-
Notifications
You must be signed in to change notification settings - Fork 63
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
AD on WaterLily using Enzyme #1647
Comments
You could try making your life a bit simpler and start with:
Edit: |
I was not aware of that issues, thanks for pointing it out. It could be related indeed, but I am not sure how to debug this in any case. |
Closed as duplicate |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Hey! I have been trying to use Enzyme to perform AD on WaterLily but haven't succeeded so far. Currently, AD works fine with ForwardDiff but we would like to switch to Enzyme to be able to have AD running on GPU (WaterLily-jl/WaterLily.jl#130). Below is the ForwardDiff MWE that works when using our SIMD backend or KernelAbstractions.jl CPU backend (note that WaterLily uses SIMD when launching
julia -t 1 ...
, and KA otherwise)When trying to switch to Enzyme, the following code would hopefully reproduce the results of ForwardDiff, but so far I only get a very long stacktrace that makes it difficult to debug. From my limited understanding, I think it is complaining about type inference on the creation of the
Simulation
struct (and something related to amemcpy
).Any hint here would be greatly appreciated. Thanks!
The text was updated successfully, but these errors were encountered: