gcc/testsuite/gcc.target/aarch64/pr110625_1.c

   1 /* { dg-do compile } */
   2 /* { dg-options "-Ofast -mcpu=neoverse-n2 -fdump-tree-vect-details -fno-tree-slp-vectorize" } */
   3 /* { dg-final { scan-tree-dump-not "reduction latency = 8" "vect" } } */
   4
   5 /* Do not increase the vector body cost due to the incorrect reduction latency
   6     Original vector body cost = 51
   7     Scalar issue estimate:
   8       ...
   9       reduction latency = 2
  10       estimated min cycles per iteration = 2.000000
  11       estimated cycles per vector iteration (for VF 2) = 4.000000
  12     Vector issue estimate:
  13       ...
  14       reduction latency = 8      <-- Too large
  15       estimated min cycles per iteration = 8.000000
  16     Increasing body cost to 102 because scalar code would issue more quickly
  17       ...
  18     missed:  cost model: the vector iteration cost = 102 divided by the scalar iteration cost = 44 is greater or equal to the vectorization factor = 2.
  19     missed:  not vectorized: vectorization not profitable.  */
  20
  21 typedef struct
  22 {
  23   unsigned short m1, m2, m3, m4;
  24 } the_struct_t;
  25 typedef struct
  26 {
  27   double m1, m2, m3, m4, m5;
  28 } the_struct2_t;
  29
  30 double
  31 bar (the_struct2_t *);
  32
  33 double
  34 foo (double *k, unsigned int n, the_struct_t *the_struct)
  35 {
  36   unsigned int u;
  37   the_struct2_t result;
  38   for (u = 0; u < n; u++, k--)
  39     {
  40       result.m1 += (*k) * the_struct[u].m1;
  41       result.m2 += (*k) * the_struct[u].m2;
  42       result.m3 += (*k) * the_struct[u].m3;
  43       result.m4 += (*k) * the_struct[u].m4;
  44     }
  45   return bar (&result);
  46 }