Talk 1: Programming for Data Locality and Parallelism, Didem Unat, Koç Univeristy.
Problems in computational finance, life-sciences, machine learning, scientific computing etc are being studied on parallel architectures, which have become increasingly complex both in terms of memory subsystems and execution units. The increasing complexity causes substantial programming challenges to all application software developers since computing devices ranging from $100M supercomputers to $100 tablets are all equipped with homogeneous or heterogeneous multicore architectures. Applications must be optimized to fully utilize a diverse pool of resources; and they must be programmed under a simple and elegant representation that permits easy code migration on the next generation computers. My research revolves around the central topic of parallel, multicore and heterogeneous computing on state-of-the-art parallel architectures. I believe data locality abstractions and asynchronous execution are among the most promising solutions for alleviating the programming and scalability challenges. While programming abstractions provide performance portability and programmer’s productivity, asynchronous execution facilitates efficient use of available resources. In this talk, I will present software solutions that I and my collaborators have developed to facilitate application development through (1) programming model and programming abstractions with TiDA, (2) runtime system with Perilla, and (3) performance monitoring tool with ComDetective.
Bio: Didem Unat is an Assistant Professor at the Computer Science and Engineering Department of Koç University, Istanbul, Turkey. She earned her Ph.D. at the University of California-San Diego. From 2012-2014, she was at the Lawrence Berkeley National Laboratory and worked at the Exascale Combustion Co-design center. She is the recipient of the prestigious Luis Alvarez Fellowship in 2012 at the Berkeley Lab, the Marie Sklodowska-Curie Individual Fellowship from the European Commission in 2015, and the Young Scientists Award in 2019 from the Science Academy of Turkey. Her research interest lies primarily in the area of high-performance computing; parallel programming models, runtime systems, compiler analysis, and performance modeling.
Talk 2: Application Characterization with Cache-Aware Roofline Model, Leonel Sousa, Universidade de Lisboa.
To overcome the issues related with the end of Dennard’s scaling and Moore’s Law, modern micro-architecture has been the subject of constant technological improvements, increasing the complexity of these systems. This complexity imposes several challenges when deriving which hardware resources are responsible to prevent application from achieving the maximum performance of the micro-architecture. To address this issue, simple models and analysis methods, such as Cache-Aware Roofline Model, were proposed in order to provide first-order insights regarding application execution, allowing software developers and hardware engineers to derive the most suitable optimization techniques to extract the maximum potential of micro-architectures. In this talk, a set of Cache-Aware Roofline Models are presented, aiming at providing a more accurate characterization of real-world applications in computer systems. To demonstrate the potential of the proposed models, a set of applications from Exascale Computing Project Benchmark Suite are characterized with the proposed models and its insights are validated against current state-of-the-art solutions.
Bio: Leonel Sousa received the PhD degree in electrical and computer engineering from the InstitutoSuperior Tecnico (IST), Universidade de Lisboa(UL), Lisbon, Portugal, in 1996. Since 1996, he has been with IST, where he is currently the chair of the Department of Electrical and Computer Engineering, and has been a full professor since 2010. In 2016, he was a visiting professor with Tsukuba University, Tsukuba, Japan, with a JSPS Invitation Fellowship for Research in Japan, and at Carnegie Mellon University, Pittsburgh, PA. His research interests include computer architectures, high performance computing, and multimedia systems. He has contributed more than 250 papers for international journals and conferences and to the organization of several international conferences. He is currently an associate editor of the IEEE Transactions on Computers. He is a distinguished scientist of the ACM.