Closed PhilMiller closed 9 years ago
Original date: 2014-07-28 22:57:05
Forgot one, now added above.
Original date: 2014-10-14 16:13:13
Most of these have been easy to resolve by extending an existing exclusion zone (typically CmiRank==0) to include the adjacent initialization. A few might require the introduction of stronger ordering methods.
Original date: 2014-11-09 16:56:51
Ping. We'd like to have a fix for this merged before the core meeting. This should be fairly simple, and would make a substantial difference for people who want to use ThreadSanitizer.
Original date: 2014-11-10 16:49:58
I have most of these fixed. Should it go in one commit?
Original date: 2014-11-10 17:06:57
One commit should be fine, unless there's something about some particular change that might need more attention
Original date: 2014-11-11 19:34:05
Most of these were pretty straightforward to implement, but there was a lot of code reading to determine the minimal safe intervention. Also found that fixes for the initialization of conv-ccs made the sockroutines problem go away and had to go digging to make sure that was correct.
Original date: 2014-12-02 18:51:24
I ran into a bug with the isomalloc fix that may take some time to unravel, that was where I left off with this before SC14. I haven't had a chance to get back to it yet.
Original date: 2015-02-10 21:20:52
The NullLB fix isn't quite right, since other threads are still going to read from _theNullLB
, and they'll just see a zero group ID, which causes a crash. I guess that assignment should be locked rather than gated on rank == 0.
Original date: 2015-09-01 18:17:03
This was fixed months ago.
Original date: 2015-11-11 03:30:57
https://charm.cs.illinois.edu/gerrit/429 https://github.com/UIUC-PPL/charm/commit/be9a0cef3e724e97a3abbc5bf19cd8ca53670cb5
I'll open a separate issue to follow up on my comment about potential problems with _theNullLB
Original issue: https://charm.cs.illinois.edu/redmine/issues/538
These probably need locks or checks for rank-0 only