The main objective of this project is the design of solutions in different areas to improve the efficiency and programming of high performance computing (HPC) systems considering application domains in various fields of science and engineering. The project proposal is a natural continuation of previous research in the HPC field, from the architecture to applications, where the group has extensive knowledge of the involved technologies and their evolution.

This project proposes a series of research lines interrelated and developed in the context of high-performance computing. The project is supported by previous research done by the group, and establishes new challenges that arise as a consequence of the rapid evolution of high-performance computing systems. In particular, our group participates in providing solutions to improve the performance of new homogeneous and heterogeneous systems at the execution model, compiler, runtime and microarchitecture levels.

This project proposes the study the currently dominant high-performance architectures, specially the new multi-core processors. In particular, our group participates in the analysis and automatic exploitation of parallelism and locality for applications based on complex data structures (pointers, indirections ...), as well as in microarchitectural support for those applications.

The main aim of this project is offering a national framework for research groups expert in supercomputing applications to collaborate together with expert hardware/software machine designers in order to design and use these machines efficiently in the near future. Our participation is based in our expertise in the analysis and automatic optimization of real-life applications.

The aim of this project is tackling some of the problems that current implementations of transactional memory systems suffer, with an important impact in their performance. Specifically, we proposes to provide solutions to the design of signatures for improving the efficiency of conflict detection, and to extend the model to heterogeneous architectures.