libvpx (probably accidentally) clears the bits if error_res is set,
along with keyframe/intraonly. This probably wasn't the intention
(since it's local data), but it's behaviour we have to copy...
Makes no real difference, but maybe scares coverity less (CID1297578)
Reviewed-by: "Ronald S. Bultje" <rsbultje@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The practical effect of this is that the scaling will wrongly not be
applied to the interpolation edge (the 3/4 constants in this patch).
In other words, we clip to the pre-scaling interpolation, even though
these should be clipped post-scaling. The resulting out-of-frame MVs
are thus automatically clipped within the visible portion of the frame,
which is probably not the intention, but is unfortunately what libvpx
does, so we need to copy that behaviour.
This reproduces libvpx behaviour. It seems like it originally only
targeted loopfilter behaviour, but this unfortunately effects following
block contexting and thus directs bitstream sync.
This also fixes intra-only MT failures (it was the same bug), see trac
4526 and 4527.
Tested-by: James Almer <jamrial@gmail.com>
Reviewed-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The unaligned size is not handled in setctx_2d(), causing edges of images
to have improper segmentation prediction, which causes visual artifacts
at image edges a few frames later.
This is required if it's subsequently re-enabled with no value updates
(which means, use values that were previously set).
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
* commit '312daa15891dc7abb77a404fe927d5ee35c52a71':
vp9: Use the correct upper bound for seg_id
Conflicts:
libavcodec/vp9.h
libavcodec/vp9block.c
Not merged, the value was correct before as far as i can see
Merged-by: Michael Niedermayer <michaelni@gmx.at>
In vp9_decode_frame function, ff_thread_finish_setup is not called
when refreshctx is equal to 0, and the next decoding thread can not
start work until the cunrrent frame has been decoded completely. So
ff_thread_finish_setup needs to be called to enable Multi-thread
decoding in this condition.
Signed-off-by: Di Wu <di1028.wu@samsung.com>
Signed-off-by: Reynaldo H. Verdejo Pinochet <reynaldo@osg.samsung.com>
* commit 'eddf95283fbba1b469b76e7518bc3941f671b193':
vp9: initialize color space and range properties
Merged-by: Michael Niedermayer <michaelni@gmx.at>
* commit '6d69f9f37689c999815a65a2d99999fad3a41705':
vp9: write uveob as 16-bit value for 16x16/32x32 transforms.
Merged-by: Michael Niedermayer <michaelni@gmx.at>
This is needed for future AVX2 implementations
Signed-off-by: James Almer <jamrial@gmail.com>
Reviewed-by: "Ronald S. Bultje" <rsbultje@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
The advantage of this is that the is32x32 division branch in
decode_coeffs_b is removed from the inner loop to outside the block
coef decoding loop in decode_coeffs. Also, it allows us to merge the
txsz branches from the block coef decoding loop, the context merge
and the context split.
The directional intra predictors either don't care about order (dc, h,
dc_left, tm), or they prefer inverted order (vr, dr, hd). This allows
more efficient SIMD implementations.
The spec doesn't describe how it should be decoded so this is probably
the safest thing to do. Fixes valgrind errors on fuzzed11.ivf and fixes
valgrind errors on fuzzed10.ivf differently.
Fixes invalid reads and crashes in vp90-2-05-resize.webm and fuzzed6.ivf.
The output is still not identical to what libvpx does (because we don't
actually scale in MC).
Reviewed-by: ubitux
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Prevents some invalid memory accesses after resolution change in
vp90-2-05-resize.webm, and libvpx does this too.
Reviewed-by: ubitux
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
- The memcpy was completely wrong because
s->prob_ctx[s->framectxid].coef is a [4][2][2][6][6][3] array, whereas
s->prob.coef is a [4][2][2][6][6][11] array.
- The additional check was committed to ffmpeg by Ronald S. Bultje.
Signed-off-by: Anton Khirnov <anton@khirnov.net>
* commit 'ca96e337169093979d7c763064ad9dae12b3108c':
vp9: drop support for real (non-emulated) edges
Conflicts:
libavcodec/vp9block.c
Merged-by: Michael Niedermayer <michaelni@gmx.at>
The reference map is never used in such cases, but we accidently copied
it anyway. This could cause crashes if this map has not yet been
allocated. Fixes trac ticket 3188.
Reviewed-by: Clément Bœsch <u@pkh.me>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
It was previously int and would return error if decode_coeffs_b()
returns an error; however, that can never happen, so refactor all
that code to make all dependent functions return void also (all the
way up to decode_coeffs_sb()).
For a random 1080p sample, decoding time went from 9.7sec (1 threads)
to 6.0sec (2 threads) and 5.2sec (4 threads) in 2-pass decoding mode.
I don't have any samples that use the parallelmode feature, but the
gains should be higher.
We need more information from last/cur_frame than from reference
buffers, so we can use a simplified structure for reference buffers,
and then store mvs and segmentation map information in last/cur.
w and h are both read as uint16 + 1 so this can not happen. A similar
change was introduced in 97962b2 / 72ca830, with the
av_log()+AVERROR_INVALIDDATA form, suggesting it could be triggerable
somehow.
Change suggested by Ronald S. Bultje.
Those should not be necessary.
Original change by one of these developers:
Anton Khirnov <anton@khirnov.net>
Diego Biurrun <diego@biurrun.de>
Luca Barbato <lu_zero@gentoo.org>
Martin Storsjö <martin@martin.st>
See 97962b2 / 72ca830
vp8_rac_get_tree() is called with a tree of size 3, so the returned
value can not be outside [0;3]. All of the [0;3] cases are handled in
the switch, so the assert should not be triggerable by any means. A
similar change was introduced in 97962b2 / 72ca830, with the
av_log()+AVERROR_INVALIDDATA form, suggesting it could be triggerable
somehow. This assert might help static analyzer, or simply the reader.
The operands of an addition can be evaluated in any order, since
the addition isn't a sequence point. The only operators that
have a defined evaluation order are &&, ||, ?: and the sequence
operator ','.
This fixes fate-vp9 on ARM RVCT.
Original change by one of these developers:
Anton Khirnov <anton@khirnov.net>
Diego Biurrun <diego@biurrun.de>
Luca Barbato <lu_zero@gentoo.org>
Martin Storsjö <martin@martin.st>
See 97962b2 / 72ca830
Originally spotted and fixed by one of these developers:
Anton Khirnov <anton@khirnov.net>
Diego Biurrun <diego@biurrun.de>
Luca Barbato <lu_zero@gentoo.org>
Martin Storsjö <martin@martin.st>
See 97962b2 / 72ca830
* commit '458446acfa1441d283dacf9e6e545beb083b8bb0':
lavc: Edge emulation with dst/src linesize
Conflicts:
libavcodec/cavs.c
libavcodec/h264.c
libavcodec/hevc.c
libavcodec/mpegvideo_enc.c
libavcodec/mpegvideo_motion.c
libavcodec/rv34.c
libavcodec/svq3.c
libavcodec/vc1dec.c
libavcodec/videodsp.h
libavcodec/videodsp_template.c
libavcodec/vp3.c
libavcodec/vp8.c
libavcodec/wmv2.c
libavcodec/x86/videodsp.asm
libavcodec/x86/videodsp_init.c
Changes to the asm are not merged, they are left for volunteers or
in their absence for later.
The changes this merge introduces are reordering of the function
arguments
See: face578d56
Merged-by: Michael Niedermayer <michaelni@gmx.at>
Originally written by Ronald S. Bultje <rsbultje@gmail.com> and
Clément Bœsch <u@pkh.me>
Further contributions by:
Anton Khirnov <anton@khirnov.net>
Diego Biurrun <diego@biurrun.de>
Luca Barbato <lu_zero@gentoo.org>
Martin Storsjö <martin@martin.st>
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
Signed-off-by: Anton Khirnov <anton@khirnov.net>
