Skip to content

Conversation

@againull
Copy link
Contributor

Don't reset command lists each time we try to get an available command
list. We should check for signalled command lists in synchronization
points like piQueueFinish, piQueueRelease, piEventsWait instead.

Don't reset command lists each time we try to get an available command
list. We should check for signalled command lists in synchronization
points like piQueueFinish, piQueueRelease, piEventsWait instead.
@againull againull requested a review from a team as a code owner May 18, 2022 13:04
@againull againull requested a review from smaslov-intel May 18, 2022 13:05
// signalled, then the command list & fence are reset and command list is
// returned to the command list cache. All events associated with command list
// are cleaned up if command list was reset.
std::scoped_lock Lock(this->Mutex);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

can we make this function non-locking (but request callers to hold a lock on the queue)?

Copy link
Contributor Author

@againull againull May 18, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I am not sure why it is better to do so. In my understanding it is better to lock a mutex for smaller amount of time because it is easier to control it this way. Otherwise we keep enabling the strategy of locking mutexes for very long time and doing everything under one lock which will result in nested locks/deadlocks.

Also, as the next step I was planning to move loop which does event->cleanup (it is currently inside resetCommandList()) out of this queue lock. Because event->cleanup locks queue itself to update it's state (update last command event and refcount) and since event->cleanup triggers cleanup of all dependent events we need to lock their queues as well which can result in deadlock situation if we have an outer queue lock (because dependent events may be from different queue). Actually, we have this deadlock situation now in the code and it needs to be fixed.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

OK, but please explain in function comment that we expect no locks on entry

@againull againull merged commit 41402c5 into intel:sycl May 18, 2022
yinyangsx pushed a commit to yinyangsx/llvm that referenced this pull request May 25, 2022
Don't reset command lists each time we try to get an available command
list. We should check for signaled command lists in synchronization
points like piQueueFinish, piQueueRelease, piEventsWait instead.
@againull againull deleted the garbage_collection branch December 3, 2022 00:14
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants