8703feedee
tty: Print full path to render node
2024-01-06 08:46:30 +04:00
1c7c5b3f28
tty: Avoid non-linear cross-device formats
2024-01-05 23:23:32 +04:00
5cd31e5730
Implement multi-GPU support
...
Rendering always happens on the primary GPU.
2024-01-03 18:16:20 +04:00
de3fc2def0
Make rendering generic over renderer
2024-01-03 18:05:15 +04:00
fd1d4b07fd
Replace renderer() with with_primary_renderer()
2024-01-03 11:38:09 +04:00
8b5acd5e6e
Rename RenderResult::Error to Skipped
2024-01-03 11:38:07 +04:00
31bb9096e2
Move Niri::render() invocation to backends
2024-01-03 11:38:02 +04:00
57a7347620
Move import_dmabuf to backends
2024-01-03 11:37:04 +04:00
628891db2c
Add disable-cursor-plane debug flag
2024-01-03 08:42:49 +04:00
be6e25f5fb
tty: Extract config variable
2024-01-03 08:42:49 +04:00
655fe413fb
tty: Don't store render formats
...
Actually, how did it even fork before? Pretty sure it was storing render
formats, not texture formats, but with render formats
weston-simple-dmabuf-feedback doesn't work?
2024-01-01 17:31:58 +04:00
ac6ff7ff41
tty: Disable dmabuf global before destroying
2024-01-01 17:31:58 +04:00
84befb4e91
tty: Move dmabuf_global up into Tty
2024-01-01 09:54:45 +04:00
d39f7bebf3
tty: Fix import sorting
...
One of the imports was too long so rustfmt bailed.
2024-01-01 09:40:41 +04:00
0dd9a42087
Move DmabufState to Niri
2023-12-31 12:02:39 +04:00
d155f5cd6c
Add a config flag to disable an output
2023-12-18 10:27:41 +04:00
0d2ed2e14a
tty: Fix clippy warning
2023-11-24 09:50:30 +04:00
42bb79d7c3
tty: Rename vblank offset plot to time since presentation
...
Makes it clearer what this is.
2023-11-24 09:19:49 +04:00
588e6d7cdd
tty: Rename presentation target offset plot to misprediction
...
Makes it clearer what this is.
2023-11-24 09:15:25 +04:00
0a2052945e
Add support for wl_compositor@v6
2023-10-29 13:58:48 +04:00
e73f33d6a3
Draw background with a solid color buffer
2023-10-29 10:45:24 +04:00
345428bce7
Drop GbmDevice in the end
...
The `GbmDevice` must be destroyed after destroying all the objects
associated with it.
2023-10-27 16:48:49 +04:00
3fd421f13f
Update Smithay
2023-10-24 15:05:14 +04:00
d59fe34243
Make clear color a const
...
The damage tracker doesn't understand clear color changes.
2023-10-13 13:30:36 +04:00
29053a807b
Return RenderResult from render()
2023-10-13 13:30:36 +04:00
0a32747495
Send dmabuf feedbacks from the backend
2023-10-13 13:30:36 +04:00
f625dede1a
Extract niri.clear_color()
2023-10-13 13:30:36 +04:00
15b6c84af0
tty: Start new tracy frame only when queueing redraw
2023-10-11 09:02:55 +04:00
082d734b2a
Make zbus optional under "dbus" feature
2023-10-10 10:25:33 +04:00
189d1bd97b
Add power-off-monitors bind
...
Implements https://github.com/YaLTeR/niri/issues/24
2023-10-09 18:37:43 +04:00
bb3fbe2e83
Make output modes configurable
2023-10-03 08:35:24 +04:00
2d3d0ce165
tty: Remove assert on needs_sync
...
cosmic-comp doesn't do anything with it.
2023-10-02 10:54:05 +04:00
5b6b3fcfbe
Avoid sending frame callbacks to invisible surfaces
2023-10-01 19:41:42 +04:00
a3aa5fca12
Refactor frame scheduling
...
Combine the redraw state variables into one enum, and refactor to get
rid of the requirement that a VBlank must queue a subsequent redraw.
Also fix the bug where ongoing animations that produced no damage could
stall the redrawing.
2023-09-30 17:13:56 +04:00
f381db8354
Make pipewire / xdp-gnome-screencast optional
2023-09-30 09:58:34 +04:00
404661ed8d
Throttle frame callbacks to once per monitor refresh
...
Under some circumstances, the compositor can get into a commit-frame
callback busy loop with a client. For example, if a client redraws on
frame callbacks, but the resulting frame has empty damage (e.g. the
damaged part of the client is outside the monitor). Or if the client
simply commits with empty damage (looking at you, Firefox).
This behavior is compliant with the Wayland specification and with the
intended idea of frame callbacks, but causes a lot of unnecessary CPU
usage in the client and the compositor.
To solve this problem, this commit introduces frame callback throttling.
Every surface may only receive a single frame callback in one
monitor refresh cycle. If a surface commits resulting in no KMS frame
submission, a timer is created, that will fire at the predicted would-
be-VBlank time, and send the accumulated frame callbacks.
This way, a surface that redraws on frame callbacks will not notice
any change in frame callback delivery, if its commits suddenly stop
producing KMS updates.
2023-09-29 13:12:50 +04:00
804e064826
shell: add support for kde server decorations
...
This should provide server side decorations for the gtk3 applications,
like firefox.
2023-09-27 18:38:38 +04:00
dc10e464ad
tty: Take metadata by value
2023-09-26 13:45:03 +04:00
906ee36a93
Skip rendering when backend is inactive
2023-09-26 13:45:03 +04:00
ac16717f4e
Avoid unwraps in more places
2023-09-26 13:45:03 +04:00
52aee11aa5
tty: Add more trace spans
2023-09-26 13:45:03 +04:00
e299328741
tty: Extract other event handlers into functions
2023-09-26 13:45:03 +04:00
0f43e90b2a
tty: Avoid panicks in the vblank callback
2023-09-26 13:45:03 +04:00
bbf707367a
tty: Extract on_vblank()
2023-09-26 13:45:03 +04:00
cfe80574df
Use new tracy-client new_leak() methods
...
I added them to avoid the use of internal module and leaks when tracing
is disabled.
2023-09-26 13:45:03 +04:00
cd4f7c0378
Remove LoopData
...
The calloop Smithay update finally lets me do this.
2023-09-26 13:45:03 +04:00
109668fa30
Add output configuration & integer scaling support
2023-09-21 13:48:50 +04:00
9fc731c115
Plot target presentation time offset
2023-09-20 11:51:25 +04:00
361b34f460
tty: Pass output name to tracy span
2023-09-19 07:39:21 +04:00
94da8b1e3e
tty: Try to create context with high priority
2023-09-18 20:09:57 +04:00
Ivan Molodetskikh