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

[jvdelisle at gcc dot gnu.org]

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

--- Comment #13 from Jerry DeLisle <jvdelisle at gcc dot gnu.org> ---
(In reply to Thomas Koenig from comment #12)
> (In reply to Jerry DeLisle from comment #11)

---snip--
> 
> May I suggest reading the docs? ;-)
> 
--- snip ---

>      The default value for N is the value specified for
>      `-fblas-matmul-limit' if this option is specified, or unlimitited
>      otherwise.

Oh gosh!, Sorry about that Thomas. I just did not see it.  And I was even
looking for it because I thought it was there! This is excellent because I am
working on a modification to the run-time libraries. This will give us
something like:

 ====================================
                 Matmul              
                 fixed               
 Size  Loops     explicit   NewMatmul
 ====================================
   16  2000      1.496      1.719    
   32  2000      2.427      1.784    
   64  2000      1.343      1.967    
  128  2000      1.657      2.113    
  256   477      2.660      2.185    
  512    59      2.027      2.195    
 1024     7      1.530      2.208    
 2048     1      1.516      2.210    

On this particular machine, the inlining at high levels of optimization has
some sweet spots at size of 32 x 32 for example, so allowing the tuning is
essential depending on users application.