Fixes "runtime error: member access within misaligned address 0xf00 for type
'struct bar', which requires # byte alignment" errors under GCC ubsan.
Reviewed-by: Nuo Mi <nuomi2021@gmail.com>
Signed-off-by: James Almer <jamrial@gmail.com>
As I understand, support for .STR files is broken for almost 10 years now (since 161442ff2c it seems).
Currently, ffmpeg fails with tons of errors like this on version 1 STRs, e.g. Wipeout 1:
[mdec @ 00000000027c72c0] ac-tex damaged at 1 9
What happens is that only the audio is present in the video file.
Anyway, that one character patch fixes the problem, video is now rendered.
Signed-off-by: aybe <aybe@users.noreply.github.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
tests/checkasm/checkasm:
C LSX LASX
hevc_idct_32x32_8_c: 1243.0 211.7 101.7
Speedup of decoding H265 4K 30FPS 30Mbps on
3A6000 with 8 threads is 1fps(56fps-->57fps).
Reviewed-by: yinshiyou-hf@loongson.cn
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
After this patch, the peformance of decoding H265 4K 30FPS 30Mbps
on 3A6000 with 8 threads improves 2fps (45fps-->47fsp).
Reviewed-by: yinshiyou-hf@loongson.cn
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
The layout for the frame flags is as follow:
chroma_format u(2)
reserved u(2)
interlace_mode u(2)
reserved u(2)
chroma_format has 2 allowed values:
0: reserved
1: reserved
2: 4:2:2
3: 4:4:4
interlace_mode has 3 allowed values:
0: progressive
1: tff
2: bff
3: reserved
0x80 is what we expect for "422 not interlaced", and the extra 0x2 from
0x82 is actually writing into the reserved bits.
This byte represents 4 reserved bits followed by 4 alpha_channel_type bits.
alpha_channel_type currently has 3 differents defined values: 0 (no
alpha), 1 (8b alpha), and 2 (16b alpha), all the other values are
reserved. The 4 initial reserved bits are expected to be 0.
This byte represents 4 reserved bits followed by 4 alpha_channel_type bits.
alpha_channel_type currently has 3 differents defined values: 0 (no
alpha), 1 (8b alpha), and 2 (16b alpha), all the other values are
reserved. This part is correctly written (alpha_bits>>3 does the correct
thing), but the 4 initial bits are reserved.
Quoting SMPTE RDD 36:2015:
A decoder shall abort if it encounters a bitstream with an unsupported
bitstream_version value. If 0, the value of the chroma_format syntax
element shall be 2 (4:2:2 sampling) and the value of the
alpha_channel_type element shall be 0 (no encoded alpha); if 1, any
permissible value may be used for those syntax elements.
So if we're not in 4:2:2 or if there is alpha, we are not allowed to use
version 0.
Quoting SMPTE RDD 36:2015:
A decoder shall abort if it encounters a bitstream with an unsupported
bitstream_version value. If 0, the value of the chroma_format syntax
element shall be 2 (4:2:2 sampling) and the value of the
alpha_channel_type element shall be 0 (no encoded alpha); if 1, any
permissible value may be used for those syntax elements.
So if we're not in 4:2:2 or if there is alpha, we are not allowed to use
version 0.
FIx warnings (soon to be errors in GCC 14, already so in Clang 15):
```
src/libavcodec/vulkan_av1.c: In function ‘vk_av1_create_params’:
src/libavcodec/vulkan_av1.c:183:43: error: initialization of ‘long long unsigned int’ from ‘void *’ makes integer from pointer without a cast [-Wint-conversion]
183 | .videoSessionParametersTemplate = NULL,
| ^~~~
src/libavcodec/vulkan_av1.c:183:43: note: (near initialization for ‘(anonymous).videoSessionParametersTemplate’)
```
Use Vulkan's VK_NULL_HANDLE instead of bare NULL.
Fix Trac ticket #10724.
Was reported downstream in Gentoo at https://bugs.gentoo.org/919067.
Signed-off-by: Sam James <sam@gentoo.org>
These functions encode a slice of alpha (1 to 8 macroblocks) which are
expected to be encoded as a repeated sequence of "[diff][run-1]", where
diff is the running difference of the alpha value and run is how many
times that value is expected to be duplicated (within the limit of a
grand total of 2048 unpacked samples, corresponding to a slice of 8 MB).
Even when run==0 (the run variable semantic is actually "run minus 1"),
there is always a diff previously encoded that needs a counter of at
least 1. This means we need to call put_alpha_run() unconditionally at
the end of the bitstream to account for the last running diff.
This commit fixes glitchy playbacks on QuickTime with M2 and M3 hardware
(but not M1 for some mysterious reason) with files generated with
commands such as:
ffmpeg -f lavfi -i testsrc2=d=5:s=912x320,chromakey -c:v prores_aw -profile:v 4 -y aw.mov
ffmpeg -f lavfi -i testsrc2=d=5:s=912x320,chromakey -c:v prores_ks -profile:v 4444 -y ks.mov
The glitch expresses itself deterministically as blinking black
rectangles on random frames (for example on frame 21, 54, 71, 79, ...).
Even with the proresdec from FFmpeg, overreads actually happens while
reading the run-minus-1 value (around val = get_bits(gb, 4) in
unpack_alpha()). This doesn't seem to cause any particular issue because
it simply overreads into the next slice, and because the decoder is
resilient, but it's still a problem.
The investigation leading to this fix was made possible because of paid
work for Jitter (https://jitter.video).
Fixes ticket #10255.
In the mov muxer (in mov_write_video_tag()), bits_per_coded_sample will
be written under certain conditions and is required to be 32 for the
transparency to be honored in QuickTime.
prores_kostya already has this setting but prores_anatoliy and
prores_videotoolbox didn't.
This commit removes the follow warning and error:
D3D12 WARNING: ID3D12CommandList::ResourceBarrier: Called on the same subresource(s) of
Resource(0x000002236E0E00D0:'Unnamed ID3D12Resource Object') in separate Barrier Descs
which is inefficient and likely unintentional. Desc[0] and Desc[1] on (subresource :
4294967295). [RESOURCE_MANIPULATION WARNING #1008: RESOURCE_BARRIER_DUPLICATE_SUBRESOURCE_TRANSITIONS]
D3D12 ERROR: ID3D12CommandList::ResourceBarrier: Before state (0x0: D3D12_RESOURCE_STATE_[COMMON|PRESENT])
of resource (0x000002236E0E00D0:'Unnamed ID3D12Resource Object') (subresource: 0) specified
by transition barrier does not match with the state (0x20000: D3D12_RESOURCE_STATE_VIDEO_DECODE_WRITE)
specified in the previous call to ResourceBarrier [RESOURCE_MANIPULATION ERROR #527:
RESOURCE_BARRIER_BEFORE_AFTER_MISMATCH]
Tested-by: Tong Wu <tong1.wu@intel.com>
Signed-off-by: Wu Jianhua <toqsxw@outlook.com>
Previous assertion was not useful. Now a warning is added to replace it.
For get_surface_index, we should return a zero index in case the index is not found.
But a warning is necessary to notify.
Signed-off-by: Tong Wu <tong1.wu@intel.com>
Previous max_num_refs was based on pp.frame_refs plus 1 and it could possibly
reaches the size limit. Actually it should be the size of pp.ref_frame_map
plus 1.
Signed-off-by: Tong Wu <tong1.wu@intel.com>
The start code is matched against 0x000001, zero_byte was treated
as last byte of last frame rather than the beginning of next frame.
Signed-off-by: Zhao Zhili <zhilizhao@tencent.com>
vvc decoder plug-in to avcodec.
split frames into slices/tiles and send them to vvc_thread for further decoding
reorder and wait for the frame decoding to be done and output the frame
Features:
+ Support I, P, B frames
+ Support 8/10/12 bits, chroma 400, 420, 422, and 444 and range extension
+ Support VVC new tools like MIP, CCLM, AFFINE, GPM, DMVR, PROF, BDOF, LMCS, ALF
+ 295 conformace clips passed
- Not support RPR, IBC, PALETTE, and other minor features yet
Performance:
C code FPS on an i7-12700K (x86):
BQTerrace_1920x1080_60_10_420_22_RA.vvc 93.0
Chimera_8bit_1080P_1000_frames.vvc 184.3
NovosobornayaSquare_1920x1080.bin 191.3
RitualDance_1920x1080_60_10_420_32_LD.266 150.7
RitualDance_1920x1080_60_10_420_37_RA.266 170.0
Tango2_3840x2160_60_10_420_27_LD.266 33.7
C code FPS on a M1 Mac Pro (ARM):
BQTerrace_1920x1080_60_10_420_22_RA.vvc 58.7
Chimera_8bit_1080P_1000_frames.vvc 153.3
NovosobornayaSquare_1920x1080.bin 150.3
RitualDance_1920x1080_60_10_420_32_LD.266 105.0
RitualDance_1920x1080_60_10_420_37_RA.266 133.0
Tango2_3840x2160_60_10_420_27_LD.266 21.7
Asm optimizations still working in progress. please check
https://github.com/ffvvc/FFmpeg/wiki#performance-data for the latest
Contributors (based on code merge order):
Nuo Mi <nuomi2021@gmail.com>
Xu Mu <toxumu@outlook.com>
Frank Plowman <post@frankplowman.com>
Shaun Loo <shaunloo10@gmail.com>
Wu Jianhua <toqsxw@outlook.com>
Thank you for reporting issues and providing performance reports:
Łukasz Czech <lukaszcz18@wp.pl>
Xu Fulong <839789740@qq.com>
Thank you for providing review comments:
Ronald S. Bultje <rsbultje@gmail.com>
James Almer <jamrial@gmail.com>
Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Co-authored-by: Xu Mu <toxumu@outlook.com>
Co-authored-by: Frank Plowman <post@frankplowman.com>
Co-authored-by: Shaun Loo <shaunloo10@gmail.com>
Co-authored-by: Wu Jianhua <toqsxw@outlook.com>
This is the main entry point for the CTU (Coding Tree Unit) decoder.
The code will divide the CTU decoder into several stages.
It will check the stage dependencies and run the stage decoder.
Co-authored-by: Xu Mu <toxumu@outlook.com>
Co-authored-by: Frank Plowman <post@frankplowman.com>
Co-authored-by: Shaun Loo <shaunloo10@gmail.com>
Co-authored-by: Wu Jianhua <toqsxw@outlook.com>
Fixes: out of array access
Fixes: 64081/clusterfuzz-testcase-minimized-ffmpeg_DEMUXER_fuzzer-6151006496620544
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Fixes: signed integer overflow: 2147478526 + 33924 cannot be represented in type 'int'
Fixes: shift exponent 32 is too large for 32-bit type 'unsigned int'
Fixes: 64243/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_JPEGLS_fuzzer-5195717848989696
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Forgot to init c->sao_edge_filter[idx] when idx=0/1/2/3.
After this patch, the speedup of decoding H265 4K 30FPS
30Mbps on 3A6000 is about 7% (42fps==>45fps).
Change-Id: I521999b397fa72b931a23c165cf45f276440cdfb
Reviewed-by: yinshiyou-hf@loongson.cn
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Rework the code a bit to speed up the 10-bit bitpacked decoding
routine. This is probably about as fast as I can get it without
switching to assembly language.
Demonstratable with:
./ffmpeg -f lavfi -i "smptehdbars=size=3840x2160" -c bitpacked -f image2 -frames:v 1 source.yuv
./ffmpeg -f bitpacked -pix_fmt yuv422p10le -s 3840x2160 -c:v bitpacked -i source.yuv -pix_fmt yuv422p10le out.yuv
On my development system, it went from 80ms for a 2160p frame
down to 20ms (i.e. a 4X speedup). Good enough for now, I hope...
Comments from Marton:
Originally on my system better performance could be achieved by simply
switching to the cached bitstream reader, but for Devin it was slower than
his direct byte operations.
I changed the order of writing output from u/y/v/y to u/v/y/y, and that made
the code faster than the cached bitstream reader on my system as well.
TIMER measurement of the decode loop on Ryzen 5 3600 with command line:
./ffmpeg -stream_loop 256 -threads 1 -f bitpacked -pix_fmt yuv422p10le -s 3840x2160 -c:v bitpacked -i source.yuv -pix_fmt yuv422p10le -f null none -loglevel error
Before: 823204127 decicycles in YUV, 256 runs, 0 skips
After: 315070524 decicycles in YUV, 256 runs, 0 skips
Signed-off-by: Devin Heitmueller <dheitmueller@ltnglobal.com>
Signed-off-by: Marton Balint <cus@passwd.hu>
The specification doesn't mention that clusters cannot have alphabet
sizes greater than 1 << bundle->log_alphabet_size, but the reference
implementation rejects these entropy streams as invalid, so we should
too. Refusing to do so can overflow a stack variable that should be
large enough otherwise.
Fixes#10738.
Found-by: Zeng Yunxiang and Li Zeyuan
Signed-off-by: Leo Izen <leo.izen@gmail.com>
If a sequence of JXL images is encapsulated in a container that has PTS
information, we should use the PTS information from the container. At
this time there is no container that does this, but if JPEG XL support
is ever added to NUT, AVTransport, or some other container, this commit
should allow the PTS information those containers provide to work as
expected.
Signed-off-by: Leo Izen <leo.izen@gmail.com>
These pixel formats have always been supported by libjxl, but at the
time this plugin was written, they were not in FFmpeg yet. Now that
they are in FFmpeg, we should support them.
Signed-off-by: Leo Izen <leo.izen@gmail.com>
These pixel formats have always been supported by libjxl, but at the
time this plugin was written, they were not in FFmpeg yet. Now that
they are in FFmpeg, we should support them.
Signed-off-by: Leo Izen <leo.izen@gmail.com>
FFmpeg doesn't support tv-range RGB throughout most of its pipeline, so
we should keep the warning. However, in case something does support it
we should at least keep it tagged properly. Additionally, the encoder
writes this tag if the space is tagged as such so this makes a round
trip work as it should.
Also, PNG doesn't support nonzero matrices but we only warn and ignore
in that case, so we have no reason to error out for illegal cICP ranges
either (i.e. greater than 1).
Signed-off-by: Leo Izen <leo.izen@gmail.com>
Reported-by: Kacper Michajłow <kasper93@gmail.com>
Unnecessary since acf63d5350adeae551d412db699f8ca03f7e76b9;
also avoids relocations.
Reviewed-by: Anton Khirnov <anton@khirnov.net>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This frame will be freed in the next line.
Reviewed-by: Zhao Zhili <quinkblack@foxmail.com>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
The current pack_output function pointer is a property of the decoder,
rather than a constant method provided by the DSP code. Indeed, except
for an unused initialisation, the field is never used in DSP code.
The penultimate loop iteration could pick any vl such that:
vlenb/4 < vl <= vlenb/2
Thus if the total length is not a multiple of vlenb/2, the vfadd.vf
on the penultimate iteration would yield corrupt values for the last
iteration.
To avoid this, force vl = vlen/2 until the last iteration. Unfortunately
this latent bug is not reproducible with either hardware or QEMU as of now.
This skips the round-trip to scalar register for the sliding 'x'
coefficients, improving performance by about 5%. The trick here is that
the vector slide-up instruction preserves elements in destination vector
until the slide offset.
The switch from vfslide1up.vf to vslideup.vi also allows the elimination
of data dependencies on consecutive slides. Since the specifications
recommend sticking to power of two offsets, we could slide as follows:
vslideup.vi v8, v0, 2
vslideup.vi v4, v0, 1
vslideup.vi v12, v8, 1
vslideup.vi v16, v8, 2
However in the device under test, this seems to make performance slightly
worse, so this is left for (in)validation with future better hardware.
This fixes the following build error:
src/libavcodec/d3d12va_decode.c:49:10: error: no previous prototype for function
'ff_d3d12va_get_surface_index' [-Werror,-Wmissing-prototypes]
49 | unsigned ff_d3d12va_get_surface_index(const AVCodecContext *avctx,
| ^
Signed-off-by: Martin Storsjö <martin@martin.st>
Same as d3d11va, this flag enables main still picture profile for
d3d12va. User should add this flag when decoding main still picture
profile.
Signed-off-by: Tong Wu <tong1.wu@intel.com>
The implementation is based on:
https://learn.microsoft.com/en-us/windows/win32/medfound/direct3d-12-video-overview
With the Direct3D 12 video decoding support, we can render or process
the decoded images by the pixel shaders or compute shaders directly
without the extra copy overhead, which is beneficial especially if you
are trying to render or post-process a 4K or 8K video.
The command below is how to enable d3d12va:
ffmpeg -hwaccel d3d12va -i input.mp4 output.mp4
Signed-off-by: Wu Jianhua <toqsxw@outlook.com>
Signed-off-by: Tong Wu <tong1.wu@intel.com>
From AOSP doc, these values are device and codec specific, but lower
values generally result in more efficient (smaller-sized) encoding.
For example, global_quality 50 on Pixel 6 results a 1080P 30 FPS
HEVC with 3744 kb/s, while global_quality 80 results 28178 kb/s.
Fix#10689
Signed-off-by: Zhao Zhili <zhilizhao@tencent.com>
With the way the runtime checks are currently set up, every single
openh264 release, no matter how minor, is considered an ABI break and
requires ffmpeg recompilation. This is unnecessarily strict because it
doesn't allow downstream distributions to ship any openh264 bug fix
version updates without breaking ffmpeg's openh264 support.
Years ago, at the time when ffmpeg's openh264 support was merged,
openh264 releases were done without a versioned soname (the library was
just libopenh264.so, unversioned). Since then, starting with version
1.3.0, openh264 has started using versioned sonames and the intent has
been to bump the soname every time there's a new release with an ABI
change.
This patch drops the exact version check and instead adds a minimum
requirement on 1.3.0 to the configure script.
Signed-off-by: Kalev Lember <klember@redhat.com>
Signed-off-by: Martin Storsjö <martin@martin.st>
The 8x4 and 4x4 use a needlessly large multiplier (unless/until we care
about embedded 64-bit-vector hardware). This is merely suboptimal.
The 8x4 case also uses an incorrect vector length, which leads to incorrect
behaviour on future/hypothetical hardware with 256-bit or larger vectors.
Pointed-out-by: Martin Storsjö <martin@martin.st>
We can't call ff_get_rv_vlenb() if we don't have RVV available
at all.
Acked-by: Rémi Denis-Courmont <remi@remlab.net>
Signed-off-by: Martin Storsjö <martin@martin.st>
The loop iterates over the length of the vector, not the order. This is
to avoid reloading the same data for each lag value. However this means
the loop only works if the maximum order is no larger than VLENB.
The loop is roughly equivalent to:
for (size_t j = 0; j < lag; j++)
autoc[j] = 1.;
while (len > lag) {
for (ptrdiff_t j = 0; j < lag; j++)
autoc[j] += data[j] * *data;
data++;
len--;
}
while (len > 0) {
for (ptrdiff_t j = 0; j < len; j++)
autoc[j] += data[j] * *data;
data++;
len--;
}
Since register pressure is only at 50%, it should be possible to implement
the same loop for order up to 2xVLENB. But this is left for future work.
Performance numbers are all over the place from ~1.25x to ~4x speedups,
but at least they are always noticeably better than nothing.
Fixes: out of array access
Fixes: 62603/clusterfuzz-testcase-minimized-ffmpeg_DEMUXER_fuzzer-5837632490569728
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Intended to replace https://patchwork.ffmpeg.org/project/ffmpeg/patch/20230802000135.26482-3-michael@niedermayer.cc/
with a more accurate block decoding magnitude bound.
Fixes: 62433/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_JPEG2000_fuzzer-5828618092937216
Fixes: 58299/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_JPEG2000_fuzzer-5828618092937216
Previous-version-reviewed-by: Tomas Härdin <git@haerdin.se>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
This patch allows the JXL parser to parse sequences of extremely small
files concatenated together. (e.g. smaller than the parser buffer)
Signed-off-by: Leo Izen <leo.izen@gmail.com>
According to ISO/IEC 14996-12, size == 1 means a 64-bit extended-size
field occurs *after* the 32-bit box type, not before. This fix should
allow correct parsing of JXL files with extended-size boxes.
Signed-off-by: Leo Izen <leo.izen@gmail.com>
The type of qsv decoders is FF_CODEC_CB_TYPE_DECODE which must not
return AVERROR(EAGAIN). commit 42b20c9 added an assertion to check the
returned value.
Signed-off-by: Haihao Xiang <haihao.xiang@intel.com>
Building iOS platform with arm64, the compiler has a warning: "instruction movi.2d with immediate #0 may not function correctly on this CPU, converting to movi.16b"
Signed-off-by: xufuji456 <839789740@qq.com>
Signed-off-by: Martin Storsjö <martin@martin.st>
This fixes corrupted audio for applications relying on ch_layout when
codec downmixing is active.
Signed-off-by: Geoffrey McRae <geoff@hostfission.com>
Signed-off-by: James Almer <jamrial@gmail.com>
Applications making use of this codec with the `downmix` option are
segfaulting unless the `ch_layout` is overridden after `avcodec_open2`
as can be seen in projects like MythTV[1]
This patch fixes this by overriding the ch_layout as done in other
decoders such as AC3.
1: af6f362a14/mythtv/libs/libmythtv/decoders/avformatdecoder.cpp (L4607)
Signed-off-by: Geoffrey McRae <geoff@hostfission.com>
Signed-off-by: James Almer <jamrial@gmail.com>
The input is laid out in 16 segments, of which 13 actually need to be
loaded. There are no really efficient ways to deal with this:
1) If we load 8 segments wit unit stride, then narrow to 16 segments with
right shifts, we can only get one half-size vector per segment, or just 2
elements per vector (EMUL=1/2) - at least with 128-bit vectors.
This ends up unsurprisingly about as fas as the C code.
2) The current approach is to load with strides. We keep that approach,
but improve it using three 4-segmented loads instead of 12 single-segment
loads. This divides the number of distinct loaded addresses by 4.
3) A potential third approach would be to avoid segmentation altogether
and splat the scalar coefficient into vectors. Then we can use a
unit-stride and maximum EMUL. But the downside then is that we have to
multiply the 3 (of 16) unused segments with zero as part of the
multiply-accumulate operations.
In addition, we also reuse vectors mid-loop so as to increase the EMUL
from 1 to 2, which also improves performance a little bit.
Oeverall the gains are quite small with the device under test, as it does
not deal with segmented loads very well. But at least the code is tidier,
and should enjoy bigger speed-ups on better hardware implementation.
Before:
ps_hybrid_analysis_c: 1819.2
ps_hybrid_analysis_rvv_f32: 1037.0 (before)
ps_hybrid_analysis_rvv_f32: 990.0 (after)
This stores the constant coefficients deinterleaved, so that they can be
loaded directly with NF=0. Unfortunately, we cannot optimise loading the
input, due to insufficient memory alignment (not 32-bit).
Before:
g722_apply_qmf_c: 82.5
g722_apply_qmf_rvv_i32: 78.2
After:
g722_apply_qmf_c: 82.5
g722_apply_qmf_rvv_i32: 65.2
Validate that a hw_frames_ctx is available before using it for
the AVHWAccel.free_frame_priv callback, and don't require it to
be present when the callback is not in use by the HWAccel.
v2: check for free_frame_priv (Hendrik)
v3: return EINVAL (Christoph Reiter)
v4: better commit message (Hendrik)
v5: fix typo with missed frames_ctx (Lynne)
See[1]: https://github.com/msys2/MINGW-packages/pull/19050
Fixes: be07145109 ("avcodec: add AVHWAccel.free_frame_priv callback")
CC: Lynne <dev@lynne.ee>
CC: Christoph Reiter <reiter.christoph@gmail.com>
Signed-off-by: Dmitry Rogozhkin <dmitry.v.rogozhkin@intel.com>
For fate-h264_mp4toannexb_ticket5927 and
fate-h264_mp4toannexb_ticket5927_2, they work by accident
previously. The sample file has two 'avc1' entries, and video
samples use the second one. It means packets should be decoded with
new extradata in side data. Before this patch, only extradata was
kept in the output, new extradata has been dropped. The output can
be decoded because the two extradata are almost the same, except
level indication. This patch fixed the issue, and add another
fate test.
Signed-off-by: Zhao Zhili <zhilizhao@tencent.com>
If there is a single group of SPS/PPS before an IDR frame, but no
SPS/PPS after that, we will miss the chance to reset
idr_sps_seen/idr_pps_seen. No SPS/PPS are inserted afterwards.
This patch saves in-band SPS/PPS and insert them before IDR frames
when necessary.
Signed-off-by: Zhao Zhili <zhilizhao@tencent.com>
start_code_size depends on whether PS comes from out-of-band or
in-band. Make the code more readable.
Signed-off-by: Zhao Zhili <zhilizhao@tencent.com>
This avoids SEI and IDR recovery flags affecting each other
Also eliminate litteral numbers from recovery handling
This should make the code clearer
Improves: tickets/4738/tickets_cut.ts
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
In this case, the inner loop computing the scalar product can be reduced
to just one multiplication and one sum even with 128-bit vectors. The
result is a lot simpler, but also brings more modest performance gains:
flac_lpc_16_13_c: 15241.0
flac_lpc_16_13_rvv_i32: 11230.0
flac_lpc_16_16_c: 17884.0
flac_lpc_16_16_rvv_i32: 12125.7
flac_lpc_16_29_c: 27847.7
flac_lpc_16_29_rvv_i32: 10494.0
flac_lpc_16_32_c: 30051.5
flac_lpc_16_32_rvv_i32: 10355.0
The entire set of 32 coefficients and corresponding past 32 samples can
fit in a single vector (with LMUL=8) exactly, but... since widening
double the needed vector sizes, we still end up too short with 128-bit
vectors. This adds a very simple version for future 256+-bit hardware,
and for pred_orders values up to 16, and a bit more involved loop for
for 128-bit hardware with pred_orders between 17 and 32.
With 128-bit hardware, the benchmarks look like this:
flac_lpc_32_13_c: 30152.0
flac_lpc_32_13_rvv_i32: 10244.7
flac_lpc_32_16_c: 37314.2
flac_lpc_32_16_rvv_i32: 10126.2
flac_lpc_32_29_c: 61910.0
flac_lpc_32_29_rvv_i32: 14495.2
flac_lpc_32_32_c: 68204.0
flac_lpc_32_32_rvv_i32: 13273.7
This commit adds support for VP8 bitstream read methods to the cbs
codec. This enables the trace_headers bitstream filter to support VP8,
in addition to AV1, H.264, H.265, and VP9. This can be useful for
debugging VP8 stream issues.
The CBS VP8 implements a simple VP8 boolean decoder using GetBitContext
to read the bitstream.
Only the read methods `read_unit` and `split_fragment` are implemented.
The write methods `write_unit` and `assemble_fragment` return the error
code AVERROR_PATCHWELCOME. This is because CBS VP8 write is unlikely to
be used by any applications at the moment. The write methods can be
added later if there is a real need for them.
TESTS: ffmpeg -i fate-suite/vp8/frame_size_change.webm -vcodec copy
-bsf:v trace_headers -f null -
Signed-off-by: Jianhui Dai <jianhui.j.dai@intel.com>
Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
This commit exports the `vp8_token_update_probs` variable to internal
library scope to facilitate its reuse within the library.
Signed-off-by: Jianhui Dai <jianhui.j.dai@intel.com>
Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
Better performance can probably be achieved with a more intricate
unrolled loop, but this is a start:
add_hfyu_left_pred_bgr32_c: 15084.0
add_hfyu_left_pred_bgr32_rvv_i32: 10280.2
This would actually be cleaner with the RISC-V P extension, but that is
not ratified yet (I think?) and usually not supported if V is supported.
With explicit unrolling, we can skip half of the sign bit flips, and
the compiler is then better able to optimise the scalar loop:
predictor_c: 31376.0 (before)
predictor_c: 23703.0 (after)
This is restricted to 128-bit vectors as larger vector sizes could read
past the end of the noise array. Support for future hardware with larger
vector sizes is left for some other time.
hf_apply_noise_0_c: 2319.7
hf_apply_noise_0_rvv_f32: 1229.0
hf_apply_noise_1_c: 2539.0
hf_apply_noise_1_rvv_f32: 1244.7
hf_apply_noise_2_c: 2319.7
hf_apply_noise_2_rvv_f32: 1232.7
hf_apply_noise_3_c: 2541.2
hf_apply_noise_3_rvv_f32: 1244.2
This commit fixes issue with missing SPS/PPS headers in video
encoded by AMF AV1 encoder.
Missing headers leads to broken seek in MPV video player.
Default value for property AV1_HEADER_INSERTION_MODE shouldn't be setup
to NONE (no headers insertion). We need to skip definition of this property,
because default value depends on USAGE property.
Signed-off-by: Dmitrii Ovchinnikov <ovchinnikov.dmitrii@gmail.com>
With 5 accumulator vectors and 6 inputs, this can only use LMUL=2.
Also the number of vector loop iterations is small, just 5 on 128-bit
vector hardware.
The vector loop is somewhat unusual in that it processes data in
descending memory order, in order to save on vector slides:
in descending order, we can extract elements to carry over to the next
iteration from the bottom of the vectors directly. With ascending order
(see in the Opus postfilter function), there are no ways to get the top
elements directly. On the downside, this requires the use of separate
shift and sub (the would-be SH3SUB instruction does not exist), with
a small pipeline stall on the vector load address.
The edge cases in scalar are done in scalar as this saves on loads
and remains significantly faster than C.
autocorrelate_c: 669.2
autocorrelate_rvv_f32: 421.0
Given the size of the data set, strided memory accesses cannot be avoided.
We can still do better than the current code.
ps_hybrid_synthesis_deint_c: 12065.5
ps_hybrid_synthesis_deint_rvv_i32: 13650.2 (before)
ps_hybrid_synthesis_deint_rvv_i64: 8181.0 (after)
Segmented loads may be slower than not. So this advantageously uses a
unit-strided load and narrowing shifts instead.
Before:
ps_add_squares_c: 60757.7
ps_add_squares_rvv_f32: 22242.5
After:
ps_add_squares_c: 60516.0
ps_add_squares_rvv_i64: 17067.7
Fixes: index -1 out of bounds for type 'CFrameBuffer [100]'
Fixes: 63877/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_FOURXM_fuzzer-5854263397711872
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
PGMYUV seems to be always limited range. This was a format originally
invented by FFmpeg at a time when YUVJ distinguished limited from full
range YUV, and this codec never appeared to output YUVJ in any
circumstance, so hard-coding limited range preserves the status quo.
The other formats are explicitly documented to be full range RGB/gray
formats. That said, don't tag them yet, due to outstanding bugs w.r.t
grayscale formats and color range handling.
This change in behavior updates a bunch of FATE tests in trivial ways
(added tagging being the only difference).
Partially fixes ticket #798
Reviewed-by: James Almer <jamrial@gmail.com>
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Peter Ross <pross@xvid.org>
This uses a more traditional approach allowing up processing of up to
period minus two elements per iteration. This also allows the algorithm
to work for all and any vector length.
As the T-Head C908 device under test can load 16 elements loop, there is
unsurprisingly a little performance drop when the period is minimal and
the parallelism is capped at 13 elements:
Before:
postfilter_15_c: 21222.2
postfilter_15_rvv_f32: 22007.7
postfilter_512_c: 20189.7
postfilter_512_rvv_f32: 22004.2
postfilter_1022_c: 20189.7
postfilter_1022_rvv_f32: 22004.2
After:
postfilter_15_c: 20189.5
postfilter_15_rvv_f32: 7057.2
postfilter_512_c: 20189.5
postfilter_512_rvv_f32: 5667.2
postfilter_1022_c: 20192.7
postfilter_1022_rvv_f32: 5667.2
As in the aligned case, we can use VLSE64.V, though the way of doing so
gets more convoluted, so the performance gains are more modest:
get_pixels_unaligned_c: 126.7
get_pixels_unaligned_rvv_i32: 145.5 (before)
get_pixels_unaligned_rvv_i64: 62.2 (after)
For the reference, those are the aligned benchmarks (unchanged) on the
same T-Head C908 hardware:
get_pixels_c: 126.7
get_pixels_rvi: 85.7
get_pixels_rvv_i64: 33.2
Fixes: out of array write
Fixes: 63520/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_FLIC_fuzzer-4876198087622656
Regression since: c7f8d42c12 (was not posted to ffmpeg-devel)
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Reviewed-by: Sean McGovern <gseanmcg@gmail.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
With 128-bit vectors, this is mostly pointless but also harmless.
Performance gains should be more noticeable with larger vector sizes.
neg_odd_64_c: 76.2
neg_odd_64_rvv_i64: 74.7
Up until now each thread had its own buffer pool for extradata
buffers when using frame-threading. Each thread can have at most
three references to extradata and in the long run, each thread's
bufferpool seems to fill up with three entries. But given
that at any given time there can be at most 2 + number of threads
entries used (the oldest thread can have two references to preceding
frames that are not currently decoded and each thread has its own
current frame, but there can be no references to any other frames),
this is wasteful. This commit therefore uses a single buffer pool
that is synced across threads.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Up until now, the VAAPI encoder uses fake data with the
AVBuffer-API: The data pointer does not point to real memory,
but is instead just a VABufferID converted to a pointer.
This has probably been copied from the VAAPI-hwcontext-API
(which presumably does it to avoid allocations).
This commit changes this without causing additional allocations
by switching to the RefStruct-pool API. This also fixes an
unchecked av_buffer_ref().
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This is in preparation for the following commit.
Reviewed-by: Anton Khirnov <anton@khirnov.net>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
It involves less allocations, in particular no allocations
after the entry has been created. Therefore creating a new
reference from an existing one can't fail and therefore
need not be checked. It also avoids indirections and casts.
Also note that nvdec_decoder_frame_init() (the callback
to initialize new entries from the pool) does not use
atomics to read and replace the number of entries
currently used by the pool. This relies on nvdec (like
most other hwaccels) not being run in a truely frame-threaded
way.
Tested-by: Timo Rothenpieler <timo@rothenpieler.org>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
It involves less allocations and therefore has the nice property
that deriving a reference from a reference can't fail,
simplifying hevc_ref_frame().
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
It involves less allocations and therefore has the nice property
that deriving a reference from a reference can't fail.
This allows for considerable simplifications in
ff_h264_(ref|replace)_picture().
Switching to the RefStruct API also allows to make H264Picture
smaller, because some AVBufferRef* pointers could be removed
without replacement.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Very similar to the AVBufferPool API, but with some differences:
1. Reusing an already existing entry does not incur an allocation
at all any more (the AVBufferPool API needs to allocate an AVBufferRef).
2. The tasks done while holding the lock are smaller; e.g.
allocating new entries is now performed without holding the lock.
The same goes for freeing.
3. The entries are freed as soon as possible (the AVBufferPool API
frees them in two batches: The first in av_buffer_pool_uninit() and
the second immediately before the pool is freed when the last
outstanding entry is returned to the pool).
4. The API is designed for objects and not naked buffers and
therefore has a reset callback. This is called whenever an object
is returned to the pool.
5. Just like with the RefStruct API, custom allocators are not
supported.
(If desired, the FFRefStructPool struct itself could be made
reference counted via the RefStruct API; an FFRefStructPool
would then be freed via ff_refstruct_unref().)
Reviewed-by: Michael Niedermayer <michael@niedermayer.cc>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This makes the code more testable as uninitialized fields are 0
and not random values from the last call
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
These entries do not correspond to VLC symbols that can be used
they do corrupt various variables like min/max bits
This also no longer assumes that there is a single non subtable
entry
Probably fixes some infinite loops too
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Fixes: shift exponent 32 is too large for 32-bit type 'int'
Fixes: 63151/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_HEVC_fuzzer-5067531154751488
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Fixes: signed integer overflow: 1900031961 + 553590817 cannot be represented in type 'int'
Fixes: 63061/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_APE_fuzzer-5166188298371072
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
The check is based on not infinite looping. It is likely
a more strict check can be done
Fixes: Infinite loop
Fixes: 62473/clusterfuzz-testcase-minimized-ffmpeg_BSF_EVC_FRAME_MERGE_fuzzer-5719883750703104
Fixes: 62765/clusterfuzz-testcase-minimized-ffmpeg_dem_EVC_fuzzer-6448531252314112
Fixes: 63378/clusterfuzz-testcase-minimized-ffmpeg_dem_MPEGPS_fuzzer-6504993844494336
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Reviewed-by: "Dawid Kozinski/Multimedia (PLT) /SRPOL/Staff Engineer/Samsung Electronics" <d.kozinski@samsung.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Everything besides VLC.table is basically write-only
and even VLC.table can be removed by accessing the
underlying table directly.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
For all VLCs here, the number of bits of the VLC is write-only,
because it is hardcoded at the call site. Therefore one can replace
these VLC structures with the only thing that is actually used:
The pointer to the VLCElem table. And in most cases one can even
avoid this.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Everything besides VLC.table is basically write-only
and even VLC.table can be removed by accessing the
underlying tables directly.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
For some VLCs here, the number of bits of the VLC is
write-only, because it is hardcoded at the call site.
Therefore one can replace these VLC structures with
the only thing that is actually used: The pointer
to the VLCElem table.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This allows to avoid the relocations inherent in an array
to individual tables; it also reduces padding.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
For all VLCs here, the number of bits of the VLC is
write-only, because it is hardcoded at the call site.
Therefore one can replace these VLC structures with
the only thing that is actually used: The pointer
to the VLCElem table.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
ff_ps_init() initializes some tables for AAC parametric stereo
and some of them are only valid for the fixed- or floating-point
decoder, whereas others (namely VLCs) are valid for both.
The latter are therefore initialized by ff_ps_init_common()
and because the two versions of ff_ps_init() can be run
concurrently, it is guarded by an AVOnce.
Yet now that there is ff_aacdec_common_init_once() there is
a better way to do this: Call ff_ps_init_common()
from ff_aacdec_common_init_once(). That way there is no need
to guard ff_ps_init_common() by an AVOnce any more.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
This allows to avoid the relocations inherent in a table
to individual tables; it also reduces padding.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
It allows to replace code tables of type uint32_t or uint16_t
by symbols of type uint8_t. It is also faster.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
For all VLCs here, the number of bits of the VLC is
write-only, because it is hardcoded at the call site.
Therefore one can replace these VLC structures with
the only thing that is actually used: The pointer
to the VLCElem table.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
The VLCs, their init code and the tables used for initialization
are currently duplicated for the floating- and fixed-point decoders.
This commit stops doing so and moves this stuff to aacdec_common.c.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
For all VLCs here, the number of bits of the VLC is
write-only, because it is hardcoded at the call site.
Therefore one can replace these VLC structures with
the only thing that is actually used: The pointer
to the VLCElem table. And in some cases one can even
avoid this.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
h->long_ref isn't guaranteed to be contiguously filled. Use the approach
from both vaapi_h264 and vdpau_h264 which goes through the 16 frames in
h->long_ref to find the LTR entries.
Fixes MR2_MW_A.264 from JVT-AVC_V1.
The fixed-point decoder actually does not use the floating-point
tables initialized by ff_aac_tableinit() at all. So don't
initialize them for it; instead merge initializing these tables
into ff_aac_float_common_init() which is already the function
for the common static initializations of the floating-point
AAC decoder and the (also floating-point) AAC encoder.
Doing so saves also one AVOnce.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
They (as well as their init code) are currently duplicated
for the floating- and fixed-point decoders.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
For all VLCs here, the number of bits of the VLC is write-only,
because it is hardcoded at the call site. Therefore one can replace
these VLC structures with the only thing that is actually used:
The pointer to the VLCElem table. And in some cases one can even
avoid this.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
For all VLCs here, the number of bits of the VLC is
write-only, because it is hardcoded at the call site.
Therefore one can replace these VLC structures with
the only thing that is actually used: The pointer
to the VLCElem table. And in some cases one can even
avoid this.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Everything besides VLC.table is basically write-only
and even VLC.table can be removed by accessing the
underlying table directly.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Everything besides VLC.table is basically write-only
and even VLC.table can be removed by accessing the
underlying table directly.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
For all VLCs here, the number of bits of the VLC is
write-only, because it is hardcoded at the call site.
Therefore one can replace these VLC structures with
the only thing that is actually used: The pointer
to the VLCElem table. And in some cases one can even
avoid this.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Everything besides VLC.table is basically write-only
and even VLC.table can be removed by accessing the
underlying table directly.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Everything besides VLC.table is basically write-only
and even VLC.table can be removed by accessing the
underlying table directly.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Everything besides VLC.table is basically write-only
and even VLC.table can be removed by accessing the
underlying table directly.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Everything besides VLC.table is basically write-only
and even VLC.table can be removed by accessing the
underlying table directly.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Everything besides VLC.table is basically write-only
and only VLC.table needs to be retained.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Everything besides VLC.table is basically write-only
and even VLC.table can be removed by accessing the
underlying table directly.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Everything besides VLC.table is basically write-only
and even VLC.table can be removed by accessing the
underlying table directly.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Everything besides VLC.table is basically write-only
and even VLC.table can be removed by accessing the
underlying table directly.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Said object is only allowed to be modified during its
initialization and is immutable afterwards.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
For most VLCs here, the number of bits of the VLC is
write-only, because it is hardcoded at the call site.
Therefore one can replace these VLC structures with
the only thing that is actually used: The pointer
to the VLCElem table.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
These VLCs are very big: The VP3 one have 164382 elements
but due to the overallocation enough memory for 313344 elements
are allocated (1.195 MiB with sizeof(VLCElem) == 4);
for VP4 the numbers are very similar, namely 311296 and 164392
elements. Since 1f4cf92cfb, each
frame thread has its own copy of these VLCs.
This commit fixes this by sharing these VLCs across threads.
The approach used here will also make it easier to support
stream reconfigurations in case of frame-multithreading
in the future.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Of all these VLCs here, only VLC.table was really used
after init, so use the ff_vlc_init_tables API
to get rid of them.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
Of all these VLCs here, only VLC.table was really used
after init, so use the ff_vlc_init_tables API
to get rid of them.
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>
