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_nmsubps ()
      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_nmsubpd ()
      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_nmsubps ()
      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_nmsubpd ()
      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  static int
      76  check_nmsubss ()
      77  {
      78    int i, j, check_fails = 0;
      79    for (i = 0; i < NUM * 4; i = i + 4)
      80      {
      81        res.f[i] = - (src1.f[i] * src2.f[i]) - src3.f[i];
      82        if (dst.f[i] != res.f[i]) 
      83  	check_fails++;
      84      }	
      85    return check_fails;
      86  }
      87  
      88  static int
      89  check_nmsubsd ()
      90  {
      91    int i, j, check_fails = 0;
      92    for (i = 0; i < NUM * 2; i = i + 2)
      93      {
      94        res.d[i] = - (src1.d[i] * src2.d[i]) - src3.d[i];
      95        if (dst.d[i] != res.d[i]) 
      96  	check_fails++;
      97      }
      98    return check_fails;
      99  }
     100  
     101  static void
     102  fma4_test (void)
     103  {
     104    int i;
     105  
     106    init_nmsubps ();
     107    
     108    for (i = 0; i < NUM; i++)
     109      dst.x[i] = _mm_nmsub_ps (src1.x[i], src2.x[i], src3.x[i]);
     110    
     111    if (check_nmsubps (&dst.x[i], &src1.f[i * 4], &src2.f[i * 4], &src3.f[i * 4])) 
     112      abort ();
     113  
     114    for (i = 0; i < NUM; i++)
     115      dst.x[i] = _mm_nmsub_ss (src1.x[i], src2.x[i], src3.x[i]);
     116    
     117    if (check_nmsubss (&dst.x[i], &src1.f[i * 4], &src2.f[i * 4], &src3.f[i * 4])) 
     118      abort ();
     119  
     120    init_nmsubpd ();
     121    
     122    for (i = 0; i < NUM; i++)
     123      dst.y[i] = _mm_nmsub_pd (src1.y[i], src2.y[i], src3.y[i]);
     124    
     125    if (check_nmsubpd (&dst.y[i], &src1.d[i * 2], &src2.d[i * 2], &src3.d[i * 2])) 
     126      abort ();
     127  
     128    for (i = 0; i < NUM; i++)
     129      dst.y[i] = _mm_nmsub_sd (src1.y[i], src2.y[i], src3.y[i]);
     130    
     131    if (check_nmsubsd (&dst.y[i], &src1.d[i * 2], &src2.d[i * 2], &src3.d[i * 2])) 
     132      abort ();
     133  }