gcc/testsuite/gcc.dg/vect/slp-reduc-sad.c

   1 /* { dg-require-effective-target vect_usad_char } */
   2
   3 #include "tree-vect.h"
   4
   5 typedef unsigned short uint16_t;
   6 typedef unsigned char uint8_t;
   7
   8 extern int abs (int);
   9 extern void abort (void);
  10
  11 int __attribute__((noinline,noclone))
  12 foo (uint8_t *pix1, uint8_t *pix2, int i_stride_pix2)
  13 {
  14   int i_sum = 5;
  15   for( int y = 0; y < 16; y++ )
  16     {
  17       i_sum += abs ( pix1[0] - pix2[0] );
  18       i_sum += abs ( pix1[1] - pix2[1] );
  19       i_sum += abs ( pix1[2] - pix2[2] );
  20       i_sum += abs ( pix1[3] - pix2[3] );
  21       i_sum += abs ( pix1[4] - pix2[4] );
  22       i_sum += abs ( pix1[5] - pix2[5] );
  23       i_sum += abs ( pix1[6] - pix2[6] );
  24       i_sum += abs ( pix1[7] - pix2[7] );
  25       i_sum += abs ( pix1[8] - pix2[8] );
  26       i_sum += abs ( pix1[9] - pix2[9] );
  27       i_sum += abs ( pix1[10] - pix2[10] );
  28       i_sum += abs ( pix1[11] - pix2[11] );
  29       i_sum += abs ( pix1[12] - pix2[12] );
  30       i_sum += abs ( pix1[13] - pix2[13] );
  31       i_sum += abs ( pix1[14] - pix2[14] );
  32       i_sum += abs ( pix1[15] - pix2[15] );
  33       pix1 += 16;
  34       pix2 += i_stride_pix2;
  35     }
  36   return i_sum;
  37 }
  38
  39 int
  40 main ()
  41 {
  42   check_vect ();
  43
  44   uint8_t X[16*16];
  45   uint8_t Y[16*16];
  46
  47   for (int i = 0; i < 16*16; ++i)
  48     {
  49       X[i] = i;
  50       Y[i] = 16*16 - i;
  51       __asm__ volatile ("");
  52     }
  53
  54   if (foo (X, Y, 16) != 32512 + 5)
  55     abort ();
  56
  57   return 0;
  58 }
  59
  60 /* { dg-final { scan-tree-dump "vect_recog_sad_pattern: detected" "vect" } } */
  61 /* { dg-final { scan-tree-dump "vectorizing stmts using SLP" "vect" } } */
  62 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" } } */