gcc/testsuite/gcc.dg/vect/vect-reduc-dot-u16b.c

   1 /* Disabling epilogues until we find a better way to deal with scans.  */
   2 /* { dg-additional-options "--param vect-epilogues-nomask=0" } */
   3 /* { dg-require-effective-target vect_int } */
   4
   5 #include <stdarg.h>
   6 #include "tree-vect.h"
   7
   8 #define N 64
   9
  10 #define DOT2 43680
  11
  12 unsigned short X[N] __attribute__ ((__aligned__(__BIGGEST_ALIGNMENT__)));
  13 unsigned short Y[N] __attribute__ ((__aligned__(__BIGGEST_ALIGNMENT__)));
  14
  15 /* ushort->int->uint dot product: the multiplication promotes the ushorts
  16    to int, and then the product is converted to uint for the addition.  */
  17 __attribute__ ((noinline)) unsigned int
  18 foo2(int len) {
  19   int i;
  20   unsigned int result = 0;
  21
  22   for (i=0; i<len; i++) {
  23     result += (X[i] * Y[i]);
  24   }
  25   return result;
  26 }
  27
  28
  29 int main (void)
  30 {
  31   unsigned int  dot2;
  32   int i;
  33
  34   check_vect ();
  35
  36   for (i=0; i<N; i++) {
  37     X[i] = i;
  38     Y[i] = 64-i;
  39     __asm__ volatile ("");
  40   }
  41
  42   dot2 = foo2 (N);
  43   if (dot2 != DOT2)
  44     abort ();
  45
  46   return 0;
  47 }
  48
  49 /* { dg-final { scan-tree-dump-times "vect_recog_dot_prod_pattern: detected(?:(?!Analysis failed).)*Analysis succeeded" 1 "vect" { target { vect_pack_trunc || vect_udot_hi } } } } */
  50
  51 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { target { vect_pack_trunc || vect_udot_hi } } } } */