-
Agenda for the Summer School at a glance:
-
Detailed Agenda for the Summer School:
-
Monday, 23rd June, a pre-school tutorial
- Afternoon: João Barbosa will introduce users to the NVidia architecture of the Fermi and Kepler architecture coupled to the CUDA programming model; hands-on exercises will be used to get the user acquainted to these computing platforms, namely in parallelization and optimization through example testing.
-
Wednesday, 25th June
- 09h00-12h30: Keshav Pingali will introduce an abstraction called the operator formulation of algorithms, which is useful for understanding the patterns of parallelism and locality in algorithms from a wide range of application areas including high-performance computing, machine learning and big-data applications.
- 14h00-14h30: Dora Blanco: will addresses the special features of n-dimensional images (2D/3D medical images or multi-/hyper- spectral images), which present new challenges in image analysis, target detection, image segmentation, classification and visualization, when aiming real time execution. Her talk shows scenarios that exploit CUDA-based GPU consumer platforms to achieve real-time execution for processing of n-dimensional images, focusing on 2D/3D medical images and remote sensing and multispectral and hyperspectral datasets. The real time applications range from medical diagnosis to the detection of shipwrecked persons in adverse climate conditions or the classification of different crops in agricultural land.
- 14h30-15h00: Basilio B. Fraguela: will address the transport of a pollutant in a shallow water system, modeled by a transport equation, with a large relevance in many ecological and environmental studies. His talk describes his experience in the implementation of a mathematical model that takes into consideration topographic data, tides and ocean currents, to predict the diffusion of a pollutant in a realistic environment. He then makes a comparative evaluation of a regular multicore CPU version with several GPU based implementations and an MPI+CUDA version, which allows to exploit heterogeneous clusters.
- 15h00-17h30: Donald Fussell will address modern high performance computing systems as well as current generation processor chips that incorporate GPU cores capable of high performance parallelism along with multiple CPU cores; he will then examine strategies for designing high performance algorithms for GPUs and CPUs and explore how to leverage a combination of the two to obtain even higher performance.
-
Thursday, 26th June
- 09h00-12h30: Mary Hall will describe the role of automatic performance tuning for current and future high-performance architectures, focusing the discussion on HPC computations that include dense and sparse linear algebra and geometric multigrid, and on both compiler transformations and code generation and programmer-directed tuning
- 14h00-14h30: Miguel Avillez will address the behaviour of interstellar gas, having a variety of temperatures, pressures and densities and permeated by magnetic fields and highly turbulent, which is dominated by the ongoing effects of stellar winds, supernovae (explosion of massive stars), and the competition of cooling and heating processes. Modeling the interstellar gas requires a multi-fluid approach to tackle both the thermal and dynamical evolutions, combined to sophisticated numerical methods/algorithms to attain the highest possible resolution and reproduce shock structures. He has been developing software to handle such multi-fluid calculations in heterogeneous computing systems (CPUs and GPUs), and he will discuss the developed software, benchmarks and performance tests.
- 14h30-15h00: Luís Paulo Santos presents a framework that addresses the development and execution of data parallel irregular applications in heterogeneous systems. A unified task-based programming and execution model is proposed, together with inter and intra-device scheduling, which aim to achieve performance scalability across multiple devices, when coupled with a data management system, while maintaining high programming productivity. Intra-device scheduling on wide SIMD/SIMT architectures resorts to consumer-producer kernels, which enable balancing irregular workloads and increase resource utilization, by allowing dynamic generation and rescheduling of new work units. Comparisons with an alternative framework that targets regular workloads, StarPU, consistently demonstrate significant speedups.
- 15h00-17h30: Jongsoo Park will overview current and future trend of many-core processor architectures as Intel Xeon Phi as the primary example; he will then discuss its implication on software design, namely, the need for low-communication, high-parallelism, and low-synchronization algorithms and implementations.
-
Friday, 27th June
- All day: John Cazes, Lucas Wilson, Carlos Rosales will introduce users to the Intel Xeon Phi architecture in a practical manner; multiple lectures and hands-on exercises will be used to get the user acquainted with the Phi platform and explore the different execution modes as well as parallelization and optimization through example testing and reports.