gcc/testsuite/gcc.dg/vect/slp-perm-4.c

   1 /* { dg-require-effective-target vect_int } */
   2 /* { dg-require-effective-target vect_perm } */
   3
   4 #include <stdarg.h>
   5 #include "tree-vect.h"
   6
   7 #define M00 100
   8 #define M10 216
   9 #define M20 23
  10 #define M30 237
  11 #define M40 437
  12
  13 #define M01 1322
  14 #define M11 13
  15 #define M21 27271
  16 #define M31 2280
  17 #define M41 284
  18
  19 #define M02 74
  20 #define M12 191
  21 #define M22 500
  22 #define M32 111
  23 #define M42 1114
  24
  25 #define M03 134
  26 #define M13 117
  27 #define M23 11
  28 #define M33 771
  29 #define M43 71
  30
  31 #define M04 334
  32 #define M14 147
  33 #define M24 115
  34 #define M34 7716
  35 #define M44 16
  36
  37 #if VECTOR_BITS > 128
  38 #define N (VECTOR_BITS * 5 / 32)
  39 #else
  40 #define N 20
  41 #endif
  42
  43 void foo (unsigned int *__restrict__ pInput, unsigned int *__restrict__ pOutput)
  44 {
  45   unsigned int i, a, b, c, d, e;
  46
  47   for (i = 0; i < N / 5; i++)
  48     {
  49        a = *pInput++;
  50        b = *pInput++;
  51        c = *pInput++;
  52        d = *pInput++;
  53        e = *pInput++;
  54
  55        *pOutput++ = M00 * a + M01 * b + M02 * c + M03 * d + M04 * e;
  56        *pOutput++ = M10 * a + M11 * b + M12 * c + M13 * d + M14 * e;
  57        *pOutput++ = M20 * a + M21 * b + M22 * c + M23 * d + M24 * e;
  58        *pOutput++ = M30 * a + M31 * b + M32 * c + M33 * d + M34 * e;
  59        *pOutput++ = M40 * a + M41 * b + M42 * c + M43 * d + M44 * e;
  60     }
  61 }
  62
  63 int main (int argc, const char* argv[])
  64 {
  65   unsigned int input[N], output[N], i;
  66
  67   check_vect ();
  68
  69   for (i = 0; i < N; i++)
  70     {
  71       input[i] = i%256;
  72       output[i] = 0;
  73       asm volatile ("" ::: "memory");
  74     }
  75
  76 #if N == 20
  77   unsigned int check_results[N]
  78     = {3208, 1334, 28764, 35679, 2789, 13028, 4754, 168364, 91254, 12399,
  79     22848, 8174, 307964, 146829, 22009, 32668, 11594, 447564, 202404, 31619};
  80 #else
  81   volatile unsigned int check_results[N];
  82
  83   for (i = 0; i < N / 5; i++)
  84     {
  85       unsigned int a = input[i * 5];
  86       unsigned int b = input[i * 5 + 1];
  87       unsigned int c = input[i * 5 + 2];
  88       unsigned int d = input[i * 5 + 3];
  89       unsigned int e = input[i * 5 + 4];
  90
  91       check_results[i * 5] = M00 * a + M01 * b + M02 * c + M03 * d + M04 * e;
  92       check_results[i * 5 + 1] = (M10 * a + M11 * b + M12 * c
  93                                   + M13 * d + M14 * e);
  94       check_results[i * 5 + 2] = (M20 * a + M21 * b + M22 * c
  95                                   + M23 * d + M24 * e);
  96       check_results[i * 5 + 3] = (M30 * a + M31 * b + M32 * c
  97                                   + M33 * d + M34 * e);
  98       check_results[i * 5 + 4] = (M40 * a + M41 * b + M42 * c
  99                                   + M43 * d + M44 * e);
 100       asm volatile ("");
 101     }
 102 #endif
 103
 104   foo (input, output);
 105
 106 #pragma GCC novector
 107   for (i = 0; i < N; i++)
 108     if (output[i] != check_results[i])
 109       abort ();
 110
 111   return 0;
 112 }
 113
 114 /* Currently interleaving is not supported for a group-size of 5.  */
 115
 116 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect" } } */
 117 /* { dg-final { scan-tree-dump-times "gaps requires scalar epilogue loop" 0 "vect" } } */
 118 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 1 "vect" } } */