MJPEG Tools Svn

Courtesy of Nicolas Boos:

Changelog :

* The "tmpgenc" quant. matrices are now the defaults.
* Fixed some audio track selection problems.
* Use ffmpeg instead of transcode for input framerate detection.
* Added new experimental "cvd_wide" output type.
* The script is a bit more verbose than before, it prints (egtools) some info about input and output streams.
* Reduced dependency to the transcode binary, now only needed for
detection of the pixel aspect ratio, and the length of streams.

The recent parallel-processing changes broke the -B switch. It's
fixed now.

Also fixed several problems related to thread termination.

Updated to explain the new "-p" option and to remove the "-s" option.

MotionSearcher.hh
SetRegion2D.hh
main.c
newdenoise.cc
newdenoise.hh

Implemented the first, easy version of parallelism. "-p 0" produces
the original behavior. "-p 1" moves the reading/writing of video
frames to separate threads. "-p 2" moves the denoising of color to a
separate thread. Some work was needed to separate their memory spaces
better, using private allocators for the set-region objects.

This does pretty well, but since color generally takes far less time to
denoise than intensity, it still doesn't fill up 2 processors. The
next step is to rewrite denoising so that 2 search-windows can process
different slices of the same frame at the same time, but that is a BIG
project.

Also, the "skip" option was removed.

Slightly more conservative defaults for the bandwidth percentages.

Add SSE code for y4mspatialfilter -- only 20% improvement :-(

Convert frame_i2f and frame_f2i to SSE, yields 20% performance improvement

use floats for intermediates (speeds things a little and changes 18% of the values!)

code reorg'd so there is one easy to vectorize function (convolveLine)

fix major bug in reversing vert/horiz filter processing order