Prohibited and Unhandled Operations in CUDA graphs

can someone elaborate following snippet on cuda graphs

“It is invalid to merge two separate capture graphs by waiting on a captured event from a stream which is being captured and is associated with a different capture graph than the event”

this is from CUDA programming guide Prohibited and Unhandled Operations


Suppose I have a graph A that is already captured. Suppose that captured graph has an event E record/complete sequence in it.

Now suppose you are stream-capturing another graph B. You cannot have a cudaStreamWaitEvent() call in the graph currently being captured (B), where the specified event for that call “belongs” to the previously captured graph (E, in A).

1 Like

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.