gcc/testsuite/gcc.dg/vect/bb-slp-24.c

   1 /* { dg-require-effective-target vect_int } */
   2
   3 #include <stdarg.h>
   4 #include "tree-vect.h"
   5
   6 #define A 3
   7 #define N 256
   8
   9 short src[N], dst[N];
  10
  11 void foo (short * __restrict__ dst, short * __restrict__ src, int h,
  12           int stride)
  13 {
  14   int i;
  15   h /= 8;
  16   for (i = 0; i < h; i++)
  17     {
  18       dst[0] += A*src[0];
  19       dst[1] += A*src[1];
  20       dst[2] += A*src[2];
  21       dst[3] += A*src[3];
  22       dst[4] += A*src[4];
  23       dst[5] += A*src[5];
  24       dst[6] += A*src[6];
  25       dst[7] += A*src[7];
  26       dst += stride;
  27       src += stride;
  28       asm volatile ("" ::: "memory");
  29     }
  30 }
  31
  32
  33 int main (void)
  34 {
  35   int i;
  36
  37   check_vect ();
  38
  39   for (i = 0; i < N; i++)
  40     {
  41       dst[i] = 0;
  42       src[i] = i;
  43     }
  44
  45   foo (dst, src, N, 8);
  46
  47 #pragma GCC novector
  48   for (i = 0; i < N; i++)
  49     {
  50       if (dst[i] != A * i)
  51         abort ();
  52     }
  53
  54   return 0;
  55 }
  56
  57 /* Exclude POWER8 (only POWER cpu for which vect_element_align is true)
  58    because loops have vectorized before SLP gets a shot.  */
  59 /* { dg-final { scan-tree-dump-times "optimized: basic block" 1 "slp1" { target { vect_element_align && { ! powerpc*-*-* } } } } } */
  60