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 "ssse3-check.h"
       7  #endif
       8  
       9  #ifndef TEST
      10  #define TEST ssse3_test
      11  #endif
      12  
      13  #include CHECK_H
      14  
      15  #include "ssse3-vals.h"
      16  
      17  #include <tmmintrin.h>
      18  
      19  #ifndef __AVX__
      20  /* Test the 64-bit form */
      21  static void
      22  ssse3_test_phsubsw (__m64 *i1, __m64 *i2, __m64 *r)
      23  {
      24    *(__m64 *) r = _mm_hsubs_pi16 (*i1, *i2);
      25    _mm_empty ();
      26  }
      27  #endif
      28  
      29  /* Test the 128-bit form */
      30  static void
      31  ssse3_test_phsubsw128 (__m128i *i1, __m128i *i2, __m128i *r)
      32  {
      33    /* Assumes incoming pointers are 16-byte aligned */
      34    *r = _mm_hsubs_epi16 (*i1, *i2);
      35  }
      36  
      37  static short
      38  signed_saturate_to_word (int x)
      39  {
      40    if (x > (int )0x7fff)
      41      return 0x7fff;
      42  
      43    if (x < (int) 0xffff8000)
      44      return 0x8000;
      45  
      46    return (short)x;
      47  }
      48  
      49  /* Routine to manually compute the results */
      50  static void
      51  compute_correct_result (short *i1, short *i2, short *r)
      52  {
      53    int i;
      54  
      55    for (i = 0; i < 4; i++)
      56      r[i] = signed_saturate_to_word (i1[2 * i] - i1[2 * i + 1]);
      57  
      58    for (i = 0; i < 4; i++)
      59      r[i + 4] = signed_saturate_to_word (i2[2 * i] - i2[2 * i + 1]);
      60  }
      61  
      62  static void
      63  TEST (void)
      64  {
      65    int i;
      66    union data r __attribute__ ((aligned(16)));
      67    union data ck;
      68    int fail = 0;
      69  
      70    for (i = 0; i < ARRAY_SIZE (vals) - 1; i++)
      71      {
      72        /* Manually compute the result */
      73        compute_correct_result (&vals[i + 0].h[0], &vals[i + 1].h[0], &ck.h[0]);
      74  
      75  #ifndef __AVX__
      76        /* Run the 64-bit tests */
      77        ssse3_test_phsubsw (&vals[i + 0].ll[0], &vals[i + 0].ll[1], &r.ll[0]);
      78        ssse3_test_phsubsw (&vals[i + 1].ll[0], &vals[i + 1].ll[1], &r.ll[1]);
      79        fail += chk_128 (ck.m[0], r.m[0]);
      80  #endif
      81  
      82        /* Run the 128-bit tests */
      83        ssse3_test_phsubsw128 (&vals[i + 0].m[0], &vals[i + 1].m[0], &r.m[0]);
      84        fail += chk_128 (ck.m[0], r.m[0]);
      85      }
      86  
      87    if (fail != 0)
      88      abort ();
      89  }