-
Notifications
You must be signed in to change notification settings - Fork 409
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 static version for linking NCCL (libnccl) #11604
Comments
So, a manual hack to get it to link static is to change the "ncl" rule to:
( But ideally this would be controlled by the user choice of compiling it statically. While looking at the code I also found out about the I tested adding |
@ddunl @PatriosTheGreat WDYT? |
I'm an author of an ML Framework using XLA.
Per issue #11596 in a recent refresh of my build, XLA build fails if I don't include NCCL. The easy fix would be to include NCCL in my build -- also good for other reasons, but my default (and only) distribution of my ML framework works for both CPU & GPU. I achieve this by linking things statically -- also because it's simpler for the end user.
The issue is that NCCL (as opposed to other CUDA libraries), doesn't have a statically linking rule, even though NVidia distributes the
libnccl_static.a
file. Relevant bazel code.I assume this is a simple fix, for someone with right "bazel-fu skillz" ... but I'm not sure. Also, there may be other considerations I'm not aware. Any help would be most appreciated!
The text was updated successfully, but these errors were encountered: