Skip to content

Upstreaming support for libnvptxcompiler in rules_ml_toolchain #34

@hugomano

Description

@hugomano

Hello,

I'm working on a project that uses the vendored CUDA PjRT plugin, and our framework supports libnvptxcompiler. To enable this, we’ve made a few patches to build and link against the static libnvptxcompiler archive.

To achieve that, I forked and modified rules_ml_toolchain to expose the static library. You can see the diff here:: https://github.com/google-ml-infra/rules_ml_toolchain/compare/main...zml:rules_ml_toolchain:zml?expand=1

Would you be open to upstreaming these changes into your repository?

Note

This currently works only with a non-hermetic CUDA/C++ toolchain, due to ABI quirks between the embedded Clang toolchain and the archive built with GCC 7.3.

For context, here's a diff showing the differences between the two builds: https://gist.github.com/hugomano/e7bc4e17d17702cdecd4640ce9c15d12

I thought this information might also be useful for these recently opened PRs:

cc: @ybaturina @yuriivcs , gentle ping 🙂

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions