Topics of Interest
In HPC software development, the top priority is always given to performance. As system-specific
optimizations are almost always required to fully exploit the potential of a system, application
programmers usually optimize their application programs for particular systems. Whenever the target
system of an application program is changed to a new one, thus, they need to adapt the program to the
new system. This is so-called legacy HPC application migration. The migration cost increases with the
hardware complexity of target systems. Since future HPC systems are expected to be extremely massive
and heterogeneous, it will be more difficult to afford the migration cost in the upcoming post-Petascale
era. Therefore, this special session, LHAM, offers an opportunity to share practices and experience of
legacy HPC application migration, and also discuss promising technologies to reduce the migration cost.
Workshop Date: September 27, 2013
"Ease of Porting Parallel HPC Codes to Intel Xeon Phi"
Michael McCool, 10:35-11:20（45min)
The Intel Xeon Phi co-processor runs Linux and supports many existing
parallel and distributed software development systems, including OpenMP and MPI.
The Intel compiler also supports offload functionality so that existing computations
running on a host processor can be annotated to offload parts of this computation to the Xeon Phi.
Both of these approaches can be used to develop Xeon Phi applications with
very little modification of existing code.
However, applications may still need to be tuned for the best performance.
The Xeon Phi has a very large number of cores and a wide vector width,
and applications need to make good use of both of these features to get the best performance.
Intel also provides a number of tools to analyze performance and identify bottlenecks.
In this talk I will go over the features of the Xeon Phi, will discuss the parallel
and distributed programming systems that work with it, and then will present some case studies of
applications that have been ported and tuned for the Xeon Phi.
Michael McCool is Intel Principal Engineer. He has degrees in Computer Engineering (University of Waterloo, BASc)
and Computer Science (University of Toronto, M.Sc. and PhD.) with specializations in mathematics (BASc) and biomedical
engineering (MSc) as well as computer graphics and parallel computing (MSc, PhD). He has research and application experience
in the areas of data mining, computer graphics (specifically sampling, rasterization, path rendering, texture hardware, antialiasing,
shading, illumination, function approximation, compression, and visualization), medical imaging, signal and image processing,
financial analysis, and parallel languages and programming platforms. In order to commercialize research work into many-core
computing platforms done while he was an Associate Professor at the University of Waterloo, in 2004 he co-founded RapidMind,
which in 2009 was acquired by Intel. Currently he is a software architect with Intel working on parallel programming languages,
applications, and mobile computing. In addition to his university teaching, he has presented numerous tutorials at Eurographics,
SIGGRAPH, and SC on graphics and/or parallel computing, and has co-authored three books. The most recent book, Structured
Parallel Programming, was co-authored with James Reinders and Arch Robison. It presents a pattern-based approach to parallel
programming using a large number of examples in Intel Cilk Plus and Intel Threading Building Blocks.
"Software Engineering of Scientific Applications for Portability,
Evolvability, and Performance"
Shirley Moore, 14:50-15:35(45min)
Execution environments and scientific applications are both in a period of rapid evolution
and are becoming increasingly complex. High performance architectures are evolving
towards combining many processors with diverse architectures ranging from multicore
chips through SIMD accelerators. Applications must execute effectively on a diversity of
architectures over their lifetimes. The increasing diversity and complexity in execution
environments and application codes is leading to a steep increase in cost and effort
in attainment of portability, evolvability and performance. Design and development
methodologies appropriate for life-cycle use with complex long-lived application systems
such as software architectures and component-based software engineering are now
well-established in mainline computer science. There exist methods and tools that when
extended and combined can yield a tool chain for partial automation of the necessary code
transformations. Adapting and extending the software engineering methodologies and
developing a tool chains for life-cycle development of complex scientific application codes
has the potential to yield a paradigm change in how these codes are mapped onto emerging
complex architectures. We describe initial steps towards establishing such a tool chain as
well as initial successes in restructuring of legacy application codes.
Shirley Moore is an Associate Professor of Computer Science at the University
of Texas at El Paso (UTEP). She is also a core faculty member in the graduate
interdisciplinary Computational Sciences Program at UTEP. She received a PhD
in Computer Sciences from Purdue University in 1990. Her research interests are
in software engineering, performance modeling, and performance optimization
of scientific applications and in hardware-software co-design. She is a Principal
Investigator on U.S. Department of Energy and Air Force Office of Scientific Research
funded projects in these areas.
"Towards an Extensible Programming Environment for Software Evolution"
Hiroyuki Takizawa, 25min
"Experience of Implementing Parallel FFTs on GPU Clusters"
Daisuke Takahashi, 25min
"The Future of Accelerator Programming:
Abstraction, Performance or Can We Have Both?"
Kamil Rocki, 25min
"An HPC Refactoring Catalog; Guidelines to Bridge The Gap between HPC Systems"
Ryusuke Egawa, 25min
General Chair: Daisuke Takahashi (University of Tsukuba)
Hiroyuki Takizawa (Tohoku University)
Reiji Suda (The University of Tokyo)
Ryusuke Egawa (Tohoku University)
Shoichi Hirasawa (Tohoku University)
Michael Resch (HLRS)
Wen-mei Hwu (University of Illinois at Urbana-Champaign)
Chisachi Kato (The University of Tokyo)
Basic Research Programs: CREST Development of System Software
Technologies for post-Peta Scale High Performance Computing.
"An evolutionary approach to construction of a software development environment for massively-parallel heterogeneous systems"
E-mail: lham2013 .at. xev.arch.is.tohoku.ac.jp (replace ".at." by "@" in the email address)