rustc_target: update x86_win64 to match the documented calling convention for f128

[durin42]

llvm/llvm-project@5ee1c0b updates llvm to match the documented calling convention to pass f128 indirectly. This change makes us do that on all versions of LLVM, not just starting with LLVM 21.

@rustbot label llvm-main

try-job: dist-x86_64-msvc
try-job: dist-x86_64-mingw
try-job: x86_64-msvc-1
try-job: x86_64-msvc-2
try-job: x86_64-mingw-1
try-job: x86_64-mingw-2

[rustbot]

r? @Mark-Simulacrum

rustbot has assigned @Mark-Simulacrum.
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

[Mark-Simulacrum]

r? @tgross35

Looks OK but want to get someone more familiar with i128 to review.

[tgross35]

I meant to send a patch once my LLVM change merged, but we should also update our codegen to match. Could you change

rust/compiler/rustc_target/src/callconv/x86_win64.rs

Lines 26 to 43 in 8279176

	if is_ret && matches!(scalar.primitive(), Primitive::Int(Integer::I128, _)) {
	if cx.target_spec().rustc_abi == Some(RustcAbi::X86Softfloat) {
	// Use the native `i128` LLVM type for the softfloat ABI -- in other words, adjust nothing.
	} else {
	// `i128` is returned in xmm0 by Clang and GCC
	// FIXME(#134288): This may change for the `-msvc` targets in the future.
	let reg = Reg { kind: RegKind::Vector, size: Size::from_bits(128) };
	a.cast_to(reg);
	}
	} else if a.layout.size.bytes() > 8
	&& !matches!(scalar.primitive(), Primitive::Float(Float::F128))
	{
	// Match what LLVM does for `f128` so that `compiler-builtins` builtins match up
	// with what LLVM expects.
	a.make_indirect();
	} else {
	a.extend_integer_width_to(32);
	}

so i128 and f128 use the same ABI? That probably means a codegen test update.

[tgross35]

You should be able to take the CC change from #137779.

[tgross35]

@bors try

r=me once that passes

[bors]

⌛ Trying commit 8399166 with merge b83a6ea...

[bors]

💔 Test failed - checks-actions

[tgross35]

The test crash seems to indicate some kind of ABI mismatch. I am guessing this happens during the libcall to __eqtf2 or __unordtf2, but I am unsure why this would be since we don't do anything special with that ABI https://github.com/rust-lang/compiler-builtins/blob/f1b9055b0d09370a474ce5e9b9d78ac4758cedf8/compiler-builtins/src/float/cmp.rs#L181-L189.

[beetrees]

Only LLVM 21+ uses the new indirect ABI; Rust currently ships with LLVM 20 which will use the old ABI when generating calls to functions in compiler-builtins.

[tgross35]

Ah right, I guess we can't switch yet without breaking things. @durin42 sorry for all the confusion here but I guess your original PR with {{(%xmm1|\(%rdx\))}} and FIXME(llvm21) was the best way to get ToT CI working.

If you want to put that into a separate PR, we can leave this one around until the LLVM21 update happens.

[durin42]

I've updated this PR to go back to the test-only edit, with a FIXME added.

[tgross35]

@bors try

[bors]

⌛ Trying commit d7e7f8b with merge e3493ce...

[bors]

☀️ Try build successful - checks-actions
Build commit: e3493ce (e3493cef53b7d8d1e2429715806de32e08f372ab)

[tgross35]

Sorry about the back and forth, thanks for the changes!

@bors r+

[bors]

📌 Commit d7e7f8b has been approved by tgross35

It is now in the queue for this repository.

[durin42]

Not a problem at all, thanks for the review!

…

On Thu, Apr 10, 2025, 20:42 bors ***@***.***> wrote: 📌 Commit d7e7f8b <d7e7f8b> has been approved by tgross35 It is now in the queue <https://bors.rust-lang.org/homu/queue/rust> for this repository. — Reply to this email directly, view it on GitHub <#138182 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAE6LIHBIV4ZQYZ3ME2H7D2Y4FWFAVCNFSM6AAAAABYR5TEMWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDOOJVGUYTMNJTGM> . You are receiving this because you were mentioned.Message ID: ***@***.***> *bors* left a comment (rust-lang/rust#138182) <#138182 (comment)> 📌 Commit d7e7f8b <d7e7f8b> has been approved by tgross35 It is now in the queue <https://bors.rust-lang.org/homu/queue/rust> for this repository. — Reply to this email directly, view it on GitHub <#138182 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAE6LIHBIV4ZQYZ3ME2H7D2Y4FWFAVCNFSM6AAAAABYR5TEMWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDOOJVGUYTMNJTGM> . You are receiving this because you were mentioned.Message ID: ***@***.***>