1  /* { dg-do run { target avx512fp16 } } */
       2  /* { dg-options "-O2 -mavx512fp16 -mavx512dq" } */
       3  
       4  
       5  #define AVX512FP16
       6  #include "avx512fp16-helper.h"
       7  
       8  #define N_ELEMS (AVX512F_LEN / 16)
       9  
      10  void NOINLINE
      11  EMULATE(fmsub_ph) (V512 * dest, V512 op1, V512 op2,
      12                   __mmask32 k, int zero_mask)
      13  {
      14      V512 v1, v2, v3, v4, v5, v6, v7, v8; 
      15      int i;
      16      __mmask16 m1, m2;
      17      
      18      m1 = k & 0xffff;
      19      m2 = (k >> 16) & 0xffff;
      20   
      21      unpack_ph_2twops(op1, &v1, &v2);
      22      unpack_ph_2twops(op2, &v3, &v4);
      23      unpack_ph_2twops(*dest, &v7, &v8);
      24  
      25      for (i = 0; i < 16; i++) {
      26          if (((1 << i) & m1) == 0) {
      27              if (zero_mask) {
      28                  v5.f32[i] = 0;
      29              }
      30              else {
      31                  v5.u32[i] = v7.u32[i];
      32              }
      33          }
      34          else {
      35             v5.f32[i] = v1.f32[i] * v3.f32[i] - v7.f32[i];
      36          }
      37  
      38          if (((1 << i) & m2) == 0) {
      39              if (zero_mask) {
      40                 v6.f32[i] = 0;
      41              }
      42              else {
      43                  v6.u32[i] = v8.u32[i];
      44              }
      45          }
      46          else {
      47              v6.f32[i] = v2.f32[i] * v4.f32[i] - v8.f32[i];
      48          }
      49  
      50      }
      51      *dest = pack_twops_2ph(v5, v6);
      52  }
      53  
      54  void NOINLINE
      55  EMULATE(m_fmsub_ph) (V512 * dest, V512 op1, V512 op2,
      56                     __mmask32 k, int zero_mask)
      57  {
      58      V512 v1, v2, v3, v4, v5, v6, v7, v8; 
      59      int i;
      60      __mmask16 m1, m2;
      61      
      62      m1 = k & 0xffff;
      63      m2 = (k >> 16) & 0xffff;
      64   
      65      unpack_ph_2twops(op1, &v1, &v2);
      66      unpack_ph_2twops(op2, &v3, &v4);
      67      unpack_ph_2twops(*dest, &v7, &v8);
      68  
      69      for (i = 0; i < 16; i++) {
      70          if (((1 << i) & m1) == 0) {
      71              if (zero_mask) {
      72                  v5.f32[i] = 0;
      73              }
      74              else {
      75                  v5.u32[i] = v7.u32[i];
      76              }
      77          }
      78          else {
      79             v5.f32[i] = v7.f32[i] * v1.f32[i] - v3.f32[i];
      80          }
      81  
      82          if (((1 << i) & m2) == 0) {
      83              if (zero_mask) {
      84                 v6.f32[i] = 0;
      85              }
      86              else {
      87                  v6.u32[i] = v8.u32[i];
      88              }
      89          }
      90          else {
      91              v6.f32[i] = v8.f32[i] * v2.f32[i] - v4.f32[i];
      92          }
      93  
      94      }
      95      *dest = pack_twops_2ph(v5, v6);
      96  }
      97  
      98  void
      99  TEST (void)
     100  {
     101    V512 res;
     102    V512 exp;
     103  
     104    init_src();
     105  
     106    init_dest(&res, &exp);
     107    EMULATE(fmsub_ph)(&exp, src1, src2,  NET_MASK, 0);
     108    HF(res) = INTRINSIC (_fmsub_ph) (HF(src1), HF(src2), HF(res));
     109    CHECK_RESULT (&res, &exp, N_ELEMS, _fmsub_ph);
     110  
     111    init_dest(&res, &exp);
     112    EMULATE(m_fmsub_ph)(&exp, src1, src2,  MASK_VALUE, 0);
     113    HF(res) = INTRINSIC (_mask_fmsub_ph) (HF(res), MASK_VALUE,
     114  					HF(src1), HF(src2));
     115    CHECK_RESULT (&res, &exp, N_ELEMS, _mask_fmsub_ph);
     116  
     117    init_dest(&res, &exp);
     118    EMULATE(fmsub_ph)(&exp, src1, src2,  MASK_VALUE, 0);
     119    HF(res) = INTRINSIC (_mask3_fmsub_ph) (HF(src1), HF(src2), HF(res), MASK_VALUE);
     120    CHECK_RESULT (&res, &exp, N_ELEMS, _mask3_fmsub_ph);
     121  
     122    init_dest(&res, &exp);
     123    EMULATE(fmsub_ph)(&exp, src1, src2,  ZMASK_VALUE, 1);
     124    HF(res) = INTRINSIC (_maskz_fmsub_ph) (ZMASK_VALUE, HF(src1), HF(src2), HF(res));
     125    CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_fmsub_ph);
     126  
     127  #if AVX512F_LEN == 512
     128    init_dest(&res, &exp);
     129    EMULATE(fmsub_ph)(&exp, src1, src2,  NET_MASK, 0);
     130    HF(res) = INTRINSIC (_fmsub_round_ph) (HF(src1), HF(src2), HF(res), _ROUND_NINT);
     131    CHECK_RESULT (&res, &exp, N_ELEMS, _fmsub_ph);
     132  
     133    init_dest(&res, &exp);
     134    EMULATE(m_fmsub_ph)(&exp, src1, src2,  MASK_VALUE, 0);
     135    HF(res) = INTRINSIC (_mask_fmsub_round_ph) (HF(res), MASK_VALUE,
     136  					HF(src1), HF(src2), _ROUND_NINT);
     137    CHECK_RESULT (&res, &exp, N_ELEMS, _mask_fmsub_ph);
     138  
     139    EMULATE(fmsub_ph)(&exp, src1, src2,  MASK_VALUE, 0);
     140    HF(res) = INTRINSIC (_mask3_fmsub_round_ph) (HF(src1), HF(src2),
     141  					 HF(res), MASK_VALUE, _ROUND_NINT);
     142    CHECK_RESULT (&res, &exp, N_ELEMS, _mask3_fmsub_ph);
     143  
     144    init_dest(&res, &exp);
     145    EMULATE(fmsub_ph)(&exp, src1, src2,  ZMASK_VALUE, 1);
     146    HF(res) = INTRINSIC (_maskz_fmsub_round_ph) (ZMASK_VALUE, HF(src1),
     147  					 HF(src2), HF(res), _ROUND_NINT);
     148    CHECK_RESULT (&res, &exp, N_ELEMS, _maskz_fmsub_ph);
     149  #endif
     150  
     151    if (n_errs != 0) {
     152      abort ();
     153    }
     154  }
     155