The init functions marked as av_cold have to be executed in any case,
so there is no gain from trying to mark paths leading to such functions
as unlikely.
Instead, only extend edges on-demand when the motion vector actually
crosses the visible decoded area using ff_emulated_edge_mc(). This
changes decoding time for cathedral from 8.722sec to 8.706sec, i.e.
0.2% faster overall. More generally (VP8 uses this also), low-motion
content gets significant speed improvements, whereas high-motion content
tends to decode in approximately the same time.
Signed-off-by: Martin Storsjö <martin@martin.st>
Instead, keep them in the bitstream buffer until we read them verbatim,
this saves a memcpy() and a subsequent clearing of the target buffer.
decode_cabac+decode_mb for a sample file (CAPM3_Sony_D.jsv) goes from
6121.4 to 6095.5 cycles, i.e. 26 cycles faster.
Signed-off-by: Martin Storsjö <martin@martin.st>
Instead, only extend edges on-demand when the motion vector actually
crosses the visible decoded area using ff_emulated_edge_mc(). This
changes decoding time for cathedral from 8.722sec to 8.706sec, i.e.
0.2% faster overall. More generally (VP8 uses this also), low-motion
content gets significant speed improvements, whereas high-motion content
tends to decode in approximately the same time.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Instead, keep them in the bitstream buffer until we read them verbatim,
this saves a memcpy() and a subsequent clearing of the target buffer.
decode_cabac+decode_mb for a sample file (CAPM3_Sony_D.jsv) goes from
6121.4 to 6095.5 cycles, i.e. 26 cycles faster.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Most of the changes are just trivial are just trivial replacements of
fields from MpegEncContext with equivalent fields in H264Context.
Everything in h264* other than h264.c are those trivial changes.
The nontrivial parts are:
1) extracting a simplified version of the frame management code from
mpegvideo.c. We don't need last/next_picture anymore, since h264 uses
its own more complex system already and those were set only to appease
the mpegvideo parts.
2) some tables that need to be allocated/freed in appropriate places.
3) hwaccels -- mostly trivial replacements.
for dxva, the draw_horiz_band() call is moved from
ff_dxva2_common_end_frame() to per-codec end_frame() callbacks,
because it's now different for h264 and MpegEncContext-based
decoders.
4) svq3 -- it does not use h264 complex reference system, so I just
added some very simplistic frame management instead and dropped the
use of ff_h264_frame_start(). Because of this I also had to move some
initialization code to svq3.
Additional fixes for chroma format and bit depth changes by
Janne Grunau <janne-libav@jannau.net>
Signed-off-by: Anton Khirnov <anton@khirnov.net>
The sh4 optimizations are removed, because the code is
100% identical to the C code, so it is unlikely to
provide any real practical benefit.
Signed-off-by: Diego Biurrun <diego@biurrun.de>
Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
ref_list is constructed from other fields per slice when needed, so do
not copy it for both frame and slice threading.
default_ref_list is constructed per frame and still needs to be copied
to per-slice contexts for slice threading, but a copy is not needed for
frame threading.
Clobbering these tables will temporarily clobber the template used
as a basis for other threads to start decoding from. If the other
decoding thread updates from the template right at that moment,
subsequent threads will get invalid (or, usually, none at all) mmco
tables. This leads to invalid reference lists and subsequent decode
failures.
Therefore, instead, decode the mmco tables only for the first slice in
a field or frame. For other slices, decode the bits and ensure they
are identical to the mmco tables in the first slice, but don't ever
clobber the context state. This prevents other threads from using a
clobbered/invalid template as starting point for decoding, and thus
fixes decoding in these cases.
This fixes occasional (~1%) failures of h264-conformance-mr1_bt_a with
frame-multithreading enabled.
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Clobbering these tables will temporarily clobber the template used
as a basis for other threads to start decoding from. If the other
decoding thread updates from the template right at that moment,
subsequent threads will get invalid (or, usually, none at all) mmco
tables. This leads to invalid reference lists and subsequent decode
failures.
Therefore, instead, decode the mmco tables only for the first slice in
a field or frame. For other slices, decode the bits and ensure they
are identical to the mmco tables in the first slice, but don't ever
clobber the context state. This prevents other threads from using a
clobbered/invalid template as starting point for decoding, and thus
fixes decoding in these cases.
This fixes occasional (~1%) failures of h264-conformance-mr1_bt_a with
frame-multithreading enabled.
Signed-off-by: Luca Barbato <lu_zero@gentoo.org>
It is not posible to call get_buffer during frame-mt codec
initialization. Libavformat might pass huge amounts of data as
extradata after parsing broken files. The 'extradata' for the fuzzed
sample sample_varPAR_s5374_r001-02.avi is 2.8M large and contains
multiple slices.
This requires to move the avcodec_default_free_buffers() call to
ff_MPV_common_end() since otherwise delayed pictures would get freed
during a size change.
This fixes out of global array reads.
The alternative solutions of checking the index or modifying the VLC tables
to prevent the index going outside are each about 1-2 cpu cyclces slower
per coded 4x4 block.
The alternative of padding the global tables directly is more ugly and
moving them to the context should benefit cache locality.
Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
Conversion of the luma intra prediction mode to one of the constrained
("alzheimer") ones can happen by crafting special bitstreams, causing
a crash because we'll call a NULL function pointer for 16x16 block intra
prediction, since constrained intra prediction functions are only
implemented for chroma (8x8 blocks).
Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org
Yusuke Nakamura
Anton Khirnov
Michael Niedermayer
Diego Biurrun
Ronald S. Bultje
Mans Rullgard
Janne Grunau
Marton Balint
Lou Logan