Compilation on the GPU?

ACM International Conference on Computing Frontiers (CF), pp. 230–236

Abstract

The emergence of highly parallel architectures has led to a renewed interest in parallel compilation. In particular, the widespread availability of GPU architectures raises the question whether compilation on the GPU is feasible. In this paper, we describe the first design and implementation of a parallel compiler from a simple imperative programming language to RISC-V machine code, that is fully executed on a GPU. To accomplish this, all stages from parsing to machine code generation were redesigned to exploit fine-grained parallelism. Experimental evaluation of the implemented prototype demonstrates our proposed parallel techniques to be effective and implementation of compilation on the GPU to be feasible. Finally, we propose a number of avenues for future work and hope to revitalize research into parallel compilation conducted in the 1980s.

Authors 3

1. Robin F. Voetter Aachen

   Leiden University

   Affiliation as printed

   Leiden University, Leiden, The Netherlands

2. Marcel Huijben Aachen

   Leiden University

   Affiliation as printed

   Leiden University, Leiden, The Netherlands

3. Kristian F. D. Rietveld Aachen

   Leiden University

   Affiliation as printed

   Leiden University, Leiden, The Netherlands