1  /* { dg-do run } */
       2  /* { dg-options "-O3 -mpower8-vector -Wno-psabi" } */
       3  /* { dg-require-effective-target p8vector_hw } */
       4  
       5  #ifndef CHECK_H
       6  #define CHECK_H "sse3-check.h"
       7  #endif
       8  
       9  #include CHECK_H
      10  
      11  #ifndef TEST
      12  #define TEST sse3_test_hsubpd_1
      13  #endif
      14  
      15  #define NO_WARN_X86_INTRINSICS 1
      16  #include <pmmintrin.h>
      17  
      18  static void
      19  sse3_test_hsubpd (double *i1, double *i2, double *r)
      20  {
      21    __m128d t1 = _mm_loadu_pd (i1);
      22    __m128d t2 = _mm_loadu_pd (i2);
      23  
      24    t1 = _mm_hsub_pd (t1, t2);
      25  
      26    _mm_storeu_pd (r, t1);
      27  }
      28  
      29  static void
      30  sse3_test_hsubpd_subsume (double *i1, double *i2, double *r)
      31  {
      32    __m128d t1 = _mm_load_pd (i1);
      33    __m128d t2 = _mm_load_pd (i2);
      34  
      35    t1 = _mm_hsub_pd (t1, t2);
      36  
      37    _mm_storeu_pd (r, t1);
      38  }
      39  
      40  static int
      41  chk_pd (double *v1, double *v2)
      42  {
      43    int i;
      44    int n_fails = 0;
      45  
      46    for (i = 0; i < 2; i++)
      47      if (v1[i] != v2[i])
      48        n_fails += 1;
      49  
      50    return n_fails;
      51  }
      52  
      53  static double p1[2] __attribute__ ((aligned(16)));
      54  static double p2[2] __attribute__ ((aligned(16)));
      55  static double p3[2];
      56  static double ck[2];
      57  
      58  static double vals[] =
      59    {
      60      100.0,  200.0, 300.0, 400.0, 5.0, -1.0, .345, -21.5,
      61      1100.0, 0.235, 321.3, 53.40, 0.3, 10.0, 42.0, 32.52,
      62      32.6,   123.3, 1.234, 2.156, 0.1, 3.25, 4.75, 32.44,
      63      12.16,  52.34, 64.12, 71.13, -.1, 2.30, 5.12, 3.785,
      64      541.3,  321.4, 231.4, 531.4, 71., 321., 231., -531.,
      65      23.45,  23.45, 23.45, 23.45, 23.45, 23.45, 23.45, 23.45,
      66      23.45,  -1.43, -6.74, 6.345, -20.1, -20.1, -40.1, -40.1,
      67      1.234,  2.345, 3.456, 4.567, 5.678, 6.789, 7.891, 8.912,
      68      -9.32,  -8.41, -7.50, -6.59, -5.68, -4.77, -3.86, -2.95,
      69      9.32,  8.41, 7.50, 6.59, -5.68, -4.77, -3.86, -2.95
      70    };
      71  
      72  //static
      73  void TEST (void)
      74  {
      75    int i;
      76    int fail = 0;
      77  
      78    for (i = 0; i < sizeof (vals) / sizeof (vals[0]); i += 4)
      79      {
      80        p1[0] = vals[i + 0];
      81        p1[1] = vals[i + 1];
      82  
      83        p2[0] = vals[i + 2];
      84        p2[1] = vals[i + 3];
      85  
      86        ck[0] = p1[0] - p1[1];
      87        ck[1] = p2[0] - p2[1];
      88  
      89        sse3_test_hsubpd (p1, p2, p3);
      90  
      91        fail += chk_pd (ck, p3);
      92  
      93        sse3_test_hsubpd_subsume (p1, p2, p3);
      94  
      95        fail += chk_pd (ck, p3);
      96      }
      97  
      98    if (fail != 0)
      99      abort ();
     100  }