gcc/testsuite/gcc.dg/vect/pr90018.c

   1 /* { dg-require-effective-target vect_double } */
   2
   3 #include "tree-vect.h"
   4
   5 void __attribute__((noinline,noclone))
   6 foo (double *a4, int n)
   7 {
   8   for (int i = 0; i < n; ++i)
   9     {
  10       /* We may not apply interleaving to the group (a), (b) because of (c).
  11          Instead group (d) and (b).  */
  12       double tem1 = a4[i*4] + a4[i*4+n*4] /* (a) */;
  13       double tem2 = a4[i*4+2*n*4+1];
  14       a4[i*4+n*4+1] = tem1; /* (c) */
  15       a4[i*4+1] = tem2;
  16       double tem3 = a4[i*4] - tem2;
  17       double tem4 = tem3 + a4[i*4+n*4] /* (d) */;
  18       a4[i*4+n*4+1] = tem4 + a4[i*4+n*4+1] /* (b) */;
  19     }
  20 }
  21 int main(int argc, char **argv)
  22 {
  23   int n = 11;
  24   double a4[4 * n * 3];
  25   double a42[4 * n * 3];
  26   check_vect ();
  27   for (int i = 0; i < 4 * n * 3; ++i)
  28     {
  29       a4[i] = a42[i] = i;
  30       __asm__ volatile ("": : : "memory");
  31     }
  32   foo (a4, n);
  33   for (int i = 0; i < n; ++i)
  34     {
  35       double tem1 = a42[i*4] + a42[i*4+n*4];
  36       double tem2 = a42[i*4+2*n*4+1];
  37       a42[i*4+n*4+1] = tem1;
  38       a42[i*4+1] = tem2;
  39       double tem3 = a42[i*4] - tem2;
  40       double tem4 = tem3 + a42[i*4+n*4];
  41       a42[i*4+n*4+1] = tem4 + a42[i*4+n*4+1];
  42       __asm__ volatile ("": : : "memory");
  43     }
  44 #pragma GCC novector
  45   for (int i = 0; i < 4 * n * 3; ++i)
  46     if (a4[i] != a42[i])
  47       __builtin_abort ();
  48   return 0;
  49 }
  50
  51 /* For v2df we try to use SLP and fail miserably.  */
  52 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { target vect_sizes_32B_16B } } } */