Add support for bottling Portable Ruby for ARM64 Linux #19111
Conversation
ZhongRuoyu
force-pushed
the
aarch64-linux
branch
from
2b8b778 to
a0d731c
Compare
ZhongRuoyu
force-pushed
the
aarch64-linux
branch
from
a0d731c to
0e89ed9
Compare
| DevelopmentTools.gcc_version(compiler.to_s) >= "9.3.1" | ||
| # Out-of-line atomics are not supported out-of-the-box on all systems. | ||
| # https://learn.arm.com/learning-paths/servers-and-cloud-computing/lse/intro/ | ||
| "#{generic_determine_optflags} -mno-outline-atomics" |
There was a problem hiding this comment.
Given we already use -march=armv8-a, is this for scenarios with a shared libgcc is used?
Is this also relevant for anything outside of Portable Ruby?
There was a problem hiding this comment.
This is not related to shared libgcc AFAIK; it's relevant for building all bottles with GCC >= 9.3.1 (where out-of-line atomics are enabled by default and not controlled by -march=armv8-a). I recall not being able to run binaries from a bottle built in standard CI environment (GCC 11) out-of-the-box on Ubuntu 18.04 (requires an extra package libc6-lse to be installed).
There was a problem hiding this comment.
That's interesting given https://community.arm.com/arm-community-blogs/b/tools-software-ides-blog/posts/making-the-most-of-the-arm-architecture-in-gcc-10 implies it can determine it at runtime using a libgcc call of __aarch64_cas4_relax.
Is it elsewhere on 18.04 and not being linked? Missing from our GCC 11 build like https://bugs.gentoo.org/868018?
There was a problem hiding this comment.
Mmm, will need to check this. Away from keyboard at the moment but will do tonight in the weekend (sorry things got in the way).
There was a problem hiding this comment.
Unless we need this to build Portable Ruby on the GitHub runner: I'd prefer if we didn't have this. Feels like something that should be formula specific.
Co-authored-by: Bo Anderson <[email protected]>
| @@ -46,15 +46,30 @@ def homebrew_extra_paths | |||
|
|
|||
| def homebrew_extra_isystem_paths | |||
| paths = [] | |||
| # Add paths for GCC headers when building against [email protected] because we have to use -nostdinc. | |||
| if deps.any? { |d| d.name == "[email protected]" } | |||
| # Add paths for GCC headers when building against [email protected] or [email protected] | |||
There was a problem hiding this comment.
Maybe this should be all versioned glibcs (glibc@.+) rather than hardcoding a list here?
| DevelopmentTools.gcc_version(compiler.to_s) >= "9.3.1" | ||
| # Out-of-line atomics are not supported out-of-the-box on all systems. | ||
| # https://learn.arm.com/learning-paths/servers-and-cloud-computing/lse/intro/ | ||
| "#{generic_determine_optflags} -mno-outline-atomics" |
There was a problem hiding this comment.
Unless we need this to build Portable Ruby on the GitHub runner: I'd prefer if we didn't have this. Feels like something that should be formula specific.
| args << "-nostdinc" if @deps.include?("[email protected]") | ||
| # Add -nostdinc when building against [email protected] or [email protected] to avoid | ||
| # mixing system and brewed glibc headers. | ||
| args << "-nostdinc" if @deps.intersect?(["[email protected]", "[email protected]"]) |
| @@ -353,19 +354,20 @@ class Cmd | |||
| end | |||
|
|
|||
| def ldflags_linux(args) | |||
| versioned_glibc_dep = (@deps & ["[email protected]", "[email protected]"]).first | |||
There was a problem hiding this comment.
And even more: same as above. Given how many times we're hardcoding these versions definitely feel like we should be using e.g. a consistent constant and a regex.
brew stylewith your changes locally?brew typecheckwith your changes locally?brew testswith your changes locally?These are the things that I identified when playing with Portable Ruby on ARM64 Linux.