Adding Support for Heterogeneous Parallel Architectures to Julia

We introduce basic terminology of GPU computing using the Heterogeneous Systems Architecture (HSA) as well as the high level structure of the Julia runtime and its foundation LLVM. We cover the steps we took to simplify using the HSA runtime library from within Julia. Building on that, we explain, in some detail, our approach for enabling Julia to compile designated parts of the program for HSA and running them on a compatible accelerator. We conclude with an evaluation of our work and thoughts on possible future development.

Autorentext

Georg's interests cover a broad range of topics. From GPU accelerated 3D object reconstruction and robot control to source-to-source transformation tools. His most recent research focuses on programming language development. Culminating in his thesis, it earned him a Master's degree in computer science.