gcc/testsuite/gcc.dg/vect/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 __attribute__((noinline))
  16 foo (unsigned int *__restrict__ pInput, unsigned int *__restrict__ pOutput)
  17 {
  18   unsigned int i, a, b;
  19
  20   for (i = 0; i < N / 2; i++)
  21     {
  22        a = *pInput++;
  23        b = *pInput++;
  24
  25        *pOutput++ = M00 * a + M01 * b;
  26        *pOutput++ = M10 * a + M11 * b;
  27     }
  28 }
  29
  30 int main (int argc, const char* argv[])
  31 {
  32   unsigned int input[N], output[N], i;
  33   unsigned int check_results[N] = {1322, 13, 4166, 471, 7010, 929, 9854, 1387, 12698, 1845, 15542, 2303, 18386, 2761, 21230, 3219};
  34
  35   check_vect ();
  36
  37   for (i = 0; i < N; i++)
  38     {
  39       input[i] = i%256;
  40       output[i] = 0;
  41       __asm__ volatile ("");
  42     }
  43
  44   foo (input, output);
  45
  46 #pragma GCC novector
  47   for (i = 0; i < N; i++)
  48     {
  49       if (output[i] != check_results[i])
  50         abort ();
  51       __asm__ volatile ("");
  52     }
  53
  54   return 0;
  55 }
  56
  57 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect"  { target vect_perm } } } */
  58 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 1 "vect" { target { vect_perm || vect_load_lanes  } } } } */
  59 /* { dg-final { scan-tree-dump "LOAD_LANES" "vect" { target vect_load_lanes } } } */
  60 /* { dg-final { scan-tree-dump "STORE_LANES" "vect" { target vect_load_lanes } } } */