(root)/
gcc-13.2.0/
gcc/
testsuite/
gcc.target/
i386/
avxvnniint8-vpdpbsuds-2.c
       1  /* { dg-do run } */
       2  /* { dg-options "-O2 -mavxvnniint8" } */
       3  /* { dg-require-effective-target avxvnniint8 } */
       4  #define AVXVNNIINT8
       5  #ifndef CHECK
       6  #define CHECK "avx-check.h"
       7  #endif
       8  
       9  #ifndef TEST
      10  #define TEST avx_test
      11  #endif
      12  
      13  #include CHECK
      14  
      15  static void
      16  CALC (int *r, int *dst, char *s1, unsigned char *s2, int size)
      17  {
      18    short tempres[32];
      19    for (int i = 0; i < size; i++) {
      20      tempres[i] = (short) s1[i] * (unsigned short) s2[i];
      21    }
      22    for (int i = 0; i < size / 4; i++) {
      23      long long test = (long long) dst[i] + tempres[i * 4] + tempres[i * 4 + 1]
      24  		    + tempres[i * 4 + 2] + tempres[i * 4 + 3];
      25      r[i] = test > 0x7FFFFFFF ? 0x7FFFFFFF : test;
      26    }
      27  }
      28  
      29  void
      30  TEST (void)
      31  {
      32    int i;
      33    union256i_d res_256;
      34    union256i_b src1_256;
      35    union256i_ub src2_256;
      36    int res_ref_256[8];
      37  
      38    for (i = 0; i < 32; i++)
      39      {
      40        int sign = i % 2 ? 1 : -1;
      41        src1_256.a[i] = 10 + 3 * i + sign;
      42        src2_256.a[i] = sign * 10 * i * i;
      43      }
      44  
      45    for (i = 0; i < 8; i++)
      46      res_256.a[i] = 0x7fffffff;
      47  
      48    CALC (res_ref_256, res_256.a, src1_256.a, src2_256.a, 32);
      49    res_256.x = _mm256_dpbsuds_epi32 (res_256.x, src1_256.x, src2_256.x);
      50    if (check_union256i_d (res_256, res_ref_256))
      51      abort ();
      52  
      53    union128i_d res_128;
      54    union128i_b src1_128;
      55    union128i_ub src2_128;
      56    int res_ref_128[4];
      57  
      58    for (i = 0; i < 16; i++)
      59      {
      60        int sign = i % 2 ? 1 : -1;
      61        src1_128.a[i] = 10 + 3 * i * i + sign;
      62        src2_128.a[i] = sign * 10 * i * i;
      63      }
      64  
      65    for (i = 0; i < 4; i++)
      66      res_128.a[i] = 0x7fffffff;
      67  
      68    CALC (res_ref_128, res_128.a, src1_128.a, src2_128.a, 16);
      69    res_128.x = _mm_dpbsuds_epi32 (res_128.x, src1_128.x, src2_128.x);
      70    if (check_union128i_d (res_128, res_ref_128))
      71      abort ();
      72  }