[vorbis-dev] vorbis_analysis() dependencies?

[Jeff Squyres]

Per Monty's suggestions from a while ago, I have [finally] gotten around
to playing with different schemes for parallel oggenc.  Monty's main
suggestion was to have a single thread loop over reading samples and
calling vorbis_analysis_blockout(), and then queueing up the resulting
blocks to be processed through vorbis_analysis() in other threads (in
parallel).

To verify that this works, I modified vanilla oggenc (current CVS head on
everything, of course) to do something similar, but in a single-threaded
way.  However, the output from this scheme differs from the output of
vanilla oggenc.  i.e., it seems that delaying the calls to
vorbis_analysis() seems to generate different output.

I'm guessing that vorbis_analysis_blockout() is somehow changing internal
DSP state, and later calls to vorbis_analysis() are based upon that DSP
state, and are therefore produce different output.  Is that more or less
that is going on?  Is this by design?

-----

Here's the general algorithm in my slightly modified oggenc (ignoring
memory leaks here):

  - allocate/initialize a new vorbis_block
  - loop until file empty
    - read some samples
    - loop while vorbis_analysis_blockout returns 1
      - save current [and now full] vorbis_block
      - allocate/initialize new vorbis_block
    - if we have > N pending vorbis_blocks
      - loop over the N pending vorbis_blocks
        - call vorbis_analyze
        - call ogg_stream_packetin
        - call ogg_stream_pageout/write page to disk if necessary

Using a short audio clip (~17sec), I get different file sizes for the
resulting .ogg file when using this algorithm vs. the vanilla oggenc
algorithm.

I've attached my code that does this.  It's C++ code (sorry -- using STL
lists made the queueing much easier):

- put it in your oggenc directory
- [re]move your current oggenc.c to prevent "make" from getting confused
- s/encode.c/encode.cc/ in Makefile.am
- rm -rf .deps
- add AC_PROG_CXX to vorbis-tools/configure.in
- run autogen.sh for the vorbis-tools

Perhaps I did something stupid in the code...?

{+} Jeff Squyres
{+} squyres at cse.nd.edu
{+} Perpetual Obsessive Notre Dame Student Craving Utter Madness
{+} "I came to ND for 4 years and ended up staying for a decade"

--- >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 'vorbis-dev-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.