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  #if VECTOR_BITS > 128
      22  #define N (VECTOR_BITS * 3 / 32 + 4)
      23  #else
      24  #define N 16
      25  #endif
      26  
      27  void foo (int *__restrict__ pInput, int *__restrict__ pOutput,
      28            int *__restrict__ pInput2, int *__restrict__ pOutput2)
      29  {
      30    int i, a, b, c, d, e;
      31  
      32    for (i = 0; i < N / 3; i++)
      33      {
      34         a = *pInput++;
      35         b = *pInput++;
      36         c = *pInput++;
      37  
      38         d = *pInput2++;
      39         e = *pInput2++;
      40  
      41         *pOutput++ = M00 * a + M01 * b + M02 * c;
      42         *pOutput++ = M10 * a + M11 * b + M12 * c;
      43         *pOutput++ = M20 * a + M21 * b + M22 * c;
      44  
      45         *pOutput2++ = K00 * d + K01 * e;
      46         *pOutput2++ = K10 * d + K11 * e;
      47      }
      48  }
      49  
      50  int main (int argc, const char* argv[])
      51  {
      52    int input[N], output[N], i;
      53    int input2[N], output2[N];
      54  
      55    check_vect ();
      56  
      57    check_vect ();
      58  
      59    for (i = 0; i < N; i++)
      60      {
      61        input[i] = i%256;
      62        input2[i] = i%256;
      63        output[i] = 0;
      64        output2[i] = 0;
      65        __asm__ volatile ("");
      66      }
      67  
      68  #if N == 16
      69    int check_results[N] = { 1470, 395, 28271, 5958, 1655, 111653, 10446, 2915,
      70  			   195035, 14934, 4175, 278417, 19422, 5435, 361799,
      71  			   0 };
      72    int check_results2[N] = { 4322, 135, 13776, 629, 23230, 1123, 32684, 1617,
      73  			    42138, 2111, 0, 0, 0, 0, 0, 0 };
      74  #else
      75    volatile int check_results[N] = {};
      76    volatile int check_results2[N] = {};
      77  
      78    for (int i = 0; i < N / 3; i++)
      79      {
      80        int a = input[i * 3];
      81        int b = input[i * 3 + 1];
      82        int c = input[i * 3 + 2];
      83        int d = input2[i * 2];
      84        int e = input2[i * 2 + 1];
      85  
      86        check_results[i * 3] = M00 * a + M01 * b + M02 * c;
      87        check_results[i * 3 + 1] = M10 * a + M11 * b + M12 * c;
      88        check_results[i * 3 + 2] = M20 * a + M21 * b + M22 * c;
      89  
      90        check_results2[i * 2] = K00 * d + K01 * e;
      91        check_results2[i * 2 + 1] = K10 * d + K11 * e;
      92  
      93        asm volatile ("" ::: "memory");
      94      }
      95  #endif
      96  
      97    foo (input, output, input2, output2);
      98  
      99    for (i = 0; i < N; i++)
     100       if (output[i] != check_results[i] || output2[i] != check_results2[i])
     101         abort ();
     102  
     103    return 0;
     104  }
     105  
     106  /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect"  { target vect_perm } } } */
     107  /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 2 "vect" { target { vect_perm3_int && { ! vect_load_lanes } } } } } */
     108  /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 0 "vect" { target vect_load_lanes } } } */
     109  /* { dg-final { scan-tree-dump "Built SLP cancelled: can use load/store-lanes" "vect" { target { vect_perm3_int && vect_load_lanes } } } } */
     110  /* { dg-final { scan-tree-dump "LOAD_LANES" "vect" { target vect_load_lanes } } } */
     111  /* { dg-final { scan-tree-dump "STORE_LANES" "vect" { target vect_load_lanes } } } */
     112