Comparative analysis of GPU parallelization frameworks for computational operations applied in CFD solvers

Higher level GPU parallelization frameworks such as OpenMP, OpenACC, Kokkos, and SYCL are examined, and a comparative analysis of their parallel computing performance for applications in computational fluid dynamics (CFD) is performed. The study will benchmark these frameworks primarily using operations commonly performed in CFD solvers, including matrix addition, matrix-scalar multiplication, and matrix-vector multiplication. The primary goal is to assess the efficiency and scalability of these frameworks on GPU based architectures for these specific problems, while also considering their ease of implementation. Each of these frameworks leverage parallel computing resources differently: OpenMP and OpenACC utilize directives (pragmas) to parallelize code, while Kokkos and SYCL employ programming abstractions. The benchmark results will be obtained using identical hardware setups and optimized code implementations for each framework, in order to provide accurate and comparable performance metrics such as execution time and resource utilization. The main goal is to provide insights into which frameworks are best suitable for the particular demands of complex CFD solvers and other high-performance computational physics applications.