docs: reconcile P2 optimization and Linux validation
This commit is contained in:
@@ -2,7 +2,7 @@
|
||||
|
||||
Faset is an independent engine project for desktop **2D and 3D games on Linux and Windows**. Its priorities are a custom editor that is comfortable to use by hand and through MCP, integration with Blender, and a path toward advanced graphics.
|
||||
|
||||
**Current status: the C++ MVP is accepted for the recorded Linux and Windows test profiles; Lua and P2 GPU visibility were added afterward.** The MVP includes the native Editor, shared GUI/MCP authoring, gameplay builds, Vulkan Player, Blender import and standalone export. Both playable games passed Release export and relocated execution on both operating systems. P2 adds optional GPU frustum and two-pass HZB occlusion modes, fixed indirect mesh bins, prepared mesh LOD selection, and an Editor HZB diagnostic view. The direct renderer remains the default and comparison reference. P2 GPU acceptance and measurements cover the Linux reference device; Windows P2 and additional physical GPUs need separate validation. The first Debug/validation benchmark found GPU culling substantially slower than direct GPU work, so the optional modes are not presented as a performance win. Windows MVP graphics acceptance used software Vulkan. See the [MVP acceptance dossier](docs/validation/mvp-acceptance.md), [P2 acceptance protocol](docs/studies/19-p2-gpu-visibility-acceptance.md), [measured P2 results](docs/studies/20-p2-gpu-visibility-benchmark-2026-09-23.md), and [implementation checkpoints](docs/IMPLEMENTATION.md) for exact scope and limits.
|
||||
**Current status: the C++ MVP is accepted for the recorded Linux and Windows test profiles; Lua and P2 GPU visibility were added afterward.** The MVP includes the native Editor, shared GUI/MCP authoring, gameplay builds, Vulkan Player, Blender import and standalone export. Both playable games passed Release export and relocated execution on both operating systems. P2 adds optional GPU frustum and two-pass HZB occlusion modes, fixed indirect mesh bins, prepared mesh LOD selection, and an Editor HZB diagnostic view. Direct remains the default and comparison reference; an exported Player can explicitly select a mode with `--visibility direct|gpu-frustum|gpu-occlusion`. P2 GPU acceptance and measurements cover the Linux reference device, with a separate software-Vulkan functional check; Windows P2 and additional physical GPUs need separate validation. The first Debug/validation benchmark exposed slow GPU culling; device-local outputs and a bounded atomic append improved the same synthetic workloads in the repeat. Neither run establishes a general game-speed benefit. Windows MVP graphics acceptance used software Vulkan. See the [MVP acceptance dossier](docs/validation/mvp-acceptance.md), [P2 Linux evidence](docs/validation/p2-gpu-visibility-2026-09-23/README.md), [P2 acceptance protocol](docs/studies/19-p2-gpu-visibility-acceptance.md), [initial benchmark](docs/studies/20-p2-gpu-visibility-benchmark-2026-09-23.md), [optimized benchmark](docs/studies/21-p2-gpu-visibility-optimization-2026-09-23.md), and [implementation checkpoints](docs/IMPLEMENTATION.md) for exact scope and limits.
|
||||
|
||||
## Start here
|
||||
|
||||
|
||||
Reference in New Issue
Block a user