Content area
Abstract
The Rust programming language seeks to unify the safety guarantees of higher-level languages with the performance of lower-level ones. One lauded safety feature of Rust is its claim of fearless concurrency for programmers, i.e., detecting concurrency errors at compile time. Prior work has studied Rust support for concurrency in open-source Rust system software and libraries. However, the interaction of Rust support and types of parallelism remains unexplored.
Does Rust always provide fearless concurrency? We answer this question by porting programs with abundant regular and irregular parallelism from C++ to Rust. With our resulting RPB suite, we classify parallelism types by their access patterns. Rust, with the Rayon library, grants parallel programmers fearlessness for regular parallelism. However, for applications with any irregular parallelism, the programmer must choose between unsafe code or high-overhead dynamic checks with errors that manifest at run time, leaving parallel programming as challenging with Rust as with its predecessors.
Details
