Re: i386 backend

[Jan Hubicka <jh at suse dot cz>]

> What do you mean by that?
Just it would be nice to have simple benchmark I can compile and run...
> 
> > it can be nice to turn it into kind of benchmark and send
> > me. The I can
> > investigate whats going wrong.
> >
> > In simple internal loops we should not be much worse unless we
> > have some good
> > purpose :)
> 
> All the chess programmers I know of, have experienced exactly the same. So
> any modern chess program would do. Mine is not the best example, being under
> heavy development at the moment. But there is Crafty, the open source master
> strength program by Dr. Robert Hyatt, which can me downloaded from his ftp
> site
> 
> ftp://ftp.cis.uab.edu/pub/hyatt/
> 
> latest version here:
> 
> ftp://ftp.cis.uab.edu/pub/hyatt/v18/crafty-18.7.tar.gz
> 
> His program behaves similarly with regards to the compilers mentioned.
OK, last time I investigated crafty, it had mainly problems with the register
alocations and use of long long, that is somewhat braindead on i386 gcc.
I can take a look how the number looks like in current gcc versus 2.95

Honza