gcc/testsuite/gcc.dg/vect/bb-slp-subgroups-3.c

   1 /* { dg-require-effective-target vect_int } */
   2 /* { dg-require-effective-target vect_int_mult } */
   3 /* PR tree-optimization/67682.  */
   4
   5 #include "tree-vect.h"
   6
   7 int __attribute__((__aligned__(8))) a[8];
   8 int __attribute__((__aligned__(8))) b[8];
   9
  10 __attribute__ ((noinline)) void
  11 test ()
  12 {
  13     a[0] = b[0] + 1;
  14     a[1] = b[1] + 2;
  15     a[2] = b[2] + 3;
  16     a[3] = b[3] + 4;
  17     a[4] = b[0] * 3;
  18     a[5] = b[2] * 4;
  19     a[6] = b[4] * 5;
  20     a[7] = b[6] * 7;
  21 }
  22
  23 int
  24 main (int argc, char **argv)
  25 {
  26   check_vect ();
  27
  28   for (int i = 0; i < 8; i++)
  29     a[i] = 1;
  30   for (int i = 0; i < 8; i++)
  31     b[i] = i + 4;
  32   __asm__ volatile ("" : : : "memory");
  33   test (a, b);
  34   __asm__ volatile ("" : : : "memory");
  35   if ((a[0] != 5) || (a[1] != 7) || (a[2] != 9) || (a[3] != 11)
  36       || (a[4] != 12) || (a[5] != 24) || (a[6] != 40) || (a[7] != 70))
  37     abort ();
  38   return 0;
  39 }
  40
  41 /* { dg-final { scan-tree-dump-times "Basic block will be vectorized using SLP" 1 "slp2" } } */
  42 /* Because we disable the cost model, targets with variable-length
  43    vectors can end up vectorizing the store to a[0..7] on its own.
  44    With the cost model we do something sensible.  */
  45 /* { dg-final { scan-tree-dump-times "optimized: basic block" 2 "slp2" { target { ! amdgcn-*-* } xfail { vect_variable_length && { ! vect128 } } } } } */
  46
  47 /* amdgcn can do this in one vector.  */
  48 /* { dg-final { scan-tree-dump-times "optimized: basic block" 1 "slp2" { target amdgcn-*-* } } } */