[MPlayer-dev-eng] Re: Mplayer compiled with different gcc ver. benchmark

[attila]

On Fri, Apr 30, 2004 at 10:49:31PM +0200, Peter Niemayer wrote:
> Piotr Neuman wrote:
> 
> >Testing cmd. :
> >./mplayer -benchmark -vo null -ao null -frames 3000 <file>
> >
> >Results:
>  [...]
> 
> mplayer is certainly one of the worst code samples to base tests
> on compiler quality for x86 on, since the most time consuming parts are
> written in assembler for that platform.
> 
> Having the time critical parts written in assembler is of course a
> good thing, but if you want to use the application for compiler
> tests, I would recommend disabling the use of any assembler.

Sorry, to come so late, but this needs to be corrected:
This here is not about testing gcc performance, it's about testing
gcc performance on a specific video application, namely MPlayer.
Yes, it's true that important parts are written in asm, but
it's also true that a lot of loops are still in c.
And unlike the start up of an office suite [1] where it doesnt
really matter whether it's 1% faster or not, it matters a lot
with MPlayer. 1% can be the difference between smoth playback
and unplayable (experienced it a lot on my K6-II 350).


[1] ok, this is not completly true, i know about a company
who compared two notebook manufactorers by measuring the
boot time and the startup time of ms office with a stop watch.
A lot more intersting then the results who was faster was
what was the company who lost came up with to speed up things.
And yes, this is/was totaly stupid^Wcrazy.


			Attila Kinali