RFC: Optimizing for 386 (Part 2)

[dragoran]

Jeff Spaleta wrote:

>On Fri, 25 Mar 2005 09:34:08 -0600, Derek Moore <derek.p.moore at gmail.com> wrote:
>  
>
>>He's not the only one that believes compiler optimizations have an
>>effect at runtime (if he was, Gentoo wouldn't exist).
>>    
>>
>
>I believe that a diet of nothing but krispy kreme doughnuts will help
>me lose weight and reduce my blood pressure.  I 'feel' lighter and my
>blood 'feels' less pressurized when i squeeze my arm with my hand. 
>I'm waiting to get a full physical after my diet book hits the best
>sellers list and my diet has become a fashionable nationwide fad.
>
>Its hard to take anybody serviously who references 3rd party "studies"
>without an attempt at citation. If he's read these studies and they
>are publicly available for review, then they should be citable if not
>directly linkable. If only red hat's internal benchmarks were
>publishable, a lot of the noise surrounding this issue would
>evaporate.
>
>-jef"my blood doesn't 'flow'.. its more like a syrupy ooze"spaleta
>
>  
>
I have written an app that messures the memory write throughtput.
I played abit with the mtune and march switches and result was that i386 
and i686 doesn't make a difference. But if I use athlonxp (my box has an 
athlonxp cpu) it shows better result.
I have an other box with a pentium 4 cpu I will test on it when I have 
some time.
Here are the results:
-mtune=pentium4 -march=i386: 958.7 MB/s
-mtune=pentium4 -march=i386: 958.7 MB/s
-mtune=pentium4 -march=pentium4: 958.7 MB/s
-mtune=athlon-xp -march=i386: 1032.44 MB/s
-mtune=athlon-xp -march=i686: 1032.44 MB/s
-mtune=athlon-xp -march=athlon-xp: 1032.44 MB/s