For the past year Code Sourcery / Mentor Graphics has been working with NVIDIA to bring OpenACC 2.0 support to GCC and to allow for this heterogeneous parallel programming API to be taken advantage of with NVIDIA GPUs from GCC. This work is closer to finally being realized for allowing OpenACC programs to be compiled with GCC and target NVIDIA GPUs on Linux.

Julian Brown of Code Sourcery is seeking approval to merge the OpenACC 2.0 runtime support and NVIDIA PTX/CUDA plug-in into their GOMP4 branch of libgomp, the runtime library by GCC traditionally used for OpenMP. The OpenACC 2.0 support is being merged into libgomp since the OpenACC support is built mostly on top of GCC's OpenMP 4.0 support. The added plug-in support allows for the OpenACC runtime support to target NVIDIA GPUs via PTX/CUDA.

More details via this mailing list post. Those interested in NVIDIA's PTX back-end for GCC can read this earlier article. This work will hopefully all materialize in mainline in time for next year's release of GCC 5.