-
Notifications
You must be signed in to change notification settings - Fork 146
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Error in index space launcher #94
Comments
That's an interesting failure mode. Do you think you can make a minimal test case and attach it to this issue? |
I have narrowed down the problem, it is in the constructor of the Index Launcher. There I try to add some Region Requirements, and if I delete them, then things work well. I will further check the validity of the Logical Regions. But in the meantime, could you please give me some hint on the possible reasons for this? Thanks a lot!! |
I don't think this is your fault, but is instead a bug in the runtime. It would be good if you could create a small test case for me to work with so I can better understand the nature of the problem. |
Actually, I have a guess as to the cause. Here are a few questions that will help narrow down the problem. There should be one region requirement in your index space launch that is causing this issue.
My hypothesis is that you are getting interfering region requirements for multiple points in your index space task launch and the runtime is not doing a good job reporting this error. However, I would like to confirm this is the issue first. Answers to those questions will help. A small test case that reproduces the problem would be even better. |
Thanks a lot for the message. • Actually I have tried adding several region requirements, some of them are read-only and some of them are read-write. Adding any of them will cause the failure. I do know that all the points within an index space task launch are required to be non-interfering with each other either because use disjoint regions or non-interfering privileges. I’ll double check, and maybe I got something wrong in my code. At the same time, I’ll try to reduce the problem and provide a small test case. Thanks again!! |
Hmm interesting. Yeah, a small test case would be very useful. |
Just an update. I have figured out that the reason for some of the failure is that I forgot to unmap the physical region before I use the aliased logical region in the subtasks. But one region requirement is still causing failure, I'll continue trying to figure it out... |
Ok, even if that is the case, the runtime should automatically be unmapping and then remapping the conflicting region around your index space task launch (it does this implicitly to avoid deadlock, but it will incur a performance cost). Play around with that to see if it is impacting the manifestation of the bug. |
Do you mean whether there are unmapping statements in my code won't affect it's correctness? But the fact is that if I do the unmapping by myself, it works; then I simply comment the single line of unmapping statement, and the error occurs... |
Yes, the unmapping statements should only impact the performance of your code. If it changes the correctness, then that is a runtime bug. If that is part of the problem, make sure to include that in the small example you are working on generating. |
This issue has been resolved as result of poor warning messages being issued for interfering region requirements. |
This is the error I get when I added the index space launch, could you please tell me what's wrong with this? Thanks a lot!
The text was updated successfully, but these errors were encountered: