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.

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

[linux-arm64] set "mkl_aarch64" bazel config for linux-arm64 platform #502

Merged
merged 1 commit into from
Sep 21, 2023

Conversation

snadampal
Copy link
Contributor

This is to enable TensorFlow core library bazel build with mkldnn (oneDNN) backend for linux-arm64 platforms.

@Craigacp
Copy link
Collaborator

Craigacp commented Sep 8, 2023

We're currently working on removing the bazel build and replacing it with the binaries built by TF for Linux & macOS platforms, so while we can merge this, it'll likely be replaced before the next release.

@snadampal
Copy link
Contributor Author

Hi @Craigacp , thanks for the review! When you say replacing the bazel build with binaries, does it mean we can build TF core outside the java repo and able to import the binaries? in that case it will automatically cover linux-arm64 platform as well, right? because it is fully supported in the TF repo.

@Craigacp
Copy link
Collaborator

Craigacp commented Sep 9, 2023

We'll repackage the TF binaries from the TF python wheels on several platforms. The Java binding itself still needs to be compiled so we're likely still limited to things we can easily build on GitHub actions as we don't have other build resources.

We're currently deciding if it will still be possible to build TF-Java entirely from scratch without downloading the binaries as we might still need to do a full build for Windows as that is packaged differently.

@saudet
Copy link
Contributor

saudet commented Sep 9, 2023

There is no problem cross compiling JNI bindings generated with JavaCPP, so given binaries for the C++ API of TF Core itself, it should already pretty much work.

@snadampal
Copy link
Contributor Author

Thanks, @saudet and @Craigacp. What is the tentative timeline to get the new build system support? if it is only few weeks away I'm fine with directly trying linux-aarch64 in the new system, otherwise I see value in having this PR on the mainline to enable source builds at least.

@Craigacp
Copy link
Collaborator

We don't have a timeline on the new build support, nor are we planning on setting up cross compiling for Linux arm64 at the moment.

@snadampal
Copy link
Contributor Author

Thanks for approving the PR, @Craigacp . Please let me know what my next step is to get this merged.

@Craigacp
Copy link
Collaborator

The CI is blocked as we're still using ubuntu-18.04 which doesn't work anymore. Once we've merged the other PR to bump the ci I'll run it on this and then we can merge it.

@karllessard karllessard merged commit 878ec6d into tensorflow:master Sep 21, 2023
@karllessard
Copy link
Collaborator

I've just merged both #503 and this one

@snadampal
Copy link
Contributor Author

Thanks, @karllessard !

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants