Sequence End units (or actually, sequence_end_codes) have up until now
not been decomposed; in fact due to a bug in cbs_mpeg2_split_fragment they
have mostly been treated as part of the preceding unit. So implement
decomposing them as preparation for fixing said bug.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Up until now, if an SEI messages was to be added to a fragment, it was
tried to add said SEI message to the first SEI NAL unit of the fragment
and if this SEI NAL unit already contained H264_NAL_SEI SEI messages (an
arbitrary limit imposed by cbs_h264), adding failed; if there was no SEI
NAL unit, a new one has been added.
With this commit, the fragment is searched for further NAL units to add
the SEI messages to. If all of them are full, a new SEI NAL unit is added.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
cbs is currently inconsistent regarding the opaque field that can be
used as a special argument to av_buffer_create in order to be used
during freeing the buffer: ff_cbs_alloc_unit_content and all the free
functions used name this parameter as if it should contain a pointer to
the unit whose content is about to be created; but both
ff_cbs_alloc_unit_content as well as ff_cbs_h264_add_sei_message
actually use a pointer to the CodedBitstreamContext as opaque. It should
actually be neither, because it is unneeded (as is evidenced by the fact
that none of the free functions use this pointer at all) and because it
ties the unit's content to the lifetime of other objects, although a
refcounted buffer is supposed to have its own lifetime that only ends
when its reference count reaches zero. This problem manifests itself in
the pointer becoming dangling.
The pointer to the unit can become dangling if another unit is added to
the fragment later as happens in the bitstream filters; in this case,
the pointer can point to the wrong unit (if the fragment's unit array
needn't be relocated) or it can point to where the array was earlier.
It can also become dangling if the unit's content is meant to survive
the resetting of the fragment it was originally read with. This applies
to the extradata of H.264 and HEVC.
The pointer to the context can become dangling if the context is closed
before the content is freed. Although this doesn't seem to happen right
now, it could happen, in particular if one uses different
CodedBitstreamContexts for in- and output.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
The minimum header is 8 bytes, the smallest bitstream that is passed to
the MB decode code is 4 bytes
Fixes: Timeout (35sec -> 18sec)
Fixes: 15800/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_EATQI_fuzzer-5684154517159936
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
The minimal size which unpack() will not fail on is 5 bytes
Fixes: Timeout (14sec -> 77ms) (testcase 15508)
Fixes: 15508/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_EATGV_fuzzer-5700053513011200
Fixes: 15996/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_EATGV_fuzzer-5751353223151616
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
since tf.pad is enabled, the conv2d(valid) changes back to its original behavior.
Signed-off-by: Guo, Yejun <yejun.guo@intel.com>
Signed-off-by: Pedro Arthur <bygrandao@gmail.com>
the reason to add this layer first is that vf_sr uses it in its
tensorflow model, and the next plan is to update the python script
to convert tf.pad into native model.
Signed-off-by: Guo, Yejun <yejun.guo@intel.com>
Signed-off-by: Pedro Arthur <bygrandao@gmail.com>
In function ff_h264_add_pixels4_8_mmi, there is no need to reset '%[ftmp0]'
to 0, because it's value has never changed since the start of the asm block.
This patch remove the redundant 'xor' and set src to zero once it was loaded.
In function ff_h264_idct_add_8_mmi, 'block' is seted to zero twice.
This patch removed the first setting zero operation and move the second one
after the load operation of block.
In function ff_h264_idct8_add_8_mmi, 'block' is seted to zero twice too.
This patch just removed the second setting zero operation.
This patch mainly simplifies the implementation of functions above,
the effect on the performance of whole h264 decoding process is not obvious.
According to the perf data, proportion of ff_h264_idct_add_8_mmi decreased from
0.29% to 0.26% and ff_h264_idct8_add_8_mmi decreased from 0.62% to 0.59% when decoding
H264 format on loongson 3A3000(For reference only , not very stable.).
Reviewed-by: Reimar Döffinger <Reimar.Doeffinger@gmx.de>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
The earlier version had three deficits:
1. It allowed to set the stream to RGB although this is not allowed when
the profile is 0 or 2.
2. If it set the stream to RGB, then it did not automatically set the
range to full range; the result was that one got a warning every time a
frame with color_config element was processed if the frame originally
had TV range and the user didn't explicitly choose PC range. Now one
gets only one warning in such a situation.
3. Intra-only frames in profile 0 are automatically BT.601, but if the
user wished another color space, he was not informed about his wishes
being unfulfillable.
The commit also improves the documentation about this.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
If the relevant elements (the color description elements for AV1 and the
VUI elements in general for H.264 (since 1156b507)) are absent, then their
correct values (usually meaning unknown) have already been inferred by
the reading process, so that it is unnecessary to initialize them again
in the av1/h264_metadata filters even when they were initially absent.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
1. The extra information in slice headers was parsed incorrectly:
In the first reading pass to derive the length of the extra information,
one should look at bits n, n + 9, n + 18, ... and check whether they
equal one (further extra information) or zero (end of extra information),
but instead bits n, n + 8, n + 16, ... were inspected. The second pass
of reading (where the length is already known and the bytes between the
length-determining bits are copied into a buffer) did not record what
was in bits n, n + 9, n + 18, ..., presuming they equal one. And during
writing, the bytes in the buffer are interleaved with set bits and
written. This means that if the detected length of the extra information
was greater than the real length, the output was corrupted. Fortunately
no sample is known that made use of this mechanism: The extra information
in slices is still marked as reserved in the specifications. cbs_mpeg2
is now ready in case this changes.
2. Furthermore, the buffer is now padded and slightly different, but
very similar code for reading resp. writing has been replaced by code
used for both. This was made possible by a new macro, the equivalent
to cbs_h2645's fixed().
3. These changes also made it possible to remove the extra_bit_slice
element from the MPEG2RawSliceHeader structure. Said element was always
zero except when the detected length of the extra information was less
than the real length.
4. The extra information in picture headers (which uses essentially the
same syntax as the extra information in slice headers) has simply been
forgotten. This meant that if this extra information was present, it was
discarded during reading; and unfortunately writing created invalid
bitstreams in this case (an extra_bit_picture - the last set bit of the
whole unit - indicated that there would be a further byte of data,
although the output didn't contain said data).
This has been fixed; both types of extra information are now parsed via
the same code and essentially passed through.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Up until now, a temporary variable was used and initialized every time a
value was read in CBS; if reading turned out to be successfull, this
value was overwritten (without having ever been looked at) with the
value read if reading was successfull; on failure the variable wasn't
touched either. Therefore these initializations can be and have been
removed.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
If a sequence display extension is read with colour_description equal to
zero, but a user wants to add one or more of the colour_description
elements, then the colour_description elements the user did not explicitly
request to be set are set to zero and not to the value equal to
unknown/unspecified (namely 2). A value of zero is not only inappropriate,
but explicitly forbidden. This is fixed by inferring the right default
values during the reading process if the elements are absent; moreover,
changing any of the colour_description elements to zero is now no longer
possible.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Fixes: null pointer passed as argument 2, which is declared to never be null
Fixes: 16008/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_SSA_fuzzer-5650582821404672 (this is a separate issue found in this testcase)
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
An incomplete image is not supported prior to this and will
not produce any output. This commit moves the failure before
time consuming operations.
Fixes: Timeout (81sec -> 76ms)
Fixes: 15723/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_BRENDER_PIX_fuzzer-5147265653538816
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
it is expected that there will be more files to support native mode,
so put all the dnn codes under libavfilter/dnn
The main change of this patch is to move the file location, see below:
modified: libavfilter/Makefile
new file: libavfilter/dnn/Makefile
renamed: libavfilter/dnn_backend_native.c -> libavfilter/dnn/dnn_backend_native.c
renamed: libavfilter/dnn_backend_native.h -> libavfilter/dnn/dnn_backend_native.h
renamed: libavfilter/dnn_backend_tf.c -> libavfilter/dnn/dnn_backend_tf.c
renamed: libavfilter/dnn_backend_tf.h -> libavfilter/dnn/dnn_backend_tf.h
renamed: libavfilter/dnn_interface.c -> libavfilter/dnn/dnn_interface.c
Signed-off-by: Guo, Yejun <yejun.guo@intel.com>
Signed-off-by: Pedro Arthur <bygrandao@gmail.com>
avio_read can successfully return even when less than the requested
amount of input was read. wavdec's bext parsing mistakenly assumed a
successful avio_read always read the full amount that was requested.
The result could be dictionary tags populated with partially
uninitialized values.
This change also fixes a broken assertion in wav_parse_bext_string that
was off-by-one, though no known current usage of that method hits that
broken case.
Chromium bug: 987270
Signed-off-by: Matt Wolenetz <wolenetz@chromium.org>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Replace the same code logic with ensure_playlist(), it's will
help reusable blocks of code.
Reviewed-by: Jun Zhao <barryjzhao@tencent.com>
Signed-off-by: vacingfang <vacingfang@tencent.com>
av_packet_unref have reseted the AVPacket, so don't need to call
reset_packet after that.
Reviewed-by: Steven Liu <lq@chinaffmpeg.org>
Signed-off-by: Jun Zhao <barryjzhao@tencent.com>
mpc8_q4_syms is an array of int8_t that is initialized using
values not in the range of an int8_t and that is only accessed via
a pointer to uint8_t in ff_init_vlc_sparse. The latter applies to all
the other *_bits and *_syms tables in mpc8huff.h, so make them all
unsigned.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Some of the VP3 arrays (namely vp31_intra_y_dequant, vp31_intra_c_dequant
and vp31_inter_dequant) are currently declared as array of (const) int8_t
despite them being only used to directly initialize an array of uint8_t.
vp31_inter_dequant even contains the value 128 which is not
representible in int8_t and might generate overflow warnings by
compilers.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com>
Reviewed-by: Peter Ross <pross@xvid.org>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
why change .4 to .25, it's for:
one scenecut(pkt_pts=20040) isn't detected by 0.4 threshold
why not change to 0.3 instead of 0.25:
it will miss the scenecut(pkt_pts=20040) after applying the next
patch which enables yuvj420
for fate testing, it's better to catch all scenecut scenes.
Reviewed-by: Marton Balint <cus@passwd.hu>
Signed-off-by: Limin Wang <lance.lmwang@gmail.com>
This patch does not make other pixel formats usable yet to make sure the test
result is the same with rgb32 format.
Reviewed-by: Marton Balint <cus@passwd.hu>
Signed-off-by: Limin Wang <lance.lmwang@gmail.com>
A value of zero for sample_count in trun box is not
prohibited by 14496-12 section 8.8.8. 4a9d32baca
disallowed this which led the demuxer to error out
when reading the header of valid files.
Andreas Rheinhardt
Michael Niedermayer
Mark Thompson
Guo, Yejun
Paul B Mahol
Steven Liu
Gyan Doshi
Shiyou Yin
Matt Wolenetz
vacingfang
Jun Zhao
Limin Wang
Baptiste Coudurier
Matthieu Bouron
Leo Zhang