Performance portability in meshless methods using a parallel implementation procedure based on meta-programming

Resumo

The present paper focuses upon the development of a parallel implementation procedure to the nu-
merical simulation of boundary value problems by using the principles of functional programming and memory

polymorphism. The proposed procedure aims to evaluate the portability of computational performance in dif-
ferent execution spaces using meta-programming and simplified representation of physical-mathematical models

through simple usage of declarative programming. For this purpose, an execution model for meshless methods is

presented, under the premise that specializations for the execution and memory spaces must occur during the com-
pilation time. This means that the declared model can be simulated in distinct architectures without any changes

in the implementation, allowing ensuring the portability of the computational performance and reducing the com-
plexity of the meshless numerical models. At the end of the paper, an example is carried out aiming to evaluate the

computational performance and the efficiency of the parallel implementation procedure applied to the Helmholtz
problem.