Upstream PCSX2 interprets the PS2's CPUs on ARM64 (pcsx2/arm64/RecStubs.cpp):
NFS Underground 2 ran at 17 fps with one core at 100%. ARMSX2 is the PCSX2 fork
with ARM64 recompilers. It is listed beside PCSX2 for the ps2 folder, installs
from the desktop catalog (package armsx2), and bannerlator-pad-defaults gives it
the PCSX2 profile under its own folder (ARMSX2/inis/PCSX2.ini).
vulkan is the default desktop renderer (settings list it first). When labwc
cannot start on a GPU renderer the desktop falls back to pixman as before, and
now records the renderer with the patched wlroots and Linux driver it failed
on, so later desktops go straight to pixman instead of waiting 20 s each time;
an app update, a different driver, or choosing a renderer again in the settings
(which deletes the record) retries.
- A program from the rail draws with the desktop's Linux driver instead of the
Steam session's. The driver's shader cache is keyed on its build, so with two
drivers an emulator compiled everything once per way of starting it; seen on
device: desktop on the imported Turnip 26.3.0, rail on the runtime's own.
- RPCS3's default shader mode builds 6650 interpreter pipelines every boot.
bannerlator-pad-defaults starts a new install on Async Recompiler
(multi-threaded) and moves an existing one still on the default once; a
marker leaves any later choice in RPCS3's settings alone. Enum strings from
rpcs3/Emu/system_config_types.cpp.
bannerlator-pad-defaults now covers DuckStation, PCSX2, Dolphin (GameCube and
Wii), Cemu, melonDS, PPSSPP and RetroArch as well as RPCS3, from each project's
source: files an emulator would otherwise fill with keyboard-only defaults are
seeded when missing (Dolphin, Cemu, PPSSPP, RPCS3), or given SDL bindings in the
config the emulator wrote itself when nothing binds a controller yet
(DuckStation, PCSX2 with its SettingsVersion, melonDS). RetroArch gets an Xbox
360 profile for its sdl2 driver and is moved off udev, which finds no pads in the
runtime. The desktop re-runs it every 30 s, so an emulator's first config is
covered for its next start.
The PC keyboard takes 47% of the screen height, header included.
- The pad keeps Steam Input's virtual-gamepad identity only in Steam sessions.
SDL ignores a Steam virtual gamepad unless run under Steam, so on the desktop
and from the rail every SDL emulator saw no controller at all; there it is a
plain Xbox 360 pad now, which SDL and evdev map by themselves.
- bannerlator-pad-defaults writes Player 1 on the controller for emulators that
have no profile yet (RPCS3 first: without one it takes the keyboard). Run at
desktop start and before a program from the rail; never overwrites.
- The desktop and everything started from it, and a program from the rail, run
on the game cores when chosen, else every core but the slowest cluster
(program_cores); BL_PIN_CORES=0 turns it off. Game cores reach every mode.
The runtime carries Termux's proot 5.1.107-70 as a prebuilt binary, built from
termux/proot 4dba3afb (every file:line site in it matches that commit). The apk
now builds that same commit with the NDK, with three patches ported from
WinNative's feature/proot-enhancements, and prefers it over the runtime's copy:
- Tracees are hashed by pid, so each ptrace stop finds its tracee without
walking a list of every thread, and terminated tracees are swept only after
a termination.
- A clean absolute path three or more directories deep under the rootfs alone
has its parent opened once and taken as canonical when the kernel's path for
it is the one proot would build, instead of an lstat per component; the final
component of a call that does not follow it is no longer lstat'ed. Extensions
still see the parent's host path, and the f2fs workaround keeps the old walk.
- A clone3 thread keeps its inherited flags when clone_args cannot be read.
talloc is linked in statically from Samba's release tarball, so libproot.so
needs only libc and libdl. On an x86_64 Android 15 emulator the same build ran
a path and thread test suite with output identical to the unpatched commit
(fake id, fake root, link2symlink, with and without seccomp), and a stat-heavy
benchmark ran 25-27% faster, 31-36% with 1500 live threads.
The drawer's Keyboard row has two buttons: Hardware opens ui/PcKeyboard, a full
PC layout - Esc, F1-F12, Ins/Del/Home/End/PgUp/PgDn, the number and letter rows,
Tab, Caps, both Shifts, Ctrl, Super, Alt, AltGr, Space and the arrows - whose
keys are evdev codes pressed and released on the compositor's wl_keyboard, as a
USB keyboard would; Android opens the Android keyboard as before. Keys go down
on touch and up on lift (the program's key repeat works); modifiers are sticky
(tap: next key, tap again: lock, again: off). A controller moves over the keys
with the d-pad and presses with A; B closes it.
droiddeck-gpu runs a program in a gamescope of its own on the app's compositor
(wayland-0), full screen over the desktop - the path the rail's play chips use,
where Vulkan works. droiddeck-desktop-gpu gives every Game/Emulator menu entry a
per-user copy through it (auto: while the desktop is on pixman), keeping 'Run in
a desktop window' as an action.
droiddeck-desktop loads /usr/local/lib/droiddeck-wlroots for the vulkan/gles2
renderers and falls back to pixman when labwc stops at start. The build stages
the wlroots release named by tools/wlroots/release.env, once published.
With a controller, credits only showed the ripple's faint box on the
dark rail, and the primary button (Play) only grew slightly with a shine
that barely shows on its light fill. Both now get the 2 dp signal-blue
outline the other controls use; credits' text turns blue as well.
With the drawer open the controller still fed the game - PadBridge
consumed every pad key and axis - and nothing in the drawer had focus.
Now, while it is open, pad events go to Android's own handling instead:
the d-pad and stick move focus, and the framework's fallbacks make A
select and B Back (which closes the drawer), as on the app's main
screen. Opening it releases the game's pad (PadBridge.releaseAll), so a
button or stick held at that moment is not left down in the game.
It opens with focus on the STEAM button (in a desktop session, the first
setting), and the STEAM and ... buttons show a focus outline like the
rest of the app's controls. ToggleRow gains a modifier for its box.
App name, app ID and code package: DroidDeck, com.droiddeck.launcher
(sources moved to com/droiddeck/launcher, the 68 JNI functions renamed
with them). The new app ID installs beside SteamDeck rather than over it.
Everywhere the name was the app's own: the sidebar title, the session
report, share zips, the shell title, wake-lock and log tags, User-Agents,
the Compose theme, the Download folder and switch files (droiddeck-env,
-osc, -driver, -tu-debug, -no-pad, -pad-log, -no-hud, -wlr-renderer; the
old names are not read), and inside the runtime (droiddeck-desktop,
Firefox's droiddeck.js, the shortcuts tag, the rumble socket, debug and
cache paths). Build scripts, CI artifact (droiddeck-apk), CMake project,
README and the progress log follow.
Kept on purpose: Valve's Steam Deck (the -steamdeck flag,
steamdeck_publicbeta/stable, 'Steam Deck mode', the steamos.cloud
package URLs) and names inside the hosted desktop package (steamdeck.png,
steamdeck-steam, steamdeck-bigpicture). 'Desktop installed' now tests for
labwc instead of the launcher the app stages itself. A committed
__pycache__ .pyc is removed.
The trash can sat inside the row's single clickable area, so with a
driver focused there was nothing to the right for a controller to move
to. Each row is now two focus targets side by side - the driver and its
trash can - linked both ways (right/left); the row's outline still
follows the driver part, so it looks the same.
Opened from the cog, the Steam/Desktop session page had nothing focused:
the cog that was focused is gone and a controller had to find its way
back in. The page now puts focus on its first control (Resolution) as it
opens, so the d-pad works at once; coming back from a driver page still
returns focus to the driver box that opened it. ChoiceRow gains a
modifier for its box.
Backing out of the Runtime or Display driver page rebuilt the session's
settings page from the top with nothing focused. Its scroll position is
now held above the page (SettingsPage takes an optional ScrollState) and
survives the driver page, and controller focus returns to the driver box
that was pressed to open it (ValueChip takes a modifier for its
FocusRequester).
- Every control shows a focus outline: the driver rows (the list at the
top had none), their trash cans (in red), the refresh button, Restore
built-in drivers, the confirmation's buttons, and the page's Back
(shared by every settings page).
- D-pad up from the first driver moves to the refresh button beside the
title; the page opens with the driver in use focused.
- Back (the controller's B too) closes the driver page only, back to the
session's settings, instead of leaving settings altogether.
- The delete confirmation opens on Cancel, so a stray A press never
deletes a driver.
Runtime driver and Display driver open a full page instead of a
drop-down, built from the settings widgets (as previewed): INSTALLED
lists every driver as a tappable row with a radio mark and a tag
(BUNDLED / DOWNLOADED / IMPORTED); AVAILABLE TO DOWNLOAD lists what the
last check of Banners-Turnip and WinNative found, with a Download button
and progress; Import stays at the bottom. The refresh button sits beside
the title (SettingsPage gains an action slot) and is the only thing that
goes online; it spins while checking.
Every driver gets a trash can with a confirmation, the bundled ones too:
deleting a bundled display driver removes its unpacked copy and hides it
(TurnipDriver.hideBundled), with 'Restore built-in drivers' to bring
them back. It stays in the apk and Auto still unpacks it on a GPU that
needs it, so a deletion can never leave a device without a driver. Auto
and Runtime default are settings, not drivers, and have no trash can.
- WinNative-Emu/Drivers joins Banners-Turnip as a source: WN-Turnip
(Android, v1.x line) and WN-Linux-Turnip (Linux, linux-v0.x line), each
in its Balanced and Performance builds for every Adreno.
- Each source's recent releases are read newest first and every variant
is offered from the newest release carrying it. Banners-Turnip stopped
publishing its A8xx Linux build after v26.3.0-20260923, so the runtime
menu listed no 8xx driver at all; it now offers that last one. The
refresh line names each source's newest release and any that could not
be reached.
- Drop-down menus are capped at the window's height and their entries
scroll under the title: a driver menu with its downloads ran off the
bottom of the screen, hiding the refresh line and Import.
Both driver menus (Runtime and Display) now offer the latest release of
The412Banner/Banners-Turnip: each asset that fits the menu and is not
installed yet is a 'Download Turnip <tag>' entry labelled by GPU
(plain / -A8xx / -710-720-Test zips for the display driver, -Linux ones
for the runtime driver; -Wayland ones are never offered). Nothing goes
online on its own: a refresh line checks GitHub when tapped and says
which release it found and when; the result is remembered. A download
goes through the same importer as a picked zip and lands as an ordinary
driver.
Imported and downloaded drivers get a trash-can button with a
confirmation instead of the small cross. Deleting the driver a mode is
set to puts that mode back on its default (Runtime default / Auto), and
a deleted release download is offered again.
The QAM offers Performance Overlay levels but nothing drew them: the
runtime has no mangoapp. CI now fetches Valve's mangohud-20260902.1 and
the five libraries mangoapp needs that the runtime lacks, at the versions
it was built against (fmt 10.2.0, spdlog 1.13.0, glfw 3.4, libtraceevent
1.8.2, libtracefs 1.8.0 from Valve's mash-20240705.2 snapshot), each
pinned by sha256 in tools/mangoapp/packages.txt. They are staged under
usr/local/lib/mangoapp with a wrapper on PATH; the runtime image is not
touched.
SessionActivity finished at once when no runtime was installed - a guard from before the loading
screen could install it - so Play (and Desktop) opened a screen that closed before anything
happened. It now logs and carries on to installThenStart.
Resolution gains Custom…: a width × height box (320×240 to 3840×2160, evened) with one-tap
presets for 4:3 (960×720, 1024×768, 1280×960), 16:10 (1280×800) and 16:9. A custom size replaces
the cap and the shape for that mode - the Shape row greys out and says so - and the compositor fits
it to the panel with bars where the shapes differ. Choosing any cap clears it. The device report
lists it for both modes.
The panel's shape was never allowed narrower than 16:9 (the foldable guard), so on a 4:3 panel
such as the Retroid Pocket's both choices gave 16:9 with bars. A third choice, Exactly this panel,
presents the panel's own aspect with no floor; the other two are unchanged and foldables still
default to 16:9. Offered in the mode settings and the in-session menu.
The package list opens in the front end's pane like the Settings pages. Each tier (Native ARM64,
ARM64 experimental, source builds) is a dropdown row with the rail's caret and an installed/total
pill; Native ARM64 starts open. Each package has its own Install/Remove button, and the one that is
installing shows its own step line and bar in its row (its tier opens when the install starts).
Leaving the page does not stop an install. The rail's Install emulators and Setup's Desktop & apps
highlight while the page is open.
Desktop is enabled without a runtime or the desktop package. The session's loading screen installs
whatever is missing first - the runtime, then the desktop ("downloading the desktop · 212 of 426
MB", checking, unpacking) - and starts the session when both are in. A failure ends on the loading
screen with the reason. Desktop UI and Desktop now give the same non-Adreno warning as Play before
any download.
Play and Desktop UI are enabled without a runtime. The session's loading screen downloads and
unpacks it first, reporting on its usual step line and bar ("downloading the Linux runtime · 332
of 791 MB", then checking, then unpacking), and starts the session when it is in. A failure ends
on the loading screen with the reason. Non-Adreno devices still get the warning before any
download. Setup > Linux runtime is unchanged.
- The client's sound goes back to the adaptive AAudio sink 0.1.5 shipped (bundled unchanged as
module-aaudio-classic-sink, sink name AAudioSink). Through the DirectAudio relay it stayed
choppy on an AYN Thor (Android 13, 20 ms bursts) even with a two-burst buffer.
- Steam cog > Audio: DirectAudio for games (on by default, unchanged) and Steam client audio
(Classic by default, or DirectAudio). The relay also starts when only the client wants it.
- The relay runs with --log and the daemon at info level, so audio.log carries the stream each
path was granted and, on the relay, Android's xrun count every 1000 callbacks.
- Device report shows both settings.
The unpack check compared a hand-bumped stamp only. Two sink changes shipped without a bump, so a
device that had run the earlier build kept the old module: the 40 ms ring floor and the 24 ms
device buffer never reached the AYN Thor they were for. The stamp now carries a CRC-32 of
pulseaudio.tzst. The sink's connect line names the buffer it asked for, and one line ten seconds
in reports the relay's device buffer, ring target and underruns.
- module-directaudio-sink keeps the relay's ring at least 40 ms deep. The relay starts at 8 ms,
which the producer inside proot missed while the client loads (two underruns on the FIT's fresh
install, steady stutter reported on another phone).
- Setup > Session logs > Share latest logs zips the newest session folder, wherever it was written
(Downloads, or app-private files when Downloads is not writable), and opens the share sheet via
a FileProvider scoped to that zip.
- Microphone defaults to on, and the app asks for RECORD_AUDIO once at first launch, together with
the storage prompt. A refusal is not asked again at start-up.
- modules/armhf held the inherited Winlator AAudio sink and two protocol modules; the app is
arm64-only and nothing loads that folder. Removed, bundle stamp bumped so installs re-unpack.
Everyone who never touched the toggle gets it; anyone who turned it off keeps it off (the setting is
only stored when changed). Off, the client's sound came from AAudio driven inside proot, which
stutters on some devices.
A second sink module: connects to the relay's socket (lazily, the relay comes up after the
daemon; a clocked null sink until then), sends the hello, maps the render ring it gets back and
keeps it topped up to the relay's target with a futex wait per burst. Every AAudio call, the
adaptive buffer and the route handling stay in the relay, outside proot. With DirectAudio on the
daemon loads this sink instead of module-aaudio-sink; the bundle carries both.
frontend/AddedGameArt: cover/poster/boxart/folder-name images, header, hero, logo and icon from
the game's folder (or its art subfolder); otherwise a one-time store search by folder name and
the capsule, header, hero and logo from Steam's CDN into the app's cache (a miss retried after a
week); an icon pulled from the .exe. The rail shows the portrait. The session's listing carries
the paths; bannerlator-steam-shortcuts copies them into userdata/<acct>/config/grid under the
client's names, never over a file it did not write itself, and sets the shortcut's icon. A toggle
in the cog turns the store lookup off.