1 : Centre Inria de l'Université de Bordeaux Institut National de Recherche en Informatique et en Automatique
Data in high performance computing are typically stored contiguously for performance considerations. In computational sciences, these data are often associated to an underlying unstructured mesh, and the stride of the data associated with each mesh element is not uniform. These data offsets are typically stored as a graph adjacency, in a dedicated array. Moreover, the developer may need to do some further sub-indexing on the data associated to one mesh element (e.g. the index of the variable on the element, the index of the degree of freedom of the variable on the element...), raising the need for storing also these arrays of sub-offsets. This presentation shows how variadic templates in modern C++ can be used to write a general, compact template class allowing to generate multidimensional offsets with an arbitrary number of indices.
Type :
:
Présentation
Thématiques
:
La programmation d’aujourd’hui et de demain
