Re: [Patch] Speedup yuvdeinterlace

Hi,

Stefan M. Fendt schrieb:
> Am Samstag, den 22.09.2007, 22:45 +0200 schrieb Burkhard Plaum:
> 
>> Is this acceptable or did I miss some deeper logic here?
> 
> Without looking to deep into the patch, no, I don't think you broke it
> --- not with the current implementation. It was recursive once (and will
> probably be again, some day...) and there it would not be a good idea to
> leave one field out... but as it is implemented now,... :-)

I looked if the reconstructed output field is used by anything else
than writing to the file and it seems not be the case.
I already guessed, that there was some recursive algorithm planned....

> but currently, I have the same problem as Andrew... no time to do
> anything serious on bugs/errors/ideas with the deinterlacer. I am filled
> up with programming @work... and so I'd rather play with my little
> daughter (1,5yrs) than doing too much with a computer @home,
> currently... sorry.)

Understandable. Hows the current state of yucdeinterlace? Is it considered
usable/finished from the algorithms point of view? By looking at the output
images, I think, that it works quite nice.

There are some problems with memory underflows: The overshoot is handled
nicely after the end of the memory areas, but not always before them.

Some lines like:

blah = malloc(size + 2 * overshoot);
blah += overshoot;

/* Use blah */

blah -= overshoot;
free(blah);

would help to fix them. Also valgrind complains a lot about
uninitialized memory, especially in the beginning, when the
saved input frames aren't there yet. memsetting the
buffers to 0x00 (luma) and 0x80 (chroma) after allocating
helps here.

FYI, my ported version is here (100% valgrind-clean):
http://gmerlin.cvs.sourceforge.net/gmerlin/gmerlin/plugins/videofilters/yuvdeinterlace.h?view=markup
http://gmerlin.cvs.sourceforge.net/gmerlin/gmerlin/plugins/videofilters/yuvdeinterlace.c?view=markup

Burkhard

-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2005.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/