The following is a list of my published works.
Preliminary work done using the CUDA library for unstructured mesh adaptation in Omega_h on NVIDIA GPUs shows a significant performance increase when using a memory pool as opposed to traditional memory management strategies. However, CUDA is a proprietary library developed by NVIDIA for NVIDIA devices. The use of vendor-specific libraries such as CUDA for GPU computing can limit the portability and flexibility of applications. As such, this research aims to achieve comparable performance gains on AMD and Intel devices using the cross-platform library, Kokkos, to implement the memory pool.
Download PDF