一句话：提出为 NPU 定制的地址转换 MMU，解耦虚拟/物理地址并仅带来极低性能开销。
随着 DNN 对计算与内存需求增长，NPU 正被广泛用于加速，并有望像 GPU 一样成为异构体系结构中的一等处理器。本文主张在 NPU 中支持地址转换，以解耦虚拟与物理内存地址空间。作者通过数据驱动的应用刻画研究，根因分析了既有以 GPU 为中心的地址转换方案的若干局限，并据此提出一种面向 NPU 定制的内存管理单元（MMU）。评估表明，与理想的 oracle MMU 设计点相比，该方案的性能开销平均仅为 0.06%。
To satisfy the compute and memory demands of deep neural networks (DNNs), neural processing units (NPUs) are widely being utilized for accelerating DNNs. Similar to how GPUs have evolved from a slave device into a mainstream processor architecture, it is likely that NPUs will become first-class citizens in this fast-evolving heterogeneous architecture space. This paper makes a case for enabling address translation in NPUs to decouple the virtual and physical memory address space. Through a careful data-driven application characterization study, we root-cause several limitations of prior GPU-centric address translation schemes and propose a memory management unit (MMU) that is tailored for NPUs. Compared to an oracular MMU design point, our proposal incurs only an average 0.06% performance overhead.
DOI 原文 ·
@inproceedings{paperbot5413,
title = {NeuMMU},
author = {Bongjoon Hyun and Youngeun Kwon and Yujeong Choi and John Kim and Minsoo Rhu},
booktitle = {ASPLOS},
year = {2020},
doi = {10.1145/3373376.3378494}
}