Any time there is a CUDA device class that has virtual methods,
nvlink will issue a warning if the resulting library is linked
into code that does not use that method, so we basically have
to suppress these warnings for any execution environment class
with virtual methods.