gcc/testsuite/gcc.dg/vect/slp-perm-7.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 /* SLP with load permutation and loop-based vectorization.  */
  28 void foo (int *__restrict__ pInput, int *__restrict__ pOutput,
  29           int *__restrict__ pInput2, int *__restrict__ pOutput2)
  30 {
  31   int i, a, b, c, d;
  32
  33   for (i = 0; i < N / 3; i++)
  34     {
  35        a = *pInput++;
  36        b = *pInput++;
  37        c = *pInput++;
  38        d = *pInput2++;
  39
  40        *pOutput++ = M00 * a + M01 * b + M02 * c;
  41        *pOutput++ = M10 * a + M11 * b + M12 * c;
  42        *pOutput++ = M20 * a + M21 * b + M22 * c;
  43
  44        /* Loop-based vectorization.  */
  45        *pOutput2++ = K00 * d;
  46     }
  47 }
  48
  49 int main (int argc, const char* argv[])
  50 {
  51   int input[N], output[N], i;
  52   int input2[N], output2[N];
  53
  54   check_vect ();
  55
  56   for (i = 0; i < N; i++)
  57     {
  58       input[i] = i%256;
  59       input2[i] = i%256;
  60       output[i] = 0;
  61       output2[i] = 0;
  62       asm volatile ("" ::: "memory");
  63     }
  64
  65 #if N == 16
  66   int check_results[N] = {1470, 395, 28271, 5958, 1655, 111653, 10446, 2915, 195035, 14934, 4175, 278417, 19422, 5435, 361799, 0};
  67   int check_results2[N] = {0, 405, 810, 1215, 1620, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
  68 #else
  69   volatile int check_results[N] = {};
  70   volatile int check_results2[N] = {};
  71
  72   for (int i = 0; i < N / 3; i++)
  73     {
  74       int a = input[i * 3];
  75       int b = input[i * 3 + 1];
  76       int c = input[i * 3 + 2];
  77       int d = input2[i];
  78
  79       check_results[i * 3] = M00 * a + M01 * b + M02 * c;
  80       check_results[i * 3 + 1] = M10 * a + M11 * b + M12 * c;
  81       check_results[i * 3 + 2] = M20 * a + M21 * b + M22 * c;
  82
  83       check_results2[i] = K00 * d;
  84
  85       asm volatile ("" ::: "memory");
  86     }
  87 #endif
  88
  89   foo (input, output, input2, output2);
  90
  91   for (i = 0; i < N; i++)
  92      if (output[i] != check_results[i] || output2[i] != check_results2[i])
  93        abort ();
  94
  95   return 0;
  96 }
  97
  98 /* { dg-final { scan-tree-dump-times "vectorized 1 loops" 1 "vect"  { target vect_perm } } } */
  99 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 1 "vect" { target { vect_perm3_int && { ! vect_load_lanes } } } } } */
 100 /* { dg-final { scan-tree-dump-times "vectorizing stmts using SLP" 0 "vect" { target vect_load_lanes } } } */
 101 /* { dg-final { scan-tree-dump "note: Built SLP cancelled: can use load/store-lanes" "vect" { target { vect_perm3_int && vect_load_lanes } } } } */
 102 /* { dg-final { scan-tree-dump "LOAD_LANES" "vect" { target vect_load_lanes } } } */
 103 /* { dg-final { scan-tree-dump "STORE_LANES" "vect" { target vect_load_lanes } } } */