Open kayceesrk opened 9 years ago
Also this example sometimes does not terminate... Anyway it is clearly non-deterministic. That's a problem...
Does this bug still occur with the new runtime?
Yep. Also, the program gets stuck at Fetching Sums
:
Domain [0], Worker [9]: Dequeuing 9998
Domain [0], Worker [9]: Dequeuing 9999
Domain [0], Worker [9]: Posting 9499500
Fetching sums
(stuck forever)
Ok. About the broken invariant, it seems that the lock freedom assumption on the CAS is wrong. (because of the current kCAS implementation) right now, we work around by changing the queue implementation, but that is WRONG.
Intermittent failures on
test/sum_queue
under the multicore scheduler.