; 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=AVX512,AVX512F
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512VL
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX512,AVX512BW
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512VLBW
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vbmi,+avx512vbmi2 | FileCheck %s --check-prefixes=AVX512,AVX512VBMI2
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vbmi,+avx512vbmi2,+avx512vl | FileCheck %s --check-prefixes=AVX512,AVX512VLVBMI2
declare <8 x i64> @llvm.fshr.v8i64(<8 x i64>, <8 x i64>, <8 x i64>)
declare <16 x i32> @llvm.fshr.v16i32(<16 x i32>, <16 x i32>, <16 x i32>)
declare <32 x i16> @llvm.fshr.v32i16(<32 x i16>, <32 x i16>, <32 x i16>)
declare <64 x i8> @llvm.fshr.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> %amt) nounwind {
; AVX512-LABEL: var_funnnel_v8i64:
; AVX512: # %bb.0:
; AVX512-NEXT: vprorvq %zmm1, %zmm0, %zmm0
; AVX512-NEXT: retq
%res = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %x, <8 x i64> %x, <8 x i64> %amt)
ret <8 x i64> %res
}
define <16 x i32> @var_funnnel_v16i32(<16 x i32> %x, <16 x i32> %amt) nounwind {
; AVX512-LABEL: var_funnnel_v16i32:
; AVX512: # %bb.0:
; AVX512-NEXT: vprorvd %zmm1, %zmm0, %zmm0
; AVX512-NEXT: retq
%res = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %x, <16 x i32> %x, <16 x i32> %amt)
ret <16 x i32> %res
}
define <32 x i16> @var_funnnel_v32i16(<32 x i16> %x, <32 x i16> %amt) nounwind {
; AVX512F-LABEL: var_funnnel_v32i16:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512F-NEXT: vpand %ymm2, %ymm1, %ymm3
; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm4 = ymm3[0],zero,ymm3[1],zero,ymm3[2],zero,ymm3[3],zero,ymm3[4],zero,ymm3[5],zero,ymm3[6],zero,ymm3[7],zero,ymm3[8],zero,ymm3[9],zero,ymm3[10],zero,ymm3[11],zero,ymm3[12],zero,ymm3[13],zero,ymm3[14],zero,ymm3[15],zero
; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm5 = 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: vpsrlvd %zmm4, %zmm5, %zmm4
; AVX512F-NEXT: vmovdqa {{.*#+}} ymm6 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
; AVX512F-NEXT: vpsubw %ymm3, %ymm6, %ymm3
; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm3 = ymm3[0],zero,ymm3[1],zero,ymm3[2],zero,ymm3[3],zero,ymm3[4],zero,ymm3[5],zero,ymm3[6],zero,ymm3[7],zero,ymm3[8],zero,ymm3[9],zero,ymm3[10],zero,ymm3[11],zero,ymm3[12],zero,ymm3[13],zero,ymm3[14],zero,ymm3[15],zero
; AVX512F-NEXT: vpsllvd %zmm3, %zmm5, %zmm3
; AVX512F-NEXT: vpord %zmm3, %zmm4, %zmm3
; AVX512F-NEXT: vpmovdw %zmm3, %ymm3
; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm1
; AVX512F-NEXT: vpand %ymm2, %ymm1, %ymm1
; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm2 = 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: vpsrlvd %zmm2, %zmm0, %zmm2
; AVX512F-NEXT: vpsubw %ymm1, %ymm6, %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: vpord %zmm0, %zmm2, %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: vmovdqa {{.*#+}} ymm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
; AVX512VL-NEXT: vpand %ymm2, %ymm1, %ymm3
; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm4 = ymm3[0],zero,ymm3[1],zero,ymm3[2],zero,ymm3[3],zero,ymm3[4],zero,ymm3[5],zero,ymm3[6],zero,ymm3[7],zero,ymm3[8],zero,ymm3[9],zero,ymm3[10],zero,ymm3[11],zero,ymm3[12],zero,ymm3[13],zero,ymm3[14],zero,ymm3[15],zero
; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm5 = 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: vpsrlvd %zmm4, %zmm5, %zmm4
; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm6 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
; AVX512VL-NEXT: vpsubw %ymm3, %ymm6, %ymm3
; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm3 = ymm3[0],zero,ymm3[1],zero,ymm3[2],zero,ymm3[3],zero,ymm3[4],zero,ymm3[5],zero,ymm3[6],zero,ymm3[7],zero,ymm3[8],zero,ymm3[9],zero,ymm3[10],zero,ymm3[11],zero,ymm3[12],zero,ymm3[13],zero,ymm3[14],zero,ymm3[15],zero
; AVX512VL-NEXT: vpsllvd %zmm3, %zmm5, %zmm3
; AVX512VL-NEXT: vpord %zmm3, %zmm4, %zmm3
; AVX512VL-NEXT: vpmovdw %zmm3, %ymm3
; AVX512VL-NEXT: vextracti64x4 $1, %zmm1, %ymm1
; AVX512VL-NEXT: vpand %ymm2, %ymm1, %ymm1
; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm2 = 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: vpsrlvd %zmm2, %zmm0, %zmm2
; AVX512VL-NEXT: vpsubw %ymm1, %ymm6, %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: vpord %zmm0, %zmm2, %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: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm2
; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
; AVX512BW-NEXT: vpsubw %zmm1, %zmm3, %zmm1
; AVX512BW-NEXT: vpsllvw %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vporq %zmm0, %zmm2, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VLBW-LABEL: var_funnnel_v32i16:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512VLBW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm2
; AVX512VLBW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16]
; AVX512VLBW-NEXT: vpsubw %zmm1, %zmm3, %zmm1
; AVX512VLBW-NEXT: vpsllvw %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: vporq %zmm0, %zmm2, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VBMI2-LABEL: var_funnnel_v32i16:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpshrdvw %zmm1, %zmm0, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLVBMI2-LABEL: var_funnnel_v32i16:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpshrdvw %zmm1, %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: retq
%res = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %x, <32 x i16> %x, <32 x i16> %amt)
ret <32 x i16> %res
}
define <64 x i8> @var_funnnel_v64i8(<64 x i8> %x, <64 x i8> %amt) nounwind {
; AVX512F-LABEL: var_funnnel_v64i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm2
; AVX512F-NEXT: vpsllw $4, %ymm2, %ymm3
; AVX512F-NEXT: vpsrlw $4, %ymm2, %ymm4
; AVX512F-NEXT: vpbroadcastq {{.*#+}} zmm5 = [1085102592571150095,1085102592571150095,1085102592571150095,1085102592571150095,1085102592571150095,1085102592571150095,1085102592571150095,1085102592571150095]
; AVX512F-NEXT: vpternlogq $226, %zmm3, %zmm5, %zmm4
; AVX512F-NEXT: vextracti64x4 $1, %zmm1, %ymm3
; AVX512F-NEXT: vpsllw $5, %ymm3, %ymm3
; AVX512F-NEXT: vpblendvb %ymm3, %ymm4, %ymm2, %ymm2
; AVX512F-NEXT: vpsllw $6, %ymm2, %ymm4
; AVX512F-NEXT: vpsrlw $2, %ymm2, %ymm6
; AVX512F-NEXT: vpbroadcastq {{.*#+}} zmm7 = [4557430888798830399,4557430888798830399,4557430888798830399,4557430888798830399,4557430888798830399,4557430888798830399,4557430888798830399,4557430888798830399]
; AVX512F-NEXT: vpternlogq $226, %zmm4, %zmm7, %zmm6
; AVX512F-NEXT: vpaddb %ymm3, %ymm3, %ymm3
; AVX512F-NEXT: vpblendvb %ymm3, %ymm6, %ymm2, %ymm2
; AVX512F-NEXT: vpsllw $7, %ymm2, %ymm4
; AVX512F-NEXT: vpsrlw $1, %ymm2, %ymm6
; AVX512F-NEXT: vpbroadcastq {{.*#+}} zmm8 = [9187201950435737471,9187201950435737471,9187201950435737471,9187201950435737471,9187201950435737471,9187201950435737471,9187201950435737471,9187201950435737471]
; AVX512F-NEXT: vpternlogq $226, %zmm4, %zmm8, %zmm6
; AVX512F-NEXT: vpaddb %ymm3, %ymm3, %ymm3
; AVX512F-NEXT: vpblendvb %ymm3, %ymm6, %ymm2, %ymm2
; AVX512F-NEXT: vpsllw $4, %ymm0, %ymm3
; AVX512F-NEXT: vpsrlw $4, %ymm0, %ymm4
; AVX512F-NEXT: vpternlogq $226, %zmm3, %zmm5, %zmm4
; AVX512F-NEXT: vpsllw $5, %ymm1, %ymm1
; AVX512F-NEXT: vpblendvb %ymm1, %ymm4, %ymm0, %ymm0
; AVX512F-NEXT: vpsllw $6, %ymm0, %ymm3
; AVX512F-NEXT: vpsrlw $2, %ymm0, %ymm4
; AVX512F-NEXT: vpternlogq $226, %zmm3, %zmm7, %zmm4
; AVX512F-NEXT: vpaddb %ymm1, %ymm1, %ymm1
; AVX512F-NEXT: vpblendvb %ymm1, %ymm4, %ymm0, %ymm0
; AVX512F-NEXT: vpsllw $7, %ymm0, %ymm3
; AVX512F-NEXT: vpsrlw $1, %ymm0, %ymm4
; AVX512F-NEXT: vpternlogq $226, %zmm3, %zmm8, %zmm4
; AVX512F-NEXT: vpaddb %ymm1, %ymm1, %ymm1
; AVX512F-NEXT: vpblendvb %ymm1, %ymm4, %ymm0, %ymm0
; AVX512F-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: var_funnnel_v64i8:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm2
; AVX512VL-NEXT: vpsllw $4, %ymm2, %ymm3
; AVX512VL-NEXT: vpsrlw $4, %ymm2, %ymm4
; AVX512VL-NEXT: vpbroadcastq {{.*#+}} ymm5 = [1085102592571150095,1085102592571150095,1085102592571150095,1085102592571150095]
; AVX512VL-NEXT: vpternlogq $226, %ymm3, %ymm5, %ymm4
; AVX512VL-NEXT: vextracti64x4 $1, %zmm1, %ymm3
; AVX512VL-NEXT: vpsllw $5, %ymm3, %ymm3
; AVX512VL-NEXT: vpblendvb %ymm3, %ymm4, %ymm2, %ymm2
; AVX512VL-NEXT: vpsllw $6, %ymm2, %ymm4
; AVX512VL-NEXT: vpsrlw $2, %ymm2, %ymm6
; AVX512VL-NEXT: vpbroadcastq {{.*#+}} ymm7 = [4557430888798830399,4557430888798830399,4557430888798830399,4557430888798830399]
; AVX512VL-NEXT: vpternlogq $226, %ymm4, %ymm7, %ymm6
; AVX512VL-NEXT: vpaddb %ymm3, %ymm3, %ymm3
; AVX512VL-NEXT: vpblendvb %ymm3, %ymm6, %ymm2, %ymm2
; AVX512VL-NEXT: vpsllw $7, %ymm2, %ymm4
; AVX512VL-NEXT: vpsrlw $1, %ymm2, %ymm6
; AVX512VL-NEXT: vpbroadcastq {{.*#+}} ymm8 = [9187201950435737471,9187201950435737471,9187201950435737471,9187201950435737471]
; AVX512VL-NEXT: vpternlogq $226, %ymm4, %ymm8, %ymm6
; AVX512VL-NEXT: vpaddb %ymm3, %ymm3, %ymm3
; AVX512VL-NEXT: vpblendvb %ymm3, %ymm6, %ymm2, %ymm2
; AVX512VL-NEXT: vpsllw $4, %ymm0, %ymm3
; AVX512VL-NEXT: vpsrlw $4, %ymm0, %ymm4
; AVX512VL-NEXT: vpternlogq $226, %ymm3, %ymm5, %ymm4
; AVX512VL-NEXT: vpsllw $5, %ymm1, %ymm1
; AVX512VL-NEXT: vpblendvb %ymm1, %ymm4, %ymm0, %ymm0
; AVX512VL-NEXT: vpsllw $6, %ymm0, %ymm3
; AVX512VL-NEXT: vpsrlw $2, %ymm0, %ymm4
; AVX512VL-NEXT: vpternlogq $226, %ymm3, %ymm7, %ymm4
; AVX512VL-NEXT: vpaddb %ymm1, %ymm1, %ymm1
; AVX512VL-NEXT: vpblendvb %ymm1, %ymm4, %ymm0, %ymm0
; AVX512VL-NEXT: vpsllw $7, %ymm0, %ymm3
; AVX512VL-NEXT: vpsrlw $1, %ymm0, %ymm4
; AVX512VL-NEXT: vpternlogq $226, %ymm3, %ymm8, %ymm4
; AVX512VL-NEXT: vpaddb %ymm1, %ymm1, %ymm1
; AVX512VL-NEXT: vpblendvb %ymm1, %ymm4, %ymm0, %ymm0
; AVX512VL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: var_funnnel_v64i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512BW-NEXT: vpxor %xmm2, %xmm2, %xmm2
; AVX512BW-NEXT: vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
; AVX512BW-NEXT: vpunpckhbw {{.*#+}} zmm4 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
; AVX512BW-NEXT: vpsrlvw %zmm3, %zmm4, %zmm3
; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm4 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
; AVX512BW-NEXT: vpandq %zmm4, %zmm3, %zmm3
; AVX512BW-NEXT: vpunpcklbw {{.*#+}} zmm1 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
; AVX512BW-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
; AVX512BW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpandq %zmm4, %zmm0, %zmm0
; AVX512BW-NEXT: vpackuswb %zmm3, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VLBW-LABEL: var_funnnel_v64i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512VLBW-NEXT: vpxor %xmm2, %xmm2, %xmm2
; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} zmm4 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
; AVX512VLBW-NEXT: vpsrlvw %zmm3, %zmm4, %zmm3
; AVX512VLBW-NEXT: vmovdqa64 {{.*#+}} zmm4 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
; AVX512VLBW-NEXT: vpandq %zmm4, %zmm3, %zmm3
; AVX512VLBW-NEXT: vpunpcklbw {{.*#+}} zmm1 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
; AVX512VLBW-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
; AVX512VLBW-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: vpandq %zmm4, %zmm0, %zmm0
; AVX512VLBW-NEXT: vpackuswb %zmm3, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VBMI2-LABEL: var_funnnel_v64i8:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512VBMI2-NEXT: vpxor %xmm2, %xmm2, %xmm2
; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} zmm4 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
; AVX512VBMI2-NEXT: vpsrlvw %zmm3, %zmm4, %zmm3
; AVX512VBMI2-NEXT: vpunpcklbw {{.*#+}} zmm1 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
; AVX512VBMI2-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
; AVX512VBMI2-NEXT: vpsrlvw %zmm1, %zmm0, %zmm1
; AVX512VBMI2-NEXT: vmovdqa64 {{.*#+}} zmm0 = [0,2,4,6,8,10,12,14,64,66,68,70,72,74,76,78,16,18,20,22,24,26,28,30,80,82,84,86,88,90,92,94,32,34,36,38,40,42,44,46,96,98,100,102,104,106,108,110,48,50,52,54,56,58,60,62,112,114,116,118,120,122,124,126]
; AVX512VBMI2-NEXT: vpermi2b %zmm3, %zmm1, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLVBMI2-LABEL: var_funnnel_v64i8:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpandq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512VLVBMI2-NEXT: vpxor %xmm2, %xmm2, %xmm2
; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} zmm3 = zmm1[8],zmm2[8],zmm1[9],zmm2[9],zmm1[10],zmm2[10],zmm1[11],zmm2[11],zmm1[12],zmm2[12],zmm1[13],zmm2[13],zmm1[14],zmm2[14],zmm1[15],zmm2[15],zmm1[24],zmm2[24],zmm1[25],zmm2[25],zmm1[26],zmm2[26],zmm1[27],zmm2[27],zmm1[28],zmm2[28],zmm1[29],zmm2[29],zmm1[30],zmm2[30],zmm1[31],zmm2[31],zmm1[40],zmm2[40],zmm1[41],zmm2[41],zmm1[42],zmm2[42],zmm1[43],zmm2[43],zmm1[44],zmm2[44],zmm1[45],zmm2[45],zmm1[46],zmm2[46],zmm1[47],zmm2[47],zmm1[56],zmm2[56],zmm1[57],zmm2[57],zmm1[58],zmm2[58],zmm1[59],zmm2[59],zmm1[60],zmm2[60],zmm1[61],zmm2[61],zmm1[62],zmm2[62],zmm1[63],zmm2[63]
; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} zmm4 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
; AVX512VLVBMI2-NEXT: vpsrlvw %zmm3, %zmm4, %zmm3
; AVX512VLVBMI2-NEXT: vpunpcklbw {{.*#+}} zmm1 = zmm1[0],zmm2[0],zmm1[1],zmm2[1],zmm1[2],zmm2[2],zmm1[3],zmm2[3],zmm1[4],zmm2[4],zmm1[5],zmm2[5],zmm1[6],zmm2[6],zmm1[7],zmm2[7],zmm1[16],zmm2[16],zmm1[17],zmm2[17],zmm1[18],zmm2[18],zmm1[19],zmm2[19],zmm1[20],zmm2[20],zmm1[21],zmm2[21],zmm1[22],zmm2[22],zmm1[23],zmm2[23],zmm1[32],zmm2[32],zmm1[33],zmm2[33],zmm1[34],zmm2[34],zmm1[35],zmm2[35],zmm1[36],zmm2[36],zmm1[37],zmm2[37],zmm1[38],zmm2[38],zmm1[39],zmm2[39],zmm1[48],zmm2[48],zmm1[49],zmm2[49],zmm1[50],zmm2[50],zmm1[51],zmm2[51],zmm1[52],zmm2[52],zmm1[53],zmm2[53],zmm1[54],zmm2[54],zmm1[55],zmm2[55]
; AVX512VLVBMI2-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
; AVX512VLVBMI2-NEXT: vpsrlvw %zmm1, %zmm0, %zmm1
; AVX512VLVBMI2-NEXT: vmovdqa64 {{.*#+}} zmm0 = [0,2,4,6,8,10,12,14,64,66,68,70,72,74,76,78,16,18,20,22,24,26,28,30,80,82,84,86,88,90,92,94,32,34,36,38,40,42,44,46,96,98,100,102,104,106,108,110,48,50,52,54,56,58,60,62,112,114,116,118,120,122,124,126]
; AVX512VLVBMI2-NEXT: vpermi2b %zmm3, %zmm1, %zmm0
; AVX512VLVBMI2-NEXT: retq
%res = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %x, <64 x i8> %x, <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> %amt) nounwind {
; AVX512-LABEL: splatvar_funnnel_v8i64:
; AVX512: # %bb.0:
; AVX512-NEXT: vpbroadcastq %xmm1, %zmm1
; AVX512-NEXT: vprorvq %zmm1, %zmm0, %zmm0
; AVX512-NEXT: retq
%splat = shufflevector <8 x i64> %amt, <8 x i64> undef, <8 x i32> zeroinitializer
%res = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %x, <8 x i64> %x, <8 x i64> %splat)
ret <8 x i64> %res
}
define <16 x i32> @splatvar_funnnel_v16i32(<16 x i32> %x, <16 x i32> %amt) nounwind {
; AVX512-LABEL: splatvar_funnnel_v16i32:
; AVX512: # %bb.0:
; AVX512-NEXT: vpbroadcastd %xmm1, %zmm1
; AVX512-NEXT: vprorvd %zmm1, %zmm0, %zmm0
; AVX512-NEXT: retq
%splat = shufflevector <16 x i32> %amt, <16 x i32> undef, <16 x i32> zeroinitializer
%res = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %x, <16 x i32> %x, <16 x i32> %splat)
ret <16 x i32> %res
}
define <32 x i16> @splatvar_funnnel_v32i16(<32 x i16> %x, <32 x i16> %amt) nounwind {
; AVX512F-LABEL: splatvar_funnnel_v32i16:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpbroadcastq {{.*#+}} xmm2 = [15,15]
; AVX512F-NEXT: vpandn %xmm2, %xmm1, %xmm3
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm4
; AVX512F-NEXT: vpsllw $1, %ymm4, %ymm5
; AVX512F-NEXT: vpsllw %xmm3, %ymm5, %ymm5
; AVX512F-NEXT: vpsllw $1, %ymm0, %ymm6
; AVX512F-NEXT: vpsllw %xmm3, %ymm6, %ymm3
; AVX512F-NEXT: vinserti64x4 $1, %ymm5, %zmm3, %zmm3
; AVX512F-NEXT: vpbroadcastw %xmm1, %xmm1
; AVX512F-NEXT: vpand %xmm2, %xmm1, %xmm1
; AVX512F-NEXT: vpsrlw %xmm1, %ymm4, %ymm2
; AVX512F-NEXT: vpsrlw %xmm1, %ymm0, %ymm0
; AVX512F-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
; AVX512F-NEXT: vporq %zmm0, %zmm3, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: splatvar_funnnel_v32i16:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpbroadcastq {{.*#+}} xmm2 = [15,15]
; AVX512VL-NEXT: vpandn %xmm2, %xmm1, %xmm3
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm4
; AVX512VL-NEXT: vpsllw $1, %ymm4, %ymm5
; AVX512VL-NEXT: vpsllw %xmm3, %ymm5, %ymm5
; AVX512VL-NEXT: vpsllw $1, %ymm0, %ymm6
; AVX512VL-NEXT: vpsllw %xmm3, %ymm6, %ymm3
; AVX512VL-NEXT: vinserti64x4 $1, %ymm5, %zmm3, %zmm3
; AVX512VL-NEXT: vpbroadcastw %xmm1, %xmm1
; AVX512VL-NEXT: vpand %xmm2, %xmm1, %xmm1
; AVX512VL-NEXT: vpsrlw %xmm1, %ymm4, %ymm2
; AVX512VL-NEXT: vpsrlw %xmm1, %ymm0, %ymm0
; AVX512VL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
; AVX512VL-NEXT: vporq %zmm0, %zmm3, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: splatvar_funnnel_v32i16:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vmovdqa {{.*#+}} xmm2 = [15,0,0,0]
; AVX512BW-NEXT: vpand %xmm2, %xmm1, %xmm3
; AVX512BW-NEXT: vpsrlw %xmm3, %zmm0, %zmm3
; AVX512BW-NEXT: vpandn %xmm2, %xmm1, %xmm1
; AVX512BW-NEXT: vpsllw $1, %zmm0, %zmm0
; AVX512BW-NEXT: vpsllw %xmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vporq %zmm3, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VLBW-LABEL: splatvar_funnnel_v32i16:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vmovdqa {{.*#+}} xmm2 = [15,0,0,0]
; AVX512VLBW-NEXT: vpand %xmm2, %xmm1, %xmm3
; AVX512VLBW-NEXT: vpsrlw %xmm3, %zmm0, %zmm3
; AVX512VLBW-NEXT: vpandn %xmm2, %xmm1, %xmm1
; AVX512VLBW-NEXT: vpsllw $1, %zmm0, %zmm0
; AVX512VLBW-NEXT: vpsllw %xmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: vporq %zmm3, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VBMI2-LABEL: splatvar_funnnel_v32i16:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpbroadcastw %xmm1, %zmm1
; AVX512VBMI2-NEXT: vpshrdvw %zmm1, %zmm0, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLVBMI2-LABEL: splatvar_funnnel_v32i16:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpbroadcastw %xmm1, %zmm1
; AVX512VLVBMI2-NEXT: vpshrdvw %zmm1, %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: retq
%splat = shufflevector <32 x i16> %amt, <32 x i16> undef, <32 x i32> zeroinitializer
%res = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %x, <32 x i16> %x, <32 x i16> %splat)
ret <32 x i16> %res
}
define <64 x i8> @splatvar_funnnel_v64i8(<64 x i8> %x, <64 x i8> %amt) nounwind {
; AVX512F-LABEL: splatvar_funnnel_v64i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm2
; AVX512F-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
; AVX512F-NEXT: vpsrlw %xmm1, %ymm3, %ymm3
; AVX512F-NEXT: vmovdqa {{.*#+}} ymm4 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
; AVX512F-NEXT: vpand %ymm4, %ymm3, %ymm3
; AVX512F-NEXT: vpunpcklbw {{.*#+}} ymm2 = ymm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
; AVX512F-NEXT: vpsrlw %xmm1, %ymm2, %ymm2
; AVX512F-NEXT: vpand %ymm4, %ymm2, %ymm2
; AVX512F-NEXT: vpackuswb %ymm3, %ymm2, %ymm2
; AVX512F-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
; AVX512F-NEXT: vpsrlw %xmm1, %ymm3, %ymm3
; AVX512F-NEXT: vpand %ymm4, %ymm3, %ymm3
; AVX512F-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
; AVX512F-NEXT: vpsrlw %xmm1, %ymm0, %ymm0
; AVX512F-NEXT: vpand %ymm4, %ymm0, %ymm0
; AVX512F-NEXT: vpackuswb %ymm3, %ymm0, %ymm0
; AVX512F-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: splatvar_funnnel_v64i8:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm2
; AVX512VL-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm2[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
; AVX512VL-NEXT: vpsrlw %xmm1, %ymm3, %ymm3
; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm4 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
; AVX512VL-NEXT: vpand %ymm4, %ymm3, %ymm3
; AVX512VL-NEXT: vpunpcklbw {{.*#+}} ymm2 = ymm2[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
; AVX512VL-NEXT: vpsrlw %xmm1, %ymm2, %ymm2
; AVX512VL-NEXT: vpand %ymm4, %ymm2, %ymm2
; AVX512VL-NEXT: vpackuswb %ymm3, %ymm2, %ymm2
; AVX512VL-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
; AVX512VL-NEXT: vpsrlw %xmm1, %ymm3, %ymm3
; AVX512VL-NEXT: vpand %ymm4, %ymm3, %ymm3
; AVX512VL-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
; AVX512VL-NEXT: vpsrlw %xmm1, %ymm0, %ymm0
; AVX512VL-NEXT: vpand %ymm4, %ymm0, %ymm0
; AVX512VL-NEXT: vpackuswb %ymm3, %ymm0, %ymm0
; AVX512VL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: splatvar_funnnel_v64i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpunpckhbw {{.*#+}} zmm2 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
; AVX512BW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512BW-NEXT: vpsrlw %xmm1, %zmm2, %zmm2
; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
; AVX512BW-NEXT: vpandq %zmm3, %zmm2, %zmm2
; AVX512BW-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
; AVX512BW-NEXT: vpsrlw %xmm1, %zmm0, %zmm0
; AVX512BW-NEXT: vpandq %zmm3, %zmm0, %zmm0
; AVX512BW-NEXT: vpackuswb %zmm2, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VLBW-LABEL: splatvar_funnnel_v64i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} zmm2 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
; AVX512VLBW-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512VLBW-NEXT: vpsrlw %xmm1, %zmm2, %zmm2
; AVX512VLBW-NEXT: vmovdqa64 {{.*#+}} zmm3 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255]
; AVX512VLBW-NEXT: vpandq %zmm3, %zmm2, %zmm2
; AVX512VLBW-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
; AVX512VLBW-NEXT: vpsrlw %xmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: vpandq %zmm3, %zmm0, %zmm0
; AVX512VLBW-NEXT: vpackuswb %zmm2, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VBMI2-LABEL: splatvar_funnnel_v64i8:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} zmm2 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
; AVX512VBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512VBMI2-NEXT: vpsrlw %xmm1, %zmm2, %zmm2
; AVX512VBMI2-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
; AVX512VBMI2-NEXT: vpsrlw %xmm1, %zmm0, %zmm1
; AVX512VBMI2-NEXT: vmovdqa64 {{.*#+}} zmm0 = [0,2,4,6,8,10,12,14,64,66,68,70,72,74,76,78,16,18,20,22,24,26,28,30,80,82,84,86,88,90,92,94,32,34,36,38,40,42,44,46,96,98,100,102,104,106,108,110,48,50,52,54,56,58,60,62,112,114,116,118,120,122,124,126]
; AVX512VBMI2-NEXT: vpermi2b %zmm2, %zmm1, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLVBMI2-LABEL: splatvar_funnnel_v64i8:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} zmm2 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
; AVX512VLVBMI2-NEXT: vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm1
; AVX512VLVBMI2-NEXT: vpsrlw %xmm1, %zmm2, %zmm2
; AVX512VLVBMI2-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
; AVX512VLVBMI2-NEXT: vpsrlw %xmm1, %zmm0, %zmm1
; AVX512VLVBMI2-NEXT: vmovdqa64 {{.*#+}} zmm0 = [0,2,4,6,8,10,12,14,64,66,68,70,72,74,76,78,16,18,20,22,24,26,28,30,80,82,84,86,88,90,92,94,32,34,36,38,40,42,44,46,96,98,100,102,104,106,108,110,48,50,52,54,56,58,60,62,112,114,116,118,120,122,124,126]
; AVX512VLVBMI2-NEXT: vpermi2b %zmm2, %zmm1, %zmm0
; AVX512VLVBMI2-NEXT: retq
%splat = shufflevector <64 x i8> %amt, <64 x i8> undef, <64 x i32> zeroinitializer
%res = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %x, <64 x i8> %x, <64 x i8> %splat)
ret <64 x i8> %res
}
;
; Constant Shifts
;
define <8 x i64> @constant_funnnel_v8i64(<8 x i64> %x) nounwind {
; AVX512-LABEL: constant_funnnel_v8i64:
; AVX512: # %bb.0:
; AVX512-NEXT: vprorvq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512-NEXT: retq
%res = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %x, <8 x i64> %x, <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) nounwind {
; AVX512-LABEL: constant_funnnel_v16i32:
; AVX512: # %bb.0:
; AVX512-NEXT: vprorvd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512-NEXT: retq
%res = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %x, <16 x i32> %x, <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) nounwind {
; AVX512F-LABEL: constant_funnnel_v32i16:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm1
; AVX512F-NEXT: vmovdqa {{.*#+}} ymm2 = [1,32768,16384,8192,4096,2048,1024,512,256,128,64,32,16,8,4,2]
; AVX512F-NEXT: vpmulhuw %ymm2, %ymm1, %ymm3
; AVX512F-NEXT: vpmulhuw %ymm2, %ymm0, %ymm4
; AVX512F-NEXT: vinserti64x4 $1, %ymm3, %zmm4, %zmm3
; AVX512F-NEXT: vpmullw %ymm2, %ymm1, %ymm1
; AVX512F-NEXT: vpmullw %ymm2, %ymm0, %ymm0
; AVX512F-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
; AVX512F-NEXT: vporq %zmm3, %zmm0, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: constant_funnnel_v32i16:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm2 = [1,32768,16384,8192,4096,2048,1024,512,256,128,64,32,16,8,4,2]
; AVX512VL-NEXT: vpmulhuw %ymm2, %ymm1, %ymm3
; AVX512VL-NEXT: vpmulhuw %ymm2, %ymm0, %ymm4
; AVX512VL-NEXT: vinserti64x4 $1, %ymm3, %zmm4, %zmm3
; AVX512VL-NEXT: vpmullw %ymm2, %ymm1, %ymm1
; AVX512VL-NEXT: vpmullw %ymm2, %ymm0, %ymm0
; AVX512VL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
; AVX512VL-NEXT: vporq %zmm3, %zmm0, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: constant_funnnel_v32i16:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
; AVX512BW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512BW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VLBW-LABEL: constant_funnnel_v32i16:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpsrlvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm1
; AVX512VLBW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512VLBW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VBMI2-LABEL: constant_funnnel_v32i16:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpshrdvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLVBMI2-LABEL: constant_funnnel_v32i16:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpshrdvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: retq
%res = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %x, <32 x i16> %x, <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) nounwind {
; AVX512F-LABEL: constant_funnnel_v64i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm1
; AVX512F-NEXT: vpunpckhbw {{.*#+}} ymm2 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
; 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: vpunpcklbw {{.*#+}} ymm1 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
; AVX512F-NEXT: vbroadcasti128 {{.*#+}} ymm4 = [1,128,64,32,16,8,4,2,1,128,64,32,16,8,4,2]
; AVX512F-NEXT: # ymm4 = mem[0,1,0,1]
; AVX512F-NEXT: vpmullw %ymm4, %ymm1, %ymm1
; AVX512F-NEXT: vpsrlw $8, %ymm1, %ymm1
; AVX512F-NEXT: vpackuswb %ymm2, %ymm1, %ymm1
; AVX512F-NEXT: vpunpckhbw {{.*#+}} ymm2 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
; AVX512F-NEXT: vpmullw %ymm3, %ymm2, %ymm2
; AVX512F-NEXT: vpsrlw $8, %ymm2, %ymm2
; AVX512F-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
; AVX512F-NEXT: vpmullw %ymm4, %ymm0, %ymm0
; AVX512F-NEXT: vpsrlw $8, %ymm0, %ymm0
; AVX512F-NEXT: vpackuswb %ymm2, %ymm0, %ymm0
; AVX512F-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: constant_funnnel_v64i8:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm1
; AVX512VL-NEXT: vpunpckhbw {{.*#+}} ymm2 = ymm1[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
; 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: vpunpcklbw {{.*#+}} ymm1 = ymm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
; AVX512VL-NEXT: vbroadcasti128 {{.*#+}} ymm4 = [1,128,64,32,16,8,4,2,1,128,64,32,16,8,4,2]
; AVX512VL-NEXT: # ymm4 = mem[0,1,0,1]
; AVX512VL-NEXT: vpmullw %ymm4, %ymm1, %ymm1
; AVX512VL-NEXT: vpsrlw $8, %ymm1, %ymm1
; AVX512VL-NEXT: vpackuswb %ymm2, %ymm1, %ymm1
; AVX512VL-NEXT: vpunpckhbw {{.*#+}} ymm2 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31]
; AVX512VL-NEXT: vpmullw %ymm3, %ymm2, %ymm2
; AVX512VL-NEXT: vpsrlw $8, %ymm2, %ymm2
; AVX512VL-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23]
; AVX512VL-NEXT: vpmullw %ymm4, %ymm0, %ymm0
; AVX512VL-NEXT: vpsrlw $8, %ymm0, %ymm0
; AVX512VL-NEXT: vpackuswb %ymm2, %ymm0, %ymm0
; AVX512VL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: constant_funnnel_v64i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
; AVX512BW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512BW-NEXT: vpsrlw $8, %zmm1, %zmm1
; AVX512BW-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
; AVX512BW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512BW-NEXT: vpsrlw $8, %zmm0, %zmm0
; AVX512BW-NEXT: vpackuswb %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VLBW-LABEL: constant_funnnel_v64i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
; AVX512VLBW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512VLBW-NEXT: vpsrlw $8, %zmm1, %zmm1
; AVX512VLBW-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
; AVX512VLBW-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512VLBW-NEXT: vpsrlw $8, %zmm0, %zmm0
; AVX512VLBW-NEXT: vpackuswb %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VBMI2-LABEL: constant_funnnel_v64i8:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
; AVX512VBMI2-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512VBMI2-NEXT: vpsrlw $8, %zmm1, %zmm1
; AVX512VBMI2-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
; AVX512VBMI2-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512VBMI2-NEXT: vpsrlw $8, %zmm0, %zmm0
; AVX512VBMI2-NEXT: vpackuswb %zmm1, %zmm0, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLVBMI2-LABEL: constant_funnnel_v64i8:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpunpckhbw {{.*#+}} zmm1 = zmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31,40,40,41,41,42,42,43,43,44,44,45,45,46,46,47,47,56,56,57,57,58,58,59,59,60,60,61,61,62,62,63,63]
; AVX512VLVBMI2-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm1, %zmm1
; AVX512VLVBMI2-NEXT: vpsrlw $8, %zmm1, %zmm1
; AVX512VLVBMI2-NEXT: vpunpcklbw {{.*#+}} zmm0 = zmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23,32,32,33,33,34,34,35,35,36,36,37,37,38,38,39,39,48,48,49,49,50,50,51,51,52,52,53,53,54,54,55,55]
; AVX512VLVBMI2-NEXT: vpsllvw {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: vpsrlw $8, %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: vpackuswb %zmm1, %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: retq
%res = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %x, <64 x i8> %x, <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) nounwind {
; AVX512-LABEL: splatconstant_funnnel_v8i64:
; AVX512: # %bb.0:
; AVX512-NEXT: vprorq $14, %zmm0, %zmm0
; AVX512-NEXT: retq
%res = call <8 x i64> @llvm.fshr.v8i64(<8 x i64> %x, <8 x i64> %x, <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) nounwind {
; AVX512-LABEL: splatconstant_funnnel_v16i32:
; AVX512: # %bb.0:
; AVX512-NEXT: vprord $4, %zmm0, %zmm0
; AVX512-NEXT: retq
%res = call <16 x i32> @llvm.fshr.v16i32(<16 x i32> %x, <16 x i32> %x, <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) nounwind {
; AVX512F-LABEL: splatconstant_funnnel_v32i16:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpsrlw $7, %ymm0, %ymm1
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm2
; AVX512F-NEXT: vpsrlw $7, %ymm2, %ymm3
; AVX512F-NEXT: vinserti64x4 $1, %ymm3, %zmm1, %zmm1
; AVX512F-NEXT: vpsllw $9, %ymm0, %ymm0
; AVX512F-NEXT: vpsllw $9, %ymm2, %ymm2
; AVX512F-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
; AVX512F-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: splatconstant_funnnel_v32i16:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpsrlw $7, %ymm0, %ymm1
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm2
; AVX512VL-NEXT: vpsrlw $7, %ymm2, %ymm3
; AVX512VL-NEXT: vinserti64x4 $1, %ymm3, %zmm1, %zmm1
; AVX512VL-NEXT: vpsllw $9, %ymm0, %ymm0
; AVX512VL-NEXT: vpsllw $9, %ymm2, %ymm2
; AVX512VL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
; AVX512VL-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: splatconstant_funnnel_v32i16:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpsrlw $7, %zmm0, %zmm1
; AVX512BW-NEXT: vpsllw $9, %zmm0, %zmm0
; AVX512BW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VLBW-LABEL: splatconstant_funnnel_v32i16:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpsrlw $7, %zmm0, %zmm1
; AVX512VLBW-NEXT: vpsllw $9, %zmm0, %zmm0
; AVX512VLBW-NEXT: vporq %zmm1, %zmm0, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VBMI2-LABEL: splatconstant_funnnel_v32i16:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpshrdw $7, %zmm0, %zmm0, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLVBMI2-LABEL: splatconstant_funnnel_v32i16:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpshrdw $7, %zmm0, %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: retq
%res = call <32 x i16> @llvm.fshr.v32i16(<32 x i16> %x, <32 x i16> %x, <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) nounwind {
; AVX512F-LABEL: splatconstant_funnnel_v64i8:
; AVX512F: # %bb.0:
; AVX512F-NEXT: vpsllw $4, %ymm0, %ymm1
; AVX512F-NEXT: vextracti64x4 $1, %zmm0, %ymm2
; AVX512F-NEXT: vpsllw $4, %ymm2, %ymm3
; AVX512F-NEXT: vinserti64x4 $1, %ymm3, %zmm1, %zmm1
; AVX512F-NEXT: vpsrlw $4, %ymm0, %ymm0
; AVX512F-NEXT: vpsrlw $4, %ymm2, %ymm2
; AVX512F-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
; AVX512F-NEXT: vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm0
; AVX512F-NEXT: retq
;
; AVX512VL-LABEL: splatconstant_funnnel_v64i8:
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vpsllw $4, %ymm0, %ymm1
; AVX512VL-NEXT: vextracti64x4 $1, %zmm0, %ymm2
; AVX512VL-NEXT: vpsllw $4, %ymm2, %ymm3
; AVX512VL-NEXT: vinserti64x4 $1, %ymm3, %zmm1, %zmm1
; AVX512VL-NEXT: vpsrlw $4, %ymm0, %ymm0
; AVX512VL-NEXT: vpsrlw $4, %ymm2, %ymm2
; AVX512VL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0
; AVX512VL-NEXT: vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm0
; AVX512VL-NEXT: retq
;
; AVX512BW-LABEL: splatconstant_funnnel_v64i8:
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vpsllw $4, %zmm0, %zmm1
; AVX512BW-NEXT: vpsrlw $4, %zmm0, %zmm0
; AVX512BW-NEXT: vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm0
; AVX512BW-NEXT: retq
;
; AVX512VLBW-LABEL: splatconstant_funnnel_v64i8:
; AVX512VLBW: # %bb.0:
; AVX512VLBW-NEXT: vpsllw $4, %zmm0, %zmm1
; AVX512VLBW-NEXT: vpsrlw $4, %zmm0, %zmm0
; AVX512VLBW-NEXT: vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm0
; AVX512VLBW-NEXT: retq
;
; AVX512VBMI2-LABEL: splatconstant_funnnel_v64i8:
; AVX512VBMI2: # %bb.0:
; AVX512VBMI2-NEXT: vpsllw $4, %zmm0, %zmm1
; AVX512VBMI2-NEXT: vpsrlw $4, %zmm0, %zmm0
; AVX512VBMI2-NEXT: vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm0
; AVX512VBMI2-NEXT: retq
;
; AVX512VLVBMI2-LABEL: splatconstant_funnnel_v64i8:
; AVX512VLVBMI2: # %bb.0:
; AVX512VLVBMI2-NEXT: vpsllw $4, %zmm0, %zmm1
; AVX512VLVBMI2-NEXT: vpsrlw $4, %zmm0, %zmm0
; AVX512VLVBMI2-NEXT: vpternlogq $216, {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %zmm1, %zmm0
; AVX512VLVBMI2-NEXT: retq
%res = call <64 x i8> @llvm.fshr.v64i8(<64 x i8> %x, <64 x i8> %x, <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
}