mirrors/gstreamer
mirrors/gstreamer
1
1
Fork 0
mirror of https://gitlab.freedesktop.org/gstreamer/gstreamer.git synced 2024-11-25 19:21:06 +00:00
Code Issues 1 Projects Releases Wiki Activity
gstreamer/subprojects/gstreamer-vaapi/NEWS
Tim-Philipp Müller f13c65d977 Release 1.22.0
2023-01-23 19:41:07 +00:00

1239 lines
50 KiB
Text
Raw Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

GStreamer 1.22 Release Notes
GStreamer 1.22.0 was originally released on 23 January 2023.
See https://gstreamer.freedesktop.org/releases/1.22/ for the latest
version of this document.
Last updated: Monday 23 January 2023, 17:00 UTC (log)
Introduction
The GStreamer team is proud to announce a new major feature release in
the stable 1.x API series of your favourite cross-platform multimedia
framework!
As always, this release is again packed with many new features, bug
fixes and other improvements.
Highlights
- AV1 video codec support improvements
- New HLS, DASH and Microsoft Smooth Streaming adaptive streaming
clients
- Qt6 support for rendering video inside a QML scene
- Minimal builds optimised for binary size, including only the
individual elements needed
- Playbin3, Decodebin3, UriDecodebin3, Parsebin enhancements and
stabilisation
- WebRTC simulcast support and support for Google Congestion Control
- WebRTC-based media server ingestion/egress (WHIP/WHEP) support
- New easy to use batteries-included WebRTC sender plugin
- Easy RTP sender timestamp reconstruction for RTP and RTSP
- ONVIF timed metadata support
- New fragmented MP4 muxer and non-fragmented MP4 muxer
- New plugins for Amazon AWS storage and audio transcription services
- New gtk4paintablesink and gtkwaylandsink renderers
- New videocolorscale element that can convert and scale in one go for
better performance
- High bit-depth video improvements
- Touchscreen event support in navigation API
- Rust plugins now shipped in macOS and Windows/MSVC binary packages
- H.264/H.265 timestamp correction elements for PTS/DTS reconstruction
before muxers
- Improved design for DMA buffer sharing and modifier handling for
hardware-accelerated video decoders/encoders/filters and
capturing/rendering on Linux
- Video4Linux2 hardware accelerated decoder improvements
- CUDA integration and Direct3D11 integration and plugin improvements
- New H.264 / AVC, H.265 / HEVC and AV1 hardware-accelerated video
encoders for AMD GPUs using the Advanced Media Framework (AMF) SDK
- applemedia: H.265 / HEVC video encoding + decoding support
- androidmedia: H.265 / HEVC video encoding support
- New “force-live” property for audiomixer, compositor, glvideomixer,
d3d11compositor etc.
- Lots of new plugins, features, performance improvements and bug
fixes
Major new features and changes
AV1 video codec support improvements
AV1 is a royalty free next-generation video codec by AOMedia and a free
alternative to H.265/HEVC.
While supported in earlier versions of GStreamer already, this release
saw a lot of improvements across the board:
- Support for hardware encoding and decoding via VAAPI/VA, AMF, D3D11,
NVCODEC, QSV and Intel MediaSDK. Hardware codecs for AV1 are slowly
becoming available in embedded systems and desktop GPUs (AMD, Intel,
NVIDIA), and these can now be used via GStreamer.
- New AV1 RTP payloader and depayloader elements.
- New encoder settings in the AOM reference encoder-based av1enc
element.
- Various improvements in the AV1 parser and in the MP4/Matroska/WebM
muxers/demuxers.
- dav1d and rav1e based software decoder/encoder elements shipped as
part of the binaries.
- AV1 parser improvements and various bugfixes all over the place.
Touchscreen event support in Navigation API
The Navigation API supports the sending of key press events and mouse
events through a GStreamer pipeline. Typically these will be picked up
by a video sink on which these events happen and then the event is
transmitted into the pipeline so it can be handled by elements inside
the pipeline if it wasn’t handled by the application.
This has traditionally been used for DVD menu support, but can also be
used to forward such inputs to source elements that render a web page
using a browser engine such as WebKit or Chromium.
This API has now gained support for touchscreen events, and this has
been implemented in various plugins such as the GTK, Qt, XV, and x11
video sinks as well as the wpevideosrc element.
GStreamer CUDA integration
- New gst-cuda library
- integration with D3D11 and NVIDIA dGPU NVMM elements
- new cudaconvertscale element
GStreamer Direct3D11 integration
- New gst-d3d11 public library
- gst-d3d11 library is not integrated with GStreamer documentation
system yet. Please refer to the examples
- d3d11screencapture: Add Windows Graphics Capture API based capture
mode, including Win32 application window capturing
- d3d11videosink and d3d11convert can support flip/rotation and crop
meta
- d3d11videosink: New emit-present property and present signal so that
applications can overlay an image on Direct3D11 swapchain’s
backbuffer via Direct3D/Direct2D APIs. See also C++ and Rust
examples
- d3d11compositor supports YUV blending/composing without intermediate
RGB(A) conversion to improve performance
- Direct3D11 video decoders are promoted to GST_RANK_PRIMARY or
higher, except for the MPEG2 decoder
H.264/H.265 timestamp correction elements
- Muxers are often picky and need proper PTS/DTS timestamps set on the
input buffers, but that can be a problem if the encoded input media
stream comes from a source that doesn’t provide proper signalling of
DTS, such as is often the case for RTP, RTSP and WebRTC streams or
Matroska container files. Theoretically parsers should be able to
fix this up, but it would probably require fairly invasive changes
in the parsers, so two new elements h264timestamper and
h265timestamper bridge the gap in the meantime and can reconstruct
missing PTS/DTS.
Easy sender timestamp reconstruction for RTP and RTSP
- it was always possible to reconstruct and retrieve the original RTP
sender timestamps in GStreamer, but required a fair bit of
understanding of the internal mechanisms and the right property
configuration and clock setup.
- rtspsrc and rtpjitterbuffer gained a new
“add-reference-timestamp-meta” property that if set puts the
original absolute reconstructed sender timestamps on the output
buffers via a meta. This is particularly useful if the sender is
synced to an NTP clock or PTP clock. The original sender timestamps
are either based on the RTCP NTP times, NTP RTP header extensions
(RFC6051) or RFC7273-style clock signalling.
Qt6 support
- new qml6glsink element for Qt6 similar to the existing Qt5 element.
Matching source and overlay elements will hopefully follow in the
near future.
OpenGL + Video library enhancements
- Support for new video formats (NV12_4L4, NV12_16L32S, NV12_8L128,
NV12_10BE_8L128) and dmabuf import in more formats (Y410, Y212_LE,
Y212_BE, Y210, NV21, NV61)
- Improved support for tiled formats with arbitrary tile dimensions,
as needed by certain hardware decoders/encoders
- glvideomixer: New “crop-left,”crop-right, “crop-top” and
“crop-bottom” pad properties for cropping inputs
- OpenGL support for gst_video_sample_convert():
- Used for video snapshotting and thumbnailing, to convert buffers
retrieved from appsinks or sink “last-sample” properties in
JPG/PNG thumbnails.
- This function can now take samples and buffers backed by GL
textures as input and will automatically plug a gldownload
element in that case.
High bit-depth support (10, 12, 16 bits per component value) improvements
- compositor can now handle any supported input format and also mix
high-bitdepth (10-16 bit) formats (naively)
- videoflip has gained support for higher bit depth formats.
- vp9enc, vp9dec now support 12-bit formats and also 10-bit 4:4:4
WebRTC
- Allow insertion of bandwidth estimation elements e.g. for Google
Congestion Control (GCC) support
- Initial support for sending or receiving simulcast streams
- Support for asynchronous host resolution for STUN/TURN servers
- GstWebRTCICE was split into base classes and implementation to make
it possible to plug custom ICE implementations
- webrtcsink: batteries-included WebRTC sender (Rust)
- whipsink: WebRTC HTTP ingest (WHIP) to a MediaServer (Rust)
- whepsrc: WebRTC HTTP egress (WHEP) from a MediaServer (Rust)
- Many other improvements and bug fixes
New HLS, DASH and MSS adaptive streaming clients
A new set of “adaptive demuxers” to support HLS, DASH and MSS adaptive
streaming protocols has been added. They provide improved performance,
new features and better stream compatibility compared to the previous
elements. These new elements require a “streams-aware” pipeline such as
playbin3, uridecodebin3 or urisourcebin.
The previous elements’ design prevented implementing several use-cases
and fixing long-standing issues. The new elements were re-designed from
scratch to tackle those:
- Scheduling Only 3 threads are present, regardless of the number of
streams selected. One in charge of downloading fragments and
manifests, one in charge of outputting parsed data downstream, and
one in charge of scheduling. This improves performance, resource
usage and latency.
- Better download control The elements now directly control the
scheduling and download of manifests and fragments using libsoup
directly instead of depending on external elements for downloading.
- Stream selection, only the selected streams are downloaded. This
improves bandwith usage. Switching stream is done in such a way to
ensure there are no gaps, meaning the new stream will be switched to
only once enough data for it has been downloaded.
- Internal parsing, the downloaded streams are parsed internally. This
allows the element to fully respect the various specifications and
offer accurate buffering, seeking and playback. This is especially
important for HLS streams which require parsing for proper
positioning of streams.
- Buffering and adaptive rate switching, the new elements handle
buffering internally which allows them to have a more accurate
visibility of which bandwith variant to switch to.
Playbin3, Decodebin3, UriDecodebin3, Parsebin improvements
The “new” playback elements introduced in 1.18 (playbin3 and its various
components) have been refactored to allow more use-cases and improve
performance. They are no longer considered experimental, so applications
using the legacy playback elements (playbin and (uri)decodebin) can
migrate to the new components to benefit from these improvements.
- Gapless The “gapless” feature allows files and streams to be
fetched, buffered and decoded in order to provide a “gapless”
output. This feature has been refactored extensively in the new
components:
- A single (uri)decodebin3 (and therefore a single set of
decoders) is used. This improves memory and cpu usage, since on
identical codecs a single decoder will be used.
- The “next” stream to play will be pre-rolled “just-in-time”
thanks to the buffering improvements in urisourcebin (see below)
- This feature is now handled at the uridecodebin3 level.
Applications that wish to have a “gapless” stream and process it
(instead of just outputting it, for example for transcoding,
retransmission, …) can now use uridecodebin3 directly. Note that
a streamsynchronizer element is required in that case.
- Buffering improvements The urisourcebin element is in charge of
fetching and (optionally) buffering/downloading the stream. It has
been extended and improved:
- When the parse-streams property is used (by default in
uridecodebin3 and playbin3), compatible streams will be demuxed
and parsed (via parsebin) and buffering will be done on the
elementary streams. This provides a more accurate handling of
buffering. Previously buffering was done on a best-effort basis
and was mostly wrong (i.e. downloading more than needed).
- Applications can use urisourcebin with this property as a
convenient way of getting elementary streams from a given URI.
- Elements can handle buffering themselves (such as the new
adaptive demuxers) by answering the GST_QUERY_BUFFERING query.
In that case urisourcebin will not handle it.
- Stream Selection Efficient stream selection was previously only
possible within decodebin3. The downside is that this meant that
upstream elements had to provide all the streams from which to chose
from, which is inefficient. With the addition of the
GST_QUERY_SELECTABLE query, this can now be handled by elements
upstream (i.e. sources)
- Elements that can handle stream selection internally (such as
the new adaptive demuxer elements) answer that query, and handle
the stream selection events themselves.
- In this case, decodebin3 will always process all streams that
are provided to it.
- Instant URI switching This new feature allows switching URIs
“instantly” in playbin3 (and uridecodebin3) without having to change
states. This mimics switching channels on a television.
- If compatible, decoders will be re-used, providing lower
latency/cpu/memory than by switching states.
- This is enabled by setting the instant-uri property to true,
setting the URI to switch to immediately, and then disabling the
instant-uri property again afterwards.
- playbin3, decodebin3, uridecodebin3, parsebin, and urisrc are no
longer experimental
- They were originally marked as ‘technology preview’ but have
since seen extensive usage in production settings, so are
considered ready for general use now.
Fraunhofer AAC audio encoder HE-AAC and AAC-LD profile support
- fdkaacenc:
- Support for encoding to HE-AACv1 and HE-AACv2 profile
- Support for encoding to AAC Low Delay (LD) profile
- Advanced bitrate control options via new “rate-control”,
“vbr-preset”, “peak-bitrate”, and “afterburner” properties
RTP rapid synchronization support in the RTP stack (RFC6051)
RTP provides several mechanisms how streams can be synchronized relative
to each other, and how absolute sender times for RTP packets can be
obtained. One of these mechanisms is via RTCP, which has the
disadvantage that the synchronization information is only distributed
out-of-band and usually some time after the start.
GStreamer’s RTP stack, specifically the rtpbin, rtpsession and
rtpjitterbuffer elements, now also have support for retrieving and
sending the same synchronization information in-band via RTP header
extensions according to RFC6051 (Rapid Synchronisation of RTP Flows).
Only 64-bit timestamps are supported currently.
This provides per packet synchronization information from the very
beginning of a stream and allows accurate inter-stream, and (depending
on setup) inter-device, synchronization at the receiver side.
ONVIF XML Timed Metadata support
The ONVIF standard implemented by various security cameras also
specifies a format for timed metadata that is transmitted together with
the audio/video streams, usually over RTSP.
Support for this timed metadata is implemented in the MP4 demuxer now as
well as the new fragmented MP4 muxer and the new non-fragmented MP4
muxer from the GStreamer Rust plugins. Additionally, the new onvif
plugin ‒ which is part of the GStreamer Rust plugins ‒ provides general
elements for handling the metadata and e.g. overlaying certain parts of
it over a video stream.
As part of this support for absolute UTC times was also implemented
according to the requirements of the ONVIF standards in the
corresponding elements.
MP3 gapless playback support
While MP3 can probably considered a legacy format at this point, a new
feature was added with this release.
When playing back plain MP3 files, i.e. outside a container format,
switches between files can now be completely gapless if the required
metadata is provided inside the file. There is no standardized metadata
for this, but the LAME MP3 encoder writes metadata that can be parsed by
the mpegaudioparse element now and forwarded to decoders for ensuring
removal of padding samples at the front and end of MP3 files.
“force-live” property for audio + video aggregators
This is a quality of life fix for playout and streaming applications
where it is common to have audio and video mixer elements that should
operate in live mode from the start and produce output continuously.
Often one would start a pipeline without any inputs hooked up to these
mixers in the beginning, and up until now there was no way to easily
force these elements into live mode from the start. One would have to
add an initial live video or audio test source as dummy input to achieve
this.
The new “force-live” property makes these audio and video aggregators
start in live mode without the need for any dummy inputs, which is
useful for scenarios where inputs are only added after starting the
pipeline.
This new property should usually be used in connection with the
“min-upstream-latency” property, i.e. you should always set a non-0
minimum upstream latency then.
This is now supported in all GstAudioAggregator and GstVideoAggregator
subclasses such as audiomixer, audiointerleave, compositor,
glvideomixer, d3d11compositor, etc.
New elements and plugins
- new cudaconvertscale element that can convert and scale in one pass
- new gtkwaylandsink element based on gtksink, but similar to
waylandsink and uses Wayland APIs directly instead of rendering with
Gtk/Cairo primitives. This approach is only compatible with Gtk3,
and like gtksink this element only supports Gtk3.
- new h264timestamper and h265timestamper elements to reconstruct
missing pts/dts from inputs that might not provide them such as
e.g. RTP/RTSP/WebRTC inputs (see above)
- mfaacdec, mfmp3dec: Windows MediaFoundation AAC and MP3 decoders
- new msdkav1enc AV1 video encoder element
- new nvcudah264enc, nvcudah265enc, nvd3d11h264enc, and nvd3d11h265enc
NVIDIA GPU encoder elements to support zero-copy encoding, via CUDA
and Direct3D11 APIs, respectively
- new nvautogpuh264enc and nvautogpuh265enc NVIDIA GPU encoder
elements: The auto GPU elements will automatically select a target
GPU instance in case multiple NVIDIA desktop GPUs are present, also
taking into account the input memory. On Windows CUDA or Direct3D11
mode will be determined by the elements automatically as well. Those
new elements are useful if target GPU and/or API mode (either CUDA
or Direct3D11 in case of Windows) is undeterminable from the encoder
point of view at the time when pipeline is configured, and therefore
lazy target GPU and/or API selection are required in order to avoid
unnecessary memory copy operations.
- new nvav1dec AV1 NVIDIA desktop GPU decoder element
- new qml6glsink element to render video with Qt6
- qsv: New Intel OneVPL/MediaSDK (a.k.a Intel Quick Sync) based
decoder and encoder elements, with gst-d3d11 (on Windows) and gst-va
(on Linux) integration
- Support multi-GPU environment, for example, concurrent video
encoding using Intel iGPU and dGPU in a single pipeline
- H.264 / H.265 / VP9 and JPEG decoders
- H.264 / H.265 / VP9 / AV1 / JPEG encoders with dynamic encoding
bitrate update
- New plugin does not require external SDK for building on Windows
- vulkanoverlaycompositor: new vulkan overlay compositor element to
overlay upstream GstVideoOverlayCompositonMeta onto the video
stream.
- vulkanshaderspv: performs operations with SPIRV shaders in Vulkan
- win32ipcvideosink, win32ipcvideosrc: new shared memory videosrc/sink
elements for Windows
- wicjpegdec, wicpngdec: Windows Imaging Component (WIC) based JPEG
and PNG decoder elements.
- Many exciting new Rust elements, see Rust section below
New element features and additions
- audioconvert: Dithering now uses a slightly slower, less biased PRNG
which results in better quality output. Also dithering can now be
enabled via the new “dithering-threshold” property for target bit
depths of more than 20 bits.
- av1enc: Add “keyframe-max-dist” property for controlling max
distance between keyframes, as well as “enc-pass”, “keyframe-mode”,
“lag-in-frames” and “usage-profile” properties.
- cccombiner: new “output-padding” property
- decklink: Add support for 4k DCI, 8k/UHD2 and 8k DCI modes
- dvbsubenc: Support for >SD resolutions is working correctly now.
- fdkaacenc: Add HE-AAC / HE-AACv2 profile support
- glvideomixer: New “crop-left,”crop-right, “crop-top” and
“crop-bottom” pad properties for cropping inputs
- gssink: new ‘content-type’ property. Useful when one wants to upload
a video as video/mp4 instead of ’video/quicktime` for example.
- jpegparse: Rewritten using the common parser library
- msdk:
- new msdkav1enc AV1 video encoder element
- msdk decoders: Add support for Scaler Format Converter (SFC) on
supported Intel platforms for hardware accelerated conversion
and scaling
- msdk encoders: support import of dmabuf, va memory and D3D11
memory
- msdk encoders: add properties for low delay bitrate control and
max frame sizes for I/P frames
- msdkh264enc, msdkh265enc: more properties to control intra
refresh
- note that on systems with multi GPUs the Windows D3D11
integration might only work reliably if the Intel GPU is the
primary GPU
- mxfdemux: Add support for Canon XF-HEVC
- openaptx: Support the freeaptx library
- qroverlay:
- new “qrcode-case-sensitive” property allows encoding case
sensitive strings like wifi SSIDs or passwords.
- added the ability to pick up data to render from an
upstream-provided custom GstQROverlay meta
- qtdemux: Add support for ONVIF XML Timed MetaData and AVC-Intra
video
- rfbsrc now supports the uri handler interface, so applications can
use RFB/VNC sources in uridecodebin(3) and playbin, with
e.g. rfb://:password@10.1.2.3:5903?shared=1
- rtponviftimestamp: Add support for using reference timestamps
- rtpvp9depay now has the same keyframe-related properties as
rtpvp8depay and rtph264depay: “request-keyframe” and
“wait-for-keyframe”
- rtspsrc: Various RTSP servers are using invalid URL operations for
constructing the control URL. Until GStreamer 1.16 these worked
correctly because GStreamer was just appending strings itself to
construct the control URL, but starting version 1.18 the correct URL
operations were used. With GStreamer 1.22, rtspsrc now first tries
with the correct control URL and if that fails it will retry with
the wrongly constructed control URL to restore support for such
servers.
- rtspsrc and rtpjitterbuffer gained a new
“add-reference-timestamp-meta” property that makes them put the
unmodified original sender timestamp on output buffers for NTP or
PTP clock synced senders
- srtsrc, srtsink: new “auto-reconnect” property to make it possible
to disable automatic reconnects (in caller mode) and make the
elements post an error immediately instead; also stats improvements
- srtsrc: new “keep-listening” property to avoid EOS on disconnect and
keep the source running while it waits for a new connection.
- videocodectestsink: added YUV 4:2:2 support
- wasapi2src: Add support for process loopback capture
- wpesrc: Add support for modifiers in key/touch/pointer events
Plugin and library moves
- The xingmux plugin has been moved from gst-plugins-ugly into
gst-plugins-good.
- The various Windows directshow plugins in gst-plugins-bad have been
unified into a single directshow plugin.
Plugin removals
- The dxgiscreencapsrc element has been removed, use
d3d11screencapturesrc instead
Miscellaneous API additions
- GST_AUDIO_FORMAT_INFO_IS_VALID_RAW() and
GST_VIDEO_FORMAT_INFO_IS_VALID_RAW() can be used to check if a
GstAudioFormatInfo or GstVideoFormatInfo has been initialised to a
valid raw format.
- Video SEI meta: new GstVideoSEIUserDataUnregisteredMeta to carry
H.264 and H.265 metadata from SEI User Data Unregistered messages.
- vulkan: Expose gst_vulkan_result_to_string()
Miscellaneous performance, latency and memory optimisations
- liborc 0.4.33 adds support for aarch64 (64-bit ARM) architecture
(not enabled by default on Windows yet though) and improvements for
32-bit ARM and should greatly enhance performance for certain
operations that use ORC.
- as always there have been plenty of performance, latency and memory
optimisations all over the place.
Miscellaneous other changes and enhancements
- the audio/video decoder base classes will not consider decoding
errors a hard error by default anymore but will continue trying to
decode. Previously more than 10 consecutive errors were considered a
hard error but this caused various partially broken streams to fail.
The threshold is configurable via the “max-errors” property.
- compatibility of the GStreamer PTP clock implementation with
different PTP server implementations was improved, and
synchronization is achieved successfully in various scenarios that
failed before.
Tracing framework and debugging improvements
New tracers
- buffer-lateness: Records lateness of buffers and the reported
latency for each pad in a CSV file. Comes with a script for
visualisation.
- pipeline-snapshot: Creates a .dot file of all pipelines in the
application whenever requested via SIGUSR1 (on UNIX systems)
- queue-levels: Records queue levels for each queue in a CSV file.
Comes with a script for visualisation.
Debug logging system improvements
- new log macros GST_LOG_ID, GST_DEBUG_ID, GST_INFO_ID,
GST_WARNING_ID, GST_ERROR_ID, and GST_TRACE_ID allow passing a
string identifier instead of a GObject. This makes it easier to log
non-gobject-based items and also has performance benefits.
Tools
- gst-play-1.0 gained a --no-position command line option to suppress
position/duration queries, which can be useful to reduce debug log
noise.
GStreamer FFMPEG wrapper
- Fixed bitrate management and timestamp inaccuracies for video
encoders
- Fix synchronization issues and errors created by the (wrong)
forwarding of upstream segment events by ffmpeg demuxers.
- Clipping meta support for gapless mp3 playback
GStreamer RTSP server
- Add RFC5576 Source-specific media attribute to the SDP media for
signalling the CNAME
- Add support for adjusting request response on pipeline errors
- Give the application the possibility to adjust the error code
when responding to a request. For that purpose the pipeline’s
bus messages are emitted to subscribers through a
“handle-message” signal. The subscribers can then check those
messages for errors and adjust the response error code by
overriding the virtual method
GstRTSPClientClass::adjust_error_code().
- Add gst_rtsp_context_set_token() method to make it possible to set
the RTSPToken on some RTSPContext from bindings such as the Python
bindings.
- rtspclientsink gained a “publish-clock-mode” property to configure
whether the pipeline clock should be published according to RFC7273
(RTP Clock Source Signalling), similar to the same API on
GstRTSPMedia.
GStreamer VA-API support
- Development activity has shifted towards the new va plugin, with
gstreamer-vaapi now basically in maintenance-only mode. Most of the
below refers to the va plugin (not gstreamer-vaapi).
- new gst-va library for GStreamer VA-API integration
- vajpegdec: new JPEG decoder
- vah264enc, vah265enc: new H.264/H.265 encoders
- vah264lpenc, vah265lpenc: new low power mode encoders
- vah265enc: Add extended formats support such as 10/12 bits, 4:2:2
and 4:4:4
- Support encoder reconfiguration
- vacompositor: Add new compositor element using the VA-API VPP
interface
- vapostproc:
- new “scale-method” property
- Process HDR caps if supported
- parse video orientation from tags
- vaapipostproc: Enable the use of DMA-Buf import and export
(gstreamer-vaapi)
GStreamer Video4Linux2 support
- Added support for Mediatek Stateless CODEC (VP8, H.264, VP9)
- Stateless H.264 interlaced decoder support
- Stateless H.265 decoder support
- Stateful decoder support for driver resolution change events
- Stateful decoding support fixes for NXP/Amphion driver
- Support for hardware crop in v4l2src
- Conformance test improvement for stateful decoders
- Fixes for Raspberry Pi CODEC
GStreamer OMX
- There were no changes in this module
GStreamer Editing Services and NLE
- Handle compositors that are bins around the actual compositor
implementation (like glvideomixers which wraps several elements)
- Add a mode to disable timeline editing API so the user can be in
full control of its layout (meaning that the user is responsible for
ensuring its validity/coherency)
- Add a new fade-in transition type
- Add support for non-1/1 PAR source videos
- Fix frame accuracy when working with very low framerate streams
GStreamer validate
- Clean up and stabilize API so we can now generate rust bindings
- Enhance the appsrc-push action type allowing to find tune the
buffers more in details
- Add an action type to verify currently configured pad caps
- Add a way to run checks from any thread after executing a ‘wait’
action. This is useful when waiting on a signal and want to check
the value of a property right when it is emited for example.
GStreamer Python Bindings
- Add a Gst.init_python() function to be called from plugins which
will initialise everything needed for the GStreamer Python bindings
but not call Gst.init() again since this will have been called
already.
- Add support for the GstURIHandlerInterface that allows elements to
advertise what URI protocols they support.
GStreamer C# Bindings
- Fix AppSrc and AppSink constructors
- The C# bindings have yet to be updated to include new 1.22 API,
which requires improvements in various places in the bindings /
binding generator stack. See issue #1718 in GitLab for more
information and to track progress.
GStreamer Rust Bindings and Rust Plugins
The GStreamer Rust bindings are released separately with a different
release cadence that’s tied to gtk-rs, but the latest release has
already been updated for the new GStreamer 1.22 API. Check the bindings
release notes for details of the changes since 0.18, which was released
around GStreamer 1.20.
gst-plugins-rs, the module containing GStreamer plugins written in Rust,
has also seen lots of activity with many new elements and plugins. A
list of all Rust plugins and elements provided with the 0.9 release can
be found in the repository.
- 33% of GStreamer commits are now in Rust (bindings + plugins), and
the Rust plugins module is also where most of the new plugins are
added these days.
- The Rust plugins are now shipped as part of the Windows MSVC + macOS
binary packages. See below for the list of shipped plugins and the
status of Rust support in cerbero.
- The Rust plugins are also part of the documentation on the GStreamer
website now.
- Rust plugins can be used from any programming language. To the
outside they look just like a plugin written in C or C++.
New Rust plugins and elements
- rtpav1pay / rtpav1depay: RTP (de)payloader for the AV1 video codec
- gtk4paintablesink: a GTK4 video sink that provides a GdkPaintable
for rendering a video in any place inside a GTK UI. Supports
zero-copy rendering via OpenGL on Linux and macOS.
- ndi: source, sink and device provider for NewTek NDI protocol
- onvif: Various elements for parsing, RTP (de)payloading, overlaying
of ONVIF timed metadata.
- livesync: Element for converting a live stream into a continuous
stream without gaps and timestamp jumps while preserving live
latency requirements.
- raptorq: Encoder/decoder elements for the RaptorQ FEC mechanism that
can be used for RTP streams (RFC6330).
WebRTC elements
- webrtcsink: a WebRTC sink (batteries included WebRTC sender with
specific signalling)
- whipsink: WebRTC HTTP ingest (WHIP) to MediaServer
- whepsrc: WebRTC HTTP egress (WHEP) from MediaServer
- rtpgccbwe: RTP bandwidth estimator based on the Google Congestion
Control algorithm (GCC), used by webrtcsink
Amazon AWS services
- awss3src / awss3sink: A source and sink element to talk to the
Amazon S3 object storage system.
- awss3hlssink: A sink element to store HLS streams on Amazon S3.
- awstranscriber: an element wrapping the AWS Transcriber service.
- awstranscribeparse: an element parsing the packets of the AWS
Transcriber service.
Video Effects (videofx)
- roundedcorners: Element to make the corners of a video rounded via
the alpha channel.
- colordetect: A pass-through filter able to detect the dominant
color(s) on incoming frames, using color-thief.
- videocompare: Compare similarity of video frames. The element can
use different hashing algorithms like Blockhash, DSSIM, and others.
New MP4 muxer + Fragmented MP4 muxer
- fmp4mux: New fragmented MP4/ISOBMFF/CMAF muxer for generating
e.g. DASH/HLS media fragments.
- isomp4mux: New non-fragmented, normal MP4 muxer.
Both plugins provides elements that replace the existing qtmux/mp4mux
element from gst-plugins-good. While not feature-equivalent yet, the new
codebase and using separate elements for the fragment and non-fragmented
case allows for easier extensability in the future.
Cerbero Rust support
- Starting this release, cerbero has support for building and shipping
Rust code on Linux, Windows (MSVC) and macOS. The Windows (MSVC) and
macOS binaries also ship the GStreamer Rust plugins in this release.
Only dynamic plugins are built and shipped currently.
- Preliminary support for Android, iOS and Windows (MinGW) exists but
more work is needed. Check the tracker issue for more details about
future work.
- The following plugins are included currently: audiofx, aws, cdg,
claxon, closedcaption, dav1d, fallbackswitch, ffv1, fmp4, gif,
hlssink3, hsv, json, livesync, lewton, mp4, ndi, onvif, rav1e,
regex, reqwest, raptorq, png, rtp, textahead, textwrap, threadshare,
togglerecord, tracers, uriplaylistbin, videofx, webrtc, webrtchttp.
Build and Dependencies
- meson 0.62 or newer is required
- GLib >= 2.62 is now required (but GLib >= 2.64 is strongly
recommended)
- libnice >= 0.1.21 is now required and contains important fixes for
GStreamer’s WebRTC stack.
- liborc >= 0.4.33 is recommended for 64-bit ARM support and 32-bit
ARM improvements
- onnx: OnnxRT >= 1.13.1 is now required
- openaptx: can now be built against libfreeaptx
- opencv: allow building against any 4.x version
- shout: libshout >= 2.4.3 is now required
- gstreamer-vaapi’s Meson build options have been switched from a
custom combo type (yes/no/auto) to the built-in Meson feature type
(enabled/disabled/auto)
- The GStreamer Rust plugins module gst-plugins-rs is now considered
an essential part of the GStreamer plugin offering and packagers and
distributors are strongly encouraged to package and ship those
plugins alongside the existing plugin modules.
- we now make use of Meson’s install tags feature which allows
selective installation of installl components and might be useful
for packagers.
Monorepo build (gst-build)
- new “orc-source” build option to allow build against a
system-installed liborc instead of forcing the use of orc as a
subproject.
- GStreamer command line tools can now be linked to the gstreamer-full
library if it’s built
Cerbero
Cerbero is a meta build system used to build GStreamer plus dependencies
on platforms where dependencies are not readily available, such as
Windows, Android, iOS, and macOS.
General improvements
- Rust support was added for all support configurations, controlled by
the rust variant; see above for more details
- All pkgconfig files are now reliably relocatable without requiring
pkg-config --define-prefix. This also fixes statically linking with
GStreamer plugins using the corresponding pkgconfig files.
- New documentation on how to build a custom GStreamer repository
using Cerbero, please see the README
- HTTPS certificate checking is enabled for downloads on all platforms
now
- Fetching now automatically retries on error for robustness against
transient errors
- Support for building the new Qt6 plugin was added
- pkgconfig files for various recipes were fixed
- Several recipes were updated to newer versions
- New plugins: adaptivedemux2 aes codectimestamper dav1d
- New libraries: cuda webrtcnice
macOS / iOS
- Added support for running Cerbero on ARM64 macOS
- GStreamer.framework and all libraries in it are now relocatable,
which means they use LC_RPATH entries to find dependencies instead
of using an absolute path. If you link to GStreamer using the
pkgconfig files, no action is necessary. However, if you use the
framework directly or link to the libraries inside the framework by
hand, then you need to pass -Wl,-rpath,<path_to_libdir> to the
linker.
- Apple bitcode support was dropped, since Apple has deprecated it
- macOS installer now correctly advertises support for both x86_64 and
arm64
- macOS framework now ships the gst-rtsp-server-1.0 library
- Various fixes were made to make static linking to gstreamer
libraries and plugins work correctly on macOS
- When statically linking to the applemedia plugin using Xcode 13, you
will need to pass -fno-objc-msgsend-selector-stubs which works
around a backwards-incompatible change in Xcode 14. This is not
required for the rest of GStreamer at present, but will be in the
future.
- macOS installer now shows the GStreamer logo correctly
Windows
- MSVC is now required by default on Windows, and the Visual Studio
variant is enabled by default
- To build with MinGW, use the mingw variant
- Visual Studio props files were updated for newer Visual Studio
versions
- Visual Studio 2015 support was dropped
- MSYS2 is now supported as the base instead of MSYS. Please see the
README for more details. Some advantages include:
- Faster build times, since parallel make works
- Faster bootstrap, since some tools are provided by MSYS2
- Other speed-ups due to using MSYS2 tools instead of MSYS
- Faster download by using powershell instead of hand-rolled Python
code
- Many recipes were ported from Autotools to Meson, speeding up the
build
- Universal Windows Platform is no longer supported, and binaries are
no longer shipped for it
- New documentation on how to force a specific Visual Studio
installation in Cerbero, please see the README
- New plugins: qsv wavpack directshow amfcodec wic win32ipc
- New libraries: d3d11
Windows MSI installer
- Universal Windows Platform prebuilt binaries are no longer available
Linux
- Various fixes for RHEL/CentOS 7 support
- Added support for running on Linux ARM64
Android
- Android support now requires Android API version 21 (Lollipop)
- Support for Android Gradle plugin 7.2
Platform-specific changes and improvements
Android
- Android SDK 21 is required now as minimum SDK version
- androidmedia: Add H.265 / HEVC video encoder mapping
- Implement JNI_OnLoad() to register static plugins etc. automatically
in case GStreamer is loaded from Java using System.loadLibrary(),
which is also useful for the gst-full deployment scenario.
Apple macOS and iOS
- The GLib version shipped with the GStreamer binaries does not
initialize an NSApp and does not run a NSRunLoop on the main thread
anymore. This was a custom GLib patch and caused it to behave
different from the GLib shipped by Homebrew or anybody else.
The change was originally introduced because various macOS APIs
require a NSRunLoop to run on the main thread to function correctly
but as this change will never get merged into GLib and it was
reverted for 1.22. Applications that relied on this behaviour should
move to the new gst_macos_main() function, which also does not
require the usage of a GMainLoop.
See e.g. gst-play.c for an example for the usage of
gst_macos_main().
- GStreamer.framework and all libraries in it are now relocatable,
which means they use LC_RPATH entries to find dependencies instead
of using an absolute path. If you link to GStreamer using the
pkgconfig files, no action is necessary. However, if you use the
framework directly or link to the libraries inside the framework by
hand, then you need to pass -Wl,-rpath,<path_to_libdir> to the
linker.
- avfvideosrc: Allow specifying crop coordinates during screen capture
- vtenc, vtdec: H.265 / HEVC video encoding + decoding support
- osxaudiosrc: Support a device as both input and output
- osxaudiodeviceprovider now probes devices more than once to
determine if the device can function as both an input AND and
output device. Previously, if the device provider detected that
a device had any output capabilities, it was treated solely as
an Audio/Sink. This caused issues for devices that have both
input and output capabilities (for example, USB interfaces for
professional audio have both input and output channels). Such
devicesare now listed as both an Audio/Sink as well as an
Audio/Source.
- osxaudio: support hidden devices on macOS
- These are devices that will not be shown in the macOS UIs and
that cannot be retrieved without having the specific UID of the
hidden device. There are cases when you might want to have a
hidden device, for example when having a virtual speaker that
forwards the data to a virtual hidden input device from which
you can then grab the audio. The blackhole project supports
these hidden devices and this change provides a way that if the
device id is a hidden device it will use it instead of checkinf
the hardware list of devices to understand if the device is
valid.
Windows
- win32ipcvideosink, win32ipcvideosrc: new shared memory videosrc/sink
elements
- wasapi2: Add support for process loopback capture for a specified
PID (requires Windows 11/Windows Server 2022)
- The Windows universal UWP build is currently non-functional and will
need updating after the recent GLib upgrade. It is unclear if anyone
is using these binaries, so if you are please make yourself known.
- wicjpegdec, wicpngdec: Windows Imaging Component (WIC) based JPEG
and PNG decoder elements.
- mfaacdec, mfmp3dec: Windows MediaFoundation AAC and MP3 decoders
- The uninstalled development environment supports PowerShell 7 now
Linux
- Improved design for DMA buffer sharing and modifier handling for
hardware-accelerated video decoders/encoders/filters and
capture/rendering on Linux and Linux-like system.
- kmssink
- new “fd” property which allows an application to provide their
own opened DRM device fd handle to kmssink. That way an
application can lease multiple fd’s from a DRM master to display
on different CRTC outputs at the same time with multiple kmssink
instances, for example.
- new “skip-vsync” property to achieve full framerate with legacy
emulation in drivers.
- HDR10 infoframe support
- va plugin and gstreamer-vaapi improvements (see above)
- waylandsink: Add “rotate-method” property and “render-rectangle”
property
- new gtkwaylandsink element based on gtksink, but similar to
waylandsink and uses Wayland APIs directly instead of rendering with
Gtk/Cairo primitives. This approach is only compatible with Gtk3,
and like gtksink this element only supports Gtk3.
Documentation improvements
- The GStreamer Rust plugins are now included and documented in the
plugin documentation.
Possibly Breaking Changes
- the Opus audio RTP payloader and depayloader no longer accept the
lower case encoding-format=multiopus but instead produce and accept
only the upper case variant encoding-format=MULTIOPUS, since those
should always be upper case in GStreamer (caps fields are always
case sensitive). This should hopefully only affect applications
where RTP caps are set manually and multi-channel audio (>= 3
channels) is used.
- wpesrc: the URI handler protocols changed from wpe:// and web:// to
web+http://, web+https://, and web+file:// which means URIs are RFC
3986 compliant and the source can simply strip the prefix from the
protocol.
- The Windows screen capture element dxgiscreencapsrc has been
removed, please use d3d11screencapturesrc instead.
- On Android the minimum supported Android API version is now version
21 and has been increased from 16.
- On macOS, the GLib version shipped with the GStreamer binaries will
no longer initialize an NSApp or run an NSRunLoop on the main
thread. See macOS/iOS section above for details.
- decklink: The decklink plugin is now using the 12.2.2 version of the
SDK and will not work with drivers older than version 12.
- On iOS Apple Bitcode support was removed from the binaries. This
feature is deprecated since XCode 14 and not used on the App Store
anymore.
- The MP4/Matroska/WebM muxers now require the “stream-format” to be
provided as part of the AV1 caps as only the original “obu-stream”
format is supported in these containers and not the “annexb” format.
Known Issues
- The Windows UWP build in Cerbero needs fixing after the recent GLib
upgrade (see above)
- The C# bindings have not been updated to include new 1.22 API yet
(see above)
Statistics
- 4072 commits
- 2224 Merge Requests
- 716 Issues
- 200+ Contributors
- ~33% of all commits and Merge Requests were in Rust modules
- 4747 files changed
- 469633 lines added
- 209842 lines deleted
- 259791 lines added (net)
Contributors
Ádám Balázs, Adam Doupe, Adrian Fiergolski, Adrian Perez de Castro, Alba
Mendez, Aleix Conchillo Flaqué, Aleksandr Slobodeniuk, Alicia Boya
García, Alireza Miryazdi, Andoni Morales Alastruey, Andrew Pritchard,
Arun Raghavan, A. Wilcox, Bastian Krause, Bastien Nocera, Benjamin
Gaignard, Bill Hofmann, Bo Elmgreen, Boyuan Zhang, Brad Hards, Branko
Subasic, Bruce Liang, Bunio FH, byran77, Camilo Celis Guzman, Carlos
Falgueras García, Carlos Rafael Giani, Célestin Marot, Christian Wick,
Christopher Obbard, Christoph Reiter, Chris Wiggins, Chun-wei Fan, Colin
Kinloch, Corentin Damman, Corentin Noël, Damian Hobson-Garcia, Daniel
Almeida, Daniel Morin, Daniel Stone, Daniels Umanovskis, Danny Smith,
David Svensson Fors, Devin Anderson, Diogo Goncalves, Dmitry Osipenko,
Dongil Park, Doug Nazar, Edward Hervey, ekwange, Eli Schwartz, Elliot
Chen, Enrique Ocaña González, Eric Knapp, Erwann Gouesbet, Evgeny
Pavlov, Fabian Orccon, Fabrice Fontaine, Fan F He, F. Duncanh, Filip
Hanes, Florian Zwoch, François Laignel, Fuga Kato, George Kiagiadakis,
Guillaume Desmottes, Gu Yanjie, Haihao Xiang, Haihua Hu, Havard Graff,
Heiko Becker, He Junyan, Henry Hoegelow, Hiero32, Hoonhee Lee, Hosang
Lee, Hou Qi, Hugo Svirak, Ignacio Casal Quinteiro, Ignazio Pillai, Igor
V. Kovalenko, Jacek Skiba, Jakub Adam, James Cowgill, James Hilliard,
Jan Alexander Steffens (heftig), Jan Lorenz, Jan Schmidt, Jianhui Dai,
jinsl00000, Johan Sternerup, Jonas Bonn, Jonas Danielsson, Jordan
Petridis, Joseph Donofry, Jose Quaresma, Julian Bouzas, Junsoo Park,
Justin Chadwell, Khem Raj, Krystian Wojtas, László Károlyi, Linus
Svensson, Loïc Le Page, Ludvig Rappe, Marc Leeman, Marek Olejnik, Marek
Vasut, Marijn Suijten, Mark Nauwelaerts, Martin Dørum, Martin Reboredo,
Mart Raudsepp, Mathieu Duponchelle, Matt Crane, Matthew Waters, Matthias
Clasen, Matthias Fuchs, Mengkejiergeli Ba, MGlolenstine, Michael Gruner,
Michiel Konstapel, Mikhail Fludkov, Ming Qian, Mingyang Ma, Myles
Inglis, Nicolas Dufresne, Nirbheek Chauhan, Olivier Crête, Pablo Marcos
Oltra, Patricia Muscalu, Patrick Griffis, Paweł Stawicki, Peter
Stensson, Philippe Normand, Philipp Zabel, Pierre Bourré, Piotr
Brzeziński, Rabindra Harlalka, Rafael Caricio, Rafael Sobral, Rafał
Dzięgiel, Raul Tambre, Robert Mader, Robert Rosengren, Rodrigo
Bernardes, Rouven Czerwinski, Ruben Gonzalez, Sam Van Den Berge,
Sanchayan Maity, Sangchul Lee, Sebastian Dröge, Sebastian Fricke,
Sebastian Groß, Sebastian Mueller, Sebastian Wick, Sergei Kovalev,
Seungha Yang, Seungmin Kim, sezanzeb, Sherrill Lin, Shingo Kitagawa,
Stéphane Cerveau, Talha Khan, Taruntej Kanakamalla, Thibault Saunier,
Tim Mooney, Tim-Philipp Müller, Tomasz Andrzejak, Tom Schuring, Tong Wu,
toor, Tristan Matthews, Tulio Beloqui, U. Artie Eoff, Víctor Manuel
Jáquez Leal, Vincent Cheah Beng Keat, Vivia Nikolaidou, Vivienne
Watermeier, WANG Xuerui, Wojciech Kapsa, Wonchul Lee, Wu Tong, Xabier
Rodriguez Calvar, Xavier Claessens, Yatin Mann, Yeongjin Jeong, Zebediah
Figura, Zhao Zhili, Zhiyuaniu, مهدي شينون (Mehdi Chinoune),
… and many others who have contributed bug reports, translations, sent
suggestions or helped testing.
Stable 1.22 branch
After the 1.22.0 release there will be several 1.22.x bug-fix releases
which will contain bug fixes which have been deemed suitable for a
stable branch, but no new features or intrusive changes will be added to
a bug-fix release usually. The 1.22.x bug-fix releases will be made from
the git 1.22 branch, which will be a stable branch.
1.22.0
1.22.0 was originally released on 23 January 2023.
Schedule for 1.24
Our next major feature release will be 1.24, and 1.23 will be the
unstable development version leading up to the stable 1.24 release. The
development of 1.23/1.24 will happen in the git main branch of the
GStreamer mono repository.
The plan for the 1.24 development cycle is yet to be confirmed.
1.24 will be backwards-compatible to the stable 1.22, 1.20, 1.18, 1.16,
1.14, 1.12, 1.10, 1.8, 1.6, 1.4, 1.2 and 1.0 release series.
------------------------------------------------------------------------
These release notes have been prepared by Tim-Philipp Müller with
contributions from Edward Hervey, Matthew Waters, Nicolas Dufresne,
Nirbheek Chauhan, Olivier Crête, Sebastian Dröge, Seungha Yang, and
Thibault Saunier.
License: CC BY-SA 4.0
Reference in a new issue View git blame Copy permalink