StreamGrid: Streaming Point Cloud Analytics via Compulsory Splitting and Deterministic Termination

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Feng, Yu, Liu, Zheng, Lin, Weikai, Liu, Zihan, Leng, Jingwen, Guo, Minyi, He, Zhezhi, Zhao, Jieru, Zhu, Yuhao
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913918479237120
author Feng, Yu
Liu, Zheng
Lin, Weikai
Liu, Zihan
Leng, Jingwen
Guo, Minyi
He, Zhezhi
Zhao, Jieru
Zhu, Yuhao
author_facet Feng, Yu
Liu, Zheng
Lin, Weikai
Liu, Zihan
Leng, Jingwen
Guo, Minyi
He, Zhezhi
Zhao, Jieru
Zhu, Yuhao
contents 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$\times$ speedup and 3.9$\times$ energy efficiency over state-of-the-art accelerators.
format Preprint
id arxiv_https___arxiv_org_abs_2503_05197
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle StreamGrid: Streaming Point Cloud Analytics via Compulsory Splitting and Deterministic Termination
Feng, Yu
Liu, Zheng
Lin, Weikai
Liu, Zihan
Leng, Jingwen
Guo, Minyi
He, Zhezhi
Zhao, Jieru
Zhu, Yuhao
Hardware Architecture
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$\times$ speedup and 3.9$\times$ energy efficiency over state-of-the-art accelerators.
title StreamGrid: Streaming Point Cloud Analytics via Compulsory Splitting and Deterministic Termination
topic Hardware Architecture
url https://arxiv.org/abs/2503.05197