Reputation: 512
cudaMemcpy in a device only supports device-to-device. I need to copy from device global to host global memory ( doesn't matter sync or async, although async is preferable ).
I would really appreciate your help!
Upvotes: 1
Views: 572
Reputation: 72372
Is there something equivalent to
cudaMemcpy
from device global memory to host global memory that I can call within a device function?
No.
As pointed out in comments, your only choice here would be to use mapped or managed host memory which can be directly addressed in device code.
Upvotes: 2