Commit Graph

14 Commits

Author SHA1 Message Date
Clément Bœsch
eb7efaa924 avfilter/dctdnoiz: fix slice_h computation
ceilf() can only work if the reminder of the division is not 0.

This fixes memory errors with for instance:
  ffmpeg -f lavfi -i testsrc=s=800x500 -threads 3 -vf dctdnoiz -frames:v 1 -f null -
2015-02-08 14:31:29 +01:00
Paul B Mahol
d0045312c3 avfilter/vf_dctdnoiz: Use av_malloc_array()
Signed-off-by: Paul B Mahol <onemda@gmail.com>
2015-02-01 16:47:36 +00:00
Michael Niedermayer
6de2f027cd avfilter/vf_dctdnoiz: use av_freep() to avoid leaving stale pointers in memory
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2014-12-14 15:56:32 +01:00
Clément Bœsch
c85e0ca68a avfilter/dctdnoiz: add slice threading 2014-08-13 19:32:00 +02:00
Clément Bœsch
d7594beede avfilter/dctdnoiz: remove a few indirections in idcts
No change in the generated asm.
2014-08-09 20:34:14 +02:00
Clément Bœsch
1ba7c6ead2 avfilter/dctdnoiz: use 32-bit (float) operations instead of 64 (double) for DCTs
This makes the code about 1.5x faster without any noticeable difference
in the output.
2014-08-08 20:00:14 +02:00
Clément Bœsch
eb16a6d229 avfilter/dctdnoiz: remove forward declarations after previous commit 2014-08-08 20:00:14 +02:00
Clément Bœsch
aaf82dc0fa avfilter/dctdnoiz: make color [de]correlation less clumsy
This has no impact on overall performance, since the block DCT taking
most of the time anyway.
2014-08-08 20:00:13 +02:00
Clément Bœsch
cec59eb63f avfilter/dctdnoiz: add 8x8 dct and make it the default
8x8 is about 5x faster than 16x16 on 1080p input. Since a block size of
8x8 makes the filter almost usable (time wise) and it's not obvious if
8x8 or 16x16 is better from a quality PoV (it really depends on the
input and parameters), the filter now defaults to 8x8, and as a result
libavfilter is micro bumped.
2014-08-07 21:24:12 +02:00
Clément Bœsch
06362ab750 avfilter/dctdnoiz: rewrite [f/i]dct
This removes the avcodec dependency and make the code almost twice as
fast. More to come.

The DCT factorization is based on "Fast and numerically stable
algorithms for discrete cosine transforms" from Gerlind Plonkaa &
Manfred Tasche (DOI: 10.1016/j.laa.2004.07.015).
2014-08-07 19:59:07 +02:00
Michael Niedermayer
325f6e0a97 Merge remote-tracking branch 'qatar/master'
* qatar/master:
  lavfi: do not export the filters from shared objects

Conflicts:
	libavfilter/af_amix.c
	libavfilter/af_anull.c
	libavfilter/asrc_anullsrc.c
	libavfilter/f_select.c
	libavfilter/f_settb.c
	libavfilter/split.c
	libavfilter/src_movie.c
	libavfilter/vf_aspect.c
	libavfilter/vf_blackframe.c
	libavfilter/vf_colorbalance.c
	libavfilter/vf_copy.c
	libavfilter/vf_crop.c
	libavfilter/vf_cropdetect.c
	libavfilter/vf_drawbox.c
	libavfilter/vf_format.c
	libavfilter/vf_framestep.c
	libavfilter/vf_frei0r.c
	libavfilter/vf_hflip.c
	libavfilter/vf_libopencv.c
	libavfilter/vf_lut.c
	libavfilter/vf_null.c
	libavfilter/vf_overlay.c
	libavfilter/vf_scale.c
	libavfilter/vf_transpose.c
	libavfilter/vf_unsharp.c
	libavfilter/vf_vflip.c

Merged-by: Michael Niedermayer <michaelni@gmx.at>
2013-10-29 11:58:11 +01:00
Clément Bœsch
b439ece51c lavfi/dctdnoiz: move DC normalization out of loops.
Make code slightly faster, simpler, clearer.

The filter is still slow as hell, and that change won't cause any
visible performance improvement (it still takes more than one minute to
process a single 1080p frame on a Core 2 here).
2013-05-26 22:41:02 +02:00
Clément Bœsch
bc1c90eabf lavfi/dctdnoiz: fix 10l typo in pr_height computation. 2013-05-15 03:19:49 +02:00
Clément Bœsch
7b43120c7c lavfi: add dctdnoiz filter. 2013-05-15 00:54:00 +02:00