| ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py |
| ; RUN: llc < %s -mtriple=x86_64-linux -mattr=+avx512bw | FileCheck %s --check-prefixes=AVX512,AVX512BW |
| ; RUN: llc < %s -mtriple=x86_64-linux -mattr=+avx512dq | FileCheck %s --check-prefixes=AVX512,AVX512DQ |
| |
| ; |
| ; trunc(abs(sub(zext(a),zext(b)))) -> abdu(a,b) |
| ; |
| |
| define <64 x i8> @abd_ext_v64i8(<64 x i8> %a, <64 x i8> %b) nounwind { |
| ; AVX512-LABEL: abd_ext_v64i8: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero |
| ; AVX512-NEXT: vextracti128 $1, %ymm2, %xmm2 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero |
| ; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm4 |
| ; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm5 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero,xmm4[4],zero,zero,zero,xmm4[5],zero,zero,zero,xmm4[6],zero,zero,zero,xmm4[7],zero,zero,zero |
| ; AVX512-NEXT: vextracti128 $1, %ymm4, %xmm4 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero,xmm4[4],zero,zero,zero,xmm4[5],zero,zero,zero,xmm4[6],zero,zero,zero,xmm4[7],zero,zero,zero |
| ; AVX512-NEXT: vextracti64x4 $1, %zmm0, %ymm0 |
| ; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm6 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm7 = xmm6[0],zero,zero,zero,xmm6[1],zero,zero,zero,xmm6[2],zero,zero,zero,xmm6[3],zero,zero,zero,xmm6[4],zero,zero,zero,xmm6[5],zero,zero,zero,xmm6[6],zero,zero,zero,xmm6[7],zero,zero,zero |
| ; AVX512-NEXT: vextracti128 $1, %ymm6, %xmm6 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm6 = xmm6[0],zero,zero,zero,xmm6[1],zero,zero,zero,xmm6[2],zero,zero,zero,xmm6[3],zero,zero,zero,xmm6[4],zero,zero,zero,xmm6[5],zero,zero,zero,xmm6[6],zero,zero,zero,xmm6[7],zero,zero,zero |
| ; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm0 |
| ; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm8 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero |
| ; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm0 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero |
| ; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm9 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm10 = xmm9[0],zero,zero,zero,xmm9[1],zero,zero,zero,xmm9[2],zero,zero,zero,xmm9[3],zero,zero,zero,xmm9[4],zero,zero,zero,xmm9[5],zero,zero,zero,xmm9[6],zero,zero,zero,xmm9[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm10, %zmm3, %zmm3 |
| ; AVX512-NEXT: vextracti128 $1, %ymm9, %xmm9 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm9 = xmm9[0],zero,zero,zero,xmm9[1],zero,zero,zero,xmm9[2],zero,zero,zero,xmm9[3],zero,zero,zero,xmm9[4],zero,zero,zero,xmm9[5],zero,zero,zero,xmm9[6],zero,zero,zero,xmm9[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm9, %zmm2, %zmm2 |
| ; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm9 |
| ; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm9 = xmm9[0],zero,xmm9[1],zero,xmm9[2],zero,xmm9[3],zero,xmm9[4],zero,xmm9[5],zero,xmm9[6],zero,xmm9[7],zero,xmm9[8],zero,xmm9[9],zero,xmm9[10],zero,xmm9[11],zero,xmm9[12],zero,xmm9[13],zero,xmm9[14],zero,xmm9[15],zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm10 = xmm9[0],zero,zero,zero,xmm9[1],zero,zero,zero,xmm9[2],zero,zero,zero,xmm9[3],zero,zero,zero,xmm9[4],zero,zero,zero,xmm9[5],zero,zero,zero,xmm9[6],zero,zero,zero,xmm9[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm10, %zmm5, %zmm5 |
| ; AVX512-NEXT: vextracti128 $1, %ymm9, %xmm9 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm9 = xmm9[0],zero,zero,zero,xmm9[1],zero,zero,zero,xmm9[2],zero,zero,zero,xmm9[3],zero,zero,zero,xmm9[4],zero,zero,zero,xmm9[5],zero,zero,zero,xmm9[6],zero,zero,zero,xmm9[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm9, %zmm4, %zmm4 |
| ; AVX512-NEXT: vextracti64x4 $1, %zmm1, %ymm1 |
| ; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm9 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm10 = xmm9[0],zero,zero,zero,xmm9[1],zero,zero,zero,xmm9[2],zero,zero,zero,xmm9[3],zero,zero,zero,xmm9[4],zero,zero,zero,xmm9[5],zero,zero,zero,xmm9[6],zero,zero,zero,xmm9[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm10, %zmm7, %zmm7 |
| ; AVX512-NEXT: vextracti128 $1, %ymm9, %xmm9 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm9 = xmm9[0],zero,zero,zero,xmm9[1],zero,zero,zero,xmm9[2],zero,zero,zero,xmm9[3],zero,zero,zero,xmm9[4],zero,zero,zero,xmm9[5],zero,zero,zero,xmm9[6],zero,zero,zero,xmm9[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm9, %zmm6, %zmm6 |
| ; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm1 |
| ; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm9 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm9, %zmm8, %zmm8 |
| ; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm1 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm1, %zmm0, %zmm0 |
| ; AVX512-NEXT: vpabsq %zmm3, %zmm1 |
| ; AVX512-NEXT: vpabsq %zmm2, %zmm2 |
| ; AVX512-NEXT: vpabsq %zmm5, %zmm3 |
| ; AVX512-NEXT: vpabsq %zmm4, %zmm4 |
| ; AVX512-NEXT: vpabsq %zmm7, %zmm5 |
| ; AVX512-NEXT: vpabsq %zmm6, %zmm6 |
| ; AVX512-NEXT: vpabsq %zmm8, %zmm7 |
| ; AVX512-NEXT: vpabsq %zmm0, %zmm0 |
| ; AVX512-NEXT: vpmovqb %zmm0, %xmm0 |
| ; AVX512-NEXT: vpmovqb %zmm7, %xmm7 |
| ; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm7[0],xmm0[0] |
| ; AVX512-NEXT: vpmovqb %zmm6, %xmm6 |
| ; AVX512-NEXT: vpmovqb %zmm5, %xmm5 |
| ; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm5 = xmm5[0],xmm6[0] |
| ; AVX512-NEXT: vinserti128 $1, %xmm0, %ymm5, %ymm0 |
| ; AVX512-NEXT: vpmovqb %zmm4, %xmm4 |
| ; AVX512-NEXT: vpmovqb %zmm3, %xmm3 |
| ; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm4[0] |
| ; AVX512-NEXT: vpmovqb %zmm2, %xmm2 |
| ; AVX512-NEXT: vpmovqb %zmm1, %xmm1 |
| ; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0] |
| ; AVX512-NEXT: vinserti128 $1, %xmm3, %ymm1, %ymm1 |
| ; AVX512-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 |
| ; AVX512-NEXT: retq |
| %aext = zext <64 x i8> %a to <64 x i64> |
| %bext = zext <64 x i8> %b to <64 x i64> |
| %sub = sub <64 x i64> %aext, %bext |
| %abs = call <64 x i64> @llvm.abs.v64i64(<64 x i64> %sub, i1 false) |
| %trunc = trunc <64 x i64> %abs to <64 x i8> |
| ret <64 x i8> %trunc |
| } |
| |
| define <64 x i8> @abd_ext_v64i8_undef(<64 x i8> %a, <64 x i8> %b) nounwind { |
| ; AVX512-LABEL: abd_ext_v64i8_undef: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm2 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm3 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero |
| ; AVX512-NEXT: vextracti128 $1, %ymm2, %xmm2 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero |
| ; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm4 |
| ; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm4 = xmm4[0],zero,xmm4[1],zero,xmm4[2],zero,xmm4[3],zero,xmm4[4],zero,xmm4[5],zero,xmm4[6],zero,xmm4[7],zero,xmm4[8],zero,xmm4[9],zero,xmm4[10],zero,xmm4[11],zero,xmm4[12],zero,xmm4[13],zero,xmm4[14],zero,xmm4[15],zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm5 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero,xmm4[4],zero,zero,zero,xmm4[5],zero,zero,zero,xmm4[6],zero,zero,zero,xmm4[7],zero,zero,zero |
| ; AVX512-NEXT: vextracti128 $1, %ymm4, %xmm4 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero,xmm4[4],zero,zero,zero,xmm4[5],zero,zero,zero,xmm4[6],zero,zero,zero,xmm4[7],zero,zero,zero |
| ; AVX512-NEXT: vextracti64x4 $1, %zmm0, %ymm0 |
| ; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm6 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm7 = xmm6[0],zero,zero,zero,xmm6[1],zero,zero,zero,xmm6[2],zero,zero,zero,xmm6[3],zero,zero,zero,xmm6[4],zero,zero,zero,xmm6[5],zero,zero,zero,xmm6[6],zero,zero,zero,xmm6[7],zero,zero,zero |
| ; AVX512-NEXT: vextracti128 $1, %ymm6, %xmm6 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm6 = xmm6[0],zero,zero,zero,xmm6[1],zero,zero,zero,xmm6[2],zero,zero,zero,xmm6[3],zero,zero,zero,xmm6[4],zero,zero,zero,xmm6[5],zero,zero,zero,xmm6[6],zero,zero,zero,xmm6[7],zero,zero,zero |
| ; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm0 |
| ; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero,xmm0[8],zero,xmm0[9],zero,xmm0[10],zero,xmm0[11],zero,xmm0[12],zero,xmm0[13],zero,xmm0[14],zero,xmm0[15],zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm8 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero |
| ; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm0 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero |
| ; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm9 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm10 = xmm9[0],zero,zero,zero,xmm9[1],zero,zero,zero,xmm9[2],zero,zero,zero,xmm9[3],zero,zero,zero,xmm9[4],zero,zero,zero,xmm9[5],zero,zero,zero,xmm9[6],zero,zero,zero,xmm9[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm10, %zmm3, %zmm3 |
| ; AVX512-NEXT: vextracti128 $1, %ymm9, %xmm9 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm9 = xmm9[0],zero,zero,zero,xmm9[1],zero,zero,zero,xmm9[2],zero,zero,zero,xmm9[3],zero,zero,zero,xmm9[4],zero,zero,zero,xmm9[5],zero,zero,zero,xmm9[6],zero,zero,zero,xmm9[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm9, %zmm2, %zmm2 |
| ; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm9 |
| ; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm9 = xmm9[0],zero,xmm9[1],zero,xmm9[2],zero,xmm9[3],zero,xmm9[4],zero,xmm9[5],zero,xmm9[6],zero,xmm9[7],zero,xmm9[8],zero,xmm9[9],zero,xmm9[10],zero,xmm9[11],zero,xmm9[12],zero,xmm9[13],zero,xmm9[14],zero,xmm9[15],zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm10 = xmm9[0],zero,zero,zero,xmm9[1],zero,zero,zero,xmm9[2],zero,zero,zero,xmm9[3],zero,zero,zero,xmm9[4],zero,zero,zero,xmm9[5],zero,zero,zero,xmm9[6],zero,zero,zero,xmm9[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm10, %zmm5, %zmm5 |
| ; AVX512-NEXT: vextracti128 $1, %ymm9, %xmm9 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm9 = xmm9[0],zero,zero,zero,xmm9[1],zero,zero,zero,xmm9[2],zero,zero,zero,xmm9[3],zero,zero,zero,xmm9[4],zero,zero,zero,xmm9[5],zero,zero,zero,xmm9[6],zero,zero,zero,xmm9[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm9, %zmm4, %zmm4 |
| ; AVX512-NEXT: vextracti64x4 $1, %zmm1, %ymm1 |
| ; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm9 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm10 = xmm9[0],zero,zero,zero,xmm9[1],zero,zero,zero,xmm9[2],zero,zero,zero,xmm9[3],zero,zero,zero,xmm9[4],zero,zero,zero,xmm9[5],zero,zero,zero,xmm9[6],zero,zero,zero,xmm9[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm10, %zmm7, %zmm7 |
| ; AVX512-NEXT: vextracti128 $1, %ymm9, %xmm9 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm9 = xmm9[0],zero,zero,zero,xmm9[1],zero,zero,zero,xmm9[2],zero,zero,zero,xmm9[3],zero,zero,zero,xmm9[4],zero,zero,zero,xmm9[5],zero,zero,zero,xmm9[6],zero,zero,zero,xmm9[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm9, %zmm6, %zmm6 |
| ; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm1 |
| ; AVX512-NEXT: vpmovzxbw {{.*#+}} ymm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero,xmm1[8],zero,xmm1[9],zero,xmm1[10],zero,xmm1[11],zero,xmm1[12],zero,xmm1[13],zero,xmm1[14],zero,xmm1[15],zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm9 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm9, %zmm8, %zmm8 |
| ; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm1 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm1, %zmm0, %zmm0 |
| ; AVX512-NEXT: vpabsq %zmm3, %zmm1 |
| ; AVX512-NEXT: vpabsq %zmm2, %zmm2 |
| ; AVX512-NEXT: vpabsq %zmm5, %zmm3 |
| ; AVX512-NEXT: vpabsq %zmm4, %zmm4 |
| ; AVX512-NEXT: vpabsq %zmm7, %zmm5 |
| ; AVX512-NEXT: vpabsq %zmm6, %zmm6 |
| ; AVX512-NEXT: vpabsq %zmm8, %zmm7 |
| ; AVX512-NEXT: vpabsq %zmm0, %zmm0 |
| ; AVX512-NEXT: vpmovqb %zmm0, %xmm0 |
| ; AVX512-NEXT: vpmovqb %zmm7, %xmm7 |
| ; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm7[0],xmm0[0] |
| ; AVX512-NEXT: vpmovqb %zmm6, %xmm6 |
| ; AVX512-NEXT: vpmovqb %zmm5, %xmm5 |
| ; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm5 = xmm5[0],xmm6[0] |
| ; AVX512-NEXT: vinserti128 $1, %xmm0, %ymm5, %ymm0 |
| ; AVX512-NEXT: vpmovqb %zmm4, %xmm4 |
| ; AVX512-NEXT: vpmovqb %zmm3, %xmm3 |
| ; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm3 = xmm3[0],xmm4[0] |
| ; AVX512-NEXT: vpmovqb %zmm2, %xmm2 |
| ; AVX512-NEXT: vpmovqb %zmm1, %xmm1 |
| ; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm1[0],xmm2[0] |
| ; AVX512-NEXT: vinserti128 $1, %xmm3, %ymm1, %ymm1 |
| ; AVX512-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 |
| ; AVX512-NEXT: retq |
| %aext = zext <64 x i8> %a to <64 x i64> |
| %bext = zext <64 x i8> %b to <64 x i64> |
| %sub = sub <64 x i64> %aext, %bext |
| %abs = call <64 x i64> @llvm.abs.v64i64(<64 x i64> %sub, i1 true) |
| %trunc = trunc <64 x i64> %abs to <64 x i8> |
| ret <64 x i8> %trunc |
| } |
| |
| define <32 x i16> @abd_ext_v32i16(<32 x i16> %a, <32 x i16> %b) nounwind { |
| ; AVX512-LABEL: abd_ext_v32i16: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm2 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm3 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero |
| ; AVX512-NEXT: vextracti64x4 $1, %zmm0, %ymm0 |
| ; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm4 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero,xmm4[4],zero,zero,zero,xmm4[5],zero,zero,zero,xmm4[6],zero,zero,zero,xmm4[7],zero,zero,zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero |
| ; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm5 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm5 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero,xmm5[2],zero,zero,zero,xmm5[3],zero,zero,zero,xmm5[4],zero,zero,zero,xmm5[5],zero,zero,zero,xmm5[6],zero,zero,zero,xmm5[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm5, %zmm2, %zmm2 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm5 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm5, %zmm3, %zmm3 |
| ; AVX512-NEXT: vextracti64x4 $1, %zmm1, %ymm1 |
| ; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm5 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm5 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero,xmm5[2],zero,zero,zero,xmm5[3],zero,zero,zero,xmm5[4],zero,zero,zero,xmm5[5],zero,zero,zero,xmm5[6],zero,zero,zero,xmm5[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm5, %zmm4, %zmm4 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm1, %zmm0, %zmm0 |
| ; AVX512-NEXT: vpabsq %zmm2, %zmm1 |
| ; AVX512-NEXT: vpabsq %zmm3, %zmm2 |
| ; AVX512-NEXT: vpabsq %zmm4, %zmm3 |
| ; AVX512-NEXT: vpabsq %zmm0, %zmm0 |
| ; AVX512-NEXT: vpmovqw %zmm0, %xmm0 |
| ; AVX512-NEXT: vpmovqw %zmm3, %xmm3 |
| ; AVX512-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm0 |
| ; AVX512-NEXT: vpmovqw %zmm2, %xmm2 |
| ; AVX512-NEXT: vpmovqw %zmm1, %xmm1 |
| ; AVX512-NEXT: vinserti128 $1, %xmm1, %ymm2, %ymm1 |
| ; AVX512-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 |
| ; AVX512-NEXT: retq |
| %aext = zext <32 x i16> %a to <32 x i64> |
| %bext = zext <32 x i16> %b to <32 x i64> |
| %sub = sub <32 x i64> %aext, %bext |
| %abs = call <32 x i64> @llvm.abs.v32i64(<32 x i64> %sub, i1 false) |
| %trunc = trunc <32 x i64> %abs to <32 x i16> |
| ret <32 x i16> %trunc |
| } |
| |
| define <32 x i16> @abd_ext_v32i16_undef(<32 x i16> %a, <32 x i16> %b) nounwind { |
| ; AVX512-LABEL: abd_ext_v32i16_undef: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm2 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm2 = xmm2[0],zero,zero,zero,xmm2[1],zero,zero,zero,xmm2[2],zero,zero,zero,xmm2[3],zero,zero,zero,xmm2[4],zero,zero,zero,xmm2[5],zero,zero,zero,xmm2[6],zero,zero,zero,xmm2[7],zero,zero,zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm3 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero |
| ; AVX512-NEXT: vextracti64x4 $1, %zmm0, %ymm0 |
| ; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm4 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm4 = xmm4[0],zero,zero,zero,xmm4[1],zero,zero,zero,xmm4[2],zero,zero,zero,xmm4[3],zero,zero,zero,xmm4[4],zero,zero,zero,xmm4[5],zero,zero,zero,xmm4[6],zero,zero,zero,xmm4[7],zero,zero,zero |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero,xmm0[2],zero,zero,zero,xmm0[3],zero,zero,zero,xmm0[4],zero,zero,zero,xmm0[5],zero,zero,zero,xmm0[6],zero,zero,zero,xmm0[7],zero,zero,zero |
| ; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm5 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm5 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero,xmm5[2],zero,zero,zero,xmm5[3],zero,zero,zero,xmm5[4],zero,zero,zero,xmm5[5],zero,zero,zero,xmm5[6],zero,zero,zero,xmm5[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm5, %zmm2, %zmm2 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm5 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm5, %zmm3, %zmm3 |
| ; AVX512-NEXT: vextracti64x4 $1, %zmm1, %ymm1 |
| ; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm5 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm5 = xmm5[0],zero,zero,zero,xmm5[1],zero,zero,zero,xmm5[2],zero,zero,zero,xmm5[3],zero,zero,zero,xmm5[4],zero,zero,zero,xmm5[5],zero,zero,zero,xmm5[6],zero,zero,zero,xmm5[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm5, %zmm4, %zmm4 |
| ; AVX512-NEXT: vpmovzxwq {{.*#+}} zmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero,xmm1[2],zero,zero,zero,xmm1[3],zero,zero,zero,xmm1[4],zero,zero,zero,xmm1[5],zero,zero,zero,xmm1[6],zero,zero,zero,xmm1[7],zero,zero,zero |
| ; AVX512-NEXT: vpsubq %zmm1, %zmm0, %zmm0 |
| ; AVX512-NEXT: vpabsq %zmm2, %zmm1 |
| ; AVX512-NEXT: vpabsq %zmm3, %zmm2 |
| ; AVX512-NEXT: vpabsq %zmm4, %zmm3 |
| ; AVX512-NEXT: vpabsq %zmm0, %zmm0 |
| ; AVX512-NEXT: vpmovqw %zmm0, %xmm0 |
| ; AVX512-NEXT: vpmovqw %zmm3, %xmm3 |
| ; AVX512-NEXT: vinserti128 $1, %xmm3, %ymm0, %ymm0 |
| ; AVX512-NEXT: vpmovqw %zmm2, %xmm2 |
| ; AVX512-NEXT: vpmovqw %zmm1, %xmm1 |
| ; AVX512-NEXT: vinserti128 $1, %xmm1, %ymm2, %ymm1 |
| ; AVX512-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 |
| ; AVX512-NEXT: retq |
| %aext = zext <32 x i16> %a to <32 x i64> |
| %bext = zext <32 x i16> %b to <32 x i64> |
| %sub = sub <32 x i64> %aext, %bext |
| %abs = call <32 x i64> @llvm.abs.v32i64(<32 x i64> %sub, i1 true) |
| %trunc = trunc <32 x i64> %abs to <32 x i16> |
| ret <32 x i16> %trunc |
| } |
| |
| define <16 x i32> @abd_ext_v16i32(<16 x i32> %a, <16 x i32> %b) nounwind { |
| ; AVX512-LABEL: abd_ext_v16i32: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vextracti64x4 $1, %zmm0, %ymm2 |
| ; AVX512-NEXT: vpmovzxdq {{.*#+}} 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 |
| ; AVX512-NEXT: vpmovzxdq {{.*#+}} 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 |
| ; AVX512-NEXT: vextracti64x4 $1, %zmm1, %ymm3 |
| ; AVX512-NEXT: vpmovzxdq {{.*#+}} 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 |
| ; AVX512-NEXT: vpsubq %zmm3, %zmm2, %zmm2 |
| ; AVX512-NEXT: vpmovzxdq {{.*#+}} 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 |
| ; AVX512-NEXT: vpsubq %zmm1, %zmm0, %zmm0 |
| ; AVX512-NEXT: vpabsq %zmm2, %zmm1 |
| ; AVX512-NEXT: vpabsq %zmm0, %zmm0 |
| ; AVX512-NEXT: vpmovqd %zmm0, %ymm0 |
| ; AVX512-NEXT: vpmovqd %zmm1, %ymm1 |
| ; AVX512-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0 |
| ; AVX512-NEXT: retq |
| %aext = zext <16 x i32> %a to <16 x i64> |
| %bext = zext <16 x i32> %b to <16 x i64> |
| %sub = sub <16 x i64> %aext, %bext |
| %abs = call <16 x i64> @llvm.abs.v16i64(<16 x i64> %sub, i1 false) |
| %trunc = trunc <16 x i64> %abs to <16 x i32> |
| ret <16 x i32> %trunc |
| } |
| |
| define <16 x i32> @abd_ext_v16i32_undef(<16 x i32> %a, <16 x i32> %b) nounwind { |
| ; AVX512-LABEL: abd_ext_v16i32_undef: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vextracti64x4 $1, %zmm0, %ymm2 |
| ; AVX512-NEXT: vpmovzxdq {{.*#+}} 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 |
| ; AVX512-NEXT: vpmovzxdq {{.*#+}} 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 |
| ; AVX512-NEXT: vextracti64x4 $1, %zmm1, %ymm3 |
| ; AVX512-NEXT: vpmovzxdq {{.*#+}} 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 |
| ; AVX512-NEXT: vpsubq %zmm3, %zmm2, %zmm2 |
| ; AVX512-NEXT: vpmovzxdq {{.*#+}} 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 |
| ; AVX512-NEXT: vpsubq %zmm1, %zmm0, %zmm0 |
| ; AVX512-NEXT: vpabsq %zmm2, %zmm1 |
| ; AVX512-NEXT: vpabsq %zmm0, %zmm0 |
| ; AVX512-NEXT: vpmovqd %zmm0, %ymm0 |
| ; AVX512-NEXT: vpmovqd %zmm1, %ymm1 |
| ; AVX512-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0 |
| ; AVX512-NEXT: retq |
| %aext = zext <16 x i32> %a to <16 x i64> |
| %bext = zext <16 x i32> %b to <16 x i64> |
| %sub = sub <16 x i64> %aext, %bext |
| %abs = call <16 x i64> @llvm.abs.v16i64(<16 x i64> %sub, i1 true) |
| %trunc = trunc <16 x i64> %abs to <16 x i32> |
| ret <16 x i32> %trunc |
| } |
| |
| define <8 x i64> @abd_ext_v8i64(<8 x i64> %a, <8 x i64> %b) nounwind { |
| ; AVX512-LABEL: abd_ext_v8i64: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: pushq %rbp |
| ; AVX512-NEXT: pushq %r15 |
| ; AVX512-NEXT: pushq %r14 |
| ; AVX512-NEXT: pushq %r13 |
| ; AVX512-NEXT: pushq %r12 |
| ; AVX512-NEXT: pushq %rbx |
| ; AVX512-NEXT: vpextrq $1, %xmm0, %rcx |
| ; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm2 |
| ; AVX512-NEXT: vmovq %xmm2, %rax |
| ; AVX512-NEXT: vpextrq $1, %xmm2, %rdx |
| ; AVX512-NEXT: vextracti64x4 $1, %zmm0, %ymm2 |
| ; AVX512-NEXT: vmovq %xmm2, %rsi |
| ; AVX512-NEXT: vpextrq $1, %xmm2, %rdi |
| ; AVX512-NEXT: vextracti128 $1, %ymm2, %xmm2 |
| ; AVX512-NEXT: vmovq %xmm2, %r8 |
| ; AVX512-NEXT: vpextrq $1, %xmm2, %r9 |
| ; AVX512-NEXT: vpextrq $1, %xmm1, %rbx |
| ; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm2 |
| ; AVX512-NEXT: vmovq %xmm2, %rbp |
| ; AVX512-NEXT: vpextrq $1, %xmm2, %r13 |
| ; AVX512-NEXT: vextracti64x4 $1, %zmm1, %ymm2 |
| ; AVX512-NEXT: vmovq %xmm2, %r15 |
| ; AVX512-NEXT: vpextrq $1, %xmm2, %r14 |
| ; AVX512-NEXT: vextracti128 $1, %ymm2, %xmm2 |
| ; AVX512-NEXT: vmovq %xmm2, %r11 |
| ; AVX512-NEXT: vpextrq $1, %xmm2, %r10 |
| ; AVX512-NEXT: xorl %r12d, %r12d |
| ; AVX512-NEXT: subq %r10, %r9 |
| ; AVX512-NEXT: movl $0, %r10d |
| ; AVX512-NEXT: sbbq %r10, %r10 |
| ; AVX512-NEXT: subq %r11, %r8 |
| ; AVX512-NEXT: movl $0, %r11d |
| ; AVX512-NEXT: sbbq %r11, %r11 |
| ; AVX512-NEXT: subq %r14, %rdi |
| ; AVX512-NEXT: movl $0, %r14d |
| ; AVX512-NEXT: sbbq %r14, %r14 |
| ; AVX512-NEXT: subq %r15, %rsi |
| ; AVX512-NEXT: movl $0, %r15d |
| ; AVX512-NEXT: sbbq %r15, %r15 |
| ; AVX512-NEXT: subq %r13, %rdx |
| ; AVX512-NEXT: movl $0, %r13d |
| ; AVX512-NEXT: sbbq %r13, %r13 |
| ; AVX512-NEXT: subq %rbp, %rax |
| ; AVX512-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill |
| ; AVX512-NEXT: movl $0, %ebp |
| ; AVX512-NEXT: sbbq %rbp, %rbp |
| ; AVX512-NEXT: subq %rbx, %rcx |
| ; AVX512-NEXT: movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill |
| ; AVX512-NEXT: movl $0, %eax |
| ; AVX512-NEXT: sbbq %rax, %rax |
| ; AVX512-NEXT: vmovq %xmm0, %rbx |
| ; AVX512-NEXT: vmovq %xmm1, %rcx |
| ; AVX512-NEXT: subq %rcx, %rbx |
| ; AVX512-NEXT: sbbq %r12, %r12 |
| ; AVX512-NEXT: sarq $63, %r12 |
| ; AVX512-NEXT: xorq %r12, %rbx |
| ; AVX512-NEXT: subq %r12, %rbx |
| ; AVX512-NEXT: sarq $63, %rax |
| ; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload |
| ; AVX512-NEXT: xorq %rax, %rcx |
| ; AVX512-NEXT: subq %rax, %rcx |
| ; AVX512-NEXT: movq %rcx, %rax |
| ; AVX512-NEXT: sarq $63, %rbp |
| ; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload |
| ; AVX512-NEXT: xorq %rbp, %rcx |
| ; AVX512-NEXT: subq %rbp, %rcx |
| ; AVX512-NEXT: sarq $63, %r13 |
| ; AVX512-NEXT: xorq %r13, %rdx |
| ; AVX512-NEXT: subq %r13, %rdx |
| ; AVX512-NEXT: sarq $63, %r15 |
| ; AVX512-NEXT: xorq %r15, %rsi |
| ; AVX512-NEXT: subq %r15, %rsi |
| ; AVX512-NEXT: sarq $63, %r14 |
| ; AVX512-NEXT: xorq %r14, %rdi |
| ; AVX512-NEXT: subq %r14, %rdi |
| ; AVX512-NEXT: sarq $63, %r11 |
| ; AVX512-NEXT: xorq %r11, %r8 |
| ; AVX512-NEXT: subq %r11, %r8 |
| ; AVX512-NEXT: sarq $63, %r10 |
| ; AVX512-NEXT: xorq %r10, %r9 |
| ; AVX512-NEXT: subq %r10, %r9 |
| ; AVX512-NEXT: vmovq %r9, %xmm0 |
| ; AVX512-NEXT: vmovq %r8, %xmm1 |
| ; AVX512-NEXT: vmovq %rdi, %xmm2 |
| ; AVX512-NEXT: vmovq %rsi, %xmm3 |
| ; AVX512-NEXT: vmovq %rdx, %xmm4 |
| ; AVX512-NEXT: vmovq %rcx, %xmm5 |
| ; AVX512-NEXT: vmovq %rax, %xmm6 |
| ; AVX512-NEXT: vmovq %rbx, %xmm7 |
| ; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0] |
| ; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm3[0],xmm2[0] |
| ; AVX512-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0 |
| ; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm5[0],xmm4[0] |
| ; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm7[0],xmm6[0] |
| ; AVX512-NEXT: vinserti128 $1, %xmm1, %ymm2, %ymm1 |
| ; AVX512-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 |
| ; AVX512-NEXT: popq %rbx |
| ; AVX512-NEXT: popq %r12 |
| ; AVX512-NEXT: popq %r13 |
| ; AVX512-NEXT: popq %r14 |
| ; AVX512-NEXT: popq %r15 |
| ; AVX512-NEXT: popq %rbp |
| ; AVX512-NEXT: retq |
| %aext = zext <8 x i64> %a to <8 x i128> |
| %bext = zext <8 x i64> %b to <8 x i128> |
| %sub = sub <8 x i128> %aext, %bext |
| %abs = call <8 x i128> @llvm.abs.v8i128(<8 x i128> %sub, i1 false) |
| %trunc = trunc <8 x i128> %abs to <8 x i64> |
| ret <8 x i64> %trunc |
| } |
| |
| define <8 x i64> @abd_ext_v8i64_undef(<8 x i64> %a, <8 x i64> %b) nounwind { |
| ; AVX512-LABEL: abd_ext_v8i64_undef: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: pushq %rbp |
| ; AVX512-NEXT: pushq %r15 |
| ; AVX512-NEXT: pushq %r14 |
| ; AVX512-NEXT: pushq %r13 |
| ; AVX512-NEXT: pushq %r12 |
| ; AVX512-NEXT: pushq %rbx |
| ; AVX512-NEXT: vpextrq $1, %xmm0, %rcx |
| ; AVX512-NEXT: vextracti128 $1, %ymm0, %xmm2 |
| ; AVX512-NEXT: vmovq %xmm2, %rax |
| ; AVX512-NEXT: vpextrq $1, %xmm2, %rdx |
| ; AVX512-NEXT: vextracti64x4 $1, %zmm0, %ymm2 |
| ; AVX512-NEXT: vmovq %xmm2, %rsi |
| ; AVX512-NEXT: vpextrq $1, %xmm2, %rdi |
| ; AVX512-NEXT: vextracti128 $1, %ymm2, %xmm2 |
| ; AVX512-NEXT: vmovq %xmm2, %r8 |
| ; AVX512-NEXT: vpextrq $1, %xmm2, %r9 |
| ; AVX512-NEXT: vpextrq $1, %xmm1, %rbx |
| ; AVX512-NEXT: vextracti128 $1, %ymm1, %xmm2 |
| ; AVX512-NEXT: vmovq %xmm2, %rbp |
| ; AVX512-NEXT: vpextrq $1, %xmm2, %r13 |
| ; AVX512-NEXT: vextracti64x4 $1, %zmm1, %ymm2 |
| ; AVX512-NEXT: vmovq %xmm2, %r15 |
| ; AVX512-NEXT: vpextrq $1, %xmm2, %r14 |
| ; AVX512-NEXT: vextracti128 $1, %ymm2, %xmm2 |
| ; AVX512-NEXT: vmovq %xmm2, %r11 |
| ; AVX512-NEXT: vpextrq $1, %xmm2, %r10 |
| ; AVX512-NEXT: xorl %r12d, %r12d |
| ; AVX512-NEXT: subq %r10, %r9 |
| ; AVX512-NEXT: movl $0, %r10d |
| ; AVX512-NEXT: sbbq %r10, %r10 |
| ; AVX512-NEXT: subq %r11, %r8 |
| ; AVX512-NEXT: movl $0, %r11d |
| ; AVX512-NEXT: sbbq %r11, %r11 |
| ; AVX512-NEXT: subq %r14, %rdi |
| ; AVX512-NEXT: movl $0, %r14d |
| ; AVX512-NEXT: sbbq %r14, %r14 |
| ; AVX512-NEXT: subq %r15, %rsi |
| ; AVX512-NEXT: movl $0, %r15d |
| ; AVX512-NEXT: sbbq %r15, %r15 |
| ; AVX512-NEXT: subq %r13, %rdx |
| ; AVX512-NEXT: movl $0, %r13d |
| ; AVX512-NEXT: sbbq %r13, %r13 |
| ; AVX512-NEXT: subq %rbp, %rax |
| ; AVX512-NEXT: movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill |
| ; AVX512-NEXT: movl $0, %ebp |
| ; AVX512-NEXT: sbbq %rbp, %rbp |
| ; AVX512-NEXT: subq %rbx, %rcx |
| ; AVX512-NEXT: movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill |
| ; AVX512-NEXT: movl $0, %eax |
| ; AVX512-NEXT: sbbq %rax, %rax |
| ; AVX512-NEXT: vmovq %xmm0, %rbx |
| ; AVX512-NEXT: vmovq %xmm1, %rcx |
| ; AVX512-NEXT: subq %rcx, %rbx |
| ; AVX512-NEXT: sbbq %r12, %r12 |
| ; AVX512-NEXT: sarq $63, %r12 |
| ; AVX512-NEXT: xorq %r12, %rbx |
| ; AVX512-NEXT: subq %r12, %rbx |
| ; AVX512-NEXT: sarq $63, %rax |
| ; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload |
| ; AVX512-NEXT: xorq %rax, %rcx |
| ; AVX512-NEXT: subq %rax, %rcx |
| ; AVX512-NEXT: movq %rcx, %rax |
| ; AVX512-NEXT: sarq $63, %rbp |
| ; AVX512-NEXT: movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload |
| ; AVX512-NEXT: xorq %rbp, %rcx |
| ; AVX512-NEXT: subq %rbp, %rcx |
| ; AVX512-NEXT: sarq $63, %r13 |
| ; AVX512-NEXT: xorq %r13, %rdx |
| ; AVX512-NEXT: subq %r13, %rdx |
| ; AVX512-NEXT: sarq $63, %r15 |
| ; AVX512-NEXT: xorq %r15, %rsi |
| ; AVX512-NEXT: subq %r15, %rsi |
| ; AVX512-NEXT: sarq $63, %r14 |
| ; AVX512-NEXT: xorq %r14, %rdi |
| ; AVX512-NEXT: subq %r14, %rdi |
| ; AVX512-NEXT: sarq $63, %r11 |
| ; AVX512-NEXT: xorq %r11, %r8 |
| ; AVX512-NEXT: subq %r11, %r8 |
| ; AVX512-NEXT: sarq $63, %r10 |
| ; AVX512-NEXT: xorq %r10, %r9 |
| ; AVX512-NEXT: subq %r10, %r9 |
| ; AVX512-NEXT: vmovq %r9, %xmm0 |
| ; AVX512-NEXT: vmovq %r8, %xmm1 |
| ; AVX512-NEXT: vmovq %rdi, %xmm2 |
| ; AVX512-NEXT: vmovq %rsi, %xmm3 |
| ; AVX512-NEXT: vmovq %rdx, %xmm4 |
| ; AVX512-NEXT: vmovq %rcx, %xmm5 |
| ; AVX512-NEXT: vmovq %rax, %xmm6 |
| ; AVX512-NEXT: vmovq %rbx, %xmm7 |
| ; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0] |
| ; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm3[0],xmm2[0] |
| ; AVX512-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0 |
| ; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm1 = xmm5[0],xmm4[0] |
| ; AVX512-NEXT: vpunpcklqdq {{.*#+}} xmm2 = xmm7[0],xmm6[0] |
| ; AVX512-NEXT: vinserti128 $1, %xmm1, %ymm2, %ymm1 |
| ; AVX512-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 |
| ; AVX512-NEXT: popq %rbx |
| ; AVX512-NEXT: popq %r12 |
| ; AVX512-NEXT: popq %r13 |
| ; AVX512-NEXT: popq %r14 |
| ; AVX512-NEXT: popq %r15 |
| ; AVX512-NEXT: popq %rbp |
| ; AVX512-NEXT: retq |
| %aext = zext <8 x i64> %a to <8 x i128> |
| %bext = zext <8 x i64> %b to <8 x i128> |
| %sub = sub <8 x i128> %aext, %bext |
| %abs = call <8 x i128> @llvm.abs.v8i128(<8 x i128> %sub, i1 true) |
| %trunc = trunc <8 x i128> %abs to <8 x i64> |
| ret <8 x i64> %trunc |
| } |
| |
| ; |
| ; sub(umax(a,b),umin(a,b)) -> abdu(a,b) |
| ; |
| |
| define <64 x i8> @abd_minmax_v64i8(<64 x i8> %a, <64 x i8> %b) nounwind { |
| ; AVX512BW-LABEL: abd_minmax_v64i8: |
| ; AVX512BW: # %bb.0: |
| ; AVX512BW-NEXT: vpminub %zmm1, %zmm0, %zmm2 |
| ; AVX512BW-NEXT: vpmaxub %zmm1, %zmm0, %zmm0 |
| ; AVX512BW-NEXT: vpsubb %zmm2, %zmm0, %zmm0 |
| ; AVX512BW-NEXT: retq |
| ; |
| ; AVX512DQ-LABEL: abd_minmax_v64i8: |
| ; AVX512DQ: # %bb.0: |
| ; AVX512DQ-NEXT: vpminub %ymm1, %ymm0, %ymm2 |
| ; AVX512DQ-NEXT: vextracti64x4 $1, %zmm1, %ymm3 |
| ; AVX512DQ-NEXT: vextracti64x4 $1, %zmm0, %ymm4 |
| ; AVX512DQ-NEXT: vpminub %ymm3, %ymm4, %ymm5 |
| ; AVX512DQ-NEXT: vpmaxub %ymm1, %ymm0, %ymm0 |
| ; AVX512DQ-NEXT: vpsubb %ymm2, %ymm0, %ymm0 |
| ; AVX512DQ-NEXT: vpmaxub %ymm3, %ymm4, %ymm1 |
| ; AVX512DQ-NEXT: vpsubb %ymm5, %ymm1, %ymm1 |
| ; AVX512DQ-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0 |
| ; AVX512DQ-NEXT: retq |
| %min = call <64 x i8> @llvm.umin.v64i8(<64 x i8> %a, <64 x i8> %b) |
| %max = call <64 x i8> @llvm.umax.v64i8(<64 x i8> %a, <64 x i8> %b) |
| %sub = sub <64 x i8> %max, %min |
| ret <64 x i8> %sub |
| } |
| |
| define <32 x i16> @abd_minmax_v32i16(<32 x i16> %a, <32 x i16> %b) nounwind { |
| ; AVX512BW-LABEL: abd_minmax_v32i16: |
| ; AVX512BW: # %bb.0: |
| ; AVX512BW-NEXT: vpminuw %zmm1, %zmm0, %zmm2 |
| ; AVX512BW-NEXT: vpmaxuw %zmm1, %zmm0, %zmm0 |
| ; AVX512BW-NEXT: vpsubw %zmm2, %zmm0, %zmm0 |
| ; AVX512BW-NEXT: retq |
| ; |
| ; AVX512DQ-LABEL: abd_minmax_v32i16: |
| ; AVX512DQ: # %bb.0: |
| ; AVX512DQ-NEXT: vpminuw %ymm1, %ymm0, %ymm2 |
| ; AVX512DQ-NEXT: vextracti64x4 $1, %zmm1, %ymm3 |
| ; AVX512DQ-NEXT: vextracti64x4 $1, %zmm0, %ymm4 |
| ; AVX512DQ-NEXT: vpminuw %ymm3, %ymm4, %ymm5 |
| ; AVX512DQ-NEXT: vpmaxuw %ymm1, %ymm0, %ymm0 |
| ; AVX512DQ-NEXT: vpsubw %ymm2, %ymm0, %ymm0 |
| ; AVX512DQ-NEXT: vpmaxuw %ymm3, %ymm4, %ymm1 |
| ; AVX512DQ-NEXT: vpsubw %ymm5, %ymm1, %ymm1 |
| ; AVX512DQ-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0 |
| ; AVX512DQ-NEXT: retq |
| %min = call <32 x i16> @llvm.umin.v32i16(<32 x i16> %a, <32 x i16> %b) |
| %max = call <32 x i16> @llvm.umax.v32i16(<32 x i16> %a, <32 x i16> %b) |
| %sub = sub <32 x i16> %max, %min |
| ret <32 x i16> %sub |
| } |
| |
| define <16 x i32> @abd_minmax_v16i32(<16 x i32> %a, <16 x i32> %b) nounwind { |
| ; AVX512-LABEL: abd_minmax_v16i32: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vpminud %zmm1, %zmm0, %zmm2 |
| ; AVX512-NEXT: vpmaxud %zmm1, %zmm0, %zmm0 |
| ; AVX512-NEXT: vpsubd %zmm2, %zmm0, %zmm0 |
| ; AVX512-NEXT: retq |
| %min = call <16 x i32> @llvm.umin.v16i32(<16 x i32> %a, <16 x i32> %b) |
| %max = call <16 x i32> @llvm.umax.v16i32(<16 x i32> %a, <16 x i32> %b) |
| %sub = sub <16 x i32> %max, %min |
| ret <16 x i32> %sub |
| } |
| |
| define <8 x i64> @abd_minmax_v8i64(<8 x i64> %a, <8 x i64> %b) nounwind { |
| ; AVX512-LABEL: abd_minmax_v8i64: |
| ; AVX512: # %bb.0: |
| ; AVX512-NEXT: vpminuq %zmm1, %zmm0, %zmm2 |
| ; AVX512-NEXT: vpmaxuq %zmm1, %zmm0, %zmm0 |
| ; AVX512-NEXT: vpsubq %zmm2, %zmm0, %zmm0 |
| ; AVX512-NEXT: retq |
| %min = call <8 x i64> @llvm.umin.v8i64(<8 x i64> %a, <8 x i64> %b) |
| %max = call <8 x i64> @llvm.umax.v8i64(<8 x i64> %a, <8 x i64> %b) |
| %sub = sub <8 x i64> %max, %min |
| ret <8 x i64> %sub |
| } |
| |
| declare <64 x i8> @llvm.abs.v64i8(<64 x i8>, i1) |
| declare <32 x i16> @llvm.abs.v32i16(<32 x i16>, i1) |
| declare <16 x i32> @llvm.abs.v16i32(<16 x i32>, i1) |
| declare <8 x i64> @llvm.abs.v8i64(<8 x i64>, i1) |
| declare <16 x i64> @llvm.abs.v16i64(<16 x i64>, i1) |
| declare <32 x i64> @llvm.abs.v32i64(<32 x i64>, i1) |
| declare <64 x i64> @llvm.abs.v64i64(<64 x i64>, i1) |
| declare <8 x i128> @llvm.abs.v8i128(<8 x i128>, i1) |
| |
| declare <64 x i8> @llvm.umax.v64i8(<64 x i8>, <64 x i8>) |
| declare <32 x i16> @llvm.umax.v32i16(<32 x i16>, <32 x i16>) |
| declare <16 x i32> @llvm.umax.v16i32(<16 x i32>, <16 x i32>) |
| declare <8 x i64> @llvm.umax.v8i64(<8 x i64>, <8 x i64>) |
| |
| declare <64 x i8> @llvm.umin.v64i8(<64 x i8>, <64 x i8>) |
| declare <32 x i16> @llvm.umin.v32i16(<32 x i16>, <32 x i16>) |
| declare <16 x i32> @llvm.umin.v16i32(<16 x i32>, <16 x i32>) |
| declare <8 x i64> @llvm.umin.v8i64(<8 x i64>, <8 x i64>) |
