GPU-Accelerated Tree-Search in Chapel Versus CUDA and HIP

In the context of exascale programming, the PGAS-based Chapel is among the rare languages targeting the holistic handling of high-performance computing issues including the productivity-aware harnessing of Nvidia and AMD GPUs. In this paper, we propose a pioneering proof-of-concept dealing with this latter issue in the context of tree-based exact optimization. Actually, we revisit the design and implementation of a generic multi-pool GPU-accelerated tree-search algorithm using Chapel. This algorithm is instantiated on the backtracking method and experimented on the N-Queens problem. For performance evaluation, the Chapel-based approach is compared to Nvidia CUDA and AMD HIP low-level counterparts. The reported results show that in a single-GPU setting, the high GPU abstraction of Chapel results in a loss of only 8% (resp. 16%) compared to CUDA (resp. HIP). In a multi-GPU setting, up to 80% (resp. 71%) of the baseline speedup is achieved for coarse-grained problem instances on Nvidia (resp. AMD) GPUs.