Software-Defined Vector Processing on Manycore Fabrics

Software-Defined Vector Processing on Manycore Fabrics
复制标题

众核结构上的软件定义矢量处理

DOI:
10.1145/3466752.3480099
复制
发表时间:
2021
期刊:
MICRO '21: MICRO-54: 54th Annual IEEE/ACM International Symposium on Microarchitecture
影响因子:
--
通讯作者:
Sampson, Adrian
Sampson, Adrian
中科院分区:
--
文献类型:
--
作者:
Bedoukian, Philip;Adit, Neil;Peguero, Edwin;Sampson, Adrian

文献摘要

参考文献

相似文献

我们描述了一个平铺架构,可以流畅地在众核(MIMD)和向量(SIMD)执行之间转换。硬件提供了一个软件定义的向量编程模型,该模型允许应用程序将多核瓦片组聚合到逻辑向量引擎中。在众核模式下,机器的行为就像一个标准的并行处理器。在向量模式中,区块组将其功能单元重新用作向量执行通道,并将暂存器重新用作向量存储体。关键机制是指令转发网络:单个瓦片获取指令并将其发送到其他尾随核心。大多数内核禁用其前端和指令缓存,因此向量组分摊了冯诺依曼控制的内在硬件成本。向量组还使用解耦的访问/执行方案来集中其内存请求并发布合并的宽负载。我们通过最小的硬件扩展来增强现有的RISC-V众核设计,以实现软件定义的向量。周期级仿真结果表明,软件定义的向量比标准MIMD执行平均提高1.7倍的性能,同时节省22%的能量。与类似配置的GPU相比,该架构将性能提高了1.9倍。
We describe a tiled architecture that can fluidly transition between manycore (MIMD) and vector (SIMD) execution. The hardware provides a software-defined vector programming model that lets applications aggregate groups of manycore tiles into logical vector engines. In manycore mode, the machine behaves as a standard parallel processor. In vector mode, groups of tiles repurpose their functional units as vector execution lanes and scratchpads as vector memory banks. The key mechanism is an instruction forwarding network: a single tile fetches instructions and sends them to other trailing cores. Most cores disable their frontends and instruction caches, so vector groups amortize the intrinsic hardware costs of von Neumann control. Vector groups also use a decoupled access/execute scheme to centralize their memory requests and issue coalesced, wide loads.We augment an existing RISC-V manycore design with a minimal hardware extension to implement software-defined vectors. Cycle-level simulation results show that software-defined vectors improve performance by an average of 1.7 × over standard MIMD execution while saving 22% of the energy. Compared to a similarly configured GPU, the architecture improves performance by 1.9 ×.
消除核心边界以获得稳健且可配置的性能
DOI: 10.1109/micro.2010.30
发表时间: 2010
期刊: 2010 43rd Annual IEEE/ACM International Symposium on Microarchitecture
影响因子: --
作者:
S. Gupta;Shuguang Feng;Amin Ansari;S. Mahlke
通讯作者: S. Mahlke
BYOC:异构 ISA 研究的“自带核心”框架
DOI: 10.1145/3373376.3378479
发表时间: 2020
期刊: Proceedings of Twenty-Fifth International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS’20
影响因子: --
作者:
Balkind, Jonathan;Lim, Katie;Schaffner, Michael;Gao, Fei;Chirkov, Grigory;Li, Ang;Lavrov, Alexey;Nguyen, Tri M.;Fu, Yaosheng;Zaruba, Florian
通讯作者: Zaruba, Florian
向量机的缓存填充/访问解耦
DOI: --
发表时间: 2004
期刊: Micro
影响因子: --
作者:
C. Batten;R. Krashinsky;S. Gerding;K. Asanović
通讯作者: K. Asanović
迷失在抽象中:在中间语言级别分析 GPU 的陷阱
DOI: --
发表时间: 2018
期刊: International Symposium on High-Performance Computer Architecture
影响因子: --
作者:
Anthony Gutierrez;Bradford M. Beckmann;A. Duțu;Joseph Gross;Michael LeBeane;J. Kalamatianos;Onur Kayiran;Matthew Poremba;Brandon Potter;Sooraj Puthoor;Matthew D. Sinclair;Mark Wyse;Jieming Yin;Xianwei Zhang;Akshay Jain;Timothy G. Rogers
通讯作者: Timothy G. Rogers