Kernel load-store vectorization

I would try the upgrade first since when I was using update 3 I had a hard time creating a case that would cause that warning to occur. That early return would cause some of the vector lanes to have nothing to do so that's most likely why you saw the warning. The compiler might have been modified to let those flow through to the end of the kernel by just putting conditionals throughout the kernel behind the scenes which would allow the vectorization to remain intact and efficient so after you upgrade you'll be able to check to see if that's the case. If that's not the case then you would need to refactor the code a bit so that all the work-items exit the kernel in the same spot (at the end)