OpenACC 2.0 Is In Good Shape For GCC 6, Complete With NVIDIA Offloading

On top of many other new features and various improvements to GCC 6 is also maturing OpenACC 2.0 support.

OpenACC is the parallel computing standard for heterogeneous CPU/GPU systems. Initial OpenACC 2.0 support was added to last year's GCC 5 release while now there is much better support in GCC 6. CodeSourcery has been leading the OpenACC/NVPTX charge in GCC for the past few years in cooperation with NVIDIA Corp.

Thomas Schwinge of CodeSourcery has provided an OpenACC status update in time for the imminent GCC 6.1 release. There is host-fallback execution support with this OpenACC 2.0a implementation, offloading is working for NVIDIA GPUs via the NVPTX, x86_64 and PowerPC 64-bit LE architectures are supported, and there is initially support for parallelized execution of OpenACC kernel constructs.

More details on the OpenACC state in GCC 6 can be found via this documentation patch. GCC 6.1 should be released within the next few days.