slp-perm-6.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 M20 23
   9 #define M01 1322
  10 #define M11 13
  11 #define M21 27271
  12 #define M02 74
  13 #define M12 191
  14 #define M22 500
  15
  16 #define K00 405
  17 #define K10 112
  18 #define K01 4322
  19 #define K11 135
  20
  21 #define N 16
  22
  23 void foo (int *__restrict__ pInput, int *__restrict__ pOutput,
  24           int *__restrict__ pInput2, int *__restrict__ pOutput2)
  25 {
  26   int i, a, b, c, d, e;
  27
  28   for (i = 0; i < N / 3; i++)
  29     {
  30        a = *pInput++;
  31        b = *pInput++;
  32        c = *pInput++;
  33
  34        d = *pInput2++;
  35        e = *pInput2++;
  36
  37        *pOutput++ = M00 * a + M01 * b + M02 * c;
  38        *pOutput++ = M10 * a + M11 * b + M12 * c;
  39        *pOutput++ = M20 * a + M21 * b + M22 * c;
  40
  41        /* Regular SLP - no permutation required.  */
  42        *pOutput2++ = K00 * d;
  43        *pOutput2++ = K10 * e;
  44     }
  45 }
  46
  47 int main (int argc, const char* argv[])
  48 {
  49   int input[N], output[N], i;
  50   int check_results[N] = {1470, 395, 28271, 5958, 1655, 111653, 10446, 2915, 195035, 14934, 4175, 278417, 19422, 5435, 361799, 0};
  51   int input2[N], output2[N];
  52   int check_results2[N] = {0, 112, 810, 336, 1620, 560, 2430, 784, 3240, 1008, 0, 0, 0, 0, 0, 0};
  53
  54   check_vect ();
  55
  56   for (i = 0; i < N; i++)
  57     {
  58       input[i] = i%256;
  59       input2[i] = i%256;
  60       output[i] = 0;
  61       output2[i] = 0;
  62       __asm__ volatile ("");
  63     }
  64
  65   foo (input, output, input2, output2);
  66
  67   for (i = 0; i < N; i++)
  68      if (output[i] != check_results[i] || output2[i] != check_results2[i])
  69        abort ();
  70
  71   return 0;
  72 }
  73
  74 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect"  { target vect_perm } } } */
  75 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 2 "vect" { target vect_perm } } } */
  76 /* { dg-final { cleanup-tree-dump "vect" } } */
  77