Presentation
CoSPARSE: A Software and Hardware Reconfigurable SpMV Framework for Graph Analytics
TimeWednesday, December 8th10:50am - 11:10am PST
Location3018
SessionScaling the data wall: Novel heterogeneous architectures for emerging data-intensive applications
Event Type
Research Manuscript
Presented In-Person
SoC, Heterogeneous, and Reconfigurable Architectures
Design
DescriptionSparse matrix-vector multiplication (SpMV) is a vital building block for graph analytics
algorithms, where the varying active vertex set across iterations can be harnessed to realize better performance. Previous solutions have focused on either dynamically switching algorithms between iterations (software) or designing custom accelerators (hardware). We propose a novel framework, CoSPARSE, that builds on a general-purpose reconfigurable hardware and considers hardware and software reconfiguration as a holistic solution to accelerate SpMV-based graph analytics algorithms. Across a suite of graph algorithms, CoSPARSE outperforms the state-of-the-art shared memory framework Ligra, with up to 3.51x better performance and 877x better energy efficiency.
algorithms, where the varying active vertex set across iterations can be harnessed to realize better performance. Previous solutions have focused on either dynamically switching algorithms between iterations (software) or designing custom accelerators (hardware). We propose a novel framework, CoSPARSE, that builds on a general-purpose reconfigurable hardware and considers hardware and software reconfiguration as a holistic solution to accelerate SpMV-based graph analytics algorithms. Across a suite of graph algorithms, CoSPARSE outperforms the state-of-the-art shared memory framework Ligra, with up to 3.51x better performance and 877x better energy efficiency.