[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Linux Kernel Mailinglist archive: Re: On K7, -march=k6 is good (Was Re: Why no -march=athlon?)

Here are some facts about k6. I didn't have time to do a benchmark myself, so 
thanks to google I found the attached message. Hope it helps those who think 
i686 is better than k6 on an Athlon.

IMO, it is possible that i686 carves better FP performance (because I haven't 
seen a test that disproves that ;), but for integer performance k6 will be 
good for you Athlon users...

And if you look at the following page which adds athlon support to our beloved 
gcc-2.95.x you will see that the gains of -march=athlon over 
-march=pentiumpro seem to be only marginal.


But there is indeed some consistent however very small speedup when using 
-march=athlon there.

The real benefit of course comes from "real" optimizations. Since these 
processors are very smart and will do out-of-order processing you don't get 
the speedups you used to get in the 80's with these "Submodel" optimizations. 

The KDE build I did were done with "-O3 -march=i686". Now if you add to those 
flags -fstrict-aliasing -malign-functions=4 -fomit-frame-pointer, and add to 
the configure flags --enable-fast-malloc=full and --enable-final you will get 
quite a bit of speedup!


Eray Ozkural (exa) <erayo@cs.bilkent.edu.tr>
Comp. Sci. Dept., Bilkent University, Ankara
www: http://www.cs.bilkent.edu.tr/~erayo
GPG public key fingerprint: 360C 852F 88B0 A745 F31B  EA0F 7C07 AE16 874D 539C

Attachment: 0658.html+athlon+gcc-2.95+optimization&hl=en
Description: Binary data

Reply to: