1  /* { dg-do compile } */
       2  /* { dg-options "-mavx512fp16 -O2" } */
       3  /* { dg-final { scan-assembler-times "vfcmulcsh\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 2 } } */
       4  /* { dg-final { scan-assembler-times "vfcmulcsh\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\[^\{\n\r]*(?:\n|\[ \\t\]+#)" 2 } } */
       5  /* { dg-final { scan-assembler-times "vfcmulcsh\[ \\t\]+\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 2 } } */
       6  /* { dg-final { scan-assembler-times "vfcmulcsh\[ \\t\]+\{rn-sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+(?:\n|\[ \\t\]+#)" 2 } } */
       7  /* { dg-final { scan-assembler-times "vfcmulcsh\[ \\t\]+\{rn-sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 2 } } */
       8  /* { dg-final { scan-assembler-times "vfcmulcsh\[ \\t\]+\{rz-sae\}\[^\{\n\]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\[^\n\r]*%xmm\[0-9\]+\{%k\[0-9\]\}\{z\}\[^\n\r]*(?:\n|\[ \\t\]+#)" 2 } } */
       9  
      10  #include <immintrin.h>
      11  
      12  volatile __m128h res, res1, res2;
      13  volatile __m128h x1, x2, x3;
      14  volatile __mmask8 m8;
      15  
      16  void extern
      17  avx512f_test (void)
      18  {
      19    res = _mm_fcmul_sch (x1, x2);
      20    res1 = _mm_mask_fcmul_sch (res1, m8, x1, x2);
      21    res2 = _mm_maskz_fcmul_sch (m8, x1, x2);
      22    res = _mm_fcmul_round_sch (x1, x2, 8);
      23    res1 = _mm_mask_fcmul_round_sch (res1, m8, x1, x2, 8);
      24    res2 = _mm_maskz_fcmul_round_sch (m8, x1, x2, 11);
      25    
      26    res = _mm_cmul_sch (x1, x2);
      27    res1 = _mm_mask_cmul_sch (res1, m8, x1, x2);
      28    res2 = _mm_maskz_cmul_sch (m8, x1, x2);
      29    res = _mm_cmul_round_sch (x1, x2, 8);
      30    res1 = _mm_mask_cmul_round_sch (res1, m8, x1, x2, 8);
      31    res2 = _mm_maskz_cmul_round_sch (m8, x1, x2, 11);
      32  }