-
-
Notifications
You must be signed in to change notification settings - Fork 5.6k
Performance regression with threads #14417
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
Comments
Cc: @JeffBezanson, @vtjnash, @tkelman, @yuyichao. |
FYI, in your second example, removing the Also the issue with that example is that the |
fwiw, with #12904 merged now, profiling is multi-thread aware and |
This regression still exists at the point @yuyichao: can you clarify on the |
The I think this is due to some tweak in inlining/type inference so it might not affect the old threading branch. My version of
|
This is not very nice. Your version of
Which doesn't seem very reasonable? |
Well, the issue is number of arguments to |
Sure, but this is pretty non-intuitive for your average programmer. But anyway, that's a separate issue. |
May be caused by #13735 |
Focusing on the Gaussian blur code here running on master, serial performance (i.e. no Ignore the threading branch; that's a red herring -- serial performance is identical to performance with The typed ASTs for the serial and multithreaded code are in the same gist and look pretty different. Do they help in understanding this? |
Fixed on master. |
Compared to the old threading branch, there is a 50% performance regression for multithreading performance on master for Laplace 3D:
For the code snippet here, the slowdown is over 100X:
I'll try and figure out where this happened.
The text was updated successfully, but these errors were encountered: