一句话：提出强制分裂与确定性终止技术，并构建StreamGrid框架，实现点云全流式处理，降低片上内存与能耗。
点云在智能应用中日益重要，但加速器中频繁的片外内存访问会造成流水线停顿和高能耗。传统行缓冲技术虽能消除片外流量，却因点云固有的计算模式而无法直接适用。为此，作者提出两种技术：强制分裂和确定性终止，从而实现完全流式处理；并进一步构建StreamGrid框架，整合这些技术并自动优化片上缓冲大小。评估表明，与未采用这些技术的基线相比，StreamGrid在精度损失很小的情况下减少61.3%的片上内存和40.5%的能耗；相较最先进加速器，还实现了10.0倍加速和3.9倍能效提升。
Point clouds are increasingly important in intelligent applications, but frequent off-chip memory traffic in accelerators causes pipeline stalls and leads to high energy consumption. While conventional line buffer techniques can eliminate off-chip traffic, they cannot be directly applied to point clouds due to their inherent computation patterns. To address this, we introduce two techniques: compulsory splitting and deterministic termination, enabling fully-streaming processing. We further propose StreamGrid, a framework that integrates these techniques and automatically optimizes on-chip buffer sizes. Our evaluation shows StreamGrid reduces on-chip memory by 61.3% and energy consumption by 40.5% with marginal accuracy loss compared to the baselines without our techniques. Additionally, we achieve 10.0× speedup and 3.9× energy efficiency over state-of-the-art accelerators.
@inproceedings{FengLL0LGHZ025,
title = {<scp>StreamGrid:</scp>
Streaming Point Cloud Analytics via Compulsory Splitting and Deterministic Termination},
author = {Yu Feng and Zheng Liu and Weikai Lin and Zihan Liu and Jingwen Leng and Minyi Guo and Zhezhi He and Jieru Zhao and Yuhao Zhu},
booktitle = {ASPLOS},
year = {2025},
doi = {10.1145/3676641.3716021}
}