During animation or other things that cause multiple frames in a row
being painted, we might skip recording frames if the max framerate is
reached.
Doing so means we might end up skipping the last frame in a series,
ending with the last frame we sent was not the last one, making things
appear to get stuck sometimes.
Handle this by creating a timeout if we ever throttle, and at the time
the timeout callback is triggered, make sure we eventually send an up to
date frame.
This is handle differently depending on the source type. A monitor
source type reports 1x1 pixel damage on each view its monitor overlaps,
while a window source type simply records a frame from the surface
directly, except without recording a timestamp, so that timestamps
always refer to when damage actually happened.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1361
Now that we don't use the record function to early out depending on
implicit state (don't record pixels if only cursor moved for example),
let it simply report an error when it fails, as we should no longer ever
return without pixels if nothing failed.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1361
Both do more or less the same but with different methods - one puts
pixels into a buffer using the CPU, the other puts pixels into a buffer
using the GPU.
However, they are behaving slightly different, which they shouldn't.
Lets first address the misleading disconnect in naming, and later we'll
make them behave more similarly.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1361
That was obviously always the intention, but it didn't work when the
display was scaled. My 3840x2160 monitor with a 3840x2160 texture was
being rendered with LINEAR filtering.
It seems the `force_bilinear` flag was TRUE when it should be FALSE.
Because a texture area that's an integer fraction of the texture
resolution is still a perfect match when that integer is the monitor
scale. We were also getting:
`meta_actor_painting_untransformed (fb, W, H, W, H, NULL, NULL) == FALSE`
when the display was scaled. Because the second W,H was not the real
sampling resolution. So with both of those issues fixed we now get
NEAREST filtering when the texture resolution matches the resolution it's
physically being rendered at.
Note: The background texture actually wasn't equal to the physical monitor
resolution prior to January 2020 (76240e24f7). So it wasn't possible to do
this before then. Since then however, the texture resolution is always
equal to the physical monitor resolution.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1346
Make clutter_actor_allocate_preferred_size() convenient to use from
layout managers by not "automatically" honouring the fixed position of
the actor, but instead allowing to pass a position to allocate the
actor at.
This way we can move the handling of fixed positions to
ClutterFixedLayout, the layout manager which is responsible for
allocating actors using fixed positions.
This also makes clutter_actor_allocate_preferred_size() more similar to
clutter_actor_allocate_available_size().
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1310
It's currently a bit hard to get the fixed position of an actor. It can
be either done by using g_object_get() with the "fixed-x"/"fixed-y"
properties or by calling clutter_actor_get_position().
Calling clutter_actor_get_position() can return the fixed position, but
it might also return the allocated position if the allocation is valid.
The latter is not the best behavior when querying the fixed position
during an allocation, so introduce a new function
clutter_actor_get_fixed_position() which always gets the fixed position
and returns FALSE in case no fixed position is set.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1310
It doesn't take all children - subsurfaces in this case - into
account, thus creating glitches if subsurfaces extend outside
of the toplevel surface.
Further more it doesn't seem to serve any special purpose - it was
added in f7315c9a36, a pretty big commit, and no discussion was
started about the code in question. So it was likely just overlooked
in the review process.
Closes https://gitlab.gnome.org/GNOME/mutter/-/issues/873
Closes https://gitlab.gnome.org/GNOME/mutter/-/issues/1316
gnome-shell displays workspace previews at one tenth scale. That's a
few binary orders of magnitude so even using a LINEAR filter was
resulting in visible jaggies. Now we apply mipmapping so they appear
smooth.
As an added bonus, the mipmaps used occupy roughly 1% the memory of
the original image (0.1 x 0.1 = 0.01) so they actually fit into GPU/CPU
caches now and rendering performance is improved. There's no need to
traverse the original texture which at 4K resolution occupies 33MB,
only a 331KB mipmap.
In my case this reduces the render time for the overview by ~10%.
Closes: https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/1416https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1347
In the case of indirect rendering like the first frame to use mutter's
background wallpaper:
Texture_A -> FBO_B (Texture_B) -> FBO_C (screen)
we would be trying to render the contents of both FBO_B and FBO_C in
the same flush, before the contents of Texture_A had made it to FBO_B.
So when FBO_C wants to use mipmaps of Texture_B they didn't exist yet
and appeared all black. And the blackness would remain for subsequent
frames as cogl has now decided the mipmaps of FBO_B are no longer
"dirty" and don't need refreshing:
FBO_B (Texture_B) (mipmaps_dirty==FALSE but black) -> FBO_C (screen)
We must flush FBO_B before referencing Texture_B for use in rendering
FBO_C. This only happens when Texture_A changes (e.g. when the user
changes their background wallpaper) so there's no ongoing performance
penalty from this flush.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1347
With the introduction of the shallow relayout mechanism another small
but severe regression sneaked into our layout machinery: We might
allocate an actor twice during the same allocation cycle, with one
allocation happening using the wrong parent.
This issue happens when reparenting an actor from a NO_LAYOUT parent to
a non-NO_LAYOUT parent, in particular it triggered a bug in gnome-shell
when DND reparents a child from the NO_LAYOUT uiGroup to the overviews
Workspace actor after a drag ended. The reason the issue happens is the
following chain of events:
1. child of a NO_LAYOUT parent queues a relayout, this child is added to
the priv->pending_relayouts list maintained by ClutterStage
2. child is reparented to a different parent which doesn't have the
NO_LAYOUT flag set, another relayout is queued, this time a different
actor is added to the priv->pending_relayouts list
3. the relayout happens and we go through the pending_relayouts list
backwards, that means the correct relayout queued during 2. happens
first, then the old one happens and we simply call
clutter_actor_allocate_preferred_size() on the actor, that allocation
overrides the other, correct one.
So fix that issue by adding a method to ClutterStage which removes
actors from the pending_relayouts list again and call this method as
soon as an actor with a NO_LAYOUT parent is detached from the stage.
With that in place, we can also remove the check whether an actor is
still on stage while looping through pending_relayouts. In case
something else is going wrong and the actor is not on stage,
clutter_actor_allocate() will warn anyway.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1356
When picking which frame clock to use, we traverse up in the actor
hierarchy until a suitable frame clock is found. ClutterTimeline
also listens to the 'stage-views-changed' to make sure it's always
attached to the correct frame clock.
However, there is one special situation where neither of them would
work: when the stage doesn't have a frame clock yet, and the actor
of the timeline is outside any stage view. When that happens, the
returned frame clock is NULL, and 'stage-views-changed' is never
emitted by the actor.
Monitor the stage for stage view changes when the frame clock is
NULL.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1285
An actor may be placed without being on any current stage view; in this
case, to get the ball rolling, walk up the actor tree to find the first
actor where a frame clock can be picked from.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1285
The frame clock owner should be able to explicitly destroy (i.e. make
defunct) a frame clock, e.g. when a stage view is destructed. This is so
that other objects can keep reference to its without it being left
around even after stopped being usable.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1285
Currently there is a point in between hot plug, and when the stage view
list is up to date. The check also tests for this behaviour; would this
ever change, the test should be adapted to deal with this too.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1285
Replace the default master clock with multiple frame clocks, each
driving its own stage view. As each stage view represents one CRTC, this
means we draw each CRTC with its own designated frame clock,
disconnected from all the others.
For example this means we when using the native backend will never need
to wait for one monitor to vsync before painting another, so e.g. having
a 144 Hz monitor next to a 60 Hz monitor, things including both Wayland
and X11 applications and shell UI will be able to render at the
corresponding monitor refresh rate.
This also changes a warning about missed frames when sending
_NETWM_FRAME_TIMINGS messages to a debug log entry, as it's expected
that we'll start missing frames e.g. when a X11 window (via Xwayland) is
exclusively within a stage view that was not painted, while another one
was, still increasing the global frame clock.
Addititonally, this also requires the X11 window actor to schedule
timeouts for _NET_WM_FRAME_DRAWN/_NET_WM_FRAME_TIMINGS event emitting,
if the actor wasn't on any stage views, as now we'll only get the frame
callbacks on actors when they actually were painted, while in the past,
we'd invoke that vfunc when anything was painted.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/903
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/3https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1285
This also changes the view construction path used by the renderer view
to use the new 'add_view()' function, meaning we have a common entry
point for views into the renderer, which will be useful later on.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1285
Before we'd create the view in init(), then continue poking at it in
realize(). Move all of the screen stage view initialization to
realize(), as that's when we have all the dependent state available.
This is possible since there is nothing needing it until realizing.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1285
The repaint callbacks are not tied to repaint, thus a bit misleading.
What the functionality in the pre/post-paint callbacks here cares about
is when actually painting; the non-painting related parts has already
moved out to a *-update signal.
This also renames the related MetaWindowActorClass vfuncs, to align with
naming convention of the signals that it listens to.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1285
Instead of going via MetaCompositor to know about when we updated
(confusingly named post-paint), use the new stage signal directly.
Note that this doesn't change the time frame callbacks are dispatched;
it's still not tied to actual painting even though it seemed so before
given the function names.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1285
The clutter "thread" repaint callback are not tied to painting, but
indirectly to updating. What the cursor renderer cares about is when we
actually painted, as this is related to the OpenGL fallback paths.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1285
We'd emit multiple "presented" signals per frame, one for "sync" and one
for "completion". Only the latter were ever used, and removing the
differentiation eases the avoidance of cogl onscreen framebuffer frame
callback details leaking into clutter.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1285
The vfunc was not tied to "paint", but was used by MetaWindowActorX11
as part of the "update" mechanisms. In order to make that more clear,
special case it in MetaWindowActorX11 by type checking the surface
actor, handling the case without MetaSurfacActor abstraction.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1285
The synchronization must happen no matter the painting, as it in itself
might result in reported damage, making the stage actually painted. Thus
move it out of the "pre-paint" handler, to something explicitly not tied
to the painting itself - ClutterStage::before-update.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1285
Instead of the 'pre-paint' signal on MetaCompositor, rely directly on
the 'before-update' signal on the stage. A reason for this is that the
callback should not only invoked in connection to painting, but updating
in general. Currently the 'pre-paint' signal is emitted no matter
whether there were any painting or not, but that's both misleading and
will go away.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1285
Right now the stage only had a signal called 'after-paint' which was not
tied to painting but updating. Change this to offer 4 signals, for the 4
different stages:
* before-update - emitted in the beginning before the actual stage
updating
* before-paint - emitted before painting if there will be any stage
painting
* after-paint - emitted after painting if there was any stage painting
* after-update - emitted as a last step of updating, no matter whether
there were any painting or not
Currently there were only one listener, that should only really have
been called if there was any painting, so no changes to listeners are
needed.
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1285
