Is it possible to load two kernels from different files and then have them be able to call each other?
When using cuModuleGetGlobal, how does cuda know which device Im refering to?
No, a kernel can only be launched by host code – not by another kernel.
Presumably it is the device associated with the current context.