5-6 March 2014
Maison de la Simulation
CET timezone
Subjects covered:
Threading, accelerators

Course Description:
OpenACC is an emerging standard for parallel computing designed to ease the use of accelerators like GPU, or coming many-core processors; OpenACC is real-world implementation of parallel programming concepts for accelerators that could be part of a future OpenMP standard.
  • presentation of the OpenACC programming model
  • hands-on session using CAPS/HMPP compiling toolchain
Hands-on sessions will use a local cluster at Maison de la Simulation. The OpenACC part of the course will be given by CAPS Entreprise.

Learning outcome: 
After this courses, participants should know the syntax of OpenACC directives, how to apply them to obtain efficient parallel codes. They will also understand the importance of memory layout.

Good knowledge in C or Fortran programming language, some knowledge about parallel programming.
GPGPU programming knowledge is not required.

    March 5
        Morning – CUDA Basics
  • Introduction to GPU computing
  • Nvidia GPU architectures
  • CUDA programming model
  • CUDA debugging
Afternoon – OpenACC (1/2)
  • OpenACC Overview & compilers
  • OpenACC Programming Model
  • Managing data with OpenACC
March 6
Morning – OpenACC (2/2)
  • OpenACC loop constructs
  • Asynchronism with OpenACC
  • OpenACC runtime API
        Afternoon – OpenHMPP Introduction
  •  What is in HMPP , what is not in OpenACC
  •  OpenHMPP compilation model
  •  Driving the code generation & gridifications
  •  Automatic unroll, split, jam, etc. transformations
  •  Computing on multiple GPUs
  •  OpenHMPP API for C, C++ and Fortran

Maison de la Simulation
Digitéo labs/Bât 565 CEA/Saclay 91191 Gif-sur-Yvette France
More info about Maison de la SImulation: http://www.maisondelasimulation.fr Getting there: http://www.maisondelasimulation.fr/Access/accessToMaisonDeLaSimulation.html List of hotels: http://www.maisondelasimulation.fr/Phocea/Page/index.php?id=53