PIMCOMP: A Universal Compilation Framework for Crossbar-based PIM DNN Accelerators

TL;DR

PIMCOMP is a universal compiler framework designed for crossbar-based PIM DNN accelerators, improving performance and latency across various workloads through a structured compilation process.

Contribution

It introduces a comprehensive compilation framework with four stages and two modes, addressing key issues like parallelism and array mapping for PIM DNN accelerators.

Findings

Achieves 1.6× throughput improvement over PUMA.

Achieves 2.4× latency reduction over PUMA.

Supports high-throughput and low-latency scenarios.

Abstract

Crossbar-based PIM DNN accelerators can provide massively parallel in-situ operations. A specifically designed compiler is important to achieve high performance for a wide variety of DNN workloads. However, some key compilation issues such as parallelism considerations, weight replication selection, and array mapping methods have not been solved. In this work, we propose PIMCOMP - a universal compilation framework for NVM crossbar-based PIM DNN accelerators. PIMCOMP is built on an abstract PIM accelerator architecture, which is compatible with the widely used Crossbar/IMA/Tile/Chip hierarchy. On this basis, we propose four general compilation stages for crossbar-based PIM accelerators: node partitioning, weight replicating, core mapping, and dataflow scheduling. We design two compilation modes with different inter-layer pipeline granularities to support high-throughput and low-latency application scenarios, respectively. Our experimental results show that PIMCMOP yields improvements of 1.6$\times$ and 2.4$\times$ in throughput and latency, respectively, relative to PUMA.