Efficient winograd-based convolution kernel implementation on edge devices

Xygkis, Athanasios; Soudris, Dimitrios; Papadopoulos, Lazaros; Yous, Sofiane; Moloney, David

doi:10.1145/3195970.3196041

Cited by 15 publications

(10 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Winograd convolution. It is an efficient algorithm to reduce multipliers for CONV3×3 and recently shows advantages on GPU [34], FPGA [38], and embedded processor [58]. However, it increases 23.5% of area in our case because the overheads of long internal bitwidths and additional pre-/post-processing become significant for our 8-bit implementation.…”

Section: Related Workmentioning

confidence: 97%

eCNN

Huang

Ding

Wang

et al. 2019

Proceedings of the 52nd Annual IEEE/ACM International Symposium on Microarchitecture

View full text Add to dashboard Cite

Convolutional neural networks (CNNs) have recently demonstrated superior quality for computational imaging applications. Therefore, they have great potential to revolutionize the image pipelines on cameras and displays. However, it is difficult for conventional CNN accelerators to support ultra-high-resolution videos at the edge due to their considerable DRAM bandwidth and power consumption. Therefore, finding a further memory-and computation-efficient microarchitecture is crucial to speed up this coming revolution.In this paper, we approach this goal by considering the inference flow, network model, instruction set, and processor design jointly to optimize hardware performance and image quality. We apply a block-based inference flow which can eliminate all the DRAM bandwidth for feature maps and accordingly propose a hardwareoriented network model, ERNet, to optimize image quality based on hardware constraints. Then we devise a coarse-grained instruction set architecture, FBISA, to support power-hungry convolution by massive parallelism. Finally, we implement an embedded processor-eCNN-which accommodates to ERNet and FBISA with a flexible processing architecture. Layout results show that it can support high-quality ERNets for super-resolution and denoising at up to 4K Ultra-HD 30 fps while using only DDR-400 and consuming 6.94W on average. By comparison, the state-of-the-art Diffy uses dualchannel DDR3-2133 and consumes 54.3W to support lower-quality VDSR at Full HD 30 fps. Lastly, we will also present application examples of high-performance style transfer and object recognition to demonstrate the flexibility of eCNN.

show abstract

Section: Related Workmentioning

confidence: 97%

eCNN

Huang

Ding

Wang

et al. 2019

Proceedings of the 52nd Annual IEEE/ACM International Symposium on Microarchitecture

View full text Add to dashboard Cite

show abstract

“…The main goal of the SDK4ED project is to minimize cost, development time, and complexity of lowenergy software development processes, by providing a set of innovative solutions (i.e., toolboxes) integrated into the form of an easy-to-use platform for automatic optimization and trade-off calculation among important design-time and run-time software quality attributes. More specifically, the outcome of the project so far showcases numerous research and technical achievements with respect to the optimization of the targeted quality attributes, namely Maintainability [3,5,16,37,50], Dependability [32,33,[51][52][53][54], Energy Consumption [23,24,42,67], as well as Quality Forecasting [60,61,63] and Decision Support [47,[55][56][57] throughout the overall software development cycle. The SDK4ED TD Forecasting tool, integrated into the TD Management (TDM) framework, aims to provide predictive forecasts regarding the evolution of the TD quality attribute.…”

Section: Rq : Can Data Clustering Improve the Accuracy Of Crossproject Td Forecasting?mentioning

confidence: 99%

A Clustering Approach Towards Cross-Project Technical Debt Forecasting

et al. 2021

View full text Add to dashboard Cite

Technical debt (TD) describes quality compromises that can yield short-term benefits but may negatively affect the quality of software products in the long run. A wide range of tools and techniques have been introduced over the years in order for the developers to be able to determine and manage TD. However, being able to also predict its future evolution is of equal importance to avoid its accumulation, and, in turn, the unlikely event of making the project unmaintainable. Although recent research endeavors have showcased the feasibility of building accurate project-specific TD forecasting models, there is a gap in the field regarding cross-project TD forecasting. Cross-project TD forecasting is of practical importance, since it would enable the application of pre-existing forecasting models on previously unknown software projects, especially new projects that do not exhibit sufficient commit history to enable the construction of project-specific models. To this end, in the present paper, we focus on cross-project TD forecasting, and we examine whether the consideration of similarities between software projects could be the key for more accurate forecasting. More specifically, we propose an approach based on data clustering. In fact, a relatively large repository of software projects is divided into clusters of similar projects with respect to their TD aspects, and specific TD forecasting models are built for each cluster, using regression algorithms. According to our approach, previously unknown software projects are assigned to one of the defined clusters and the cluster-specific TD forecasting model is applied to predict future TD values. The approach was evaluated through several experiments based on real-world applications. The results of the analysis suggest that the proposed approach comprises a promising solution for accurate cross-project TD forecasting.

show abstract

“…[40], [86] utilized the SIMT architecture of GPGPU to process CNN using Winograd convolution in parallel, and [40] also added support for Zero-Skip. [87], [88] used high-efficiency Winograd convolution on IoT devices to achieve high performance. [41], [89], [90] used random calculation and approximate calculation to complete the implementation.…”

Section: Cpumentioning

confidence: 99%

Fast Convolution based on Winograd Minimum Filtering: Introduction and Development

Gan¹,

Huang²

2021

Computer Science and Information Technology Trends

View full text Add to dashboard Cite

Convolutional Neural Network (CNN) has been widely used in various fields and played an important role. Convolution operators are the fundamental component of convolutional neural networks, and it is also the most time-consuming part of network training and inference. In recent years, researchers have proposed several fast convolution algorithms including FFT and Winograd. Among them, Winograd convolution significantly reduces the multiplication operations in convolution, and it also takes up less memory space than FFT convolution. Therefore, Winograd convolution has quickly become the first choice for fast convolution implementation within a few years. At present, there is no systematic summary of the convolution algorithm. This article aims to fill this gap and provide detailed references for follow-up researchers. This article summarizes the development of Winograd convolution from the three aspects of algorithm expansion, algorithm optimization, implementation, and application, and finally makes a simple outlook on the possible future directions.

show abstract

Efficient winograd-based convolution kernel implementation on edge devices

Cited by 15 publications

References 6 publications

eCNN

eCNN

A Clustering Approach Towards Cross-Project Technical Debt Forecasting

Fast Convolution based on Winograd Minimum Filtering: Introduction and Development

Contact Info

Product

Resources

About