[Speex-dev] speex on embedded processor
jean-marc.valin at usherbrooke.ca
Fri Feb 1 14:27:34 PST 2008
Alain M. a écrit :
> Hi Jean-Marc,
> will you include these optimizations for minimal ram usage in the main
The optimisations are already there. There's nothing to add.
> Maybe if Peter sends a Patch...
> Jean-Marc Valin escreveu:
>> Sounds like you did most of what you needed to do. Using a coverage
>> analysis tool might help you remove bits of code here and there (e.g.
>> unused _ctl() cases), but the main part is done. About the preprocessor,
>> there are some of the state variables that could instead be allocated on
>> the stack, but that's about as far as you can shrink things.
>> Peter Ruckle a écrit :
>>> I need to shrink speex so it will fit on an embedded processor along
>>> with some other code. The chip has only 20k of ram. I have done
>>> some work with libspeex.dsp by removing references to all modes but
>>> narrow band 8000 hz (submode 3). I used the defines: FIXED_POINT,
>>> DISABLE_FLOATING_POINT_API, DISABLE_VBR. I also removed the
>>> structures for the other submodes and the functions that they called.
>>> I reduced the NB_DEC_STACK and NB_ENC_STACK to the smallest values
>>> and minimized the heap based on all calls to calloc. I haven't
>>> optimized the call stack yet. I reduced the ram usage to around 6k.
>>> The problem is when I added the preprocess functionality, the ram
>>> usage went to around 20k. I need to know what can be removed or
>>> reduced from the S[eexPreprocessState structure. The only
>>> functionality we want to use is the denoise option. We don't want
>>> vbr or echo cancellation. I only see about 1k that is directly
>>> related that can be removed from the structure. Are there any other
>>> buffers in the structure that can be removed or a way to reduce the
>>> size of some buffers.
>>> Peter Ruckle
>>> _______________________________________________ Speex-dev mailing
>>> list Speex-dev at xiph.org http://lists.xiph.org/mailman/listinfo/speex-dev
>> Speex-dev mailing list
>> Speex-dev at xiph.org
> Speex-dev mailing list
> Speex-dev at xiph.org
More information about the Speex-dev