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
; AVX512VL-NEXT: vpandnq %zmm3, %zmm2, %zmm4
; AVX512VL-NEXT: vpsrlq $1, %zmm1, %zmm1
; AVX512VL-NEXT: vpsrlvq %zmm4, %zmm1, %zmm1
; AVX512VL-NEXT: vpandq %zmm3, %zmm2, %zmm2
; AVX512VL-NEXT: vpsllvq %zmm2, %zmm0, %zmm0
; AVX512VL-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: var_funnnel_v8i64:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
; AVX512BW-NEXT: vpandnq %zmm3, %zmm2, %zmm4
; AVX512BW-NEXT: vpsrlq $1, %zmm1, %zmm1
; AVX512BW-NEXT: vpsrlvq %zmm4, %zmm1, %zmm1
; AVX512BW-NEXT: vpandq %zmm3, %zmm2, %zmm2
; AVX512BW-NEXT: vpsllvq %zmm2, %zmm0, %zmm0
; AVX512BW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VBMI2-LABEL: var_funnnel_v8i64:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpshldvq %zmm2, %zmm1, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLBW-LABEL: var_funnnel_v8i64:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpbroadcastq {{.*#+}} zmm3 = [63,63,63,63,63,63,63,63]
; AVX512VLBW-NEXT: vpandnq %zmm3, %zmm2, %zmm4
; AVX512VLBW-NEXT: vpsrlq $1, %zmm1, %zmm1
; AVX512VLBW-NEXT: vpsrlvq %zmm4, %zmm1, %zmm1
; AVX512VLBW-NEXT: vpandq %zmm3, %zmm2, %zmm2
; AVX512VLBW-NEXT: vpsllvq %zmm2, %zmm0, %zmm0
; AVX512VLBW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VLVBMI2-LABEL: var_funnnel_v8i64:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpshldvq %zmm2, %zmm1, %zmm0
; AVX512VLVBMI2-NEXT: retq
%res = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %x, <8 x i64> %y, <8 x i64> %amt)
ret <8 x i64> %res
}
define <16 x i32> @var_funnnel_v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> %amt) nounwind {
; AVX512F-LABEL: var_funnnel_v16i32:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
; AVX512F-NEXT: vpandnd %zmm3, %zmm2, %zmm4
; AVX512F-NEXT: vpsrld $1, %zmm1, %zmm1
; AVX512F-NEXT: vpsrlvd %zmm4, %zmm1, %zmm1
; AVX512F-NEXT: vpandd %zmm3, %zmm2, %zmm2
; AVX512F-NEXT: vpsllvd %zmm2, %zmm0, %zmm0
; AVX512F-NEXT: vpord %zmm1, %zmm0, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: var_funnnel_v16i32:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
; AVX512VL-NEXT: vpandnd %zmm3, %zmm2, %zmm4
; AVX512VL-NEXT: vpsrld $1, %zmm1, %zmm1
; AVX512VL-NEXT: vpsrlvd %zmm4, %zmm1, %zmm1
; AVX512VL-NEXT: vpandd %zmm3, %zmm2, %zmm2
; AVX512VL-NEXT: vpsllvd %zmm2, %zmm0, %zmm0
; AVX512VL-NEXT: vpord %zmm1, %zmm0, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: var_funnnel_v16i32:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
; AVX512BW-NEXT: vpandnd %zmm3, %zmm2, %zmm4
; AVX512BW-NEXT: vpsrld $1, %zmm1, %zmm1
; AVX512BW-NEXT: vpsrlvd %zmm4, %zmm1, %zmm1
; AVX512BW-NEXT: vpandd %zmm3, %zmm2, %zmm2
; AVX512BW-NEXT: vpsllvd %zmm2, %zmm0, %zmm0
; AVX512BW-NEXT: vpord %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VBMI2-LABEL: var_funnnel_v16i32:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpshldvd %zmm2, %zmm1, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLBW-LABEL: var_funnnel_v16i32:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpbroadcastd {{.*#+}} zmm3 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31]
; AVX512VLBW-NEXT: vpandnd %zmm3, %zmm2, %zmm4
; AVX512VLBW-NEXT: vpsrld $1, %zmm1, %zmm1
; AVX512VLBW-NEXT: vpsrlvd %zmm4, %zmm1, %zmm1
; AVX512VLBW-NEXT: vpandd %zmm3, %zmm2, %zmm2
; AVX512VLBW-NEXT: vpsllvd %zmm2, %zmm0, %zmm0
; AVX512VLBW-NEXT: vpord %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VLVBMI2-LABEL: var_funnnel_v16i32:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpshldvd %zmm2, %zmm1, %zmm0
; AVX512VLVBMI2-NEXT: retq
%res = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> %amt)
ret <16 x i32> %res
}
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: 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: 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: 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: retq
;
; AVX512BW-LABEL: var_funnnel_v32i16:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512BW-NEXT: vpandnq %zmm3, %zmm2, %zmm4
; AVX512BW-NEXT: vpsrlw $1, %zmm1, %zmm1
; AVX512BW-NEXT: vpsrlvw %zmm4, %zmm1, %zmm1
; AVX512BW-NEXT: vpandq %zmm3, %zmm2, %zmm2
; AVX512BW-NEXT: vpsllvw %zmm2, %zmm0, %zmm0
; AVX512BW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VBMI2-LABEL: var_funnnel_v32i16:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpshldvw %zmm2, %zmm1, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLBW-LABEL: var_funnnel_v32i16:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512VLBW-NEXT: vpandnq %zmm3, %zmm2, %zmm4
; AVX512VLBW-NEXT: vpsrlw $1, %zmm1, %zmm1
; AVX512VLBW-NEXT: vpsrlvw %zmm4, %zmm1, %zmm1
; AVX512VLBW-NEXT: vpandq %zmm3, %zmm2, %zmm2
; AVX512VLBW-NEXT: vpsllvw %zmm2, %zmm0, %zmm0
; AVX512VLBW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VLVBMI2-LABEL: var_funnnel_v32i16:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpshldvw %zmm2, %zmm1, %zmm0
; AVX512VLVBMI2-NEXT: retq
%res = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %amt)
ret <32 x i16> %res
}
define <64 x i8> @var_funnnel_v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> %amt) nounwind {
; AVX512F-LABEL: var_funnnel_v64i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm3
; AVX512F-NEXT: vpsrlw $1, %ymm3, %ymm3
; AVX512F-NEXT: vmovdqa {{.*#+}} ymm4 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
; AVX512F-NEXT: vpand %ymm4, %ymm3, %ymm5
; AVX512F-NEXT: vpsrlw $4, %ymm5, %ymm3
; AVX512F-NEXT: vmovdqa {{.*#+}} ymm6 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512F-NEXT: vpand %ymm6, %ymm3, %ymm7
; AVX512F-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
; AVX512F-NEXT: vextracti64x4 $1, %zmm2, %ymm3
; AVX512F-NEXT: vmovdqa {{.*#+}} ymm8 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
; AVX512F-NEXT: vpxor %ymm3, %ymm8, %ymm9
; AVX512F-NEXT: vpsllw $5, %ymm9, %ymm9
; AVX512F-NEXT: vpblendvb %ymm9, %ymm7, %ymm5, %ymm5
; AVX512F-NEXT: vpsrlw $2, %ymm5, %ymm7
; AVX512F-NEXT: vmovdqa {{.*#+}} ymm10 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
; AVX512F-NEXT: vpand %ymm7, %ymm10, %ymm7
; AVX512F-NEXT: vpaddb %ymm9, %ymm9, %ymm9
; AVX512F-NEXT: vpblendvb %ymm9, %ymm7, %ymm5, %ymm5
; AVX512F-NEXT: vpsrlw $1, %ymm5, %ymm7
; AVX512F-NEXT: vpand %ymm4, %ymm7, %ymm7
; AVX512F-NEXT: vpaddb %ymm9, %ymm9, %ymm9
; AVX512F-NEXT: vpblendvb %ymm9, %ymm7, %ymm5, %ymm5
; AVX512F-NEXT: vpsrlw $1, %ymm1, %ymm1
; AVX512F-NEXT: vpand %ymm4, %ymm1, %ymm1
; AVX512F-NEXT: vpsrlw $4, %ymm1, %ymm7
; AVX512F-NEXT: vpand %ymm6, %ymm7, %ymm6
; AVX512F-NEXT: vpxor %ymm2, %ymm8, %ymm7
; AVX512F-NEXT: vpsllw $5, %ymm7, %ymm7
; AVX512F-NEXT: vpblendvb %ymm7, %ymm6, %ymm1, %ymm1
; AVX512F-NEXT: vpsrlw $2, %ymm1, %ymm6
; AVX512F-NEXT: vpand %ymm6, %ymm10, %ymm6
; AVX512F-NEXT: vpaddb %ymm7, %ymm7, %ymm7
; AVX512F-NEXT: vpblendvb %ymm7, %ymm6, %ymm1, %ymm1
; AVX512F-NEXT: vpsrlw $1, %ymm1, %ymm6
; AVX512F-NEXT: vpand %ymm4, %ymm6, %ymm4
; AVX512F-NEXT: vpaddb %ymm7, %ymm7, %ymm6
; AVX512F-NEXT: vpblendvb %ymm6, %ymm4, %ymm1, %ymm1
; AVX512F-NEXT: vinserti64x4 $1, %ymm5, %zmm1, %zmm1
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm4
; AVX512F-NEXT: vpsllw $4, %ymm4, %ymm5
; AVX512F-NEXT: vmovdqa {{.*#+}} ymm6 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
; AVX512F-NEXT: vpand %ymm6, %ymm5, %ymm5
; AVX512F-NEXT: vpsllw $5, %ymm3, %ymm3
; AVX512F-NEXT: vpblendvb %ymm3, %ymm5, %ymm4, %ymm4
; AVX512F-NEXT: vpsllw $2, %ymm4, %ymm5
; AVX512F-NEXT: vmovdqa {{.*#+}} ymm7 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
; AVX512F-NEXT: vpand %ymm7, %ymm5, %ymm5
; AVX512F-NEXT: vpaddb %ymm3, %ymm3, %ymm3
; AVX512F-NEXT: vpblendvb %ymm3, %ymm5, %ymm4, %ymm4
; AVX512F-NEXT: vpaddb %ymm4, %ymm4, %ymm5
; AVX512F-NEXT: vpaddb %ymm3, %ymm3, %ymm3
; AVX512F-NEXT: vpblendvb %ymm3, %ymm5, %ymm4, %ymm3
; AVX512F-NEXT: vpsllw $4, %ymm0, %ymm4
; AVX512F-NEXT: vpand %ymm6, %ymm4, %ymm4
; AVX512F-NEXT: vpsllw $5, %ymm2, %ymm2
; AVX512F-NEXT: vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
; AVX512F-NEXT: vpsllw $2, %ymm0, %ymm4
; AVX512F-NEXT: vpand %ymm7, %ymm4, %ymm4
; AVX512F-NEXT: vpaddb %ymm2, %ymm2, %ymm2
; AVX512F-NEXT: vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
; AVX512F-NEXT: vpaddb %ymm0, %ymm0, %ymm4
; AVX512F-NEXT: vpaddb %ymm2, %ymm2, %ymm2
; AVX512F-NEXT: vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
; AVX512F-NEXT: vinserti64x4 $1, %ymm3, %zmm0, %zmm0
; AVX512F-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: var_funnnel_v64i8:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vextracti64x4 $1, %zmm1, %ymm3
; AVX512VL-NEXT: vpsrlw $1, %ymm3, %ymm3
; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm4 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127]
; AVX512VL-NEXT: vpand %ymm4, %ymm3, %ymm5
; AVX512VL-NEXT: vpsrlw $4, %ymm5, %ymm3
; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm6 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512VL-NEXT: vpand %ymm6, %ymm3, %ymm7
; AVX512VL-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
; AVX512VL-NEXT: vextracti64x4 $1, %zmm2, %ymm3
; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm8 = [7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7]
; AVX512VL-NEXT: vpxor %ymm3, %ymm8, %ymm9
; AVX512VL-NEXT: vpsllw $5, %ymm9, %ymm9
; AVX512VL-NEXT: vpblendvb %ymm9, %ymm7, %ymm5, %ymm5
; AVX512VL-NEXT: vpsrlw $2, %ymm5, %ymm7
; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm10 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63]
; AVX512VL-NEXT: vpand %ymm7, %ymm10, %ymm7
; AVX512VL-NEXT: vpaddb %ymm9, %ymm9, %ymm9
; AVX512VL-NEXT: vpblendvb %ymm9, %ymm7, %ymm5, %ymm5
; AVX512VL-NEXT: vpsrlw $1, %ymm5, %ymm7
; AVX512VL-NEXT: vpand %ymm4, %ymm7, %ymm7
; AVX512VL-NEXT: vpaddb %ymm9, %ymm9, %ymm9
; AVX512VL-NEXT: vpblendvb %ymm9, %ymm7, %ymm5, %ymm5
; AVX512VL-NEXT: vpsrlw $1, %ymm1, %ymm1
; AVX512VL-NEXT: vpand %ymm4, %ymm1, %ymm1
; AVX512VL-NEXT: vpsrlw $4, %ymm1, %ymm7
; AVX512VL-NEXT: vpand %ymm6, %ymm7, %ymm6
; AVX512VL-NEXT: vpxor %ymm2, %ymm8, %ymm7
; AVX512VL-NEXT: vpsllw $5, %ymm7, %ymm7
; AVX512VL-NEXT: vpblendvb %ymm7, %ymm6, %ymm1, %ymm1
; AVX512VL-NEXT: vpsrlw $2, %ymm1, %ymm6
; AVX512VL-NEXT: vpand %ymm6, %ymm10, %ymm6
; AVX512VL-NEXT: vpaddb %ymm7, %ymm7, %ymm7
; AVX512VL-NEXT: vpblendvb %ymm7, %ymm6, %ymm1, %ymm1
; AVX512VL-NEXT: vpsrlw $1, %ymm1, %ymm6
; AVX512VL-NEXT: vpand %ymm4, %ymm6, %ymm4
; AVX512VL-NEXT: vpaddb %ymm7, %ymm7, %ymm6
; AVX512VL-NEXT: vpblendvb %ymm6, %ymm4, %ymm1, %ymm1
; AVX512VL-NEXT: vinserti64x4 $1, %ymm5, %zmm1, %zmm1
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm4
; AVX512VL-NEXT: vpsllw $4, %ymm4, %ymm5
; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm6 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240]
; AVX512VL-NEXT: vpand %ymm6, %ymm5, %ymm5
; AVX512VL-NEXT: vpsllw $5, %ymm3, %ymm3
; AVX512VL-NEXT: vpblendvb %ymm3, %ymm5, %ymm4, %ymm4
; AVX512VL-NEXT: vpsllw $2, %ymm4, %ymm5
; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm7 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252]
; AVX512VL-NEXT: vpand %ymm7, %ymm5, %ymm5
; AVX512VL-NEXT: vpaddb %ymm3, %ymm3, %ymm3
; AVX512VL-NEXT: vpblendvb %ymm3, %ymm5, %ymm4, %ymm4
; AVX512VL-NEXT: vpaddb %ymm4, %ymm4, %ymm5
; AVX512VL-NEXT: vpaddb %ymm3, %ymm3, %ymm3
; AVX512VL-NEXT: vpblendvb %ymm3, %ymm5, %ymm4, %ymm3
; AVX512VL-NEXT: vpsllw $4, %ymm0, %ymm4
; AVX512VL-NEXT: vpand %ymm6, %ymm4, %ymm4
; AVX512VL-NEXT: vpsllw $5, %ymm2, %ymm2
; AVX512VL-NEXT: vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
; AVX512VL-NEXT: vpsllw $2, %ymm0, %ymm4
; AVX512VL-NEXT: vpand %ymm7, %ymm4, %ymm4
; AVX512VL-NEXT: vpaddb %ymm2, %ymm2, %ymm2
; AVX512VL-NEXT: vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
; AVX512VL-NEXT: vpaddb %ymm0, %ymm0, %ymm4
; AVX512VL-NEXT: vpaddb %ymm2, %ymm2, %ymm2
; AVX512VL-NEXT: vpblendvb %ymm2, %ymm4, %ymm0, %ymm0
; AVX512VL-NEXT: vinserti64x4 $1, %ymm3, %zmm0, %zmm0
; AVX512VL-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: var_funnnel_v64i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm0[8],zmm1[9],zmm0[9],zmm1[10],zmm0[10],zmm1[11],zmm0[11],zmm1[12],zmm0[12],zmm1[13],zmm0[13],zmm1[14],zmm0[14],zmm1[15],zmm0[15],zmm1[24],zmm0[24],zmm1[25],zmm0[25],zmm1[26],zmm0[26],zmm1[27],zmm0[27],zmm1[28],zmm0[28],zmm1[29],zmm0[29],zmm1[30],zmm0[30],zmm1[31],zmm0[31],zmm1[40],zmm0[40],zmm1[41],zmm0[41],zmm1[42],zmm0[42],zmm1[43],zmm0[43],zmm1[44],zmm0[44],zmm1[45],zmm0[45],zmm1[46],zmm0[46],zmm1[47],zmm0[47],zmm1[56],zmm0[56],zmm1[57],zmm0[57],zmm1[58],zmm0[58],zmm1[59],zmm0[59],zmm1[60],zmm0[60],zmm1[61],zmm0[61],zmm1[62],zmm0[62],zmm1[63],zmm0[63]
; AVX512BW-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
; AVX512BW-NEXT: vpxor %xmm4, %xmm4, %xmm4
; AVX512BW-NEXT: vpunpckhbw {{.*#+}} zmm5 = zmm2[8],zmm4[8],zmm2[9],zmm4[9],zmm2[10],zmm4[10],zmm2[11],zmm4[11],zmm2[12],zmm4[12],zmm2[13],zmm4[13],zmm2[14],zmm4[14],zmm2[15],zmm4[15],zmm2[24],zmm4[24],zmm2[25],zmm4[25],zmm2[26],zmm4[26],zmm2[27],zmm4[27],zmm2[28],zmm4[28],zmm2[29],zmm4[29],zmm2[30],zmm4[30],zmm2[31],zmm4[31],zmm2[40],zmm4[40],zmm2[41],zmm4[41],zmm2[42],zmm4[42],zmm2[43],zmm4[43],zmm2[44],zmm4[44],zmm2[45],zmm4[45],zmm2[46],zmm4[46],zmm2[47],zmm4[47],zmm2[56],zmm4[56],zmm2[57],zmm4[57],zmm2[58],zmm4[58],zmm2[59],zmm4[59],zmm2[60],zmm4[60],zmm2[61],zmm4[61],zmm2[62],zmm4[62],zmm2[63],zmm4[63]
; AVX512BW-NEXT: vpsllvw %zmm5, %zmm3, %zmm3
; AVX512BW-NEXT: vpsrlw $8, %zmm3, %zmm3
; AVX512BW-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm1[0],zmm0[0],zmm1[1],zmm0[1],zmm1[2],zmm0[2],zmm1[3],zmm0[3],zmm1[4],zmm0[4],zmm1[5],zmm0[5],zmm1[6],zmm0[6],zmm1[7],zmm0[7],zmm1[16],zmm0[16],zmm1[17],zmm0[17],zmm1[18],zmm0[18],zmm1[19],zmm0[19],zmm1[20],zmm0[20],zmm1[21],zmm0[21],zmm1[22],zmm0[22],zmm1[23],zmm0[23],zmm1[32],zmm0[32],zmm1[33],zmm0[33],zmm1[34],zmm0[34],zmm1[35],zmm0[35],zmm1[36],zmm0[36],zmm1[37],zmm0[37],zmm1[38],zmm0[38],zmm1[39],zmm0[39],zmm1[48],zmm0[48],zmm1[49],zmm0[49],zmm1[50],zmm0[50],zmm1[51],zmm0[51],zmm1[52],zmm0[52],zmm1[53],zmm0[53],zmm1[54],zmm0[54],zmm1[55],zmm0[55]
; AVX512BW-NEXT: vpunpcklbw {{.*#+}} zmm1 = zmm2[0],zmm4[0],zmm2[1],zmm4[1],zmm2[2],zmm4[2],zmm2[3],zmm4[3],zmm2[4],zmm4[4],zmm2[5],zmm4[5],zmm2[6],zmm4[6],zmm2[7],zmm4[7],zmm2[16],zmm4[16],zmm2[17],zmm4[17],zmm2[18],zmm4[18],zmm2[19],zmm4[19],zmm2[20],zmm4[20],zmm2[21],zmm4[21],zmm2[22],zmm4[22],zmm2[23],zmm4[23],zmm2[32],zmm4[32],zmm2[33],zmm4[33],zmm2[34],zmm4[34],zmm2[35],zmm4[35],zmm2[36],zmm4[36],zmm2[37],zmm4[37],zmm2[38],zmm4[38],zmm2[39],zmm4[39],zmm2[48],zmm4[48],zmm2[49],zmm4[49],zmm2[50],zmm4[50],zmm2[51],zmm4[51],zmm2[52],zmm4[52],zmm2[53],zmm4[53],zmm2[54],zmm4[54],zmm2[55],zmm4[55]
; AVX512BW-NEXT: vpsllvw %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpsrlw $8, %zmm0, %zmm0
; AVX512BW-NEXT: vpackuswb %zmm3, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VBMI2-LABEL: var_funnnel_v64i8:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm0[8],zmm1[9],zmm0[9],zmm1[10],zmm0[10],zmm1[11],zmm0[11],zmm1[12],zmm0[12],zmm1[13],zmm0[13],zmm1[14],zmm0[14],zmm1[15],zmm0[15],zmm1[24],zmm0[24],zmm1[25],zmm0[25],zmm1[26],zmm0[26],zmm1[27],zmm0[27],zmm1[28],zmm0[28],zmm1[29],zmm0[29],zmm1[30],zmm0[30],zmm1[31],zmm0[31],zmm1[40],zmm0[40],zmm1[41],zmm0[41],zmm1[42],zmm0[42],zmm1[43],zmm0[43],zmm1[44],zmm0[44],zmm1[45],zmm0[45],zmm1[46],zmm0[46],zmm1[47],zmm0[47],zmm1[56],zmm0[56],zmm1[57],zmm0[57],zmm1[58],zmm0[58],zmm1[59],zmm0[59],zmm1[60],zmm0[60],zmm1[61],zmm0[61],zmm1[62],zmm0[62],zmm1[63],zmm0[63]
; AVX512VBMI2-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
; AVX512VBMI2-NEXT: vpxor %xmm4, %xmm4, %xmm4
; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} zmm5 = zmm2[8],zmm4[8],zmm2[9],zmm4[9],zmm2[10],zmm4[10],zmm2[11],zmm4[11],zmm2[12],zmm4[12],zmm2[13],zmm4[13],zmm2[14],zmm4[14],zmm2[15],zmm4[15],zmm2[24],zmm4[24],zmm2[25],zmm4[25],zmm2[26],zmm4[26],zmm2[27],zmm4[27],zmm2[28],zmm4[28],zmm2[29],zmm4[29],zmm2[30],zmm4[30],zmm2[31],zmm4[31],zmm2[40],zmm4[40],zmm2[41],zmm4[41],zmm2[42],zmm4[42],zmm2[43],zmm4[43],zmm2[44],zmm4[44],zmm2[45],zmm4[45],zmm2[46],zmm4[46],zmm2[47],zmm4[47],zmm2[56],zmm4[56],zmm2[57],zmm4[57],zmm2[58],zmm4[58],zmm2[59],zmm4[59],zmm2[60],zmm4[60],zmm2[61],zmm4[61],zmm2[62],zmm4[62],zmm2[63],zmm4[63]
; AVX512VBMI2-NEXT: vpsllvw %zmm5, %zmm3, %zmm3
; AVX512VBMI2-NEXT: vpsrlw $8, %zmm3, %zmm3
; AVX512VBMI2-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm1[0],zmm0[0],zmm1[1],zmm0[1],zmm1[2],zmm0[2],zmm1[3],zmm0[3],zmm1[4],zmm0[4],zmm1[5],zmm0[5],zmm1[6],zmm0[6],zmm1[7],zmm0[7],zmm1[16],zmm0[16],zmm1[17],zmm0[17],zmm1[18],zmm0[18],zmm1[19],zmm0[19],zmm1[20],zmm0[20],zmm1[21],zmm0[21],zmm1[22],zmm0[22],zmm1[23],zmm0[23],zmm1[32],zmm0[32],zmm1[33],zmm0[33],zmm1[34],zmm0[34],zmm1[35],zmm0[35],zmm1[36],zmm0[36],zmm1[37],zmm0[37],zmm1[38],zmm0[38],zmm1[39],zmm0[39],zmm1[48],zmm0[48],zmm1[49],zmm0[49],zmm1[50],zmm0[50],zmm1[51],zmm0[51],zmm1[52],zmm0[52],zmm1[53],zmm0[53],zmm1[54],zmm0[54],zmm1[55],zmm0[55]
; AVX512VBMI2-NEXT: vpunpcklbw {{.*#+}} zmm1 = zmm2[0],zmm4[0],zmm2[1],zmm4[1],zmm2[2],zmm4[2],zmm2[3],zmm4[3],zmm2[4],zmm4[4],zmm2[5],zmm4[5],zmm2[6],zmm4[6],zmm2[7],zmm4[7],zmm2[16],zmm4[16],zmm2[17],zmm4[17],zmm2[18],zmm4[18],zmm2[19],zmm4[19],zmm2[20],zmm4[20],zmm2[21],zmm4[21],zmm2[22],zmm4[22],zmm2[23],zmm4[23],zmm2[32],zmm4[32],zmm2[33],zmm4[33],zmm2[34],zmm4[34],zmm2[35],zmm4[35],zmm2[36],zmm4[36],zmm2[37],zmm4[37],zmm2[38],zmm4[38],zmm2[39],zmm4[39],zmm2[48],zmm4[48],zmm2[49],zmm4[49],zmm2[50],zmm4[50],zmm2[51],zmm4[51],zmm2[52],zmm4[52],zmm2[53],zmm4[53],zmm2[54],zmm4[54],zmm2[55],zmm4[55]
; AVX512VBMI2-NEXT: vpsllvw %zmm1, %zmm0, %zmm0
; AVX512VBMI2-NEXT: vpsrlw $8, %zmm0, %zmm0
; AVX512VBMI2-NEXT: vpackuswb %zmm3, %zmm0, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLBW-LABEL: var_funnnel_v64i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm0[8],zmm1[9],zmm0[9],zmm1[10],zmm0[10],zmm1[11],zmm0[11],zmm1[12],zmm0[12],zmm1[13],zmm0[13],zmm1[14],zmm0[14],zmm1[15],zmm0[15],zmm1[24],zmm0[24],zmm1[25],zmm0[25],zmm1[26],zmm0[26],zmm1[27],zmm0[27],zmm1[28],zmm0[28],zmm1[29],zmm0[29],zmm1[30],zmm0[30],zmm1[31],zmm0[31],zmm1[40],zmm0[40],zmm1[41],zmm0[41],zmm1[42],zmm0[42],zmm1[43],zmm0[43],zmm1[44],zmm0[44],zmm1[45],zmm0[45],zmm1[46],zmm0[46],zmm1[47],zmm0[47],zmm1[56],zmm0[56],zmm1[57],zmm0[57],zmm1[58],zmm0[58],zmm1[59],zmm0[59],zmm1[60],zmm0[60],zmm1[61],zmm0[61],zmm1[62],zmm0[62],zmm1[63],zmm0[63]
; AVX512VLBW-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
; AVX512VLBW-NEXT: vpxor %xmm4, %xmm4, %xmm4
; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} zmm5 = zmm2[8],zmm4[8],zmm2[9],zmm4[9],zmm2[10],zmm4[10],zmm2[11],zmm4[11],zmm2[12],zmm4[12],zmm2[13],zmm4[13],zmm2[14],zmm4[14],zmm2[15],zmm4[15],zmm2[24],zmm4[24],zmm2[25],zmm4[25],zmm2[26],zmm4[26],zmm2[27],zmm4[27],zmm2[28],zmm4[28],zmm2[29],zmm4[29],zmm2[30],zmm4[30],zmm2[31],zmm4[31],zmm2[40],zmm4[40],zmm2[41],zmm4[41],zmm2[42],zmm4[42],zmm2[43],zmm4[43],zmm2[44],zmm4[44],zmm2[45],zmm4[45],zmm2[46],zmm4[46],zmm2[47],zmm4[47],zmm2[56],zmm4[56],zmm2[57],zmm4[57],zmm2[58],zmm4[58],zmm2[59],zmm4[59],zmm2[60],zmm4[60],zmm2[61],zmm4[61],zmm2[62],zmm4[62],zmm2[63],zmm4[63]
; AVX512VLBW-NEXT: vpsllvw %zmm5, %zmm3, %zmm3
; AVX512VLBW-NEXT: vpsrlw $8, %zmm3, %zmm3
; AVX512VLBW-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm1[0],zmm0[0],zmm1[1],zmm0[1],zmm1[2],zmm0[2],zmm1[3],zmm0[3],zmm1[4],zmm0[4],zmm1[5],zmm0[5],zmm1[6],zmm0[6],zmm1[7],zmm0[7],zmm1[16],zmm0[16],zmm1[17],zmm0[17],zmm1[18],zmm0[18],zmm1[19],zmm0[19],zmm1[20],zmm0[20],zmm1[21],zmm0[21],zmm1[22],zmm0[22],zmm1[23],zmm0[23],zmm1[32],zmm0[32],zmm1[33],zmm0[33],zmm1[34],zmm0[34],zmm1[35],zmm0[35],zmm1[36],zmm0[36],zmm1[37],zmm0[37],zmm1[38],zmm0[38],zmm1[39],zmm0[39],zmm1[48],zmm0[48],zmm1[49],zmm0[49],zmm1[50],zmm0[50],zmm1[51],zmm0[51],zmm1[52],zmm0[52],zmm1[53],zmm0[53],zmm1[54],zmm0[54],zmm1[55],zmm0[55]
; AVX512VLBW-NEXT: vpunpcklbw {{.*#+}} zmm1 = zmm2[0],zmm4[0],zmm2[1],zmm4[1],zmm2[2],zmm4[2],zmm2[3],zmm4[3],zmm2[4],zmm4[4],zmm2[5],zmm4[5],zmm2[6],zmm4[6],zmm2[7],zmm4[7],zmm2[16],zmm4[16],zmm2[17],zmm4[17],zmm2[18],zmm4[18],zmm2[19],zmm4[19],zmm2[20],zmm4[20],zmm2[21],zmm4[21],zmm2[22],zmm4[22],zmm2[23],zmm4[23],zmm2[32],zmm4[32],zmm2[33],zmm4[33],zmm2[34],zmm4[34],zmm2[35],zmm4[35],zmm2[36],zmm4[36],zmm2[37],zmm4[37],zmm2[38],zmm4[38],zmm2[39],zmm4[39],zmm2[48],zmm4[48],zmm2[49],zmm4[49],zmm2[50],zmm4[50],zmm2[51],zmm4[51],zmm2[52],zmm4[52],zmm2[53],zmm4[53],zmm2[54],zmm4[54],zmm2[55],zmm4[55]
; AVX512VLBW-NEXT: vpsllvw %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: vpsrlw $8, %zmm0, %zmm0
; AVX512VLBW-NEXT: vpackuswb %zmm3, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VLVBMI2-LABEL: var_funnnel_v64i8:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm0[8],zmm1[9],zmm0[9],zmm1[10],zmm0[10],zmm1[11],zmm0[11],zmm1[12],zmm0[12],zmm1[13],zmm0[13],zmm1[14],zmm0[14],zmm1[15],zmm0[15],zmm1[24],zmm0[24],zmm1[25],zmm0[25],zmm1[26],zmm0[26],zmm1[27],zmm0[27],zmm1[28],zmm0[28],zmm1[29],zmm0[29],zmm1[30],zmm0[30],zmm1[31],zmm0[31],zmm1[40],zmm0[40],zmm1[41],zmm0[41],zmm1[42],zmm0[42],zmm1[43],zmm0[43],zmm1[44],zmm0[44],zmm1[45],zmm0[45],zmm1[46],zmm0[46],zmm1[47],zmm0[47],zmm1[56],zmm0[56],zmm1[57],zmm0[57],zmm1[58],zmm0[58],zmm1[59],zmm0[59],zmm1[60],zmm0[60],zmm1[61],zmm0[61],zmm1[62],zmm0[62],zmm1[63],zmm0[63]
; AVX512VLVBMI2-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
; AVX512VLVBMI2-NEXT: vpxor %xmm4, %xmm4, %xmm4
; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} zmm5 = zmm2[8],zmm4[8],zmm2[9],zmm4[9],zmm2[10],zmm4[10],zmm2[11],zmm4[11],zmm2[12],zmm4[12],zmm2[13],zmm4[13],zmm2[14],zmm4[14],zmm2[15],zmm4[15],zmm2[24],zmm4[24],zmm2[25],zmm4[25],zmm2[26],zmm4[26],zmm2[27],zmm4[27],zmm2[28],zmm4[28],zmm2[29],zmm4[29],zmm2[30],zmm4[30],zmm2[31],zmm4[31],zmm2[40],zmm4[40],zmm2[41],zmm4[41],zmm2[42],zmm4[42],zmm2[43],zmm4[43],zmm2[44],zmm4[44],zmm2[45],zmm4[45],zmm2[46],zmm4[46],zmm2[47],zmm4[47],zmm2[56],zmm4[56],zmm2[57],zmm4[57],zmm2[58],zmm4[58],zmm2[59],zmm4[59],zmm2[60],zmm4[60],zmm2[61],zmm4[61],zmm2[62],zmm4[62],zmm2[63],zmm4[63]
; AVX512VLVBMI2-NEXT: vpsllvw %zmm5, %zmm3, %zmm3
; AVX512VLVBMI2-NEXT: vpsrlw $8, %zmm3, %zmm3
; AVX512VLVBMI2-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm1[0],zmm0[0],zmm1[1],zmm0[1],zmm1[2],zmm0[2],zmm1[3],zmm0[3],zmm1[4],zmm0[4],zmm1[5],zmm0[5],zmm1[6],zmm0[6],zmm1[7],zmm0[7],zmm1[16],zmm0[16],zmm1[17],zmm0[17],zmm1[18],zmm0[18],zmm1[19],zmm0[19],zmm1[20],zmm0[20],zmm1[21],zmm0[21],zmm1[22],zmm0[22],zmm1[23],zmm0[23],zmm1[32],zmm0[32],zmm1[33],zmm0[33],zmm1[34],zmm0[34],zmm1[35],zmm0[35],zmm1[36],zmm0[36],zmm1[37],zmm0[37],zmm1[38],zmm0[38],zmm1[39],zmm0[39],zmm1[48],zmm0[48],zmm1[49],zmm0[49],zmm1[50],zmm0[50],zmm1[51],zmm0[51],zmm1[52],zmm0[52],zmm1[53],zmm0[53],zmm1[54],zmm0[54],zmm1[55],zmm0[55]
; AVX512VLVBMI2-NEXT: vpunpcklbw {{.*#+}} zmm1 = zmm2[0],zmm4[0],zmm2[1],zmm4[1],zmm2[2],zmm4[2],zmm2[3],zmm4[3],zmm2[4],zmm4[4],zmm2[5],zmm4[5],zmm2[6],zmm4[6],zmm2[7],zmm4[7],zmm2[16],zmm4[16],zmm2[17],zmm4[17],zmm2[18],zmm4[18],zmm2[19],zmm4[19],zmm2[20],zmm4[20],zmm2[21],zmm4[21],zmm2[22],zmm4[22],zmm2[23],zmm4[23],zmm2[32],zmm4[32],zmm2[33],zmm4[33],zmm2[34],zmm4[34],zmm2[35],zmm4[35],zmm2[36],zmm4[36],zmm2[37],zmm4[37],zmm2[38],zmm4[38],zmm2[39],zmm4[39],zmm2[48],zmm4[48],zmm2[49],zmm4[49],zmm2[50],zmm4[50],zmm2[51],zmm4[51],zmm2[52],zmm4[52],zmm2[53],zmm4[53],zmm2[54],zmm4[54],zmm2[55],zmm4[55]
; AVX512VLVBMI2-NEXT: vpsllvw %zmm1, %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: vpsrlw $8, %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: vpackuswb %zmm3, %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: retq
%res = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> %amt)
ret <64 x i8> %res
}
;
; Uniform Variable Shifts
;
define <8 x i64> @splatvar_funnnel_v8i64(<8 x i64> %x, <8 x i64> %y, <8 x i64> %amt) nounwind {
; AVX512F-LABEL: splatvar_funnnel_v8i64:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vmovdqa {{.*#+}} xmm3 = [63,63]
; AVX512F-NEXT: vpandn %xmm3, %xmm2, %xmm4
; AVX512F-NEXT: vpsrlq $1, %zmm1, %zmm1
; AVX512F-NEXT: vpsrlq %xmm4, %zmm1, %zmm1
; AVX512F-NEXT: vpand %xmm3, %xmm2, %xmm2
; AVX512F-NEXT: vpsllq %xmm2, %zmm0, %zmm0
; AVX512F-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: splatvar_funnnel_v8i64:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpbroadcastq {{.*#+}} xmm3 = [63,63]
; AVX512VL-NEXT: vpandn %xmm3, %xmm2, %xmm4
; AVX512VL-NEXT: vpsrlq $1, %zmm1, %zmm1
; AVX512VL-NEXT: vpsrlq %xmm4, %zmm1, %zmm1
; AVX512VL-NEXT: vpand %xmm3, %xmm2, %xmm2
; AVX512VL-NEXT: vpsllq %xmm2, %zmm0, %zmm0
; AVX512VL-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: splatvar_funnnel_v8i64:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm3 = [63,63]
; AVX512BW-NEXT: vpandn %xmm3, %xmm2, %xmm4
; AVX512BW-NEXT: vpsrlq $1, %zmm1, %zmm1
; AVX512BW-NEXT: vpsrlq %xmm4, %zmm1, %zmm1
; AVX512BW-NEXT: vpand %xmm3, %xmm2, %xmm2
; AVX512BW-NEXT: vpsllq %xmm2, %zmm0, %zmm0
; AVX512BW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VBMI2-LABEL: splatvar_funnnel_v8i64:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpbroadcastq %xmm2, %zmm2
; AVX512VBMI2-NEXT: vpshldvq %zmm2, %zmm1, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLBW-LABEL: splatvar_funnnel_v8i64:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpbroadcastq {{.*#+}} xmm3 = [63,63]
; AVX512VLBW-NEXT: vpandn %xmm3, %xmm2, %xmm4
; AVX512VLBW-NEXT: vpsrlq $1, %zmm1, %zmm1
; AVX512VLBW-NEXT: vpsrlq %xmm4, %zmm1, %zmm1
; AVX512VLBW-NEXT: vpand %xmm3, %xmm2, %xmm2
; AVX512VLBW-NEXT: vpsllq %xmm2, %zmm0, %zmm0
; AVX512VLBW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VLVBMI2-LABEL: splatvar_funnnel_v8i64:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpbroadcastq %xmm2, %zmm2
; AVX512VLVBMI2-NEXT: vpshldvq %zmm2, %zmm1, %zmm0
; AVX512VLVBMI2-NEXT: retq
%splat = shufflevector <8 x i64> %amt, <8 x i64> undef, <8 x i32> zeroinitializer
%res = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %x, <8 x i64> %y, <8 x i64> %splat)
ret <8 x i64> %res
}
define <16 x i32> @splatvar_funnnel_v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> %amt) nounwind {
; AVX512F-LABEL: splatvar_funnnel_v16i32:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpunpckhdq {{.*#+}} zmm3 = zmm1[2],zmm0[2],zmm1[3],zmm0[3],zmm1[6],zmm0[6],zmm1[7],zmm0[7],zmm1[10],zmm0[10],zmm1[11],zmm0[11],zmm1[14],zmm0[14],zmm1[15],zmm0[15]
; AVX512F-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512F-NEXT: vpsllq %xmm2, %zmm3, %zmm3
; AVX512F-NEXT: vpunpckldq {{.*#+}} zmm0 = zmm1[0],zmm0[0],zmm1[1],zmm0[1],zmm1[4],zmm0[4],zmm1[5],zmm0[5],zmm1[8],zmm0[8],zmm1[9],zmm0[9],zmm1[12],zmm0[12],zmm1[13],zmm0[13]
; AVX512F-NEXT: vpsllq %xmm2, %zmm0, %zmm0
; AVX512F-NEXT: vshufps {{.*#+}} zmm0 = zmm0[1,3],zmm3[1,3],zmm0[5,7],zmm3[5,7],zmm0[9,11],zmm3[9,11],zmm0[13,15],zmm3[13,15]
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: splatvar_funnnel_v16i32:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpunpckhdq {{.*#+}} zmm3 = zmm1[2],zmm0[2],zmm1[3],zmm0[3],zmm1[6],zmm0[6],zmm1[7],zmm0[7],zmm1[10],zmm0[10],zmm1[11],zmm0[11],zmm1[14],zmm0[14],zmm1[15],zmm0[15]
; AVX512VL-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512VL-NEXT: vpsllq %xmm2, %zmm3, %zmm3
; AVX512VL-NEXT: vpunpckldq {{.*#+}} zmm0 = zmm1[0],zmm0[0],zmm1[1],zmm0[1],zmm1[4],zmm0[4],zmm1[5],zmm0[5],zmm1[8],zmm0[8],zmm1[9],zmm0[9],zmm1[12],zmm0[12],zmm1[13],zmm0[13]
; AVX512VL-NEXT: vpsllq %xmm2, %zmm0, %zmm0
; AVX512VL-NEXT: vshufps {{.*#+}} zmm0 = zmm0[1,3],zmm3[1,3],zmm0[5,7],zmm3[5,7],zmm0[9,11],zmm3[9,11],zmm0[13,15],zmm3[13,15]
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: splatvar_funnnel_v16i32:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpunpckhdq {{.*#+}} zmm3 = zmm1[2],zmm0[2],zmm1[3],zmm0[3],zmm1[6],zmm0[6],zmm1[7],zmm0[7],zmm1[10],zmm0[10],zmm1[11],zmm0[11],zmm1[14],zmm0[14],zmm1[15],zmm0[15]
; AVX512BW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512BW-NEXT: vpsllq %xmm2, %zmm3, %zmm3
; AVX512BW-NEXT: vpunpckldq {{.*#+}} zmm0 = zmm1[0],zmm0[0],zmm1[1],zmm0[1],zmm1[4],zmm0[4],zmm1[5],zmm0[5],zmm1[8],zmm0[8],zmm1[9],zmm0[9],zmm1[12],zmm0[12],zmm1[13],zmm0[13]
; AVX512BW-NEXT: vpsllq %xmm2, %zmm0, %zmm0
; AVX512BW-NEXT: vshufps {{.*#+}} zmm0 = zmm0[1,3],zmm3[1,3],zmm0[5,7],zmm3[5,7],zmm0[9,11],zmm3[9,11],zmm0[13,15],zmm3[13,15]
; AVX512BW-NEXT: retq
;
; AVX512VBMI2-LABEL: splatvar_funnnel_v16i32:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpbroadcastd %xmm2, %zmm2
; AVX512VBMI2-NEXT: vpshldvd %zmm2, %zmm1, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLBW-LABEL: splatvar_funnnel_v16i32:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpunpckhdq {{.*#+}} zmm3 = zmm1[2],zmm0[2],zmm1[3],zmm0[3],zmm1[6],zmm0[6],zmm1[7],zmm0[7],zmm1[10],zmm0[10],zmm1[11],zmm0[11],zmm1[14],zmm0[14],zmm1[15],zmm0[15]
; AVX512VLBW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512VLBW-NEXT: vpsllq %xmm2, %zmm3, %zmm3
; AVX512VLBW-NEXT: vpunpckldq {{.*#+}} zmm0 = zmm1[0],zmm0[0],zmm1[1],zmm0[1],zmm1[4],zmm0[4],zmm1[5],zmm0[5],zmm1[8],zmm0[8],zmm1[9],zmm0[9],zmm1[12],zmm0[12],zmm1[13],zmm0[13]
; AVX512VLBW-NEXT: vpsllq %xmm2, %zmm0, %zmm0
; AVX512VLBW-NEXT: vshufps {{.*#+}} zmm0 = zmm0[1,3],zmm3[1,3],zmm0[5,7],zmm3[5,7],zmm0[9,11],zmm3[9,11],zmm0[13,15],zmm3[13,15]
; AVX512VLBW-NEXT: retq
;
; AVX512VLVBMI2-LABEL: splatvar_funnnel_v16i32:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpbroadcastd %xmm2, %zmm2
; AVX512VLVBMI2-NEXT: vpshldvd %zmm2, %zmm1, %zmm0
; AVX512VLVBMI2-NEXT: retq
%splat = shufflevector <16 x i32> %amt, <16 x i32> undef, <16 x i32> zeroinitializer
%res = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> %splat)
ret <16 x i32> %res
}
define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %amt) nounwind {
; AVX512F-LABEL: splatvar_funnnel_v32i16:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpbroadcastq {{.*#+}} xmm3 = [15,15]
; AVX512F-NEXT: vpandn %xmm3, %xmm2, %xmm4
; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm5
; AVX512F-NEXT: vpsrlw $1, %ymm5, %ymm5
; AVX512F-NEXT: vpsrlw %xmm4, %ymm5, %ymm5
; AVX512F-NEXT: vpsrlw $1, %ymm1, %ymm1
; AVX512F-NEXT: vpsrlw %xmm4, %ymm1, %ymm1
; AVX512F-NEXT: vinserti64x4 $1, %ymm5, %zmm1, %zmm1
; AVX512F-NEXT: vpbroadcastw %xmm2, %xmm2
; AVX512F-NEXT: vpand %xmm3, %xmm2, %xmm2
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm3
; AVX512F-NEXT: vpsllw %xmm2, %ymm3, %ymm3
; AVX512F-NEXT: vpsllw %xmm2, %ymm0, %ymm0
; AVX512F-NEXT: vinserti64x4 $1, %ymm3, %zmm0, %zmm0
; AVX512F-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: splatvar_funnnel_v32i16:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpbroadcastq {{.*#+}} xmm3 = [15,15]
; AVX512VL-NEXT: vpandn %xmm3, %xmm2, %xmm4
; AVX512VL-NEXT: vextracti64x4 $1, %zmm1, %ymm5
; AVX512VL-NEXT: vpsrlw $1, %ymm5, %ymm5
; AVX512VL-NEXT: vpsrlw %xmm4, %ymm5, %ymm5
; AVX512VL-NEXT: vpsrlw $1, %ymm1, %ymm1
; AVX512VL-NEXT: vpsrlw %xmm4, %ymm1, %ymm1
; AVX512VL-NEXT: vinserti64x4 $1, %ymm5, %zmm1, %zmm1
; AVX512VL-NEXT: vpbroadcastw %xmm2, %xmm2
; AVX512VL-NEXT: vpand %xmm3, %xmm2, %xmm2
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm3
; AVX512VL-NEXT: vpsllw %xmm2, %ymm3, %ymm3
; AVX512VL-NEXT: vpsllw %xmm2, %ymm0, %ymm0
; AVX512VL-NEXT: vinserti64x4 $1, %ymm3, %zmm0, %zmm0
; AVX512VL-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: splatvar_funnnel_v32i16:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm3 = [15,0,0,0]
; AVX512BW-NEXT: vpandn %xmm3, %xmm2, %xmm4
; AVX512BW-NEXT: vpsrlw $1, %zmm1, %zmm1
; AVX512BW-NEXT: vpsrlw %xmm4, %zmm1, %zmm1
; AVX512BW-NEXT: vpand %xmm3, %xmm2, %xmm2
; AVX512BW-NEXT: vpsllw %xmm2, %zmm0, %zmm0
; AVX512BW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VBMI2-LABEL: splatvar_funnnel_v32i16:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpbroadcastw %xmm2, %zmm2
; AVX512VBMI2-NEXT: vpshldvw %zmm2, %zmm1, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLBW-LABEL: splatvar_funnnel_v32i16:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vmovdqa {{.*#+}} xmm3 = [15,0,0,0]
; AVX512VLBW-NEXT: vpandn %xmm3, %xmm2, %xmm4
; AVX512VLBW-NEXT: vpsrlw $1, %zmm1, %zmm1
; AVX512VLBW-NEXT: vpsrlw %xmm4, %zmm1, %zmm1
; AVX512VLBW-NEXT: vpand %xmm3, %xmm2, %xmm2
; AVX512VLBW-NEXT: vpsllw %xmm2, %zmm0, %zmm0
; AVX512VLBW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VLVBMI2-LABEL: splatvar_funnnel_v32i16:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpbroadcastw %xmm2, %zmm2
; AVX512VLVBMI2-NEXT: vpshldvw %zmm2, %zmm1, %zmm0
; AVX512VLVBMI2-NEXT: retq
%splat = shufflevector <32 x i16> %amt, <32 x i16> undef, <32 x i32> zeroinitializer
%res = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> %splat)
ret <32 x i16> %res
}
define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> %amt) nounwind {
; AVX512F-LABEL: splatvar_funnnel_v64i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm3
; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm4
; AVX512F-NEXT: vpunpckhbw {{.*#+}} ymm5 = ymm4[8],ymm3[8],ymm4[9],ymm3[9],ymm4[10],ymm3[10],ymm4[11],ymm3[11],ymm4[12],ymm3[12],ymm4[13],ymm3[13],ymm4[14],ymm3[14],ymm4[15],ymm3[15],ymm4[24],ymm3[24],ymm4[25],ymm3[25],ymm4[26],ymm3[26],ymm4[27],ymm3[27],ymm4[28],ymm3[28],ymm4[29],ymm3[29],ymm4[30],ymm3[30],ymm4[31],ymm3[31]
; AVX512F-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512F-NEXT: vpsllw %xmm2, %ymm5, %ymm5
; AVX512F-NEXT: vpsrlw $8, %ymm5, %ymm5
; AVX512F-NEXT: vpunpcklbw {{.*#+}} ymm3 = ymm4[0],ymm3[0],ymm4[1],ymm3[1],ymm4[2],ymm3[2],ymm4[3],ymm3[3],ymm4[4],ymm3[4],ymm4[5],ymm3[5],ymm4[6],ymm3[6],ymm4[7],ymm3[7],ymm4[16],ymm3[16],ymm4[17],ymm3[17],ymm4[18],ymm3[18],ymm4[19],ymm3[19],ymm4[20],ymm3[20],ymm4[21],ymm3[21],ymm4[22],ymm3[22],ymm4[23],ymm3[23]
; AVX512F-NEXT: vpsllw %xmm2, %ymm3, %ymm3
; AVX512F-NEXT: vpsrlw $8, %ymm3, %ymm3
; AVX512F-NEXT: vpackuswb %ymm5, %ymm3, %ymm3
; AVX512F-NEXT: vpunpckhbw {{.*#+}} ymm4 = 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]
; AVX512F-NEXT: vpsllw %xmm2, %ymm4, %ymm4
; AVX512F-NEXT: vpsrlw $8, %ymm4, %ymm4
; AVX512F-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]
; AVX512F-NEXT: vpsllw %xmm2, %ymm0, %ymm0
; AVX512F-NEXT: vpsrlw $8, %ymm0, %ymm0
; AVX512F-NEXT: vpackuswb %ymm4, %ymm0, %ymm0
; AVX512F-NEXT: vinserti64x4 $1, %ymm3, %zmm0, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: splatvar_funnnel_v64i8:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm3
; AVX512VL-NEXT: vextracti64x4 $1, %zmm1, %ymm4
; AVX512VL-NEXT: vpunpckhbw {{.*#+}} ymm5 = ymm4[8],ymm3[8],ymm4[9],ymm3[9],ymm4[10],ymm3[10],ymm4[11],ymm3[11],ymm4[12],ymm3[12],ymm4[13],ymm3[13],ymm4[14],ymm3[14],ymm4[15],ymm3[15],ymm4[24],ymm3[24],ymm4[25],ymm3[25],ymm4[26],ymm3[26],ymm4[27],ymm3[27],ymm4[28],ymm3[28],ymm4[29],ymm3[29],ymm4[30],ymm3[30],ymm4[31],ymm3[31]
; AVX512VL-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512VL-NEXT: vpsllw %xmm2, %ymm5, %ymm5
; AVX512VL-NEXT: vpsrlw $8, %ymm5, %ymm5
; AVX512VL-NEXT: vpunpcklbw {{.*#+}} ymm3 = ymm4[0],ymm3[0],ymm4[1],ymm3[1],ymm4[2],ymm3[2],ymm4[3],ymm3[3],ymm4[4],ymm3[4],ymm4[5],ymm3[5],ymm4[6],ymm3[6],ymm4[7],ymm3[7],ymm4[16],ymm3[16],ymm4[17],ymm3[17],ymm4[18],ymm3[18],ymm4[19],ymm3[19],ymm4[20],ymm3[20],ymm4[21],ymm3[21],ymm4[22],ymm3[22],ymm4[23],ymm3[23]
; AVX512VL-NEXT: vpsllw %xmm2, %ymm3, %ymm3
; AVX512VL-NEXT: vpsrlw $8, %ymm3, %ymm3
; AVX512VL-NEXT: vpackuswb %ymm5, %ymm3, %ymm3
; AVX512VL-NEXT: vpunpckhbw {{.*#+}} ymm4 = 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]
; AVX512VL-NEXT: vpsllw %xmm2, %ymm4, %ymm4
; AVX512VL-NEXT: vpsrlw $8, %ymm4, %ymm4
; AVX512VL-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]
; AVX512VL-NEXT: vpsllw %xmm2, %ymm0, %ymm0
; AVX512VL-NEXT: vpsrlw $8, %ymm0, %ymm0
; AVX512VL-NEXT: vpackuswb %ymm4, %ymm0, %ymm0
; AVX512VL-NEXT: vinserti64x4 $1, %ymm3, %zmm0, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: splatvar_funnnel_v64i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm0[8],zmm1[9],zmm0[9],zmm1[10],zmm0[10],zmm1[11],zmm0[11],zmm1[12],zmm0[12],zmm1[13],zmm0[13],zmm1[14],zmm0[14],zmm1[15],zmm0[15],zmm1[24],zmm0[24],zmm1[25],zmm0[25],zmm1[26],zmm0[26],zmm1[27],zmm0[27],zmm1[28],zmm0[28],zmm1[29],zmm0[29],zmm1[30],zmm0[30],zmm1[31],zmm0[31],zmm1[40],zmm0[40],zmm1[41],zmm0[41],zmm1[42],zmm0[42],zmm1[43],zmm0[43],zmm1[44],zmm0[44],zmm1[45],zmm0[45],zmm1[46],zmm0[46],zmm1[47],zmm0[47],zmm1[56],zmm0[56],zmm1[57],zmm0[57],zmm1[58],zmm0[58],zmm1[59],zmm0[59],zmm1[60],zmm0[60],zmm1[61],zmm0[61],zmm1[62],zmm0[62],zmm1[63],zmm0[63]
; AVX512BW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512BW-NEXT: vpsllw %xmm2, %zmm3, %zmm3
; AVX512BW-NEXT: vpsrlw $8, %zmm3, %zmm3
; AVX512BW-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm1[0],zmm0[0],zmm1[1],zmm0[1],zmm1[2],zmm0[2],zmm1[3],zmm0[3],zmm1[4],zmm0[4],zmm1[5],zmm0[5],zmm1[6],zmm0[6],zmm1[7],zmm0[7],zmm1[16],zmm0[16],zmm1[17],zmm0[17],zmm1[18],zmm0[18],zmm1[19],zmm0[19],zmm1[20],zmm0[20],zmm1[21],zmm0[21],zmm1[22],zmm0[22],zmm1[23],zmm0[23],zmm1[32],zmm0[32],zmm1[33],zmm0[33],zmm1[34],zmm0[34],zmm1[35],zmm0[35],zmm1[36],zmm0[36],zmm1[37],zmm0[37],zmm1[38],zmm0[38],zmm1[39],zmm0[39],zmm1[48],zmm0[48],zmm1[49],zmm0[49],zmm1[50],zmm0[50],zmm1[51],zmm0[51],zmm1[52],zmm0[52],zmm1[53],zmm0[53],zmm1[54],zmm0[54],zmm1[55],zmm0[55]
; AVX512BW-NEXT: vpsllw %xmm2, %zmm0, %zmm0
; AVX512BW-NEXT: vpsrlw $8, %zmm0, %zmm0
; AVX512BW-NEXT: vpackuswb %zmm3, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VBMI2-LABEL: splatvar_funnnel_v64i8:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm0[8],zmm1[9],zmm0[9],zmm1[10],zmm0[10],zmm1[11],zmm0[11],zmm1[12],zmm0[12],zmm1[13],zmm0[13],zmm1[14],zmm0[14],zmm1[15],zmm0[15],zmm1[24],zmm0[24],zmm1[25],zmm0[25],zmm1[26],zmm0[26],zmm1[27],zmm0[27],zmm1[28],zmm0[28],zmm1[29],zmm0[29],zmm1[30],zmm0[30],zmm1[31],zmm0[31],zmm1[40],zmm0[40],zmm1[41],zmm0[41],zmm1[42],zmm0[42],zmm1[43],zmm0[43],zmm1[44],zmm0[44],zmm1[45],zmm0[45],zmm1[46],zmm0[46],zmm1[47],zmm0[47],zmm1[56],zmm0[56],zmm1[57],zmm0[57],zmm1[58],zmm0[58],zmm1[59],zmm0[59],zmm1[60],zmm0[60],zmm1[61],zmm0[61],zmm1[62],zmm0[62],zmm1[63],zmm0[63]
; AVX512VBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512VBMI2-NEXT: vpsllw %xmm2, %zmm3, %zmm3
; AVX512VBMI2-NEXT: vpsrlw $8, %zmm3, %zmm3
; AVX512VBMI2-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm1[0],zmm0[0],zmm1[1],zmm0[1],zmm1[2],zmm0[2],zmm1[3],zmm0[3],zmm1[4],zmm0[4],zmm1[5],zmm0[5],zmm1[6],zmm0[6],zmm1[7],zmm0[7],zmm1[16],zmm0[16],zmm1[17],zmm0[17],zmm1[18],zmm0[18],zmm1[19],zmm0[19],zmm1[20],zmm0[20],zmm1[21],zmm0[21],zmm1[22],zmm0[22],zmm1[23],zmm0[23],zmm1[32],zmm0[32],zmm1[33],zmm0[33],zmm1[34],zmm0[34],zmm1[35],zmm0[35],zmm1[36],zmm0[36],zmm1[37],zmm0[37],zmm1[38],zmm0[38],zmm1[39],zmm0[39],zmm1[48],zmm0[48],zmm1[49],zmm0[49],zmm1[50],zmm0[50],zmm1[51],zmm0[51],zmm1[52],zmm0[52],zmm1[53],zmm0[53],zmm1[54],zmm0[54],zmm1[55],zmm0[55]
; AVX512VBMI2-NEXT: vpsllw %xmm2, %zmm0, %zmm0
; AVX512VBMI2-NEXT: vpsrlw $8, %zmm0, %zmm0
; AVX512VBMI2-NEXT: vpackuswb %zmm3, %zmm0, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLBW-LABEL: splatvar_funnnel_v64i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm0[8],zmm1[9],zmm0[9],zmm1[10],zmm0[10],zmm1[11],zmm0[11],zmm1[12],zmm0[12],zmm1[13],zmm0[13],zmm1[14],zmm0[14],zmm1[15],zmm0[15],zmm1[24],zmm0[24],zmm1[25],zmm0[25],zmm1[26],zmm0[26],zmm1[27],zmm0[27],zmm1[28],zmm0[28],zmm1[29],zmm0[29],zmm1[30],zmm0[30],zmm1[31],zmm0[31],zmm1[40],zmm0[40],zmm1[41],zmm0[41],zmm1[42],zmm0[42],zmm1[43],zmm0[43],zmm1[44],zmm0[44],zmm1[45],zmm0[45],zmm1[46],zmm0[46],zmm1[47],zmm0[47],zmm1[56],zmm0[56],zmm1[57],zmm0[57],zmm1[58],zmm0[58],zmm1[59],zmm0[59],zmm1[60],zmm0[60],zmm1[61],zmm0[61],zmm1[62],zmm0[62],zmm1[63],zmm0[63]
; AVX512VLBW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512VLBW-NEXT: vpsllw %xmm2, %zmm3, %zmm3
; AVX512VLBW-NEXT: vpsrlw $8, %zmm3, %zmm3
; AVX512VLBW-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm1[0],zmm0[0],zmm1[1],zmm0[1],zmm1[2],zmm0[2],zmm1[3],zmm0[3],zmm1[4],zmm0[4],zmm1[5],zmm0[5],zmm1[6],zmm0[6],zmm1[7],zmm0[7],zmm1[16],zmm0[16],zmm1[17],zmm0[17],zmm1[18],zmm0[18],zmm1[19],zmm0[19],zmm1[20],zmm0[20],zmm1[21],zmm0[21],zmm1[22],zmm0[22],zmm1[23],zmm0[23],zmm1[32],zmm0[32],zmm1[33],zmm0[33],zmm1[34],zmm0[34],zmm1[35],zmm0[35],zmm1[36],zmm0[36],zmm1[37],zmm0[37],zmm1[38],zmm0[38],zmm1[39],zmm0[39],zmm1[48],zmm0[48],zmm1[49],zmm0[49],zmm1[50],zmm0[50],zmm1[51],zmm0[51],zmm1[52],zmm0[52],zmm1[53],zmm0[53],zmm1[54],zmm0[54],zmm1[55],zmm0[55]
; AVX512VLBW-NEXT: vpsllw %xmm2, %zmm0, %zmm0
; AVX512VLBW-NEXT: vpsrlw $8, %zmm0, %zmm0
; AVX512VLBW-NEXT: vpackuswb %zmm3, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VLVBMI2-LABEL: splatvar_funnnel_v64i8:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm0[8],zmm1[9],zmm0[9],zmm1[10],zmm0[10],zmm1[11],zmm0[11],zmm1[12],zmm0[12],zmm1[13],zmm0[13],zmm1[14],zmm0[14],zmm1[15],zmm0[15],zmm1[24],zmm0[24],zmm1[25],zmm0[25],zmm1[26],zmm0[26],zmm1[27],zmm0[27],zmm1[28],zmm0[28],zmm1[29],zmm0[29],zmm1[30],zmm0[30],zmm1[31],zmm0[31],zmm1[40],zmm0[40],zmm1[41],zmm0[41],zmm1[42],zmm0[42],zmm1[43],zmm0[43],zmm1[44],zmm0[44],zmm1[45],zmm0[45],zmm1[46],zmm0[46],zmm1[47],zmm0[47],zmm1[56],zmm0[56],zmm1[57],zmm0[57],zmm1[58],zmm0[58],zmm1[59],zmm0[59],zmm1[60],zmm0[60],zmm1[61],zmm0[61],zmm1[62],zmm0[62],zmm1[63],zmm0[63]
; AVX512VLVBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm2, %xmm2
; AVX512VLVBMI2-NEXT: vpsllw %xmm2, %zmm3, %zmm3
; AVX512VLVBMI2-NEXT: vpsrlw $8, %zmm3, %zmm3
; AVX512VLVBMI2-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm1[0],zmm0[0],zmm1[1],zmm0[1],zmm1[2],zmm0[2],zmm1[3],zmm0[3],zmm1[4],zmm0[4],zmm1[5],zmm0[5],zmm1[6],zmm0[6],zmm1[7],zmm0[7],zmm1[16],zmm0[16],zmm1[17],zmm0[17],zmm1[18],zmm0[18],zmm1[19],zmm0[19],zmm1[20],zmm0[20],zmm1[21],zmm0[21],zmm1[22],zmm0[22],zmm1[23],zmm0[23],zmm1[32],zmm0[32],zmm1[33],zmm0[33],zmm1[34],zmm0[34],zmm1[35],zmm0[35],zmm1[36],zmm0[36],zmm1[37],zmm0[37],zmm1[38],zmm0[38],zmm1[39],zmm0[39],zmm1[48],zmm0[48],zmm1[49],zmm0[49],zmm1[50],zmm0[50],zmm1[51],zmm0[51],zmm1[52],zmm0[52],zmm1[53],zmm0[53],zmm1[54],zmm0[54],zmm1[55],zmm0[55]
; AVX512VLVBMI2-NEXT: vpsllw %xmm2, %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: vpsrlw $8, %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: vpackuswb %zmm3, %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: retq
%splat = shufflevector <64 x i8> %amt, <64 x i8> undef, <64 x i32> zeroinitializer
%res = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> %splat)
ret <64 x i8> %res
}
;
; Constant Shifts
;
define <8 x i64> @constant_funnnel_v8i64(<8 x i64> %x, <8 x i64> %y) nounwind {
; AVX512F-LABEL: constant_funnnel_v8i64:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpsrlvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512F-NEXT: vpsllvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512F-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: constant_funnnel_v8i64:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpsrlvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512VL-NEXT: vpsllvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512VL-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: constant_funnnel_v8i64:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpsrlvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512BW-NEXT: vpsllvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512BW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VBMI2-LABEL: constant_funnnel_v8i64:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpshldvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLBW-LABEL: constant_funnnel_v8i64:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpsrlvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512VLBW-NEXT: vpsllvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512VLBW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VLVBMI2-LABEL: constant_funnnel_v8i64:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpshldvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm0
; AVX512VLVBMI2-NEXT: retq
%res = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %x, <8 x i64> %y, <8 x i64> <i64 4, i64 14, i64 50, i64 60, i64 4, i64 14, i64 50, i64 60>)
ret <8 x i64> %res
}
define <16 x i32> @constant_funnnel_v16i32(<16 x i32> %x, <16 x i32> %y) nounwind {
; AVX512F-LABEL: constant_funnnel_v16i32:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpsrlvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512F-NEXT: vpsllvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512F-NEXT: vpord %zmm1, %zmm0, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: constant_funnnel_v16i32:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpsrlvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512VL-NEXT: vpsllvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512VL-NEXT: vpord %zmm1, %zmm0, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: constant_funnnel_v16i32:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpsrlvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512BW-NEXT: vpsllvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512BW-NEXT: vpord %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VBMI2-LABEL: constant_funnnel_v16i32:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpshldvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLBW-LABEL: constant_funnnel_v16i32:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpsrlvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512VLBW-NEXT: vpsllvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512VLBW-NEXT: vpord %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VLVBMI2-LABEL: constant_funnnel_v16i32:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpshldvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm0
; AVX512VLVBMI2-NEXT: retq
%res = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>)
ret <16 x i32> %res
}
define <32 x i16> @constant_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y) nounwind {
; AVX512F-LABEL: constant_funnnel_v32i16:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm2
; AVX512F-NEXT: vpsrlw $1, %ymm2, %ymm2
; AVX512F-NEXT: vmovdqa {{.*#+}} ymm3 = <2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768,u>
; AVX512F-NEXT: vpmulhuw %ymm3, %ymm2, %ymm4
; AVX512F-NEXT: vpblendw {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5,6],ymm2[7],ymm4[8,9,10,11,12,13,14],ymm2[15]
; AVX512F-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7]
; AVX512F-NEXT: vpsrlw $1, %ymm1, %ymm1
; AVX512F-NEXT: vpmulhuw %ymm3, %ymm1, %ymm3
; AVX512F-NEXT: vpblendw {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5,6],ymm1[7],ymm3[8,9,10,11,12,13,14],ymm1[15]
; AVX512F-NEXT: vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
; AVX512F-NEXT: vinserti64x4 $1, %ymm2, %zmm1, %zmm1
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm2
; AVX512F-NEXT: vmovdqa {{.*#+}} ymm3 = [1,2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768]
; AVX512F-NEXT: vpmullw %ymm3, %ymm2, %ymm2
; AVX512F-NEXT: vpmullw %ymm3, %ymm0, %ymm0
; AVX512F-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
; AVX512F-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: constant_funnnel_v32i16:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vextracti64x4 $1, %zmm1, %ymm2
; AVX512VL-NEXT: vpsrlw $1, %ymm2, %ymm2
; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm3 = <2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768,u>
; AVX512VL-NEXT: vpmulhuw %ymm3, %ymm2, %ymm4
; AVX512VL-NEXT: vpblendw {{.*#+}} ymm2 = ymm4[0,1,2,3,4,5,6],ymm2[7],ymm4[8,9,10,11,12,13,14],ymm2[15]
; AVX512VL-NEXT: vpblendd {{.*#+}} ymm2 = ymm4[0,1,2,3],ymm2[4,5,6,7]
; AVX512VL-NEXT: vpsrlw $1, %ymm1, %ymm1
; AVX512VL-NEXT: vpmulhuw %ymm3, %ymm1, %ymm3
; AVX512VL-NEXT: vpblendw {{.*#+}} ymm1 = ymm3[0,1,2,3,4,5,6],ymm1[7],ymm3[8,9,10,11,12,13,14],ymm1[15]
; AVX512VL-NEXT: vpblendd {{.*#+}} ymm1 = ymm3[0,1,2,3],ymm1[4,5,6,7]
; AVX512VL-NEXT: vinserti64x4 $1, %ymm2, %zmm1, %zmm1
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm2
; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm3 = [1,2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768]
; AVX512VL-NEXT: vpmullw %ymm3, %ymm2, %ymm2
; AVX512VL-NEXT: vpmullw %ymm3, %ymm0, %ymm0
; AVX512VL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
; AVX512VL-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: constant_funnnel_v32i16:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512BW-NEXT: vpsrlw $1, %zmm1, %zmm1
; AVX512BW-NEXT: vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512BW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VBMI2-LABEL: constant_funnnel_v32i16:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpshldvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLBW-LABEL: constant_funnnel_v32i16:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512VLBW-NEXT: vpsrlw $1, %zmm1, %zmm1
; AVX512VLBW-NEXT: vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512VLBW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VLVBMI2-LABEL: constant_funnnel_v32i16:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpshldvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm0
; AVX512VLVBMI2-NEXT: retq
%res = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>)
ret <32 x i16> %res
}
define <64 x i8> @constant_funnnel_v64i8(<64 x i8> %x, <64 x i8> %y) nounwind {
; AVX512F-LABEL: constant_funnnel_v64i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm2
; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm3
; AVX512F-NEXT: vpunpckhbw {{.*#+}} ymm4 = ymm3[8],ymm2[8],ymm3[9],ymm2[9],ymm3[10],ymm2[10],ymm3[11],ymm2[11],ymm3[12],ymm2[12],ymm3[13],ymm2[13],ymm3[14],ymm2[14],ymm3[15],ymm2[15],ymm3[24],ymm2[24],ymm3[25],ymm2[25],ymm3[26],ymm2[26],ymm3[27],ymm2[27],ymm3[28],ymm2[28],ymm3[29],ymm2[29],ymm3[30],ymm2[30],ymm3[31],ymm2[31]
; AVX512F-NEXT: vbroadcasti128 {{.*#+}} ymm5 = [1,128,64,32,16,8,4,2,1,128,64,32,16,8,4,2]
; AVX512F-NEXT: # ymm5 = mem[0,1,0,1]
; AVX512F-NEXT: vpmullw %ymm5, %ymm4, %ymm4
; AVX512F-NEXT: vpsrlw $8, %ymm4, %ymm4
; AVX512F-NEXT: vpunpcklbw {{.*#+}} ymm2 = ymm3[0],ymm2[0],ymm3[1],ymm2[1],ymm3[2],ymm2[2],ymm3[3],ymm2[3],ymm3[4],ymm2[4],ymm3[5],ymm2[5],ymm3[6],ymm2[6],ymm3[7],ymm2[7],ymm3[16],ymm2[16],ymm3[17],ymm2[17],ymm3[18],ymm2[18],ymm3[19],ymm2[19],ymm3[20],ymm2[20],ymm3[21],ymm2[21],ymm3[22],ymm2[22],ymm3[23],ymm2[23]
; AVX512F-NEXT: vbroadcasti128 {{.*#+}} ymm3 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]
; AVX512F-NEXT: # ymm3 = mem[0,1,0,1]
; AVX512F-NEXT: vpmullw %ymm3, %ymm2, %ymm2
; AVX512F-NEXT: vpsrlw $8, %ymm2, %ymm2
; AVX512F-NEXT: vpackuswb %ymm4, %ymm2, %ymm2
; AVX512F-NEXT: vpunpckhbw {{.*#+}} ymm4 = 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]
; AVX512F-NEXT: vpmullw %ymm5, %ymm4, %ymm4
; AVX512F-NEXT: vpsrlw $8, %ymm4, %ymm4
; AVX512F-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]
; AVX512F-NEXT: vpmullw %ymm3, %ymm0, %ymm0
; AVX512F-NEXT: vpsrlw $8, %ymm0, %ymm0
; AVX512F-NEXT: vpackuswb %ymm4, %ymm0, %ymm0
; AVX512F-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: constant_funnnel_v64i8:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm2
; AVX512VL-NEXT: vextracti64x4 $1, %zmm1, %ymm3
; AVX512VL-NEXT: vpunpckhbw {{.*#+}} ymm4 = ymm3[8],ymm2[8],ymm3[9],ymm2[9],ymm3[10],ymm2[10],ymm3[11],ymm2[11],ymm3[12],ymm2[12],ymm3[13],ymm2[13],ymm3[14],ymm2[14],ymm3[15],ymm2[15],ymm3[24],ymm2[24],ymm3[25],ymm2[25],ymm3[26],ymm2[26],ymm3[27],ymm2[27],ymm3[28],ymm2[28],ymm3[29],ymm2[29],ymm3[30],ymm2[30],ymm3[31],ymm2[31]
; AVX512VL-NEXT: vbroadcasti128 {{.*#+}} ymm5 = [1,128,64,32,16,8,4,2,1,128,64,32,16,8,4,2]
; AVX512VL-NEXT: # ymm5 = mem[0,1,0,1]
; AVX512VL-NEXT: vpmullw %ymm5, %ymm4, %ymm4
; AVX512VL-NEXT: vpsrlw $8, %ymm4, %ymm4
; AVX512VL-NEXT: vpunpcklbw {{.*#+}} ymm2 = ymm3[0],ymm2[0],ymm3[1],ymm2[1],ymm3[2],ymm2[2],ymm3[3],ymm2[3],ymm3[4],ymm2[4],ymm3[5],ymm2[5],ymm3[6],ymm2[6],ymm3[7],ymm2[7],ymm3[16],ymm2[16],ymm3[17],ymm2[17],ymm3[18],ymm2[18],ymm3[19],ymm2[19],ymm3[20],ymm2[20],ymm3[21],ymm2[21],ymm3[22],ymm2[22],ymm3[23],ymm2[23]
; AVX512VL-NEXT: vbroadcasti128 {{.*#+}} ymm3 = [1,2,4,8,16,32,64,128,1,2,4,8,16,32,64,128]
; AVX512VL-NEXT: # ymm3 = mem[0,1,0,1]
; AVX512VL-NEXT: vpmullw %ymm3, %ymm2, %ymm2
; AVX512VL-NEXT: vpsrlw $8, %ymm2, %ymm2
; AVX512VL-NEXT: vpackuswb %ymm4, %ymm2, %ymm2
; AVX512VL-NEXT: vpunpckhbw {{.*#+}} ymm4 = 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]
; AVX512VL-NEXT: vpmullw %ymm5, %ymm4, %ymm4
; AVX512VL-NEXT: vpsrlw $8, %ymm4, %ymm4
; AVX512VL-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]
; AVX512VL-NEXT: vpmullw %ymm3, %ymm0, %ymm0
; AVX512VL-NEXT: vpsrlw $8, %ymm0, %ymm0
; AVX512VL-NEXT: vpackuswb %ymm4, %ymm0, %ymm0
; AVX512VL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: constant_funnnel_v64i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpunpckhbw {{.*#+}} zmm2 = zmm1[8],zmm0[8],zmm1[9],zmm0[9],zmm1[10],zmm0[10],zmm1[11],zmm0[11],zmm1[12],zmm0[12],zmm1[13],zmm0[13],zmm1[14],zmm0[14],zmm1[15],zmm0[15],zmm1[24],zmm0[24],zmm1[25],zmm0[25],zmm1[26],zmm0[26],zmm1[27],zmm0[27],zmm1[28],zmm0[28],zmm1[29],zmm0[29],zmm1[30],zmm0[30],zmm1[31],zmm0[31],zmm1[40],zmm0[40],zmm1[41],zmm0[41],zmm1[42],zmm0[42],zmm1[43],zmm0[43],zmm1[44],zmm0[44],zmm1[45],zmm0[45],zmm1[46],zmm0[46],zmm1[47],zmm0[47],zmm1[56],zmm0[56],zmm1[57],zmm0[57],zmm1[58],zmm0[58],zmm1[59],zmm0[59],zmm1[60],zmm0[60],zmm1[61],zmm0[61],zmm1[62],zmm0[62],zmm1[63],zmm0[63]
; AVX512BW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
; AVX512BW-NEXT: vpsrlw $8, %zmm2, %zmm2
; AVX512BW-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm1[0],zmm0[0],zmm1[1],zmm0[1],zmm1[2],zmm0[2],zmm1[3],zmm0[3],zmm1[4],zmm0[4],zmm1[5],zmm0[5],zmm1[6],zmm0[6],zmm1[7],zmm0[7],zmm1[16],zmm0[16],zmm1[17],zmm0[17],zmm1[18],zmm0[18],zmm1[19],zmm0[19],zmm1[20],zmm0[20],zmm1[21],zmm0[21],zmm1[22],zmm0[22],zmm1[23],zmm0[23],zmm1[32],zmm0[32],zmm1[33],zmm0[33],zmm1[34],zmm0[34],zmm1[35],zmm0[35],zmm1[36],zmm0[36],zmm1[37],zmm0[37],zmm1[38],zmm0[38],zmm1[39],zmm0[39],zmm1[48],zmm0[48],zmm1[49],zmm0[49],zmm1[50],zmm0[50],zmm1[51],zmm0[51],zmm1[52],zmm0[52],zmm1[53],zmm0[53],zmm1[54],zmm0[54],zmm1[55],zmm0[55]
; AVX512BW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512BW-NEXT: vpsrlw $8, %zmm0, %zmm0
; AVX512BW-NEXT: vpackuswb %zmm2, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VBMI2-LABEL: constant_funnnel_v64i8:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} zmm2 = zmm1[8],zmm0[8],zmm1[9],zmm0[9],zmm1[10],zmm0[10],zmm1[11],zmm0[11],zmm1[12],zmm0[12],zmm1[13],zmm0[13],zmm1[14],zmm0[14],zmm1[15],zmm0[15],zmm1[24],zmm0[24],zmm1[25],zmm0[25],zmm1[26],zmm0[26],zmm1[27],zmm0[27],zmm1[28],zmm0[28],zmm1[29],zmm0[29],zmm1[30],zmm0[30],zmm1[31],zmm0[31],zmm1[40],zmm0[40],zmm1[41],zmm0[41],zmm1[42],zmm0[42],zmm1[43],zmm0[43],zmm1[44],zmm0[44],zmm1[45],zmm0[45],zmm1[46],zmm0[46],zmm1[47],zmm0[47],zmm1[56],zmm0[56],zmm1[57],zmm0[57],zmm1[58],zmm0[58],zmm1[59],zmm0[59],zmm1[60],zmm0[60],zmm1[61],zmm0[61],zmm1[62],zmm0[62],zmm1[63],zmm0[63]
; AVX512VBMI2-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
; AVX512VBMI2-NEXT: vpsrlw $8, %zmm2, %zmm2
; AVX512VBMI2-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm1[0],zmm0[0],zmm1[1],zmm0[1],zmm1[2],zmm0[2],zmm1[3],zmm0[3],zmm1[4],zmm0[4],zmm1[5],zmm0[5],zmm1[6],zmm0[6],zmm1[7],zmm0[7],zmm1[16],zmm0[16],zmm1[17],zmm0[17],zmm1[18],zmm0[18],zmm1[19],zmm0[19],zmm1[20],zmm0[20],zmm1[21],zmm0[21],zmm1[22],zmm0[22],zmm1[23],zmm0[23],zmm1[32],zmm0[32],zmm1[33],zmm0[33],zmm1[34],zmm0[34],zmm1[35],zmm0[35],zmm1[36],zmm0[36],zmm1[37],zmm0[37],zmm1[38],zmm0[38],zmm1[39],zmm0[39],zmm1[48],zmm0[48],zmm1[49],zmm0[49],zmm1[50],zmm0[50],zmm1[51],zmm0[51],zmm1[52],zmm0[52],zmm1[53],zmm0[53],zmm1[54],zmm0[54],zmm1[55],zmm0[55]
; AVX512VBMI2-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512VBMI2-NEXT: vpsrlw $8, %zmm0, %zmm0
; AVX512VBMI2-NEXT: vpackuswb %zmm2, %zmm0, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLBW-LABEL: constant_funnnel_v64i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} zmm2 = zmm1[8],zmm0[8],zmm1[9],zmm0[9],zmm1[10],zmm0[10],zmm1[11],zmm0[11],zmm1[12],zmm0[12],zmm1[13],zmm0[13],zmm1[14],zmm0[14],zmm1[15],zmm0[15],zmm1[24],zmm0[24],zmm1[25],zmm0[25],zmm1[26],zmm0[26],zmm1[27],zmm0[27],zmm1[28],zmm0[28],zmm1[29],zmm0[29],zmm1[30],zmm0[30],zmm1[31],zmm0[31],zmm1[40],zmm0[40],zmm1[41],zmm0[41],zmm1[42],zmm0[42],zmm1[43],zmm0[43],zmm1[44],zmm0[44],zmm1[45],zmm0[45],zmm1[46],zmm0[46],zmm1[47],zmm0[47],zmm1[56],zmm0[56],zmm1[57],zmm0[57],zmm1[58],zmm0[58],zmm1[59],zmm0[59],zmm1[60],zmm0[60],zmm1[61],zmm0[61],zmm1[62],zmm0[62],zmm1[63],zmm0[63]
; AVX512VLBW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
; AVX512VLBW-NEXT: vpsrlw $8, %zmm2, %zmm2
; AVX512VLBW-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm1[0],zmm0[0],zmm1[1],zmm0[1],zmm1[2],zmm0[2],zmm1[3],zmm0[3],zmm1[4],zmm0[4],zmm1[5],zmm0[5],zmm1[6],zmm0[6],zmm1[7],zmm0[7],zmm1[16],zmm0[16],zmm1[17],zmm0[17],zmm1[18],zmm0[18],zmm1[19],zmm0[19],zmm1[20],zmm0[20],zmm1[21],zmm0[21],zmm1[22],zmm0[22],zmm1[23],zmm0[23],zmm1[32],zmm0[32],zmm1[33],zmm0[33],zmm1[34],zmm0[34],zmm1[35],zmm0[35],zmm1[36],zmm0[36],zmm1[37],zmm0[37],zmm1[38],zmm0[38],zmm1[39],zmm0[39],zmm1[48],zmm0[48],zmm1[49],zmm0[49],zmm1[50],zmm0[50],zmm1[51],zmm0[51],zmm1[52],zmm0[52],zmm1[53],zmm0[53],zmm1[54],zmm0[54],zmm1[55],zmm0[55]
; AVX512VLBW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512VLBW-NEXT: vpsrlw $8, %zmm0, %zmm0
; AVX512VLBW-NEXT: vpackuswb %zmm2, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VLVBMI2-LABEL: constant_funnnel_v64i8:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} zmm2 = zmm1[8],zmm0[8],zmm1[9],zmm0[9],zmm1[10],zmm0[10],zmm1[11],zmm0[11],zmm1[12],zmm0[12],zmm1[13],zmm0[13],zmm1[14],zmm0[14],zmm1[15],zmm0[15],zmm1[24],zmm0[24],zmm1[25],zmm0[25],zmm1[26],zmm0[26],zmm1[27],zmm0[27],zmm1[28],zmm0[28],zmm1[29],zmm0[29],zmm1[30],zmm0[30],zmm1[31],zmm0[31],zmm1[40],zmm0[40],zmm1[41],zmm0[41],zmm1[42],zmm0[42],zmm1[43],zmm0[43],zmm1[44],zmm0[44],zmm1[45],zmm0[45],zmm1[46],zmm0[46],zmm1[47],zmm0[47],zmm1[56],zmm0[56],zmm1[57],zmm0[57],zmm1[58],zmm0[58],zmm1[59],zmm0[59],zmm1[60],zmm0[60],zmm1[61],zmm0[61],zmm1[62],zmm0[62],zmm1[63],zmm0[63]
; AVX512VLVBMI2-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm2, %zmm2
; AVX512VLVBMI2-NEXT: vpsrlw $8, %zmm2, %zmm2
; AVX512VLVBMI2-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm1[0],zmm0[0],zmm1[1],zmm0[1],zmm1[2],zmm0[2],zmm1[3],zmm0[3],zmm1[4],zmm0[4],zmm1[5],zmm0[5],zmm1[6],zmm0[6],zmm1[7],zmm0[7],zmm1[16],zmm0[16],zmm1[17],zmm0[17],zmm1[18],zmm0[18],zmm1[19],zmm0[19],zmm1[20],zmm0[20],zmm1[21],zmm0[21],zmm1[22],zmm0[22],zmm1[23],zmm0[23],zmm1[32],zmm0[32],zmm1[33],zmm0[33],zmm1[34],zmm0[34],zmm1[35],zmm0[35],zmm1[36],zmm0[36],zmm1[37],zmm0[37],zmm1[38],zmm0[38],zmm1[39],zmm0[39],zmm1[48],zmm0[48],zmm1[49],zmm0[49],zmm1[50],zmm0[50],zmm1[51],zmm0[51],zmm1[52],zmm0[52],zmm1[53],zmm0[53],zmm1[54],zmm0[54],zmm1[55],zmm0[55]
; AVX512VLVBMI2-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: vpsrlw $8, %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: vpackuswb %zmm2, %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: retq
%res = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1>)
ret <64 x i8> %res
}
;
; Uniform Constant Shifts
;
define <8 x i64> @splatconstant_funnnel_v8i64(<8 x i64> %x, <8 x i64> %y) nounwind {
; AVX512F-LABEL: splatconstant_funnnel_v8i64:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpsrlq $50, %zmm1, %zmm1
; AVX512F-NEXT: vpsllq $14, %zmm0, %zmm0
; AVX512F-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: splatconstant_funnnel_v8i64:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpsrlq $50, %zmm1, %zmm1
; AVX512VL-NEXT: vpsllq $14, %zmm0, %zmm0
; AVX512VL-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: splatconstant_funnnel_v8i64:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpsrlq $50, %zmm1, %zmm1
; AVX512BW-NEXT: vpsllq $14, %zmm0, %zmm0
; AVX512BW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VBMI2-LABEL: splatconstant_funnnel_v8i64:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpshldq $14, %zmm1, %zmm0, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLBW-LABEL: splatconstant_funnnel_v8i64:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpsrlq $50, %zmm1, %zmm1
; AVX512VLBW-NEXT: vpsllq $14, %zmm0, %zmm0
; AVX512VLBW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VLVBMI2-LABEL: splatconstant_funnnel_v8i64:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpshldq $14, %zmm1, %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: retq
%res = call <8 x i64> @llvm.fshl.v8i64(<8 x i64> %x, <8 x i64> %y, <8 x i64> <i64 14, i64 14, i64 14, i64 14, i64 14, i64 14, i64 14, i64 14>)
ret <8 x i64> %res
}
define <16 x i32> @splatconstant_funnnel_v16i32(<16 x i32> %x, <16 x i32> %y) nounwind {
; AVX512F-LABEL: splatconstant_funnnel_v16i32:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpsrld $28, %zmm1, %zmm1
; AVX512F-NEXT: vpslld $4, %zmm0, %zmm0
; AVX512F-NEXT: vpord %zmm1, %zmm0, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: splatconstant_funnnel_v16i32:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpsrld $28, %zmm1, %zmm1
; AVX512VL-NEXT: vpslld $4, %zmm0, %zmm0
; AVX512VL-NEXT: vpord %zmm1, %zmm0, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: splatconstant_funnnel_v16i32:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpsrld $28, %zmm1, %zmm1
; AVX512BW-NEXT: vpslld $4, %zmm0, %zmm0
; AVX512BW-NEXT: vpord %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VBMI2-LABEL: splatconstant_funnnel_v16i32:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpshldd $4, %zmm1, %zmm0, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLBW-LABEL: splatconstant_funnnel_v16i32:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpsrld $28, %zmm1, %zmm1
; AVX512VLBW-NEXT: vpslld $4, %zmm0, %zmm0
; AVX512VLBW-NEXT: vpord %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VLVBMI2-LABEL: splatconstant_funnnel_v16i32:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpshldd $4, %zmm1, %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: retq
%res = call <16 x i32> @llvm.fshl.v16i32(<16 x i32> %x, <16 x i32> %y, <16 x i32> <i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4, i32 4>)
ret <16 x i32> %res
}
define <32 x i16> @splatconstant_funnnel_v32i16(<32 x i16> %x, <32 x i16> %y) nounwind {
; AVX512F-LABEL: splatconstant_funnnel_v32i16:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpsrlw $9, %ymm1, %ymm2
; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm1
; AVX512F-NEXT: vpsrlw $9, %ymm1, %ymm1
; AVX512F-NEXT: vinserti64x4 $1, %ymm1, %zmm2, %zmm1
; AVX512F-NEXT: vpsllw $7, %ymm0, %ymm2
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm0
; AVX512F-NEXT: vpsllw $7, %ymm0, %ymm0
; AVX512F-NEXT: vinserti64x4 $1, %ymm0, %zmm2, %zmm0
; AVX512F-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: splatconstant_funnnel_v32i16:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpsrlw $9, %ymm1, %ymm2
; AVX512VL-NEXT: vextracti64x4 $1, %zmm1, %ymm1
; AVX512VL-NEXT: vpsrlw $9, %ymm1, %ymm1
; AVX512VL-NEXT: vinserti64x4 $1, %ymm1, %zmm2, %zmm1
; AVX512VL-NEXT: vpsllw $7, %ymm0, %ymm2
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
; AVX512VL-NEXT: vpsllw $7, %ymm0, %ymm0
; AVX512VL-NEXT: vinserti64x4 $1, %ymm0, %zmm2, %zmm0
; AVX512VL-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: splatconstant_funnnel_v32i16:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpsrlw $9, %zmm1, %zmm1
; AVX512BW-NEXT: vpsllw $7, %zmm0, %zmm0
; AVX512BW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VBMI2-LABEL: splatconstant_funnnel_v32i16:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpshldw $7, %zmm1, %zmm0, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLBW-LABEL: splatconstant_funnnel_v32i16:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpsrlw $9, %zmm1, %zmm1
; AVX512VLBW-NEXT: vpsllw $7, %zmm0, %zmm0
; AVX512VLBW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VLVBMI2-LABEL: splatconstant_funnnel_v32i16:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpshldw $7, %zmm1, %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: retq
%res = call <32 x i16> @llvm.fshl.v32i16(<32 x i16> %x, <32 x i16> %y, <32 x i16> <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>)
ret <32 x i16> %res
}
define <64 x i8> @splatconstant_funnnel_v64i8(<64 x i8> %x, <64 x i8> %y) nounwind {
; AVX512F-LABEL: splatconstant_funnnel_v64i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpsllw $4, %ymm0, %ymm2
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm0
; AVX512F-NEXT: vpsllw $4, %ymm0, %ymm0
; AVX512F-NEXT: vinserti64x4 $1, %ymm0, %zmm2, %zmm2
; AVX512F-NEXT: vpsrlw $4, %ymm1, %ymm0
; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm1
; AVX512F-NEXT: vpsrlw $4, %ymm1, %ymm1
; AVX512F-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
; AVX512F-NEXT: vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm2, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: splatconstant_funnnel_v64i8:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpsllw $4, %ymm0, %ymm2
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm0
; AVX512VL-NEXT: vpsllw $4, %ymm0, %ymm0
; AVX512VL-NEXT: vinserti64x4 $1, %ymm0, %zmm2, %zmm2
; AVX512VL-NEXT: vpsrlw $4, %ymm1, %ymm0
; AVX512VL-NEXT: vextracti64x4 $1, %zmm1, %ymm1
; AVX512VL-NEXT: vpsrlw $4, %ymm1, %ymm1
; AVX512VL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
; AVX512VL-NEXT: vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm2, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: splatconstant_funnnel_v64i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpsllw $4, %zmm0, %zmm2
; AVX512BW-NEXT: vpsrlw $4, %zmm1, %zmm0
; AVX512BW-NEXT: vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm2, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VBMI2-LABEL: splatconstant_funnnel_v64i8:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpsllw $4, %zmm0, %zmm2
; AVX512VBMI2-NEXT: vpsrlw $4, %zmm1, %zmm0
; AVX512VBMI2-NEXT: vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm2, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLBW-LABEL: splatconstant_funnnel_v64i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpsllw $4, %zmm0, %zmm2
; AVX512VLBW-NEXT: vpsrlw $4, %zmm1, %zmm0
; AVX512VLBW-NEXT: vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm2, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VLVBMI2-LABEL: splatconstant_funnnel_v64i8:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpsllw $4, %zmm0, %zmm2
; AVX512VLVBMI2-NEXT: vpsrlw $4, %zmm1, %zmm0
; AVX512VLVBMI2-NEXT: vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm2, %zmm0
; AVX512VLVBMI2-NEXT: retq
%res = call <64 x i8> @llvm.fshl.v64i8(<64 x i8> %x, <64 x i8> %y, <64 x i8> <i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4, i8 4>)
ret <64 x i8> %res
}