Research Goal
In real-world applications, numerical methods based on unstructured meshes, such as Finite Element and Finite Volume, are time-demanding. My research is about studying and developing optimisations at different layers of abstractions for these computations. The key difficulty is represented by the presence of indirect memory accesses (e.g. A[B[i]]) that prevent many standard (e.g. polyhedral) optimisations.
Research Interests
- Vectorisation of Finite Element assembly kernels
- Tiling techniques for unstructured mesh computations
Publications
- T. De Matteis, F. Luporini, G. Mencagli and M. Vanneschi. Evaluation of Architectural Supports for Fine-Grained Synchronization Mechanisms. Proceedings of the 11th IASTED International Conference on Parallel Distributed Computing and Networks, Innsbruck, Austria, 2013. isbn: 978-0-88986-955-4 (Print) 978-0-88986-943-1 (CD), doi: 0.2316/P.2013.795-023.
Talks