arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2511.04684cs.AR
收藏

RAS:一种面向高性能神经无损压缩的位精确 rANS 加速器

RAS: A Bit-Exact rANS Accelerator For High-Performance Neural Lossless Compression

Institute for Artificial Intelligence, Peking University(北京大学人工智能研究院);School of Electronic Engineering and Computer Science, EECS, Peking University(北京大学电子工程与计算机科学学院);School of Integrated Circuits, Peking University(北京大学集成电路学院)

Yuchao Qin, Anjunyi Fan, Bonan Yan

AI总结:

本文提出 RAS 硬件加速器,通过将 rANS 算法与概率生成器集成、两阶段更新、预测引导解码和多通道设计,在图像无损压缩中实现 121.2 倍编码和 70.9 倍解码加速,并减少约 55% 的二分搜索步数,优于 CPU/GPU rANS 实现。

摘要:

数据中心处理着海量数据,需要高效的无损压缩,然而新兴的基于概率模型的方法往往计算速度较慢。为了解决这一问题,我们提出了 RAS,即 Range Asymmetric Numeral System 加速系统,这是一种将 rANS 算法集成到无损压缩流水线中并消除关键瓶颈的硬件架构。RAS 将 rANS 核心与概率生成器耦合,以 BF16 格式存储分布,并将其一次性转换到由统一的除法/取模数据通路共享的定点域中。具有字节级重归一化的两阶段 rANS 更新降低了逻辑成本和内存访问量,而预测引导的解码路径则投机性地缩小累积分布函数(CDF)搜索窗口,并在必要时安全回退以保持位精确性。多通道组织扩展了吞吐量,并支持细粒度时钟门控以实现高效调度。在图像工作负载上,我们的 RTL 仿真原型相比 Python rANS 基线实现了 121.2 倍的编码加速和 70.9 倍的解码加速,将平均解码器二分搜索步数从 7.00 步减少到 3.15 步(约减少 55%)。当与神经概率模型配合使用时,RAS 能够保持比经典编解码器更高的压缩比,并优于 CPU/GPU 上的 rANS 实现,为快速神经无损压缩提供了一种实用的方法。

英文摘要:

Data centers handle vast volumes of data that require efficient lossless compression, yet emerging probabilistic models based methods are often computationally slow. To address this, we introduce RAS, the Range Asymmetric Numeral System Acceleration System, a hardware architecture that integrates the rANS algorithm into a lossless compression pipeline and eliminates key bottlenecks. RAS couples an rANS core with a probabilistic generator, storing distributions in BF16 format and converting them once into a fixed-point domain shared by a unified division/modulo datapath. A two-stage rANS update with byte-level re-normalization reduces logic cost and memory traffic, while a prediction-guided decoding path speculatively narrows the cumulative distribution function (CDF) search window and safely falls back to maintain bit-exactness. A multi-lane organization scales throughput and enables fine-grained clock gating for efficient scheduling. On image workloads, our RTL-simulated prototype achieves 121.2x encode and 70.9x decode speedups over a Python rANS baseline, reducing average decoder binary-search steps from 7.00 to 3.15 (approximately 55% fewer). When paired with neural probability models, RAS sustains higher compression ratios than classical codecs and outperforms CPU/GPU rANS implementations, offering a practical approach to fast neural lossless compression.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑