[tremor] Re: [vorbis-dev] Optimizing Vorbis for a DSP

[Johannes Sandvall]

On Sun 28 Mar 2004 23:23:46 CEST, Graham Mitchell wrote:
> > I have recently ported Vorbis to a Texas instruments DSP (TMS320C5510),
> > The problem I have is that it is not decoding the audio in real time(
> > about 50% of real time). Could any one give me a clue in which area I
> > could start to optimize the code.
> 
> Well, a few weeks ago Johannes Sandvall posted to the list about "replacing 
> the IMDCT with a N/4 FFT", which I seem to recall gave a large speedup; his 
> report is here:
> 	http://www.sandvall.nu/thesis.pdf
> 
> And my crufty brain seems to recall that profiling has determined that more 
> time is spent in the function "mdct_backward" than any other, but I could be 
> way off base.
> 
> Others on the list are much more qualified than I am, but sometimes a faster, 
> mediocre response is better than a slower, perfect one.  At least that's 
> what I'm banking on, anyway.
> 

Here is maybe a patch of the code. Should include both optimized TI
assembler and reference c code. Patch available at:
http://www.sandvall.nu/patch

It's for the lowmem branch of Tremor. One of the ways to install the
patch is:

dir_of_tremor_checkout$ cat patch | patch -p1

/ Johannes Sandvall
--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'tremor-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.