Presentation
PIMCOMP: A Universal Compilation Framework for Crossbar-based PIM DNN Accelerators
DescriptionA specifically designed compiler is important to PIM DNN accelerators. However some key compilation considerations are neglected by the state-of-the-art solution (PUMA). In this work, we propose PIMCOMP - a universal compilation framework for crossbar-based PIM DNN accelerators. PIMCOMP is built on an abstract PIM accelerator architecture and an execution model. We present four compilation stages: node partitioning, weight replicating, core mapping and dataflow scheduing. We design two pipeline modes to support high-throughput and low-latency application scenarios respectively. Our experimental results show that PIMCOMP yields improvements of 1.6× and 2.4× in throughput and latency respectively, relative to PUMA.
Event Type
Research Manuscript
TimeWednesday, July 12th3:55pm - 4:10pm PDT
Location3003, 3rd Floor
Design
In-memory and Near-memory Computing Architectures, Applications and Systems