Are kernels with zero thread handles cleanly?

Accelerated Computing CUDA CUDA Programming and Performance
1

Does the CUDA runtime properly “do nothing” if a kernel is invoked with zero threads, i.e. blocksize == 0?

I’d like to avoid a test for zero, if CUDA is going to do that anyway.

2

Passing a block or grid size containing a zero for any dimension will generate an execution parameter error IIRC.