[theora-dev] Hardware ReconRefFrames!

[Felipe Portavales Goldstein]

We have now the ReconRefFrames routine (and all sub-routines called
from it) running on a synthesizable VHDL model.

It means that we can make the most time-consuming part of the
algorithm entirely in hardware.

As I said in other emails the ReconRefFrames and its subroutines
consumes about 60% to 70% of cpu-time.

The subroutines are:

ReconRefFrames
    ExpandBlock
        IDCT
        Recon Filters (all)
    LoopFilter (all vert and horz)
    Copy Routines (CopyRecon etc..)
    UpdateUMVBorder


So,
This last part of the algorithm can be done completly in hardware and
the resulting decoded Frame can be sent directly to the D/A converter
and shown in the Video Monitor.

It means that the software only sends data to the hardware and never
needs to read back, because the hardware can send directly to the
Video Output.

The VHDL models are on the SVN, but the top-level module is still
under validation tests.


-- 
________________________________________
Felipe Portavales <portavales at gmail.com>
Undergraduate Student - IC-UNICAMP
Computer Systems Laboratory
http://www.lsc.ic.unicamp.br