Scientific Computing in Rust 2023

, ,

Last week I attended the Scientific Computing in Rust 2023 virtual conference. The organizers put together a well-stirred mixture of talks covering a wide range of topics. It was good to see so many other enthusiastic researchers using Rust as their main tool to build software surrounding computational needs.

The invited speakers Gonzalo Brito, Sarah El Kazdadi and Manuel Drehwald had particularly interesting talks. Gonzalo presented a Lattice-Boltzmann CFD solver as an application of rayon, Rusts multithreading library. Sarah presented faer-rs, an implementation of low-dimensional linear algebra routines which could in the future be used as a backend to popular crates such as nalgebra or ndarray. Manuel’s talk evolved around a heavily modified version of the rust compiler called rust-enzyme to bring auto-differentiation to the Rust language.

A common problem and question of many attendees was regarding GPU-specific functionality and libraries. At the moment, there does not seem to be a unified crate which the community has agreed upon to use or that even offers every feature desirable. wgpu aims to provide a useful graphics API which can also be exploited for numerical computations. rust-cuda and rust-gpu are two other crates in this space. While wgpu and rust-gpu compile native Rust code to SPIR-V, rust-cuda provides direct access to low-level bindings of the CUDA framework by NVIDIA. Similarly, there exist libraries functioning as wrappers for popular C++ graphics libraries such as arrayfire-rust. For the next conference, I am looking forward to more interesting talks tackling this problem.

All together, I can only say positive things about the event and I hope to be attending again in the near future.

Image Credit: scientificcomputing.rs/ferris

Leave a Reply

Your email address will not be published. Required fields are marked *

Categories