gcc/testsuite/gcc.dg/vect/vect-reduc-in-order-4.c

   1 /* { dg-do run { xfail { { i?86-*-* x86_64-*-* } && ia32 } } } */
   2 /* { dg-require-effective-target vect_double } */
   3 /* { dg-add-options ieee } */
   4 /* { dg-additional-options "-fno-fast-math" } */
   5
   6 #include "tree-vect.h"
   7
   8 #define N (VECTOR_BITS * 17)
   9
  10 double __attribute__ ((noinline, noclone))
  11 reduc_plus_double (double *a)
  12 {
  13   double r1 = 0;
  14   double r2 = 0;
  15   double r3 = 0;
  16   double r4 = 0;
  17   for (int i = 0; i < N; i += 4)
  18     {
  19       r1 += a[i];
  20       r2 += a[i + 1];
  21       r3 += a[i + 2];
  22       r4 += a[i + 3];
  23     }
  24   return r1 * r2 * r3 * r4;
  25 }
  26
  27 int __attribute__ ((optimize (1)))
  28 main ()
  29 {
  30   double a[N];
  31   double r[4] = {};
  32   for (int i = 0; i < N; i++)
  33     {
  34       a[i] = (i * 0.1) * (i & 1 ? 1 : -1);
  35       r[i % 4] += a[i];
  36       asm volatile ("" ::: "memory");
  37     }
  38   double res = reduc_plus_double (a);
  39   if (res != r[0] * r[1] * r[2] * r[3])
  40     __builtin_abort ();
  41   return 0;
  42 }
  43
  44 /* { dg-final { scan-tree-dump {in-order unchained SLP reductions not supported} "vect" } } */
  45 /* { dg-final { scan-tree-dump-not {vectorizing stmts using SLP} "vect" } } */