This adds a few missing gst_object_unref calls for the opengl context in
gstglwindow_gbm_egl.c, as well as the missing close call for the
drm node fd in gst_gl_display_gbm_finalize.
While this creates a circular reference between the pipeline and the
context, this ensures that the context stays alive for as long as any
callbacks could be called on it. The circular reference is broken once
the conversion is finished (or error, or timeout), which will then cause
everything to be freed.
Previously it was possible that a callback could be called on the
context right after it was freed already.
Also use only a single context structure, the second structure does not
simplify anything and duplicates storage.
This patch adds API in the audio decoder base class for setting the arbitrary
caps on the source pad. Previously only caps converted from audio info were
possible. This is particularly useful when subclass wants to set caps features
for audio decoder producing metadata.
The Y210 format was added in the middle of the formats enum and list,
introducing an ABI break.
This issue was detected thanks to the gstreamer-rs test harness.
We assume here the same data format for the user data as for the
DID/SDID: 10 bits with parity in the upper 2 bits. In theory some
standards could define this differently and even have full 10 bits of
user data but there does not seem to be a single such standard after
all these years.
Currently in Python it would become a signed 64 bit value but should
actually be an unsigned 32 bit value with all bits set.
This is the same problem as with GST_MESSAGE_TYPE_ANY.
See https://bugzilla.gnome.org/show_bug.cgi?id=732633
Rather then letting gst_gl_memory_setup_buffer guess the GL format used
for an eglimage after importing a dmabuf be explicit about it. This
fixes issues where dmabuf import may have used another format then
gst_gl_format_from_video_info would guess on the basis of the available
GL extensions.
In particular on etnaviv the gst_gl_format_from_video_info would
assuming a luminance + alpha GL format is used for YUY2, but the dmabuf
import will always use RG88. Which causes images to end up somewhat pink when
displayed on the screen.
When importing an egl image from dmabuf gst_gl_format_from_video_info
was used to work what the result GL format will be. Unfortunately that
will only work if the conventional format and the choosen DRM fourcc for
the format match up.
On etnaviv platforms there is no support for GL_EXT_texture_rg, so the
GL format chosen for YUY2 ends up being GST_GL_LUMINANCE_ALPHA. However
DRM does not do luminance + alpha as it's a legacy GL thing, so the
dmabuf import ends up using DRM_FORMAT_GR88.
To fix this, tie the DRM_FORMAT and the GL format together so they
always match up.
There is new code that ensures that we renegotiate after an
uploader transition if the negotiated caps have changed.
The problem is that the raw uploader will not really try and
fixate the input caps, but instead of return a subset with the
only the supported target texture.
This had two effect, raw uploads was always done renegotiated
once and the raw upload unit test was now failing as it didn't
expect a renegotiation.
As it's a valid check, simply relax the gst_caps_is_equal() check
and use a gst_caps_is_subset() instead.
https://bugzilla.gnome.org/show_bug.cgi?id=783521
The direct dmabuf upload does color conversion, so when it transforms
the caps, it replaces the format with all formats found through the
format query. When this uploader can't be used, it makes the upstream
source pick a unsupported format.
To fix this, we only append the caps with a list of format. So the
source will only pick one of these formats if the downstream preferred
format is not supported. A negotiation failure after this would be
normal.
This fixes pipelines without a glcolorconvert element.
https://bugzilla.gnome.org/show_bug.cgi?id=783521
The idea is that some GPUs (like the Vivante series) can actually
perform the YUV->RGB conversion internally, so no custom conversion
shaders are needed. To make use of this feature, we need an additional
uploader that can import DMABUF FDs and also directly pass the pixel
format, relying on the GPU to do the conversion.
Based on patches from Nicolas Dufresne <nicolas.dufresne@collabora.com> and
Carlos Rafael Giani <dv@pseudoterminal.org>.
https://bugzilla.gnome.org/show_bug.cgi?id=783521
If a upload method is selected then use it exclusively in transform_caps().
Also, reconfigure if the current caps don't match the current upload
method.
https://bugzilla.gnome.org/show_bug.cgi?id=783521
This should not be necessary, but currently not all plugins that provide
dmabuf memory announce this with caps features, e.g. v4l2.
The static caps already contain the system memory. It didn't break before
because other upload methods provide the necessary transformation.
https://bugzilla.gnome.org/show_bug.cgi?id=783521
Reconfigure will trigger a set_caps which clears the upload method.
Remember the method in this case and start with it.
Wrap around once to try all methods if necessary.
https://bugzilla.gnome.org/show_bug.cgi?id=783521
The colorspace conversion happens during the upload so the necessary hints
must be provided to ensure that the conversion works correctly.
At least the Mesa Intel driver will create a texture without error but
produces an incorrect result. Use eglQueryDmaBufModifiersEXT() to check if
non-external upload is supported for the given format.
Based on a patch from Carlos Rafael Giani <dv@pseudoterminal.org>.
https://bugzilla.gnome.org/show_bug.cgi?id=783521
gst_gl_memory_setup_buffer() was not properly using the number
of pointers to wrapped. This also fixes the validation, as we
only support 1 wrapper per view, or num_planes * views wrapper.
https://bugzilla.gnome.org/show_bug.cgi?id=783521
rtsp_connection_send takes care of adding those already,
and some reverse proxies such as nginx will reject the request
altogether if the Authorization header is present twice,
even with the same value.
https://bugzilla.gnome.org/show_bug.cgi?id=797272
Add a source-info property that will read/write meta to the buffers
about RTP source information. The GstRTPSourceMeta can be used to
transport information about the origin of a buffer, e.g. the sources
that is included in a mixed audio buffer.
A new function gst_rtp_base_payload_allocate_output_buffer() is added
for payloaders to use to allocate the output RTP buffer with the correct
number of CSRCs according to the meta and fill it.
RTPSourceMeta does not make sense on RTP buffers since the information
is in the RTP header. So the payloader will strip the meta from the
output buffer.
https://bugzilla.gnome.org/show_bug.cgi?id=761947
Using the correct blend modes for each case or converting to
premultipled in the very unlikely case that separate blend modes are
unavailable on ancient opengl hardware.
It was checking for GST_IS_CAPS only and that would fail if the new
restriction caps was NULL and its documentation says it accepts NULL as
valid input.
If we have an upstream GST_EVENT_STREAM_START, use that one instead
of creating a new one which could be completely different from the
upstream one and drop information (like the stream flags and stream
object).
Only create a new event if we don't already have one from upstream
https://bugzilla.gnome.org/show_bug.cgi?id=797215
For each lib we build export its own API in headers when we're
building it, otherwise import the API from the headers.
This fixes linker warnings on Windows when building with MSVC.
The problem was that we had defined all GST_*_API decorators
unconditionally to GST_EXPORT. This was intentional and only
supposed to be temporary, but caused linker warnings because
we tell the linker that we want to export all symbols even
those from externall DLLs, and when the linker notices that
they were in external DLLS and not present locally it warns.
What we need to do when building each library is: export
the library's own symbols and import all other symbols. To
this end we define e.g. BUILDING_GST_FOO and then we define
the GST_FOO_API decorator either to export or to import
symbols depending on whether BUILDING_GST_FOO is set or not.
That way external users of each library API automatically
get the import.
While we're at it, add new GST_API_EXPORT in config.h and use
that for GST_*_API decorators instead of GST_EXPORT.
The right export define depends on the toolchain and whether
we're using -fvisibility=hidden or not, so it's better to set it
to the right thing directly than hard-coding a compiler whitelist
in the public header.
We put the export define into config.h instead of passing it via the
command line to the compiler because it might contain spaces and brackets
and in the autotools scenario we'd have to pass that through multiple
layers of plumbing and Makefile/shell escaping and we're just not going
to be *that* lucky.
The export define is only used if we're compiling our lib, not by external
users of the lib headers, so it's not a problem to put it into config.h
Also, this means all .c files of libs need to include config.h
to get the export marker defined, so fix up a few that didn't
include config.h.
This commit depends on a common submodule commit that makes gst-glib-gen.mak
add an #include "config.h" to generated enum/marshal .c files for the
autotools build.
https://bugzilla.gnome.org/show_bug.cgi?id=797185
There are a lot of symbols in GstGL-1.0.gir generated by automake that
are not when it is generated by meson, because a lot of headers were
not included in meson's gir generation.
https://bugzilla.gnome.org/show_bug.cgi?id=797158
The purpose of value_name in GEnumValue structure is to express in
text format the name of the enum, not a description of the value, so
it can be use later for the gir file generation.
https://bugzilla.gnome.org/show_bug.cgi?id=797144
Reconfigure will trigger a set_caps which clears the upload method.
Remember the method in this case and start with it.
Wrap around once to try all methods if necessary.
https://bugzilla.gnome.org/show_bug.cgi?id=783521
The idea is that some GPUs (like the Vivante series) can actually
perform the YUV->RGB conversion internally, so no custom conversion
shaders are needed. To make use of this feature, we need an additional
uploader that can import DMABUF FDs and also directly pass the pixel
format, relying on the GPU to do the conversion.
Based on patches from Nicolas Dufresne <nicolas.dufresne@collabora.com> and
Carlos Rafael Giani <dv@pseudoterminal.org>.
https://bugzilla.gnome.org/show_bug.cgi?id=783521
The colorspace conversion happens during the upload so the necessary hints
must be provided to ensure that the conversion works correctly.
At least the Mesa Intel driver will create a texture without error but
produces an incorrect result. Use eglQueryDmaBufModifiersEXT() to check if
non-external upload is supported for the given format.
Based on a patch from Carlos Rafael Giani <dv@pseudoterminal.org>.
https://bugzilla.gnome.org/show_bug.cgi?id=783521
If a upload method is selected then use it exclusively in transform_caps().
Also, reconfigure if the current caps don't match the current upload
method.
https://bugzilla.gnome.org/show_bug.cgi?id=783521
This should not be necessary, but currently not all plugins that provide
dmabuf memory announce this with caps features, e.g. v4l2.
The static caps already contain the system memory. It didn't break before
because other upload methods provide the necessary transformation.
https://bugzilla.gnome.org/show_bug.cgi?id=783521
gst_gl_memory_setup_buffer() was not properly using the number
of pointers to wrapped. This also fixes the validation, as we
only support 1 wrapper per view, or num_planes * views wrapper.
https://bugzilla.gnome.org/show_bug.cgi?id=783521
The gst_video_decoder_clip_and_push_buf() now drops the internal stream
lock while pushing. This means, the output_queued list could be modififed
during that time. To make the code safe again, we delete the link before
pushing the data. The walk pointer will later be updated with the list
head, which makes it safe in case the list was modififed.
https://bugzilla.gnome.org/show_bug.cgi?id=715192
On Windows, the ringbuffer thread function must have the "Pro Audio"
priority set, otherwise it sometimes doesn't get scheduled for
200-300ms, which will immediately cause an underrun unless you set
a very high latency-time and buffer-time.
This has no compile-time deps since it tries to load avrt.dll at
runtime to set the thread priority.
Release STREAM_LOCK before calling gst_pad_push() and take it
back afterward so that upstream isn't blocked while output
buffer is being pushed downstream.
https://bugzilla.gnome.org/show_bug.cgi?id=715192
Release STREAM_LOCK before calling gst_pad_push() and take it
back afterward so that upstream isn't blocked while output
buffer is being pushed downstream.
https://bugzilla.gnome.org/show_bug.cgi?id=715192
Add a new macro that gives you the rate of the fields, which is the
numerator of the field-rate for ALTERNATE interlacing video and FPS for
progressive and other interlacing formats.
https://bugzilla.gnome.org/show_bug.cgi?id=796106
Add a variant of gst_video_decoder_set_output_state() that allows the user
to pass an interlacing mode as well. This is needed to ensure that
gst_video_info_set_interlaced_format() is used instead so that
GstVideoInfo.size is correctly initialized.
https://bugzilla.gnome.org/show_bug.cgi?id=796106
Add a new macro that gives you the height of a field. It returns the
height of the full frame unless split-field (alternate) interlacing is
in use. Also GST_VIDEO_INFO_COMP_HEIGHT macro now uses this new macro to
get the height for its calculation.
https://bugzilla.gnome.org/show_bug.cgi?id=796106
Add a helper to set the interlacing mode while creating the GstVideoInfo
in addition to format and resolution. Using this helper will ensure that
size is correctly calculated for split-field interlacing mode.
https://bugzilla.gnome.org/show_bug.cgi?id=796106
Add a new interlace mode enum to represent buffers containing a single
field of an interlaced video in a buffer. The name is based on the
equivalent video format in the V4L2 API, V4L2_FIELD_ALTERNATE:
https://01.org/linuxgraphics/gfx-docs/drm/media/uapi/v4l/field-order.html
Since caps fields are optional, we also introduce a new caps feature,
"format:Interlaced" that always goes with "alternate" interlace mode to ensure
that caps for this incompatible format are incompatible with other interlaced
and progressive video caps.
https://bugzilla.gnome.org/show_bug.cgi?id=796106
gst_gl_window_dispmanx_egl_set_window_handle() removes native window handle
(dispmanx element), regardless it was foreign window handle
(set via gst_video_overlay_set_window_handle()) or not.
This problem prevents glimagesink reusable.
(PAUSED -> READY -> PAUSED does not work)
This patch corrects it comparing the native window handle with foreign window
handle. This behavior is same as gst_gl_window_dispmanx_egl_close().
https://bugzilla.gnome.org/show_bug.cgi?id=785199
The glimagesink wayland backend lacks the implementation of
gst_gl_window_wayland_egl_set_preferred_size. Because of this, glimagesink windows on
wayland are created with a fixed window size of 320x240.
[Matthew Waters]: gst-indent sources
https://bugzilla.gnome.org/show_bug.cgi?id=789384
This allows consumers of the gstgl dependency where gstgl is optional
to do things like:
config_data.set('HAVE_GST_GL', gstgl_dep.found())
deps = [gstvideo_dep, gstgl_dep]
meaning they can still use the dep unconditionally. With the
disabler we would just disable the whole target even if the
gstgl part was an optional extra. We can add an option to
dependency() later to let users/consumers of the dep decide
if they want a not-found dependency or a disabler instead.
It's also needed by the generic/states test and the variable
is currently checked as part of the opengl lib tests so wouldn't
be available if opengl was disabled.
Function body must only be removed if compiling with
GST_REMOVE_DEPRECATED. Function declaration must not
be skipped if GST_DISABLE_DEPRECATED is defined, since
it contains our export decorator and we wouldn't export
the symbol as public API any more then since we compile
with GST_DISABLE_DEPRECATED.
Building an image with yocto for an 'am335x' processor will fail,
because 'gbm.h' can not be found and 'EGL_PLATFORM_GBM_MESA' is not defined.
So simply adding define of 'EGL_PLATFORM_GBM_MESA' in 'gstgldisplay_egl.c'
and adding 'GBM_CFLAGS' in 'gst-gl.m4' to get rid of this errors and complete
a full build process.
ERRORs were:
-----------
gst-plugins-base-1.14.1/gst-libs/gst/gl/gbm/gstglwindow_gbm_egl.h:24:10:
fatal error: gbm.h: No such file or directory
#include <gbm.h>
^~~~~~~
gst-plugins-base-1.14.1/gst-libs/gst/gl/egl/gstgldisplay_egl.c:157:39:
error: 'EGL_PLATFORM_GBM_MESA' undeclared (first use in this function);
did you mean 'EGL_PLATFORM_WAYLAND'?
ret = _gst_eglGetPlatformDisplay (EGL_PLATFORM_GBM_MESA, (gpointer) display,
^~~~~~~~~~~~~~~~~~~~~
https://bugzilla.gnome.org/show_bug.cgi?id=796885
This is useful if the output buffers are planar and have extra padding
on each plane, in which case size/bpf does not represent the number of
valid samples.
https://bugzilla.gnome.org/show_bug.cgi?id=705977
Aggregation will break the layout, as it concatenates buffers,
and fixing it here would be much more inefficient than configuring
the actual decoder implementation to output larger buffers.
https://bugzilla.gnome.org/show_bug.cgi?id=705977
This was miss-leading VAAPI which was not forcing linear back buffers,
would still export DMABuf. That caused bad rendering as the buffers are
tiled and most likely compressed.
https://bugzilla.gnome.org/show_bug.cgi?id=796822
This was miss-leading VAAPI which was not forcing linear back buffers,
would still export DMABuf. That caused bad rendering as the buffers are
tiled and most likely compressed.
V4L2 and OMX decoder don't support draining and keeping reference
frames. As a side effect, these decoder just stops working on
gaps/discont. When this drain was introduced, the commit stated that
this was for TRICKMODE_KEY_UNITS, so only drain if running in this mode.
https://bugzilla.gnome.org/show_bug.cgi?id=796771
e4bf9ed8f0 was not quite right and changed
the wrong thing. Intead we needed to change the multiplication order
and should have kept the previous to/from matrices as is done in this
patch.
We would make the subproject conditional on the wrap-mode=nodownload
however get_option('wrap-mode') or similar is not available from
meson.build files as the wrap-mode is meant to be used automatically.
Instead use the dependency fallback mechanism to get at the subproject
where possible i.e. when downloading is allows and only add the compat
includes when we have a valid internal dependency from the gl-headers
subproject.
https://bugzilla.gnome.org/show_bug.cgi?id=796534
Helper function for bindings, in python for example
users can now replace:
res, msg = GstSdp.SDPMessage.new()
GstSdp.sdp_message_parse_buffer(bytes(text.encode()), msg)
with:
res, msg = GstSdp.SDPMessage.new_from_text(text)
https://bugzilla.gnome.org/show_bug.cgi?id=796563
This reverts commit 82c43ff9a3.
Reverting for the time being, since it trips up build bots
that don't have network connectivity.
We ideally shouldn't have a subproject that's used unconditionally.
The external time should be moved only as much as needed
to get back to the ideal center point, so that the clock
is still allowed to drift both directions after the correction.
This reduces excessive back and forth corrections that were
caused by the assumption of a linear drift.
https://bugzilla.gnome.org/show_bug.cgi?id=788006
The matrices were converting the wrong values with non-diagonal-only matrices.
e.g. a typical yflip matrix in [-1,1]^3 such as
1 0 0 0
0 -1 0 0
0 0 1 0
0 0 0 1
Would have actually required a matrix like this in [0,1]^3
1 0 0 0
0 -1 0 0
0 0 1 0
0 -2 0 1
Which is
1. not consistent with our multiplication convention and would require
transposing matrices or changing our multiplication order (from what is
generally used on opengl matrix guides/tutorials).
2. Produces incorrect values when input with actual vertices accounting for
the difference in multiplication order. e.g. some vertices multiplied by
the yflip matrix using vertex * yflip(== transpose(yflip) * vertex):
vertex: -> result: expected:
vec4(1,0,1,1) -> vec4(1,-2,1,1) vec4(1,1,1,1)
vec4(1,1,1,1) -> vec4(1,-3,1,1) vec4(1,0,1,1)
With the updated values, we now get the expected values.
Includes a test for this behaviour and the example above
The autotools build installs the file and sets the LICENSE_TRANSLATIONS_PATH
macro which licences.c uses as a fallback path. Do the same with meson.
https://bugzilla.gnome.org/show_bug.cgi?id=796274
This pixel format is a fully packed variant of NV12_10LE32,
a luma pixel would take 10bits in memory, without any
filled bits between pixels in a stride. The color range
follows the BT.2020 standard.
In order to get a better performance in hardware memory
operation, it may expend the stride, append zero data at the
end of echo lines.
Pack function by Nicolas Dufresne.
https://bugzilla.gnome.org/show_bug.cgi?id=795462
Signed-off-by: Nicolas Dufresne <nicolas@ndufresne.ca>
Signed-off-by: ayaka <ayaka@soulik.info>
This moves all the conversion related code to a single place, allows
less code-duplication inside compositor and makes the glmixer code less
awkward. It's also the same pattern as used by GstAudioAggregator.
For a single format in the caps, we were returning some weird answers,
like only RGB formats for a RGB input when we can also support YUV from
RGB.
Fixup of 3cfff727b1 where I thought my
previous (~6month) self had got this right. Don't trust your previous
self people!
Also yield common options to the outer project (gst-build in our case)
so that they don't have to be set manually and use array types for some
options.
Otherwise subclasses might accidentially use the old audioinfo/caps.
None of the subclasses currently uses the audioinfo/caps, but future
subclasses might.
https://bugzilla.gnome.org/show_bug.cgi?id=795827
This is only used for caching reasons and should never actually be in
the public API. If this is ever a bottleneck later, caching around a
class private struct could be implemented.
The aggregated_frame is now called prepared_frame and passed to the
prepare_frame and cleanup_frame virtual methods directly. For the
currently queued buffer there is a method on the video aggregator pad
now.
Instead of having special cases at each GL texture creation, upload,
readback or copy for all non-8-bits-per-components.
Simply store the more specific format and retrieve the generic
component/type tuple from that.
Introduce a helper function for retrieving the generic GL format (RGBA,
RGB, RG, R, L, A) and type (BYTE, SHORT, SHORT_5_6_5) from a sized
GL format enum (RGBA8, RGB565, RG8, etc).
The problem is that even though the functions we are calling are
in-place transformation, orc automatically puts the restrict keyword
on all arguments. To silence that warning just create yet-another
variable containing the same value.
https://bugzilla.gnome.org/show_bug.cgi?id=795765
It is possible that both application and the stream are waiting
currently, if for example the following happens:
1) app is waiting because no buffer in appsink
2) appsink providing a buffer and waking up app
3) appsink getting another buffer and waiting because it's full now
4) app thread getting back control
Previously step 4 would overwrite that the appsink is currently waiting,
so it would never be signalled again.
https://bugzilla.gnome.org/show_bug.cgi?id=795551
Instead of trying to guess what profile to build, just get the possible
elements to use with the specified caps and determine the
EncodingProfile from it.
https://bugzilla.gnome.org/show_bug.cgi?id=795490
If we can guarantee the lifetime of the fd is longer than
the memory, we can use DONT_CLOSE flag not to close when release.
But it's not provided in gstdmabuf yet while gstfdmemory does.
For example, in case of using VA-API or MSDK, we would need this api.
Otherwise we should call dup to duplicate the fd.
https://bugzilla.gnome.org/show_bug.cgi?id=794829
This pixel format is a fully packed variant of NV12, a luma
pixel would take 10bits in memory, without any filled bits
between pixels in a stride. The color range follows
the BT.2020 standard.
In order to get a performance in hardware memory
operation, it may expend the stride, append zero data at the
end of echo lines.
Signed-off-by: ayaka <ayaka@soulik.info>
https://bugzilla.gnome.org/show_bug.cgi?id=795462
In the situation described in
https://bugzilla.gnome.org/show_bug.cgi?id=795397,
downstream_caps consists of two structures, the first with
the preferred rate, if at all possible (44100), the second
containing the full range of allowed rates, as audioresample
correctly tries to negotiate passthrough caps.
As audioaggregator cannot perform rate conversion, it wants
to return a fixated rate in its getcaps implementation,
however it previously directly used the first structure in
the caps allowed downstream, without taking the filter into
consideration, to determine the rate to fixate to.
With this, we first intersect our downstream caps with the
filter, in order not to fixate to an unsupported rate.
Otherwise it won't be possible to specify some profiles such as
video/x-h264,profile=(string)high-4:4:4
With this patch, we can do
video/x-h264,profile=(string)high-4\:4\:4
The default implementation for packet loss handling previously
always sent a gap event.
While this is correct as long as we know the packet that was
lost was actually a media packet, with ULPFEC this becomes
a bit more complicated, as we do not know whether the packet
that was lost was a FEC packet, in which case it is better
to not actually send any gap events in the default implementation.
Some payloaders can be more clever about, for example VP8 can
use the picture-id, and the M and S bits to determine whether
the missing packet was inside an encoded frame or outside,
and thus whether if it was a media packet or a FEC packet,
which is why ulpfecdec still lets these lost events go through,
though stripping them of their seqnum, and appending a new
"might-have-been-fec" field to them.
This is all a bit terrible, but necessary to have ULPFEC
integrate properly with the rest of our RTP stack.
https://bugzilla.gnome.org/show_bug.cgi?id=794909
the meta initialization function is provided *after* the base implementation
fields have been set so do *NOT* reset them otherwise it would result
in corrupted GstMeta.
Instead explicitely set our fields to the default values we actually want.
