gcc/testsuite/gcc.dg/vect/slp-perm-1.c

   1 /* { dg-require-effective-target vect_int } */
   2
   3 #include <stdarg.h>
   4 #include <stdio.h>
   5 #include "tree-vect.h"
   6
   7 #define M00 100
   8 #define M10 216
   9 #define M20 23
  10 #define M01 1322
  11 #define M11 13
  12 #define M21 27271
  13 #define M02 74
  14 #define M12 191
  15 #define M22 500
  16
  17 #define N 16
  18
  19 void foo (unsigned int *__restrict__ pInput, unsigned int *__restrict__ pOutput)
  20 {
  21   unsigned int i, a, b, c;
  22
  23   for (i = 0; i < N / 3; i++)
  24     {
  25        a = *pInput++;
  26        b = *pInput++;
  27        c = *pInput++;
  28
  29        *pOutput++ = M00 * a + M01 * b + M02 * c;
  30        *pOutput++ = M10 * a + M11 * b + M12 * c;
  31        *pOutput++ = M20 * a + M21 * b + M22 * c;
  32     }
  33 }
  34
  35 int main (int argc, const char* argv[])
  36 {
  37   unsigned int input[N], output[N], i;
  38   unsigned int check_results[N] = {1470, 395, 28271, 5958, 1655, 111653, 10446, 2915, 195035, 14934, 4175, 278417, 19422, 5435, 361799, 0};
  39
  40   check_vect ();
  41
  42   for (i = 0; i < N; i++)
  43     {
  44       input[i] = i%256;
  45       output[i] = 0;
  46       __asm__ volatile ("");
  47     }
  48
  49   foo (input, output);
  50
  51   for (i = 0; i < N; i++)
  52     {
  53       if (output[i] != check_results[i])
  54         abort ();
  55       __asm__ volatile ("");
  56     }
  57
  58   return 0;
  59 }
  60
  61 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect"  { target vect_perm } } } */
  62 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 1 "vect" { target vect_perm } } } */
  63 /* { dg-final { cleanup-tree-dump "vect" } } */
  64