[Speex-dev] DSP c5x

Jean-Marc Valin Jean-Marc.Valin at USherbrooke.ca
Fri Mar 11 11:05:17 PST 2005

Hi, (trying to get back in touch with the rest of the world)

You shouldn't use ENABLE_VALGRIND unless you're actually debugging using
valgrind. As for reducing the encoder/decoder stack, you can usually
decrease that value when you're not using high complexity settings. If
your compiler supports C99 variable arrays or a non-braindead alloca,
you can try the code in svn and define either VAR_ARRAYS or USE_ALLOCA.
Note however that this is still experimental and may break.

As for the remaining float values when you define FIXED_POINT, it's
because not all the code has been completely converted to fixed-point.
There are still some float operations left in e.g. initialization and
packet loss concealment.


Le vendredi 11 mars 2005 à 12:11 -0600, Paul Gryting a écrit :
> I'm finding it dificult to get the code to fit within c5x dsp, especially
>    due to the large heap necessary for the huge allocations in
> nb_encoder_init and nb_decoder_init.
> Are the people that have this running using additonal external ram?
> Has anyone experimented with reducing the internal encoder/decoder stack
> sizes
>     for narrow band... is there a practical minimum that is usable?
> Also, Should ENABLE_VALGRIND be enabled?
> Any assistance always appreciated, thanks,
> Paul
> _______________________________________________
> Speex-dev mailing list
> Speex-dev at xiph.org
> http://lists.xiph.org/mailman/listinfo/speex-dev
Jean-Marc Valin <Jean-Marc.Valin at USherbrooke.ca>
Université de Sherbrooke

More information about the Speex-dev mailing list