I have a very complex kernel, but respecting all the OpenCL 1.0 limits ( I use 1 texture sampler, 1 texture, 1 constant block of 416 bytes and 7 kernel arguments ).
My kernel executes ok in both NVIDIA and ATI cards(5XXX).
However, when I paste the CL code into the SKA sometimes ( not always ) displays the register and throughput as N/A as you can see here:
If I simplify a bit the CL code then the SKA outputs the values correctly.
ps: Why I cannot attach/link images???