Native Wayland/X11 playback and NCNN/Vulkan upscaling

[sirok1]

I added full Linux x86_64 support with native Wayland/X11 video rendering, cross-vendor NCNN/Vulkan AI upscaling, and release packaging while preserving the existing Windows behavior.

Linux Playback

• Added native Wayland and X11 support without requiring XWayland.
• Video is rendered through mpv_render_context into a GTK GLArea.
• WebKitGTK and the native video surface are arranged using GtkOverlay.
• Added HiDPI-aware framebuffer sizing and OpenGL render context cleanup.
• Fixed absolute Linux paths passed through the command line.
• Initial media loading is deferred until the OpenGL render context has been created.
• Added PipeWire, PulseAudio, and ALSA audio output fallbacks.
• Linux user data follows the XDG Base Directory specification:
  • $XDG_CONFIG_HOME/l-mpv;
  • ~/.config/l-mpv by default.
• MediaInfo uses the CLI implementation on Linux.
• Track extraction uses the bundled or system FFmpeg binary.
• Windows-only features are hidden or disabled on Linux:
  • WASAPI audio visualizer;
  • DirectML and TensorRT;
  • portable executable auto-updates;
  • Windows registry and Explorer integration.

NCNN/Vulkan AI Upscaling

• Added an NCNN Vulkan backend for NVIDIA, AMD, and Intel GPUs.
• Added support for preconverted NCNN model pairs:
  • *.param;
  • *.bin.
• Linux models are stored in ~/.config/l-mpv/models/ncnn/.
• Added an in-app download for the curated AnimeJaNai V2 model with SHA-256 verification.
• Added a custom VapourSynth API 4 plugin with tiled NCNN/Vulkan inference.
• Hardware decoding is temporarily disabled while the AI filter is active because VapourSynth currently transfers frames through system memory.
• Added a dropped-frame watchdog. When NCNN cannot maintain smooth playback, it:
  • removes the NCNN filter;
  • enables the packaged FSRCNNX/Anime4K shaders;
  • displays an OSD notification.

Private Linux Runtime

Added a reproducible private runtime build with pinned revisions of:

• mpv 0.41;
• VapourSynth R79;
• NCNN with Vulkan support;
• a minimal bundled Python runtime;
• FSRCNNX and Anime4K GLSL shaders.

The runtime is isolated from the system mpv installation and loaded dynamically by the application.

Packaging and Releases

Added:

• a self-contained AppImage build;
• an Arch Linux PKGBUILD;
• .SRCINFO;
• a desktop entry with MIME types;
• GitHub Actions release jobs for:
  • building an AppImage on Ubuntu 22.04;
  • building an Arch Linux pkg.tar.zst package.

Generated AppImage and Arch package artifacts are excluded from Git.

UI Changes

• Added runtime platform detection.
• Exposed the NCNN upscaling settings on Linux.
• Hidden settings that are only available on Windows.
• Updated the backend selector, model types, and runtime status reporting.
• Kept controls accessible on X11 to account for native child-window compositing and pointer-event limitations.

Documentation

The README now documents:

• Windows and Linux feature differences;
• the native Wayland/X11 rendering architecture;
• AppImage and Arch Linux installation;
• NCNN models and automatic shader fallback;
• XDG user-data paths;
• private runtime and package build instructions;
• the updated project structure.

Compatibility

Platform	Video Rendering	AI Backend
Windows 10/11 x64	HWND + D3D11	DirectML / TensorRT
Linux Wayland	GTK GLArea + OpenGL/EGL	NCNN/Vulkan
Linux X11	GTK GLArea + OpenGL	NCNN/Vulkan

The existing Windows implementation and portable behavior remain unchanged.

Verification

• Frontend production build
• Rust release build
• All 7 Rust unit tests
• Shell syntax validation
• GitHub Actions workflow YAML validation
• git diff --check
• AppImage launch
• Extracted Arch package launch
• Native Wayland video playback
• Native X11 video playback
• NCNN plugin and AnimeJaNai V2 loading
• Vulkan device detection on Intel Iris Xe and NVIDIA RTX 3050

Additional Testing Needed

• Benchmark 1080p → 4K upscaling at 24 FPS across representative GPUs.
• Force dropped frames during real playback and verify the visible watchdog fallback.
• Perform deeper resize, fullscreen, drag-and-drop, transparency, and HiDPI testing.
• Inspect the final AppImage produced by the Ubuntu 22.04 release workflow.

Notes

The AppImage bundles the application libraries and private AI runtime, but it does not include GPU drivers. NCNN/Vulkan upscaling requires a working Vulkan ICD installed on the host system. Also, after the merge, I’d be happy to help with publishing it to the AUR.