Skip to content

Rustc pull update - #2264

Merged
folkertdev merged 4 commits into
mainfrom
rustc-pull
Nov 15, 2026
Merged

folkertdev merged 4 commits into
mainfrom
rustc-pull

Conversation

@workflows-stdarch

Copy link
Copy Markdown

Latest update from rustc.

@rustbot

rustbot commented Nov 5, 2026

Copy link
Copy Markdown
Collaborator

Thanks for the pull request, and welcome! The Rust Project has assigned @Amanieu (or someone else) to review your changes, you should hear from them (or someone else) within the next two weeks.

Why was this reviewer chosen?

The reviewer was selected based on:

  • Owners of files modified in this PR: @Amanieu, @adamgemmell, @davidtwco, @folkertdev
  • @Amanieu, @adamgemmell, @davidtwco, @folkertdev expanded to Amanieu, adamgemmell, davidtwco, folkertdev
  • Random selection from Amanieu, adamgemmell, davidtwco, folkertdev

@folkertdev

Copy link
Copy Markdown
Contributor

@sayantn I'm a bit stumped on this

   Compiling core_arch v0.1.5 (D:\a\stdarch\stdarch\crates\core_arch)
   Compiling simd-test-macro v0.1.0 (D:\a\stdarch\stdarch\crates\simd-test-macro)
   Compiling stdarch-test v0.1.0 (D:\a\stdarch\stdarch\crates\stdarch-test)
error: stdarch_test_shim___tile_cmmimfp16ps_tcmmimfp16ps: Failed to config tile register, please define the shape earlier

error: could not compile `core_arch` (lib test) due to 1 previous error

I think the only thing that changed is the LLVM version got bumped for nightly?

@sayantn

sayantn commented Nov 10, 2026

Copy link
Copy Markdown
Contributor

I have no clue either. Most probably it's some llvm bug, or it's also possible that we need to do a similar preparation routine (as we do in linux) for windows-gnu. Will take a look. In the meantime can you do a ci run with fail-fast disabled? I want to see if other targets stumble across this too

@adamgemmell

Copy link
Copy Markdown
Contributor

It's reproducible on linux: https://godbolt.org/z/eGaf7YKGf

opt-level 2 & 3 fail, but 1 and 0 pass. Setting -opt-bisect-limit doesn't fix anything however which seems a little odd.

@sayantn

sayantn commented Nov 11, 2026 •

Copy link
Copy Markdown
Contributor

Found a smaller repro, apparently LLVM doesn't like if the tiles are loaded before anything else https://godbolt.org/z/KqeWaeKKE
The only difference between the 2 is that the tile corresponding to %a is loaded before loading all fields of %b in the RHS (RHS is what rustc generates). Apparently the RHS never used to compile on LLC (since 20.1). So this means the problem is in rustc, maybe some load order changed recently?

@folkertdev

Copy link
Copy Markdown
Contributor

We changed loads to no longer use the LLVM type, maybe that matters?

among others

This updates the rust-version file to ee8268e23d0e2e308b52e4f63df69afb69bc61a4.
@folkertdev

Copy link
Copy Markdown
Contributor

Actually given those symptoms, maybe it's

So it's an error that was always there, but we recently stopped ignoring it?


Anyway, given that the IR we generate is wrong in a way that we can't easily change, what do we do here?

@adamgemmell

Copy link
Copy Markdown
Contributor

A bisect pointed me at that PR, f90a2d27c2a1e18f10f4fd84c0dcd02710c37de8. For this particular intrinsic, 92998d2 avoids the LLVM error but then the assertion test fails, and there's a few other intrinsics with different implementation patterns that also fail anyway.

I filed rust-lang/rust#165931, we could maybe work around the malformed IR but it's a compiler issue really. I think we should skip the tests for now to unblock merges.

panstromek and others added 3 commits November 15, 2026 16:24
They have the same signature now, so we can unify them. The implementation for tuple is selected based on whether the concatenated string contains field names.
they run into rust-lang/rust#165931, which we can't fix here.
@folkertdev

Copy link
Copy Markdown
Contributor

I've disabled the assembly tests on functions that take __tile1024i by value now. That passes in the docker container for me locally, so hopefully that unblocks us.

@folkertdev
folkertdev added this pull request to the merge queue Nov 15, 2026
Merged via the queue into main with commit 349454b Nov 15, 2026
87 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants