Skip to content
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.

Sign up for GitHub

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

Jump to bottom

Error messages when CUDNN is not loaded. #287

Closed
MikeInnes opened this issue Jun 7, 2018 · 4 comments
Closed

Error messages when CUDNN is not loaded. #287

MikeInnes opened this issue Jun 7, 2018 · 4 comments

Comments

@MikeInnes
Copy link
Member

e.g. when running conv (at least until we have Julia implementations).
@MikeInnes MikeInnes mentioned this issue Oct 3, 2018
Closed
@maleadt
Copy link
Collaborator

maleadt commented Jan 5, 2019

https://discourse.julialang.org/t/flux-failing-on-gpu/18948
I'm not sure who is responsible for triggering conversion to pointer not defined for CuArray{Float32,2} when cuDNN isn't available, but it's pretty tough to debug for users and there's been multiple reports already.

@MikeInnes
Copy link
Member Author

At a guess, the gemm! inside conv in NNlib. We should just have a generic conv fallback in CuArrays, even if it only errors out.

@MikeInnes MikeInnes mentioned this issue Jan 10, 2019
Merged
bors bot added a commit to JuliaGPU/CuArrays.jl that referenced this issue Jan 15, 2019
@bors @MikeInnes
Merge #250
008e6f7 
250: NNlib without CUDNN r=maleadt a=MikeInnes

This patch refactors our NNlib wrappers so that more things are independent of CUDNN.

Right now the main benefit is to give better errors when CUDNN isn't installed (FluxML/Flux.jl#287). Of course we'd ideally have generic, if non-optimal, implementations of these functions available in future.

Co-authored-by: Mike J Innes <[email protected]>
@tbenst tbenst mentioned this issue Apr 11, 2019
Closed
@darsnack
Copy link
Member

Doesn't CUDA.jl do something like this now?

@maleadt
Copy link
Collaborator

maleadt commented Feb 12, 2021

Yes, you should always get a nice error when trying to use CUDNN when it isn't available.

@maleadt maleadt closed this as completed Feb 12, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
3 participants
@maleadt @MikeInnes @darsnack