tqchen left a comment (apache/tvm-ffi#895)
It would be first good to understand what is the overall convention, in
general, i think we would need the following elements (consistently across
frameworks)
- A global index to device, so when we do `xpu:index` we know what the device
is and being consistent across the frameworks
- Would be good to know what is Torch's convention
- Would also be good to know what Intel's general convention
- A consistent way to know what is stream, in this case seems queue is the way,
we generally don not add extra wrapprs like GetSYCLQueueFromDevice and instead
just let downstream do the cast
--
Reply to this email directly or view it on GitHub:
https://github.com/apache/tvm-ffi/issues/895#issuecomment-6588805663
You are receiving this because you are subscribed to this thread.
Message ID: <apache/tvm-ffi/issues/895/[email protected]>
