Capstan: A vector RDA for sparsity

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-oforder execution to increase SRAM random-access throughput from 32% to 80%. For a variety of sparse applications, Capstan with DDR4 memory is 18× faster than a multi-core CPU baseline, while Capstan with HBM2 memory is 16× faster than an Nvidia V100 GPU. For sparse applications that can be mapped to Plasticine, a recent dense RDA, Capstan is 7.6× to 365× faster and only 16% larger.