Struct NVCVWorkspaceRec
Defined in File Workspace.h
Struct Documentation
-
struct NVCVWorkspaceRec
Aggregates multiple resources into a single workspace objects
Public Members
-
NVCVWorkspaceMem hostMem
Plain host memory. This should not be used in any GPU code.
On systems with a discrete GPU, this kind of memory doesn’t need a CUDA event. On systems with integrated GPU or HMM systems, there’s no difference between plain and pinned host memory with respect to synchronization.
-
NVCVWorkspaceMem pinnedMem
Pinned host memory.
cudaXxxAsync operations on this kind of memory are performed truly asynchronously, which calls for synchronization. When used as a staging buffer for passing data to a CUDA kernel, a typical synchronization scheme would be to wait for the
ready
event on host (cudaEventSynchronize), issue H2D copy and record theready
event.
-
NVCVWorkspaceMem cudaMem
GPU memory
-
NVCVWorkspaceMem hostMem