Re: Problem with optimizations

[Olivier Scalbert <olivier.scalbert@algosyn.com>]

Gautier wrote:
> For your short-cut version of Is_Solved you might want to write:
> 
>         for i in Face_Index_T'range loop
>           for c in Column_T loop
>             for r in Row_T loop
>               if Cube(i)(c,r) /= Final_Position(i)(c,r) then
>                 return False;
>               end if;
> ...
> 
> Just nicer looking, probably as fast as the "unrolled" version you have. 
> Anyway, I strongly recommend adding -funroll-loops along with your -O2 
> switch. Also -fpeel-loops, -ftracer, -funswitch-loops might help.
> _________________________________________________________
> Gautier's Ada programming -- http://sf.net/users/gdemont/
> NB: For a direct answer, e-mail address on the Web site!

Thanks for your help.

Your version, which is nicer and more generic (4*4, 5*5 cube) on my machine:
gnatmake -f -O3 -W -gnatp -fpeel-loops -ftracer -funswitch-loops rubikmain
real    0m3.057s
user    0m2.988s
sys     0m0.004s

Mine (manually unrolled):
real    0m2.935s
user    0m2.876s
sys     0m0.004s

So, nearly the same. So I'll go for the nice one!

But with:
  function Is_Solved1(Cube: Cube_T) return Boolean is
     begin
         return Cube = Final_position;
     end Is_Solved1;

I have:
real    0m4.775s
user    0m4.660s
sys     0m0.004s

That is much longer !

Does it have the same semantic ?

Olivier