[oc] video processors (was: think different)

[Jecel Assumpcao Jr <jecel@merlintec.com>]

Marko,
> Yes, Java is far more useful language just because it is more specific
> (higher) language...
> But I found out that you can find some parallelism even in C. And if
> you set (reasonable) limit, that programs doesn't cause out of array
> or structure pointers, you can conclude many things.

A major problem with C is address aliasing: given two pointers into
memory it is hard to determine at compile time that they will always
point to different addresses. Without this information, the dataflow
you extract from the source code will have to be very conservative, and
so not very parallel.

Here is some experiments that I have done in extracting parallelism
from Self code (a Smalltalk dialect, and the really good language that
Sun developed ;-):

   http://www.lsi.usp.br/~jecel/tiny.html

> or2k would be very suitable for this. Or2k project was kinda abandoned
> because arhitecture is too difenent and 2.5Dhrystones/MIPS doesn't
> balance that for desktop computer. I really think having or2k matrix 4x4
> and maybe or1k to program it, would be perfect solution for video.
> AFAIK there is no better processor for lo-res video aplications like you
> mentioned (and still being special purpose).

Do you know the:

  - Philips Trimedia?

     http://www-us.semiconductors.philips.com/trimedia/

  - Stanford Imagine Chip?

     http://cva.stanford.edu/imagine/cva_imagine.html

  - BOPS ManArray?

     http://bopsnet.com/cores/manarray.shtml

If not, you might find them (particularly the last one) interesting as
sources of inspiration for the Or2K. I would include my own Tachyon CPU
design in this list, but I haven't finished writing the pages
describing it yet.

-- Jecel