Quoting from openmp.org: OpenMP, the de-facto standard for parallel programming on shared memory systems, continues to extend its reach beyond pure HPC to include embedded systems, real time systems, and accelerators. Release Candidate 1 of the OpenMP 4.0 API specifications currently under development is now available for public discussion. This update includes thread affinity, initial support for Fortran 2003, SIMD constructs to vectorize both serial and parallelized loops, user-defined reductions, and sequentially consistent atomics. The OpenMP ARB plans to integrate the Technical Report on directives for attached accelerators, as well as more new features, in a final Release Candidate 2, to appear sometime in the first Quarter of 2013, followed by the finalized full 4.0 API specifications soon thereafter.
Expect big news on OpenMP 4.0 for next week’s SC12. The OpenMP Language Committee – responsible for developing the standard – always planned to release the next version of the standard as a draft for public comment in time for SC12. We worked very hard during the last weeks to stay within our schedule. And we will do the following:
Release OpenMP 4.0 RC1 as a draft for public review. This document will be in a pretty good shape and will represent the foundation of OpenMP 4.0. It will contain several new features, to be discussed and explained during SC12 at our booth and/or the OpenMP BoF. Among these new features is the SIMD construct, to vectorize both serial as well as parallelized loops, taskgroups (no task dependencies yet), thread binding via places (I talked a lot on this already), array sectioning, basic support for Fortran 2003, and some other minor corrections and improvements.
Publish a Technical Report on OpenMP for Accelerators, more specifically on “Directives for Attached Accelerators”. This was always planned to be the major addition for OpenMP 4.0. However, integrating support for accelerators with the rest of OpenMP is a hard task and a lot of work, and it is not 100% done yet. There were many discussion on how to deal with this situation: do as outlined here, wait for just some more weeks, come up with a completely new schedule and wait until we are completely done, … . Almost all technical aspects have been discussed and answered. But the wording is not yet completed. And support for NVIDIA-like GPUs might not be optimal. However, I personally think the proposal is really good and the big opportunity in making the current state of work public is that the HPC community can take a look at it, think about it, comment on it, and possibly improve it. It is already online: http://openmp.org/wp/openmp-specifications/.
Hoping for constructive feedback and taking the additional time to work on the OpenMP for Accelerator extension, the current plan is to come up with a second draft for public comment (RC2) in January 2013 and then finalize the standard quickly after. Quickly in terms of a few weeks. This plan is still ambitious, but I think this is a good plan.
If you want to learn more, come to the OpenMP booth, and come to the BoF on Tuesday afternoon, 17:30h, which unluckily I will not be able to attend myself :-/. Listen to what the people will show you and let us know what you like and what you dislike.