No, because the calls will always be the same: all threads (with 1.x half- computing power) warp will simultaneously read the same parameter, which is transmitted to all threads in one transmission.
For completeness, I mentioned that only the computing power of 1.x devices stores parameters in shared memory. Higher computing capabilities store them in read-only memory, where the same considerations of uniformity apply.
tera
source share