The OpenACC 1.0 API has been public for more than one year as an open standard to simplify parallel programming on CPUs and GPUs, but to this point it's basically only backed by commercial compilers. OpenACC is similar to OpenMP in terms of using PRAGMA compiler directives and special functions for tapping multiple CPU cores in an easy and straight-forward manner with C/C++ and Fortran code, but unlike OpenMP, OpenACC is also aware of GPUs.
The main compilers supporting OpenACC 1.0 are PGI, Cray, and CAPS/HMPP. Even with the forthcoming releases of GCC 4.8 and LLVM/Clang 3.2, there's still no OpenACC support. (LLVM/Clang is also still lacking mainline support for OpenMP.)
Introduced last month at the Super Computing '12 conference in Salt Lake City, a draft version of OpenACC 2.0 was also made public for review. OpenACC introduces new controls for handling data movement, better handling of unstructured data, improvements in dealing with non-contiguous memory, support for explicit function calls, and other new functionality to make happy HPC programmers. The tentative OpenACC 2.0 specification can be found on the standard's website.
Hopefully in 2013 we will begin to see open-source compilers like LLVM/Clang and GCC beginning to support OpenACC for exposing new parallel computing capabilities. On a semi-related note, last month Intel showcased Shevlin Park, an effort to make Microsoft's C++ AMP (Accelerated Massive Parallelism) multi-platform by porting it to LLVM/Clang.