Now that dynamic max render time uses a new algorithm and takes dispatch
lateness into account, this seems worth a shot. We'll see how it works
out in the wild.
The net result compared to before these changes is still slightly higher
(by ~0.5 ms) minimum latency for me, as measured by
weston-presentation-shm. It should be less vulnerable to frame drops
though.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2500>
Store only two values per kind of duration: The short term and long term
maximum.
The short term maximum is updated in each frame clock dispatch. The long
term maximum is updated at most once per second: If the short term
maximum is higher, the long term maximum is updated to match it.
Otherwise, a fraction of the delta between the two maxima is subtracted
from the long term maximum.
Compared to the previous algorithm:
* The calculcations are simpler.
* The calculated max render time has a slow exponential drop-off (by at
most a few milliseconds every second) instead of potentially abruptly
dropping after as few as 16 frames.
This should fix https://gitlab.gnome.org/GNOME/gnome-shell/-/issues/4830
since the short term maximum should always include a sample from the
clock's second tick.
v2:
* Use divisor 2 instead of 4.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2500>
Dispatch lateness is the difference between when we wanted frame clock
dispatch to run and when it actually started running. This can be up to
1ms even under normal circumstances due to process scheduling
granularity, or even higher under load.
This keeps track of dispatch lateness of the last 16 frame clock
dispatches, and incorporates the maximum into the dynamic render time
estimate.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2500>
There are two tests; one checks that clearing with a color that cannot
be represented using 8 bits per channel doesn't loose precision when
painted, then read back using glReadPixels(). Would the texture backing
store have 8 bits per channel instead of 10, we'd get a different value
back.
The other test checks that painting from one fbo to another also doesn't
loose that precision.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2461>
OpenGL requires more hand holding in the driver regarding what pixel
memory layouts can be written when calling glReadPixels(), compared to
GLES2. Lets move the details of this logic to the corresponding
backends, so in the future, the GLES2 backend can be adapted to handle
more formats, without placing that logic in the generic layer.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2461>
COGL_PIXEL_FORMAT_ABGR_2101010 is defined to mean the 2 A bits are
placed in a 32 bit unsigned integer on the bits with highest
significance, followed by B on the following 10 bits, and so on, until R
on the 10 least significant bits.
UNSIGNED_INT_2_10_10_10_REV_EXT is defined to represent color channels
as
```
31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0
-------------------------------------------------------------------------------------
| a | b | g | r |
-------------------------------------------------------------------------------------
```
As can be seen, this matches COGL_PIXEL_FORMAT_ABGR_2101010, meaning
that's the format we can directly read and write.
In Cogl, when finding the GL formats, we get the tuple with the GL
format given the format we pass, but we also get returned "required
format" (CoglPixelFormat). This required format represents the format
that is required when reading actual pixels from GLES. In GLES, the
above mentioned format is the only one supported by the
EXT_texture_type_2_10_10_10_REV extension, thus for other types, we need
to do the CPU side conversion ourselves. To achieve this, correctly
return COGL_PIXEL_FORMAT_ABGR_2101010 as the required format.
The internal format should also be GL_RGB10_A2, not GL_RGBA.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2461>
Commit bf84b24 created meta-enums.h but it's pretty empty so far, the
vast majority of enum definitions is still in common.h.
Move the Meta enum definitions to meta-enums.h as one would expect them
to be found.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2467>
This replaces the v1 implementation, which is now renamed to
legacy-xdg-foreign. Both implementations use the same data structures
internally, so that protocol version mismatches between
the importer client and exporter client don't fail.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2770>
Sysprof's build options have changed recently. This both bumps the sysprof
version and updates the configuration options for the subproject.
If now is not a good time to bump this, that is totally fine, but I wanted
to give you a MR with the necessary changes all in one.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2428>
Things like meta_compositor_destroy() and meta_compositor_add_window()
isn't intended to be used externally, and if they was, things would
probably fall apart rather quickly.
MetaCompositor also isn't introspected, meaning things that technically
belong to the compositing parts isn't easily available via some object,
but much take detours via other objects like MetaDisplay.
So move the API intended for internal usage to compositor-private.h, and
leave API that is meant to be expose in the public compositor.h.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2718>
The "later" API is used to queue actions in relation to compositing,
thus is owned by the MetaCompositor instance. Make users of this
functionality get MetaLaters instance from the compositor, and stop
using the global meta_later() API.
display: Use non-singleton MetaLater API
tests: Use non-singleton MetaLater API
meta/common: Make docs refer to context aware MetaLater API
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2718>
This means we can eliminate the use of scattered singletons that isn't
added by the tests or the test framework itself.
tests: Don't get backend from old singleton getter
Either use the ownership chain, or the explicit test context instance
pointer.
tests/wayland: Pass context to test client constructor
So that we can get the Wayland compositor directly from the context.
tests: Don't get display from singleton
tests/client: Make test client carry a context pointer
tests/runner: Have test cases carry a context pointer
tests/wayland/test-driver: Get backend from context
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2718>
Instead of passing around state using GINT_TO_POINTER() pass around a
state struct that also carries a pointer to the context. This allows
avoiding using old singletons for getting a window list.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2718>
On the path towards clear ownership chains and always using them to find
other components, do the same for X11 client support paths too.
x11-display: Don't get backend from signleton
x11/selection: Don't get display from singleton
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2718>
The API has no concept of user data, and requires the user to some how
get an instance without context, i.e. via static globals. Limit this to
the file where this is needed.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2718>
As with the backend commit, this means all objects can reach the
MetaContext by walking up the chain, thus can e.g. get the backend from
the context, instead of the global singleton.
This also is a squashed commit containing:
compositor: Get backend via the context
The MetaCompositor instance is owned by MetaDisplay, which is owned by
MetaContext. Get the backend via that chain of ownership.
dnd: Don't get backend from singleton
window-actor: Don't get backend from singleton
dnd: Don't get Wayland compositor via singleton
background: Don't get the monitor manager from the singleton
plugins: Don't get backend from singleton
This applies to MetaPlugin, it's manager class, and the default plugin.
feedback-actor: Pass a compositor pointer when constructing
This allows getting to the display.
later: Keep a pointer to the manager object
This allows using the non-singleton API in idle callbacks.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2718>
As elsewhere, make sure objects that need to have a ownership up to the
context, and use this ownership chain to find relevant components, such
as the backend or the Wayland compositor object instance.
wayland/data-device: Hook up data devices to seats
They are tied to a seat - make that connection in struct fields too, so
that related objects can get to the context via it.
wayland: Don't get Wayland compositor via singleton getter
This means via the ownership chain or equivalent.
xwayland: Hook up manager to Wayland compositor
Same applies to the drag-n-drop struct.
xwayland: Make X11 event handling compositor instance aware
This avoids finding it via singletons in the callee.
xwayland: Don't get Wayland compositor from singleton
xwayland: Pass manager when handling dnd event
window/xwayland: Don't get Wayland compositor from singleton
xwayland/grab-keyboard: Don't get backend from singleton
xwayland: Don't get backend from singleton
wayland: Always get the backend from the context
This means traveling up the ownership chain or equivalent when
necessary.
wayland: Hook up data devices, offers and sources to the compositor
This allows tying them to a context without going through any
singletons.
wayland: Don't get display from singleton
xwayland: Don't get display from singleton
tablet: Don't get display from singleton
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2718>
As with other parts, make objects have the ability to walk up the
ownership chain to the context, to get things like the Wayland
compositor or backend instances.
Contains these squashed commits:
display: Don't get backend from singleton
window: Don't get backend from singleton
keybindings: Don't get backend from singleton
workspace: Don't get backend from singleton
display: Don't get Wayland compositor from singleton
selection: Add display getter
context/main: Get backend directly from the context
clipboard-manager: Don't get display from singleton
stack-tracker: Don't use singleton MetaLater API
startup-notification: Hook up sequences and activations to display
This allows using context aware API directly.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2718>
This means objects have an owner, where the chain eventually always
leads to a MetaContext. This also means that all objects can find their
way to other object instances via the chain, instead of scattered global
singletons.
This is a squashed commit originally containing the following:
cursor-tracker: Don't get backend from singleton
idle-manager: Don't get backend from singleton
input-device: Pass pointer to backend during construction
The backend is needed during construction to get the wacom database.
input-mapper: Pass backend when constructing
monitor: Don't get backend from singleton
monitor-manager: Get backend directly from monitor manager
remote: Get backend from manager class
For the remote desktop and screen cast implementations, replace getting
the backend from singletons with getting it via the manager classes.
launcher: Pass backend during construction
device-pool: Pass backend during construction
Instead of passing the (maybe null) launcher, pass the backend, and get
the launcher from there. That way we always have a way to some known
context from the device pool.
drm-buffer/gbm: Get backend via device pool
cursor-renderer: Get backend directly from renderer
input-device: Get backend getter
input-settings: Add backend construct property and getter
input-settings/x11: Don't get backend from singleton
renderer: Get backend from renderer itself
seat-impl: Add backend getter
seat/native: Get backend from instance struct
stage-impl: Get backend from stage impl itself
x11/xkb-a11y: Don't get backend from singleton
backend/x11/nested: Don't get Wayland compositor from singleton
crtc: Add backend property
Adding a link to the GPU isn't enough; the virtual CRTCs of virtual
monitors doesn't have one.
cursor-tracker: Don't get display from singleton
remote: Don't get display from singleton
seat: Don't get display from singleton
backend/x11: Don't get display from singleton
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2718>
While already cleaning up API, if this should ever be more non-static
than a constant, it's better if its a function on the monitor manager
instance than something static.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2718>
