slp-perm-2.c

   1 /* { dg-require-effective-target vect_int } */
   2
   3 #include <stdarg.h>
   4 #include "tree-vect.h"
   5
   6 #define M00 100
   7 #define M10 216
   8 #define M01 1322
   9 #define M11 13
  10 #define M02 74
  11 #define M12 191
  12
  13 #define N 16
  14
  15 void foo (unsigned int *__restrict__ pInput, unsigned int *__restrict__ pOutput)
  16 {
  17   unsigned int i, a, b;
  18
  19   for (i = 0; i < N / 2; i++)
  20     {
  21        a = *pInput++;
  22        b = *pInput++;
  23
  24        *pOutput++ = M00 * a + M01 * b;
  25        *pOutput++ = M10 * a + M11 * b;
  26     }
  27 }
  28
  29 int main (int argc, const char* argv[])
  30 {
  31   unsigned int input[N], output[N], i;
  32   unsigned int check_results[N] = {1322, 13, 4166, 471, 7010, 929, 9854, 1387, 12698, 1845, 15542, 2303, 18386, 2761, 21230, 3219};
  33
  34   check_vect ();
  35
  36   for (i = 0; i < N; i++)
  37     {
  38       input[i] = i%256;
  39       output[i] = 0;
  40       __asm__ volatile ("");
  41     }
  42
  43   foo (input, output);
  44
  45   for (i = 0; i < N; i++)
  46     {
  47       if (output[i] != check_results[i])
  48         abort ();
  49       __asm__ volatile ("");
  50     }
  51
  52   return 0;
  53 }
  54
  55 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect"  { target vect_perm } } } */
  56 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 1 "vect" { target vect_perm } } } */
  57 /* { dg-final { cleanup-tree-dump "vect" } } */
  58