Rust's Standard Library on the GPU

Summary

At VectorWare, we are building the first GPU-native software company. Today, we are excited to announce that we can successfully use Rust's standard library from GPUs. This milestone marks a significant step towards our vision of enabling developers to write complex, high-performance applications that leverage the full power of GPU hardware using familiar Rust abstractions. This post is a preview of what we've built. We're preparing our work for potential upstreaming and will share deeper technical details in future posts. Rust's std library Rust's standard library is organized as a set of layered abstractions: core defines the language's foundation and assumes neither a heap nor an operating system. alloc builds on core by adding heap allocation. std sits at the top of core and alloc. It adds APIs for operating system concerns such as files, networking, threads, and processes. A defining feature of Rust is that layers 2 and 3 are optional. Code can opt out of std via the #![no_std] annotation, relying only on core and, when needed, alloc. This makes Rust usable in domains such as embedded, firmware, and drivers, which lack a traditional operating system. Why std isn't supported on GPUs today We are the maintainers of rust-cuda and rust-gpu, open source projects that enable Rust code to run on the GPU. When targeting the GPU with these projects, Rust code is compiled with #![no_std] as GPUs do not have operating systems and therefore cannot support std. Because #![no_std] is part of the Rust language, #![no_std] libraries on crates.io written for other purposes can generally run on the GPU without modification. This ability to reuse existing open source libraries is much better than what exists in other (non-Rust) GPU ecosystems. Still, there is a cost to opting out of std. Many of Rust's most useful and ergonomic abstractions live in the standard library and the majority of open source libraries assume std. Enabling meaningful std support on GPUs unlocks a much lar...