Yes, makes sense... These units are accessible via (undocumented) intrinsics in some OpenCL implementations (or in GLSL), but, unfortunately, there is no portab

Yes, makes sense... These units are accessible via (undocumented) intrinsics in some OpenCL implementations (or in GLSL), but, unfortunately, there is no portable solution. And the FP precision requirements in the OpenCL standard are way too high, even for the FP16 extension.