OpenACC (OpenMP-Like GPU Accelerator)

http://www.openacc.org/ https://gcc.gnu.org/wiki/OpenACC

This seems like a god-send for those of us which need somewhat portable code that's parallel. OpenCL is fantastic... if you are writing a new program from scratch, it is very tough (in my experience) to fit the OpenCL paradigm in legacy code.

Does anyone know if any compiler has successfully implemented a 'performant' solution of the specification? Is there any independent analysis of the GCC6 implementation worth reading?