At the moment, our frame scheduling infrastructure is still heavily
based on Xinerama-style rendering. Specifically, we assume that painting
is driven by a single timer, etc.
This change introduces a new type - RenderLoop. Its main purpose is to
drive compositing on a specific output, or in case of X11, on the
overlay window.
With RenderLoop, compositing is synchronized to vblank events. It
exposes the last and the next estimated presentation timestamp. The
expected presentation timestamp can be used by effects to ensure that
animations are synchronized with the upcoming vblank event.
On Wayland, every outputs has its own render loop. On X11, per screen
rendering is not possible, therefore the platform exposes the render
loop for the overlay window. Ideally, the Scene has to expose the
RenderLoop, but as the first step towards better compositing scheduling
it's good as is for the time being.
The RenderLoop tries to minimize the latency by delaying compositing as
close as possible to the next vblank event. One tricky thing about it is
that if compositing is too close to the next vblank event, animations
may become a little bit choppy. However, increasing the latency reduces
the choppiness.
Given that, there is no any "silver bullet" solution for the choppiness
issue, a new option has been added in the Compositing KCM to specify the
amount of latency. By default, it's "Medium," but if a user is not
satisfied with the upstream default, they can tweak it.
With the new compositing timing, we want to start compositing some time
later after a vsync event. This doesn't go along with the video sync
based method to synchronize buffer swaps with vblank.
Since practically all drivers nowadays provide support for the swap
control extensions (GLX_EXT_swap_control, GLX_SGI_swap_control, or
GLX_MESA_swap_control), it's safe to rely on them for the purpose of
synchronizing buffer swaps to vblank.
The compositing timing algorithm assumes that glXSwapBuffers() and
eglSwapBuffers() block. While this was true long time ago with NVIDIA
drivers, nowadays, it's not the case. The NVIDIA driver queues
several buffers in advance and if the application runs out of them,
it will block. With Mesa driver, swapping buffer was never blocking.
This change makes the render backends swap buffers right after ending
a compositing cycle. This may potentially block, but it shouldn't be
an issue with modern drivers. In case it gets proven, we can move
glXSwapBuffers() and eglSwapBuffers() in a separate thread.
Note that this change breaks the compositing timing algorithm, but
it's already sort of broken with Mesa drivers.
Currently, the OpenGLBackend and the QPainterBackend have hooks to
indicate the start and the end of compositing cycle, but in both cases,
the hooks have different names. This change fixes that inconsistency.
In order to allow per screen rendering, we need the Compositor to be
able to drive rendering on each screen. Currently, it's not possible
because Scene::paint() paints all screen.
With this change, the Compositor will be able to ask the Scene to paint
only a screen with the specific id.
The main advantage of SPDX license identifiers over the traditional
license headers is that it's more difficult to overlook inappropriate
licenses for kwin, for example GPL 3. We also don't have to copy a
lot of boilerplate text.
In order to create this change, I ran licensedigger -r -c from the
toplevel source directory.
Summary: No need to keep them around for no reason.
Test Plan: Tested the plugins I thought could be affected. Have been using it for a couple of days without problems
Reviewers: #kwin, zzag
Reviewed By: #kwin, zzag
Subscribers: zzag, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D28062
Summary:
The GLX backend might need a combination of swap and composite timer events for
continous painting.
The reason for that is that if the buffer age extension is not available we
fall back to copies in case not the whole screen is repainted.
The timer logic is adapted to make this possible in a lean way what cleans up
the Compositor class in several ways.
Test Plan: Tested on X11 (with/without swap events, buffer age enabled) and Wayland.
Reviewers: #kwin
Subscribers: hurikhan77, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D26216
Summary:
Activate intel swap events by default if available. They were hidden behind an
environmental variable because of some critical release blocking issue years
ago.
Manual testing indicates that there are no issues anymore with this extension.
Since it allows us to use swap events with MESA drivers for optimized repaints
enable swap events by default again.
For now leave a modified environment variable to switch back to using no swap
events easily.
CCBUG: 342582
Test Plan: i915
Reviewers: #kwin
Subscribers: zzag, broulik, kwin
Tags: #kwin
Maniphest Tasks: T11071
Differential Revision: https://phabricator.kde.org/D25300
Summary:
Add a small getter to query information internally if the backend supports
swap events. Defaults to true as it is the default in the GBM Wayland backend.
Test Plan: i915
Reviewers: #kwin
Subscribers: kwin
Tags: #kwin
Maniphest Tasks: T11071
Differential Revision: https://phabricator.kde.org/D25298
Summary:
KDE is known for having a strong view on the client-side decorations vs
server-side decorations issue. The main argument raised against CSD is
that desktop will look less consistent when clients start drawing window
decorations by themselves, which is somewhat true. It all ties to how
well each toolkit is integrated with the desktop environment.
KDE doesn't control the desktop market on Linux. Another big "player"
is GNOME. Both KDE and GNOME have very polarized views on in which
direction desktop should move forward. The KDE community is pushing more
toward server-side decorations while the GNOME community is pushing
more toward client-side decorations. Both communities have developed
great applications and it's not rare to see a GNOME application being
used in KDE Plasma. The only problem is that these different views are
not left behind the curtain and our users pay the price. Resizing GTK
clients in Plasma became practically impossible due to resize borders
having small hit area.
When a client draws its window decoration, it's more likely that it also
draws the drop-shadow around the decoration. The compositor must know
the extents of the shadow so things like snapping and so on work as
expected. And here lies the problem... While the xdg-shell protocol has
a way to specify such things, the NetWM spec doesn't have anything like
that. There's _GTK_FRAME_EXTENTS in the wild, however the problem with
it is that it's a proprietary atom, which is specific only to GTK apps.
Due to that, _GTK_FRAME_EXTENTS wasn't implemented because implementing
anything like that would require major changes in how we think about
geometry.
Recent xdg-shell window geometry patches adjusted geometry abstractions
in kwin to such a degree that it's very easy to add support for client
side decorated clients on X11. We just have to make sure that the
X11Client class provides correct buffer geometry and frame geometry when
the gtk frame extents are set.
Even though the X11 code is feature frozen, I still think it's worth
to have _GTK_FRAME_EXTENTS support in kwin because it will fix the resize
issues. Also, because KWin/Wayland is unfortunately far from becoming
default, it will help us with testing some implementation bits of the
window geometry from xdg-shell.
BUG: 390550
FIXED-IN: 5.18.0
Test Plan:
Things like quick tiling, maximizing, tiling scripts and so on work as
expected with GTK clients.
Reviewers: #kwin, davidedmundson
Reviewed By: #kwin, davidedmundson
Subscribers: cblack, trmdi, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D24660
Summary:
Compositing in X11 was done time shifted, meaning that we paint first, then
wait one vblank interval length and present on prepareRenderingFrame the
previous paint result. This is supposed to make sure we don't miss the vblank
and in case of block till retrace be able to continue issuing commands and
only shortly before next vblank present.
This is counter-intuitiv, not how we do it on Wayland or even on MESA with X.
The reason seems to be that the GLX backend was in the beginning written
against Nvidia proprietary driver which needed this but nowadays even this
driver defaults to non-blocking behavior on buffer swap.
Therefore remove this legacy anomaly fully and directly present after paint.
We then wait one refresh cycle and in the future can optimize this by delaying
the paint and present till shortly before vsync.
Test Plan: kwin_x11 tested on i915 and Nvidia proprietary driver.
Reviewers: #kwin
Subscribers: zzag, alexeymin, kwin
Tags: #kwin
Maniphest Tasks: T11071
Differential Revision: https://phabricator.kde.org/D23514
Summary:
Selecting not to vsync does not make sense for an X11 compositor. In the end
we want clients to be able to present async if they want to but the compositor
is supposed to send swaps with vsync to the XServer in order to not generate
tearing artifacts.
There was also a detection logic which did some questionable things in case
vsync was not available. I don't think this is necessary at all since we can
just always run a timer to present with or without vsync.
Test Plan: kwin_x11 tested on i915.
Reviewers: #kwin, zzag
Subscribers: zzag, kwin
Tags: #kwin
Maniphest Tasks: T11071
Differential Revision: https://phabricator.kde.org/D23511
Summary:
Current hardware should suppport either GLX_EXT_swap_control or
GLX_MESA_swap_control. To simplify code remove the usage of SGI extensions.
Test Plan: kwin_x11 tested on i915.
Reviewers: #kwin, zzag
Reviewed By: #kwin, zzag
Subscribers: davidedmundson, kwin
Tags: #kwin
Maniphest Tasks: T11071
Differential Revision: https://phabricator.kde.org/D23510
Summary:
It is not clear what the advantage of triple buffering is for KWin. An X11
compositor is meant to swap buffers once every monitor cycle. For that triple
buffering is not necessary.
The functionality is not maintained, does not reliably work as displayed by
the existence of an environment variable to force some behavior, pollutes
our code and every compositing-related problem that might be mitigated with
triple buffering should find a simpler and more fitting solution with other
means.
There is one caveat which is if we shall block for retrace. We set it
currently according to the result of the swap profiler and in the most common
case with double buffering it is set to true. But on Nvidia systems this might
be actual the wrong behavior. Instead of trying to work around this ignore
the issue for now and move the overall architecture to something less complex
by presenting after paint how we do it in the Wayland DRM backend and with
double buffering on GLX (although this is at the moment also borken because
we actually present then twice).
Test Plan: kwin_x11 tested on i915.
Reviewers: #kwin, zzag
Reviewed By: #kwin, zzag
Subscribers: zzag, fredrik, kwin
Tags: #kwin
Maniphest Tasks: T11071
Differential Revision: https://phabricator.kde.org/D23504
Summary:
This is necessary to keep openQA working, which uses LLVMpipe as a
renderer on a Cirrus device that operates in depth 16.
LLVMpipe advertises 24/32 bit sRGB configurations on this setup, but
they cannot be presented.
CCBUG: 408594
Test Plan: Compile tested only.
Reviewers: fvogt, #kwin, zzag
Reviewed By: fvogt, #kwin, zzag
Subscribers: romangg, sbergeron, fvogt, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D22203
Summary:
EffectQuickView/Scene is a convenient class to render a QtQuick
scenegraph into an effect.
Current methods (such as present windows) involve creating an underlying
platform window which is expensive, causes a headache to filter out
again in the rest of the code, and only works as an overlay.
The new class exposes things more natively to an effect where we don't
mess with real windows, we can perform the painting anywhere in the view
and we don't have issues with hiding/closing.
QtQuick has both software and hardware accelerated modes, and kwin also
has 3 render backends. Every combination is supported.
* When used in OpenGL mode for both, we render into an FBO export the
texture ID then it's up to the effect to render that into a scene.
* When using software QtQuick rendering we blit into an image, upload
that into a KWinGLTexture which serves as an abstraction layer and
render that into the scene.
* When using GL for QtQuick and XRender/QPainter in kwin everything is
rendered into the internal FBO, blit and exported as an image.
* When using software rendering for both an image gets passed directly.
Mouse and keyboard events can be forwarded, only if the effect
intercepts them.
The class is meant to be generic enough that we can remove all the
QtQuick code from Aurorae.
The intention is also to replace EffectFrameImpl using this backend and
we can kill all of the EffectFrame code throughout the scenes.
The close button in present windows will also be ported to this,
simplifiying that code base.
Classes that handle the rendering and handling QML are intentionally
split so that in the future we can have a declarative effects API create
overlays from within the same context. Similar to how one can
instantiate windows from a typical QML scene.
Notes:
I don't like how I pass the kwin GL context from the backends into the
effect, but I need something that works with the library separation. It
also currently has wayland problem if I create a QOpenGLContext before
the QPA is set up with a scene - but I don't have anything better?
I know for the EffectFrame we need an API to push things through the
effects stack to handle blur/invert etc. Will deal with that when we
port the EffectFrame.
Test Plan: Used in an effect
Reviewers: #kwin, zzag
Reviewed By: #kwin, zzag
Subscribers: zzag, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D24215
Summary:
Because KWin is a very old project, we use three kinds of null pointer
literals: 0, NULL, and nullptr. Since C++11, it's recommended to use
nullptr keyword.
This change converts all usages of 0 and NULL literal to nullptr. Even
though it breaks git history, we need to do it in order to have consistent
code as well to ease code reviews (it's very tempting for some people to
add unrelated changes to their patches, e.g. converting NULL to nullptr).
Test Plan: Compiles.
Reviewers: #kwin, davidedmundson, romangg
Reviewed By: #kwin, davidedmundson, romangg
Subscribers: romangg, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D23618
Summary:
Switch to Q_ASSERT in order to make code a bit more consistent. We have
places where both assert and Q_ASSERT are used next to each other. Also,
distributions like Ubuntu don't strip away assert(), let's hope that
things are a bit different with Q_ASSERT.
Test Plan: Compiles.
Reviewers: #kwin, davidedmundson
Reviewed By: #kwin, davidedmundson
Subscribers: romangg, davidedmundson, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D23605
Summary:
This patch is a first take at splitting up of the Compositor class into
Wayland and X11 child classes.
In this first patch we mostly deal with setup and teardown procedures.
A future goal is to further differentiate the compositing part itself too.
Test Plan: Manually X from VT and Wayland nested. Autotests pass.
Reviewers: #kwin
Subscribers: sbergeron, anthonyfieroni, zzag, kwin
Tags: #kwin
Maniphest Tasks: T11071
Differential Revision: https://phabricator.kde.org/D22195
Summary:
QRegion::rects was deprecated in Qt 5.11. It is advised to use begin()
and end() methods instead.
Reviewers: #kwin, romangg
Reviewed By: #kwin, romangg
Subscribers: kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D22353
Summary:
The NVIDIA implementation of glXSwapBuffers will, by default, queue up
to two frames for presentation before blocking. KWin's compositor,
however, assumes that calls to glXSwapBuffers will always block until
the next vblank when rendering double buffered. This assumption isn't
valid, as glXSwapBuffers is specified as being an implicit glFlush,
not an implicit glFinish, and so it isn't required to block. When this
assumption is violated, KWin's frame timing logic will
break. Specifically, there will be extraneous calls to
setCompositeTimer with a waitTime of 0 after the non-blocking buffer
swaps, dramatically reducing desktop responsiveness. To remedy this,
a call to glXWaitGL was added by Thomas Luebking after glXSwapBuffers
in 2015 (see bug 346275, commit
8bea96d701). That glXWaitGL call is
equivalent to a glFinish call in direct rendering, so it was a good
way to make glXSwapBuffers behave as though it implied a glFinish
call.
However, the NVIDIA driver will by default do a busy wait in glFinish,
for reduced latency. Therefore that change dramatically increased CPU
usage. GL_YIELD can be set to USLEEP (case insensitive) to change
the behavior and use usleep instead. When using the NVIDIA driver,
KWin will disable vsync entirely if GL_YIELD isn't set to USLEEP
(case sensitive, a bug in KWin).
However, the NVIDIA driver supports another environment variable,
__GL_MaxFramesAllowed, which can be used to control how many frames
may be queued by glXSwapBuffers. If this is set to 1 the function
will always block until retrace, in line with KWin's expectations.
This allows the now-unnecessary call to glXWaitGL to be removed along
with the logic to conditionally disable vsync, providing a better
experience on NVIDIA hardware.
Reviewers: #kwin, davidedmundson, zzag
Reviewed By: #kwin, davidedmundson, zzag
Subscribers: kwin, davidedmundson, zzag
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D19867
Summary:
Mesa requires XESetWireToEvent xlib callbacks to be called
when DRI2 is used. This is done by the GLX integration in
the Qt's xcb plugin, but Qt 5.12 initializes the GLX integration
only when required, e.g. when a window with OpenGL support is
created or when availability of OpenGL is checked.
So force initialization of the GLX integration by calling
QOpenGLContext::supportsThreadedOpenGL().
https://codereview.qt-project.org/#/c/6557/https://bugzilla.opensuse.org/show_bug.cgi?id=1120090
Reviewers: #kwin, graesslin
Reviewed By: #kwin, graesslin
Subscribers: davidedmundson, graesslin, fvogt, filipf, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D18366
Mesa's glXChooseFBConfig will not return any sRGB capable fbconfig when it
is not explicitly asked for. On some systems, the only ARGB32 visual is
paired with an sRGB capable fbconfig, so application windows using ARGB32
visuals would fail to display.
BUG: 387159
FIXED-IN: 5.11.4
Summary:
Unfortunately a rather large change which required more refactoring than
initially expected. The main problem was that some parts needed to go
into platformsupport so that the platform plugins can link them. Due to
the rather monolithic nature of scene_opengl.h a few changes were
required:
* SceneOpenGL::Texture -> SceneOpenGLTexture
* SceneOpenGL::TexturePrivate -> SceneOpenGLTexturePrivate
* texture based code into dedicated files
* SwapProfiler code into dedicated files
* SwapProfiler only used in x11 variants
* Safety checks for OpenGL scene moved into the new plugin
* signal declared in SceneOpenGL moved to Scene, so that we don't need
to include SceneOpenGL in composite
Test Plan: Nested OpenGL compositor works
Reviewers: #kwin, #plasma
Subscribers: plasma-devel, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D7740
Summary:
Based on the work of 3f4995fb9b this change
introduces a GlxContextAttributeBuilder to make the requesting of context
attributes cleaner, more verbose and less error prone copy and paste.
Test Plan:
Switched between Core and legacy and verified the output;
extended auto test
Reviewers: #kwin, #plasma
Subscribers: plasma-devel
Tags: #plasma
Differential Revision: https://phabricator.kde.org/D6411
Summary:
The overlay window is only needed for the X11 based compositors. Given
that it is better suited in the X11 platform. Unfortunately it is not
possible to completely move it into the platform plugin as it is still
referenced in KWin core (e.g. SceneXRender). Due to that the
OverlayWindow in KWin core is turned into a pure virtual class with the
implementation being moved into the plugin.
The platform API gains a new virtual factory method which is only
implemented in the X11 platform.
Test Plan: Compiles
Reviewers: #kwin, #plasma
Subscribers: plasma-devel, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D7193
Summary:
KWin already used C++14 constructs in a conditional way. This doesn't
make much sense today, it's better to just require C++14.
For KWin only gcc and clang are currently compilers of relevance. Gcc
supports C++14 since version 5 and defaults to C++14 since 6.1 [1].
Clang supports C++14 since version 3.4 [2].
An overview of compiler support in various distributions:
* Debian stable (stretch): gcc 6.3, clang 3.8
* Debian oldstable (jessie): 4.9, clang 3.5
* Ubuntu 17.04: gcc 6.1, clang 3.8
* Ubuntu 16.04: gcc 5.3, clang 3.8
* openSUSE Tumbleweed: gcc 7.1, clang 4.0
* openSUSE Leap 42.3: gcc ?, clang ? [3]
* FreeBSD: clang >= 34 in ports
* Slackware 14.2: gcc 5.3
This overview shows that every distro out there has at least one
supported compiler which can still compile KWin with this change.
[1] https://gcc.gnu.org/projects/cxx-status.html#cxx14
[2] https://clang.llvm.org/cxx_status
[3] Sorry I fail to understand openSUSE's package repository.
It seems that there is gcc 7 available, but gcc package is 4.8
Test Plan: Compiles on my neon system
Reviewers: #plasma
Subscribers: plasma-devel, kwin, #kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D6634
Summary:
Mismatch in the major/minor version. Requesting 1.2 doesn't make any
sense given that KWin requires 2.1.
Reviewers: #kwin, #plasma
Subscribers: plasma-devel, kwin
Tags: #kwin
Differential Revision: https://phabricator.kde.org/D6401
