-
-
Notifications
You must be signed in to change notification settings - Fork 53
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
Add missing libcuda.so on 9.2 #93
Conversation
Hi! This is the friendly automated conda-forge-linting service. I just wanted to let you know that I linted all conda-recipes in your PR ( |
@conda-forge-admin, please rerender |
ab01f50
to
34a9ee5
Compare
@conda-forge-admin, please rerender |
@conda-forge-admin, please rerender |
@conda-forge-admin, please rerender |
@conda-forge-admin, please rerender |
@conda-forge-admin, please rerender |
@isuruf @beckermr |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Nit: allow future extensions.
@conda-forge-admin, please rerender |
Hi! This is the friendly automated conda-forge-webservice. |
Ping @isuruf |
@isuruf |
|
OK, clearly I should stop trying to do these PRs late at night - sorry for the cumulation of oversights. In any case, if we don't have the rights to register another rpm, then I guess we have to download directly? |
22f72d2
to
ee896eb
Compare
OK, I give up. Even with
|
rm cuda-repo-*.rpm && \ | ||
sudo yum install -y cuda-compat-10-0.x86_64 && \ | ||
# note that $CUDA_HOME is just a symlink to /usr/local/cuda-${CUDA_VER}, and yum installs in that folder | ||
sudo mv /usr/local/cuda-10.0/* /usr/local/cuda-9.2 && \ |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Could we try the following approach:
- keep /usr/local/cuda-10.0/compat as it is
- in linux-anvil-cuda/Dockerfile append /usr/local/cuda-10.0/compat to /etc/ld.so.conf.d/cuda-$CUDA_VER.conf only when CUDA_VER=9.2
?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@isuruf
What do you think about this idea? I dislike not being able to fix this correctly here, but without proper elevation - maybe we could add these two specific commands to the sudoer-list in the image?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The sudoers-route works (I tested it; see conda-forge/docker-images#138), and should lead to IMO the cleanest solution all-around.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I don't think this is safe to do. If a user builds a CUDA 9.2 package with this CUDA 10.0 compat libcuda.so
this can potentially break things on a driver 396 install for example as this compat I believe is based on driver 410. I could be mistaken about this though.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The user is not building against CUDA 10.0
. It's just loading the application using cuda compat 10.0. This is useful for checking that packages can be loaded on CI without GPUs.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yes, it's added to ld.conf which has the almost the same effect as LD_LIBRARY_PATH
@h-vetinari, how about not moving that and adding /usr/local/cuda-10.0/compat
to ld.conf in cuda 9.2 image?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Yes, it's added to ld.conf which has the almost the same effect as
LD_LIBRARY_PATH
@h-vetinari, how about not moving that and adding
/usr/local/cuda-10.2/compat
to ld.conf in cuda 9.2 image?
Won't this basically cause symbols from a newer driver library to be used during building packages which aren't backwards compatible? I believe CUDA 10.0 compat is based on driver 410, CUDA 10.2 compat is based on driver 440, but CUDA 9.2 can use driver 396 which presents possible issues from lack of backwards compatibility.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Nope. We are using the stub library from 9.2 to build and the 10.0 compat is used only for loading
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@h-vetinari, how about not moving that and adding /usr/local/cuda-10.0/compat to ld.conf in cuda 9.2 image?
This is what @pearu suggested at the top of this review thread. Fine by me.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
See here: conda-forge/docker-images#139
This need to be combined with conda-forge/docker-images#139 to have its desired effect. |
5ce56f1
to
e4d00f1
Compare
…da-forge-pinning 2020.05.14.17.14.20
@isuruf |
@kkraus14, any objections? |
I'm still not clear on how this being added to |
|
Thanks @h-vetinari for the PR and @kkraus14 for the review. |
This is complementary to conda-forge/docker-images#135; aiming at unblocking conda-forge/pyarrow-feedstock#101 & conda-forge/faiss-split-feedstock#1.