Open bnolan opened 9 years ago
+1
@schteppe how do you feel about branching the codebase so that you can accept PRs & run some benchmarks while this is a draft specification ? (without adversely affecting the current stable build)
Sounds cool! I made a new branch for SIMD.js stuff: https://github.com/schteppe/cannon.js/tree/simd
Some thoughts from me:
We can use a polyfill to emulate SIMD in non-compatible browsers (hopefully if it's implemented well using float32arrays it'll still be just as fast), and it might be helpful porting some of cannon to use array buffers anyway!
I think we should endeavour to not change the API at all, the cannon api is great fun to use. :)
With simd.js support getting closer to released browsers, it might be worth investigating what parts of cannon can benefit from SIMD instructions.
https://blog.mozilla.org/javascript/2015/03/10/state-of-simd-js-performance-in-firefox/