Thank you all, for useful tips.<br><br>Niranjan<br><br><br><br><div class="gmail_quote">On Fri, Oct 23, 2009 at 7:02 AM, Gregory Maxwell <span dir="ltr"><<a href="mailto:gmaxwell@gmail.com">gmaxwell@gmail.com</a>></span> wrote:<br>
<blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;"><div class="im">On Thu, Oct 22, 2009 at 8:40 PM, Niranjan Udipi <<a href="mailto:ucniranjan@gmail.com">ucniranjan@gmail.com</a>> wrote:<br>
> Thank you very much for the detailed description. I request you to elaborate<br>
> what is 'fine-grained parallelism' in Tremor, with an example.<br>
><br>
> Another thought is to split the OggVorbis encoded data in to four nearly<br>
> equal PCM length parts, once all the Headers have been decoded. Then, decode<br>
> each part separately in a new thread. In this case, only at the beginning<br>
> and end of the divided data parts, i.e. the first and last packets need the<br>
> previous and next packet respectively, for decoding. Please comment on this<br>
> idea. Please suggest me what functions I will have to use, to implement this<br>
> scheme.<br>
<br>
</div>Refer to section 4.3.8 of the Vorbis specification:<br>
<a href="http://www.xiph.org/vorbis/doc/Vorbis_I_spec.html#x1-790004.3.8" target="_blank">http://www.xiph.org/vorbis/doc/Vorbis_I_spec.html#x1-790004.3.8</a><br>
<br>
The vorbis frames represent overlapping portions of audio.<br>
<br>
You could decode the frames in parallel but must then sum the<br>
overlapping parts. This will require fairly extensive modification to<br>
the decoder. I do not believe that anyone here has investigated making<br>
this change so you would need to determine the best approach yourself.<br>
</blockquote></div><br>