GrateTile:用于CNN处理的高效稀疏张量切片
GrateTile: Efficient Sparse Tensor Tiling for CNN Processing
摘要:
我们提出了GrateTile,一个有效的,硬件友好的数据存储方案稀疏CNN特征图(激活)。它将数据划分为大小不等的子张量,并以较小的索引开销将其存储在压缩但可随机访问的格式中。这种设计使现代CNN加速器能够以平铺处理方式即时获取和解压缩子张量。GrateTile适用于支持对齐、合并数据访问的架构,并且只需要对整体架构设计进行最小的更改。我们使用最先进的CNN模拟GrateTile,并显示平均55%的DRAM带宽减少,同时仅使用0.6%的特征图大小进行索引存储。
英文摘要:
We propose GrateTile, an efficient, hardwarefriendly data storage scheme for sparse CNN feature maps (activations). It divides data into uneven-sized subtensors and, with small indexing overhead, stores them in a compressed yet randomly accessible format. This design enables modern CNN accelerators to fetch and decompressed sub-tensors on-the-fly in a tiled processing manner. GrateTile is suitable for architectures that favor aligned, coalesced data access, and only requires minimal changes to the overall architectural design. We simulate GrateTile with state-of-the-art CNNs and show an average of 55% DRAM bandwidth reduction while using only 0.6% of feature map size for indexing storage.
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。