gcc/testsuite/gcc.target/i386/gfni-2.c

   1 /* { dg-do compile } */
   2 /* { dg-options "-mgfni -mavx512bw -mavx512vl -O2" } */
   3 /* { dg-final { scan-assembler-times "vgf2p8affineinvqb\[ \\t\]+\[^\{\n\]*\\\$3\[^\n\r]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
   4 /* { dg-final { scan-assembler-times "vgf2p8affineinvqb\[ \\t\]+\[^\{\n\]*\\\$3\[^\n\r]*%ymm\[0-9\]+\[^\\n\\r]*%ymm\[0-9\]+\[^\\n\\r\]*%ymm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
   5 /* { dg-final { scan-assembler-times "vgf2p8affineinvqb\[ \\t\]+\[^\{\n\]*\\\$3\[^\n\r]*%ymm\[0-9\]+\[^\\n\\r]*%ymm\[0-9\]+\[^\\n\\r\]*%ymm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)" 1 } } */
   6 /* { dg-final { scan-assembler-times "vgf2p8affineinvqb\[ \\t\]+\[^\{\n\]*\\\$3\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
   7 /* { dg-final { scan-assembler-times "vgf2p8affineinvqb\[ \\t\]+\[^\{\n\]*\\\$3\[^\n\r]*%xmm\[0-9\]+\[^\\n\\r]*%xmm\[0-9\]+\[^\\n\\r\]*%xmm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
   8 /* { dg-final { scan-assembler-times "vgf2p8affineinvqb\[ \\t\]+\[^\{\n\]*\\\$3\[^\n\r]*%xmm\[0-9\]+\[^\\n\\r]*%xmm\[0-9\]+\[^\\n\\r\]*%xmm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)" 1 } } */
   9 /* { dg-final { scan-assembler-times "vgf2p8affineqb\[ \\t\]+\[^\{\n\]*\\\$3\[^\n\r]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
  10 /* { dg-final { scan-assembler-times "vgf2p8affineqb\[ \\t\]+\[^\{\n\]*\\\$3\[^\n\r]*%ymm\[0-9\]+\[^\\n\\r]*%ymm\[0-9\]+\[^\\n\\r\]*%ymm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
  11 /* { dg-final { scan-assembler-times "vgf2p8affineqb\[ \\t\]+\[^\{\n\]*\\\$3\[^\n\r]*%ymm\[0-9\]+\[^\\n\\r]*%ymm\[0-9\]+\[^\\n\\r\]*%ymm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)" 1 } } */
  12 /* { dg-final { scan-assembler-times "vgf2p8affineqb\[ \\t\]+\[^\{\n\]*\\\$3\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
  13 /* { dg-final { scan-assembler-times "vgf2p8affineqb\[ \\t\]+\[^\{\n\]*\\\$3\[^\n\r]*%xmm\[0-9\]+\[^\\n\\r]*%xmm\[0-9\]+\[^\\n\\r\]*%xmm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
  14 /* { dg-final { scan-assembler-times "vgf2p8affineqb\[ \\t\]+\[^\{\n\]*\\\$3\[^\n\r]*%xmm\[0-9\]+\[^\\n\\r]*%xmm\[0-9\]+\[^\\n\\r\]*%xmm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)" 1 } } */
  15 /* { dg-final { scan-assembler-times "vgf2p8mulb\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
  16 /* { dg-final { scan-assembler-times "vgf2p8mulb\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\\n\\r]*%ymm\[0-9\]+\[^\\n\\r\]*%ymm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
  17 /* { dg-final { scan-assembler-times "vgf2p8mulb\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\\n\\r]*%ymm\[0-9\]+\[^\\n\\r\]*%ymm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)" 1 } } */
  18 /* { dg-final { scan-assembler-times "vgf2p8mulb\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
  19 /* { dg-final { scan-assembler-times "vgf2p8mulb\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\\n\\r]*%xmm\[0-9\]+\[^\\n\\r\]*%xmm\[0-9\]+\{%k\[1-7\]\}(?:\n|\[ \\t\]+#)" 1 } } */
  20 /* { dg-final { scan-assembler-times "vgf2p8mulb\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\\n\\r]*%xmm\[0-9\]+\[^\\n\\r\]*%xmm\[0-9\]+\{%k\[1-7\]\}\{z\}(?:\n|\[ \\t\]+#)" 1 } } */
  21
  22 #include <x86intrin.h>
  23
  24 int *p;
  25 volatile __m256i x3, x4;
  26 volatile __m128i x5, x6;
  27 volatile __mmask32 m32;
  28 volatile __mmask16 m16;
  29
  30 void extern
  31 avx512vl_test (void)
  32 {
  33     x3 = _mm256_gf2p8affineinv_epi64_epi8(x3, x4, 3);
  34     x3 = _mm256_mask_gf2p8affineinv_epi64_epi8(x3, m32, x4, x3, 3);
  35     x3 = _mm256_maskz_gf2p8affineinv_epi64_epi8(m32, x3, x4, 3);
  36     x5 = _mm_gf2p8affineinv_epi64_epi8(x5, x6, 3);
  37     x5 = _mm_mask_gf2p8affineinv_epi64_epi8(x5, m16, x6, x5, 3);
  38     x5 = _mm_maskz_gf2p8affineinv_epi64_epi8(m16, x5, x6, 3);
  39     x3 = _mm256_gf2p8affine_epi64_epi8(x3, x4, 3);
  40     x3 = _mm256_mask_gf2p8affine_epi64_epi8(x3, m32, x4, x3, 3);
  41     x3 = _mm256_maskz_gf2p8affine_epi64_epi8(m32, x3, x4, 3);
  42     x5 = _mm_gf2p8affine_epi64_epi8(x5, x6, 3);
  43     x5 = _mm_mask_gf2p8affine_epi64_epi8(x5, m16, x6, x5, 3);
  44     x5 = _mm_maskz_gf2p8affine_epi64_epi8(m16, x5, x6, 3);
  45     x3 = _mm256_gf2p8mul_epi8(x3, x4);
  46     x3 = _mm256_mask_gf2p8mul_epi8(x3, m32, x4, x3);
  47     x3 = _mm256_maskz_gf2p8mul_epi8(m32, x3, x4);
  48     x5 = _mm_gf2p8mul_epi8(x5, x6);
  49     x5 = _mm_mask_gf2p8mul_epi8(x5, m16, x6, x5);
  50     x5 = _mm_maskz_gf2p8mul_epi8(m16, x5, x6);
  51 }