This commit implements a psychoacoustic system for the native Opus
encoder. Its unlike any other psychoacoustic system known since its
capable of using a lookahead to make better choices on how to treat the
current frame and how many bits to allocate for it (and future frames).
Also, whilst the main bulk of the analysis function has to run in a
single thread, the per-frame anaylsis functions does not modify the main
psychoacoustic context, so in the future it will be fairly trivial to
run those as slice threads.
Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
This patchset enhances Alexis Ballier's original patch and validates
it using Qualcomm's Venus hardware (driver recently landed upstream
[1]).
This has been tested on Qualcomm's DragonBoard 410c and 820c
Configure/make scripts have been validated on Ubuntu 10.04 and
16.04.
Tested decoders:
- h264
- h263
- mpeg4
- vp8
- vp9
- hevc
Tested encoders:
- h264
- h263
- mpeg4
Tested transcoding (concurrent encoding/decoding)
Some of the changes introduced:
- v4l2: code cleanup and abstractions added
- v4l2: follow the new encode/decode api.
- v4l2: fix display size for NV12 output pool.
- v4l2: handle EOS (EPIPE and draining)
- v4l2: vp8 and mpeg4 decoding and encoding.
- v4l2: hevc and vp9 support.
- v4l2: generate EOF on dequeue errors.
- v4l2: h264_mp4toannexb filtering.
- v4l2: fixed make install and fate issues.
- v4l2: codecs enabled/disabled depending on pixfmt defined
- v4l2: pass timebase/framerate to the context
- v4l2: runtime decoder reconfiguration.
- v4l2: add more frame information
- v4l2: free hardware resources on last reference being released
- v4l2: encoding: disable b-frames for upstreaming (patch required)
[1] https://lwn.net/Articles/697956/
System Level view:
v42l_m2m_enc/dec --> v4l2_m2m --> v4l2_context --> v4l2_buffers
Reviewed-by: Jorge Ramirez <jorge.ramirez-ortiz@linaro.org>
Reviewed-by: Alexis Ballier <aballier@gentoo.org>
Tested-by: Jorge Ramirez <jorge.ramirez-ortiz@linaro.org>
Signed-off-by: wm4 <nfxjfg@googlemail.com>
Fixes: runtime error: signed integer overflow: 161 * 13872281 cannot be represented in type 'int'
Fixes: 3295/clusterfuzz-testcase-minimized-4738998142500864
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Fixes: runtime error: signed integer overflow: 2147483647 + 1 cannot be represented in type 'int'
Fixes: 3361/clusterfuzz-testcase-minimized-5065842955911168
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Fixes: runtime error: signed integer overflow: 2147392585 + 524288 cannot be represented in type 'int'
Fixes: 3348/clusterfuzz-testcase-minimized-4809500517203968
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
VP9SharedContext needs to be the first member so its properties can be
safely accessed from hardware accelerators, without the need to share
the full VP9Context.
Fixes ticket #6674.
This is the equivalent to what 7d317d4706
did for the codec-specific options.
av_opt_copy has specific handling so it's fine that we already copied
the whole context before.
Signed-off-by: Reimar Döffinger <Reimar.Doeffinger@gmx.de>
Preload data in band filter 0-8 for better pipeline parallelization.
Signed-off-by: Kaustubh Raste <kaustubh.raste@imgtec.com>
Reviewed-by: Manojkumar Bhosale <Manojkumar.Bhosale@imgtec.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Remove loops and unroll as block sizes are known.
Signed-off-by: Kaustubh Raste <kaustubh.raste@imgtec.com>
Reviewed-by: Manojkumar Bhosale <Manojkumar.Bhosale@imgtec.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Optimize luma intra case by reducing conditional cases.
Signed-off-by: Kaustubh Raste <kaustubh.raste@imgtec.com>
Reviewed-by: Manojkumar Bhosale <Manojkumar.Bhosale@imgtec.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Seperate the filter processing in all strong, all weak and strong + weak cases.
Signed-off-by: Kaustubh Raste <kaustubh.raste@imgtec.com>
Reviewed-by: Manojkumar Bhosale <Manojkumar.Bhosale@imgtec.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Intended for use with hardware frames for which rawvideo is not
sufficient. Requires the trusted packet flag to be set - decoding
fails if not to avoid security issues (the wrapped AVFrame can
contain pointers to arbitrary data).
Whether the udu string should be freed depends on whether the SEI it
gets added to was created internally by cbs or externally by the bsf.
The current code frees it twice in the former case.
Should fix the following tsan warning:
WARNING: ThreadSanitizer: data race (pid=19806)
Read of size 4 at 0x7b84000012f0 by thread T9:
#0 worker src/libavcodec/frame_thread_encoder.c:66 (ffmpeg+0x0000007f349e)
[..]
Previous write of size 4 at 0x7b84000012f0 by main thread (mutexes: write M1395):
#0 ff_frame_thread_encoder_free src/libavcodec/frame_thread_encoder.c:239 (ffmpeg+0x0000007f379e)
[..]
Fixes: runtime error: index 5 out of bounds for type 'int8_t const[5]'
Fixes:3175/clusterfuzz-testcase-minimized-4736774054084608
Found-by: continuous fuzzing process https://github.com/google/oss-fuzz/tree/master/projects/ffmpeg
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
They are time consuming operations, performing them after the other checks
improves the speed with damaged input dramatically.
Fixes: Timeout
Fixes: 2928/clusterfuzz-testcase-4992812120539136
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>
