actually, this might be sufficiently proper documentation:
http://focus.ti.com/dsp/docs/dspsupp...ocCategoryId=6
for example:
http://www.ti.com/litv/pdf/spru732j (C64x+ DSP CPU and Instruction Set Reference Guide)

or it may as well not be sufficient.

im not sure what IVA2 represents... if its software or hardware based?
maybe its TI's name for a software suite for the C64+ DSP?
maybe its standalone silicon in the SoC?

personally i think IVA2 mostly reffers to TI's software that comes bundled with the C64+ DSP - everything ranging from compilers/SDK's to pre-built codecs (coded with those SDK's).
but like u said - it seems quite undocumented, unless the link above provides some info reffering to the IVA2.

im still investigating if its worth undertaking the task to create a clean "Camera OS" for the N900, and code some lightweight capture codecs...
Im not even sure that i can learn all relevant documents, since its so much.

...about undertaking the task to coding a new codec:

* the DSP have very limited cache resources and it cant process a complete 1280x720 picture in one sweep (of course).
* if regular RAM is used instead - it will hog the BUS.

therefore, i think the best approach is to compress each frame - segment by segment (which will cause blockiness), and save them one by one to RAM... writing to the eMMC will have to wait until the video is finished.
Using this approach, will put a time-limit on how long the recorded videos can become, but it will atleast provide plenty of resources/speed for capturing in HQ and high fps.


...still i know too little to make qualified judgements - so all of the above should be taken with a grain of salt... we'll see if i manage to learn enough to undertake this task.