ISC'14

June 22–26, 2014
Leipzig, Germany

Session Details

 
Name: BoF 08: Experiences with & the Future of OpenACC
 
Time: Tuesday, June 24, 2014
10:00 am - 11:00 am
 
Room:   Hall 5
CCL - Congress Center Leipzig
 
Breaks:07:30 am - 11:00 am Welcome Coffee
 
Presenter:   Fernanda Foertter, ORNL
  Guido Juckeland, TU Dresden
  Duncan Poole, NVIDIA
  Will Sawyer, CSCS
  Thomas Schulthess, CSCS
  Nathan Sidwell, Mentor Graphics
 
Abstract:   This BoF brings together the European OpenACC user community to discuss the recent and future changes to the OpenACC specification. The focus of this BoF is the gathering of feedback from users on OpenACC 2.0 and the discussion of upcoming features (such as deep copy, performance tools API, etc) in the next release. We also plan to discuss the future road map, and in more detail, needs of the community to be considered for future releases. OpenACC has gained momentum, with support on competing architectures, highlighted by being included in the very competitive SPEC benchmark suite. Also represented will be developers working on OpenACC in GCC and other open source compiler research projects and tools for the OpenACC standard, where it is seen as a complementary alternative to coding in CUDA or OpenCL.

Target Audience
The design of super computers is being heavily influenced with power-efficient and heterogeneous node architectures. The challenge is to map scientific code to such architectures with only modest changes to the original code. HPC researchers and programmers (including Oil & Gas, Computational Chemistry) have been adopting OpenACC vigorously because it is getting easier to harness hybrid compute nodes, and substantially accelerate applications. Certain scientific for e.g. climate modeling, the scientists do not want to create and maintain several versions of the code. Similar codes from other domains have not even been explored on GPUs, fearing the time and effort needed to do so. Through the BoF, we plan to broaden the user community and make it easier for programmers to utilize OpenACC across an array of (competing) processors and accelerators.