Barcelona Supercomputing center Co-Supervisor:

Heterogeneous computing combines general purpose CPUs with accelerators to efficiently execute both: sequential control-intensive and data parallel phases of applications. Exist-ing programming models for heterogeneous computing rely on programmers to explicitly manage the different memories in the system and manage data transfers between the CPU system memory and accelerator memory. To facilitate the programming of accelerator-based systems, has created a library called GMAC. This model basically abstracts the developer from the peculiarities of each partic-ular system and proposes a simple API to replace the host code for devices management for a more generic one. Programming with GMAC, the programmer does not need to manage data transfers between different devices. This action is done internally by GMAC. In this project we improve GMAC with new capabilities and also improve already existing functions. First, we have developed a utility that divide the shared data space into similar blocks in length. Secondly, we have developed a memory coherence protocol to maintain all shared data coherent along all the devices. Also, we have checked that all the functions