gcc/testsuite/gcc.dg/vect/slp-19c.c

   1 /* { dg-require-effective-target vect_int } */
   2
   3 #include <stdarg.h>
   4 #include "tree-vect.h"
   5
   6 #if VECTOR_BITS > 128
   7 #define N (VECTOR_BITS * 2 / 32)
   8 #else
   9 #define N 16
  10 #endif
  11
  12 int
  13 main1 ()
  14 {
  15   unsigned int i;
  16   unsigned int out[N*8];
  17 #if N == 16
  18   unsigned int in[N*8] = {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63};
  19 #else
  20   unsigned int in[N * 8];
  21   for (unsigned int i = 0; i < N * 8; ++i)
  22     {
  23       in[i] = i & 63;
  24       asm volatile ("" ::: "memory");
  25     }
  26 #endif
  27   unsigned int ia[N*2], a0, a1, a2, a3;
  28
  29   /* The last stmt requires interleaving of not power of 2 size - not
  30      vectorizable.  */
  31   for (i = 0; i < N/2; i++)
  32     {
  33       out[i*12] = in[i*12];
  34       out[i*12 + 1] = in[i*12 + 1];
  35       out[i*12 + 2] = in[i*12 + 2];
  36       out[i*12 + 3] = in[i*12 + 3];
  37       out[i*12 + 4] = in[i*12 + 4];
  38       out[i*12 + 5] = in[i*12 + 5];
  39       out[i*12 + 6] = in[i*12 + 6];
  40       out[i*12 + 7] = in[i*12 + 7];
  41       out[i*12 + 8] = in[i*12 + 8];
  42       out[i*12 + 9] = in[i*12 + 9];
  43       out[i*12 + 10] = in[i*12 + 10];
  44       out[i*12 + 11] = in[i*12 + 11];
  45
  46       ia[i] = in[i*12 + 7];
  47     }
  48
  49   /* check results:  */
  50 #pragma GCC novector
  51   for (i = 0; i < N/2; i++)
  52     {
  53       if (out[i*12] !=  in[i*12]
  54          || out[i*12 + 1] != in[i*12 + 1]
  55          || out[i*12 + 2] != in[i*12 + 2]
  56          || out[i*12 + 3] != in[i*12 + 3]
  57          || out[i*12 + 4] != in[i*12 + 4]
  58          || out[i*12 + 5] != in[i*12 + 5]
  59          || out[i*12 + 6] != in[i*12 + 6]
  60          || out[i*12 + 7] != in[i*12 + 7]
  61          || out[i*12 + 8] != in[i*12 + 8]
  62          || out[i*12 + 9] != in[i*12 + 9]
  63          || out[i*12 + 10] != in[i*12 + 10]
  64          || out[i*12 + 11] != in[i*12 + 11]
  65          || ia[i] != in[i*12 + 7])
  66         abort ();
  67     }
  68
  69   /* Hybrid SLP with unrolling by 2.  */
  70   for (i = 0; i < N; i++)
  71     {
  72       out[i*6] = in[i*6];
  73       out[i*6 + 1] = in[i*6 + 1];
  74       out[i*6 + 2] = in[i*6 + 2];
  75       out[i*6 + 3] = in[i*6 + 3];
  76       out[i*6 + 4] = in[i*6 + 4];
  77       out[i*6 + 5] = in[i*6 + 5];
  78
  79       ia[i] = i;
  80     }
  81
  82   /* check results:  */
  83 #pragma GCC novector
  84   for (i = 0; i < N/2; i++)
  85     {
  86       if (out[i*6] !=  in[i*6]
  87          || out[i*6 + 1] != in[i*6 + 1]
  88          || out[i*6 + 2] != in[i*6 + 2]
  89          || out[i*6 + 3] != in[i*6 + 3]
  90          || out[i*6 + 4] != in[i*6 + 4]
  91          || out[i*6 + 5] != in[i*6 + 5]
  92          || ia[i] != i)
  93         abort ();
  94     }
  95
  96   return 0;
  97 }
  98
  99 int main (void)
 100 {
 101   check_vect ();
 102
 103   main1 ();
 104
 105   return 0;
 106 }
 107
 108 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { target { ! vect64 } } } } */
 109 /* The unsupported interleaving works fine with V2SImode vectors given we
 110    can always combine that from two vectors.  */
 111 /* { dg-final { scan-tree-dump-times "vectorized 2 loops" 1 "vect" { target vect64 } } } */
 112 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 2 "vect" { target { ! vect64 } } } } */
 113 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 4 "vect" { target vect64 } } } */