Update to LLVM 23

[nikic]

View all comments

Changes in this PR (apart from the LLVM update):

• Updated cov-map tests due to minor ordering changes introduced in llvm/llvm-project@5931034. The affected tests are limited to LLVM >= 23 now.
• Set LLVM_VERSIONED_DYLIB_NAME_ON_DARWIN=OFF, which restores the previous unversioned naming of the LLVM dylib on Darwin. I expect we'll want to change this for consistency with Linux, but given the amount of trouble Darwin dylibs have caused in the past, I'm not trying to do it in this PR.
• Distribute the llvm-project/libc subproject, which is now a build dependency of LLVM.
• Pull in f128 Windows ABI changes from adjust f128 windows abi for LLVM 23 #158778. These need to land in sync with the LLVM update (as ABI of libcalls is controlled by LLVM, not our ABI lowering).
• Disable RISCV backend on dist-i686-mingw, because the build OOMs on a large generated file. This is a temporary measure until host tools are removed by Drop tier 2 i686-pc-windows-gnu host tools compiler-team#1020.
• The llvm-project fork includes a new downstream patch for mingw: rust-lang/llvm-project@b89cd8e It works around mingw having broken TLS prior to GCC 16. (We're currently on GCC 14.)

Issues:

• [X86] Assertion failure in tile copy lowering llvm/llvm-project#209512
• [X86] New VectorCombine regresses vtestps matching llvm/llvm-project#209714
• [X86] InstCombine change breaks blsic matching llvm/llvm-project#209718
• Invalid sse4a tests stdarch#2190 (synced in stdarch subtree update #159517)
• [LTO] Sort DefinedGlobals in LTO cache key llvm/llvm-project#210025
• Update tests for LLVM 23 #159391
• Fix compatibility with LLVM 23 EnzymeAD/Enzyme#2951
• Update Enzyme to handle LLVM23 #159301
• [X86] checkTCRetEnoughRegs() assertion failure for i686-windows-msvc with PIC llvm/llvm-project#210300

[nikic]

@bors try @rust-timer queue

[rust-bors]

💔 Test for b003a3f failed: CI. Failed job:

• try - dist-x86_64-linux-quick (web logs, enhanced plaintext logs)

[nikic]

@bors try

[rust-bors]

💔 Test for 0259978 failed: CI. Failed job:

• try - dist-x86_64-linux-quick (web logs, enhanced plaintext logs)

[nikic]

@bors try

[rust-bors]

💔 Test for 4973bfc failed: CI. Failed job:

• try - dist-x86_64-linux-quick (web logs, enhanced plaintext logs)

[ZuseZ4]

perfect, I dropped the last commit and only left my earlier gcc fix, as well as Nikita's PR in here.
Once the proper fix lands someone can then just approve this one, I'll probably not be on my computer.

[Kobzol]

#160149 has landed, let's try.

@bors r=cuviper

@bors try jobs=aarch64-msvc-1

[rust-bors]

📌 Commit 60d1bbf has been approved by cuviper

It is now in the queue for this repository.

[rust-bors]

💔 Test for af2b20b failed: CI. Failed job:

• auto - dist-x86_64-illumos (web logs, enhanced plaintext logs)

[RalfJung]

@jclulow @pfmooney looks like the Illumos runner uses an ancient GCC that cannot build LLVM 23

[rust-bors]

☀️ Try build successful (CI)
Build commit: db7ba90 (db7ba9036aa814f321e65bd65bb212a0a50dc47c)
Base parent: 922325b (922325bb13bfea5b41454318563f2a65e83c2336)

[ZuseZ4]

I've updated it to 10.1.0, the same that I also used for netbsd. Let's see if it passes: #160042 (comment)

[jclulow]

Please be sure to heed the comment about verifying the correctness of the binaries produced by the cross compiler if you're going to change the GCC version!

[RalfJung]

We would appreciate your help with that verification :)

[jclulow]

We would appreciate your help with that verification :)

I understand. It's midnight on Friday night and I just heard about this, so I wanted to make sure we're not just going to bump the number and move on before we take a look in the next couple of days. Thanks for the heads up.

[nikic]

LLVM theoretically supports GCC >= 7.4, so submitting a workaround for the issue upstream may work as well (i.e. use uint64_t instead of the enum class plus LLVM_PREFERRED_TYPE -- most likely this is also needed to actually get bit packing on MSVC).

Also I'm kind of confused why I didn't hit this when running the docker images locally before. I may have done something wrong, because this doesn't look like a problem when would have been introduced in the time since I tested it.

[dianqk]

Per https://stackoverflow.com/a/73511549, I think we can also disable -Werror for PassWrapper.cpp when building with GCC 7.x.

[nikic]

Oh, I didn't realize this is our own -Werror. We should definitely disable it for builds using GCC. (LLVM is not supposed to be warning free when using GCC, only when using recent Clang. And this can affect building our bindings via headers.)

[Kobzol]

I'll take a look into that.

[tgross35]

@tgross35 I don't need to do anything special for compiler-rt files that don't exist in previous LLVM versions, right? People not building against the vendored compiler-rt must use the precompiled compiler-rt functionality (optimized-compiler-builtins set to a path), correct?

It should just work here; these paths are relative to RUST_COMPILER_RT_ROOT which gets set by bootstrap

rust/src/bootstrap/src/core/build_steps/compile.rs

Line 636 in 65dd30f

cargo.env("RUST_COMPILER_RT_ROOT", &compiler_builtins_root);

.

https://github.com/rust-lang/compiler-builtins/blob/842b28fe3bc09b643ddaf639f2648f33ecbc0634/ci/download-compiler-rt.sh will need to be updated to get CI in that repo to work, but I can do that later.

c-b changes LGTM

View changes since the review

[Kobzol]

#160421 might fix this. @ZuseZ4 Can you please add it to your test PR and run the Illumos job?

[ZuseZ4]

@bors try jobs=dist-x86_64-illumos

[ZuseZ4]

runner is unexpectedly slow, restarting
@bors try jobs=dist-x86_64-illumos

[rust-bors]

⌛ Trying commit 8937199 with merge 9523e21…

(The previously running try build was automatically cancelled.)

To cancel the try build, run the command @bors try cancel.

Workflow: https://github.com/rust-lang/rust/actions/runs/30843935591