Re: [gnugo-devel] multi-board branch

[Marco Scheurer]

On Apr 28, 2005, at 5:48 PM, Paul Pogonyshev wrote:

> Arend wrote:
> > On Mon, 25 Apr 2005, Paul Pogonyshev wrote:
> > > > > Unfortunately, I was wrong.  Performance hit is around 10%: 6885
> > > > > seconds CPU time against 6106 seconds.  Timing may be not very
> > > > > precise as I used the box during regression runs, but the main
> > > > > picture is clear.
> > > >
> > > > Which CPU, compiler? I may try to run some tests myself.
> > >
> > > Athlon XP 2500+, GCC 3.2.
> >
> > On my Pentium M with GCC 3.4, I get about 7-8% performance penalty.
>
> So what do people think about this?  Is it too much?

I'm a bit surprised by these numbers. I was under the impression that 
the overhead of accessing global variables would somehow compensate for 
the cost of dereferencing.  I still think it is a good thing to have, 
but maybe the idea of using accessor macros to be able to switch that 
on or off could be given another thought?  Maybe something like:

#define board(engine) engine->board

vs

#define board(engine) g_board

Are you interested in numbers using GCC 4.0 on a G4 and /or a G5? If so 
could someone remind me what's needed to make measurements?

Thanks,

marco