GitLab

Filters were reverted for the release so the binaries wouldn't include
libavfilter (which wasn't yet used).

This reverts commit 1e6a4716.

Change-Id: Ib9790a544b442066265b3fe1ea96313be35eff62
Reviewed-by: Sebastien Blin <sebastien.blin@savoirfairelinux.com>

FFmpeg may check channels instead of channel_layout, so set both.

Change-Id: I25094b104529e7d5ce3476d580a43fc5a445f48d

Refactors the MediaEncoder class so it can be reused elsewhere in the
codebase.

Concretely, MediaEncoder now supports multiple streams and file formats.

Adds a unit test for MediaEncoder.

Change-Id: I27d6f2e8538809e01fef5c35978a0c35dcf8cae4
Reviewed-by: Sebastien Blin <sebastien.blin@savoirfairelinux.com>

Refactors out the common code from video and audio encoding/flushing
into a single method. This reuses code instead of duplicating it.

The method is public in the hopes of being able to reuse MediaEncoder
elsewhere in the codebase.

Allow caller to specify the stream index for the future case that the
AVFormatContext contains multiple streams (recording video and audio).

Change-Id: Ieae52bc453ef66d141a40819ca10fedbd38d9a86
Reviewed-by: Sebastien Blin <sebastien.blin@savoirfairelinux.com>

This reverts commit ea2933d4.

Change-Id: I624c07255f60186a0d1e6610a58956775f91b342

Change-Id: Ia84ed72a0ba48778fcdef404d87fd35f717f64ab

Adds dependency to libavfilter, which provides many useful utilities
such as hardware encoding, mixing audio or rotating video.

Access to libavfilter is done through the MediaFilter class, which takes
a filter graph string and AVFrames as input and outputs filtered
AVFrames.

Adds unit tests for MediaFilter.

Change-Id: I0bf3869a9f19ad9cbbe1fc5e2d37eb83ece66ccc
Reviewed-by: Sebastien Blin <sebastien.blin@savoirfairelinux.com>

Change-Id: I6ab8a0078f7b67b2342859e216ab4ff02d721736
Reviewed-by: Sebastien Blin <sebastien.blin@savoirfairelinux.com>

Change-Id: I86d2bfcf88a2a2b0eac50e4bb2b1c74359f84f76

FFmpeg has the convenient macro av_err2str, but it cannot compile using
g++ (https://ffmpeg.org/pipermail/libav-user/2013-January/003440.html).
Adds an equivalent function for C++.

Change-Id: Ie9eb2c0dfe68c58fd6a5ff82a3729ff63b736a69

Prevent deadlock by stop waiting on the
condition variable after a second.

Change-Id: If707fac389615333d1074b0973d693a84c9a974e

FFmpeg docs state that if avcodec_send_packet returns EAGAIN,
avcodec_receive_frame must be called; it is not an error.

Change-Id: I5e84a8a0103fe868056915b9c4ea2fc8d090f7ee

In newer versions, AVIOContext's read_packet cb no longer considers 0
as being EOF. Changes the read callbacks for SDP and RTP to return
AVERROR_EOF instead of 0.

Change-Id: Ica5bdc95913b4475f1e954a1141700fc88638a59

- Handles downsampling in the event of an core layer samplerate
  reduction.
- Handles urgent frames in case a call is incoming during a call.

Change-Id: I0bd3b97ef96595576bc5daef7bea24c3bb32cf26

Level 2.0 (0x14) can only handle up to 352x288 at 30 fps, which can trip
up the decoder (especially a hardware decoder) if the video is of a
higher resolution. Increase the default level to 4.1 (0x29), which can
handle up to 2,048x1,024 at 30 fps.

Change-Id: Icd3fea81a4ea45796bab9d5f985a8ae1e1b36cfd

This reverts commit c4b580a2.

Change-Id: If4a00bde338aa657c482ae18cc59902ed1be593c

The function yuv422_clear_to_black is called up to 3 times per frame.
Remove the superfluous calls. This function only does useful in a
conference, when there can be blank space in a frame.

Change-Id: I64588adba59daaf3677bad419a519432eec41dff
Reviewed-by: Sébastien Blin <sebastien.blin@savoirfairelinux.com>

Change-Id: I5cd9cc6dd57d3c6447b397cc486cb82c48b2ea85

Merges setupFromAudioData and setupFromVideoData to reduce redundancy

Change-Id: Id543d0775a7803b778796b9abc02c9decd2c7e15
Reviewed-by: Guillaume Roguez <guillaume.roguez@savoirfairelinux.com>

Change-Id: I2e23a83ae992c4ed0063dafff34d3a49ee006681
Reviewed-by: Andreas Traczyk <andreas.traczyk@savoirfairelinux.com>

some devices are providing their framerate ratios in a format
incompatible with FFmpeg encoder.
This patch sanitizes these entries to fit the FFmpeg code.

Change-Id: Ib25c33699749ddcdfdf9cefc853f24caae433609
Reviewed-by: Guillaume Roguez <guillaume.roguez@savoirfairelinux.com>

- Seperates coreaudio implementations for iOS and OS X.
- Manages input/ouput port overrides when layer is reset via calls
  to Dring::setAudioOutputDevice in response to an audio route override
  event listened for by the iOS client.
- TODO: use AVAudioSession API with graphs in Objective-C++, or
  optimally, in Swift in the iOS/OS X clients.

Change-Id: I6ba34a8bdce7bedd3b9e9ca9bbcd2cc8caccf3f4

This fixes the Debian stable build.

Change-Id: I5a59beb34e3379c8e950a4325ef62d3810be84f4

AVBufferRef leaked its contents after call to openDevice.
- openDevice renamed to initDevice to better reflect its purpose
- initDevice now responsible for all setup and cleanup of device

Change-Id: I2fc36b6b0fe84158bb62fa2dac0cf441716ac540
Reviewed-by: Guillaume Roguez <guillaume.roguez@savoirfairelinux.com>

Change-Id: I1ef88bc640f79790aad2b5285ea2c776aa8cd366

Change-Id: I2794c1d374654870985a655a42ac647ff66c97a1
Reviewed-by: Philippe Gorley <philippe.gorley@savoirfairelinux.com>

Change-Id: Ic59bdd5104d3516f6aadf66b6b92d63a7f0dd561

Incorrect condition was leading to bad pixel format conversion before
sending frames to client on macOS (BGR instead of correct RGB).

Change-Id: Ie76a488c4151d00249ed1adb5b0b02e43d2972eb
Reviewed-by: Andreas Traczyk <andreas.traczyk@savoirfairelinux.com>

Change-Id: Iacf37f3b6fc7016b3225066e164046b3db45c138

Falling back from hardware to software decoding requires only a
reinitialization of the AVCodecContext. Hardware decoding failure is no
longer bubbled up to the RTP session, which makes it much faster.

Change-Id: I79fdfcfa41f822b3299d74ac654146789fcfd97b

- adds a slightly altered version of the android video implementation,
  as a swift AVFoundation implementation will be used in the ios client
  to enumerate the video input devices.

Change-Id: I7c16fc39c5d22e54fb146881150b6cbcfddb7806

Replaces empty methods and std::bind calls with lambdas.

Change-Id: I0d9c876a9ba0bbdc6a77a85e0e2d75766312f952

libavutil now covers the simple cases for creating and managing devices.
As Ring does not need to fine-tune these processes, most of the accel
code can be and has been removed.

Most hardware decoders output NV12, so skip extra conversions by
outputting NV12. Said pixel format is supported by everything that isn't
excessively old.

Change-Id: I10c440026fc3b289dbba7ecbca47e55c57147207
Reviewed-by: Anthony Léonard <anthony.leonard@savoirfairelinux.com>

Bumps FFmpeg to match the other platforms. Modifies the acceleration
code to work under the new API.

Enables bitstream filters, as some of libavcodec's components, such as
VideoToolbox, use them.

Change-Id: I3a6cee2cf06881bba4602c0ed635ff45908e7b85
Reviewed-by: Kateryna Kostiuk <kateryna.kostiuk@savoirfairelinux.com>

Audio encoding is usually mono-threaded and doesn't require
as much CPU as video encoding. Limit to 4 threads since more
would not be useful and might introduce latency and decrease performance.

Video encoding takes significantly more resources, however for live encoding,
using more than 16 threads might introduce unreasonable latency.
Since new customer CPUs (AMD Ryzen and Intel Core i7/i9) now have
more than 16 logical cores, setting a higher bound of 16 threads.

Change-Id: I5dc2d3f51019c563b8cec3fbddffa13cbafdb9a5

Recent CPUs should be able to encode higher quality h264 in real-time
without any problems.

Allows users to encode h264 with higher quality if they so choose.

Change-Id: Ibbb01cbc55e20f497f99f56c6c0e6616ff0734bc

This attribute was useful with the old decoding API. Refcounted frames
are always used with the avcodec_send_packet/avcodec_receive_frame API.

Change-Id: I0904ed8731c329e0f8f3d59d1872bff6a26543be
Reviewed-by: Guillaume Roguez <guillaume.roguez@savoirfairelinux.com>

Handles RestartRequired return code on hardware acceleration failure.

Change-Id: I170928e23ed607c2255e20729a71adc1364bbb6a
Reviewed-by: Guillaume Roguez <guillaume.roguez@savoirfairelinux.com>

Fixes a regression where keyframe requests were sent by the wrong peer.

Corrects artifacts in a shorter amount of time.

Change-Id: I2b981b46c07422a4289f378e2a5deaaba0047a3b
Reviewed-by: Guillaume Roguez <guillaume.roguez@savoirfairelinux.com>

AV_INPUT_BUFFER_MIN_SIZE is the new name for FF_MIN_BUFFER_SIZE, which is
not available anymore in new releases.

Change-Id: I79abb3a8192d8572323f705eecc44edc974026f8
Reviewed-by: Philippe Gorley <philippe.gorley@savoirfairelinux.com>