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_movshdup_1
      13  #endif
      14  
      15  #define NO_WARN_X86_INTRINSICS 1
      16  #include <pmmintrin.h>
      17  
      18  static void
      19  sse3_test_movshdup_reg (float *i1, float *r)
      20  {
      21    __m128 t1 = _mm_loadu_ps (i1);
      22    __m128 t2 = _mm_movehdup_ps (t1);
      23  
      24    _mm_storeu_ps (r, t2);
      25  }
      26  
      27  static void
      28  sse3_test_movshdup_reg_subsume (float *i1, float *r)
      29  {
      30    __m128 t1 = _mm_load_ps (i1);
      31    __m128 t2 = _mm_movehdup_ps (t1);
      32  
      33    _mm_storeu_ps (r, t2);
      34  }
      35  
      36  static int
      37  chk_ps (float *v1, float *v2)
      38  {
      39    int i;
      40    int n_fails = 0;
      41  
      42    for (i = 0; i < 4; i++)
      43      if (v1[i] != v2[i])
      44        n_fails += 1;
      45  
      46    return n_fails;
      47  }
      48  
      49  static float p1[4] __attribute__ ((aligned(16)));
      50  static float p2[4];
      51  static float ck[4];
      52  
      53  static float vals[] =
      54    {
      55      100.0,  200.0, 300.0, 400.0, 5.0, -1.0, .345, -21.5,
      56      1100.0, 0.235, 321.3, 53.40, 0.3, 10.0, 42.0, 32.52,
      57      32.6,   123.3, 1.234, 2.156, 0.1, 3.25, 4.75, 32.44,
      58      12.16,  52.34, 64.12, 71.13, -.1, 2.30, 5.12, 3.785,
      59      541.3,  321.4, 231.4, 531.4, 71., 321., 231., -531.,
      60      23.45,  23.45, 23.45, 23.45, 23.45, 23.45, 23.45, 23.45,
      61      23.45,  -1.43, -6.74, 6.345, -20.1, -20.1, -40.1, -40.1,
      62      1.234,  2.345, 3.456, 4.567, 5.678, 6.789, 7.891, 8.912,
      63      -9.32,  -8.41, -7.50, -6.59, -5.68, -4.77, -3.86, -2.95,
      64      9.32,  8.41, 7.50, 6.59, -5.68, -4.77, -3.86, -2.95
      65    };
      66  
      67  //static
      68  void
      69  TEST (void)
      70  {
      71    int i;
      72    int fail = 0;
      73  
      74    for (i = 0; i < sizeof (vals) / sizeof (vals[0]); i += 2)
      75      {
      76        p1[0] = 0.0;
      77        p1[1] = vals[i+0];
      78        p1[2] = 1.0;
      79        p1[3] = vals[i+1];
      80  
      81        ck[0] = p1[1];
      82        ck[1] = p1[1];
      83        ck[2] = p1[3];
      84        ck[3] = p1[3];
      85  
      86        sse3_test_movshdup_reg (p1, p2);
      87  
      88        fail += chk_ps (ck, p2);
      89  
      90        sse3_test_movshdup_reg_subsume (p1, p2);
      91  
      92        fail += chk_ps (ck, p2);
      93      }
      94  
      95    if (fail != 0)
      96      abort ();
      97  }