/MarkovJuniorWeb

Typescript version of MarkovJunior, runs on the web (with pathtracer)

Primary LanguageTypeScriptMIT LicenseMIT

MarkovJuniorWeb

Typescript version of MarkovJunior, runs in browser (also in node.js).

  • Everything have been implemented including isometric rendering, exporting the output as a .vox file, and node tree visualization.
  • Every model from the original repository can be loaded with this project, but the output would be different due to different random seed implementation (dotnet builtin vs seededrandom).

demo RTX=on

Development

Install dependensies:

npm i

Start development server on localhost:

npm start

Build static site:

npm build

Run in node (writes result to /output):

npm run cli

TODO

  • node editor
  • import from & export to xml
  • voxel editor
  • file system on the web
  • optimization

Performance

This port is around 2x slower than the original repo (JS vs C#), but it doesn't affect the page much; even with 200 steps per frame there's hardly any FPS drop on most models. However, the slowdown is quite noticable on computation expensive calculations, e.g. uni/bi-direction inference.

Random Notes

SokobanLevel1 takes ~10 seconds for the original C# code on my pc to reach the desired state, while it takes 20+ seconds on the web. I've tried JIT/unroll the rules into webassembly with generated AssemblyScript and it actually works: it gains a x2 speedup and the performance almost match the native C# version. The only problem is the load & compile time is terrilbe and it's incredibly hard to debug WebAssembly. I rolled back the commits on main and put the experimental stuff in the optimization branch, but I'm still pretty proud of this MarkovJunior rules -> AssemblyScript -> Wasm JIT compiler I wrote.

Update: I wrote a precompiled wasm version and it works fine, and the runtime is reduced from 20+ seconds on SokobanLevel1 to ~13 seconds (not too bad I guess ¯\_(ツ)_/¯ ).