gcc/testsuite/gcc.dg/vect/bb-slp-32.c

   1 /* { dg-require-effective-target vect_int } */
   2 /* { dg-additional-options "-fvect-cost-model=dynamic" } */
   3
   4 #include "tree-vect.h"
   5
   6 int __attribute__((noipa))
   7 foo (int * __restrict__ x, int *p, int a, int b)
   8 {
   9   p = __builtin_assume_aligned (p, __BIGGEST_ALIGNMENT__);
  10   x = __builtin_assume_aligned (x, __BIGGEST_ALIGNMENT__);
  11   int tem0, tem1, tem2, tem3;
  12   int sum = 0;
  13   tem0 = p[0] + 1 + a;
  14   sum += tem0;
  15   x[0] = tem0;
  16   tem1 = p[1] + 2 + b;
  17   sum += tem1;
  18   x[1] = tem1;
  19   tem2 = p[2] + 3 + b;
  20   sum += tem2;
  21   x[2] = tem2;
  22   tem3 = p[3] + 4 + a;
  23   sum += tem3;
  24   x[3] = tem3;
  25   return sum;
  26 }
  27
  28 int x[4] __attribute__((aligned(__BIGGEST_ALIGNMENT__)));
  29 int p[4] __attribute__((aligned(__BIGGEST_ALIGNMENT__))) = { 0, 1, 2, 3 };
  30
  31 int main()
  32 {
  33   check_vect ();
  34
  35   if (foo (x, p, 7, 13) != 56)
  36     abort ();
  37   if (x[0] != 8 || x[1] != 16 || x[2] != 18 || x[3] != 14)
  38     abort ();
  39   return 0;
  40 }
  41
  42 /* This is a weak test but we want to re-use the arithmetic for both the
  43    store and the reduction.  */
  44 /* { dg-final { scan-tree-dump "re-using SLP tree" "slp2" { target { x86_64-*-* i?86-*-* } } } } */
  45 /* On i386 we vectorize both the store and the reduction.  */
  46 /* { dg-final { scan-tree-dump-times "basic block part vectorized" 2 "slp2" { target { x86_64-*-* i?86-*-* } } } } */