Setting host memory via an hostnode before a memcopy node to device is not reflected in the device kernel execution

[Robert_Crovella]

my expectation is that is a doc oversight.

If its of concern you can can always request CUDA documentation updates by filing a bug.