OpenCL spec 1.2 is already there and NVIDIA has no plan so far to adopt it.
I am planning to extend an existing API for OpenCL. Since i am provided with the nvidia resources for the development, how do i make the API compatible for old, current and future opencl specs ?
I think this is an open question to every API design.
Some guidelines will be helpful from the forum.
Look at cl.hpp as an example. We have tried to build this support in. You have to be a little careful to do both compile-time checks (which let it link against the right opencl.dll) and also run-time checks that account for a 1.1 device hiding behind a 1.2 linked OpenCL library.