# Efficient Winograd or Cook-Toom Convolution Kernel Implementation on Widely Used Mobile CPUs

## TL;DR

This paper presents optimized implementation strategies for Winograd and Cook-Toom convolution algorithms on mobile ARM Cortex-A CPUs, achieving up to 60% faster inference in CNNs by leveraging SIMD instructions and resource utilization improvements.

## Contribution

It introduces efficient implementation techniques for Winograd and Cook-Toom convolutions on embedded ARM CPUs, addressing a gap in optimized CNN deployment on mobile devices.

## Key findings

- Up to 60% inference latency reduction compared to existing methods
- Effective utilization of ARMv8-A NEON SIMD instructions
- Region-wise multi-channel implementation improves performance

## Abstract

The Winograd or Cook-Toom class of algorithms help to reduce the overall compute complexity of many modern deep convolutional neural networks (CNNs). Although there has been a lot of research done on model and algorithmic optimization of CNN, little attention has been paid to the efficient implementation of these algorithms on embedded CPUs, which usually have very limited memory and low power budget. This paper aims to fill this gap and focuses on the efficient implementation of Winograd or Cook-Toom based convolution on modern Arm Cortex-A CPUs, widely used in mobile devices today. Specifically, we demonstrate a reduction in inference latency by using a set of optimization strategies that improve the utilization of computational resources, and by effectively leveraging the ARMv8-A NEON SIMD instruction set. We evaluated our proposed region-wise multi-channel implementations on Arm Cortex-A73 platform using several representative CNNs. The results show significant performance improvements in full network, up to 60%, over existing im2row/im2col based optimization techniques