pitrou commented on PR #34972: URL: https://github.com/apache/arrow/pull/34972#issuecomment-1507166234
Thanks @gmarkall for the detailed insight! (also, hi :-)) > I also think it would also have been more difficult to have the consumer supply the stream for CAI, because it is just a property containing a dict This is also how the C Data Interface works, though of course it's a C struct rather than a Python dict. This is also why I would originally prefer the CUDA stream to be provider-supplied, because it can be part of that interface... So, to sum it up, we have: * in favour of the producer-supplied CUDA stream approach: allows to expose the CUDA stream in the C Data Interface (not only the C Stream Interface, which is more specialized - for example, it's sync not async) * in favour of the consumer-supplied CUDA stream approach: allows the consumer to ignore synchronization issues entirely, especially by passing a default stream Am I missing something in this bullet list? -- 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]
