I seriously doubt that you'll see an "official" push to add an OpenCL API to the core of perl(either 5 or 6). It's just not something the majority of perl applications would need to make use of.

What you might see eventually is perl bindings for OpenCl in the vein of PyOpenCL. But for that to happen you'll need an individual, or a group of individuals to be inspired to actually write those bindings.

It might be worth asking this on the PDL mailing lists. PDL is a project that involves the efficient manipulation of multidimensional arrays in perl. So people working with that might be inspired to experiment with GPU programming.

CUDA is also supported on MacOS as well as Windows and Linux. It's free-as-in-beer and, IMAO, much easier to use than opencl. The downside, of course, is that it runs only on nVidia hardware.

I'm teaching myself CUDA programming but it's early days yet. When I've reached the stage of being able to do something useful to people other than myself I'll share it. There's a good chance that a Perl interface to the "something useful" will be one of the ways in which it will be shared. At the moment, I envisage something rather closer to the use of GMP in Math::BigInt than, say, the complexity of Perl/Tk. Perhaps a group could be started to discuss what a Perl interface should look like and to build a preliminary specification.

Paul