How to ensure that a child kernel finished processing before the parent kernel continues in Cuda

I am learning Cuda and decided to do a basic image box blur demo as a way to get familiar.