I am trying to develop a multi-threaded application that accesses CUDA from 2 threads. I use cuCtxPopCurrent and cuCtxPushCurrent functions to pass the CUDA context between the threads.
My understanding is that I need Driver API for this, Runtime API will not do the trick.
Is it possible to call CUFFT from one of the threads? NVIDIA CUDA C Programming Best Practices Guide explicitly states that it is not (p. 1.3.3).
However, CUDA Toolkit 3.0 beta announcement promised “CUDA Driver / Runtime Buffer Interoperability, which allows applications using the CUDA Driver API to also use libraries implemented using the CUDA C Runtime”.
I still cannot make it work under ver. 3.0 beta (cufftExecC2C returns error code 1).
Does anybody now how to do it?
You can’t use context migration with the runtime API (which CUFFT uses) at this time.
Thank you. Hope it will be there in final version.
Hello, since I urgently need this feature can you tell when this will enter a next version?