On the use of GPGPU-enabled scientific libraries for Earth Science applications: two cases studies from oceanography

Robust and consolidated scientific libraries are versatile instruments capable of adapting themselves to different execution environments. They provide a large variety of numerical tools to solve problems of interest for a wide range of scientific communities. Using such tools in design of numerical applications, involves advantages concerning both the "software quality" improvement (e.g. in terms of robustness, portability, resilience, scalability, etc.) and time reduction both in software development and in maintenance. If the initial effort for the introduction of scientific computing libraries in high level software may seem large, on the other hand, it ensures the chance to give a long life to the software (by amortizing the initial time investment) and prevents the user from the introduction of changes to the lowest level necessary to incorporate any evolution of the hardware execution environment. Here we present two case studies as proof-of-concept of what we said above.