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

[jvdelisle at gcc dot gnu.org]

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

Jerry DeLisle <jvdelisle at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jvdelisle at gcc dot gnu.org

--- Comment #4 from Jerry DeLisle <jvdelisle at gcc dot gnu.org> ---
Created attachment 36869
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=36869&action=edit
Thomas program with a modified dgemm.

The dgemm in this example is a stripped out version of an "optimized for cache"
version from netlib.org.  I stripped out a lot of the unused code.

Results show better performance for larger arrays.  Maybe we could model the
library routines after this and invoke for larger arrays.

 Size     Loops          Matmul           dgemm         Matmul          Matmul
                      fixed explicit                    assumed      variable
explicit
 ==============================================================================
    2    200000           1.752           0.042           0.124           0.295
    4    200000           2.172           0.314           0.434           0.704
    8    200000           2.293           1.071           0.721           1.127
   16    200000           2.826           1.533           0.972           1.468
   32    200000           2.707           1.666           1.184           2.154
   64     30757           2.726           1.853           1.192           2.299
  128      3829           2.641           1.965           1.379           2.542
  256       477           2.661           2.001           1.384           2.594
  512        59           1.740           2.011           1.147           1.746
 1024         7           1.344           2.024           1.070           1.355
 2048         1           1.305           2.026           1.088           1.312