Because CUDA threads and the host process are all in the same share group, you can create pending or sliding breakpoints on source lines and functions in the GPU code before the code is loaded onto the GPU. If TotalView can’t locate code associated with a particular line in the source view, you can still plant a breakpoint there, if you know that there will be code there once the CUDA kernel loads.

Pending and sliding breakpoints are not specific to CUDA and are discussed in more detail in "Setting Source-Level Breakpoints".