PIM-FW: Hardware-Software Co-Design of All-pairs Shortest Paths in DRAM

TL;DR

This paper introduces PIM-FW, a hardware-software co-designed architecture leveraging processing-in-memory to significantly accelerate the Floyd-Warshall algorithm for all-pairs shortest paths, reducing data movement and energy consumption.

Contribution

It presents a novel PIM architecture and dataflow for the FW algorithm, enabling high parallelism and eliminating data movement bottlenecks in graph processing.

Findings

Achieves 18.7x speedup on 8192x8192 graphs

Consumes 3200x less DRAM energy than GPU implementations

Demonstrates effective hardware-software co-design for PIM-based graph algorithms

Abstract

All-pairs shortest paths (APSP) is a fundamental algorithm used for routing, logistics, and network analysis, but the cubic time complexity and heavy data movement of the canonical Floyd-Warshall (FW) algorithm severely limits its scalability on conventional CPUs or GPUs. In this paper, we propose PIM-FW, a novel co-designed hardware architecture and dataflow that leverages processing in and near memory architecture designed to accelerate blocked FW algorithm on an HBM3 stack. To enable fine-grained parallelism, we propose a massively parallel array of specialized bit-serial bank PE and channel PE designed to accelerate the core min-plus operations. Our novel dataflow complements this hardware, employing an interleaved mapping policy for superior load balancing and hybrid in and near memory computing model for efficient computation and reduction. The novel in-bank computing approach allows all distance updates to be performed and stored in memory bank, a key contribution is that eliminates the data movement bottleneck inherent in GPU-based approaches. We implement a full software and hardware co-design using a cycle-accurate simulator to simulate an 8-channel, 4-Hi HBM3 PIM stack on real road-network traces. Experimental results show that, for a 8192 x 8192 graph, PIM-FW achieves a 18.7x speedup in end-to-end execution, and consumes 3200x less DRAM energy compared to a state-of-the-art GPU-only Floyd-Warshall.