d3d11: Serialize ensure_element_data() call

It can be called any time from random thread. Thus, multiple
device objects can be created for the same physical device
if the call is not serialized, that's not a scenario we expect.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2727>

proxysink: Fix GstProxySrc leak

Clear weak pointer to peer src when disposing.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1552>

proxysink: Make sure stream-start and caps events are forwarded

There might be a sequence of event and buffer flow:
- Got stream-start/caps/segment events
- Got flush events
- And then buffers with a new segment event

In the above case, stream-start and caps event might not be reached to
peer proxysrc if peer proxysrc is not ready to receive them.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1552>

splitmuxsink: Don't crash on EOS without buffer

Fix a case where upstream pushed EOS without buffers.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2174>

webrtc: implement support for asynchronous host resolution

Doesn't block anymore if a mdns host resolution takes multiple seconds
to complete in e.g. stun/turn/ice candidate usage.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1961>

videoconvertscale: ensure writable caps when fixating format

gst_video_convert_scale_get_fixed_format() receives 'othercaps' from
basetransforms' fixate_caps() vmethod which explicitly mentions that
'`othercaps` may not be writable'.

The gst_caps_intersect() call just before may or may not produce new
caps. Particularly in cases like EMPTY or ANY caps on either of the
inputs, only a ref is taken and returned to the caller.

As a result, gst_video_convert_scale_fixate_format() may have attempted
to modify a non-writable caps structure.

Fix by adding a gst_caps_make_writable().

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2709>

rtprtx: Fix copying extension headers

There was a typo leading to reading memory from the buffer we were
writing to.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2696>

d3d11converter: Remove convert() method

Expose only single convert_buffer() method. It's safer approach than
passing arbitrary list of SRV/RTV pointers.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2713>

d3d11testsrc: Use convert_buffer() method

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2713>

d3d11: Remove video processor

Processing using ID3D11VideoProcessor was integrated into
GstD3D11Converter and GstD3D11VideoProcessor implementation is
not used anymore.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2713>

d3d11memory: Update alloc_wrapped() API to avoid staging texture alloc

Add size parameter and use it for CPU accessible memory size
instead of allocating staging texture per API call.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2713>

wasapi: Implement default audio channel mask

Some multichannel capture devices does not provide a channel mask value
which will result in a pipeline failure due to the empty channel mask.

Implemented the same fix used for wasapi2

Fixes https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1204

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2692>

examples/webrtc/signalling: Fix compatibility with Python 3.10

- ssl module requires an explicit TLS_SERVER role
- asyncio throws a deprecation warning when using
  asyncio.get_event_loop().  Remove custom event loop handling entirely
- No need to keep the websocket server in a member variable, can use
  a future to signal exit case along with the async with context manager
  of websockets.serve()

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2698>

v4l2videodec: replace multiple decoder bug warnings with single one

Achieve this by dropping frames after a drain if the driver failed to so.
This works around RaspberryPi driver issue [1].

[1] https://github.com/raspberrypi/linux/issues/5059#issuecomment-

Fixes #1103

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2518>

d3d11videosink: Add support for rotation

Adding "rotate-method" property

Fixes: https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/issues/1396
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2708>

d3d11videosink: Protect window with lock at every place

Access to the object should be thread safe to support runtime
property update

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2708>

d3d11convert: Add support for GstVideoOrientationMethod

Add flip/rotation support via GstVideoDirectionInterface

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2708>

d3d11converter: Add support for GstVideoOrientationMethod

Shader can handle rotation/flip with other conversions
at once in a single pipeline, and depending on device
capability, videoprocessor may support it too.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2708>

tests: udpsink: make test work in environments without IPv6

Part-fixes https://gitlab.freedesktop.org/gstreamer/gst-plugins-good/-/issues/939

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2659>

d3d11videosink: Use single GstD3D11Converter object

GstD3D11Converter supports videoprocessor and texture upload.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2697>

d3d11memory: Add private method for texture wrapped memory allocation

Unlike public method gst_d3d11_allocator_alloc_wrapped(), newly
added method by this commit will not calculate CPU accessible memory
size, since staging texture must be allocated to calculate the size.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2697>

d3d11: Use library private header in plugin

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2697>

d3d11compositor: Use single GstD3D11Converter object

... and remove unnessary videoprocessor/upload/copy code

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2697>

d3d11convert: Add border-color property

Use user specified border-color instead of black color

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2697>

d3d11convert: Use single GstD3D11Converter object

... and don't use GstD3D11VideoProcessor. Now GstD3D11Converter will
be able to convert using videoprocessor, and texture upload is also supported by
GstD3D11Converter. All the noisy code can be removed therefore.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2697>

d3d11converter: Add support conversion using videoprocessor

* Add videoprocessor feature to d3d11converter, in order to unifiy
  conversion flow.
* Add convert_buffer() method to support automatic shader/videoprocessor
  selection. The method also supports texture upload if input memory
  cannot be used for conversion (e.g., system memory or so)

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2697>

d3d11overlaycompositor: Objectify overlay render implementation

Be consistent with other implementations

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2697>

d3d11: Don't use const pointer to GstDxgiColorSpace

Instead, fill values of passed GstDxgiColorSpace struct

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2697>

samiparse: clean up some GString usage

There's no need to re-assign the return value of
g_string_append_*() functions and such to the variable
holding the GString. These return values are just for
convenience so function calls can be chained. The actual
GString pointer won't change, it's not a GList after all.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2685>

samiparse: micro-optimise entity handling

Avoid relocations and hard-code entity string length
in the struct, since we basically get it for free here.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2685>

samiparse: fix handling of self-closing tags

We would check the wrong string (rest of line rather than element)
for the / suffix of self-closing tags, which is not only wrong but
also has atrocious performance with certain strings like the garbled
nonsense clusterfuzz feeds us, which might cause discoverer to time
out when processing garbled SAMI files.

Fixes https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=47461

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2685>

Revert "glupload: Fix caps query with no filter"

This reverts commit 6f9ae5d7580763b5d18badb76f2166ff0012886a.

The _transform_caps() function can't tell the difference
between the caller wanting to know the output caps
for the current method, or all possible output caps. If
it includes caps for all possible methods, glupload can
end up negotiating and sending the wrong output caps
downstream.

Partially reverts !2687
Fixes #1310

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2699>

decklinkvideosink: Pass video frames directly to the Decklink SDK without copying

If the video frame is stored in PBO memory then we need to copy anyway as
it might be stored in CPU-accessible GPU memory that can't be accessed
from the Decklink driver.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2674>

msdkmjpegdec: Add support for error report

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2401>

glvideomixer: Only consider property changes a geometry change if there as an actual change

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2669>

glvideomixer: Add crop-{left,right,top,bottom} pad properties for cropping inputs

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2669>

nvencoder: Add support for dynamic GPU device selection

Adding nvautogpu{h264,h265}enc class which will accept upstream logical
GPU device object (GstCudaContext or GstD3D11Device) instead of
using pre-assigned GPU instance.

If upstream logical GPU device object is not NVENC compatible
(e.g., D3D11 device of non-NVIDIA GPU) or it's system memory,
then user specified "cuda-device-id" or "adapter-luid" property
will be used for GPU device selection.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2666>

cuda: Protect gst_cuda_ensure_element_context() with lock

Serialize the order of context setup since it can be called
from multiple threads

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2666>

cuda: Add YUY2 and UYVY formats for GL interop

Those formats are not supported conversion formats by CUDA convert/scale elements
but would be useful for GL <-> CUDA interop use case.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2691>

cudadownload: Passthrough in case of CUDA -> SYSTEM

GstCudaMemory supports CPU access via CUDA pinned host memory already
and it would show faster memory transfer performance between
GPU and CPU than copying from/to normal system memory.
If downstream supports video meta, we can passthrough CUDA memory.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2690>

webrtcbin: Reject caps that are not valid for creating an SDP media.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2689>

sdpmessage: Don't set SDP medias from caps without media/payload/clock-rate fields

Previously it would've silently failed reading the payload/clock-rate
and instead would've used some random value that happened to be on the
stack.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2689>

glupload: Fix caps query with no filter

If no filter caps are provided with a caps query, always
generate a full set of all caps from all upload methods,
not just the configured one. This is needed to handle
renegotiation when dealing with raw sysmem caps - as the upload
method might accept raw sysmem caps, but only the raw data
uploader adds those to the caps query.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2687>

Revert "glupload: allow system memory for dmabuf in transform_caps"

This reverts commit f3292dc1561a8d62812c3f1a2bb3de5c5bb6a807.

Only the raw data uploader should add sysmem caps to the
actual caps query, because we want them to be at the
lowest priority. If upstream does select to send raw
caps, then the correct upload method will still
be chosen because the accept_caps implementation
will accept them

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2687>

gl: Don't use the full transform_caps() method for reconfiguration check

When checking if we need to reconfigure when uploading, check
specifically the output caps of the current method will
result in compatible/incompatible caps, not the full set
of output caps from all upload methods.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2687>

cudaformat: add support for planar 4:2:2 YUV formats in CUDA D3D11 interop

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2269>

cudamemory: add support for planar 4:2:2 YUV formats

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2269>

video: Include new video-sei.h in video.h

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2688>

tracing: add hooks for gst_pad_chain() / gst_pad_chain_list()

This allows tracing buffers when they arrive in a pad instead of just
when they are pushed out of a pad.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2682>

v4l2videodec: replace multiple decoder bug warnings by a single warning
(warning due to incorrectly dropped frames at initial caps adjustment)

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2517>

v4l2codecs: enumerate all possible formats

Some decoder may also allow to (down) scale the decoded video.
Use VIDIOC_ENUM_FRAMESIZES iotcl to enumerate the possible output
formats resolutions.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2589>

nvcodec: Add cuda-device-id read-only property to stateless decoders

... and remove unnecessary intermediate subclass from class hierarchy

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2665>

nvcodec: Add cuda-device-id read-only property to CUVID decoders

Similar to the other hardware decoder plugins, provides assigned
device id via property

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2665>

adaptivedemux2: Ignore stopped stream flow state

When calculating the combined stream flow state
for a period, don't consider stopped streams.

Fixes switching to the next period in DASH streams
with multiple video/audio/subtitle streams.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2684>

adaptivedemux2: Clean up stream parsebins on finalize

Remove the parsebin for a stream from the overall
bin when cleaning up the stream, to avoid
keeping around old ones when moving between periods

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2676>

adaptivedemux2: Fix memory leaks and use-after-free

Fix various small memory leaks, and an invalid
access to GstEvent after giving away the ref
via gst_pad_push_event()

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2676>

hlsdemux2: Fix potential segfault

Fix a potential segfault if we receive a ISO-FF stream
with moof before moov.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2676>

hlsdemux2: Fix memory leaks

Clean up various memory leaks

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2676>

hlsdemux2: Free current_segment on finalize

Avoid a memory leak by making sure to release the
current segment on exit if non-NULL

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2676>

coding style: allow declarations after statement

See https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1243/
and https://gitlab.freedesktop.org/gstreamer/gstreamer-project/-/issues/78

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2683>

d3d11convert: Add support for GstVideoCropMeta

Performs crop, scale, and color space conversion all in
a single render pipeline. Note that cropping related property is not
added in this element (which will make negotiation very complicated),
but user can configure videocrop element for crop meta to be attached
on each buffer.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2678>

d3d11videosink: Fix for force-aspect-ratio setting when rendering on shared texture

Set specified force-aspect-ratio value on window object
in case of shared texture rendering as well

Fixes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1304
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2671>

mfvideoenc: Fix broken encoding when resolution is not an even number

Width and height values of 4:2:0 subsampled YUV format should be even number,
and if it's not the case, there should be padding which is not a contiguous memory layout.
Do copy input frames to MediaFoundation's memory in that case for now.

Fixes: https://gitlab.freedesktop.org/gstreamer/gstreamer/-/issues/1165
Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2661>

videorate: remove property-value quotes

Fixes warnings like:
Received a structure string that contains '="0.5"'. Reading as a gdouble value, rather than a string value. This is undesired behaviour, and with GStreamer 1.22  onward, this will be interpreted as a string value instead because it is wrapped in '"' quotes. If you want to guarantee this value is read as a string, before this change, use '=(string)"0.5"' instead. If you want to read in a gdouble value, leave its value unquoted.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2621>

adaptivedemux2: track: Fix buffering time calc before output

Use the lowest track input time as the output
time when calculating track buffering levels
before anything has been dequeued.

Fixes multi-period DASH not advancing to the
next period in some cases.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2650>

adaptivedemux2: track: Add period number to most debug statements

Store the period number the track belongs to, and
add it in various debug statements

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2650>

adaptivedemux2: stream: Don't take TRACKS_LOCK when sending EOS event

The stream tracks list can't change while we're
iterating it from the scheduling thread,
and the event handler immediately takes the
tracks lock, causing a deadlock.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2650>

hlsdemux2: Improve media playlist updates

* When dealing with rendition streams, we attempt to synchronize the media
  playlist against the variant stream. This helps with speeding up the correct
  initial fragment search and avoids issues when streams at activated at a much
  later time.
* Also add checks for variant stream existence before attempting to use them

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2679>

hlsdemux2: Only seek on selected streams

When handling seeks, there is no need to seek on unselected streams. If they
later get activated they will be properly seek onto

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2679>

tests: check: Update hlsdemux2 tests for playlist changes

We no longer do auto-magic fallbacks when synchronizing a disconnected
playlist. It is handled at a higher level.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2679>

hlsdemux2: Fix debug return statement

Due to latest commits res could have been NULL.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2679>

hlsdemux2: No longer re-add segments from before the playlist

When updating playlists, there is a possibility that the playlists don't
perfectly align, but the last entry of the previous playlist is *just* before
the first entry of the new playlist.

In those cases, we still can transfer the timing information from one playlist
to another, but we do not want to return that segment as being the matching one.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2679>

hlsdemux2: Use variant stream as support for synchronizing playlists

When matching playlists, there is a possibility that rendition streams will not
have been updated in time (for example because that stream started later, or
playback was paused). This would cause several playback failures and seeking
failures.

In order to still fall back on our feet, attempt to synchronize that rendition
playlist against the current variant playlist. This will attempt to match the
stream time using SN/DNS/PDT/...

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2679>

hlsdemux2: Detect synchronization loss

If we have been updating too slowly and have gone out of the current live
window, inform the baseclass accordingly.

This is different from the case where we have been updating quicker than what
the server provides.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2679>

hlsdemux2: More improvement/fixes to position tracking

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2679>

adaptivedemux2: Handle loss of synchronization

Add a new custom GstFlowReturn so that subclasses can notify that they have lost
live synchronization.

When that happens, do a flushing seek back to the live position

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2679>

adaptivedemux2: Fix SNAP seek handling

* Since only flushing seeks are allowed, the "current" position is always the
  global output position (and not "some" stream current position).

* In terms of figuring out to which stream to "snap" to, we can send it to any
  selected stream. Removes the requirement of this function to a specific output
  pad.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2679>

hlsdemux2: Fix position tracking

Remove the "pending advance" hack and instead rely on the base stream current
position to track our position (instead of a potentially NULL "current
segment").

Also ensure the media playlists are always refreshed with valid stream time,
even if there is no current segment.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2679>

hlsdemux2: Add a new method to synchronize two media playlist

This allows transfering the stream time of the playlist to an updated
variant/rendition.

This allows updating that information without having a "current segment"

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2679>

adaptivedemux2: Initialize and use stream start/current position

The stream start and current position would be properly set when seeking or
activating a stream after playback started. But it would never be properly
initialized.

Set it to NONE initially to indicate to subclasses that no position has been
tracked yet. This will allow them to detect initial stream usage.

Futhermore, once the initial streams setup is done, make sure that it is set to
a valid initial value:
* The minimum stream time in live
* Or else the period start

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2679>

v4l2videodec: Fix activation of internal pool

If the driver does not support VIDIOC_CREATE_BUFS ioctl, the pool
configuration may get changed, which requires a validation. This would
fail to activate a pool in a case it shouldn't normally fail unless we
are out of memory.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2456>

dv, opusparse: fix duplicate symbols in static build

Fixes https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/issues/1262

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2670>

d3d11: Enumerate devices via native DXGI API

D3D11CreateDevice() can fail for some reason (e.g., buggy GPU driver)
and then gst_d3d11_device_new() will return nullptr.
Use DXGI API directly instead of relying on the gst_d3d11_device_new()

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2663>

examples: Add an example for application texture sharing

This example shows GstD3D11BufferPool usage and a way of
D3D11 texture sharing between application and GStreamer via appsrc.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2646>

examples: Add d3d11decoder zero-copy rendering example

This example code demonstrates D3D11 device sharing between
application and GStreamer. Application can access texture
using appsink and it can be rendered on application's window without
any copy operation.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2646>

d3d11: Add a helper method for GstContext creation

Add gst_d3d11_context_new() method for application to create
GstContext using GstD3D11Device.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2646>

d3d11memory: Add support for wrapping application texture

Add gst_d3d11_allocator_alloc_wrapped() method to support wrapping
application texture using GstD3D11Memory without copy.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2646>

d3d11: Install library headers

Our Direct3D11 abstraction layer has been improved and
it gained good shape from API point of view.
Also, On Windows, GstD3D11 has various advantages over GstGL
in terms of compatibility/stability/feature/performance.
Note that WGL implementation is known to be buggy for some
drivers/vendors/scenario (that's a reason why Google implemented ANGLE).
Moreover, GstGL is not fully optimized for Windows unfortunately.

It's the time to open this interface to application developers
for various optimized processing using our Direct3D11
infrastructure.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2646>

d3d11: Update gst_d3d11_allocation_params_new() interface

Define GST_D3D11_ALLOCATION_FLAG_DEFAULT (0) value instead of
casting enum type everywhere. And pass D3D11_RESOURCE_MISC_FLAG value

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2646>

va: postproc: add scale-method property

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2654>

va: filter/postproc: move unconditional props to vavpp

Only conditional/dynamic properties should be installed/handled
by vafilter.  Thus, move and install the unconditional/static
properties in vavpp.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2654>

va: compositor/filter: use scale method setter

Use the filter's scale method member instead of the sample's member.

Removes the filter_flags member from the composite sample.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2654>

va: filter: add scale method field and setter

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2654>

va: Add vajpegdec element.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1575>

va: allocator: Add hack for no fourcc when surface creation.

This patch adds general mechanism for handling specific hacks. In this
case for jpeg decoder in i965 driver, which cannot create surfaces
with fourcc specified.

From jpeg decoder to the allocator, which creates the surfaces,
there's a non-simple path: basedec pseudo-class adds a hacks guint32
which will be set by actual elements (vajpegdec, in this case) and
basedec will always set the hack to the allocator when the allocator
is instantiated.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1575>

va: Add RGBP format definition.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1575>

va: caps: Add jpeg fields in coded caps.

Given the supported rt formats in a profile/entrypoint config it's
possible to know the supported JPEG colorspace and subsampling. This
patch adds this information in coded caps to a safer autoplugging
after jpegparser.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1575>

va: Add jpegdecoder base class.

This base class is intented for hardware accelerated decoders, but since
only VA uses it, it will be kept internally in va plugin.

It follows the same logic as the others video decoders in the library but.
as JPEG are independet images, there's no need to handle a DBP so no need
of a picture object. Instead a scan object is added with all the structures
required to decode the image (huffman and quant tables, mcus, etc.).

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1575>

va: build: Remove duplicated source.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/1575>

cudamemorycopy: Make sure writable caps before removing fields

The caps to be modified may not be writable when D3D11/GL/NVMM
are all disabled.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gstreamer/-/merge_requests/2664>