vect-reduc-dot-s8a.c

   1 /* { dg-require-effective-target vect_int } */
   2
   3 #include <stdarg.h>
   4 #include "tree-vect.h"
   5
   6 #define N 64
   7
   8 #define DOT1 43680
   9
  10 signed char X[N] __attribute__ ((__aligned__(__BIGGEST_ALIGNMENT__)));
  11 signed char Y[N] __attribute__ ((__aligned__(__BIGGEST_ALIGNMENT__)));
  12
  13 /* char->short->int dot product.
  14    The dot-product pattern should be detected.
  15    Vectorizable on vect_sdot_qi targets (targets that support dot-product of
  16    signed chars).
  17
  18    In the future could also be vectorized as widening-mult + widening-summation,
  19    or with type-conversion support.
  20  */
  21 __attribute__ ((noinline)) int
  22 foo1(int len) {
  23   int i;
  24   int result = 0;
  25   short prod;
  26
  27   for (i=0; i<len; i++) {
  28     prod = X[i] * Y[i];
  29     result += prod;
  30   }
  31   return result;
  32 }
  33
  34 int main (void)
  35 {
  36   int i, dot1;
  37
  38   check_vect ();
  39
  40   for (i=0; i<N; i++) {
  41     X[i] = i;
  42     Y[i] = 64-i;
  43     __asm__ volatile ("");
  44   }
  45
  46   dot1 = foo1 (N);
  47   if (dot1 != DOT1)
  48     abort ();
  49
  50   return 0;
  51 }
  52
  53 /* { dg-final { scan-tree-dump-times "vect_recog_dot_prod_pattern: detected" 1 "vect" } } */
  54 /* { dg-final { scan-tree-dump-times "vect_recog_widen_mult_pattern: detected" 1 "vect" } } */
  55 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { target vect_sdot_qi } } } */
  56 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { target { vect_widen_mult_qi_to_hi && vect_widen_sum_hi_to_si } } } } */
  57
  58 /* { dg-final { cleanup-tree-dump "vect" } } */