Previously, there was no way to flush an encoder such that after
draining, the encoder could be used again. We generally suggested
that clients teardown and replace the encoder instance in these
situations. However, for at least some hardware encoders, the cost of
this tear down/replace cycle is very high, which can get in the way of
some use-cases - for example: segmented encoding with nvenc.
To help address that use case, we added support for calling
avcodec_flush_buffers() to nvenc and things worked in practice,
although it was not clearly documented as to whether this should work
or not. There was only one previous example of an encoder implementing
the flush callback (audiotoolboxenc) and it's unclear if that was
intentional or not. However, it was clear that calling
avocdec_flush_buffers() on any other encoder would leave the encoder in
an undefined state, and that's not great.
As part of cleaning this up, this change introduces a formal capability
flag for encoders that support flushing and ensures a flush call is a
no-op for any other encoder. This allows client code to check if it is
meaningful to call flush on an encoder before actually doing it.
I have not attempted to separate the steps taken inside
avcodec_flush_buffers() because it's not doing anything that's wrong
for an encoder. But I did add a sanity check to reject attempts to
flush a frame threaded encoder because I couldn't wrap my head around
whether that code path was actually safe or not. As this combination
doesn't exist today, we'll deal with it if it ever comes up.
Currently the frame pool used by the default get_buffer2()
implementation is a single struct, allocated when opening the decoder.
A pointer to it is simply copied to each frame thread and we assume that
no thread attempts to modify it at an unexpected time. This is rather
fragile and potentially dangerous.
With this commit, the frame pool is made refcounted, with the reference
being propagated across threads along with other context variables. The
frame pool is now also immutable - when the stream parameters change we
drop the old reference and create a new one.
Fixes: signed integer overflow: 1677721600 * 32 cannot be represented in type 'int'
Fixes: 18885/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_FFWAVESYNTH_fuzzer-5741242185154560
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Some decoders may not need a writable buffer in some specific cases, but only
a reference to the existing buffer with updated frame properties instead, for
the purpose of returning duplicate frames. For this, the
FF_REGET_BUFFER_FLAG_READONLY flag is added, which will prevent potential
allocations and buffer copies when they are not needed.
Signed-off-by: James Almer <jamrial@gmail.com>
-1 will be map to error number "EPERM", and will be map to the error
message like "Error while decoding stream #0:0: Operation not permitted",
it's a strange error message when debug update_frame_pool fail,
now only return the error code from av_image_fill_pointers in case
of av_image_fill_pointers failure.
Signed-off-by: Jun Zhao <mypopydev@gmail.com>
This reverts commit 662558f985.
The avcodec_parameters_to_context() call was freeing and reallocating
AVCodecContext->extradata, essentially taking ownership of it, which according
to the doxy is user owned. This is an API break and has produces crashes in
some library users like Firefox.
Revert until a better solution is found to internally propagate the filtered
extradata back into the decoder context.
Signed-off-by: James Almer <jamrial@gmail.com>
This reverts commit f631c328e6.
The avcodec_parameters_to_context() call was freeing and reallocating
AVCodecContext->extradata, essentially taking ownership of it, which according
to the doxy is user owned. This is an API break and has produced crashes in
some library users like Firefox[1].
Revert until a better solution is found to internally propagate the filtered
extradata back into the decoder context, or a decision is made to change the
API.
[1] https://bugzilla.mozilla.org/show_bug.cgi?id=1486080
Signed-off-by: James Almer <jamrial@gmail.com>
Certain AVCodecParameters, like the contents of the extradata, may be changed
by the init() function of any of the bitstream filters in the chain.
Signed-off-by: James Almer <jamrial@gmail.com>
Initialize the bsfs once when opening the codec and uninitialize them once when
closing it, instead of at every codec flush/seek.
Signed-off-by: James Almer <jamrial@gmail.com>
Fixes a bug that would prevent using multiple comma-separated filters,
and allows options to be passed to each filter.
Based on similar loop in ffmpeg_opt.c's new_output_stream().
Signed-off-by: Aman Gupta <aman@tmm1.net>
STRIDE_ALIGN is not known in libavutil so av_image_check_size* cannot consider it
Fixes: OOM
Fixes: 8291/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_SNOW_fuzzer-5176528009691136
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
The pseudo palette allocation is optional now. But if it's still
allocated (like the internal get_buffer2 implementation does, for
compatibility), it shouldn't print a warning.
PSEUDOPAL pixel formats are not paletted, but carried a palette with the
intention of allowing code to treat unpaletted formats as paletted. The
palette simply mapped the byte values to the resulting RGB values,
making it some sort of LUT for RGB conversion.
It was used for 1 byte formats only: RGB4_BYTE, BGR4_BYTE, RGB8, BGR8,
GRAY8. The first 4 are awfully obscure, used only by some ancient bitmap
formats. The last one, GRAY8, is more common, but its treatment is
grossly incorrect. It considers full range GRAY8 only, so GRAY8 coming
from typical Y video planes was not mapped to the correct RGB values.
This cannot be fixed, because AVFrame.color_range can be freely changed
at runtime, and there is nothing to ensure the pseudo palette is
updated.
Also, nothing actually used the PSEUDOPAL palette data, except xwdenc
(trivially changed in the previous commit). All other code had to treat
it as a special case, just to ignore or to propagate palette data.
In conclusion, this was just a very strange old mechnaism that has no
real justification to exist anymore (although it may have been nice and
useful in the past). Now it's an artifact that makes the API harder to
use: API users who allocate their own pixel data have to be aware that
they need to allocate the palette, or FFmpeg will crash on them in
_some_ situations. On top of this, there was no API to allocate the
pseuo palette outside of av_frame_get_buffer().
This patch not only deprecates AV_PIX_FMT_FLAG_PSEUDOPAL, but also makes
the pseudo palette optional. Nothing accesses it anymore, though if it's
set, it's propagated. It's still allocated and initialized for
compatibility with API users that rely on this feature. But new API
users do not need to allocate it. This was an explicit goal of this
patch.
Most changes replace AV_PIX_FMT_FLAG_PSEUDOPAL with FF_PSEUDOPAL. I
first tried #ifdefing all code, but it was a mess. The FF_PSEUDOPAL
macro reduces the mess, and still allows defining FF_API_PSEUDOPAL to 0.
Passes FATE with FF_API_PSEUDOPAL enabled and disabled. In addition,
FATE passes with FF_API_PSEUDOPAL set to 1, but with allocation
functions manually changed to not allocating a palette.
This is for applications which want to explicitly check for invalid
UTF-8 manually, and take actions that are better than dropping invalid
subtitles silently. (It's pretty much silent because sporadic avcodec
error messages are so common that you can't reasonably display them in a
prominent and meaningful way in a application GUI.)
And remove the function altogether while at it. It's a duplicate of
another.
Reviewed-by: wm4 <nfxjfg@googlemail.com>
Signed-off-by: James Almer <jamrial@gmail.com>
This number is definitely required when frame threading is enabled, so
add it here rather than forcing all users to handle it themselves.
DXVA2 contained this addition in specific code as well (therefore being
added twice in the internal case) - just remove it from there.
AVCodecContext.extra_hw_frames is added to the size of hardware frame
pools created by libavcodec for APIs which require fixed-size pools.
This allows the user to keep references to a greater number of frames
after decode, which may be necessary for some use-cases.
It is also added to the initial_pool_size value returned by
avcodec_get_hw_frames_parameters() if a fixed-size pool is required.
This removes the dependency that hardware pixel formats previously had on
AVHWAccel instances, meaning only those which actually do something need
exist after this patch.
Also updates avcodec_default_get_format() to be able to choose hardware
formats if either a matching device has been supplied or no additional
external configuration is required, and avcodec_get_hw_frames_parameters()
to use the hardware config rather than searching the old hwaccel list.
In commit 061a0c14bb ("decode: restructure the core decoding code"), the
deprecated avcodec_decode_* APIs were reworked so that they called into the
new avcodec_send_packet / avcodec_receive_frame API. This had the side effect
of prohibiting sending new packets containing data after a drain
packet, but in previous versions of FFmpeg this "worked" and some
applications relied on it.
To restore some compatibility, reset the codec if we receive a new non-drain
packet using the old API after draining has completed. While this does
not give the same behaviour as the old API did, in the majority of cases
it works and it does not require changes to any other part of the decoding
code.
Fixes ticket #6775
Signed-off-by: James Cowgill <jcowgill@debian.org>
Signed-off-by: Marton Balint <cus@passwd.hu>
This removes the dependency that hardware pixel formats previously had on
AVHWAccel instances, meaning only those which actually do something need
exist after this patch.
Also updates avcodec_default_get_format() to be able to choose hardware
formats if either a matching device has been supplied or no additional
external configuration is required, and avcodec_get_hw_frames_parameters()
to use the hardware config rather than searching the old hwaccel list.
The FF_CODEC_CAP_HWACCEL_REQUIRE_CLASS mechanism is deleted because it
no longer does anything (the codec already contains the pointers to the
matching hwaccels).
There is no reason to keep this intact when decoding failed, specially
as private_ref is supposed to always be NULL when a frame is returned to
the user.
Currently, AVHWAccels are looked up using a (codec_id, pixfmt) tuple.
This means it's impossible to have 2 decoders for the same codec and
using the same opaque hardware pixel format.
This breaks merging Libav's CUVID hwaccel. FFmpeg has its own CUVID
support, but it's a full stream decoder, using NVIDIA's codec parser.
The Libav one is a true hwaccel, which is based on the builtin software
decoders.
Fix this by introducing another field to disambiguate AVHWAccels, and
use it for our CUVID decoders. FF_CODEC_CAP_HWACCEL_REQUIRE_CLASS makes
this mechanism backwards compatible and optional.
This will be useful in the CUVID hwaccel. It should also eventually
replace current decoder-specific mechanisms used by various other
hwaccels.
Merges Libav commit 704311b294.
If the get_buffer() call fails, the frame might have some side data
already set. Make sure it gets freed.
Merges Libav commit de77671438.
Signed-off-by: James Almer <jamrial@gmail.com>
Commit b46a77f19d accidentally broke this (requested change that was
added to the patch later and which was not fully tested).
Signed-off-by: Mark Thompson <sw@jkqxz.net>
James Almer
Michael Niedermayer
Marton Balint
Mark Thompson
Philip Langdale
Anton Khirnov
Limin Wang
vectronic
Jun Zhao
Gyan Doshi
Carl Eugen Hoyos
Jun Zhao
Aman Gupta
wm4
James Cowgill
Timo Rothenpieler