Hi all!
I have a simple question regarding the CUFFT library.
I couldnt manage to find anything, telling which stream CUFFT will use, when executing a plan. I know this question is somewhat rhetorical and that using any other than the zero-stream does only make sense when using multiple ones to overlap kernel execution and memcopies.
The reason I ask is following:
Consider you have a kernel launched before executing your CUFFT plan. Now do you have to call a syncfoo() in order to wait for this kernel to finish in case that otherwise both kernels might run in parallel on future hardware or is this neglectable?
I didnt see any undesired behavior in my apps without the sync and would really wonder if CUFFT would make use of anything else than zero-stream. But just for being able to correctly test my app and to ensure this might never happen (some funny guy at nvidia could have made it use stream 42 or sth like that ;) ) it would be nice if things like this would be stated somewhere in the docs etc (working on radar signal processing where you have some strict guidelines).