Site icon R-bloggers

R-universe now builds MacOS ARM64 binaries for use on Apple Silicon (aka M1/M2/M3) systems

[This article was first published on rOpenSci - open tools for open science, and kindly contributed to R-bloggers]. (You can report issue about the content on this page here)
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.

Abstract / TLDR

R-universe now provides MacOS arm64 binaries for all R packages. This means that MacOS users on Apple Silicon hardware (aka M1/M2/M3) can install the very latest builds of any R package without the need for any compilation:

install.packages('arrow',
 repos = c('https://apache.r-universe.dev', 'https://cloud.r-project.org'))

R-universe uses cross-compiling for arm64 binaries, though this should not make much of a difference for package authors and R users. Packages with C/C++/Fortran/Rust code are all supported.

Why cross compiling

Because GitHub Actions currently does not offer arm64 runners for OSS projects, the arm64 binaries are cross-compiled on the MacOS intel runners. The cross build environment is set up to mimic a native arm64 machine, such that most R packages do not need any modification to work. We found only a small number of packages with a buggy configure script that may need a fix to allow cross-compilation.

The r-universe workflow only builds arm64 binaries when needed, i.e. for packages that include compiled code (C/C++/Fortran/Rust). Packages that do not include any compiled code are portable by design, so for these packages the binaries built for MacOS on intel are served both in the x86_64 and arm64 cranlike repositories, without the need for an additional cross compile step.

Just like CRAN, the r-universe package homepage shows a link to the r-4.3-x86_64 and r-4.3-arm64 binaries. Packages without compiled code have a r-4.3-any binary which is used for either architecture.

To have a look at the build logs, click on the little apple icon next to these links. Alternatively, you can use the /buildlog shortlink, for example https://apache.r-universe.dev/arrow/buildlog will take you to the latest arrow builds.

On this page you can find the arm64 build log specifically by expanding the r-release-macos job and then under the section “Cross Compile for MacOS ARM64”. If this section is disabled, it means it was skipped because this package does not have compiled code, and does not need cross compilation.

Some technical details

For those interested how the cross compilation is set up, here are the main ingredients:

With this setup, almost any R package can be built in the cross environment exactly the same way they do on normal arm64 hardware. But if your package does not work and you need some help fixing it, please feel free to open an issue.

Universal binaries

Finally, some R packages download precompiled binaries for libraries too big or complicated to build on the fly. It is safest to distribute such libraries in the form of universal binaries which contain both the x86_64 and arm64 libs. This way your download script does not need to make any guesses about the target architecture it is building for: the same libs can be linked on either target.

Creating a universal binary can be done for both static and dynamic libraries and is really easy. If you have a x86_64 and arm64 version of libfoo.a you can glue them together with lipo:

lipo -create x86_64/libfoo.a arm64/libfoo.a -output universal/libfoo.a

And this new libfoo.a can be used when building either for intel or arm. Again if you need any help with this feel free to reach out.

To leave a comment for the author, please follow the link and comment on their blog: rOpenSci - open tools for open science.

R-bloggers.com offers daily e-mail updates about R news and tutorials about learning R and many other topics. Click here if you're looking to post or find an R/data-science job.
Want to share your content on R-bloggers? click here if you have a blog, or here if you don't.
Exit mobile version