heaust/yawn
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Build Rust render graph core

Browse Source 
Keep hot render state in a generic shared SOA arena and use worker messages only for allocation, graph compilation, loadout changes, and render pacing. Compile external S-expression graphs into upfront WebGPU resources, transient aliases, and reusable render bundles.

Amp-Thread-ID: https://ampcode.com/threads/T-01a01380-b478-77d0-84a0-102880a5c5ae
Co-authored-by: Heaust Azure <heaust.azure@gmail.com>
This commit is contained in:
Amp
2026-08-20 05:35:24 +00:00
co-authored by heaust
parent d095ee7849
commit 239b1d25b0
17 changed files with 2697 additions and 667 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+5 -5
View File
@@ -1,14 +1,14 @@
# Yawn
Yawn Core is two things: a generic structure-of-arrays arena in a `SharedArrayBuffer`, and a render-graph worker that turns externally supplied WGSL into an up-front WebGPU loadout.
Yawn Core is two things: a generic structure-of-arrays arena in a `SharedArrayBuffer`, and a Rust render graph that turns externally supplied WGSL into an up-front WebGPU loadout.
```text
JSO or FXNode → AST → S-expression → graph worker → WebGPU
↑
any thread → direct shared row writes ┘
JSO or FXNode → AST → S-expression → worker messages → Rust/WebGPU
↑
any thread → direct shared row writes ─────────────┘
```
Messages allocate `{ name, rows, stride, format }` arrays and load graphs. Existing render data is changed by writing `f32`, `u32`, or `i32` rows directly. Allocations are 64-byte aligned, row strides are multiples of 16 bytes, and compatible non-overlapping transient textures share physical allocations.
The arena starts with only one eight-float `info` row for frame timing and direct SAB render skipping. Messages create or delete other `{ name, rows, stride, format }` arrays, allocate named slots, compile graphs, switch loadouts, and control render pacing. Existing render data is changed by writing `f32`, `u32`, or `i32` rows directly. Allocations are 64-byte aligned, row strides are multiples of 16 bytes, and compatible non-overlapping transient textures share physical allocations.
WGSL, pipelines, glTF import, and conventional mesh/camera/material handles live in `addons/`; core contains no shader or scene model.