Are the NPP 10.1 npp**_Ctx functions asynchronous in the same way that functions like cudaMemcpyAsync are?
I don’t think this would be correct across the board.
For example, an NPP get buffer size function call is returning data to the host. That data must be valid when the CPU thread proceeds to the next line of code. It cannot be fully asynchronous.
There may be other examples. I wouldn’t be able to provide a list of all the synchronous or asynchronous characteristics of NPP functions.
One of the fundamental purposes and motivations of the NPP stream context is to help the NPP library move towards fully stateless operation. This is not the same thing as synchronous/asynchronous.
In my case I am interested in the image conversion functions nppi**_Ctx, like resize, format conversion.
These sound like operations that can be asynchronous. Are most of them?