As heterogeneous node architectures are becoming more prevalent for mainstream computing systems, the requirement for simple and platform portable programming methods manifests in a number of different programming strategies. Application developers that want to migrate existing to or develop new applications for these novel platforms, are looking for best–practice approaches that offer a reasonable results in a short period of development time. As a result domain specific languages as well as directive based programming methods are welcome strategies when targeting heterogeneous system architectures, since they can abstract the underlying hardware specific details of low–level programming approaches and have shown to offer very good productivity. The holy grail for all such strategies is to achieve not only a platform portability but also offer an application performance the is comparable to the low–level programming models on multiple platforms without code changes. The proposed workshop will solicit papers and presentations both on the methodology of these approaches as well as best–practice experiences from early adopters.<br>Topics of interest for workshop submissions include (but are not limited to):<br>Directives–based programming models for current and emerging systems<br>Ideas and implementations for language extensions<br>Language extensions or programming strategies for memory hierarchies<br>Interoperability and Composability with low–level APIs<br>Users’ experiences and implementations<br>Scientific libraries interoperability with directive–based programming models<br>Exploring high–level prescriptive or descriptive programming approaches<br>Hybrid programming approaches (OpenMP + MPI, MPI +X, MPI + OpenACC and so on)<br>Task–based programming (asynchronous execution, scheduling)<br>Translating to low–level API, design approaches and challenges<br>Experiences developing applications for current and emerging systems<br>Challenges in maintaining a portable yet single code base<br>Debugging, performance evaluation and benchmark studies<br>Energy/Power–aware compiler and runtime optimizations for current and emerging systems<br>
Abbrevation
HPM4MMS
City
Prague
Country
Czechia
Deadline Paper
Start Date
End Date
Abstract