The default implementation was required when superframe parsing
was handled by vp9decoder. For instance, if a superframe consists
of multiple frames with show_existing_frame header, it was vague
that which GstVp9Picture should consume GstVideoCodecFrame.
After 1.18 release, we introduced vp9parse element and
superframe should be handled by upstream vp9parse elemenet now.
So, we don't need to care about the superframe at vp9decoder class
level anymore. Simply, a frame corresponding to show_existing_frame
can be dropped if subclass doesn't implement duplicate_picture().
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2151>
DMABuf allocator already implements DMABuf Sync, meaning that doing
mmap/munmap (unless the mode have changed) is not required. In fact, on
systems with IOMMU it makes the kernel redo the mmu table which is visible
in the CPU usage.
This change reduces CPU usage when decoding
bbb_sunflower_2160p_60fps_normal.mp4 on RK3399 SoC from over 30% to
around 15%.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2152>
Implementation of mem_copy() virtual method for GstVaAllocator.
It's a deep copy where a new VA memory is popped out from the pool or,
if pool is empty, a new memory is allocated. The original memory is
mapped to read, and if its VAImage is not derived and size to copy is
the whole surface, the mapped VAImage of the original memory is put in
the new memory. Otherwise a slow memcpy is done between both memories.
Fixes: #1568
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2136>
Existing VP9 parser implementation doesn't provide information
required by other stateless decoding APIs (i.e., DXVA and NVDEC),
specifically loop filter and segmentation parameters might not exist
current frame. So parser needs to fill the information by using previously
parsed information.
We can update the gstvp9parser implementation so that it can provide
all information required by stateless decoding APIs with a huge API break,
or adding more ugly struct in it.
Instead doing as such, this commit introduce a new VP9 parser implementation.
What is different from existing one?
* All variables will follow the specification as much as possible:
VP9 Bitstream & Decoding Process Specification - v0.6 31st March 2016
* Parser will fill all the required information for decoding frame
to GstVp9FrameHeader struct. In case of old VP9 parser,
user needs to read additional data from parser's member variables.
* GstVp9StatefulParser object struct is completely completely opaque
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2112>
The picture buffer (V4L2 CAPTURE buffer) was being released immediatly
when the request was done. This was problematic since even after the
request is done, the picture buffer might still be used as a reference
and should not be reused for further decoding yet.
This change effectively bind the picture buffer lifetime to the request.
So that if the picture is never showned (decode only frame) or the request
queue is full before the buffer is displayed, the picture buffer will
remain alive.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2142>
Different VA drives might have different definitions for RGB32 color
formats because different bit interpretation. Sadly the specification
doesn't clarify these interpretations. So VA users have to figure out
what's the correct mapping with it's rendering color format
definition.
This patch aims to fix the static map structure after the
VAImageFormats are queried. There is another static map with the
different interpretations of the RGB32 formats, and compare them with
the given VAImageFormat, then with the GStreamer color format, update
the mapping table.
Finally, some RGB32 color formats were added.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2129>
Add missing include for sys/ioctl.h so that these warnings dissapear
when compiling:
../subprojects/gst-plugins-bad/sys/v4l2codecs/gstv4l2decoder.c:179:9:
warning: implicit declaration of function ‘ioctl’
[-Wimplicit-function-declaration]
Signed-off-by: Daniel Almeida <daniel.almeida@collabora.com>
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2140>
If the application did not define one yet, define our own
VK_DEFINE_NON_DISPATCHABLE_HANDLE that is independent of the
architecture.
Vulkan, by default, provides a define that depends on the architecture,
which causes the symbol type to be different. This causes an
architecture dependent .gir file, which then causes multilib
installation problems because the .gir files can't be shared.
Make it possible to override the format specifier and provide
a default one that is compatible with the default non dispatchable
handle.
Return VK_NULL_HANDLE from functions that return a non-dispatchable
handle.
Fixes#1566
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2130>
vapostproc tries to be in passthrough mode as much as possible. But
they might be situations where the user might force to process the
frames. For example, when upstream sets the crop meta and the user
wants VA do that cropping, rather than downstream.
For those situations this property will disable the passthrough mode,
if it's enabled.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2058>
If incoming buffers have crop meta it's done by vapostproc, iif
vapostproc is not in passthrough mode and downstream doesn't handle
it.
This patch announces the crop meta API in proposed bufferpool, while
it stops filtering meta APIs, since it was only filter crop api.
Also if downstream supports crop and video metas, vapostporoc
announces both meta APIs in upstream bufferpool.
Finally, the meta is removed from the buffer if the crop is enabled.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2058>
This new struct describes the input and output GstBuffers to
post-process, including VA flags. It also contains the VASurfaceID and
VARectangle, but those are private, completed inside GstVaFilter.
It is used for pass arguments to gst_va_filter_convert_surface() function.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2058>
When an input buffer needs to be copied into a VA memory, it's
required to create a buffer pool. This patch uses the
propose_allocation() caps to instantiate the allocator and pool,
instead of the negotiated caps, which rather represents the resolution
to display.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2058>
Instead of requiring interlaced video, simply skip CC detection
when the input is progressive.
This allows placing line21decoder unconditionally in pipelines,
without having to worry about whether the input stream will be
interlaced, or even worse interlacing just in case!
+ update doc cache
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/1885>
Derived images are direct maps to surfaces bits, but in Intel Gen7 to
Gen9, that memory is not cachable, thus reading can be very slow (it
might produce timeout is tests such as fluster).
This patch tries first to define if derived images are possible, and
later use them only if mapping is not for reading.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2128>
This plugin, for decoders more concretely, assumes that a VA config
can do certain color conversions when mapping frames onto CPU's
memory.
This assumption was valid for i965 and Gallium drivers which generates
valid outputs in bitstreams testers (v.gr. fluster). Nonetheless, iHD,
even when it generates acceptable rendered frames, output's MD5 of
tests weren't valid.
This patch append the image formats, for color conversion when mapping
to memory, for non-iHD drivers.
Part-of: <https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/merge_requests/2128>
