[Question] Performance: manifold3d python

[AlbertoVPersonal]

Hi,

I would like to know whether the team has some reports with performance results using mainfold3d Python package.
We are using the JS library and the performance is not good as we need.
And we are thinking to move the code to Python or Rust.

Regards,
Alberto

[pca006132]

A native build would be faster, and it can use multiple threads. For general performance, it depends on how you use the library, we cannot comment on this without looking at your use case, and how fast you need it to be.

[pca006132]

We optimize on several consecutive mesh boolean operations, but to benefit from that you have to avoid frequent conversion to/from meshes. See https://github.com/elalish/manifold/wiki/Performance-Considerations#users-related

And what level of performance do you want? How much faster than wasm? Are you doing many small operations (e.g. expecting 10k operations in 1s) or just a few larger operations (10 slow operations in 1s)? Native will speed up those larger operations due to parallelization, I would not expect small operations to benefit significantly.

[AlbertoVPersonal]

I think that my colleague will reply with the details. I will talk with him tomorrow.

[elalish]

Also #1079 hasn't made it to an npm release yet, but that parallelism in WASM could have a significant speedup as well. You may want to try building that locally - I'd love to know what effect is has on performance for your use case.

[AlbertoVPersonal]

Ok. I will talk with him about your comment. Thanks!

[pca006132]

Note that #1079 is not really stable, we still encounter weird stack issue that is impossible to debug due to lack of sanitizer support in emscripten. And regular builds don't encounter any stack issues...