quic-sanirudh commented on PR #16557: URL: https://github.com/apache/tvm/pull/16557#issuecomment-1951771170
Hi @tqchen, I've gone through the OpenCL code a little bit and as far as I understood, that `BufferDescriptor` metadata stored there is similar to the physical shape we use in this PR. We could refactor the code to follow a similar structure, but the main intention in this PR was to enable this `alloc_discontiguous_tensor` builtin as a runtime call to support these pointer offsets, because we would like to allocate tensors as a discontiguous chunk in VTCM memory because of its size restrictions. If it's okay, could you provide feedback on the introduction of `copy_tensor` part as well, and perhaps we can merge this one if that looks okay and we'll refactor the hexagon_device_api in a follow-up PR so that it's not blocking some related work downstream. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
