You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The recently added (revamped) jacobi_smp exposes a problem in the HPX thread scheduling system. There appears to be a deadlock in hpx/runtime/threads/policies/thread_queue.hpp line 470. In the application, all worker threads eventually wait on the same mutex which is tried to get locked here. The cause for this unclear to me (might be due to some race condition between spawning new threads from within hpx threads and the scheduling loop).
When changing the mutex type of the local_priority_scheduler to hpx::lcos::local::spinlock the deadlock in this particular line disappears, but another deadlock in cleanup_terminated (in the same file) appears to happen.
Something is fishy within the scheduler.
To reproduce the problem, run the following command:
The recently added (revamped) jacobi_smp exposes a problem in the HPX thread scheduling system. There appears to be a deadlock in hpx/runtime/threads/policies/thread_queue.hpp line 470. In the application, all worker threads eventually wait on the same mutex which is tried to get locked here. The cause for this unclear to me (might be due to some race condition between spawning new threads from within hpx threads and the scheduling loop).
When changing the mutex type of the local_priority_scheduler to hpx::lcos::local::spinlock the deadlock in this particular line disappears, but another deadlock in cleanup_terminated (in the same file) appears to happen.
Something is fishy within the scheduler.
To reproduce the problem, run the following command:
The text was updated successfully, but these errors were encountered: