Proceeding of the Faculty of technical Sciences, Novi Sad

Electrotechnical and Computer Engineering

Vol. 38 No. 12 (2023): Proceedings of the Faculty of Technical Sciences

DEVELOPMENT OF PYTHON BINDINGS, LIBRARY AND APPLICATION FOR A HARDWARE CNN ACCELERATOR

  • Nebojša Pilipović
DOI:
https://doi.org/10.24867/25BE28Pilipovic
Submitted
September 14, 2023
Published
2023-12-06

Abstract

This paper presents a development process of the required software stack, and the application, for the control of a hardware neural network accelerator, based on a modified CoNNa architecture. The existing C++ library was modified, a Python interface to it was developed, and lastly the user application itself was developed, with the ability to run multiple networks concurrently, on an arbitrary number of available cores.

References

[1] R. Struharik, B. Vukobratović, A. Erdeljan, i D. Rakanović, „Conna–compressed cnn hardware accelerator“, u 2018 21st Euromicro Conference on Digital System Design (DSD), IEEE, 2018, str. 365–372.
[2] „ctypes — A foreign function library for Python“, Python documentation. https://docs.python.org/3/library/ctypes.html (pristupljeno 30. avgust 2023.).
[3] F. Doglio, Mastering Python High Performance. Packt Publishing, 2015.