It looks like you may not be working on this anymore. Anyway, I have an AMD card with OpenCL that I write with quite a bit. I'd like to compile your code, but you use the KernelFunctor features of OpenCL that have changed several times, essentially with every release. I've tried substituting every cl.hpp that I can find, but I get compiler errors somewhere in your usage of KernelFunctor or the argument assignment functions on every combination I've tried. I've experimented with several obvious variations on your code using different functor recommendations from the Khronos documents. The main issue there is that you set the args from the context structure which is not present when the functor is called. The way your code works is great, and I would prefer to set the args and trigger the functor separately, as you do. I may restructure your code, but perhaps you can tell me what OpenCL you used to compile.
It looks like you may not be working on this anymore. Anyway, I have an AMD card with OpenCL that I write with quite a bit. I'd like to compile your code, but you use the KernelFunctor features of OpenCL that have changed several times, essentially with every release. I've tried substituting every cl.hpp that I can find, but I get compiler errors somewhere in your usage of KernelFunctor or the argument assignment functions on every combination I've tried. I've experimented with several obvious variations on your code using different functor recommendations from the Khronos documents. The main issue there is that you set the args from the context structure which is not present when the functor is called. The way your code works is great, and I would prefer to set the args and trigger the functor separately, as you do. I may restructure your code, but perhaps you can tell me what OpenCL you used to compile.