gcc/testsuite/gcc.target/aarch64/sve/slp_7_run.c

   1 /* { dg-do run { target aarch64_sve_hw } } */
   2 /* { dg-options "-O2 -ftree-vectorize -ffast-math" } */
   3
   4 #include "slp_7.c"
   5
   6 #define N (54 * 4)
   7
   8 #define HARNESS(TYPE)                                                   \
   9   {                                                                     \
  10     TYPE a[N], b[4] = { 40, 22, 75, 19 };                               \
  11     for (unsigned int i = 0; i < N; ++i)                                \
  12       {                                                                 \
  13         a[i] = i * 2 + i % 5;                                           \
  14         asm volatile ("" ::: "memory");                                 \
  15       }                                                                 \
  16     vec_slp_##TYPE (a, b, N / 4);                                       \
  17     TYPE x0 = 40;                                                       \
  18     TYPE x1 = 22;                                                       \
  19     TYPE x2 = 75;                                                       \
  20     TYPE x3 = 19;                                                       \
  21     for (unsigned int i = 0; i < N; i += 4)                             \
  22       {                                                                 \
  23         x0 += a[i];                                                     \
  24         x1 += a[i + 1];                                                 \
  25         x2 += a[i + 2];                                                 \
  26         x3 += a[i + 3];                                                 \
  27         asm volatile ("" ::: "memory");                                 \
  28       }                                                                 \
  29     /* _Float16 isn't precise enough for this.  */                      \
  30     if ((TYPE) 0x1000 + 1 != (TYPE) 0x1000                              \
  31         && (x0 != b[0] || x1 != b[1] || x2 != b[2] || x3 != b[3]))      \
  32       __builtin_abort ();                                               \
  33   }
  34
  35 int __attribute__ ((optimize (1)))
  36 main (void)
  37 {
  38   TEST_ALL (HARNESS)
  39 }