Evaluating graph coloring on GPUs

Abstract

This paper evaluates features of graph coloring algorithms implemented on graphics processing units (GPUs), comparing coloring heuristics and thread decompositions. As compared to prior work on graph coloring for other parallel architectures, we find that the large number of cores and relatively high global memory bandwidth of a GPU lead to different strategies for the parallel implementation. Specifically, we find that a simple uniform block partitioning is very effective on GPUs and our parallel coloring heuristics lead to the same or fewer colors than prior approaches for distributed-memory cluster architecture. Our algorithm resolves many coloring conflicts across partitioned blocks on the GPU by iterating through the coloring process, before returning to the CPU to resolve remaining conflicts. With this approach we get as few color (if not fewer) than the best sequential graph coloring algorithm and performance is close to the fastest sequential graph coloring algorithms which have poor color quality.

Parallel graph coloring algorithms for distributed GPU environments

“…Distributedmemory algorithms such as those in [3,19] use the speculate and iterate approach. Grosset et al [20] present a hybrid speculate and iterate approach that splits computations between the CPU and a single GPU, but does not operate on multiple GPUs in a distributed memory context. Sallinen et al [21] demonstrated how to color very large, dynamic graphs efficiently.…”

Feluca: A Two-Stage Graph Coloring Algorithm With Color-Centric Paradigm on GPU

“…Ç atalyürek et al proposed a multi-threaded dataflow algorithm for Cray XMT [47], which replies on the hardware support and it is not suitable for the current GPU. Grosset et al implemented the G-M algorithm [51] on GPU [21]. But the authors left part of the conflict resolution work to CPU.…”

“…A significant amount of work has been carried out to develop new data layout models [14], graph programming models (GAS, BSP), memory access patterns, workload mapping in order to optimize graph processing on GPU [15], [16], [17], [18], [19]. In graph coloring, although recent attempts [20], [21] have been made, unleashing the full power of GPU to achieve high-performance graph coloring still remains a great challenge.…”

High-Performance Parallel Graph Coloring with Strong Guarantees on Work, Depth, and Quality

“…Some of these schemes were implemented within frameworks or libraries [82], [122]- [124]. Another line of schemes incorporates GPUs and vectorization [41], [44], [45], [125]- [129]. Other schemes use recoloring [130], [131] in which one improves an already existing coloring.…”