cudaMalloc and kernel call do they need to be in the same thread

[bog]

Is it necessary that memory allocation (on device) and kernel calls belong to the same host thread?

Regards,
Bogdan

[tmurray]

unless you use the driver API and the thread migration calls, yes