On one hand my research interests concern the performance of many-core micro-architectures and their programming models. Today's many-core designs are based on more than twenty years of single core incremental design implementations. Consequently micro-architecture components are well suited for single thread performance and centralized model of execution but not for parallel executions. We believe that many-core designs need to be revisited to efficiently exploit parallelism.
On the other hand my research focus on numerical accuracy of high performance computing. High performance computing software, such as numerical simulation may suffer from floating-point errors. Such errors affect reproducibility and accuracy. We aim to provide useful libraries to improve both reproducibility and accuracy.
Download a short CV: short_cv_dp.pdf