Unit:
Κατεύθυνση Σχεδίασης Ολοκληρωμένων ΚυκλωμάτωνΠληροφορική
Author:
Zouzoula Stavroula
Supervisors info:
Δημήτριος Σούντρης, Καθηγητής, Ηλεκτρολόγοι Μηχανικοί και Μηχανικοί Υπολογιστών, Εθνικό Μετσόβιο Πολυτεχνείο
Αντώνης Πασχάλης, Καθηγητής, Πληροφορικής και Τηλεπικοινωνιών, Εθνικό και Καποδιστριακό Πανεπιστήμιο Αθηνών
Λάζαρος Παπαδόπουλος, Διδάκτωρ, Ηλεκτρολόγοι Μηχανικοί και Μηχανικοί Υπολογιστών, Εθνικό Μετσόβιο Πολυτεχνείο
Original Title:
Implementation of Linear Algebra Kernels in Heterogeneous Systems using StarPU
Translated title:
Implementation of Linear Algebra Kernels in Heterogeneous Systems using StarPU
Summary:
Manycore and heterogeneous computing systems led application programmers to turn to parallelism. A lot of parallel models have been proposed, with the most promising being task-based models. Many task-based runtime systems have been developed in the effort to better exploit parallelism and achieve better performance.
In this master thesis, StarPU's runtime system is presented. StarPU was applied to some Polybench benchmarks that were executed for different input sizes in an heterogeneous node. From the obtained results, we reached conclusions on StarPU's scheduling decisions.
Main subject category:
Technology - Computer science
Keywords:
Parallel programming, task-based programming models, StarPU
