As I know, we can allocate a Pinned memory area within kernel memory. (From KGPU)
Then, allocate linux kernel data in Pinned memory and transfer to GPU.
But problem is that linux kernel data should be arranged as array.
Today, a case that is a tree.
I have tried pass it from Pinned memory to GPU.
But when a node access next node, memory access error occured.
I was wondering is Unified Memory can be allocated as Pinned memory area in kernel memory?
So tree can be builded in Unified Memory area and used by GPU without other runtime API like cudaMallocMaganed.
Or is that Unified memory must only use cudaMallocMaganed?