Intel Is Landing More OpenMP 4.0 Functionality In LLVM Clang

In May the LLVM Clang compiler reached the state of complete OpenMP 3.1 support after the liberally licensed compiler saw its OpenMP support stagger for so long out-of-tree. Intel had been leading the recent efforts for getting OpenMP working in Clang and they've continued in working towards OpenMP 4.0 compliance.

The full OpenMP 3.1 support in Clang will be found with the next release, LLVM 3.7, and there's also some work of OMP4. While the OpenMP 4.0 support isn't yet complete, Alexey Bataev of Intel Russia has continued landing various functionality into the mainline code.

Today the latest OMP4 features for Clang are support for the "omp for simd" directive, support for the "omp parallel for" directive, "#pragma omp taskgroup" directive, and code generation for OMP 4.0's proc_bind clause. This is on top of other recent commits by Bataev.

OpenMP 4.0 brings support for accelerators, atomics, error handling, thread affinity, tasking extensions, SIMD support, and more.

I haven't seen it communicated publicly when they hope to have full OpenMP 4.0 support in Clang, but hopefully it will manage to become a reality this year. GCC 5 is already shipping OpenMP 4.0 with offloading support and is already working on OpenMP 4.1.