Our context is non-persistent, and we propagate it throughout the
pipeline. This means that if we try to reuse any gstmsdk element by
removing it from the pipeline and then re-adding it, we'll clone the
mfxSession and create a new gstmsdk context as a child of the old one
inside `gst_msdk_context_new_with_parent()`.

Normally this only allocates a few KB inside the driver, but on
Windows it seems to allocate tens of MBs which leads to linearly
increasing memory usage for each PLAYING->NULL->PLAYING state cycle
for the process. The contexts will only be freed when the pipeline
itself goes to `NULL`, which would defeat the purpose of dynamic
pipelines.

Essentially, we need to optimize the case in which the element is
removed from the pipeline and re-added and the same context is re-set
on it. To detect that case, we set the context on `old_context`, and
compare it to the new one when preparing the context. If they're the
same, we don't need to do anything.

Fixes gstreamer/gst-plugins-bad#946

Split it out into a separate function with early exits to make the
flow clearer, and document what the function is doing clearly.
No functional changes.

If we're using the software implementation, we should not use video
memory because that will be too slow.

`gst_object_replace()` is not supposed to be used for unreffing and
NULLing objects.

Copied and pasted from gstreamer-vaapi and did a few changes for gst-msdk.

Allow the base class to copy GstVideoRegionOfInterestMeta data to the
output buffer when calling gst_video_decoder_finish_frame

A reconfig is needed when ROI is changed, otherwise the ROI parameters won't
take effect

set_extra_params is added to allow sub class to add extra mfx parameters
for changed input frame or meta data

need_reconfig is added to allow sub class requires a reconfig when
the input frame or the MetaData (e.g. GstVideoRegionOfInterestMeta)
attached to the input frame is changed.

Seungha Yang authored 5 years ago and GStreamer Marge Bot committed 5 years ago

... and handle EOS and EOB nals.

Only the first CRA picture should be associated with NoRaslOutputFlag
as the comment in code.

`@videolibdir@` isn't used anymore.

`pipe()` isn't used since 15927b65,
and `socketpair()` from `#include <sys/socket.h>` is used only in the
examples. In practice, you can use probably also use anything that
allows you to create fd pairs, such as named pipes or anonymous pipes.

We use the cross-platform GstPollFD API in the plugin.

Sebastian Dröge authored 5 years ago and GStreamer Marge Bot committed 5 years ago

Default to 150ms instead of 8 frames, which seems to work in the
majority of cases.

Use consistent memory layout between dxva and other shader use case.
For example, use DXGI_FORMAT_NV12 texture format instead of
two textures with DXGI_FORMAT_R8_UNORM and DXGI_FORMAT_R8G8_UNORM.

This reverts commit ddd13fc7

Dynamic usage can reduce the number of copy per frame but make
things complicated and the benefit seems to not significant.
Also since we don't provide _map() method for the dynamic usage,
application cannot read buffers which make "last-sample" property
unusable in case of d3d11videosink.

Add any ICE candidates from the SDP before adding pending
trickle ICE candidates to support non-trickle peers

Fixes gstreamer/gst-plugins-bad#678

DXVA requires a hardware interface but may not be available,
such as in the case of VMs or when the GPU vendor does not provide a decoder interface.

... and change some debug levels since initialization failure might not be fatal.

Shinya Saito authored 5 years ago and Nicolas Dufresne committed 5 years ago

xdg_shell fullscreen mode doesn't work for committing
xdg_surface without configure acknowledgement.

In addition, we can't set different surface setting from
acknowledged config in this mode.

Josep Torra authored 5 years ago and GStreamer Marge Bot committed 5 years ago

Some raw h264 encoded files trigger the assignment of wrong PTS to buffers
when some SEI data is provided. This change prevents it to happen.

Also ensure this behavior is being tested.

We might have some old timecodes that are in the future now and have to
drop those to make sure that our queue is correctly ordered and we don't
have multiple timecodes for the same running time.

Directly read them out of the decoder as soon as we passed audio and
then store them in a queue that we handle internally together with their
timestamps. This cleans up memory management and gives us proper control
over the queue instead of guessing how the queue inside the LTC decoder
actually works and when it overflows.

And don't introduce any latency at all if not LTC audio pad was
requested.

And also introduce 6 instead of 2 frames of latency compared to the LTC
audio input as that seems to be an upper bound for how much the LTC
library is lagging behind.

Otherwise we don't interpolate between LTC timecodes but only ever put
an LTC timecode on buffers once we actually received one.

Nicolas Dufresne authored 5 years ago and Sebastian Dröge committed 5 years ago

This g_once() is called everywhere, even in provider. This cause
spurious error when device monitor is used. Just silence or remove
the spurious logs.

AES128 support was added since nettle version 3.0

../subprojects/gst-plugins-bad/ext/hls/gsthlsdemux.h:110:10: error: field ‘ctx’ has incomplete type
   struct CBC_CTX (struct aes128_ctx, AES_BLOCK_SIZE) aes_ctx;

Commit a1584b6f caused big performance drop if the downstream element
is not a msdk element because it is very slow to read data from video
memory directly.

This reverts commit a1584b6f.

By default new test cases are disabled since it might be timed out
or test environment might not have render device.

P016 format is no different than P010. Not much things to add code.

... and remove code for RGBx since it's not supported format
by our d3d11 implementation for now.