mirror of
https://github.com/Droid-Deck/DroidDeck.git
synced 2026-11-20 13:26:02 +00:00
Merge current main into PR #1061
This commit is contained in:
commit
b4a9dec170
138 files changed
+6127
-257
No files matched your search
@@ -0,0 +1,80 @@
|
||||
# Experimental Adreno 850 detection and driver imports
|
||||
|
||||
Reports [#342](https://github.com/Droid-Deck/DroidDeck/issues/342) and
|
||||
[#723](https://github.com/Droid-Deck/DroidDeck/issues/723) describe Turnip failing
|
||||
on Adreno 850, including an SM8975 / Adreno850v2 device reporting
|
||||
`VK_ERROR_INCOMPATIBLE_DRIVER`. These reports motivate the detection changes;
|
||||
they are not proof that the application's existing Turnip builds support A850.
|
||||
|
||||
## Detection
|
||||
|
||||
The application parses a GPU model only from an Adreno device name. The raw
|
||||
KGSL `gpu_chipid` is recorded separately in device diagnostics: a decimal or
|
||||
hexadecimal chip ID is not an Adreno model number.
|
||||
|
||||
Detection prefers the KGSL name, then an already cached Vulkan device name,
|
||||
then a platform fallback. An explicit `SM8975` or `QTI SM8975` SoC identifies
|
||||
Adreno 850 when neither device name is available, including when the board
|
||||
property still uses an older platform name. Driver auto selection uses the
|
||||
same detection, with existing external-driver and Quest overrides retained.
|
||||
This does not start a new Vulkan query during driver selection.
|
||||
|
||||
A850 is labelled experimental. The SM8975 platform identity retains that
|
||||
classification even when the selected driver reports Adreno 840; the displayed
|
||||
model still follows the KGSL/Vulkan name. Identifying the hardware does not add native
|
||||
A850 support to Turnip or prove that Steam and games will run.
|
||||
|
||||
## Optional community driver bundle
|
||||
|
||||
The community attachments uploaded to #342 on 2026-10-12 contain experimental
|
||||
Android display and Linux game drivers. Their metadata describes a binary
|
||||
patch that treats A850 as A840 and reuses some Adreno X2 entries. These are
|
||||
not native A850 drivers; they may fail to start, hang, or render incorrectly,
|
||||
and have not been validated across A850 revisions.
|
||||
|
||||
Download the original attachments separately:
|
||||
|
||||
| Platform | Original attachment | SHA-256 |
|
||||
| --- | --- | --- |
|
||||
| Android | [AdrenoTools-A850-experimental.zip](https://github.com/user-attachments/files/33319268/AdrenoTools-A850-experimental.zip) | `0cf849d068c875f6c469380ca1f74da1f757575bce53d4ef9c1f6d38677cbe8b` |
|
||||
| Linux | [A850-experimental.zip](https://github.com/user-attachments/files/33319267/A850-experimental.zip) | `40fc59506620001531b9bdd1c9152e104c6fb1f4c07773c4743b9d70deb87eec` |
|
||||
|
||||
Prepare a bundle accepted by DroidDeck's existing DriverBundle importer:
|
||||
|
||||
```sh
|
||||
mkdir -p build/a850-experimental
|
||||
python3 tools/prepare_a850_driver_bundle.py \
|
||||
--android /absolute/path/AdrenoTools-A850-experimental.zip \
|
||||
--linux /absolute/path/A850-experimental.zip \
|
||||
--output build/a850-experimental/A850-community-experimental.zip
|
||||
```
|
||||
|
||||
The tool accepts only the pinned originals, validates their ZIP entries,
|
||||
metadata and ELF platform, and preserves the original files byte for byte.
|
||||
It does not download, execute or rebuild drivers. Neither binary is included
|
||||
in the source tree or installed automatically by these changes.
|
||||
|
||||
To try the bundle, stop the session and wait for any automatic driver check
|
||||
or download to finish. Note the existing driver selections, then use
|
||||
**Components → GPU drivers → Mix and match → Import .zip**. This is a GPU
|
||||
driver bundle, not a Decky plugin ZIP. Confirm that both the Android display
|
||||
and Linux game drivers are selected, then fully exit and reopen the app.
|
||||
Keep **Session → Performance → Session repair → Turnip: sysmem rendering**
|
||||
enabled as advised by the original Linux driver metadata.
|
||||
|
||||
Test entering Steam separately from game rendering. If it fails, restore the
|
||||
previous drivers and restart the app. Preserve the **Share logs** ZIP,
|
||||
including `device.txt`, `session.log`, the exact chip ID and Android version.
|
||||
|
||||
## Validation
|
||||
|
||||
```sh
|
||||
python3 -m unittest discover -s tools/tests -p test_a850_driver_bundle.py -v
|
||||
./gradlew -PskipRust=true :app:testDebugUnitTest --tests 'com.droiddeck.launcher.gpu.*'
|
||||
```
|
||||
|
||||
The tests cover detection priority, chip-ID rejection, SoC normalization,
|
||||
experimental status, driver-generation selection and bundle integrity. They
|
||||
do not replace Vulkan, Steam or game testing on the target phone. A user
|
||||
confirmed that importing the experimental bundle succeeded; target-device
|
||||
Steam and game compatibility remain unverified.
|
||||
@@ -0,0 +1,25 @@
|
||||
# Themes
|
||||
|
||||
A theme is a `Palette` (`ui/Theme.kt`): a dark background scale (`background`, `surface`,
|
||||
`surfaceVariant`, `line`, `line2`), two text colours, and the accent: `primary` / `signal`
|
||||
(flat accent, also Material's primary), `primary2` (the darker or second accent) and
|
||||
`onPrimary`. `PrimaryButton` and the launch flood already draw a gradient from `primary` to
|
||||
`primary2`. The choice is stored as an id (`SessionPrefs.theme`) and picked in Setup › Launcher.
|
||||
|
||||
## Gradient theme
|
||||
|
||||
`Themes.GRADIENT` is the one theme whose colours the person picks: a primary and a secondary,
|
||||
kept in `SessionPrefs.gradientColors` and in `Themes.gradientColors`, a Compose state, so a change
|
||||
restyles the whole app at once. `App.onCreate` loads the saved colours; code outside a composition
|
||||
uses `Themes.current(context)`.
|
||||
|
||||
`ThemeColors.palette` builds the full palette from the two colours: primary becomes
|
||||
`primary` / `signal`, secondary becomes `primary2`, the background scale is a dark neutral tinted with
|
||||
the primary's hue (plain grey for a grey primary), and `onPrimary` is dark or white by the
|
||||
luminance of the pair. A colour too dark to show on the dark background is lifted
|
||||
(`ThemeColors.readable`) rather than refused.
|
||||
|
||||
Setup shows Primary and Secondary under the theme when Gradient is chosen, each a small colour
|
||||
square. It opens `ColorPickerDialog`: a wheel (hue and saturation, touch), hue / saturation /
|
||||
brightness sliders (they work with a controller) and a hex box, with the old and new colour side by
|
||||
side. OK applies it.
|
||||
Vendored
+21
@@ -0,0 +1,21 @@
|
||||
MIT License
|
||||
|
||||
Copyright (c) 2026 Holger Bartnick
|
||||
|
||||
Permission is hereby granted, free of charge, to any person obtaining a copy
|
||||
of this software and associated documentation files (the "Software"), to deal
|
||||
in the Software without restriction, including without limitation the rights
|
||||
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
||||
copies of the Software, and to permit persons to whom the Software is
|
||||
furnished to do so, subject to the following conditions:
|
||||
|
||||
The above copyright notice and this permission notice shall be included in all
|
||||
copies or substantial portions of the Software.
|
||||
|
||||
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
||||
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
||||
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
||||
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
||||
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
||||
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
||||
SOFTWARE.
|
||||
Vendored
+47
@@ -0,0 +1,47 @@
|
||||
# VITURE protocol V2 attribution and scope
|
||||
|
||||
The frame builder, parser, pose/raw layouts, and commands in
|
||||
`VitureV2Protocol.kt` are adapted from [elasticjava/viture-v2](https://github.com/elasticjava/viture-v2)
|
||||
at commit `ece8c30485f7d357fefa362ae74f3eabac9b3899` (MIT).
|
||||
Copyright (c) 2026 Holger Bartnick. The full license is in `LICENSE` here.
|
||||
|
||||
Sources used are upstream `PROTOCOL.md`, `src/lib.rs`, and `src/xr.rs`.
|
||||
Upstream developed and measured its driver on VITURE Pro 2 XR `35ca:1301`.
|
||||
Its research used captures of the proprietary SDK's USB traffic and logging;
|
||||
the resulting implementation communicates directly over USB without that SDK.
|
||||
This is an SDK-free implementation, not a claim of SDK-independent protocol research.
|
||||
Our earlier local proprietary SDK trace was not used for this adaptation.
|
||||
|
||||
The Android transport and application integration are implemented in Kotlin.
|
||||
Building this path requires no Rust toolchain, vendor SDK, vendor binary,
|
||||
developer account, private download, or repository secret. It uses Android's
|
||||
USB host permission and `UsbRequest` APIs. To explicitly build without the
|
||||
optional vendor driver, pass the Gradle property `-PvitureSdkDir=`.
|
||||
|
||||
The application enables this experimental transport only for `35ca:1301`.
|
||||
It requires valid pose samples before advertising 3DoF, uses the upstream
|
||||
120 Hz pose mode, and treats the device timestamp as an uninterpreted counter.
|
||||
Delivered timestamps use host `System.nanoTime()` receipt times. Bootstrap and
|
||||
resume reject samples collected before the enable write completes. This host-time
|
||||
cutoff does not establish when hardware produced a sample or purge device/kernel
|
||||
buffers. The pose
|
||||
basis conversion follows upstream's Y-up rendering convention. Hardware captures
|
||||
verify changing, normalized poses during head motion; the relative rotation mapping
|
||||
is unit-tested. The user also reported a successful Pro 2 game test; see the
|
||||
validation notes for the scope of that report.
|
||||
|
||||
Raw gyro and accelerometer packets can be decoded with their complete payload
|
||||
preserved, but raw data is not fused or used by the application. Upstream reports
|
||||
that enabling pose and raw streams together can ACK successfully while sending
|
||||
no samples, so the application requests pose alone. Accelerometer values remain
|
||||
in g; gyroscope values are in radians/second according to upstream.
|
||||
|
||||
Brightness, volume, and display queries are supported. Setters, display switching,
|
||||
6DoF, cameras, and hand tracking are not implemented by this path. Upstream's
|
||||
README/protocol document and source disagree about setter coverage; this
|
||||
adaptation does not use those setters without device validation.
|
||||
|
||||
The standalone `tools/viture/viture-v2-probe` captures descriptors, commands,
|
||||
replies, and pose/raw packets for device compatibility testing. Its provenance
|
||||
and license are included beside its source. Pro 2 hardware checks and their
|
||||
limitations are recorded in [VALIDATION.md](VALIDATION.md).
|
||||
+102
@@ -0,0 +1,102 @@
|
||||
# Pro 2 direct USB validation
|
||||
|
||||
These tests used a Samsung Galaxy Z Fold7 (SM-F976U), Android 17, with glasses
|
||||
connected to the phone and wireless ADB used for installation and log collection.
|
||||
The tested Pro 2 identifies as `35ca:1301`, with interface 0 exposing 64-byte
|
||||
interrupt IN `0x81` and interrupt OUT `0x01`. Its firmware query returned
|
||||
`30.0.00.001_20260714`. Results apply to this device and firmware.
|
||||
|
||||
## Protocol probe
|
||||
|
||||
The SDK-free Java probe in `tools/viture/viture-v2-probe` uses the attributed MIT
|
||||
protocol. Every run queried firmware, enabled one stream, and sent stream-off at
|
||||
completion. Each Pro 2 run received a successful firmware response and successful
|
||||
start/stop acknowledgments. All three had zero malformed frames or samples.
|
||||
|
||||
| Run | Valid samples | Measured rate | Observation |
|
||||
| --- | ---: | ---: | --- |
|
||||
| Pose, initial 15-second window | 1,762 | 120.67 Hz | Normalized quaternions; components changed |
|
||||
| Pose, 30-second motion window | 3,571 | 120.63 Hz | User turned, looked up/down, and tilted; all components varied |
|
||||
| Raw, separate 15-second window | 1,471 | 100.77 Hz | Finite gyro and accelerometer values; acceleration magnitude 0.882–1.129 g |
|
||||
|
||||
Rates use host receipt time between the first and last samples. Query/setup time
|
||||
reduces the sampling interval within each window. Although both stream requests
|
||||
specified the upstream 120 Hz setting, the raw stream delivered approximately
|
||||
101 Hz. Device counters are retained without assigning time units. Pose quaternion
|
||||
norms were within 0.00000007 of one. These measurements do not establish accuracy,
|
||||
drift, physical axis signs in a rendered scene, or 6DoF.
|
||||
|
||||
Raw gyro/accelerometer decoding is available for future work. The application uses
|
||||
only the device's pose stream. Combined pose/raw streaming is deliberately excluded
|
||||
because upstream observed acknowledgments without subsequent samples in that mode.
|
||||
|
||||
## Reproducing the checks
|
||||
|
||||
Build the application with `-PvitureSdkDir=` to explicitly exclude the optional
|
||||
vendor SDK. No vendor SDK download, vendor account, binary upload, or repository
|
||||
secret is required for Pro 2. See each tool's README/provenance document for Android
|
||||
SDK/JDK requirements, build commands, USB permission, and log collection.
|
||||
|
||||
The protocol probe is separate from the application driver. The
|
||||
`tools/viture/viture-v2-driver-check` harness packages the actual compiled
|
||||
`VitureV2Link`, `VitureV2Protocol`, and `GlassesLink` classes and records their hashes.
|
||||
It verifies readiness from poses and a stop/restart/stop/close sequence.
|
||||
|
||||
On the same Pro 2, the final production-driver check **passed**: 1,446 poses in the
|
||||
12-second resumed capture at 120.65 Hz, plus three bootstrap poses. It reported one
|
||||
ready callback, two tracking-on and two tracking-off callbacks, zero errors and
|
||||
zero bad frames. Both 500 ms pause windows delivered zero further poses. After
|
||||
close, readiness and tracking were both false. The complete sequence took 13.03 s.
|
||||
This validates the production USB driver; it does not by itself exercise the full
|
||||
DroidDeck UI or an in-game head-tracking session.
|
||||
|
||||
All 115 selected host-side tests passed. They cover wire framing/checksums,
|
||||
bounded parsing, pose and raw layouts,
|
||||
invalid samples, readiness without trusting ACKs, host-receipt cutoffs and
|
||||
duplicate/backward device-tick rejection,
|
||||
stop/restart behavior, transport failures, relative coordinate conversion, exact
|
||||
USB model routing, consumer removal before the first restarted pose, and UI hiding
|
||||
of unsupported hardware/display controls.
|
||||
|
||||
## Full application check
|
||||
|
||||
The complete debug APK built successfully with `:app:assembleDebug -PvitureSdkDir=`
|
||||
and passed the repository's runtime-input verification and APK signature checks.
|
||||
The APK contains the upstream MIT license and contains none of `libglasses.so`,
|
||||
`libcarina_vio.so`, `libcloud_protocol.so`, or `libdroiddeck_viture.so`.
|
||||
|
||||
Installed as `com.droiddeck.launcher.xr`, version 0.5.0, on the same phone, it
|
||||
recognized Pro 2 and exposed `supportsScreen3dof=true` after USB permission and
|
||||
valid bootstrap samples. A five-second sample through DroidDeck's own `glasses`
|
||||
agent endpoint returned **595 poses (118.98 Hz including startup)**, with quaternion
|
||||
norms within 0.00000006 of one and zero bad frames. Tracking was false before the
|
||||
sample, true during it, and false afterward; the connection remained ready with
|
||||
no control error. This checks application integration and consumer-driven streaming,
|
||||
not an in-game rendering or SteamVR session.
|
||||
|
||||
Tested APK SHA-256:
|
||||
`f5a2484ea1708a143cb94a96e084188cb0d2b122154a2645d3127c66f6d04f21`.
|
||||
|
||||
## User game test
|
||||
|
||||
After installing the tested build, the user reported that the Pro 2 game test
|
||||
passed. This adds user-reported in-game validation to the automated USB and
|
||||
application checks above. The game title, test duration, and separate recentering
|
||||
or SteamVR checks were not recorded.
|
||||
|
||||
## Local capture fingerprints
|
||||
|
||||
Full motion logs are local diagnostic artifacts, not required build inputs. These
|
||||
SHA-256 values identify the captures behind the table above:
|
||||
|
||||
| Capture | SHA-256 |
|
||||
| --- | --- |
|
||||
| `pro2-pose-initial.jsonl` | `beb09072d8d7ce2b1378c41992b88787e0b09f3ab3ea330a413b3ef9e9f5b762` |
|
||||
| `pro2-pose-motion.jsonl` | `d72f3bbd6564061e9bf4e06fec7f4c6dcdbfabeee626a99a7bff279b416ccc61` |
|
||||
| `pro2-raw.jsonl` | `944ac2fe30c7b1a817bc4d3281ead620000be32aa2dbd96e6b104cf4d7063091` |
|
||||
| `pro2-production-driver-final.jsonl` | `0f1c1073963a7c3052bf8bdbcb39e3ef8ec332367ab1a3e39b622d36fcd06db2` |
|
||||
| `pro2-full-app.json` | `3e48d9c3e8a230ec84e09bb51395b8566618406abb93937b8e2b69c40b4854c5` |
|
||||
|
||||
The provenance in [README.md](README.md) applies to all these checks: the upstream
|
||||
MIT project used SDK traffic captures in its research. This adaptation contains
|
||||
and loads no vendor SDK; it does not claim independent protocol discovery.
|
||||
Reference in new issue
Block a user