Programming GPU Devices using OpenACC Directives on the Cray XK6
March, 6-7  2012
CSCS in Manno

The registration fee is of CHF 150 including the coffee breaks.

Registration and agenda »

Contents

Attendees of this HP2C training event will learn about the Cray XK6 hybrid multi-core and GPU architecture and its programming environment.

They will learn about the OpenACC directives, which were designed to help users develop and port applications to run on heterogeneous systems. They will have an understanding on how to use the Cray Performance tools to identify “hot areas” in the code to focus the use of OpenACC directives. They will have the opportunity to experiment the OpenACC directives with the Cray Compilation Environment (CCE). In addition, they will learn about the Cray scientific libraries for accelerators and will learn and experiment Allinea’s DDT and Cray’s Performance Tools for debugging and performance tuning of heterogeneous applications on the Cray XK6 systems.

Attendees are encouraged to bring in their own applications and codes for the hands-on sessions.  Experts from Cray PE, OpenACC and libsci development and performance tools and Allinea DDT debugger will be present at the meeting for discussions and feedback.  We also invite current users who have their applications running successfully on the Cray XK6 system to present brief user experience talks.

Agenda

- Welcome
- Overview of the Cray XK6 system
- Introduction to Cray XK6 Programming Environment
- Support for GPU application development and execution

• GPU development environments (CUDA C & Fortran, OpenCL & OpenACC from Cray & PGI)
• GPU accelerated libraries
• Message passing communication (MPI)

- Introduction to OpenACC
- Development cycle of application porting

• Static analysis of the application
• Find hot loops
• Scoping Analysis
• Add OpenMP
• Create OpenACC regions from OpenMP regions

- Using libsci_acc
- Debugging
- Performance tuning

• Profile application
• Using the accelerator hardware counters
• Analysis of data transfers
• Add data regions

Abstract

We consider regularization methods for the nonconvex unconstrained and convexely constrained optimization problems.  After motivating these algorithms, we review known convergence results and emphasize their remarkabke complexity properties, that is the number of function evaluations that are needed for the algorithm to produce an epsilon-critical point.  We also discuss the complexity of the well-known steepest-descent and Newton’s method in the unconstrained case and report some surprising conclusions regarding their relative complexity.

Bio

Philippe Toint is director of the  Department of Mathematics of the University of Namur (Belgium), co-director of the  Numerical Analysis Research Unit, director of the  Transportation Research Group. Chairman elect (2010-2012) of the Mathematical Optimization Society , SIAM fellow (class 2009) and Honorary Professor at the University of Edinburgh.

His research interests are smooth nonlinear optimization, with an emphasis on the algorithmic viewpoint, ranging from convergence theory to numerical considerations and software development ( LANCELOT, CUTEr, GALAHAD ). Practical and multidisciplinary applications of optimization techniques.

Multiple scales analyses and numerics for atmospheric flows

Regime(s) of validity of sound-proof models

Abstract

Asymptotic techniques generalize the classical approach of scale analysis in theoretical meteorology. Through, e.g., matched asymptotic and multiple scales expansions they allow us to systematically study interactions across separated length and times scales. This will be demonstrated drawing from recent work on hurricane-like concentrated vortices and on cloud–internal wave interactions.

Whereas the classical theory of anelastic mostions by Ogura and Phillips (1962) is naturally captured in an asymptotics-based framework, its subsequent extensions, e.g., by Dutton & Fichtl (1969), Lipps & Hemler (1982), Bannon (1996), as well as Durran’s pseudo-incompressible model (1989,2008) pose a particular challenge. I will show that their systematic theoretical justification will require techniques that go beyond scale analysis and single or multiple scales expansions.

From these analytical developments one can draw a series of conclusions regarding the construction of robust numerical method for atmospheric flow simulations. The last part of the lecture will highlight how multiscale asymptotics ideas may be combined with numerical multigrid techniques to generate a novel scale-dependent time integration scheme for weakly compressible flows.

BIO

Rupert Klein holds a professorship for “Scientific Computing/Modelling and Simulation of Global Environment Systems” at the Institute for Mathematics and Computer Science at the Freie Universität Berlin. His research is characterised by the merger of applied mathematical modelling and modern computational techniques. During his 20-year academic career he has addressed problems in theoretical and computational fluid mechanics, ranging from high-speed and low-speed combustion, via the dynamics of slender vortices, to multiplescale phenomena in atmospheric flows.

Dr. Klein was born in Wuppertal, Germany, and studied Mechanical Engineering at RWTH Aachen, Germany, where he also received his doctoral degree in 1988. A two-year postdoctoral research fellowship with the Program in Applied and Computational Mathematics at Princeton University, USA, was followed by an assistant professorship with the Department of Mechanical Engineering of RWTH Aachen, Germany. His interest in environmental problems and in man-environment-machine systems led to a professorship with the department of Safety-Technology at Wuppertal University in 1995. Soon afterwards he was jointly appointed by the Potsdam Institute for Climate Impact Research, where he then headed the Data & Computation Department, and Freie Universität Berlin. This appointment placed him at the interface between climate research and modern applied and computational mathematics. In 2007 he joined the Department of Mathematics and Computer Science with a full-time appointment.

Klein was awarded the Horning Memorial Award and the Arch T Colwell Merit Award from the Society of Automotive Engineers (SAE) in 1990, the Bennigsen–Förder Prize from the state of North-Rhine-Westphalia in 1995, and the International Fellow Award from Johns Hopkins University in 1995/96. More recently he was awarded the Gottfried-Wilhelm-Leibniz-Preis of the Deutscheforschungsgemeinschaft, and became a member of the Berlin-Brandenburg Academy of Sciences. Currently, he is a member of the editorial boards of Theoretical and Computational Fluid Dynamics, Computers and Fluids, and Communications in Applied Mathematics and Computational Science.

(Cross posting from ETH Life)

The Swiss platform for High-Performance and High-Productivity Computing (HP2C) is the world’s first and only project with the aim of developing optimized scientific simulations for high-performance computers. Seismologists from ETH Zurich are also involved, namely in the «Petaquake» project.

What is the exact structure of the Earth’s interior? What are the processes that take place there? Where and how do earthquakes originate? These are some of the central questions concerning our planet that we have not yet been able to answer with certainty. A view into the Earth’s interior similar to computed tomography for a human being could provide these answers, thus helping to improve seismic risk maps. This would be an important basis for assessing the risk of the locations of nuclear power plants or hospitals in Switzerland for example.

Rread the full article on ETH Life »

These projects should focus on advances in mathematical methods, algorithms and computational techniques to better exploit next generation supercomputing platforms for the analysis and forecast of global challenges, including climate change, regional weather prediction, natural hazards, but also social challenges like major disruptions in financial markets. Given the short duration and the limitation of budget, projects are expected to focus on methodological development and testing to prove the feasibility and impact of the new methods proposed. Projects focused merely on methods and algorithms for current low-end computer platforms will not be supported.

Projects will be developed by one or more groups in a Swiss higher education institution, in close cooperation with the HP2C core group at CSCS and USI, which will provide scientific computing expertise (computational mathematics and computer science). The application developers from the project will also have access to the prototype computing hardware operated and maintained at CSCS.

Applicants are requested to notify their intention to submit an application by August 31, 2010. Complete project proposals must be submitted by September 15, 2010.

Additional information »

Have a look to the different presentations of the consortium members and in particular to the presentation of HP2C by Schulthess.

As stated in the presentation, the overarching goal of HP2C is to:

Prepare computational sciences to make effective use of next generation supercomputers

And the specific goal is to:

Emerge with several high-impact scientific applications that scale and run efficiently on leadership computing platforms in 2012/13 timeframe

In December 2009, the steering committee of HP2C selected the following 8 projects to be supported:

• Advanced Gyrokinetic – Advanced gyrokinetic numerical simulations of turbulence in fusion plasmas; Prof. Laurent Villard, EPF Lausanne
• CP2K – New Frontiers in ab initio Molecular Dynamics; Prof. Dr. Juerg Hutter, Uni Zürich
• Computational Cosmology Computational Cosmology on the Petascale; Prof. Dr. George Lake, Uni Zürich
• Selectome – Selectome, looking for Darwinian evolution in the tree of life; Prof. Dr. Marc Robinson-Rechavi, Uni Lausanne
• Cardiovascular System – HPC for Cardiovascular System Simulations; Prof. Alfio Quarteroni, EPFL
• MAQUIS – Modern Algorithms for Quantum Interacting Systems; Prof. Thierry Giamarchi, University of Geneva
• PETAQUAKE – Large-Scale Parallel Nonlinear Optimization for High Resolution 3D-Seismic Imaging; Dr. Olaf Schenk, Uni Basel
• Stellar Explosions – Productive 3D Models of Stellar Explosions; Dr. Matthias Liebendörfer, Uni Basel

A decision about 4 additional projects is expected in the next weeks.

For additional information about HP2C and its projects see »

Cray press release »