ConvBench：用于2D卷积原语评估的综合基准
摘要
卷积是卷积神经网络(CNN)中核心计算密集型操作，导致许多高性能算法的发展，如Im2col-GEMM、Winograd和Direct-Convolution。然而，比较不同卷积算法是一项容易出错的任务，因为它需要特定的数据布局和系统资源。未能满足这些要求可能导致不必要的延迟。因此，考虑卷积算法中所有处理步骤对于全面评估和公正比较其性能至关重要。此外，已知大多数卷积基准采用有限覆盖和手工操作的临时测试套件。本文提出ConvBench，用于卷积算法的评估和比较。它评估了从1097个真实世界深度学习模型中派生的9243个卷积运算， resulting in performance and execution breakdown graphs for detailed evaluation。ConvBench capability is evaluated across the Sliced Convolution (SConv) algorithm. The experiments showed results faster than Im2col-GEMM in 93.6% of the convolutions. However, the use of ConvBench allowed the delving into the remaining 6.4% underperforming convolutions, uncovering a critical slowdown of 79.5% on average of SConv's packing step. This analysis underscores a potential source of optimization for SConv, opening up new paths for convolution designers to improve their algorithms.
关键词
引用
@article{arxiv.2407.10730,
title = {ConvBench: A Comprehensive Benchmark for 2D Convolution Primitive Evaluation},
author = {Lucas Alvarenga and Victor Ferrari and Rafael Souza and Marcio Pereira and Guido Araujo},
journal= {arXiv preprint arXiv:2407.10730},
year = {2024}
}
备注
5 pages, 3 figures, presented on MLArchSys workshop of ISCA'2024