gcc/testsuite/gcc.dg/vect/slp-perm-8.c

   1 /* { dg-require-effective-target vect_int } */
   2
   3 #include <stdarg.h>
   4 #include "tree-vect.h"
   5
   6 #if VECTOR_BITS > 512
   7 #define N (VECTOR_BITS * 6 / 16)
   8 #else
   9 #define N 200
  10 #endif
  11
  12 void __attribute__((noinline))
  13 foo (unsigned char *__restrict__ pInput, unsigned char *__restrict__ pOutput)
  14 {
  15   unsigned char a, b, c;
  16   unsigned int i;
  17
  18   for (i = 0; i < N / 3; i++)
  19     {
  20        a = *pInput++;
  21        b = *pInput++;
  22        c = *pInput++;
  23
  24        *pOutput++ = a + b + c + 3;
  25        *pOutput++ = a + b + c + 12;
  26        *pOutput++ = a + b + c + 1;
  27     }
  28 }
  29
  30 int main (int argc, const char* argv[])
  31 {
  32   unsigned char input[N], output[N];
  33   unsigned char check_results[N];
  34   unsigned int i;
  35
  36   check_vect ();
  37
  38   for (i = 0; i < N; i++)
  39     {
  40       input[i] = i;
  41       output[i] = 0;
  42       __asm__ volatile ("");
  43     }
  44
  45   for (i = 0; i < N / 3; i++)
  46     {
  47       check_results[3*i] = 9 * i + 6;
  48       check_results[3*i+1] = 9 * i + 15;
  49       check_results[3*i+2] = 9 * i + 4;
  50       __asm__ volatile ("" : : : "memory");
  51     }
  52
  53   foo (input, output);
  54
  55 #pragma GCC novector
  56   for (i = 0; i < N - (N % 3); i++)
  57      if (output[i] != check_results[i])
  58        abort ();
  59
  60   return 0;
  61 }
  62
  63 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" { target { vect_perm_byte } } } } */
  64 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 1 "vect" { target { vect_perm3_byte && { { ! vect_load_lanes } && { { ! vect_partial_vectors_usage_1 } || s390_vx } } } } } } */
  65 /* The epilogues are vectorized using partial vectors.  */
  66 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 2 "vect" { target { vect_perm3_byte && { { ! vect_load_lanes } && { vect_partial_vectors_usage_1 && { ! s390_vx } } } } } } } */
  67 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 0 "vect" { target vect_load_lanes } } } */
  68 /* { dg-final { scan-tree-dump "Built SLP cancelled: can use load/store-lanes" "vect" { target { vect_perm3_byte && vect_load_lanes } } } } */
  69 /* { dg-final { scan-tree-dump "LOAD_LANES" "vect" { target vect_load_lanes } } } */
  70 /* { dg-final { scan-tree-dump "STORE_LANES" "vect" { target vect_load_lanes } } } */