gcc/testsuite/gcc.dg/vect/slp-perm-6.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 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        /* Regular SLP - no permutation required.  */
  46        *pOutput2++ = K00 * d;
  47        *pOutput2++ = K10 * e;
  48     }
  49 }
  50
  51 int main (int argc, const char* argv[])
  52 {
  53   int input[N], output[N], i;
  54   int input2[N], output2[N];
  55
  56   check_vect ();
  57
  58   for (i = 0; i < N; i++)
  59     {
  60       input[i] = i%256;
  61       input2[i] = i%256;
  62       output[i] = 0;
  63       output2[i] = 0;
  64       __asm__ volatile ("");
  65     }
  66
  67 #if N == 16
  68   int check_results[N] = { 1470, 395, 28271, 5958, 1655, 111653, 10446, 2915,
  69                            195035, 14934, 4175, 278417, 19422, 5435, 361799,
  70                            0 };
  71   int check_results2[N] = { 0, 112, 810, 336, 1620, 560, 2430, 784, 3240, 1008,
  72                             0, 0, 0, 0, 0, 0 };
  73 #else
  74   volatile int check_results[N] = {};
  75   volatile int check_results2[N] = {};
  76
  77   for (int i = 0; i < N / 3; i++)
  78     {
  79       int a = input[i * 3];
  80       int b = input[i * 3 + 1];
  81       int c = input[i * 3 + 2];
  82       int d = input2[i * 2];
  83       int e = input2[i * 2 + 1];
  84
  85       check_results[i * 3] = M00 * a + M01 * b + M02 * c;
  86       check_results[i * 3 + 1] = M10 * a + M11 * b + M12 * c;
  87       check_results[i * 3 + 2] = M20 * a + M21 * b + M22 * c;
  88
  89       check_results2[i * 2] = K00 * d;
  90       check_results2[i * 2 + 1] = K10 * e;
  91
  92       asm volatile ("" ::: "memory");
  93     }
  94 #endif
  95
  96   foo (input, output, input2, output2);
  97
  98   for (i = 0; i < N; i++)
  99      if (output[i] != check_results[i] || output2[i] != check_results2[i])
 100        abort ();
 101
 102   return 0;
 103 }
 104
 105 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect"  { target vect_perm } } } */
 106 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 2 "vect" { target { vect_perm3_int && { ! vect_load_lanes } } } } } */
 107 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 1 "vect" { target vect_load_lanes } } } */
 108 /* { dg-final { scan-tree-dump "note: Built SLP cancelled: can use load/store-lanes" "vect" { target { vect_perm3_int && vect_load_lanes } } } } */
 109 /* { dg-final { scan-tree-dump "LOAD_LANES" "vect" { target vect_load_lanes } } } */
 110 /* { dg-final { scan-tree-dump "STORE_LANES" "vect" { target vect_load_lanes } } } */