In case that pipeline is like ".. ! decoder ! encoder ! ..." with using
video memory,
decoder needs to know the async depth of the following msdk element so
that it could
allocate the correct number of video memory.
Otherwise, decoder's memory is exhausted while processing.
https://bugzilla.gnome.org/show_bug.cgi?id=790752
How to share/create GstMsdkcontext is the following:
- Search GstMsdkContext if there's in the pipeline.
- If found, check if it's decoder, encoder or vpp by job type.
- If it's same job type, it creates another instance of
GstMsdkContext
with joined-session.
- Otherwise just use the shared GstMsdkContext.
- If not found, just creates new instance of GstMsdkContext.
https://bugzilla.gnome.org/show_bug.cgi?id=790752
According to the driver's instruction, if there are two or more encoders
or decoders in a process, the session should be joined by
MFXJoinSession.
To achieve this successfully by GstContext, this patch adds job type
specified if it's encoder, decoder or vpp.
If a msdk element gets to know if joining session is needed by the
shared context,
it should create another instance of GstContext with joined session,
which
is not shared.
https://bugzilla.gnome.org/show_bug.cgi?id=790752
1\ In decide_allocation, it makes its own msdk bufferpool.
- If downstream supports video meta, it just replace it with the msdk
bufferpool.
- If not, it uses the msdk bufferpool as a side pool, which will be
decoded into.
and will copy it to downstream's bufferpool.
2\ Decide if using video memory or system memory.
- This is not completed in this patch.
- It might be decided in update_src_caps.
- But tested for both system memory and video memory cases.
https://bugzilla.gnome.org/show_bug.cgi?id=790752
1\ Proposes msdk bufferpool to upstream.
- If upstream has accepted the proposed msdk bufferpool,
encoder can get msdk surface from the buffer directly.
- If not, encoder get msdk surface its own msdk bufferpool
and copy from upstream's frame to the surface.
2\ Replace arrays of surfaces with msdk bufferpool.
3\ In case of using VPP, there should be another msdk bufferpool
with NV12 info so that it could convert first and encode.
Calls gst_msdk_set_frame_allocator and uses video memory only on linux.
and uses system memory on Windows until d3d allocator is implemented.
https://bugzilla.gnome.org/show_bug.cgi?id=790752
Implements 2 memory allocators:
1\ GstMsdkSystemAllocator: This will allocate system memory.
2\ GstMsdkVideoAllocator: This will allocate device memory depending
on the platform. (eg. VASurface)
Currently GstMsdkBufferPool uses video allocator currently by default
only on linux. On Windows, we should use system memory until d3d
allocator
is implemented.
https://bugzilla.gnome.org/show_bug.cgi?id=790752
Implements msdk frame allocator which is required from the driver.
Also makes these functions global so that GstMsdkAllocator could use
the allocated video memory later and couple with GstMsdkMemory.
GstMsdkContext keeps allocation information such as mfxFrameAllocRequest
and mfxFrameAllocResponse after allocation.
https://bugzilla.gnome.org/show_bug.cgi?id=790752
Makes GstMsdkContext to be a descendant of GstObject so that
we could track the life-cycle of the session of the driver.
Also replaces MsdkContext with this one.
Keeps msdk_d3d.c alive for the future.
https://bugzilla.gnome.org/show_bug.cgi?id=790752
Same changes as done for wasapisink in cbe2fc40a. Turns out this is
sometimes also needed for capture. Reported by Mathieu_Du.
Also improve logging in that case for easier debugging.
Sometimes the minimum period advertised by a card results in an
unaligned buffer size error during initialization in exclusive mode.
In that case, we can fetch the actual buffer size in frames and
calculate the period from that.
We can't do this pre-emptively because we can't call GetBufferSize
till Initialize has been called at least once.
https://bugzilla.gnome.org/show_bug.cgi?id=793289
This reduces the chances of startup glitches, and also reduces the
chances that we'll get garbled output due to driver bugs.
Recommended by the WASAPI documentation.
https://bugzilla.gnome.org/show_bug.cgi?id=793289
So far, we have been completely discarding the values of latency-time
and buffer-time and trying to always open the device in the lowest
latency mode possible. However, sometimes this is a bad idea:
1. When we want to save power/CPU and don't want low latency
2. When the lowest latency setting causes glitches
3. Other audio-driver bugs
Now we will try to follow the user-set values of latency-time and
buffer-time in shared mode, and only latency-time in exclusive mode (we
have no control over the hardware buffer size, and there is no use in
setting GstAudioRingBuffer size to something larger).
The elements will still try to open the devices in the lowest latency
mode possible if you set the "low-latency" property to "true".
https://bugzilla.gnome.org/show_bug.cgi?id=793289
This requires using allocated strings, but it's the best option. For
instance, a call could fail because CoInitialize() wasn't called, or
because some other thing in the stack failed.
https://bugzilla.gnome.org/show_bug.cgi?id=793289
This is particularly important when running in exclusive mode because
any delays will immediately cause glitching.
The MinGW version in Cerbero is too old, so we can only enable this when
building with MSVC or when people build GStreamer for MSYS2 or other
MinGW-based distributions.
To force-enable this code when building with MinGW, build with
CFLAGS="-DGST_FORCE_WIN_AVRT -lavrt".
https://bugzilla.gnome.org/show_bug.cgi?id=793289
This provides much lower latency compared to opening in shared mode,
but it also means that the device cannot be opened by any other
application. The advantage is that the achievable latency is much
lower.
In shared mode, WASAPI's engine period is 10ms, and so that is the
lowest latency achievable.
In exclusive mode, the limit is the device period itself, which in my
testing with USB DACs, on-board PCI sound-cards, and HDMI cards is
between 2ms and 3.33ms.
We set our audioringbuffer limits to match the device, so the
achievable sink latency is 6-9ms. Further improvements can be made if
needed.
https://bugzilla.gnome.org/show_bug.cgi?id=793289
We will use ->device for storing a pointer to the IMMDevice structure
which is needed for fetching the caps supported by devices in
exclusive mode.
https://bugzilla.gnome.org/show_bug.cgi?id=793289
This will set the actual-latency-time and actual-buffer-time of the sink
and source.
We completely ignore the latency-time/buffer-time values set
on the element because WASAPI is happiest when it is reading/writing at
the default period. Improving this will likely require the use of the
IAudioClient3 interfaces which are not available in MinGW yet.
https://bugzilla.gnome.org/show_bug.cgi?id=792897
Currently only does probing and does not handle messages from
endpoints/devices. In the future we want to do proper monitoring which
is well-supported in WASAPI.
https://bugzilla.gnome.org/show_bug.cgi?id=792897
We need to parse the WAVEFORMATEXTENSIBLE structure, figure out what
positions the channels have (if they are positional), and reorder them
as necessary.
https://bugzilla.gnome.org/show_bug.cgi?id=792897
There is no fixed limitation for the number of devices on the
decklink API side according to BlackMagic. Many PC motherboards
are able support 6 decklink cards each with up to 8 inputs so
a limit of 16 might well be too low.
https://bugzilla.gnome.org/show_bug.cgi?id=777239
Both the source and the sink elements were broken in a number of ways:
* prepare() was assuming that the format was always S16LE 2ch 44.1KHz.
We now probe the preferred format with GetMixFormat().
* Device initialization was done with the wrong buffer size
(buffer_time is in microseconds, not nanoseconds).
* sink_write() and src_read() were just plain wrong and would never
write or read anything useful.
* Some functions in prepare() were always returning FALSE which meant
trying to use the elements would *always* fail.
* get_caps() and delay() were not implemented at all.
TODO: support for >2 channels
TODO: pro-audio low-latency
TODO: SPDIF and other encoded passthroughs
Three new properties are now implemented: role, mute, and device.
* 'role' designates the stream role of the initialized device, see:
https://msdn.microsoft.com/en-us/library/windows/desktop/dd370842(v=vs.85).aspx
* 'device' is a system-wide GUIDesque string for a specific device.
* 'mute' is a sink property and simply mutes it.
On my Windows 8.1 system, the lowest latency that works is:
wasapisrc buffer-time=20000
wasapisink buffer-time=10000
aka, 20ms and 10ms respectively. These values are close to the lowest
possible with the IAudioClient interface. Further improvements require
porting to IAudioClient2 or IAudioClient3.
https://docs.microsoft.com/en-us/windows-hardware/drivers/audio/low-latency-audio
Sometimes we might get an audio packet without a corresponding video
frame. In these cases, the stream and hardware reference timestamps
would be missing, because they're called on the video frame. Instead of
potentially breaking stuff downstream that might depend on these, we now
extrapolate them.
https://bugzilla.gnome.org/show_bug.cgi?id=792042
When we receive a video or audio buffer, we calculate the next stream
time based on the current stream time + buffer duration. If the next
buffer's stream time is after that, we issue a warning.
This happens because the stream time incoming from Decklink should be
really constant and without gaps. If there is a gap, it means that
something went wrong, e.g. the internal buffer pool is empty (too many
buffers queued up downstream).
https://bugzilla.gnome.org/show_bug.cgi?id=781776
Sometimes we might get an audio packet without a corresponding video
frame. In these cases, the stream and hardware reference timestamps
would be missing, because they're called on the video frame. Instead of
potentially breaking stuff downstream that might depend on these, we now
extrapolate them.
https://bugzilla.gnome.org/show_bug.cgi?id=792042
The correct behaviour of anything stuck in the ->render() function
between ->unlock() and ->unlock_stop() is to call
gst_base_sink_wait_preroll() and only return an error if this returns an
error, otherwise, it must continue where it left off!
https://bugzilla.gnome.org/show_bug.cgi?id=774950
