Skip to content

Don't compute FnAbi for LLVM intrinsics in backends #150768

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
rust-bors merged 3 commits into from
Feb 12, 2026
Merged

Don't compute FnAbi for LLVM intrinsics in backends #150768

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
acc8c0b
Reduce usage of FnAbi in codegen_llvm_intrinsic_call
bjorn3 Dec 28, 2025
f1ab003
Don't compute FnAbi for LLVM intrinsics in backends
bjorn3 Jan 8, 2026
fe9715b
Remove support for ScalarPair unadjusted arguments
bjorn3 Jan 8, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 8 additions & 11 deletions compiler/rustc_codegen_gcc/src/intrinsic/mod.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,13 +22,18 @@ use rustc_codegen_ssa::traits::{
ArgAbiBuilderMethods, BaseTypeCodegenMethods, BuilderMethods, ConstCodegenMethods,
IntrinsicCallBuilderMethods, LayoutTypeCodegenMethods,
};
use rustc_data_structures::fx::FxHashSet;
use rustc_middle::bug;
use rustc_middle::ty::layout::{FnAbiOf, LayoutOf};
#[cfg(feature = "master")]
use rustc_middle::ty::layout::FnAbiOf;
use rustc_middle::ty::layout::LayoutOf;
use rustc_middle::ty::{self, Instance, Ty};
use rustc_span::{Span, Symbol, sym};
use rustc_target::callconv::{ArgAbi, PassMode};

use crate::abi::{FnAbiGccExt, GccType};
#[cfg(feature = "master")]
use crate::abi::FnAbiGccExt;
use crate::abi::GccType;
use crate::builder::Builder;
use crate::common::{SignType, TypeReflection};
use crate::context::CodegenCx;
Expand Down Expand Up @@ -620,8 +625,6 @@ impl<'a, 'gcc, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'a, 'gcc, 'tc
*func
} else {
self.linkage.set(FunctionType::Extern);
let fn_abi = self.fn_abi_of_instance(instance, ty::List::empty());
let fn_ty = fn_abi.gcc_type(self);

let func = match sym {
"llvm.fma.f16" => {
Expand All @@ -634,13 +637,7 @@ impl<'a, 'gcc, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'a, 'gcc, 'tc

self.intrinsics.borrow_mut().insert(sym.to_string(), func);

self.on_stack_function_params
.borrow_mut()
.insert(func, fn_ty.on_stack_param_indices);
#[cfg(feature = "master")]
for fn_attr in fn_ty.fn_attributes {
func.add_attribute(fn_attr);
}
self.on_stack_function_params.borrow_mut().insert(func, FxHashSet::default());
Copy link
Member

@Mark-Simulacrum Mark-Simulacrum Feb 8, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I guess this is the same ABI question as below - we're assuming there's no on-stack parameter passing since it's an intrinsic?

Copy link
Member Author

@bjorn3 bjorn3 Feb 8, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

extern "unadjusted" previously forced PassMode::Direct for all arguments, which means that on-stack parameters would never occur.


crate::attributes::from_fn_attrs(self, func, instance);

Expand Down
33 changes: 27 additions & 6 deletions compiler/rustc_codegen_llvm/src/intrinsic.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -646,10 +646,32 @@ impl<'ll, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> {
) -> Self::Value {
let tcx = self.tcx();

// FIXME remove usage of fn_abi
let fn_abi = self.fn_abi_of_instance(instance, ty::List::empty());
assert!(!fn_abi.ret.is_indirect());
let fn_ty = fn_abi.llvm_type(self);
let fn_ty = instance.ty(tcx, self.typing_env());
let fn_sig = match *fn_ty.kind() {
ty::FnDef(def_id, args) => {
tcx.instantiate_bound_regions_with_erased(tcx.fn_sig(def_id).instantiate(tcx, args))
}
_ => unreachable!(),
};
assert!(!fn_sig.c_variadic);

let ret_layout = self.layout_of(fn_sig.output());
let llreturn_ty = if ret_layout.is_zst() {
self.type_void()
} else {
ret_layout.immediate_llvm_type(self)
};

let mut llargument_tys = Vec::with_capacity(fn_sig.inputs().len());
for &arg in fn_sig.inputs() {
let arg_layout = self.layout_of(arg);
if arg_layout.is_zst() {
continue;
}
llargument_tys.push(arg_layout.immediate_llvm_type(self));
}

let fn_ty = self.type_func(&llargument_tys, llreturn_ty);

let fn_ptr = if let Some(&llfn) = self.intrinsic_instances.borrow().get(&instance) {
llfn
Expand All @@ -665,12 +687,11 @@ impl<'ll, 'tcx> IntrinsicCallBuilderMethods<'tcx> for Builder<'_, 'll, 'tcx> {
let llfn = declare_raw_fn(
self,
sym,
fn_abi.llvm_cconv(self),
llvm::CCallConv,
Copy link
Member

@Mark-Simulacrum Mark-Simulacrum Feb 8, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is this effectively just a lie? Or how does LLVM model the ABI for intrinsics? Is there a hidden assumption here that intrinsics only ever take ~integers/pointers that don't have any special handling in terms of ABI lowering of the arguments?

Copy link
Member Author

@bjorn3 bjorn3 Feb 8, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This matches what was effectively done previously. I don't know if LLVM actually cares about the calling convention for intrinsics. It recognizes intrinsics based on a builtin list of symbol names. (if you declare an llvm.* function that doesn't match a known intrinsic, it will treat it like a regular function and emit an import of said function in the emitted object file)

llvm::UnnamedAddr::Global,
llvm::Visibility::Default,
fn_ty,
);
fn_abi.apply_attrs_llfn(self, llfn, Some(instance));

llfn
};
Expand Down
Loading