SDAG tracing counts time waiting for a when to be satisfied as overhead
When a when block is unsatisfied, SDAG still traces it as SDAG overhead, even though at this point the runtime is free to do other things. Tracing of SDAG overhead should stop when a when goes unsatisfied.
- Status changed from New to Implemented
Fix has been implemented. Needs to undergo testing.
I don't see the patch posted on Gerrit. Could you push it and post a link?
Could we get that patch posted for review & integration? It'd be nice to have it integrated before 6.7, if it's already done.
Ping - please post the patch.
Working on putting together the patch.
Ping. Is there a fix for this? Even if it's only partial, we can hand that off to someone else to finish up.
- Target version set to 6.7.1
Ping, again. Is there actually fixed code for this sitting in a repository somewhere, or is that still a work in progress?
What's the status of this?
Jonathan said he has a partial fix but that it wasn't entirely correct, so he'll look at it again.
- Status changed from Implemented to In Progress
- Target version changed from 6.7.1 to 6.8.0
- Assignee changed from Jonathan Lifflander to Eric Mikida
- Category set to Tracing
- Status changed from In Progress to New
- Tags set to SDAG, tracing, charmxi
- Status changed from New to Closed
I can't seem to reproduce this bug, and the code looks pretty sound from what I can tell. The only way TRACE_EXECUTE_BEGIN is called with SDAG_RTS as the entry point will always be from within an invokeEntry call. These calls end with the corresponding TRACE_EXECUTE_END() from either ck.C:651 or cklocation.C:1981.
Also available in: Atom