Changing CFLAGS for i386 packages on x86_64(new benchmarks included)

[dragoran]

Kevin Kofler wrote:
> dragoran <dragoran <at> feuerpokemon.de> writes:
>   
>> the last one (-m64) is weird (much slower!!) no -OX was used.
>>     
>
> Then your benchmarks are essentially worthless. Sorry, but since GCC defaults 
> to -O0, which means no optimization whatsoever, i.e. very bad code, you should 
> NEVER compile production code without an -O flag (generally -O2 or -Os), and 
> especially not benchmarks!
>
> So please rerun your benchmarks with -O2 to get more useful results.
>
>   
ok done results see attachment
native 64bit one is much faster now (than without -O) and its now ~ 
equal to 32bit perfomance
as for the other result there are better then the old one but the diff 
between them is not the same as before all of them have almost equal 
perfomance.
so far the old benchmarks with -O2 added...
I have also tryed -mfpmath=sse and -ftree-vectorice and they seem to 
have a positive effect on module 1 which is (looking at the sourcecode):
/*******************************************************/
/* Module 1.  Calculate integral of df(x)/f(x) defined */
/*            below.  Result is ln(f(1)). There are 14 */
/*            double precision operations per loop     */
/*            ( 7 +, 0 -, 6 *, 1 / ) that are included */
/*            in the timing.                           */
/*            50.0% +, 00.0% -, 42.9% *, and 07.1% /   */
/*******************************************************/
For x86_64 (the -m64 benchmarks) it seems that -Os is better than -O2 
(maybe because on x86_64 the binarys are generally bigger and this 
reduces this effect == less cache misses?)
>         Kevin Kofler
>
>   

-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: benchmarks2.txt
Url: http://lists.fedoraproject.org/pipermail/devel/attachments/20060718/f0edad6f/attachment-0002.txt