Is it possible to use an NDRange to parallelize a strlen() function for a dynamically allocated character array? This unknown size doesn't seem to fit with the buffer model in OpenCL. What happens if I try to create a buffer longer than host_ptr? How would it know, and when? What if I use CL_MEM_USE_HOST_PTR? What happens if inside the kernel it accesses beyond the bounds of host_ptr? Has anyone tried this before in OpenCL?