pr45752.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 M30 237
  10 #define M40 437
  11
  12 #define M01 1322
  13 #define M11 13
  14 #define M21 27271
  15 #define M31 2280
  16 #define M41 284
  17
  18 #define M02 74
  19 #define M12 191
  20 #define M22 500
  21 #define M32 111
  22 #define M42 1114
  23
  24 #define M03 134
  25 #define M13 117
  26 #define M23 11
  27 #define M33 771
  28 #define M43 71
  29
  30 #define M04 334
  31 #define M14 147
  32 #define M24 115
  33 #define M34 7716
  34 #define M44 16
  35
  36 #define N 16
  37
  38 void foo (unsigned int *__restrict__ pInput,
  39           unsigned int *__restrict__ pOutput,
  40           unsigned int *__restrict__ pInput2,
  41           unsigned int *__restrict__ pOutput2)
  42 {
  43   unsigned int i, a, b, c, d, e;
  44
  45   for (i = 0; i < N / 5; i++)
  46     {
  47        a = *pInput++;
  48        b = *pInput++;
  49        c = *pInput++;
  50        d = *pInput++;
  51        e = *pInput++;
  52
  53        *pOutput++ = M00 * a + M01 * b + M02 * c + M03 * d + M04 * e;
  54        *pOutput++ = M10 * a + M11 * b + M12 * c + M13 * d + M14 * e;
  55        *pOutput++ = M20 * a + M21 * b + M22 * c + M23 * d + M24 * e;
  56        *pOutput++ = M30 * a + M31 * b + M32 * c + M33 * d + M34 * e;
  57        *pOutput++ = M40 * a + M41 * b + M42 * c + M43 * d + M44 * e;
  58
  59
  60        a = *pInput2++;
  61        b = *pInput2++;
  62        c = *pInput2++;
  63        d = *pInput2++;
  64        e = *pInput2++;
  65
  66        *pOutput2++ = M00 * a + M01 * b + M02 * c + M03 * d + M04 * e;
  67        *pOutput2++ = M10 * a + M11 * b + M12 * c + M13 * d + M14 * e;
  68        *pOutput2++ = M20 * a + M21 * b + M22 * c + M23 * d + M24 * e;
  69        *pOutput2++ = M30 * a + M31 * b + M32 * c + M33 * d + M34 * e;
  70        *pOutput2++ = M40 * a + M41 * b + M42 * c + M43 * d + M44 * e;
  71
  72     }
  73 }
  74
  75 int main (int argc, const char* argv[])
  76 {
  77   unsigned int input[N], output[N], i, input2[N], output2[N];
  78   unsigned int check_results[N] = {3208, 1334, 28764, 35679, 2789, 13028,
  79     4754, 168364, 91254, 12399, 22848, 8174, 307964, 146829, 22009, 0};
  80   unsigned int check_results2[N] = {7136, 2702, 84604, 57909, 6633, 16956,
  81     6122, 224204, 113484, 16243, 26776, 9542, 363804, 169059, 25853, 0};
  82
  83   check_vect ();
  84
  85   for (i = 0; i < N; i++)
  86     {
  87       input[i] = i%256;
  88       input2[i] = i + 2;
  89       output[i] = 0;
  90       output2[i] = 0;
  91       __asm__ volatile ("");
  92     }
  93
  94   foo (input, output, input2, output2);
  95
  96   for (i = 0; i < N; i++)
  97     if (output[i] != check_results[i]
  98         || output2[i] != check_results2[i])
  99       abort ();
 100
 101   return 0;
 102 }
 103
 104 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 0 "vect"  } } */
 105 /* { dg-final { scan-tree-dump-times "permutation requires at least three vectors" 2 "vect" { target vect_perm } } } */
 106 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 0 "vect"  } } */
 107 /* { dg-final { cleanup-tree-dump "vect" } } */
 108