Blog post by: Jim Cownie, principal engineer, Intel Corporation (UK).

The first UK OpenMP Users’ Group Meeting took place at St Catherine’s College Oxford on May 21st and 22nd.  Over 40 people attended, and the ~50% of people who filled in feedback forms were positive about the experience: all would recommend it to colleagues, and rated it as either “Excellent”, or “Very Good” (the top two possibilities on a five point scale).

So, if you weren’t there, what did you miss!?

uk-openmp-users-2018-attendeees

 

Day 1 – Tutorials, Workshops and Panel Discussion

The first day consisted mostly of tutorials; Mark Bull presented his “Hands-On-Introduction to OpenMP”, Michael Klemm and I presented “Advanced OpenMP: Performance and 4.5 Features”, and Simon McIntosh-Smith gave his “Programming Your GPU with OpenMP: A Hands-On Introduction”.

We finished the technical part of the day with a panel session “OpenMP is Twenty: Where is it Going?” which allowed participants to question members of the OpenMP ARB.  Feedback from this included

  • How do I find where to report bugs in OpenMP compilers?
    • Answer: We’ll investigate adding a bug-report email (or link) to the compiler listing at OpenMP.org
  • OpenMP provides good control of vectorization, but when dealing with conditions inside vectorized code, there is no way to ask the compiler either to generate masked vector code, or, instead, introduce control flow divergence. Please will the compiler vendors and language committee consider the value of such a directive?

In the evening we reassembled for a fine dinner at St Hilda’s college which allowed participants to discuss OpenMP (and other subjects) with the speakers and OpenMP ARB members.

Day 2 – Keynote and Technical Sessions

We started our second day with a keynote from Paul Selwood of the UK Met Office.

uk-openmp-users-2018-met-office-talk

We followed that with technical presentations

  • ParaFormance: Democratizing Parallel Software Development
  • Fortran 2008, 2018 Coarrays and OpenMP
  • Reducing the  Burden  of  Parallel  Loop  Schedulers  for  Many-Core Processors
  • Implicit Mapping of Pointers Inside C++ Lambda Closure Objects in OpenMP Target Offload Regions
  • Heterogeneous and Distributed Programming in Tomorrow’s C++ Today
  • The Importance of Performance Analysis in OpenMP Code
  • Tuning OpenMP Codes with Intel® Vtune Amplifier Performance Analyzer
  • The TORUS radiative transfer code: OpenMP implementation and performance
  • Enabling OpenMP for the Arm architecture

Conclusions

There is a lively OpenMP community in the UK, which seems to value the chance to get together here (rather than in the US) to discuss what we are doing, share knowledge, and provide feedback to the OpenMP language designers.

If you missed it this year – come next year. The next event is provisionally scheduled to take place in late May in Edinburgh.  More details to follow.

Thanks to the Sponsors

The event would not have been possible without the tremendous support from the sponsors.  Thanks to all of them.

 

uk-openmp-sponsors-201800405