[Bug fortran/68600] Inlined MATMUL is too slow.

[tkoenig at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68600

--- Comment #3 from Thomas Koenig <tkoenig at gcc dot gnu.org> ---
Created attachment 36868
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=36868&action=edit
Modified benchmark (really this time)

Hi Dominique,

I think you are seeing the effects of inefficiencies of assumed-shape arrays.

If you want to use matmul on very small matrix sizes, it is best to
use fixed-size explicit arrays.

Below the results of the modified benchmark (some changes to keep
the optimizer honest, such as a call to a dummy subroutine) on my
rather dated home box:

 Size     Loops          Matmul           dgemm         Matmul          Matmul
                      fixed explicit                    assumed      variable
explicit

=====================================================================================
    2    200000          11.948           0.072           0.142           0.411
    4    200000           1.711           0.417           0.534           0.861
    8    200000           2.314           0.953           0.858           1.076
   16    200000           1.745           1.276           0.918           1.000
   32    200000           1.459           1.456           1.371           1.436
   64     30757           1.501           1.440           1.360           1.393
  128      3829           1.586           1.544           1.557           1.529
  256       477           1.531           1.519           1.544           1.507
  512        59           1.315           1.290           1.263           1.231
 1024         7           1.110           1.081           1.069           1.053
 2048         1           1.095           1.086           1.081           1.058