Towards Hybrid CPU-GPU Computing in a Backtracking-based Load Balancing Framework

Abstract

General-purpose computing on graphics processing units (GPGPUs) has become increasingly prevalent, with hybrid CPU-GPU systems at the forefront of parallel computing. Dynamic load balancing is highly effective for maximizing CPU and GPU utilization in such environments. Backtrackingbased load balancing, utilizing work-stealing, offers a promising strategy for task parallelism. However, Tascell, a task-parallel language implementing this mechanism, currently lacks GPU support, limiting its potential for hybrid CPU-GPU parallelism and constraining its application in computationally intensive tasks. In this paper, we propose enabling Tascell to fully utilize the computational power of CPU-GPU hybrid environments by writing both CPU-oriented and GPU-oriented code for workers to execute, allowing any worker to run GPU-oriented code based on task size and GPU availability. Using this technique, we implemented hybrid CPU-GPU programs for three applications using Tascell: recursive block matrix multiplication, 2D stencil computations and Mandelbrot Set calculations. The GPUoriented code was implemented using OpenACC or the NVBLAS library. We conducted performance evaluations on both highperformance and workstation-grade CPU-GPU hybrid computing environments. Results demonstrated that in the workstationgrade environment, the hybrid approach outperformed both CPU-only and GPU-only configurations. Notably, hybrid CPUGPU executions achieved performance improvements of up to 12.85% and 25.19% in 2D stencil applications compared to GPU-only and CPU-only executions, respectively. These findings provide valuable insights into effectively leveraging hybrid CPUGPU systems within a backtracking-based load balancing framework.