A Case Study: Exploiting Neural Machine Translation to Translate CUDA to OpenCL

TL;DR

This paper demonstrates how a seq2seq neural machine translation model can be used to translate CUDA programs into OpenCL, involving specific data preparation and processing steps, validated through a case study on benchmark suites.

Contribution

It introduces a detailed methodology for applying seq2seq models to translate CUDA to OpenCL, including input set generation and processing techniques.

Findings

Successful translation of CUDA to OpenCL in case studies

Effective training input generation method

Potential for automated code translation between GPU programming languages

Abstract

The sequence-to-sequence (seq2seq) model for neural machine translation has significantly improved the accuracy of language translation. There have been new efforts to use this seq2seq model for program language translation or program comparisons. In this work, we present the detailed steps of using a seq2seq model to translate CUDA programs to OpenCL programs, which both have very similar programming styles. Our work shows (i) a training input set generation method, (ii) pre/post processing, and (iii) a case study using Polybench-gpu-1.0, NVIDIA SDK, and Rodinia benchmarks.