Support long double intrinsics in any aarch64 linux

[richkadel]

Expands the support added in #377 from just musl to any linux.

Fixes #428

[richkadel]

@alexcrichton - Here's the PR I proposed in zulip.

cc: @tmandry @petrhosek

[richkadel]

@Amanieu - Alex referred me to you for reviewing this change. Do you have time to take a look at this?

(There is some urgency for resolving this for Fuchsia, so I appreciate any help you can provide. Thank you!)

[Amanieu]

Arguably this should apply to all targets using aarch64, so we can just go ahead and remove the if. Looking at compiler-rt's CMakeLists.txt shows that the *tf intrinsics are always included.

[richkadel]

I just pushed an amended commit to address the CI error:

checking /target/aarch64-unknown-linux-gnu/debug/deps/libcompiler_builtins-9c67e34fdec82140.rlib for duplicate symbols
================================================================
0000000000000000 T __fe_getround
0000000000000000 T __fe_raise_inexact
Error: Process completed with exit code 1.

I wasn't sure if I should include the entire block from the musl condition or not, but apparently those symbols should be in the musl condition only.

[richkadel]

Arguably this should apply to all targets using aarch64, so we can just go ahead and remove the if. Looking at compiler-rt's CMakeLists.txt shows that the *tf intrinsics are always included.

Got it. Will do. Thanks.

[richkadel]

@Amanieu - Done. Thanks!

[Amanieu]

The CI error happens because fp_mode.c is compiled and included twice. The fix is to avoid compiling the same file multiple times on line 510 of build.rs.

This wasn't caught before because only aarch64-gnu is tested on CI, not aarch64-musl.

[richkadel]

Ah, I can add a hashset lookup on line 510 to avoid adding dups, and move those last two symbols with the others under any aarch64.

I think that's what you're suggesting, so I'll start that. Let me know if I misunderstand.

[Amanieu]

Yes, that's what I had in mind.

[richkadel]

Done (I think) but I'm not sure how to launch a test of that code locally on my machine (cargo test from both the root dir and the testcrate dir didn't even attempt to compile it). Apologies if CI finds a typo/syntax error.

If there's a better way to test locally, let me know and I'll check it.

[richkadel]

@Amanieu thank you for your quick response!

Can I help with any steps in https://github.com/rust-lang/compiler-builtins/blob/master/PUBLISHING.md in some way?

[Amanieu]

I just published a new release.

[richkadel]

Awesome! Thanks so much!

[richkadel]

@Amanieu - I just wanted to let you know that this change did resolve the problems we had 👍 .

Thanks again!