Skip to content

Latest commit

 

History

History
65 lines (54 loc) · 3.33 KB

README.md

File metadata and controls

65 lines (54 loc) · 3.33 KB
Raw

Assessing the suitability of Rust for performant and productive implementations of HPC codebases

A third year project in partial requirement for a Computer Science BSc at the University of Warwick, supervised by Dr Richard Kirk.

Abstract

Rust is a type-safe programming language originally developed by Mozilla in 2010. Since its first release, it has gained a large community of dedicated followers. Designed to be memory safe and have foundations in functional programming, it claims to be highly performant, safe, and provide a robust development toolchain.

The Mantevo Suite is a collection of mini-apps, largely written in C++ and FORTRAN. Mini-apps are small software codebases with performance characteristics representative of full-scale applications. They are traditionally used for hardware-software co-design, allowing simple but accurate estimates of application performance on hardware. However, they can also be used to assess the software stack they are run on, for example the language in which they are implemented.

This project assesses the suitability of Rust for performant and productive implementations of HPC codebases through the example of HPCCG, "the original Mantevo mini-app". We present a Rust implementation of HPCCG, showing the possibility of applying both shared and distributed memory parallelism in Rust to HPC codebases using Rayon and MPI, along with a workflow for translation including a novel approach for equivalence checking between Rust and C++. We further present a performance analysis within a novel framework empowering reproducibility, which empirically shows Rust closely approaches C++ in performance on real-world applications. We conclude that Rust is a viable language for High-Performance Computing applications, despite its performance not matching C++, as it provides commensurate benefits in developer productivity such as its compiler guarantees of memory and thread safety.

Keywords: High-Performance Computing, Parallel Computing, Mini-application, Mantevo, HPCCG, Rust, C++, OpenMP, Rayon, MPI

Download

This is a meta-repository of the written and software deliverables for this project, and as such contains nested submodules. To clone a copy containing all nested submodules, consider running the following command:

git clone --recurse-submodules -j8 https://github.com/EdmundGoodman/rust-in-hpc

Directory structure

Written project deliverables are typeset in LaTeX, and can be found in the submodule directories dissertation/, presentation/, progress-report/, and specification/

Software components are contained in the code/ directory, which has four subdirectories:

  • helper-scripts/, software written by the author to acheive corollary goals such as aggregating and selecting mini-apps, or proofs-of-concept for data races in OpenMP and MPI
  • hpc-multibench/, the main software component of the project, which contains many nested subdirectories
  • hpccg-rs/, a softlink to within hpc-multibench/ containing the Rust translations of the HPCCG mini-app
  • open-source/, forks of software projects which were not written by me, but I have interacted with in the form of pull requests or issues in the course of the project. These are not claimed to be my work.