In high performance computing heterogeneity is becoming more and more important. It aims at exploiting the capability of the different units in obtaining better energy efficiency and performances given different workloads.
Usually in this context the starting point are application designed to work with MPI in a multi node environment, and in the single node OpenMP is used to exploit the shared memory. In the latest years GPU started to enter in this context, with custom libraries or kernels offloaded using low level languages like CUDA.
In this seminar, based on a course followed at the CINECA supercomputing center,  the speaker will introduce a directive based language, OpenACC, that is inspired by OpenMP but targets different accelerators and supports the creation of kernel that can be offloaded with the minimum effort by the programmer.