On 20 Oct 2002, Atlantic Tech Solutions wrote:

>Thanks for the succinct and very clear explanation. I have come to many
>of the same conclusions. There are few apps which do benefit from
>-march=i686, but if one aims to increase performance for a given
>application other compiler optimizations will have a more dramatic
>benefit. 

I agree.  Also, the best optimizations will vary greatly from 
application to application, etc.

Another problem is knowing what optimizations are actually stable 
enough to use regularly.  I've never trusted a lot of gcc's 
optimizations personally.  Perhaps it has improved over the last 
few years a lot in this area though.


>Apps which use lots of floating point code can <<sometimes>> benefit
>from options like -ffast-math. My short bit of testing shows
>-march=athlon or -march=athlon-xp makes a bigger difference on
>8.0(because of gcc 3.2), than variations on x86.  

Interesting...   Perhaps an athlon optimized Mesa would be nice.  
;o)

>Simply upgrading from gcc 2.96 from 3.2 probably gains the most. :O

Yeah, that probably gives the largest benefit for the general 
case.

TTYL


-- 
Mike A. Harris          ftp://people.redhat.com/mharris
OS Systems Engineer
XFree86 maintainer
Red Hat Inc.




Reply via email to