Parflow 3.9: development of lightweight embedded DSLs for geoscientific models

Abstract

Abstract. Recognizing the leap in high-performance computing with accelerated co-processors, we propose a lightweight approach to adapt legacy codes to next generation hardware and achieve efficiently a high degree of performance portability. We focus on abstracting the computing kernels at the loop levels based on the lightweight, preprocessor-based embedded Domain Specific Language (eDSL) concept in conjunction with Unified Memory management. We outline a set of code pre-adaptations that facilitate the proposed abstraction. In two geophysical code applications programmed in C and Fortran, we demonstrate the efficiency of the eDSL approach in adaptation to NVIDIA GPUs with: native CUDA and Kokkos eDSL backends achieving up to 10–30 fold speedup. Our experience suggests that the proposed lightweight eDSL code adaptation is less expensive in terms of Full Time Equivalent of effort than adaptation based on complex DSL approaches, even if no earlier GPU competence exists.

Challenges of applying an embedded domain specific language for performance portability to Earth system models

“…The submodule CLM (Common Land Model) that is coupled with ParFlow is based on the Fortran programming language [6,7,8]. After decades of development of ParFlow, which originally supported only the Multiple Processing Interface (MPI), ParFlow now supports utilization of GPUs via the eDSL approach [9,10].…”

“…The details of applying the eDSL approach to ParFlow are outlined by Piotrowski et al [10]. The key points of the eDSL approach for ParFlow can be summarized as follows:…”

“…The manpower and development time for implementing the eDSL approach should also be considered. At least one year of development time was required to implement the eDSL approach into other codes [9,10]. The development of more advanced DSL approaches for ICON took multiple years [4].…”