[Tremor] optimizing for coldfire

[Pedro Vasconcelos]

On Thu, 26 May 2005 09:48:36 +0200
Balatoni Denes <dbalatoni at programozo.hu> wrote:

> Hi!
> 
> I too profiled only on a desktop machine. I have found that the second
> most timeconsuming operation after the imdct is the residue decoding. That
> means that all(or maybe only the most important parts) data used in huffman 
> code decoding should be in sram too imho.
> 
> bye
> Denes
> 

Hi Denes, Thanks for your reply. I did look at res012.c but I couldn't
find an obvious candidate for sram allocation. Could you elaborate on
what data structures should i look for? 

BTW, I found out that even with just the simple optimizations, Tremor
does actually run at 130% real-time provided the Coldfire is uprated to
120mhz rather than the default 50mhz. But extra optimizations will help
reduce the power consumption...

Cheers,

Pedro


-- 
Pedro Vasconcelos, School of Computer Science, University of St Andrews
-----------------------------------------------------------------------
"The difference between Theory and Practice 
is greater in Practice than in Theory."