-
Notifications
You must be signed in to change notification settings - Fork 35
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
no-op context overdubbing overhead #91
Comments
using:
on:
Here is the timing info:
Here is a plot of the output
Here is the same, byt with Log scale on Y. I really should have put a |
To we have any traction on this? I fixing/improving would be a big help to oxinabox/MagneticReadHead.jl#27 |
I don't think anyone has worked on this. Would be good to know where the overhead is coming from, e.g. what is the difference between the executed code is it an inlining failure? |
A strange thing I have noticed, the number of allocations is not constant
Timing (after I have ran it a few times getting the compilation done)
|
I can't reproduce the original behaviour on Julia 1.1, (which landed many important performance enhancements for Cassette). I am still seeing the weird oscillation in allocation counts. |
Ok, this is weird, my 1.0.2 results are also a ton better. With Cassette 0.2.2. Original tests were on Cassette master, which would have been roughly 0.1.4 so it could be that. I guess until have alernate evidence we must assume it is that. |
From @oxinabox on Slack:
The text was updated successfully, but these errors were encountered: