| ; 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 |
| |
| ; |
| ; Variable Rotates |
| ; |
| |
| define <8 x i64> @var_rotate_v8i64(<8 x i64> %a, <8 x i64> %b) nounwind { |
| ; AVX512-LABEL: var_rotate_v8i64: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vprolvq %zmm1, %zmm0, %zmm0 |
| ; AVX512-NEXT: retq |
| %b64 = sub <8 x i64> <i64 64, i64 64, i64 64, i64 64, i64 64, i64 64, i64 64, i64 64>, %b |
| %shl = shl <8 x i64> %a, %b |
| %lshr = lshr <8 x i64> %a, %b64 |
| %or = or <8 x i64> %shl, %lshr |
| ret <8 x i64> %or |
| } |
| |
| define <16 x i32> @var_rotate_v16i32(<16 x i32> %a, <16 x i32> %b) nounwind { |
| ; AVX512-LABEL: var_rotate_v16i32: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vprolvd %zmm1, %zmm0, %zmm0 |
| ; AVX512-NEXT: retq |
| %b32 = sub <16 x i32> <i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32>, %b |
| %shl = shl <16 x i32> %a, %b |
| %lshr = lshr <16 x i32> %a, %b32 |
| %or = or <16 x i32> %shl, %lshr |
| ret <16 x i32> %or |
| } |
| |
| define <32 x i16> @var_rotate_v32i16(<32 x i16> %a, <32 x i16> %b) nounwind { |
| ; AVX512F-LABEL: var_rotate_v32i16: |
| ; AVX512F: # %bb.0: |
| ; 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: 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: vpsllvd %zmm4, %zmm0, %zmm4 |
| ; AVX512F-NEXT: vpmovdw %zmm4, %ymm4 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm5 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16] |
| ; AVX512F-NEXT: vpsubw %ymm2, %ymm5, %ymm2 |
| ; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero |
| ; AVX512F-NEXT: vpsrlvd %zmm2, %zmm0, %zmm0 |
| ; AVX512F-NEXT: vpmovdw %zmm0, %ymm0 |
| ; AVX512F-NEXT: vpor %ymm0, %ymm4, %ymm0 |
| ; AVX512F-NEXT: vpmovzxwd {{.*#+}} zmm2 = 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 {{.*#+}} 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 %zmm2, %zmm1, %zmm2 |
| ; AVX512F-NEXT: vpmovdw %zmm2, %ymm2 |
| ; AVX512F-NEXT: vpsubw %ymm3, %ymm5, %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: vpsrlvd %zmm3, %zmm1, %zmm1 |
| ; AVX512F-NEXT: vpmovdw %zmm1, %ymm1 |
| ; AVX512F-NEXT: vpor %ymm1, %ymm2, %ymm1 |
| ; AVX512F-NEXT: retq |
| ; |
| ; AVX512VL-LABEL: var_rotate_v32i16: |
| ; AVX512VL: # %bb.0: |
| ; 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: 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: vpsllvd %zmm4, %zmm0, %zmm4 |
| ; AVX512VL-NEXT: vpmovdw %zmm4, %ymm4 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm5 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16] |
| ; AVX512VL-NEXT: vpsubw %ymm2, %ymm5, %ymm2 |
| ; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm2 = ymm2[0],zero,ymm2[1],zero,ymm2[2],zero,ymm2[3],zero,ymm2[4],zero,ymm2[5],zero,ymm2[6],zero,ymm2[7],zero,ymm2[8],zero,ymm2[9],zero,ymm2[10],zero,ymm2[11],zero,ymm2[12],zero,ymm2[13],zero,ymm2[14],zero,ymm2[15],zero |
| ; AVX512VL-NEXT: vpsrlvd %zmm2, %zmm0, %zmm0 |
| ; AVX512VL-NEXT: vpmovdw %zmm0, %ymm0 |
| ; AVX512VL-NEXT: vpor %ymm0, %ymm4, %ymm0 |
| ; AVX512VL-NEXT: vpmovzxwd {{.*#+}} zmm2 = 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 {{.*#+}} 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 %zmm2, %zmm1, %zmm2 |
| ; AVX512VL-NEXT: vpmovdw %zmm2, %ymm2 |
| ; AVX512VL-NEXT: vpsubw %ymm3, %ymm5, %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: vpsrlvd %zmm3, %zmm1, %zmm1 |
| ; AVX512VL-NEXT: vpmovdw %zmm1, %ymm1 |
| ; AVX512VL-NEXT: vpor %ymm1, %ymm2, %ymm1 |
| ; AVX512VL-NEXT: retq |
| ; |
| ; AVX512BW-LABEL: var_rotate_v32i16: |
| ; AVX512BW: # %bb.0: |
| ; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm2 = [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, %zmm2, %zmm2 |
| ; AVX512BW-NEXT: vpsllvw %zmm1, %zmm0, %zmm1 |
| ; AVX512BW-NEXT: vpsrlvw %zmm2, %zmm0, %zmm0 |
| ; AVX512BW-NEXT: vporq %zmm0, %zmm1, %zmm0 |
| ; AVX512BW-NEXT: retq |
| ; |
| ; AVX512VLBW-LABEL: var_rotate_v32i16: |
| ; AVX512VLBW: # %bb.0: |
| ; AVX512VLBW-NEXT: vmovdqa64 {{.*#+}} zmm2 = [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, %zmm2, %zmm2 |
| ; AVX512VLBW-NEXT: vpsllvw %zmm1, %zmm0, %zmm1 |
| ; AVX512VLBW-NEXT: vpsrlvw %zmm2, %zmm0, %zmm0 |
| ; AVX512VLBW-NEXT: vporq %zmm0, %zmm1, %zmm0 |
| ; AVX512VLBW-NEXT: retq |
| %b16 = sub <32 x i16> <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>, %b |
| %shl = shl <32 x i16> %a, %b |
| %lshr = lshr <32 x i16> %a, %b16 |
| %or = or <32 x i16> %shl, %lshr |
| ret <32 x i16> %or |
| } |
| |
| define <64 x i8> @var_rotate_v64i8(<64 x i8> %a, <64 x i8> %b) nounwind { |
| ; AVX512F-LABEL: var_rotate_v64i8: |
| ; AVX512F: # %bb.0: |
| ; AVX512F-NEXT: vpsrlw $4, %ymm0, %ymm4 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm5 = [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 %ymm5, %ymm4, %ymm4 |
| ; AVX512F-NEXT: vpsllw $4, %ymm0, %ymm6 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm7 = [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 %ymm7, %ymm6, %ymm6 |
| ; AVX512F-NEXT: vpor %ymm4, %ymm6, %ymm4 |
| ; AVX512F-NEXT: vpsllw $5, %ymm2, %ymm2 |
| ; AVX512F-NEXT: vpblendvb %ymm2, %ymm4, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpsrlw $6, %ymm0, %ymm4 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm6 = [3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3] |
| ; AVX512F-NEXT: vpand %ymm6, %ymm4, %ymm4 |
| ; AVX512F-NEXT: vpsllw $2, %ymm0, %ymm8 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm9 = [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 %ymm9, %ymm8, %ymm8 |
| ; AVX512F-NEXT: vpor %ymm4, %ymm8, %ymm4 |
| ; AVX512F-NEXT: vpaddb %ymm2, %ymm2, %ymm2 |
| ; AVX512F-NEXT: vpblendvb %ymm2, %ymm4, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpsrlw $7, %ymm0, %ymm4 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm8 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1] |
| ; AVX512F-NEXT: vpand %ymm8, %ymm4, %ymm4 |
| ; AVX512F-NEXT: vpaddb %ymm0, %ymm0, %ymm10 |
| ; AVX512F-NEXT: vpor %ymm4, %ymm10, %ymm4 |
| ; AVX512F-NEXT: vpaddb %ymm2, %ymm2, %ymm2 |
| ; AVX512F-NEXT: vpblendvb %ymm2, %ymm4, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpsrlw $4, %ymm1, %ymm2 |
| ; AVX512F-NEXT: vpand %ymm5, %ymm2, %ymm2 |
| ; AVX512F-NEXT: vpsllw $4, %ymm1, %ymm4 |
| ; AVX512F-NEXT: vpand %ymm7, %ymm4, %ymm4 |
| ; AVX512F-NEXT: vpor %ymm2, %ymm4, %ymm2 |
| ; AVX512F-NEXT: vpsllw $5, %ymm3, %ymm3 |
| ; AVX512F-NEXT: vpblendvb %ymm3, %ymm2, %ymm1, %ymm1 |
| ; AVX512F-NEXT: vpsrlw $6, %ymm1, %ymm2 |
| ; AVX512F-NEXT: vpand %ymm6, %ymm2, %ymm2 |
| ; AVX512F-NEXT: vpsllw $2, %ymm1, %ymm4 |
| ; AVX512F-NEXT: vpand %ymm9, %ymm4, %ymm4 |
| ; AVX512F-NEXT: vpor %ymm2, %ymm4, %ymm2 |
| ; AVX512F-NEXT: vpaddb %ymm3, %ymm3, %ymm3 |
| ; AVX512F-NEXT: vpblendvb %ymm3, %ymm2, %ymm1, %ymm1 |
| ; AVX512F-NEXT: vpsrlw $7, %ymm1, %ymm2 |
| ; AVX512F-NEXT: vpand %ymm8, %ymm2, %ymm2 |
| ; AVX512F-NEXT: vpaddb %ymm1, %ymm1, %ymm4 |
| ; AVX512F-NEXT: vpor %ymm2, %ymm4, %ymm2 |
| ; AVX512F-NEXT: vpaddb %ymm3, %ymm3, %ymm3 |
| ; AVX512F-NEXT: vpblendvb %ymm3, %ymm2, %ymm1, %ymm1 |
| ; AVX512F-NEXT: retq |
| ; |
| ; AVX512VL-LABEL: var_rotate_v64i8: |
| ; AVX512VL: # %bb.0: |
| ; AVX512VL-NEXT: vpsrlw $4, %ymm0, %ymm4 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm5 = [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 %ymm5, %ymm4, %ymm4 |
| ; AVX512VL-NEXT: vpsllw $4, %ymm0, %ymm6 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm7 = [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 %ymm7, %ymm6, %ymm6 |
| ; AVX512VL-NEXT: vpor %ymm4, %ymm6, %ymm4 |
| ; AVX512VL-NEXT: vpsllw $5, %ymm2, %ymm2 |
| ; AVX512VL-NEXT: vpblendvb %ymm2, %ymm4, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpsrlw $6, %ymm0, %ymm4 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm6 = [3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3] |
| ; AVX512VL-NEXT: vpand %ymm6, %ymm4, %ymm4 |
| ; AVX512VL-NEXT: vpsllw $2, %ymm0, %ymm8 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm9 = [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 %ymm9, %ymm8, %ymm8 |
| ; AVX512VL-NEXT: vpor %ymm4, %ymm8, %ymm4 |
| ; AVX512VL-NEXT: vpaddb %ymm2, %ymm2, %ymm2 |
| ; AVX512VL-NEXT: vpblendvb %ymm2, %ymm4, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpsrlw $7, %ymm0, %ymm4 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm8 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1] |
| ; AVX512VL-NEXT: vpand %ymm8, %ymm4, %ymm4 |
| ; AVX512VL-NEXT: vpaddb %ymm0, %ymm0, %ymm10 |
| ; AVX512VL-NEXT: vpor %ymm4, %ymm10, %ymm4 |
| ; AVX512VL-NEXT: vpaddb %ymm2, %ymm2, %ymm2 |
| ; AVX512VL-NEXT: vpblendvb %ymm2, %ymm4, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpsrlw $4, %ymm1, %ymm2 |
| ; AVX512VL-NEXT: vpand %ymm5, %ymm2, %ymm2 |
| ; AVX512VL-NEXT: vpsllw $4, %ymm1, %ymm4 |
| ; AVX512VL-NEXT: vpand %ymm7, %ymm4, %ymm4 |
| ; AVX512VL-NEXT: vpor %ymm2, %ymm4, %ymm2 |
| ; AVX512VL-NEXT: vpsllw $5, %ymm3, %ymm3 |
| ; AVX512VL-NEXT: vpblendvb %ymm3, %ymm2, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: vpsrlw $6, %ymm1, %ymm2 |
| ; AVX512VL-NEXT: vpand %ymm6, %ymm2, %ymm2 |
| ; AVX512VL-NEXT: vpsllw $2, %ymm1, %ymm4 |
| ; AVX512VL-NEXT: vpand %ymm9, %ymm4, %ymm4 |
| ; AVX512VL-NEXT: vpor %ymm2, %ymm4, %ymm2 |
| ; AVX512VL-NEXT: vpaddb %ymm3, %ymm3, %ymm3 |
| ; AVX512VL-NEXT: vpblendvb %ymm3, %ymm2, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: vpsrlw $7, %ymm1, %ymm2 |
| ; AVX512VL-NEXT: vpand %ymm8, %ymm2, %ymm2 |
| ; AVX512VL-NEXT: vpaddb %ymm1, %ymm1, %ymm4 |
| ; AVX512VL-NEXT: vpor %ymm2, %ymm4, %ymm2 |
| ; AVX512VL-NEXT: vpaddb %ymm3, %ymm3, %ymm3 |
| ; AVX512VL-NEXT: vpblendvb %ymm3, %ymm2, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: retq |
| ; |
| ; AVX512BW-LABEL: var_rotate_v64i8: |
| ; AVX512BW: # %bb.0: |
| ; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8] |
| ; AVX512BW-NEXT: vpsubb %zmm1, %zmm2, %zmm2 |
| ; AVX512BW-NEXT: vpsllw $4, %zmm0, %zmm3 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm3, %zmm3 |
| ; AVX512BW-NEXT: vpsllw $5, %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512BW-NEXT: vpblendmb %zmm3, %zmm0, %zmm3 {%k1} |
| ; AVX512BW-NEXT: vpsllw $2, %zmm3, %zmm4 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm4, %zmm4 |
| ; AVX512BW-NEXT: vpaddb %zmm1, %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512BW-NEXT: vmovdqu8 %zmm4, %zmm3 {%k1} |
| ; AVX512BW-NEXT: vpaddb %zmm1, %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512BW-NEXT: vpaddb %zmm3, %zmm3, %zmm3 {%k1} |
| ; AVX512BW-NEXT: vpsllw $5, %zmm2, %zmm1 |
| ; AVX512BW-NEXT: vpaddb %zmm1, %zmm1, %zmm2 |
| ; AVX512BW-NEXT: vpmovb2m %zmm2, %k1 |
| ; AVX512BW-NEXT: vpmovb2m %zmm1, %k2 |
| ; AVX512BW-NEXT: vpsrlw $4, %zmm0, %zmm1 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vmovdqu8 %zmm1, %zmm0 {%k2} |
| ; AVX512BW-NEXT: vpsrlw $2, %zmm0, %zmm1 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vmovdqu8 %zmm1, %zmm0 {%k1} |
| ; AVX512BW-NEXT: vpsrlw $1, %zmm0, %zmm1 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vpaddb %zmm2, %zmm2, %zmm2 |
| ; AVX512BW-NEXT: vpmovb2m %zmm2, %k1 |
| ; AVX512BW-NEXT: vmovdqu8 %zmm1, %zmm0 {%k1} |
| ; AVX512BW-NEXT: vporq %zmm0, %zmm3, %zmm0 |
| ; AVX512BW-NEXT: retq |
| ; |
| ; AVX512VLBW-LABEL: var_rotate_v64i8: |
| ; AVX512VLBW: # %bb.0: |
| ; AVX512VLBW-NEXT: vmovdqa64 {{.*#+}} zmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8] |
| ; AVX512VLBW-NEXT: vpsubb %zmm1, %zmm2, %zmm2 |
| ; AVX512VLBW-NEXT: vpsllw $4, %zmm0, %zmm3 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm3, %zmm3 |
| ; AVX512VLBW-NEXT: vpsllw $5, %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512VLBW-NEXT: vpblendmb %zmm3, %zmm0, %zmm3 {%k1} |
| ; AVX512VLBW-NEXT: vpsllw $2, %zmm3, %zmm4 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm4, %zmm4 |
| ; AVX512VLBW-NEXT: vpaddb %zmm1, %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512VLBW-NEXT: vmovdqu8 %zmm4, %zmm3 {%k1} |
| ; AVX512VLBW-NEXT: vpaddb %zmm1, %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512VLBW-NEXT: vpaddb %zmm3, %zmm3, %zmm3 {%k1} |
| ; AVX512VLBW-NEXT: vpsllw $5, %zmm2, %zmm1 |
| ; AVX512VLBW-NEXT: vpaddb %zmm1, %zmm1, %zmm2 |
| ; AVX512VLBW-NEXT: vpmovb2m %zmm2, %k1 |
| ; AVX512VLBW-NEXT: vpmovb2m %zmm1, %k2 |
| ; AVX512VLBW-NEXT: vpsrlw $4, %zmm0, %zmm1 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vmovdqu8 %zmm1, %zmm0 {%k2} |
| ; AVX512VLBW-NEXT: vpsrlw $2, %zmm0, %zmm1 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vmovdqu8 %zmm1, %zmm0 {%k1} |
| ; AVX512VLBW-NEXT: vpsrlw $1, %zmm0, %zmm1 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vpaddb %zmm2, %zmm2, %zmm2 |
| ; AVX512VLBW-NEXT: vpmovb2m %zmm2, %k1 |
| ; AVX512VLBW-NEXT: vmovdqu8 %zmm1, %zmm0 {%k1} |
| ; AVX512VLBW-NEXT: vporq %zmm0, %zmm3, %zmm0 |
| ; AVX512VLBW-NEXT: retq |
| %b8 = sub <64 x i8> <i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8>, %b |
| %shl = shl <64 x i8> %a, %b |
| %lshr = lshr <64 x i8> %a, %b8 |
| %or = or <64 x i8> %shl, %lshr |
| ret <64 x i8> %or |
| } |
| |
| ; |
| ; Uniform Variable Rotates |
| ; |
| |
| define <8 x i64> @splatvar_rotate_v8i64(<8 x i64> %a, <8 x i64> %b) nounwind { |
| ; AVX512-LABEL: splatvar_rotate_v8i64: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vpbroadcastq %xmm1, %zmm1 |
| ; AVX512-NEXT: vprolvq %zmm1, %zmm0, %zmm0 |
| ; AVX512-NEXT: retq |
| %splat = shufflevector <8 x i64> %b, <8 x i64> undef, <8 x i32> zeroinitializer |
| %splat64 = sub <8 x i64> <i64 64, i64 64, i64 64, i64 64, i64 64, i64 64, i64 64, i64 64>, %splat |
| %shl = shl <8 x i64> %a, %splat |
| %lshr = lshr <8 x i64> %a, %splat64 |
| %or = or <8 x i64> %shl, %lshr |
| ret <8 x i64> %or |
| } |
| |
| define <16 x i32> @splatvar_rotate_v16i32(<16 x i32> %a, <16 x i32> %b) nounwind { |
| ; AVX512-LABEL: splatvar_rotate_v16i32: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vpbroadcastd %xmm1, %zmm1 |
| ; AVX512-NEXT: vprolvd %zmm1, %zmm0, %zmm0 |
| ; AVX512-NEXT: retq |
| %splat = shufflevector <16 x i32> %b, <16 x i32> undef, <16 x i32> zeroinitializer |
| %splat32 = sub <16 x i32> <i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32, i32 32>, %splat |
| %shl = shl <16 x i32> %a, %splat |
| %lshr = lshr <16 x i32> %a, %splat32 |
| %or = or <16 x i32> %shl, %lshr |
| ret <16 x i32> %or |
| } |
| |
| define <32 x i16> @splatvar_rotate_v32i16(<32 x i16> %a, <32 x i16> %b) nounwind { |
| ; AVX512F-LABEL: splatvar_rotate_v32i16: |
| ; AVX512F: # %bb.0: |
| ; AVX512F-NEXT: vpbroadcastw %xmm2, %ymm3 |
| ; AVX512F-NEXT: vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero |
| ; AVX512F-NEXT: vpsllw %xmm2, %ymm0, %ymm4 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm5 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16] |
| ; AVX512F-NEXT: vpsubw %ymm3, %ymm5, %ymm3 |
| ; AVX512F-NEXT: vpmovzxwq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero |
| ; AVX512F-NEXT: vpsrlw %xmm3, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpor %ymm0, %ymm4, %ymm0 |
| ; AVX512F-NEXT: vpsllw %xmm2, %ymm1, %ymm2 |
| ; AVX512F-NEXT: vpsrlw %xmm3, %ymm1, %ymm1 |
| ; AVX512F-NEXT: vpor %ymm1, %ymm2, %ymm1 |
| ; AVX512F-NEXT: retq |
| ; |
| ; AVX512VL-LABEL: splatvar_rotate_v32i16: |
| ; AVX512VL: # %bb.0: |
| ; AVX512VL-NEXT: vpbroadcastw %xmm2, %ymm3 |
| ; AVX512VL-NEXT: vpmovzxwq {{.*#+}} xmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero |
| ; AVX512VL-NEXT: vpsllw %xmm2, %ymm0, %ymm4 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm5 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16] |
| ; AVX512VL-NEXT: vpsubw %ymm3, %ymm5, %ymm3 |
| ; AVX512VL-NEXT: vpmovzxwq {{.*#+}} xmm3 = xmm3[0],zero,zero,zero,xmm3[1],zero,zero,zero |
| ; AVX512VL-NEXT: vpsrlw %xmm3, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpor %ymm0, %ymm4, %ymm0 |
| ; AVX512VL-NEXT: vpsllw %xmm2, %ymm1, %ymm2 |
| ; AVX512VL-NEXT: vpsrlw %xmm3, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: vpor %ymm1, %ymm2, %ymm1 |
| ; AVX512VL-NEXT: retq |
| ; |
| ; AVX512BW-LABEL: splatvar_rotate_v32i16: |
| ; AVX512BW: # %bb.0: |
| ; AVX512BW-NEXT: vpbroadcastw %xmm1, %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 %zmm2, %zmm3, %zmm2 |
| ; AVX512BW-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero |
| ; AVX512BW-NEXT: vpsllw %xmm1, %zmm0, %zmm1 |
| ; AVX512BW-NEXT: vpsrlvw %zmm2, %zmm0, %zmm0 |
| ; AVX512BW-NEXT: vporq %zmm0, %zmm1, %zmm0 |
| ; AVX512BW-NEXT: retq |
| ; |
| ; AVX512VLBW-LABEL: splatvar_rotate_v32i16: |
| ; AVX512VLBW: # %bb.0: |
| ; AVX512VLBW-NEXT: vpbroadcastw %xmm1, %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 %zmm2, %zmm3, %zmm2 |
| ; AVX512VLBW-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero |
| ; AVX512VLBW-NEXT: vpsllw %xmm1, %zmm0, %zmm1 |
| ; AVX512VLBW-NEXT: vpsrlvw %zmm2, %zmm0, %zmm0 |
| ; AVX512VLBW-NEXT: vporq %zmm0, %zmm1, %zmm0 |
| ; AVX512VLBW-NEXT: retq |
| %splat = shufflevector <32 x i16> %b, <32 x i16> undef, <32 x i32> zeroinitializer |
| %splat16 = sub <32 x i16> <i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16, i16 16>, %splat |
| %shl = shl <32 x i16> %a, %splat |
| %lshr = lshr <32 x i16> %a, %splat16 |
| %or = or <32 x i16> %shl, %lshr |
| ret <32 x i16> %or |
| } |
| |
| define <64 x i8> @splatvar_rotate_v64i8(<64 x i8> %a, <64 x i8> %b) nounwind { |
| ; AVX512F-LABEL: splatvar_rotate_v64i8: |
| ; AVX512F: # %bb.0: |
| ; AVX512F-NEXT: vpbroadcastb %xmm2, %ymm2 |
| ; AVX512F-NEXT: vpsrlw $4, %ymm0, %ymm3 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm4 = [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 %ymm4, %ymm3, %ymm3 |
| ; AVX512F-NEXT: vpsllw $4, %ymm0, %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: vpor %ymm3, %ymm5, %ymm3 |
| ; AVX512F-NEXT: vpsllw $5, %ymm2, %ymm2 |
| ; AVX512F-NEXT: vpblendvb %ymm2, %ymm3, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpsrlw $6, %ymm0, %ymm3 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm5 = [3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3] |
| ; AVX512F-NEXT: vpand %ymm5, %ymm3, %ymm3 |
| ; AVX512F-NEXT: vpsllw $2, %ymm0, %ymm7 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm8 = [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 %ymm8, %ymm7, %ymm7 |
| ; AVX512F-NEXT: vpor %ymm3, %ymm7, %ymm3 |
| ; AVX512F-NEXT: vpaddb %ymm2, %ymm2, %ymm7 |
| ; AVX512F-NEXT: vpblendvb %ymm7, %ymm3, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpsrlw $7, %ymm0, %ymm3 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm9 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1] |
| ; AVX512F-NEXT: vpand %ymm9, %ymm3, %ymm3 |
| ; AVX512F-NEXT: vpaddb %ymm0, %ymm0, %ymm10 |
| ; AVX512F-NEXT: vpor %ymm3, %ymm10, %ymm3 |
| ; AVX512F-NEXT: vpaddb %ymm7, %ymm7, %ymm10 |
| ; AVX512F-NEXT: vpblendvb %ymm10, %ymm3, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpsrlw $4, %ymm1, %ymm3 |
| ; AVX512F-NEXT: vpand %ymm4, %ymm3, %ymm3 |
| ; AVX512F-NEXT: vpsllw $4, %ymm1, %ymm4 |
| ; AVX512F-NEXT: vpand %ymm6, %ymm4, %ymm4 |
| ; AVX512F-NEXT: vpor %ymm3, %ymm4, %ymm3 |
| ; AVX512F-NEXT: vpblendvb %ymm2, %ymm3, %ymm1, %ymm1 |
| ; AVX512F-NEXT: vpsrlw $6, %ymm1, %ymm2 |
| ; AVX512F-NEXT: vpand %ymm5, %ymm2, %ymm2 |
| ; AVX512F-NEXT: vpsllw $2, %ymm1, %ymm3 |
| ; AVX512F-NEXT: vpand %ymm8, %ymm3, %ymm3 |
| ; AVX512F-NEXT: vpor %ymm2, %ymm3, %ymm2 |
| ; AVX512F-NEXT: vpblendvb %ymm7, %ymm2, %ymm1, %ymm1 |
| ; AVX512F-NEXT: vpsrlw $7, %ymm1, %ymm2 |
| ; AVX512F-NEXT: vpand %ymm9, %ymm2, %ymm2 |
| ; AVX512F-NEXT: vpaddb %ymm1, %ymm1, %ymm3 |
| ; AVX512F-NEXT: vpor %ymm2, %ymm3, %ymm2 |
| ; AVX512F-NEXT: vpblendvb %ymm10, %ymm2, %ymm1, %ymm1 |
| ; AVX512F-NEXT: retq |
| ; |
| ; AVX512VL-LABEL: splatvar_rotate_v64i8: |
| ; AVX512VL: # %bb.0: |
| ; AVX512VL-NEXT: vpbroadcastb %xmm2, %ymm2 |
| ; AVX512VL-NEXT: vpsrlw $4, %ymm0, %ymm3 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm4 = [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 %ymm4, %ymm3, %ymm3 |
| ; AVX512VL-NEXT: vpsllw $4, %ymm0, %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: vpor %ymm3, %ymm5, %ymm3 |
| ; AVX512VL-NEXT: vpsllw $5, %ymm2, %ymm2 |
| ; AVX512VL-NEXT: vpblendvb %ymm2, %ymm3, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpsrlw $6, %ymm0, %ymm3 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm5 = [3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3] |
| ; AVX512VL-NEXT: vpand %ymm5, %ymm3, %ymm3 |
| ; AVX512VL-NEXT: vpsllw $2, %ymm0, %ymm7 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm8 = [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 %ymm8, %ymm7, %ymm7 |
| ; AVX512VL-NEXT: vpor %ymm3, %ymm7, %ymm3 |
| ; AVX512VL-NEXT: vpaddb %ymm2, %ymm2, %ymm7 |
| ; AVX512VL-NEXT: vpblendvb %ymm7, %ymm3, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpsrlw $7, %ymm0, %ymm3 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm9 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1] |
| ; AVX512VL-NEXT: vpand %ymm9, %ymm3, %ymm3 |
| ; AVX512VL-NEXT: vpaddb %ymm0, %ymm0, %ymm10 |
| ; AVX512VL-NEXT: vpor %ymm3, %ymm10, %ymm3 |
| ; AVX512VL-NEXT: vpaddb %ymm7, %ymm7, %ymm10 |
| ; AVX512VL-NEXT: vpblendvb %ymm10, %ymm3, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpsrlw $4, %ymm1, %ymm3 |
| ; AVX512VL-NEXT: vpand %ymm4, %ymm3, %ymm3 |
| ; AVX512VL-NEXT: vpsllw $4, %ymm1, %ymm4 |
| ; AVX512VL-NEXT: vpand %ymm6, %ymm4, %ymm4 |
| ; AVX512VL-NEXT: vpor %ymm3, %ymm4, %ymm3 |
| ; AVX512VL-NEXT: vpblendvb %ymm2, %ymm3, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: vpsrlw $6, %ymm1, %ymm2 |
| ; AVX512VL-NEXT: vpand %ymm5, %ymm2, %ymm2 |
| ; AVX512VL-NEXT: vpsllw $2, %ymm1, %ymm3 |
| ; AVX512VL-NEXT: vpand %ymm8, %ymm3, %ymm3 |
| ; AVX512VL-NEXT: vpor %ymm2, %ymm3, %ymm2 |
| ; AVX512VL-NEXT: vpblendvb %ymm7, %ymm2, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: vpsrlw $7, %ymm1, %ymm2 |
| ; AVX512VL-NEXT: vpand %ymm9, %ymm2, %ymm2 |
| ; AVX512VL-NEXT: vpaddb %ymm1, %ymm1, %ymm3 |
| ; AVX512VL-NEXT: vpor %ymm2, %ymm3, %ymm2 |
| ; AVX512VL-NEXT: vpblendvb %ymm10, %ymm2, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: retq |
| ; |
| ; AVX512BW-LABEL: splatvar_rotate_v64i8: |
| ; AVX512BW: # %bb.0: |
| ; AVX512BW-NEXT: vpbroadcastb %xmm1, %zmm1 |
| ; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8] |
| ; AVX512BW-NEXT: vpsubb %zmm1, %zmm2, %zmm2 |
| ; AVX512BW-NEXT: vpsllw $4, %zmm0, %zmm3 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm3, %zmm3 |
| ; AVX512BW-NEXT: vpsllw $5, %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512BW-NEXT: vpblendmb %zmm3, %zmm0, %zmm3 {%k1} |
| ; AVX512BW-NEXT: vpsllw $2, %zmm3, %zmm4 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm4, %zmm4 |
| ; AVX512BW-NEXT: vpaddb %zmm1, %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512BW-NEXT: vmovdqu8 %zmm4, %zmm3 {%k1} |
| ; AVX512BW-NEXT: vpaddb %zmm1, %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512BW-NEXT: vpaddb %zmm3, %zmm3, %zmm3 {%k1} |
| ; AVX512BW-NEXT: vpsllw $5, %zmm2, %zmm1 |
| ; AVX512BW-NEXT: vpaddb %zmm1, %zmm1, %zmm2 |
| ; AVX512BW-NEXT: vpmovb2m %zmm2, %k1 |
| ; AVX512BW-NEXT: vpmovb2m %zmm1, %k2 |
| ; AVX512BW-NEXT: vpsrlw $4, %zmm0, %zmm1 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vmovdqu8 %zmm1, %zmm0 {%k2} |
| ; AVX512BW-NEXT: vpsrlw $2, %zmm0, %zmm1 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vmovdqu8 %zmm1, %zmm0 {%k1} |
| ; AVX512BW-NEXT: vpsrlw $1, %zmm0, %zmm1 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vpaddb %zmm2, %zmm2, %zmm2 |
| ; AVX512BW-NEXT: vpmovb2m %zmm2, %k1 |
| ; AVX512BW-NEXT: vmovdqu8 %zmm1, %zmm0 {%k1} |
| ; AVX512BW-NEXT: vporq %zmm0, %zmm3, %zmm0 |
| ; AVX512BW-NEXT: retq |
| ; |
| ; AVX512VLBW-LABEL: splatvar_rotate_v64i8: |
| ; AVX512VLBW: # %bb.0: |
| ; AVX512VLBW-NEXT: vpbroadcastb %xmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vmovdqa64 {{.*#+}} zmm2 = [8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8] |
| ; AVX512VLBW-NEXT: vpsubb %zmm1, %zmm2, %zmm2 |
| ; AVX512VLBW-NEXT: vpsllw $4, %zmm0, %zmm3 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm3, %zmm3 |
| ; AVX512VLBW-NEXT: vpsllw $5, %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512VLBW-NEXT: vpblendmb %zmm3, %zmm0, %zmm3 {%k1} |
| ; AVX512VLBW-NEXT: vpsllw $2, %zmm3, %zmm4 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm4, %zmm4 |
| ; AVX512VLBW-NEXT: vpaddb %zmm1, %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512VLBW-NEXT: vmovdqu8 %zmm4, %zmm3 {%k1} |
| ; AVX512VLBW-NEXT: vpaddb %zmm1, %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512VLBW-NEXT: vpaddb %zmm3, %zmm3, %zmm3 {%k1} |
| ; AVX512VLBW-NEXT: vpsllw $5, %zmm2, %zmm1 |
| ; AVX512VLBW-NEXT: vpaddb %zmm1, %zmm1, %zmm2 |
| ; AVX512VLBW-NEXT: vpmovb2m %zmm2, %k1 |
| ; AVX512VLBW-NEXT: vpmovb2m %zmm1, %k2 |
| ; AVX512VLBW-NEXT: vpsrlw $4, %zmm0, %zmm1 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vmovdqu8 %zmm1, %zmm0 {%k2} |
| ; AVX512VLBW-NEXT: vpsrlw $2, %zmm0, %zmm1 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vmovdqu8 %zmm1, %zmm0 {%k1} |
| ; AVX512VLBW-NEXT: vpsrlw $1, %zmm0, %zmm1 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vpaddb %zmm2, %zmm2, %zmm2 |
| ; AVX512VLBW-NEXT: vpmovb2m %zmm2, %k1 |
| ; AVX512VLBW-NEXT: vmovdqu8 %zmm1, %zmm0 {%k1} |
| ; AVX512VLBW-NEXT: vporq %zmm0, %zmm3, %zmm0 |
| ; AVX512VLBW-NEXT: retq |
| %splat = shufflevector <64 x i8> %b, <64 x i8> undef, <64 x i32> zeroinitializer |
| %splat8 = sub <64 x i8> <i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8>, %splat |
| %shl = shl <64 x i8> %a, %splat |
| %lshr = lshr <64 x i8> %a, %splat8 |
| %or = or <64 x i8> %shl, %lshr |
| ret <64 x i8> %or |
| } |
| |
| ; |
| ; Constant Rotates |
| ; |
| |
| define <8 x i64> @constant_rotate_v8i64(<8 x i64> %a) nounwind { |
| ; AVX512-LABEL: constant_rotate_v8i64: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vprolvq {{.*}}(%rip), %zmm0, %zmm0 |
| ; AVX512-NEXT: retq |
| %shl = shl <8 x i64> %a, <i64 4, i64 14, i64 50, i64 60, i64 4, i64 14, i64 50, i64 60> |
| %lshr = lshr <8 x i64> %a, <i64 60, i64 50, i64 14, i64 4, i64 60, i64 50, i64 14, i64 4> |
| %or = or <8 x i64> %shl, %lshr |
| ret <8 x i64> %or |
| } |
| |
| define <16 x i32> @constant_rotate_v16i32(<16 x i32> %a) nounwind { |
| ; AVX512-LABEL: constant_rotate_v16i32: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vprolvd {{.*}}(%rip), %zmm0, %zmm0 |
| ; AVX512-NEXT: retq |
| %shl = shl <16 x i32> %a, <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> |
| %lshr = lshr <16 x i32> %a, <i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21, i32 28, i32 27, i32 26, i32 25, i32 24, i32 23, i32 22, i32 21> |
| %or = or <16 x i32> %shl, %lshr |
| ret <16 x i32> %or |
| } |
| |
| define <32 x i16> @constant_rotate_v32i16(<32 x i16> %a) nounwind { |
| ; AVX512F-LABEL: constant_rotate_v32i16: |
| ; AVX512F: # %bb.0: |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm2 = [1,2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768] |
| ; AVX512F-NEXT: vpmulhuw %ymm2, %ymm0, %ymm3 |
| ; AVX512F-NEXT: vpmullw %ymm2, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpor %ymm3, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpmulhuw %ymm2, %ymm1, %ymm3 |
| ; AVX512F-NEXT: vpmullw %ymm2, %ymm1, %ymm1 |
| ; AVX512F-NEXT: vpor %ymm3, %ymm1, %ymm1 |
| ; AVX512F-NEXT: retq |
| ; |
| ; AVX512VL-LABEL: constant_rotate_v32i16: |
| ; AVX512VL: # %bb.0: |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm2 = [1,2,4,8,16,32,64,128,256,512,1024,2048,4096,8192,16384,32768] |
| ; AVX512VL-NEXT: vpmulhuw %ymm2, %ymm0, %ymm3 |
| ; AVX512VL-NEXT: vpmullw %ymm2, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpor %ymm3, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpmulhuw %ymm2, %ymm1, %ymm3 |
| ; AVX512VL-NEXT: vpmullw %ymm2, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: vpor %ymm3, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: retq |
| ; |
| ; AVX512BW-LABEL: constant_rotate_v32i16: |
| ; AVX512BW: # %bb.0: |
| ; AVX512BW-NEXT: vpsllvw {{.*}}(%rip), %zmm0, %zmm1 |
| ; AVX512BW-NEXT: vpsrlvw {{.*}}(%rip), %zmm0, %zmm0 |
| ; AVX512BW-NEXT: vporq %zmm0, %zmm1, %zmm0 |
| ; AVX512BW-NEXT: retq |
| ; |
| ; AVX512VLBW-LABEL: constant_rotate_v32i16: |
| ; AVX512VLBW: # %bb.0: |
| ; AVX512VLBW-NEXT: vpsllvw {{.*}}(%rip), %zmm0, %zmm1 |
| ; AVX512VLBW-NEXT: vpsrlvw {{.*}}(%rip), %zmm0, %zmm0 |
| ; AVX512VLBW-NEXT: vporq %zmm0, %zmm1, %zmm0 |
| ; AVX512VLBW-NEXT: retq |
| %shl = shl <32 x i16> %a, <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> |
| %lshr = lshr <32 x i16> %a, <i16 16, i16 15, i16 14, i16 13, i16 12, i16 11, i16 10, i16 9, i16 8, i16 7, i16 6, i16 5, i16 4, i16 3, i16 2, i16 1, i16 16, i16 15, i16 14, i16 13, i16 12, i16 11, i16 10, i16 9, i16 8, i16 7, i16 6, i16 5, i16 4, i16 3, i16 2, i16 1> |
| %or = or <32 x i16> %shl, %lshr |
| ret <32 x i16> %or |
| } |
| |
| define <64 x i8> @constant_rotate_v64i8(<64 x i8> %a) nounwind { |
| ; AVX512F-LABEL: constant_rotate_v64i8: |
| ; AVX512F: # %bb.0: |
| ; AVX512F-NEXT: vpsrlw $4, %ymm0, %ymm2 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm3 = [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 %ymm3, %ymm2, %ymm2 |
| ; AVX512F-NEXT: vpsllw $4, %ymm0, %ymm4 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm5 = [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 %ymm5, %ymm4, %ymm4 |
| ; AVX512F-NEXT: vpor %ymm2, %ymm4, %ymm2 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm4 = [8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256] |
| ; AVX512F-NEXT: vpblendvb %ymm4, %ymm2, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpsrlw $6, %ymm0, %ymm2 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm6 = [3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3] |
| ; AVX512F-NEXT: vpand %ymm6, %ymm2, %ymm2 |
| ; AVX512F-NEXT: vpsllw $2, %ymm0, %ymm7 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm8 = [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 %ymm8, %ymm7, %ymm7 |
| ; AVX512F-NEXT: vpor %ymm2, %ymm7, %ymm2 |
| ; AVX512F-NEXT: vpaddb %ymm4, %ymm4, %ymm7 |
| ; AVX512F-NEXT: vpblendvb %ymm7, %ymm2, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpsrlw $7, %ymm0, %ymm2 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm9 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1] |
| ; AVX512F-NEXT: vpand %ymm9, %ymm2, %ymm2 |
| ; AVX512F-NEXT: vpaddb %ymm0, %ymm0, %ymm10 |
| ; AVX512F-NEXT: vpor %ymm2, %ymm10, %ymm2 |
| ; AVX512F-NEXT: vpaddb %ymm7, %ymm7, %ymm10 |
| ; AVX512F-NEXT: vpblendvb %ymm10, %ymm2, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpsrlw $4, %ymm1, %ymm2 |
| ; AVX512F-NEXT: vpand %ymm3, %ymm2, %ymm2 |
| ; AVX512F-NEXT: vpsllw $4, %ymm1, %ymm3 |
| ; AVX512F-NEXT: vpand %ymm5, %ymm3, %ymm3 |
| ; AVX512F-NEXT: vpor %ymm2, %ymm3, %ymm2 |
| ; AVX512F-NEXT: vpblendvb %ymm4, %ymm2, %ymm1, %ymm1 |
| ; AVX512F-NEXT: vpsrlw $6, %ymm1, %ymm2 |
| ; AVX512F-NEXT: vpand %ymm6, %ymm2, %ymm2 |
| ; AVX512F-NEXT: vpsllw $2, %ymm1, %ymm3 |
| ; AVX512F-NEXT: vpand %ymm8, %ymm3, %ymm3 |
| ; AVX512F-NEXT: vpor %ymm2, %ymm3, %ymm2 |
| ; AVX512F-NEXT: vpblendvb %ymm7, %ymm2, %ymm1, %ymm1 |
| ; AVX512F-NEXT: vpsrlw $7, %ymm1, %ymm2 |
| ; AVX512F-NEXT: vpand %ymm9, %ymm2, %ymm2 |
| ; AVX512F-NEXT: vpaddb %ymm1, %ymm1, %ymm3 |
| ; AVX512F-NEXT: vpor %ymm2, %ymm3, %ymm2 |
| ; AVX512F-NEXT: vpblendvb %ymm10, %ymm2, %ymm1, %ymm1 |
| ; AVX512F-NEXT: retq |
| ; |
| ; AVX512VL-LABEL: constant_rotate_v64i8: |
| ; AVX512VL: # %bb.0: |
| ; AVX512VL-NEXT: vpsrlw $4, %ymm0, %ymm2 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm3 = [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 %ymm3, %ymm2, %ymm2 |
| ; AVX512VL-NEXT: vpsllw $4, %ymm0, %ymm4 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm5 = [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 %ymm5, %ymm4, %ymm4 |
| ; AVX512VL-NEXT: vpor %ymm2, %ymm4, %ymm2 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm4 = [8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256] |
| ; AVX512VL-NEXT: vpblendvb %ymm4, %ymm2, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpsrlw $6, %ymm0, %ymm2 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm6 = [3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3] |
| ; AVX512VL-NEXT: vpand %ymm6, %ymm2, %ymm2 |
| ; AVX512VL-NEXT: vpsllw $2, %ymm0, %ymm7 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm8 = [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 %ymm8, %ymm7, %ymm7 |
| ; AVX512VL-NEXT: vpor %ymm2, %ymm7, %ymm2 |
| ; AVX512VL-NEXT: vpaddb %ymm4, %ymm4, %ymm7 |
| ; AVX512VL-NEXT: vpblendvb %ymm7, %ymm2, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpsrlw $7, %ymm0, %ymm2 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm9 = [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1] |
| ; AVX512VL-NEXT: vpand %ymm9, %ymm2, %ymm2 |
| ; AVX512VL-NEXT: vpaddb %ymm0, %ymm0, %ymm10 |
| ; AVX512VL-NEXT: vpor %ymm2, %ymm10, %ymm2 |
| ; AVX512VL-NEXT: vpaddb %ymm7, %ymm7, %ymm10 |
| ; AVX512VL-NEXT: vpblendvb %ymm10, %ymm2, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpsrlw $4, %ymm1, %ymm2 |
| ; AVX512VL-NEXT: vpand %ymm3, %ymm2, %ymm2 |
| ; AVX512VL-NEXT: vpsllw $4, %ymm1, %ymm3 |
| ; AVX512VL-NEXT: vpand %ymm5, %ymm3, %ymm3 |
| ; AVX512VL-NEXT: vpor %ymm2, %ymm3, %ymm2 |
| ; AVX512VL-NEXT: vpblendvb %ymm4, %ymm2, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: vpsrlw $6, %ymm1, %ymm2 |
| ; AVX512VL-NEXT: vpand %ymm6, %ymm2, %ymm2 |
| ; AVX512VL-NEXT: vpsllw $2, %ymm1, %ymm3 |
| ; AVX512VL-NEXT: vpand %ymm8, %ymm3, %ymm3 |
| ; AVX512VL-NEXT: vpor %ymm2, %ymm3, %ymm2 |
| ; AVX512VL-NEXT: vpblendvb %ymm7, %ymm2, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: vpsrlw $7, %ymm1, %ymm2 |
| ; AVX512VL-NEXT: vpand %ymm9, %ymm2, %ymm2 |
| ; AVX512VL-NEXT: vpaddb %ymm1, %ymm1, %ymm3 |
| ; AVX512VL-NEXT: vpor %ymm2, %ymm3, %ymm2 |
| ; AVX512VL-NEXT: vpblendvb %ymm10, %ymm2, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: retq |
| ; |
| ; AVX512BW-LABEL: constant_rotate_v64i8: |
| ; AVX512BW: # %bb.0: |
| ; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm1 = [8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256] |
| ; AVX512BW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512BW-NEXT: vpsllw $4, %zmm0, %zmm2 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm2, %zmm2 |
| ; AVX512BW-NEXT: vpblendmb %zmm2, %zmm0, %zmm2 {%k1} |
| ; AVX512BW-NEXT: vpsllw $2, %zmm2, %zmm3 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm3, %zmm3 |
| ; AVX512BW-NEXT: vpaddb %zmm1, %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512BW-NEXT: vmovdqu8 %zmm3, %zmm2 {%k1} |
| ; AVX512BW-NEXT: vpaddb %zmm1, %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512BW-NEXT: vpaddb %zmm2, %zmm2, %zmm2 {%k1} |
| ; AVX512BW-NEXT: vmovdqa64 {{.*#+}} zmm1 = [57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536] |
| ; AVX512BW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512BW-NEXT: vpsrlw $4, %zmm0, %zmm3 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm3, %zmm3 |
| ; AVX512BW-NEXT: vmovdqu8 %zmm3, %zmm0 {%k1} |
| ; AVX512BW-NEXT: vpsrlw $2, %zmm0, %zmm3 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm3, %zmm3 |
| ; AVX512BW-NEXT: vpaddb %zmm1, %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512BW-NEXT: vmovdqu8 %zmm3, %zmm0 {%k1} |
| ; AVX512BW-NEXT: vpsrlw $1, %zmm0, %zmm3 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm3, %zmm3 |
| ; AVX512BW-NEXT: vpaddb %zmm1, %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512BW-NEXT: vmovdqu8 %zmm3, %zmm0 {%k1} |
| ; AVX512BW-NEXT: vporq %zmm0, %zmm2, %zmm0 |
| ; AVX512BW-NEXT: retq |
| ; |
| ; AVX512VLBW-LABEL: constant_rotate_v64i8: |
| ; AVX512VLBW: # %bb.0: |
| ; AVX512VLBW-NEXT: vmovdqa64 {{.*#+}} zmm1 = [8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256] |
| ; AVX512VLBW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512VLBW-NEXT: vpsllw $4, %zmm0, %zmm2 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm2, %zmm2 |
| ; AVX512VLBW-NEXT: vpblendmb %zmm2, %zmm0, %zmm2 {%k1} |
| ; AVX512VLBW-NEXT: vpsllw $2, %zmm2, %zmm3 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm3, %zmm3 |
| ; AVX512VLBW-NEXT: vpaddb %zmm1, %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512VLBW-NEXT: vmovdqu8 %zmm3, %zmm2 {%k1} |
| ; AVX512VLBW-NEXT: vpaddb %zmm1, %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512VLBW-NEXT: vpaddb %zmm2, %zmm2, %zmm2 {%k1} |
| ; AVX512VLBW-NEXT: vmovdqa64 {{.*#+}} zmm1 = [57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536,57600,41152,24704,8256,8192,24640,41088,57536] |
| ; AVX512VLBW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512VLBW-NEXT: vpsrlw $4, %zmm0, %zmm3 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm3, %zmm3 |
| ; AVX512VLBW-NEXT: vmovdqu8 %zmm3, %zmm0 {%k1} |
| ; AVX512VLBW-NEXT: vpsrlw $2, %zmm0, %zmm3 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm3, %zmm3 |
| ; AVX512VLBW-NEXT: vpaddb %zmm1, %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512VLBW-NEXT: vmovdqu8 %zmm3, %zmm0 {%k1} |
| ; AVX512VLBW-NEXT: vpsrlw $1, %zmm0, %zmm3 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm3, %zmm3 |
| ; AVX512VLBW-NEXT: vpaddb %zmm1, %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vpmovb2m %zmm1, %k1 |
| ; AVX512VLBW-NEXT: vmovdqu8 %zmm3, %zmm0 {%k1} |
| ; AVX512VLBW-NEXT: vporq %zmm0, %zmm2, %zmm0 |
| ; AVX512VLBW-NEXT: retq |
| %shl = shl <64 x i8> %a, <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> |
| %lshr = lshr <64 x i8> %a, <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, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7> |
| %or = or <64 x i8> %shl, %lshr |
| ret <64 x i8> %or |
| } |
| |
| ; |
| ; Uniform Constant Rotates |
| ; |
| |
| define <8 x i64> @splatconstant_rotate_v8i64(<8 x i64> %a) nounwind { |
| ; AVX512-LABEL: splatconstant_rotate_v8i64: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vprolq $14, %zmm0, %zmm0 |
| ; AVX512-NEXT: retq |
| %shl = shl <8 x i64> %a, <i64 14, i64 14, i64 14, i64 14, i64 14, i64 14, i64 14, i64 14> |
| %lshr = lshr <8 x i64> %a, <i64 50, i64 50, i64 50, i64 50, i64 50, i64 50, i64 50, i64 50> |
| %or = or <8 x i64> %shl, %lshr |
| ret <8 x i64> %or |
| } |
| |
| define <16 x i32> @splatconstant_rotate_v16i32(<16 x i32> %a) nounwind { |
| ; AVX512-LABEL: splatconstant_rotate_v16i32: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vprold $4, %zmm0, %zmm0 |
| ; AVX512-NEXT: retq |
| %shl = shl <16 x i32> %a, <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> |
| %lshr = lshr <16 x i32> %a, <i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28> |
| %or = or <16 x i32> %shl, %lshr |
| ret <16 x i32> %or |
| } |
| |
| define <32 x i16> @splatconstant_rotate_v32i16(<32 x i16> %a) nounwind { |
| ; AVX512F-LABEL: splatconstant_rotate_v32i16: |
| ; AVX512F: # %bb.0: |
| ; AVX512F-NEXT: vpsrlw $9, %ymm0, %ymm2 |
| ; AVX512F-NEXT: vpsllw $7, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpor %ymm2, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpsrlw $9, %ymm1, %ymm2 |
| ; AVX512F-NEXT: vpsllw $7, %ymm1, %ymm1 |
| ; AVX512F-NEXT: vpor %ymm2, %ymm1, %ymm1 |
| ; AVX512F-NEXT: retq |
| ; |
| ; AVX512VL-LABEL: splatconstant_rotate_v32i16: |
| ; AVX512VL: # %bb.0: |
| ; AVX512VL-NEXT: vpsrlw $9, %ymm0, %ymm2 |
| ; AVX512VL-NEXT: vpsllw $7, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpor %ymm2, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpsrlw $9, %ymm1, %ymm2 |
| ; AVX512VL-NEXT: vpsllw $7, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: vpor %ymm2, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: retq |
| ; |
| ; AVX512BW-LABEL: splatconstant_rotate_v32i16: |
| ; AVX512BW: # %bb.0: |
| ; AVX512BW-NEXT: vpsllw $7, %zmm0, %zmm1 |
| ; AVX512BW-NEXT: vpsrlw $9, %zmm0, %zmm0 |
| ; AVX512BW-NEXT: vporq %zmm0, %zmm1, %zmm0 |
| ; AVX512BW-NEXT: retq |
| ; |
| ; AVX512VLBW-LABEL: splatconstant_rotate_v32i16: |
| ; AVX512VLBW: # %bb.0: |
| ; AVX512VLBW-NEXT: vpsllw $7, %zmm0, %zmm1 |
| ; AVX512VLBW-NEXT: vpsrlw $9, %zmm0, %zmm0 |
| ; AVX512VLBW-NEXT: vporq %zmm0, %zmm1, %zmm0 |
| ; AVX512VLBW-NEXT: retq |
| %shl = shl <32 x i16> %a, <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> |
| %lshr = lshr <32 x i16> %a, <i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9, i16 9> |
| %or = or <32 x i16> %shl, %lshr |
| ret <32 x i16> %or |
| } |
| |
| define <64 x i8> @splatconstant_rotate_v64i8(<64 x i8> %a) nounwind { |
| ; AVX512F-LABEL: splatconstant_rotate_v64i8: |
| ; AVX512F: # %bb.0: |
| ; AVX512F-NEXT: vpsrlw $4, %ymm0, %ymm2 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm3 = [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 %ymm3, %ymm2, %ymm2 |
| ; AVX512F-NEXT: vpsllw $4, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm4 = [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 %ymm4, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpor %ymm2, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpsrlw $4, %ymm1, %ymm2 |
| ; AVX512F-NEXT: vpand %ymm3, %ymm2, %ymm2 |
| ; AVX512F-NEXT: vpsllw $4, %ymm1, %ymm1 |
| ; AVX512F-NEXT: vpand %ymm4, %ymm1, %ymm1 |
| ; AVX512F-NEXT: vpor %ymm2, %ymm1, %ymm1 |
| ; AVX512F-NEXT: retq |
| ; |
| ; AVX512VL-LABEL: splatconstant_rotate_v64i8: |
| ; AVX512VL: # %bb.0: |
| ; AVX512VL-NEXT: vpsrlw $4, %ymm0, %ymm2 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm3 = [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 %ymm3, %ymm2, %ymm2 |
| ; AVX512VL-NEXT: vpsllw $4, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm4 = [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 %ymm4, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpor %ymm2, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpsrlw $4, %ymm1, %ymm2 |
| ; AVX512VL-NEXT: vpand %ymm3, %ymm2, %ymm2 |
| ; AVX512VL-NEXT: vpsllw $4, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: vpand %ymm4, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: vpor %ymm2, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: retq |
| ; |
| ; AVX512BW-LABEL: splatconstant_rotate_v64i8: |
| ; AVX512BW: # %bb.0: |
| ; AVX512BW-NEXT: vpsllw $4, %zmm0, %zmm1 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vpsrlw $4, %zmm0, %zmm0 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm0, %zmm0 |
| ; AVX512BW-NEXT: vporq %zmm0, %zmm1, %zmm0 |
| ; AVX512BW-NEXT: retq |
| ; |
| ; AVX512VLBW-LABEL: splatconstant_rotate_v64i8: |
| ; AVX512VLBW: # %bb.0: |
| ; AVX512VLBW-NEXT: vpsllw $4, %zmm0, %zmm1 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vpsrlw $4, %zmm0, %zmm0 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm0, %zmm0 |
| ; AVX512VLBW-NEXT: vporq %zmm0, %zmm1, %zmm0 |
| ; AVX512VLBW-NEXT: retq |
| %shl = shl <64 x i8> %a, <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> |
| %lshr = lshr <64 x i8> %a, <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> |
| %or = or <64 x i8> %shl, %lshr |
| ret <64 x i8> %or |
| } |
| |
| ; |
| ; Masked Uniform Constant Rotates |
| ; |
| |
| define <8 x i64> @splatconstant_rotate_mask_v8i64(<8 x i64> %a) nounwind { |
| ; AVX512-LABEL: splatconstant_rotate_mask_v8i64: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vprolq $15, %zmm0, %zmm0 |
| ; AVX512-NEXT: vpandq {{.*}}(%rip), %zmm0, %zmm0 |
| ; AVX512-NEXT: retq |
| %shl = shl <8 x i64> %a, <i64 15, i64 15, i64 15, i64 15, i64 15, i64 15, i64 15, i64 15> |
| %lshr = lshr <8 x i64> %a, <i64 49, i64 49, i64 49, i64 49, i64 49, i64 49, i64 49, i64 49> |
| %rmask = and <8 x i64> %lshr, <i64 255, i64 127, i64 127, i64 255, i64 255, i64 127, i64 127, i64 255> |
| %lmask = and <8 x i64> %shl, <i64 33, i64 65, i64 129, i64 257, i64 33, i64 65, i64 129, i64 257> |
| %or = or <8 x i64> %lmask, %rmask |
| ret <8 x i64> %or |
| } |
| |
| define <16 x i32> @splatconstant_rotate_mask_v16i32(<16 x i32> %a) nounwind { |
| ; AVX512-LABEL: splatconstant_rotate_mask_v16i32: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vprold $4, %zmm0, %zmm0 |
| ; AVX512-NEXT: vpandq {{.*}}(%rip), %zmm0, %zmm0 |
| ; AVX512-NEXT: retq |
| %shl = shl <16 x i32> %a, <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> |
| %lshr = lshr <16 x i32> %a, <i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28, i32 28> |
| %rmask = and <16 x i32> %lshr, <i32 3, i32 7, i32 15, i32 31, i32 63, i32 127, i32 255, i32 511, i32 3, i32 7, i32 15, i32 31, i32 63, i32 127, i32 255, i32 511> |
| %lmask = and <16 x i32> %shl, <i32 511, i32 255, i32 127, i32 63, i32 31, i32 15, i32 7, i32 3, i32 511, i32 255, i32 127, i32 63, i32 31, i32 15, i32 7, i32 3> |
| %or = or <16 x i32> %lmask, %rmask |
| ret <16 x i32> %or |
| } |
| |
| define <32 x i16> @splatconstant_rotate_mask_v32i16(<32 x i16> %a) nounwind { |
| ; AVX512F-LABEL: splatconstant_rotate_mask_v32i16: |
| ; AVX512F: # %bb.0: |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm2 = [55,55,55,55,55,55,55,55,55,55,55,55,55,55,55,55] |
| ; AVX512F-NEXT: vpsrlw $11, %ymm0, %ymm3 |
| ; AVX512F-NEXT: vpsllw $5, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpor %ymm3, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpand %ymm2, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpsrlw $11, %ymm1, %ymm3 |
| ; AVX512F-NEXT: vpsllw $5, %ymm1, %ymm1 |
| ; AVX512F-NEXT: vpor %ymm3, %ymm1, %ymm1 |
| ; AVX512F-NEXT: vpand %ymm2, %ymm1, %ymm1 |
| ; AVX512F-NEXT: retq |
| ; |
| ; AVX512VL-LABEL: splatconstant_rotate_mask_v32i16: |
| ; AVX512VL: # %bb.0: |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm2 = [55,55,55,55,55,55,55,55,55,55,55,55,55,55,55,55] |
| ; AVX512VL-NEXT: vpsrlw $11, %ymm0, %ymm3 |
| ; AVX512VL-NEXT: vpsllw $5, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpor %ymm3, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpand %ymm2, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpsrlw $11, %ymm1, %ymm3 |
| ; AVX512VL-NEXT: vpsllw $5, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: vpor %ymm3, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: vpand %ymm2, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: retq |
| ; |
| ; AVX512BW-LABEL: splatconstant_rotate_mask_v32i16: |
| ; AVX512BW: # %bb.0: |
| ; AVX512BW-NEXT: vpsllw $5, %zmm0, %zmm1 |
| ; AVX512BW-NEXT: vpsrlw $11, %zmm0, %zmm0 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm0, %zmm0 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vporq %zmm0, %zmm1, %zmm0 |
| ; AVX512BW-NEXT: retq |
| ; |
| ; AVX512VLBW-LABEL: splatconstant_rotate_mask_v32i16: |
| ; AVX512VLBW: # %bb.0: |
| ; AVX512VLBW-NEXT: vpsllw $5, %zmm0, %zmm1 |
| ; AVX512VLBW-NEXT: vpsrlw $11, %zmm0, %zmm0 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm0, %zmm0 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vporq %zmm0, %zmm1, %zmm0 |
| ; AVX512VLBW-NEXT: retq |
| %shl = shl <32 x i16> %a, <i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5, i16 5> |
| %lshr = lshr <32 x i16> %a, <i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11, i16 11> |
| %rmask = and <32 x i16> %lshr, <i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55, i16 55> |
| %lmask = and <32 x i16> %shl, <i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33, i16 33> |
| %or = or <32 x i16> %lmask, %rmask |
| ret <32 x i16> %or |
| } |
| |
| define <64 x i8> @splatconstant_rotate_mask_v64i8(<64 x i8> %a) nounwind { |
| ; AVX512F-LABEL: splatconstant_rotate_mask_v64i8: |
| ; AVX512F: # %bb.0: |
| ; AVX512F-NEXT: vpsrlw $4, %ymm0, %ymm2 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm3 = [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 %ymm3, %ymm2, %ymm2 |
| ; AVX512F-NEXT: vpsllw $4, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm4 = [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 %ymm4, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpor %ymm2, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vmovdqa {{.*#+}} ymm2 = [39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39] |
| ; AVX512F-NEXT: vpand %ymm2, %ymm0, %ymm0 |
| ; AVX512F-NEXT: vpsrlw $4, %ymm1, %ymm5 |
| ; AVX512F-NEXT: vpand %ymm3, %ymm5, %ymm3 |
| ; AVX512F-NEXT: vpsllw $4, %ymm1, %ymm1 |
| ; AVX512F-NEXT: vpand %ymm4, %ymm1, %ymm1 |
| ; AVX512F-NEXT: vpor %ymm3, %ymm1, %ymm1 |
| ; AVX512F-NEXT: vpand %ymm2, %ymm1, %ymm1 |
| ; AVX512F-NEXT: retq |
| ; |
| ; AVX512VL-LABEL: splatconstant_rotate_mask_v64i8: |
| ; AVX512VL: # %bb.0: |
| ; AVX512VL-NEXT: vpsrlw $4, %ymm0, %ymm2 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm3 = [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 %ymm3, %ymm2, %ymm2 |
| ; AVX512VL-NEXT: vpsllw $4, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm4 = [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 %ymm4, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpor %ymm2, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vmovdqa {{.*#+}} ymm2 = [39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39,39] |
| ; AVX512VL-NEXT: vpand %ymm2, %ymm0, %ymm0 |
| ; AVX512VL-NEXT: vpsrlw $4, %ymm1, %ymm5 |
| ; AVX512VL-NEXT: vpand %ymm3, %ymm5, %ymm3 |
| ; AVX512VL-NEXT: vpsllw $4, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: vpand %ymm4, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: vpor %ymm3, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: vpand %ymm2, %ymm1, %ymm1 |
| ; AVX512VL-NEXT: retq |
| ; |
| ; AVX512BW-LABEL: splatconstant_rotate_mask_v64i8: |
| ; AVX512BW: # %bb.0: |
| ; AVX512BW-NEXT: vpsllw $4, %zmm0, %zmm1 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vpsrlw $4, %zmm0, %zmm0 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm0, %zmm0 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm0, %zmm0 |
| ; AVX512BW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512BW-NEXT: vporq %zmm0, %zmm1, %zmm0 |
| ; AVX512BW-NEXT: retq |
| ; |
| ; AVX512VLBW-LABEL: splatconstant_rotate_mask_v64i8: |
| ; AVX512VLBW: # %bb.0: |
| ; AVX512VLBW-NEXT: vpsllw $4, %zmm0, %zmm1 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vpsrlw $4, %zmm0, %zmm0 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm0, %zmm0 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm0, %zmm0 |
| ; AVX512VLBW-NEXT: vpandq {{.*}}(%rip), %zmm1, %zmm1 |
| ; AVX512VLBW-NEXT: vporq %zmm0, %zmm1, %zmm0 |
| ; AVX512VLBW-NEXT: retq |
| %shl = shl <64 x i8> %a, <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> |
| %lshr = lshr <64 x i8> %a, <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> |
| %rmask = and <64 x i8> %lshr, <i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55, i8 55> |
| %lmask = and <64 x i8> %shl, <i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33, i8 33> |
| %or = or <64 x i8> %lmask, %rmask |
| ret <64 x i8> %or |
| } |