This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

RE: i386 backend


>
> > If you want an instance where the difference is pronounced, compile
> Crafty,
> > which I have posted a link to before. You will see that Crafty is
> > significantly slower on gcc. If that isn't a testcase, I don't know
> what is.
> Crafty is well enough known that it is the object of specific
> optimizations in commercial compilers.  gcc tends not to have so much
> specific tuning to benchmarks known to marketing.  Did you try all the

I don't know if what you are saying is true, but if you are, I can say that
the stuff aimed at Crafty also helps all other chess programs I know, and
probably a more general class of applications as well. There is nothing
artificial about that. That's just.. well, good.

> versions of gcc available to you, profile with all feasible choices of
> options, and pick the best for each section?

I've tried quite a few things, yes. People other than me, have tried much
harder to squeeze all the performance they could out of Crafty using
different compilers, profiler guided compilation etc. and have come up with
similar result.

David


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]