Hi guys: I have a question about ndarray, if I get my ndarray through ndarray = m_graph_runtime->GetOutput(index). Suppose this ndarray’s device type is GPU context which is KDLGPU, when I was trying to get auto* dltensor = ndarray.operator->(); I found dltensor-> data is just null pointer, and it cannot copy data from ndarray to dltensor unless I change ndarray’s context to cpu context before executing auto* dltensor = ndarray.operator->(). Can someone tell me why? What if I want to directly get a cuda buffer from Ndarray.operator->()? Is there some way I can do that? Thank you !!!
Hi, have you solve the problem?I met the same problem with you, looking for your advice!