Open AccessA method of accelerating CNN computation on DSP
Author(s) -
Bin Li,
Luo Ying-guang,
Luo Guo-bin Fu,
Peng Ye,
Jia Hu
Publication year - 2020
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1684/1/012063
Subject(s) - digital signal processing , computer science , convolution (computer science) , computation , multiplication (music) , parallelism (grammar) , parallel computing , algorithm , computer engineering , computer hardware , computational science , artificial intelligence , mathematics , artificial neural network , combinatorics
Because convolution network contains a lot of multiplication and addition operations, there is still room for optimization after it is implemented on the hardware platform. This paper mainly studies the method of accelerating the deep learning algorithm on DSP, proposes to improve the convolution layer calculation by using Winograd algorithm, and implements it on GPDSP. The final results show that it can increase the parallelism of data calculation and significantly reduce the computing time.