This thesis aims to integrate the performance portability library Kokkos, into ALPACA, a finite-volume, multiresolution, multi-phase CFD code framework. The primary motivation behind this work is to make ALPACA hardware-agnostic and to enable its execution on different architectures. The thesis discusses the implementation aspects in detail, including the data structure and algorithmic changes and the issues faced during the porting process. The implementation is tested against the original ALPACA code to ensure algorithmic correctness. The performance of CUDA, OpenMP and the serial backend is analyzed in detail. A significant performance improvement is observed, especially in the case of the CUDA backend. The performance of the OpenMP backend is comparable to the original ALPACA code with MPI implementation. The serial backend is observed to be slower than the original ALPACA code. Also, various experimental optimizations are explored to identify possible avenues of parallelism. The thesis concludes with a discussion of future work and the potential improvements that can be made to the current implementation. Overall, the thesis lays the building block for a performance portable version of ALPACA using Kokkos.     «

This thesis aims to integrate the performance portability library Kokkos, into ALPACA, a finite-volume, multiresolution, multi-phase CFD code framework. The primary motivation behind this work is to make ALPACA hardware-agnostic and to enable its execution on different architectures. The thesis discusses the implementation aspects in detail, including the data structure and algorithmic changes and the issues faced during the porting process. The implementation is tested against the original...     »