LLVM Clang Is Moving Closer To Full OpenMP 4.5 Support

[phoronix]

Phoronix: LLVM Clang Is Moving Closer To Full OpenMP 4.5 Support

While it took LLVM's Clang C/C++ compiler initially a long time to supporting OpenMP, the code continues to mature in supporting the latest updates to this parallel programming specification...

http://www.phoronix.com/scan.php?page=news_item&px=Clang-OpenMP-4.5-Closer

[davidbepo]

good, this is one of the last clang weak points left

[coder111]

What's the state of OpenACC in GCC and llvm? In theory, OpenACC should be more useful for compute workloads, as it should be able to offload work to GPU as well as the CPU.

[elldekaa]

What's the state of OpenACC in GCC and llvm? In theory, OpenACC should be more useful for compute workloads, as it should be able to offload work to GPU as well as the CPU.

Since its version 4.0, the OpenMP standard supports offloading to GPU/DSP (or any other accelerating device). The problem is then to add the support of the compilers... Clang currently supports CPU offloading (for prototyping but also used by IBM for powerpc acceleration in their supercomputers) and will support offloading to NVidia GPUs (cuda) when this patch is merged: https://reviews.llvm.org/D14254 (almost ready). There is several other experimental fork to support OpenMP offloading for various type of devices, like OpenCL devices: https://omp2ocl.github.io/aclang/

I think there is no support for OpenACC in Clang but there is a new project trying to transform OpenACC into OpenMP code: http://lists.llvm.org/pipermail/cfe-...er/056222.html
In my opinion, OpenACC is going to die slowly to let OpenMP take the place...