Re: [Schrodinger-devel] Question about schro_upsampled_frame_upsample

[Wladimir v. d. L. <la...@gm...>]

>
> Cool.  I've wanted to try this out to see how well it works.
> I'd be interested to know of any architectural problems in schro that
> would make it difficult to port to a massively parallel environment.


I ported all the image operations (conversion, frame arithmetic, wavelets,
motion compensation) but not the bitstream stuff. I think the serial nature
of decoding bitstreams and arithmetic decoding will make it hard to port to
a parralel environment, but I'm still thinking about it.

One way we're going to test decoding is to use the auxiliary bitstream
> information to encode an SHA1 sum of each picture reconstructed by the
> encoder.  Again, it's on the list.


This would be nice, the way I verified it was to decode a dirac movie to raw
yuv using the GPU and CPU implementation, then compare it with md5sum :')
Intra-frame now fully works.

Er, um, oops.  Yeah.  The default engine is intra-only.  I generally
> use misc/pydvdrip or misc/gst-transcode to create streams (usually from
> DVD rips), and there are several parameters that I fiddle with to get
> what I want.  You want engine 3.


Engine 3 crashes after a few frames on my testing movie for some reason,
with the error:

*** glibc detected *** gst-launch-0.10: double free or corruption (!prev):
0x000000000073bfc0 ***

Some other thread keeps waiting and the decoding process hangs indefinitly.

Maybe I should make some test streams available...


A small testing stream would be very useful.

> Is it possible use the output of the reference Dirac implementation as
> input
> > to Schrodinger? (I read that the bitstreams are compatible now).
>
> Give or take a few bugs, yes.


I'll give that a try then, it will be my only option until I get engine 3 to
work.

Wladimir