Add new GCC shard for iains' GCC port to Darwin Aarch64

[Keno]

Iain Sandoe is working on a port of GCC to Darwin Aaarch64 over at
https://github.com/iains/gcc-darwin-arm64. This is effort is what
Apple recommended we use for Fortran support. To help with testing
the port, define a new GCC shard that builds Iain's fork of GCC,
as well as using a local copy of the macOS SDK, so it can be built
against the NDA'd version of the Darwin AArch64 SDK for testing.

Depends on #1625

[fxcoudert]

On the Fortran side of that build, if you have any questions, please let me know, or open an issue on Iain's repo and CC me. I'll be trying to test as much Fortran-based software (and submit patches to their upstream) before the release.

[Keno]

Will do. I built this yesterday and ran through openblas 0.3.10 which built fine. We'll build the rest of the ecosystem once the shard it up on our build server.

[fxcoudert]

There were already 2 issues fixed for OpenBlas: OpenMathLib/OpenBLAS#2804 & OpenMathLib/OpenBLAS#2805
Are there other projects that you depend on? I've tried FFTW as well, for now.

Edit: I see you have gmp, we carry a patch for it in Homebrew:

    # Remove when upstream fix is released
    # https://gmplib.org/list-archives/gmp-bugs/2020-July/004837.html
    # arm64-darwin patch
    url "https://raw.githubusercontent.com/Homebrew/formula-patches/c53834b4/gmp/6.2.0-arm.patch"

[Keno]

Yes, we already carry the same GMP patch (JuliaLang/julia#36616) though we need to apply it here too once the shard is merged. The OpenBLAS fixes didn't seem to be necessary on 0.3.10. I guess that doesn't have the ARMv8 dynamic arch code yet and we manually force the cmake settings to aarch64 in our build system so that change wasn't needed either.

[Keno]

Are there other projects that you depend on? I've tried FFTW as well, for now.

You can see the full list of things that packages in the ecosystem depend on at the top level in this repo. We'll shortly do a world build of everything, which'll tell us what's broken.

[Keno]

Artifacts registered here JuliaPackaging/BinaryBuilderBase.jl#29 and are in the process of uploading to the build machine. Once that's through, I think we should try triggering a bulk rebuild to see what fails.