I noticed that there is a statement about Xavier like “GPU can access to System Memory via CPU Cache” in some references, for example, p27 in “https://on-demand.gputechconf.com/gtc/2018/presentation/s8868-cuda-on-xavier-what-is-new.pdf”. Can I rely on this feature for data coherence between CPU and GPU？I’m curious about that In what scenarios can this feature be used？
Actually，I need some communication between the GPU kernel and the CPU thread in an application. If I can’t rely on that feature, is there any other mechanism work for that?
Please noticed that the file you shared is for DRIVE system rather than Jetson platform.
Xavier do support IO coherence memory. It allow GPU to snoop CPU cache but not vice versa.
You can find some information about Jetson memory here:
The main concept is that the memory can shared via CPU and GPU but cannot be accessed concurrently.