1  /* { dg-do run } */
       2  /* { dg-require-effective-target fma4 } */
       3  /* { dg-options "-O0 -mfma4" } */
       4  
       5  #include "fma4-check.h"
       6  
       7  #include <x86intrin.h>
       8  #include <string.h>
       9  
      10  #define NUM 20
      11  
      12  union
      13  {
      14    __m128 x[NUM];
      15    float f[NUM * 4];
      16    __m128d y[NUM];
      17    double d[NUM * 2];
      18  } dst, res, src1, src2, src3;
      19  
      20  /* Note that in macc*,msub*,mnmacc* and mnsub* instructions, the intermdediate 
      21     product is not rounded, only the addition is rounded. */
      22  
      23  static void
      24  init_maccps ()
      25  {
      26    int i;
      27    for (i = 0; i < NUM * 4; i++)
      28      {
      29        src1.f[i] = i;
      30        src2.f[i] = i + 10;
      31        src3.f[i] = i + 20;
      32      }
      33  }
      34  
      35  static void
      36  init_maccpd ()
      37  {
      38    int i;
      39    for (i = 0; i < NUM * 4; i++)
      40      {
      41        src1.d[i] = i;
      42        src2.d[i] = i + 10;
      43        src3.d[i] = i + 20;
      44      }
      45  }
      46  
      47  static int
      48  check_maccps ()
      49  {
      50    int i, j, check_fails = 0;
      51    for (i = 0; i < NUM * 4; i = i + 4)
      52      for (j = 0; j < 4; j++)
      53        {
      54  	res.f[i + j] = (src1.f[i + j] * src2.f[i + j]) + src3.f[i + j];
      55  	if (dst.f[i + j] != res.f[i + j]) 
      56  	  check_fails++;
      57        }
      58    return check_fails;
      59  }
      60  
      61  static int
      62  check_maccpd ()
      63  {
      64    int i, j, check_fails = 0;
      65    for (i = 0; i < NUM * 2; i = i + 2)
      66      for (j = 0; j < 2; j++)
      67        {
      68  	res.d[i + j] = (src1.d[i + j] * src2.d[i + j]) + src3.d[i + j];
      69  	if (dst.d[i + j] != res.d[i + j]) 
      70  	  check_fails++;
      71        }
      72    return check_fails;
      73  }
      74  
      75  
      76  static int
      77  check_maccss ()
      78  {
      79    int i, j, check_fails = 0;
      80    for (i = 0; i < NUM * 4; i= i + 4)
      81      {
      82        res.f[i] = (src1.f[i] * src2.f[i]) + src3.f[i];
      83        if (dst.f[i] != res.f[i]) 
      84  	check_fails++;
      85      }	
      86    return check_fails;
      87  }
      88  
      89  static int
      90  check_maccsd ()
      91  {
      92    int i, j, check_fails = 0;
      93    for (i = 0; i < NUM * 2; i = i + 2)
      94      {
      95        res.d[i] = (src1.d[i] * src2.d[i]) + src3.d[i];
      96        if (dst.d[i] != res.d[i]) 
      97  	check_fails++;
      98      }
      99    return check_fails;
     100  }
     101  
     102  static void
     103  fma4_test (void)
     104  {
     105    int i;
     106  
     107    init_maccps ();
     108    
     109    for (i = 0; i < NUM; i++)
     110      dst.x[i] = _mm_macc_ps (src1.x[i], src2.x[i], src3.x[i]);
     111  
     112    if (check_maccps ()) 
     113      abort ();
     114  
     115    for (i = 0; i < NUM; i++)
     116      dst.x[i] = _mm_macc_ss (src1.x[i], src2.x[i], src3.x[i]);
     117    
     118    if (check_maccss ()) 
     119      abort ();
     120  
     121    init_maccpd ();
     122    
     123    for (i = 0; i < NUM; i++)
     124      dst.y[i] = _mm_macc_pd (src1.y[i], src2.y[i], src3.y[i]);
     125    
     126    if (check_maccpd ()) 
     127      abort ();
     128  
     129    for (i = 0; i < NUM; i++)
     130      dst.y[i] = _mm_macc_sd (src1.y[i], src2.y[i], src3.y[i]);
     131    
     132    if (check_maccsd ()) 
     133      abort ();
     134  }