Re: [mpg123-devel] fixed point decoders

[Thomas O. <tho...@or...>]

Am Tue, 02 Jun 2009 02:54:08 +0900
schrieb Taihei Monma <tm...@ma...>: 

> > But, well, this Core2 U7600 seems to be different enough from your  
> > average desktop Core2.
> 
> Hm, then a mystery... Thinking that the bottleneck of our decoder on  
> Core2 is the memory access, maybe smaller (2MB) L2 cache and slower  
> (533MHz) FSB have a huge negative impact on the performance.

Well, that explains that it's slower than a normal Core2 ... but slower per-MHz than an old K7?
The K7 has a lot less Cache and also slower RAM... but perhaps Core2 is just a bit overhyped for it's architecture as it's better than Netburst, which was really .... not good. K7 is also better than that;-)

Sidenote on MAD comparison: On my Core2 laptop, madplay is consistently faster than generic mpg123 code... perhaps we are entering compiler (flags) specific territory here?
But mpg123 has the lead in quality. Again, great work there.


Alrighty then,

Thomas.

PS: I hope you agree that we should not put the integer changes into 1.8.0... it's exciting and developing, changing code at the decoder core (although #ifdefed)... and because of that it should probably really stay in SVN until an even greater mpg123 1.9.0 arises. I hate to introduce subtle regressions (however far-fetched).