03-30-2012, 11:42 PM
(03-30-2012, 10:49 PM)gat3way_ Wrote: atom, did you try vectorized kernels like those for sm_21? From what I understood, 680GTX may turn out (from OpenCL perspective) something very much resembling a VLIW card, just in another way. I mean at least uint4 for best resource utilization
yes its with vectorized code (uint2)