What's the returned memory space for "cublasSdot"?

Hello, all!

I’m using CUBLAS 3.2.

I tried to return the function call “cublasSdot” to a variable in a device memory, then I had a segmentation fault.

But return “cublasSdot” to a variable in host memory worked fine. So “cublasSdot” does NOT support return to a device memory space.

Is there any tricks to return “cublasSdot” to a device memory without copy the data from host to device?

Thanks in advance!

I believe what you are asking is available in CUDA 4.0.

Thanks! I will try to find the answer.