When color management is disabled, the fragment shader was still first
ensuring straight alpha and then immediately just going back to
pre-multiplied. This is near-impossible for a shader compiler to
optimize out, I guess because of the if-statement to handle division by
zero. Having view alpha applied in between certainly didn't make it
easier.
That causes extra fragment computations that are unnecessary. In the
issue report this was found to cause a notable performance regression.
Fix the performance regression by introducing special-case paths for
when straight alpha is not needed. This skips the unnecessary
computations.
Fixes: https://gitlab.freedesktop.org/wayland/weston/-/issues/623
Fixes: 9a6a4e7032
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
Currently, the opaque is discarded for all transformations other than a simple
translation, because correctly transforming the opaque area is not possible in
general.
However, there is one simple case that is probably the most common one: A fully
opaque surface that is translated and scaled. In this case the opaque area is
simply the new bounding box. So set the transformed opaque area accordingly.
Signed-off-by: Michael Olbrich <m.olbrich@pengutronix.de>
If surface->is_opaque is set then we can assume that the whole surface is
opaque. In the trivial case (no transformation or translation only) this means
that transform.boundingbox is exactly the view area and is fully opaque. So it
can be used for transform.opaque.
This is important because damage calculation uses transform.opaque. Without
this, anything underneath a surface without an explicit opaque region but a
pixel format without alpha channel is drawn unnecessarily.
Signed-off-by: Michael Olbrich <m.olbrich@pengutronix.de>
The drm_device is initialized as a side effect of the (badly named)
drm_device_is_kms function. Explicitly pass the drm_device to be able to
initialize kms devices that are not the main drm device of the drm backend.
Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
If we have multiple drm devices, we cannot use the drm device from the backend,
because we would only get the primary device and not the device of the output.
Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
If Weston receives a hotplug event, it has to check if the hotplug device
actually belongs to the drm device before updating the heads of the device. The
hotplug event should only remove heads that belong to the device and must not
change heads of other devices.
Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
The compositor lists the heads from all devices, but we must only disable the
connectors that belong to the current device. Therefore, other heads must be
ignored.
Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
The outputs, heads, crtcs, and connectors are specific to a drm device and not
the backend in general.
Link them to the device that they belong to to be able to retrieve the
respective device.
Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
The commits happen per device instead of per backend. The pending state is
therefore per device as well. Allow to retrieve the device from the pending
state.
Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
The scanout format for the dma-buf feedback are specific to the kms device that
is used for scanout. Therefore, we have to pass the device of the output when
retrieving the scanout formats.
Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
The fbs are specific to the device on which they will be displayed. Therefore,
we have to tell which device shall be used when we are creating the fb.
Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
The atomic commit is device specific. If we have multiple kms devices, we need
to know which device was used for the atomic commit.
Pass the device instead of the backend through the atomic commit.
Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
Extract the kms device from the backend to allow a better separation of the
backend and the kms device. This will allow to handle multiple kms devices with
a single drm backend.
Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
Get the backend at the beginning of the function instead of retrieving it from
another object in the debug statement. This simplifies refactoring, as the debug
statement is not affected by changes how the backend is retrieved.
Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
The gbm_format is the same as the drm format used by the pixel format.
Print the format name using the pixel format in the error message to make the
error message easier to understand for humans.
Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
The fb already contains a DRM fd for later use. So just use that one instead of
fetching it from the backend.
This is necessary if the fbs are allocated on different devices, since otherwise
the wrong device might be used to get the fd of the passed fb.
Signed-off-by: Michael Olbrich <m.olbrich@pengutronix.de>
Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
Looks like we missed this one during the conversion to
weston_signal_emit_mutable.
Found by running weston under valgrind and running/killing
weston-simple-dmabuf-egl
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
This reverts commit 1618697dc3.
The original commit was a workaround for
https://gitlab.freedesktop.org/mesa/mesa/-/issues/2219 which was fixed
in Mesa:
- c7617d8908a970124321ce731b43d5996c3c5775 released as 20.1.0-rc1
- a0e6341fe4417e41cda0b19e4fa7f8bbe4e1dba1 released as 19.3.5
- f27e5d9df5bc9c85d45c2cb1f2a4997b453365fe released as 20.0.0
This workaround should not be necessary anymore, we don't use it in our
CI, and it was manual to begin with. Therefore remove it.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
The RDP spec says we can trust x, y position on all messages except
PTR_FLAGS_WHEEL and PTR_FLAGS_HWHEEL, so let's do that to ensure
proper sync with the RDP client.
Co-authored-by: Steve Pronovost <spronovo@microsoft.com>
Co-authored-by: Brenton DeGeer <brdegeer@microsoft.com>
Signed-off-by: Hideyuki Nagase <hideyukn@microsoft.com>
Signed-off-by: Steve Pronovost <spronovo@microsoft.com>
Signed-off-by: Brenton DeGeer <brdegeer@microsoft.com>
And use it to get a feedback event for when adding scanout tranche.
With this change, I get back a feedback event for dmabuf-feedback
on VC4:
���� tranche: target device /dev/dri/card0, scanout
� ���� format ABGR2101010, modifier LINEAR (0x0)
� ���� format XBGR2101010, modifier LINEAR (0x0)
� ���� format ARGB8888, modifier LINEAR (0x0)
� ���� format ABGR8888, modifier LINEAR (0x0)
� ���� format XRGB8888, modifier LINEAR (0x0)
� ���� format XBGR8888, modifier LINEAR (0x0)
� ���� format RGB565, modifier LINEAR (0x0)
� ���� format YUV420, modifier LINEAR (0x0)
� ���� format YUV422, modifier LINEAR (0x0)
� ���� format YVU420, modifier LINEAR (0x0)
� ���� format YVU422, modifier LINEAR (0x0)
� ���� format NV12, modifier LINEAR (0x0)
� ���� format NV12, modifier BROADCOM_SAND128 (0x700000000000004)
� ���� format NV16, modifier LINEAR (0x0)
� ���� end of tranche
Besides that, it can place a fullscreen state of simple-egl on the
primary plane, which without this change wasn't possible.
Signed-off-by: Marius Vlad <marius.vlad@collabora.com>
I guess this reverts commit 73bdc0ce85
"xwm: Fix fd leak in weston_wm_send_data()"
That commit closes the send half of a pipe in weston_wm_send_data,
claiming that it's dup()licated later, and we'll leak the fd if
we don't close it.
That may have been true at the time? But currently that fd is only
duplicated by wl_event_loop_add_fd() in its normal operation, and
closing our original before that fd handler ever fires results
in an EBADF on write, and the data never reaching its intended
destination.
Worse, by the time that handler is called there might be another
use for that fd, and we could push data into it and close it.
To provoke the problem, launch an app like FireFox over Xwayland,
cut something to the clipboard, then close the app (this is the
path where the wm has stored the clipboard contents and the
app has gone away). relaunch it and paste the clipboard content
back in. clipboard_client_data() will EBADF on write, and the
data won't be pasted.
Reported-by: Hideyuki Nagase <hideyukn@microsoft.com>
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
This a new matrix inversion test written from scratch to be suitable for
running in CI: quick to run and automatically detects success/failure.
This all is a result of what I learnt while working on
https://gitlab.freedesktop.org/pq/fourbyfour
Computing the residual error with infinity norm comes straight from
fourbyfour documentation on how to evaluate matrix inversion error.
Most of the hard-coded test matrices have been generated with fourbyfour
project as well, as it contains the generator code. The matrices are
hard-coded here also to make testing faster, but primarily because the
generator code needs BLAS and LAPACK, and having those as Weston
dependencies would be far too much just for this.
Now, if someone wants to modify weston_matrix stuff, we should at least
detect matrix inversion and multiplication bugs.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
This #define was used only by the matrix-test program, which was removed
in the previous commit.
Remove it as unused and fold away MATRIX_TEST_EXPORT.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
This test program was useful a decade ago when weston_matrix_invert()
was being developed. It was a manual test program that ran for a certain
number of seconds and required human interpretation of numbers to see if
results were acceptable or not. Hence it was foundamentally unsuitable
for CI.
The way it generated random matrices for inversion testing was also very
naive, and it used the determinant value to determine invertability
which is completely bogus. This made it also a bad test for correctness.
Much better speed and correctness testing is implemented in
https://gitlab.freedesktop.org/pq/fourbyfour
with documented testing procedures. It has a copy of the weston_matrix
implementation.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
Buffer scale is common enough in the modern desktop space to
expect average GL clients to handle it. Thus lets include it into
our main example client.
While on it, also handle buffer transforms. It's essentially free
for GL clients in terms of computing power but may increase the
chance that Wayland compositors are able to hit scanout fast paths.
Thus having an example client for it is likely valueabel for client
and compositor developers.
Signed-off-by: Robert Mader <robert.mader@collabora.com>
Setting the opaque region correctly is common source of error for
clients that simply want to express that a whole surface is opaque.
This is especially true once buffer_scale and buffer_transform come
into play, as unlike for damage, where buffer_damage is the
encouraged and user friendly way today, opaque regions are always
in logical coordinates.
As faulty opaque regions don't have a visual impact in these cases
but only increase resource consumption, these errors often remain
for long times. See
1e2bc68171
for one of many examples.
Give an easy example how to set the opaque region in a conformant
and reliable way.
Signed-off-by: Robert Mader <robert.mader@collabora.com>
Obviously the first allocation is always leaked, there is a second
zalloc() right below. Fix the leak.
Found by code inspection.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
As we could have situations where dmabuf import failed when attempting
to figure it the framebuffer is scanout-capable, make sure we also have
a way to store that information. Otherwise, we could end up
NULL-dereferencing, as we don't provide a valid storage for it.
Further more, with this, we also print out the reason why it failed, to
aid in further debugging.
Observed on platforms where GBM_BO_HANDLE failed + in combination w/
direct-display proto extension.
Signed-off-by: Marius Vlad <marius.vlad@collabora.com>
Output color profile may be changed in flight. Output basic color
characteristics and EOTF mode cannot yet be changed in flight, but it is
reasonable to assume they could in the future. Therefore the color
outcome data may change in flight as well, which is the basis for HDR
metadata, which needs to be updated as well.
Track the changes to color outcome data with a serial number.
DRM-backend checks the serial number to see if it needs to re-create the
HDR metadata blob. This allows the changes to propagate all the way to
KMS.
The code added here is more of a reminder of what should happen than a
tested path.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
Forward the HDR Static Metadata Type 1 to the video sink. This makes the
sink aware of our video content parameters and may be able to produce a
better picture. This type of metadata is used only with the ST 2084 HDR
mode a.k.a PQ.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
This creates a new file for KMS related color code, to avoid making
drm.c even longer.
The moved code was just added in 5151f9fe9e
so the new file copyrights are written based on that.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
'color_characteristics_config_error' test ensures that all code paths in
parse_color_characteristics() and wet_output_set_color_characteristics()
get exercised. The return value and logged error messages are checked.
Other cases test the weston_hdr_metadata_type1 validation.
These are for the sake of test coverage, but also an example of how to
test a function from main.c, and how to capture messages from
weston_log().
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
Check that weston.ini settings to eotf-mode and basic color
characteristics are correctly parsed.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
This is the beginnings of creating composited content HDR metadata for
the ST2084 HDR mode. The immediate goal is to allow essentially setting
the HDR metadata from weston.ini, so that it can be experimented with.
Setting an output ICC profile will stop weston.ini metadata from taking
effect, but using an ICC profile in HDR mode is an open question anyway.
maxDML, maxCLL, and minDML are set based on the assumption that we want
to make use of the full sink/monitor dynamic range.
This also adds several TODOs about how we should handle output profiles,
basic output color characteristics, and HDR metadata. Implementing these
properly will take more thought and effort.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
This adds hdr_meta field in weston_output_color_outcome. This field is
intended to be set by color manager modules, and read by backends which
will send the information to the video sink in SMPTE ST 2084 mode a.k.a
Perceptual Quantizer HDR system.
Such metadata is essential in ST 2084 mode for the video sink to produce
a good picture.
The validation of the data and the group split is based on the HDR
Static Metata Type 1 definition in CTA-861-G specification.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
This adds an option to program basic display color characteristics from
weston.ini. In the future there will be a way to set this information
from EDID, but because EDID is unreliable that will probably not be the
default. An ICC profile will likely override most or all of this. The
main reason to add this option is to be able to characterise HDR
monitors.
An 'output' section can have a key 'color_characteristics' (string)
set to a name. The name refers to any 'color_characteristics' section
with 'name' set to the same string.
The 'name' key of a 'color_characteristics' section cannot contain a
colon ':'. Names with colon in 'output' section key
'color_characteristics' value are reserved for future use, e.g. to
indicate that the metadata is to be taken from EDID instead of a
weston.ini section.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
This adds color_chracteristics field in weston_output. This field is
intended to be set by compositor frontends and read by color managers.
Color managers can use this information when choosing the output color
space and dynamic range, particularly when no ICC profile has been set.
This is most useful for HDR outputs, where the HDR static metadata for
PQ mode or the display luminance parameters for HLG mode can be based on
color_characteristics.
The fields of weston_color_characteristics mirror the information
available in EDID. However, using EDID information as-is has several
caveats, so the decision to use EDID for this is left for the frontend
and ultimately to the end user.
There are no defined ranges or validity checks for this data. The color
manager will have to validate the values against whatever it is using
them for.
Signed-off-by: Pekka Paalanen <pekka.paalanen@collabora.com>
Allow the front end to register audio setup and teardown functions. These
functions should use FreeRDP's rdpsnd_server_context or
audin_server_context and set up their own handler threads.
The backend remains mostly ignorant to any audio details beyond setting up
and tearing down.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
Instead of a meson option or hidden define, just run these checks always.
It is not Weston's style to add build options for specific asserts, and
currently weston's codebase is expected to always run with asserts
enabled.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
It is used in Mesa. Lets switch to it as well in order to provide
good examples and encourage proper API usage.
Signed-off-by: Robert Mader <robert.mader@collabora.com>
Compositors may choose to send multiple scanout or non-scanout
tranches. So instead of assuming that the first respective tranche
contains the format/modifier we're looking for, check all tranches.
While on it, make sure that in case a compositor sends scanout
tranches on the initial feedback, `pick_format_from_scanout_tranche()`
does not unintentionally pick `INITIAL_BUFFER_FORMAT`.
Signed-off-by: Robert Mader <robert.mader@collabora.com>
It doesn't and can't build, because it depends on cairo-gl. We already
have simple-egl which shows how to use EGL/GLESv2 on Wayland.
Signed-off-by: Daniel Stone <daniels@collabora.com>
It's three planes, not two.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Fixes: 8b167a1703 ("gl-renderer: Store EGL buffer state in weston_buffer")
There's just no good reason to do this.
The query entrypoints already tell us if we need to use
GL_TEXTURE_EXTERNAL_OES for a particular format/modifier. We also have
RGB -> YUV fallbacks which should be able to work well with TEXTURE_2D.
TEXTURE_EXTERNAL pessimises quite hard, forcing GPU-side reloads as well
as bad filtering. Allowing multi-planar formats to use TEXTURE_2D should
thus result in performance and quality improvements.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Pekka Paalanen
Michael Olbrich
Michael Tretter
Derek Foreman
Hideyuki Nagase
Marius Vlad
Robert Mader
Daniel Stone