For documentation see

RaijinCL is a library for numerical computing in OpenCL, primarily focusing on BLAS.
Author: Rahul Garg, PhD student at McGill University. Work done under supervision of Prof. Laurie Hendren.
Written in C++. Essentially only depends on OpenCL.
Licensed under Apache 2.0
Currently focusing on Linux and tested on GPUs from AMD and Nvidia. 
Let me know if you are interested in sponsoring a port for a particular hardware or OS.