SSCRAP: An Environment for Coarse Grained Algorithms
Résumé
In this paper we present SSCRAP an environment for the development of coarse grained algorithms which is well suited for regular as well as irregular problems. SSCRAP implements the currently known coarse grained models on distributed memory architectures as well as on shared memory architectures. It allows for a high-level algorithm design and handles communication and synchronization aspects efficiently at the same time. By making the communication level transparent for the user it guarantees a good portability. SSCRAP has also a fine execution analysis: users are able to compare efficiency of algorithms and to check the predictability level of the considered theoretical complexity.