Add native support for Apple Silicon #12900

meteorcloudy · 2021-01-26T13:36:00Z

After this PR, we can use a x86 version of Bazel to build an ARM64 Bazel binary for Apple Silicon by

bazel build --cpu=darwin_arm64 //src:bazel

The Bazel client, embedded tools and embedded JDK are all native ARM64 binaries.

Changes in this PR:

Add OpenJDK definitions for ARM64 macOS
Manually created the embedded JDK binaries on an Apple Silicon machine and uploaded them to the bazel mirror site.
AutoCpuConverter.java: identify the cpu value correctly
cc_toolchain_config.bzl: Explicitly specify the target platform also in the x86 cc toolchain, so that we can also do cross-platform compile from ARM64 to x86.
lib_cc_configure.bzl: return darwin_x86_64 or darwin_arm64 instead of just darwin for cpu value on macOS.

With the native arm64 Bazel binary, building Bazel itself takes only half of the time compared with the x86 Bazel binary on an Apple DTK machine:

$ bazel build --cpu=darwin_arm64 --disk_cache= --repository_cache= //src:bazel
Starting local Bazel server and connecting to it...
...
Target //src:bazel up-to-date:
  bazel-bin/src/bazel
INFO: Elapsed time: 1044.462s, Critical Path: 180.08s
INFO: 2984 processes: 179 internal, 2092 darwin-sandbox, 1 local, 712 worker.
INFO: Build completed successfully, 2984 total actions

cp  ./bazel-bin/src/bazel ~/bin/bazel-arm64

$ bazel-arm64 clean --expunge

$ bazel-arm64 build --disk_cache= --repository_cache= //src:bazel
Starting local Bazel server and connecting to it...
...
Target //src:bazel up-to-date:
  bazel-bin/src/bazel
INFO: Elapsed time: 464.470s, Critical Path: 114.88s
INFO: 2984 processes: 179 internal, 2092 darwin-sandbox, 1 local, 712 worker.
INFO: Build completed successfully, 2984 total actions

Closes: #11628

meisterT

cc @comius

meteorcloudy · 2021-01-26T13:43:29Z

FYI @aiuto

comius

Do android tests pass (with a bootstrapped version of bazel on arm architecture)? See #12880

Consider using jdk11.0.9 otherwise, if it's not too much effort.

comius · 2021-01-26T15:39:10Z

src/main/java/com/google/devtools/build/lib/analysis/config/AutoCpuConverter.java

-          return "darwin";
+          switch (CPU.getCurrent()) {
+            case X86_64:
+              return "darwin_x86_64";


Changing "darwin" to "darwin_x86_64" might cause some issues. @katre is probably more familiar with this.

It indeed broke the test_disable_cc_toolchain_detection test on macOS, check the latest commit for the fix. I'm not sure if there are more issues I missed.

Long term this will be great because there's existing ambiguity between darwin and darwin_x86_64

meteorcloudy · 2021-01-26T16:13:06Z

Do android tests pass (with a bootstrapped version of bazel on arm architecture)? See #12880

Consider using jdk11.0.9 otherwise, if it's not too much effort.

I haven't tried to install the Android SDK yet, looks like they don't have official Apple Silicon support for now. And there are many other tests still broken, so it will probably take a while to enable tests for Apple Silicon. Hopefully we can get that issue resolved by then, otherwise I can revert the JDK version to 11.0.9.

meteorcloudy · 2021-01-26T16:20:48Z

One problem I noticed from the failing test is that netty doesn't work on Apple Silicon because the jar file actually contains some native shared libraries that're not built for Apple Silicon. This means remote caching/remote execution won't work on Apple Silicon, and I'm not sure what to do..

keith

Awesome!

aiuto · 2021-01-27T05:17:07Z

cc: @susinmotion

mahenzon · 2021-01-29T01:55:12Z

Hi!
I'm trying to build bazel for darwin_arm64, I cloned repo (latest commit is db69c9f, so of course it includes "Add native support for Apple Silicon" (8c7e11a)
Also, I applied this patch bazelbuild/rules_apple#980 (comment)

So, I use this command:
bazel build --cpu=darwin_arm64 //src:bazel

But I'm getting this:

<...>
Use --sandbox_debug to see verbose messages from the sandbox
In file included from src/main/java/net/starlark/java/eval/cpu_profiler_posix.cc:20:
bazel-out/darwin_arm64-fastbuild/bin/external/bazel_tools/tools/jdk/include/jni.h:45:10: fatal error: 'jni_md.h' file not found
#include "jni_md.h"
         ^~~~~~~~~~
1 error generated.
Target //src:bazel failed to build
Use --verbose_failures to see the command lines of failed build steps.
INFO: Elapsed time: 1.730s, Critical Path: 1.19s
INFO: 10 processes: 6 internal, 4 darwin-sandbox.
FAILED: Build did NOT complete successfully

I also tried to checkout 8c7e11a in case it can help, but it doesn’t . Getting the same error

Is there something else I should patch?

meteorcloudy · 2021-01-29T11:22:44Z

@mahenzon Which Bazel version were you using? Can you try with the latest 4.0.0?

meteorcloudy · 2021-01-29T11:35:13Z

@mahenzon And can you check if jni_md.h exist in your JDK installation? Somewhere like /Library/Java/JavaVirtualMachines/zulu-13.jdk/Contents/Home/include/darwin/jni_md.h

mahenzon · 2021-01-29T16:14:50Z

yeah, there's file /Library/Java/JavaVirtualMachines/adoptopenjdk-8.jdk/Contents/Home/include/darwin/jni_md.h

I used 3.7.0.
This time I tried with 4.0.0 and it worked! Thanks!

So, full list of steps:

Download 4.0.0 curl -O -L https://github.com/bazelbuild/bazel/releases/download/4.0.0/bazel-4.0.0-darwin-x86_64
mv bazel-4.0.0-darwin-x86_64 bazel
chmod +x bazel
download source and do any patches you need
build source using bazel 4.0.0 ~/bazel-dist-4.0.0/bazel build --cpu=darwin_arm64 //src:bazel

Your build will be in bazel-bin/src/bazel. Copy it where you need it and run chmod +x bazel
Note that this build's version will be no_version:

➜ ./bazel --version
bazel no_version

UPD:
it's recommended to build Bazel in release mode by adding -c opt:
~/bazel-dist-4.0.0/bazel build --cpu=darwin_arm64 -c opt //src:bazel
© TelegramMessenger/Telegram-iOS#467 (comment)

philwo · 2021-02-27T10:43:25Z

It also works on my Mac Mini M1 and gives a nice ~2x speed increase! 🚀

$ cd ~/bin
$ curl -LO https://github.com/bazelbuild/bazel/releases/download/4.0.0/bazel-4.0.0-darwin-x86_64
$ chmod +x ~/bin/bazel-4.0.0-darwin-x86_64

$ cd ~/src && git clone https://github.com/bazelbuild/bazel.git && cd bazel
$ bazel-4.0.0-darwin-x86_64 build --cpu=darwin_arm64 -c opt //src:bazel
INFO: Elapsed time: 495.900s, Critical Path: 148.84s
INFO: Build completed successfully, 2884 total actions

$ mv bazel-bin/src/bazel ~/bin/bazel
$ bazel clean --expunge
$ bazel build --cpu=darwin_arm64 -c opt //src:bazel
INFO: Elapsed time: 255.364s, Critical Path: 68.15s
INFO: Build completed successfully, 2884 total actions

qthequartermasterman · 2021-04-12T23:45:33Z

It also works on my Mac Mini M1 and gives a nice ~2x speed increase! 🚀

$ cd ~/bin
$ curl -LO https://github.com/bazelbuild/bazel/releases/download/4.0.0/bazel-4.0.0-darwin-x86_64
$ chmod +x ~/bin/bazel-4.0.0-darwin-x86_64

$ cd ~/src && git clone https://github.com/bazelbuild/bazel.git && cd bazel
$ bazel-4.0.0-darwin-x86_64 build --cpu=darwin_arm64 -c opt //src:bazel
INFO: Elapsed time: 495.900s, Critical Path: 148.84s
INFO: Build completed successfully, 2884 total actions

$ mv bazel-bin/src/bazel ~/bin/bazel
$ bazel clean --expunge
$ bazel build --cpu=darwin_arm64 -c opt //src:bazel
INFO: Elapsed time: 255.364s, Critical Path: 68.15s
INFO: Build completed successfully, 2884 total actions

I have successfully used these steps to build bazel 4.0.0 on Apple Silicon. However, another application I am trying to build is throwing an error because this bazel build is unversioned, as @mahenzon mentions (#12900 (comment)). Is there any way to build bazel in a way so that it returns a correct version number?

meteorcloudy · 2021-04-13T07:34:24Z

@qthequartermasterman
--embed_label <version> is the build flag you are looking for.

qthequartermasterman · 2021-04-18T17:54:24Z

Thank you!

…

On Apr 13, 2021, at 2:34 AM, Yun Peng ***@***.***> wrote: @qthequartermasterman --embed_label <version> is the build flag you are looking for. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

After this PR, we can use a x86 version of Bazel to build an ARM64 Bazel binary for Apple Silicon by ``` bazel build --cpu=darwin_arm64 //src:bazel ``` The Bazel client, embedded tools and embedded JDK are all native ARM64 binaries. Changes in this PR: 1. Add OpenJDK definitions for ARM64 macOS 2. Manually created the embedded JDK binaries on an Apple Silicon machine and uploaded them to the bazel mirror site. 3. AutoCpuConverter.java: identify the cpu value correctly 4. cc_toolchain_config.bzl: Explicitly specify the target platform also in the x86 cc toolchain, so that we can also do cross-platform compile from ARM64 to x86. With the native arm64 Bazel binary, building Bazel itself takes only half of the time compared with the x86 Bazel binary on an Apple DTK machine: ``` $ bazel build --cpu=darwin_arm64 --disk_cache= --repository_cache= //src:bazel Starting local Bazel server and connecting to it... ... Target //src:bazel up-to-date: bazel-bin/src/bazel INFO: Elapsed time: 1044.462s, Critical Path: 180.08s INFO: 2984 processes: 179 internal, 2092 darwin-sandbox, 1 local, 712 worker. INFO: Build completed successfully, 2984 total actions cp ./bazel-bin/src/bazel ~/bin/bazel-arm64 $ bazel-arm64 clean --expunge $ bazel-arm64 build --disk_cache= --repository_cache= //src:bazel Starting local Bazel server and connecting to it... ... Target //src:bazel up-to-date: bazel-bin/src/bazel INFO: Elapsed time: 464.470s, Critical Path: 114.88s INFO: 2984 processes: 179 internal, 2092 darwin-sandbox, 1 local, 712 worker. INFO: Build completed successfully, 2984 total actions ``` Closes: #11628 Closes #12900. PiperOrigin-RevId: 354058336

After this PR, we can use a x86 version of Bazel to build an ARM64 Bazel binary for Apple Silicon by ``` bazel build --cpu=darwin_arm64 //src:bazel ``` The Bazel client, embedded tools and embedded JDK are all native ARM64 binaries. Changes in this PR: 1. Add OpenJDK definitions for ARM64 macOS 2. Manually created the embedded JDK binaries on an Apple Silicon machine and uploaded them to the bazel mirror site. 3. AutoCpuConverter.java: identify the cpu value correctly 4. cc_toolchain_config.bzl: Explicitly specify the target platform also in the x86 cc toolchain, so that we can also do cross-platform compile from ARM64 to x86. With the native arm64 Bazel binary, building Bazel itself takes only half of the time compared with the x86 Bazel binary on an Apple DTK machine: ``` $ bazel build --cpu=darwin_arm64 --disk_cache= --repository_cache= //src:bazel Starting local Bazel server and connecting to it... ... Target //src:bazel up-to-date: bazel-bin/src/bazel INFO: Elapsed time: 1044.462s, Critical Path: 180.08s INFO: 2984 processes: 179 internal, 2092 darwin-sandbox, 1 local, 712 worker. INFO: Build completed successfully, 2984 total actions cp ./bazel-bin/src/bazel ~/bin/bazel-arm64 $ bazel-arm64 clean --expunge $ bazel-arm64 build --disk_cache= --repository_cache= //src:bazel Starting local Bazel server and connecting to it... ... Target //src:bazel up-to-date: bazel-bin/src/bazel INFO: Elapsed time: 464.470s, Critical Path: 114.88s INFO: 2984 processes: 179 internal, 2092 darwin-sandbox, 1 local, 712 worker. INFO: Build completed successfully, 2984 total actions ``` Closes: #11628 Closes #12900. PiperOrigin-RevId: 354058336 Revert default cpu value on x86 macOS to "darwin" Many downstream projects are still depending on the default cpu value on x86 macOS being "darwin" instead of "darwin_x86_64" https://buildkite.com/bazel/bazel-auto-sheriff-face-with-cowboy-hat/builds/412 Related: #11628 Closes #12918. PiperOrigin-RevId: 354279409

After this PR, we can use a x86 version of Bazel to build an ARM64 Bazel binary for Apple Silicon by ``` bazel build --cpu=darwin_arm64 //src:bazel ``` The Bazel client, embedded tools and embedded JDK are all native ARM64 binaries. Changes in this PR: 1. Add OpenJDK definitions for ARM64 macOS 2. Manually created the embedded JDK binaries on an Apple Silicon machine and uploaded them to the bazel mirror site. 3. AutoCpuConverter.java: identify the cpu value correctly 4. cc_toolchain_config.bzl: Explicitly specify the target platform also in the x86 cc toolchain, so that we can also do cross-platform compile from ARM64 to x86. With the native arm64 Bazel binary, building Bazel itself takes only half of the time compared with the x86 Bazel binary on an Apple DTK machine: ``` $ bazel build --cpu=darwin_arm64 --disk_cache= --repository_cache= //src:bazel Starting local Bazel server and connecting to it... ... Target //src:bazel up-to-date: bazel-bin/src/bazel INFO: Elapsed time: 1044.462s, Critical Path: 180.08s INFO: 2984 processes: 179 internal, 2092 darwin-sandbox, 1 local, 712 worker. INFO: Build completed successfully, 2984 total actions cp ./bazel-bin/src/bazel ~/bin/bazel-arm64 $ bazel-arm64 clean --expunge $ bazel-arm64 build --disk_cache= --repository_cache= //src:bazel Starting local Bazel server and connecting to it... ... Target //src:bazel up-to-date: bazel-bin/src/bazel INFO: Elapsed time: 464.470s, Critical Path: 114.88s INFO: 2984 processes: 179 internal, 2092 darwin-sandbox, 1 local, 712 worker. INFO: Build completed successfully, 2984 total actions ``` Closes: #11628 Closes #12900. PiperOrigin-RevId: 354058336 Revert default cpu value on x86 macOS to "darwin" Many downstream projects are still depending on the default cpu value on x86 macOS being "darwin" instead of "darwin_x86_64" https://buildkite.com/bazel/bazel-auto-sheriff-face-with-cowboy-hat/builds/412 Related: #11628 Closes #12918. PiperOrigin-RevId: 354279409 Update platform version to 0.0.4 Use configuration flags for //src/conditions:darwin_arm64 Partially reverts 6d637f4, to allow cross-compiling Bazel from darwin_x86_64 to darwin_arm64 with the `--cpu=darwin_arm64` flag. The only source that is different between darwin_x86_64 and darwin_arm64 now is the embedded JDK. Closes #12653. PiperOrigin-RevId: 346298051 Make sure Bazel selects correct remote JDK on Apple Silicon

lexuanquynh · 2021-06-30T16:41:20Z

I got error:

Required bazel version is "4.0.0", but "no_version"" is reported by /Users/xuanquynhle/bin/bazel

meteorcloudy added 2 commits January 26, 2021 14:22

Make Bazel work on Apple Silicon

5693a62

Add embedded JDK for macOS aarch64

86bc5fe

google-cla bot added the cla: yes label Jan 26, 2021

meteorcloudy requested review from philwo, comius and meisterT January 26, 2021 13:36

Fix comment

1f0316f

meisterT approved these changes Jan 26, 2021

View reviewed changes

meteorcloudy added 2 commits January 26, 2021 14:53

Fix artifact name for arm64 macOS JDK

72e3ac3

Add more missing mentions of the ARM64 macOS JDK

a41a794

comius approved these changes Jan 26, 2021

View reviewed changes

comius reviewed Jan 26, 2021

View reviewed changes

Fix cpu value check for macOS in cc auto configure

78327ee

keith approved these changes Jan 26, 2021

View reviewed changes

bazel-io closed this in 8c7e11a Jan 27, 2021

mahenzon mentioned this pull request Jan 29, 2021

Cannot run simulator on M1 Mac TelegramMessenger/Telegram-iOS#467

Closed

tkoeppe mentioned this pull request Jan 29, 2021

Error on M1 Mac (aarch64) google-deepmind/lab2d#5

Closed

wojtekmach mentioned this pull request Feb 18, 2021

XLA doesn't support Mac ARM elixir-nx/nx#217

Closed

bvirlet mentioned this pull request Mar 10, 2021

Compile TensorFlow Lite for iOS Simulator on Apple Silicon tensorflow/tensorflow#47400

Closed

RinkiyaKeDad mentioned this pull request Mar 26, 2021

support building for and from ARM cert-manager/cert-manager#3804

Closed

AnjeyNov mentioned this pull request Jul 29, 2021

Can't generate Project on M1 TelegramMessenger/Telegram-iOS#587

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add native support for Apple Silicon #12900

Add native support for Apple Silicon #12900

meteorcloudy commented Jan 26, 2021 •

edited

Loading

meisterT left a comment

meteorcloudy commented Jan 26, 2021

comius left a comment

comius Jan 26, 2021

meteorcloudy Jan 26, 2021

keith Jan 26, 2021

meteorcloudy commented Jan 26, 2021

meteorcloudy commented Jan 26, 2021 •

edited

Loading

keith left a comment

aiuto commented Jan 27, 2021

mahenzon commented Jan 29, 2021 •

edited

Loading

meteorcloudy commented Jan 29, 2021

meteorcloudy commented Jan 29, 2021

mahenzon commented Jan 29, 2021 •

edited

Loading

philwo commented Feb 27, 2021

qthequartermasterman commented Apr 12, 2021

meteorcloudy commented Apr 13, 2021 •

edited

Loading

qthequartermasterman commented Apr 18, 2021 via email

lexuanquynh commented Jun 30, 2021

Add native support for Apple Silicon #12900

Add native support for Apple Silicon #12900

Conversation

meteorcloudy commented Jan 26, 2021 • edited Loading

meisterT left a comment

Choose a reason for hiding this comment

meteorcloudy commented Jan 26, 2021

comius left a comment

Choose a reason for hiding this comment

comius Jan 26, 2021

Choose a reason for hiding this comment

meteorcloudy Jan 26, 2021

Choose a reason for hiding this comment

keith Jan 26, 2021

Choose a reason for hiding this comment

meteorcloudy commented Jan 26, 2021

meteorcloudy commented Jan 26, 2021 • edited Loading

keith left a comment

Choose a reason for hiding this comment

aiuto commented Jan 27, 2021

mahenzon commented Jan 29, 2021 • edited Loading

meteorcloudy commented Jan 29, 2021

meteorcloudy commented Jan 29, 2021

mahenzon commented Jan 29, 2021 • edited Loading

philwo commented Feb 27, 2021

qthequartermasterman commented Apr 12, 2021

meteorcloudy commented Apr 13, 2021 • edited Loading

qthequartermasterman commented Apr 18, 2021 via email

lexuanquynh commented Jun 30, 2021

meteorcloudy commented Jan 26, 2021 •

edited

Loading

meteorcloudy commented Jan 26, 2021 •

edited

Loading

mahenzon commented Jan 29, 2021 •

edited

Loading

mahenzon commented Jan 29, 2021 •

edited

Loading

meteorcloudy commented Apr 13, 2021 •

edited

Loading