OpenCNN: A Winograd Minimal Filtering Algorithm Implementation in CUDA

被引：8
作者：
Castro, Roberto L. [1 ]
Andrade, Diego [1 ]
Fraguela, Basilio B. [1 ]
机构：
[1] Univ A Coruna, Ctr Invest CITIC, Campus Elvina, La Coruna 15071, Spain
来源：
MATHEMATICS | 2021年 / 9卷 / 17期
关键词：
deep learning; convolution; Winograd; CUDA; CONVOLUTION;
D O I：
10.3390/math9172033
中图分类号：
O1 [数学];
学科分类号：
0701 ; 070101 ;
摘要：
Improving the performance of the convolution operation has become a key target for High Performance Computing (HPC) developers due to its prevalence in deep learning applied mainly to video processing. The improvement is being pushed by algorithmic and implementation innovations. Algorithmically, the convolution can be solved as it is mathematically enunciated, but other methods allow to transform it into a Fast Fourier Transform (FFT) or a GEneral Matrix Multiplication (GEMM). In this latter group, the Winograd algorithm is a state-of-the-art variant that is specially suitable for smaller convolutions. In this paper, we present openCNN, an optimized CUDA C++ implementation of the Winograd convolution algorithm. Our approach achieves speedups of up to 1.76x on Turing RTX 2080Ti and up to 1.85x on Ampere RTX 3090 with respect to Winograd convolution in cuDNN 8.2.0. OpenCNN is released as open-source software.
引用
收藏
页数：19
相关论文
共 27 条
[11]  
Jia Z., 2019, ARXIV190307486
[12]   Optimizing N-Dimensional, Winograd-Based Convolution for Manycore CPUs [J].
Jia, Zhen ;
Zlateski, Aleksandar ;
Durand, Fredo ;
Li, Kai .
ACM SIGPLAN NOTICES, 2018, 53 (01) :109-123
[13]  
Khan J., 2019, ARXIV191000078
[14]   Fast Algorithms for Convolutional Neural Networks [J].
Lavin, Andrew ;
Gray, Scott .
2016 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2016, :4013-4021
[15]  
Long J, 2015, PROC CVPR IEEE, P3431, DOI 10.1109/CVPR.2015.7298965
[16]  
Mathieu M., 2014, 2 INT C LEARN REPR I, P1
[17]  
Nervana Systems, NEON
[18]  
NVIDIA, CUTL
[19]  
NVIDIA, 2018, NVIDIA Turing GPU Architecture
[20]  
NVIDIA, CUDA C++ programming guide
123