nvcodec: nvdecoder: Move to refcount based GstNvDecoderFrame

This refcount based way would be helpful for sharing nvdec frame among
multiple codec pictures and later zero-copy use case.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1738>

nvcodec: nvdecoder: Get rid of G_GNUC_INTERNAL

default is visibility=hidden. Don't need to use G_GNUC_INTERNAL
for new code therefore.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1738>

nvcodec: Add VP8 stateless decoder element

Like other nvcodec stateless decoders, the rank of this new nvvp8sldec
element will be secondary for now.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1738>

nvcodec: nvsldec: Fix typo in debug message

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1738>

mpegtspacketizer: Handle PCR issues with adaptive streams

A lot of content producers out there targetting "adaptive streaming" are riddled
with non-compliant PCR streams (essentially all the players out there just use
PTS/DTS and don't care about the PCR).

In order to gracefully cope with these, we detect them appropriately and any
small (< 15s) PCR resets get gracefully ignored.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1785>

nvcodec: leave g_once_init when all quarks are initialized

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1782>

codecs: h264decoder: Fix missing drain handling in bumping

Should've included in the commit 5527cc4a2e7ce8eeee1d8a717f99252477d6015f

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1783>

codecs: h264decoder: Try reference picture marking process in any case

... even if there is some invalid conditions
(because of broken stream, our implementation fault or so).
Otherwise baseclass will keep such reference pictures and
it would result to DPB full.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1783>

examples: Properly handle extended descriptors

By checking the extended tag. Provides a bit more information (if extended tag
is known)

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1781>

d3d11h264dec: Fix for MbaffFrameFlag and FrameNumList

As per spec 7.4.3 Slice header semantics, the flag value is derived as
MbaffFrameFlag = (mb_adaptive_frame_field_flag && !field_pic_flag)
and DXVA uses the value.
Regarding FrameNumList, in case of long-term ref, FrameNumList[i]
value should be long_term_frame_idx not long_term_pic_num.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1780>

codecs: h264decoder: Reset frame number per MMCO type 5

It should be cleared so that avoid wrong frame gap detection
for following pictures.

Passing 4 more conformance bitstream tests
* MR2_TANDBERG_E
* MR3_TANDBERG_B
* MR4_TANDBERG_C
* MR5_TANDBERG_C

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1768>

codecs: h264decoder: Fix for MMCO type 2

As per 8.2.5.4.2, we should mark a picture which has
LongTermPicNum == long_term_pic_num as "unused for reference",
not pic_num.

Passing conformance bitstream test with MR2_MW_A

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1768>

codecs: h264picture: Add more trace log

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1768>

decklink: correct framerate 2KDCI 23.98

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1771>

doc: player: mention that get_pipeline method needs unref

All other methods in docs clearly mention that an unref is needed, so should `get_pipeline()`. #1450

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1764>

player: call ref_sink on pipeline

Otherwise `gst_player_get_pipeline()` will return a floating reference which may confuse bindings and lead to crash.

Fixes #1450

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1763>

meson: Enable some MSVC warnings for parity with GCC/Clang

This makes it easier to do development with MSVC by making it warn
on common issues that GCC/Clang error out for in our CI configuration.

Continuation from https://gitlab.freedesktop.org/gstreamer/gst-build/-/merge_requests/223

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1760>

va: Add HEVC decoding support

This add HEVC decoding support into the new VA plugin. This implementation has
been tested using the ITU comformance test (through fluster). It fails all
MAIN10 tests, as this is not implemented yet along with the following:

CONFWIN_A_Sony_1 (looks fine, but md5sum is incorrect)
PICSIZE_A_Bossen_1 (height too high)
PICSIZE_B_Bossen_1 (same)
VPSSPSPPS_A_MainConcept_1 (parser issue)

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1714>

h265parser: Fix wrong warning message

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1714>

h265decoder: Remove unsued WpOffsetHalfRangeC

This is only needed for VA implementation of weight tables and isn't used
within the base class.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1714>

codecs: h264decoder: Rework for DPB management

Sync with recent h265decoder DPB implementation.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1761>

codecs: h264decoder: Remove unused pts variable

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1761>

aiffparse: Also set a channel mask for 2 channels

And only do add debug output at FIXME level when using the fallback
channel mask, not for those defined in the AIFF spec.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1756>

webrtc: Add properties to change the socket buffer sizes to ice object

libnice doesn't touch the kernel buffer sizes. When dealing with RTP data,
it's generally advisable to increase them to avoid dropping packets locally.
This is especially important when running multiple higher bitrate streams at
the same time.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1366>

vkdeviceprovider: Avoid deadlock on physical device

Don't hold the object lock on the vk physical device while
constructing a GstVulkanDevice around it, as
GstVulkanDevice can make calls on the physical device that
require the object lock.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1754>

codecs: h265picture: Minor update for coding style

It's GstH265Dpb, not GstH265Decoder

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1755>

codecs: h265decoder: Make GstVideoCodecFrame hold the last reference of the buffer

The functionality of passing the last reference of GstH265Picture
was silently dropped by the commit eeffd91109a409063e866337452eedd392649775
This commit will make it work again.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1755>

codecs: h265decoder: Clear GstVideoCodecFrame on DPB clear if needed

h265decoder might need to clear DPB depending on context even if
it's not flushing case. So associated GstVideoCodecFrame needs to be
released in case non-flushing case.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1755>

codecs: h265decoder: Don't drain DPB on EOB/EOS/IDR nalu

DPB bumping decision per end-of-bitstream, end-of-sequence or IDR nal
should done by spec. In short, draining on EOB/EOS/IDR is undefined
behavior as per spec.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1755>

h265decoder: Complete dependent slice header

This will save the last independent slice and fill in the missing
information for dependent slices. This was left over during the porting
from gstreamer-vaapi. The private variable prev_independent_slice was
already there.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1750>

h265decoder: Prevent possible infinite loop

Theoretically, one could produce a broken stream that would lead to
infinite in the specified algorithm to calculate l0/l1 reference lists.
This patch will pearly exit if this condition is met.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1750>

h265parse: Add missing const qualifier

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1750>

Revert "d3d11decoder: Use D3D11/DXGI standard colorimetry"

This reverts commit a52fc6deeda203add520cb59ae0026d109ecda95.

The change breaks H264/HEVC conformance bitstream tests

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1753>

wlvideoformat: fix DMA format convertor

In the most of case, this typo would work. But for
ARGB8888 and XRGB8888, which shm format is not based on fourcc,
which would never appear in format enumeration.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1751>

d3d11decoder: Use D3D11/DXGI standard colorimetry

D3D11/DXGI supports smaller set of colorimetry than all possible
combination. This restriction would make more streams convertible
by using ID3D11VideoProcessor

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1743>

d3d11window: Use ID3D11VideoProcessor only if device supports corresponding conversion

... and drop support for ID3D11VideoProcessor if device doesn't
support ID3D11VideoContext1 interface and therefore we cannot
query conversion supportability.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1743>

d3d11h{264,265}dec: Submit picture level parameters only once

Submit PICTURE_PARAMETERS and INVERSE_QUANTIZATION_MATRIX
buffers only once per picture. Multiple submission is redundant.
Also this modification would fix broken hevc decoding with
dependent slice.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1749>

codecs: h265decocer: Rework for DPB management

* Move all DPB bumping process into GstH265Dpb internal
* Handle DPB add process in GstH265Dpb struct
* Make implementation to be 1:1 mappable with hevc specification
* Fix wrong DPB bumping implementation especially when no_output_of_prior_pics_flag
  was specified.

With fixes from Nicolas Dufresne <nicolas.dufresne@collabora.com>

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1748>

d3d11decoder: Get rid of framerate field from pad template

Framerate is optional value and we don't have any framerate
related restriction for those elements. This commit is to fix
negotiation failure when upstream doesn't set framerate on caps.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1747>

tests: Don't set dtlsenc state before linking.

Link the dtlsenc in the testsuite before setting it to paused, as it
starts a pad task that can generate a not-linked error otherwise.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1744>

dtls: Catch bus errors and fail instead of hanging.

If the DTLS elements fail, they post a bus error and don't signal any
key negotiation. Catch the bus error and fail the test early instead
of letting it hang and time out.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1741>

sctp: Do downward state change logic after chaining up.

Call the parent state_change function first when changing state
downward, to make sure that the element has stopped before cleaning
it up.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1741>

dtls: Avoid bio_buffer assertion on shutdown.

On shutdown, a previous iteration of dtsl_connection_process()
might be incomplete and leave a partial bio_buffer behind.

If the DTLS connection is already marked closed, drop out
of dtls_connection_process early without asserting.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1741>

webrtc: Fix a race on shutdown.

The main context can disappear in gst_webrtc_bin_enqueue_task()
between checking the is_closed flag and enqueueing a source on the
main context. Protect the main context with the object lock instead
of the PC lock, and hold a ref briefly to make sure it stays alive.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1741>

webrtc: Set the DSCP markings based on the priority

This matches how the WebRTC javascript API works and the Chrome implementation.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1707>

rtpsender: Add API to set the priority

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1707>

rtptransceiver: Store the SSRC of the current stream

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1707>

webrtc: Save the media kind in the transceiver

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1707>

webrtcbin: Remove unused function

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1707>

webrtc: Document more objects

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1707>

d3d11decoder: Allow 10bits only profiles

HEVC_VLD_Main10 and VP9_VLD_10bit_Profile2 might not support
8bit format (i.e., NV12) depending on GPU vendor.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1742>

decklink: Remove \n from debug output

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1721>

decklink: Correctly indent everything

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1721>

decklink: Add a default profile id

This causes no changes to the profile but keeps the existing settings.
The profile can also be changed from e.g. the card's configuration
application and in that case probably should be left alone.

The default is the new value as it keeps the profile setting as it is,
which is consistent with the previous behaviour in 1.18.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1721>

decklink: Mark internal function as static

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1721>

decklink: Remove some dead code

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1721>

decklink: Update enum value bounds check in gst_decklink_get_mode()

The widescreen modes moved after GST_DECKLINK_MODE_2160p60 and using
them now would cause an assertion. This is a regression from
309f6187fef890c7ffa49305f38e89beac3b1423.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1737>

ttmlparse: Handle whitespace before XML declaration

When ttmlparse is in, e.g., an MPEG-DASH pipeline, there may be
whitespace between successive TTML documents in ttmlparse's accumulated
input. As libxml2 will fail to parse documents that have whitespace
before the opening XML declaration, ensure that any preceding whitespace
is not passed to libxml2.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1539>

ttmlparse: Ensure only single TTML doc parsed

The parser handles only one TTML file at a time, therefore if there are
multiple TTML documets in the input, parse only the first.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1539>

amc: Fix crash when encoding AVC

gstamcvideoenc.c calls gst_amc_avc_profile_to_string() with alternatives
set to NULL which causes a crash.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1732>

isac: add iSAC plugin

Wrapper on the iSAC reference encoder and decoder from webrtc,
see https://en.wikipedia.org/wiki/Internet_Speech_Audio_Codec

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1124>

transcodebin: Create the decodebin in _init

This way user can request pads right from the beginning

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1151>

transcodebin: Accept more than one stream

Look-up the stream matching the given ID also after building the stream list
from the received collection. Without this change the transcoder would discard
the second incoming stream.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1151>

transcodebin: Port to decodebin3

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1151>

uritranscodebin: Move to using a urisourcebin for our source.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1151>

transcoder: Base sync transcoding variant on a GMainLoop

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1151>

waylandsink: release frame callback when destroyed

We would use a frame callback from the surface to indicate
that last buffer is rendered, but when we destroy the surface
and that callback is not back yet, it may cause the wayland event
queue crash.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1729>

rtmp2src: Set buffer timestamp on output buffer

This timestamp information would be useful for queue2 element
when calculating time level and also it makes buffering decision
more reliable.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1727>

d3d11videoprocessor: Fix wrong input/output supportability check

The flag argument of ID3D11VideoProcessorEnumerator::CheckVideoProcessorFormat
method is output value, not input.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1726>

nvcodec: nvsldec: Add support for CUDA memory

Add CUDA memory support. Note that zero copying is not supported yet

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1720>

d3d11memory: Adjust log level for some spammy debug messages

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1723>

d3d11convert: Use ID3D11VideoProcessor only if device supports colorspace

Check whether conversion with given combination of input/output
format and dxgi colorspace is supported or not by driver.
If not, we should use shader.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1723>

d3d11videoprocessor: Add a method for device's conversion caps check

Add a wrapper method for
ID3D11VideoProcessorEnumerator1::CheckVideoProcessorFormatConversion.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1723>

d3d11format: Map more colorimetry with dxgi colorspace

Map more logically identical set of GstVideoColorimetry formats
with dxgi color space.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1723>

va: pool: Set the video_alignment after we get image info.

The set_format() of the allocator may change the stride of the
alloc_info. We should update the video_align.stride_align based
on it. Or, we get a warning in gst_video_meta_validate_alignment().

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1698>

va: h264dec: Set the padding to VideoAlignment's right.

In our va implemenation, we just use image's info to map the buffer.
The padding info just plays a role as a place holder to expand the
allocation size in caps when decoding size is bigger than display
size. So the padding_right or padding_left does not change the result.
But we find if using padding_left, it is hard to meet the requirement
of gst_video_meta_validate_alignment(), when the video meta's stride
is different from the allocation width.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1698>

va: basedec: No need to call base class' decide_allocation().

We have already done the jobs in gst_va_base_dec_decide_allocation()
and no need to call base class' decide_allocation() again. The base
class' decide_allocation() will set_format() again and let use do the
image/surface testing again, which is low performance and no needed.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1698>

va: allocator: use _update_image_info() to set allocator parameters.

Use this standalone function to update the allocator info and make
all ensure_image() and mem_alloc() API clean.
We also change the default way of using image. We now set the non
derive manner as the default manner, and if it fails, then fallback
to the derived image manner.
On a lot of platforms, the derived image does not have caches, so the
read and write operations have very low performance.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1698>

va: allocator: Add a helper function to update the image info.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1698>

va: allocator: Decide the allocator's parameters when set_format().

Moving the parameters testing and setting from the allocator_alloc_full()
to the allocator_try(). The allocator_alloc_full() will be called every
time when we need to allocate a new memory. But all these parameters such
as the surface and the image format, rt_format, etc, are unchanged during
the whole allocator lifetime. Just setting them in set_format() is enough.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1698>

opencv: allow compilation against 4.5.x

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1709>

jpeg2000parse: sub-sampling parse should take component into account

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1653>

jpeg2000parse: no pts interpolation with subframe.

The jpeg2000parser must not interpolate PTS with subframes.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1653>

jpeg2000parse: support frame and stripe alignment in caps

forward alignment and num-stripes caps properties

Use caps height when setting caps for subframe

We want downstream to use full frame height, not subframe height

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1653>

decklink: Reset skip counters when starting the sources

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/378>

decklink*src: Post a warning message on the bus about dropped frames

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/378>

decklink*src: Aggregate dropped frame/packet logging

decklink*src currently prints a log entry for every dropped frame and
audio packet. That completely spams the logs.

This change aggregates information about dropped packets and only prints
a message once when dropping starts, and a summary when dropping ends.

Fixes https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/issues/705

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/378>

d3d11memory: Protect view object with lock

Make resource allocation more thread-safe

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1718>

d3d11convert: Add support for conversion using ID3D11VideoProcessor

Output texture of d3d11 decoder cannot have the bind flag
D3D11_BIND_SHADER_RESOURCE (meaning that it cannot be used for shader
input resource). So d3d11convert (and it's subclasses) was copying
texture into another internal texture to use d3d11 shader.
It's obviously overhead and we can avoid texture copy for
colorspace conversion or resizing via ID3D11VideoProcessor
as it supports decoder output texture.
This commit would be a visible optimization for d3d11 decoder with
d3d11compositor use case because we can avoid texture copy per frame.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1718>

d3d11memory: Store ID3D11VideoProcessorOutputView object

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1718>

codecs: h265decoder: Fix picture leaks because of reference set.

The last frame's reference set has no one to cleanup. We need to
clean all pictures in the stop() func.
We also add a helper function to cleanup all the pictures in the
reference picture set.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1717>

codecs: h265decoder: Fix 3 ref array leaks in finalize.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1717>

d3d11window: Reuse ID3D11VideoProcessorInputView if possible

GstMemory object could be disposed if GstBuffer is not allocated
by GstD3D11BufferPool such as via gst_buffer_copy() and/or
gst_buffer_make_writable(). So attaching qdata on GstMemory
object would cause unnecessary view alloc/free.
By using view pool which is implemented in GstD3D11Allocator,
we can avoid redundant view alloc/free.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1716>

d3d11memory: Implement ID3D11VideoProcessorInputView pool

Similar to ID3D11VideoDecoderOutputView pool implementation

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1716>

Meson: Use pkg-config generator

h265decoder: Add support for l0/l1

Add support for reference list needed for VA-API and some V4L2 decoders.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1713>

h265decoder: Sync with the H264 implementation

This ensures that we get the last reference to picture being outputed,
avoiding GstBuffer structure copies and simplifying the buffer management.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1713>

d3d11decoder: Directly access ID3D11VideoDecoderOutputView for decoding

Decoder output view is stored in GstD3D11Memory object instead of
wrapper struct now. So qdata is no more required.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1712>

d3d11memory: Implement ID3D11VideoDecoderOutputView pool

Similar to texture-array pool, we can reuse decoder output view
since the life time of output view is identical to that of texture-array.
In this way, we can avoid frequent output view alloc/free.

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1712>

d3d11memory: Move to GArray to store texture-array status

The size D3D11_REQ_TEXTURE2D_ARRAY_AXIS_DIMENSION is 2048
which is too large in practice especially for a texture
of dpb

Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1712>