Commit Graph
4 Commits
Author SHA1 Message Date
MaxsTechReview 4349563fd1 feat(display): zero-copy presentation toggle and compositor display buffers (#2905)
* feat(display): zero-copy presentation toggle and compositor display buffers

Adds a Zero-copy presentation switch under Steam Settings, Display,
Session Display (on by default), and a display-buffer allocator in the
compositor: a Vulkan driver that states displayBuffers asks it for the
memory of the images it exports, so frames presented through the nested
compositor are gralloc buffers that go on the display layer without a
copy. Compressed buffers are offered only where the device's allocator
matches the layout the drivers render to; everything else stays on the
copy path.

The private buffer protocol is renamed to droiddeck-ahb-v1.

* fix(display): zero-copy frames keep the full refresh rate

Two buffers set on the game layer inside one refresh replaced each other, and a frame that slipped into the next refresh made the compositor hold and drop the on-time frame behind it. The layer now queues buffers, and a refresh takes a second frame when the one before it took none. The pacing tool no longer counts its dump time in the frame rate.

* Keep cropped scenes on the display layer and pace layer frames evenly

Accept any in-bounds source crop for the display layer and map the draw
geometry the same way the copy path does. Time layer frames against the
display's frame timeline, and when the game's rate is a steady whole
fraction of the refresh rate show each frame a fixed number of refreshes
after the last so uneven arrivals no longer reach the screen.

* List zero-copy in the settings map

* Hold display buffers until the display releases them

A buffer the game or gamescope rendered straight into went back to its
client as soon as the next one was latched, with the display's release
fence attached to the dma-buf. The KGSL driver does not wait on that
fence, so the client could draw into a buffer the panel was still
reading. Display buffers now wait for the release fence before the
release event, and gamescope composites only into output images the
host has released.

* Keep the game picture copy-free when an overlay has no display layer

When the display cannot give a second layer of its own to an overlay,
the compositor tells gamescope how many planes it can show directly.
gamescope then composes its overlays into the game picture, which still
reaches the display without a copy.

* feat(session): the panel refresh rate follows the session's frame rate

The session learns the panel's refresh rates and reports the refresh rate and
frame rate it wants. The panel moves to the highest rate the frame rate divides
evenly, frame generation counts at its output rate, and a panel whose rate the
system holds offers only that rate.

* feat(session): the panel follows a frame rate the game holds on its own

The compositor reports the game's frame rate once a second. A rate that
holds for three seconds and divides a panel rate becomes the content
rate when neither the app's limit nor the session's limiter is setting
the pace, so a game capped in its own settings gets a panel rate its
frames divide into. A rate that was dropped has to hold twice as long
before it is used again.

* Show generated frames on the game's display layer

With frame generation on, the game's own frames stay on their display
layer untouched and each generated frame is placed on the same layer
between them, timed across the game's frame interval. The compositor
pass is kept when effects, HDR routing or a cursor need it, or when the
panel cannot show the multiplied rate.

* Keep scaling, generated frames and HDR on the display layer

A scaler or frame generation with zero-copy on now draws once into the
display layer instead of falling back to the copied surface. HDR output
goes to the layer when the panel supports it; on a panel without HDR the
picture is tone-mapped on the same path, and the toggle is labelled to
match what the panel can do.

* Spread a frame limit the refresh rate does not divide evenly across refreshes

* Cover the spare-tick and panel-rate paths in the compositor source tests

* Pin the gamescope build with the display-plane, release, refresh and limiter patches

* Give the layer and terminal-caller harnesses the black fill state they compile against

* feat(display): switch HDR during a session without restarting the app

The compositor now offers HDR whenever the device can present it, and the
HDR setting only decides whether the session uses it. Both HDR switches set
gamescope's hdr_enabled, the compositor's HDR output and zero-copy for the
running session. gamescope patch 0139 lets hdr_enabled change after start.
The 'Restart the app to apply.' hint is removed from every locale.

* Remake the output images in the new format when HDR is switched during a session

* Pin the gamescope build that switches HDR during a session

* Apply a saved HDR opt-out at session start and keep HDR retries with their session

* Serialise the HDR output switch with the gate opening

* feat(display): keep the client's HDR setting and the session's in step

The session publishes its HDR state with the other device controls and the
client helper applies it to the client's own display setting. A change made in
the client is reported back and goes through the same switch the drawer and
Session Display use, so all three always show one value.

* fix(desktop): keep the display-buffer socket when the runtime directory is cleared

The desktop launcher removed everything in the runtime directory except the
compositor's Wayland socket, which took the display-buffer socket with it, so
no desktop program could ask for display buffers and every frame was copied.

* fix(display): the client's HDR setting follows the session in every client mode

The bridge only ran when a variable nothing sets was present, so the client's
HDR setting and the session's never met. It now runs whenever the session
publishes an HDR state, and a client that has no such setting is asked again
only every ten seconds.

* fix(display): harden display-buffer hand-off, layer fallbacks and HDR sync

- Stop closing a buffer handle's descriptor the allocator still owns, and free an allocation whose reply could not be sent.
- Rebuild the advertised buffer formats when zero-copy is switched during a session.
- Fall back to the copy path when a layer buffer cannot be allocated, and hide the game layer when it is skipped for a frame.
- Keep a release fence that has not signalled instead of rendering into a buffer the display still reads.
- Ignore a refresh-target reset from a client that does not own the target.
- Restart the content-rate measurement after a pause, clear it when the game stops, and lengthen engagement only for a rate that was released before.
- Compare colour conversions by their values, and read padded cursor buffers at their row width.
- Use the running session's mode for the HDR and zero-copy settings, probe the display before screen effects are pushed, retry a held panel mode, and ignore callbacks after the activity is destroyed.
- Write a newer HDR value while an earlier write is settling, and reapply the value after the bridge is reinstalled.
2026-11-21 00:26:50 -05:00
Kurt Himebauch 758e7983fa feat(display): black frame insertion on 100 Hz and faster screens (#2653)
* feat(display): black frame insertion on 100 Hz and faster screens

* feat(display): black frame insertion lives in Screen effects

* test(compositor): cover black frame insertion in the render-on-commit harness

* fix(display): keep HDR and pacing intact around black frame insertion

* fix(display): gate black frame insertion on the panel's refresh, not the tick rate

BFI checked the 100 Hz minimum against g_refresh_ns, an average of the time
between handled vsync ticks. BFI presents on every tick, so under load the
ticks run late, the average fell to 95-100 Hz on a 120 Hz panel, and BFI
switched off and on again many times a second (3793 times in one session on
a Retroid Pocket Duo). That showed as uneven flicker.

Use the display mode's refresh (g_output_refresh_mhz) instead, which only
changes when the mode does.

* fix(display): offer black frame insertion for the bottom screen and while it is on

* fix(display): capture a picture, not a black refresh, while black frame insertion strobes

* fix(display): hold black frame insertion off until the last capture ends

* fix(display): keep the game composed while a capture holds black frame insertion

* test(display): a picture hold presents only pictures while black frame insertion runs
2026-11-16 07:33:39 -05:00
MaxsTechReview b91e8fdeab fix: a session in picture-in-picture draws at 30 fps instead of the full refresh (#1842)
* fix: a session in picture-in-picture draws at 30 fps instead of the full refresh

Since auto picture-in-picture became the default (#1727), pressing Home leaves the session running in a small window that takes no input. A game kept drawing at its full rate there: on a REDMAGIC Astra 2 Among Us in picture-in-picture drew 3.31 W, the same as playing it and about 1.05 W more than a suspended session. While in picture-in-picture the compositor now paces the session at 30 fps (or a lower limit already set). Mirroring to glasses keeps the glasses rate, and the display mode request ignores the picture-in-picture limit so the app in front is not slowed.

* fix: keep the no-argument output limit the glasses pacing tests read

* fix: hold scene renders to the output limit and keep frame generation off in picture-in-picture

A frame limit only paced dmabuf buffer releases, so a shared-memory client was still drawn on every vsync with its frame callbacks sent at the panel's rate. The vsync tick now waits until the limit's interval, less half a refresh, has passed since the last scene. Frame generation is disarmed while picture-in-picture is shown and armed again when it ends; frame generation requests are applied in order.

* Give the commit render harness the frame limit hold

* Hold scene renders to a frame limit deadline, also on the fallback timer, and keep frame generation for glasses mirroring in picture in picture
2026-10-30 13:29:10 -04:00
The412Banner 8549ac9ce8 perf(compositor): draw a new frame when it arrives (#1707)
A new game frame waited for the next screen tick before the compositor
drew it: 1.9-3.5 ms on average at 144 Hz, measured with the pacing
counters. With frames going straight back to the game (no compositor
limit), the compositor now draws as soon as the clients' requests are
handled, at most once per refresh, and presentation times stay on the
tick.

gamescope delivers about one frame per refresh, so the tick leaves a
frame younger than one refresh for the next arrival to draw, unless the
tick itself changed the window, effects or headset pose; a frame no
other frame follows is drawn at the tick after. Once a held frame is
still waiting at the next tick (a client paced by frame callbacks), the
tick draws every frame for a second before holding one again. The
queued draw checks again that drawing on arrival is still on.

Paced output (glasses), streaming and frame generation keep drawing on
the tick. The hidden pref renderOnCommit (default on) turns it off. The
pacing log line counts the scenes drawn on arrival. On the device new
frames wait 0.12 ms instead of ~2 ms.
2026-10-28 09:21:07 +00:00