Preprint / Version 1

Proposal for offloading large-grained calculation using software engineering technology

##article.authors##

  • Yoji Yamato Network Service Systems Labs., NTT, Inc.

DOI:

https://doi.org/10.31224/8870

Keywords:

Environment-Adaptive Software, Automatic Offloading, GPGPU, Functional Block, Code Clone Detection

Abstract

We propose environment-adaptive software that automatically converts code written by programmers for typical CPUs according to the operating environment to perform high-performance processing. This paper focuses on the automatic offloading of functional blocks that perform computational processing such as discrete 3D Fourier transform and MRI image processing. Existing applications to be offloaded are analyzed using machine learning-based code detection to search for offloadable parts, and then analyzed using abstract syntax tree-based code detection to process the offloaded parts with a high-speed implementation and perform the offloading. The ability to automatically offload using the proposed method is confirmed by measuring the processing time using an actual NVIDIA GeForce 3090 GPU.

IEICE Technical Report, ICM2026-18, Nov. 2026, Copyright(C)2026 IEICE

Downloads

Download data is not yet available.

Downloads

Posted

2026-11-20