gcc/testsuite/gcc.dg/vect/vect-reduc-dot-u16.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 #define DOT2 43680
  10
  11 unsigned short X[N] __attribute__ ((__aligned__(16)));
  12 unsigned short Y[N] __attribute__ ((__aligned__(16)));
  13
  14 /* short->short->int dot product.
  15    Not detected as a dot-product pattern.
  16    Not vectorized due to presence of type-conversions. */
  17 unsigned int
  18 foo1(int len) {
  19   int i;
  20   unsigned int result = 0;
  21   unsigned short prod;
  22
  23   for (i=0; i<len; i++) {
  24     prod = X[i] * Y[i];
  25     result += prod;
  26   }
  27   return result;
  28 }
  29
  30 /* short->int->int dot product.
  31    Currently not detected as a dot-product pattern: the multiplication
  32    promotes the ushorts to int, and then the product is promoted to unsigned
  33    int for the addition.  Which results in an int->unsigned int cast, which
  34    since no bits are modified in the cast should be trivially vectorizable.  */
  35 unsigned int
  36 foo2(int len) {
  37   int i;
  38   unsigned int result = 0;
  39
  40   for (i=0; i<len; i++) {
  41     result += (X[i] * Y[i]);
  42   }
  43   return result;
  44 }
  45
  46
  47 int main (void)
  48 {
  49   unsigned int dot1, dot2;
  50   int i;
  51
  52   check_vect ();
  53
  54   for (i=0; i<N; i++) {
  55     X[i] = i;
  56     Y[i] = 64-i;
  57   }
  58
  59   dot1 = foo1 (N);
  60   if (dot1 != DOT1)
  61     abort ();
  62
  63   dot2 = foo2 (N);
  64   if (dot2 != DOT2)
  65     abort ();
  66
  67   return 0;
  68 }
  69
  70 /* { dg-final { scan-tree-dump-times "vect_recog_dot_prod_pattern: detected" 1 "vect" { xfail *-*-* } } } */
  71
  72 /* Once the dot-product pattern is detected in the second loop, we expect
  73    that loop to be vectorized on vect_udot_hi targets (targets that support
  74    dot-product of unsigned shorts).  */
  75 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { xfail *-*-* } } } */
  76
  77 /* { dg-final { cleanup-tree-dump "vect" } } */