gcc/testsuite/gcc.target/i386/avx512bf16vl-vdpbf16ps-1.c

   1 /* { dg-do compile } */
   2 /* { dg-options "-mavx512bf16 -mavx512vl -O2" } */
   3 /* { dg-final { scan-assembler-times "vdpbf16ps\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\[^\{\n\]*%ymm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
   4 /* { dg-final { scan-assembler-times "vdpbf16ps\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\{%k\[0-9\]\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 2 } } */
   5 /* { dg-final { scan-assembler-times "vdpbf16ps\[ \\t\]+\[^\{\n\]*%ymm\[0-9\]+\[^\n\r]*%ymm\[0-9\]+\[^\{\n\]*%ymm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
   6 /* { dg-final { scan-assembler-times "vdpbf16ps\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\{\n\]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 1 } } */
   7 /* { dg-final { scan-assembler-times "vdpbf16ps\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 2 } } */
   8 /* { dg-final { scan-assembler-times "vdpbf16ps\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\{\n\]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 1 } } */
   9
  10 #include <immintrin.h>
  11
  12 volatile __m256 res1;
  13 volatile __m256bh x1, x2;
  14 volatile __m128 res2;
  15 volatile __m128bh x3, x4;
  16 volatile __mmask8 m8;
  17
  18 void extern
  19 avx512bf16_test (void)
  20 {
  21   res1 = _mm256_dpbf16_ps (res1, x1, x2);
  22   res1 = _mm256_mask_dpbf16_ps (res1, m8, x1, x2);
  23   res1 = _mm256_maskz_dpbf16_ps (m8, res1, x1, x2);
  24
  25   res2 = _mm_dpbf16_ps (res2, x3, x4);
  26   res2 = _mm_mask_dpbf16_ps (res2, m8, x3, x4);
  27   res2 = _mm_maskz_dpbf16_ps (m8, res2, x3, x4);
  28 }