Capstan: A Vector RDA for Sparsity
Alexander Rucker, Matthew Vilim, Tian Zhao, Yaqi Zhang, Raghu Prabhakar, Kunle Olukotun
Abstract
This paper proposes Capstan: a scalable, parallel-patterns-based, reconfigurable dataflow accelerator (RDA) for sparse and dense tensor applications. Instead of designing for one application, we start with common sparse data formats, each of which supports multiple applications. Using a declarative programming model, Capstan supports application-independent sparse iteration and memory primitives that can be mapped to vectorized, high-performance hardware. We optimize random-access sparse memories with configurable out-of-order execution to increase SRAM random-access throughput from 32% to 80%.
Topics & Concepts
Computer scienceDataflowScalabilityRandom accessParallel computingSparse matrixStatic random-access memoryThroughputComputer architectureComputer hardwareProgramming languageGaussianOperating systemPhysicsQuantum mechanicsWirelessParallel Computing and Optimization TechniquesEmbedded Systems Design TechniquesInterconnection Networks and Systems