LLVM: LLVM-20.0 removes MMX types

[maurer]

See llvm/llvm-project#98505

@rustbot label: +llvm-main

[rustbot]

r? @cuviper

rustbot has assigned @cuviper.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[cuviper]

Can we just remove it entirely, unconditionally? I don't see any use of the Rust TypeKind::X86_MMX, and it seems better to remove the temptation since this will soon be unavailable anyway.

[maurer]

I've removed it entirely. The risk with this is that if LLVM does generate an x86_mmx type ID somewhere, we'll error out if we ask what kind it is. The specific risk I'm thinking of looks like:

fn foo<'ll>(cx: &CodegenCx<'ll>, ty: &'ll type) {
  match cx.type_kind(ty) {
    TypeKind::Array => do_array_thing();
    _ => do_default_thing();
  }
}

This would error out on the generation of the type_kind for mmx, even if the default case was going to cover things in every occurrence of mmx, as the generation of the kind would produce an error "Rust does not support the TypeId: blah"

We can see what test suites think about it though.

[cuviper]

This would error out on the generation of the type_kind for mmx, even if the default case was going to cover things in every occurrence of mmx, as the generation of the kind would produce an error "Rust does not support the TypeId: blah"

Ack. It passed PR CI, at least, and we're still early in the release cycle -- if something comes up outside of CI, we can always revert to your initial narrower change.

@bors r+ rollup

[bors]

📌 Commit 38931cd has been approved by cuviper

It is now in the queue for this repository.