From 25234f8652f59a796617c44688e8d4696611c314 Mon Sep 17 00:00:00 2001
From: Guillaume Gomez <guillaume1.gomez@gmail.com>
Date: Mon, 26 Jan 2026 18:24:52 +0100
Subject: [PATCH 1/2] Add two new target-specific intrinsics mapping

---
 src/intrinsic/llvm.rs | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/src/intrinsic/llvm.rs b/src/intrinsic/llvm.rs
index 10183c5ee52..fbf3050a8a1 100644
--- a/src/intrinsic/llvm.rs
+++ b/src/intrinsic/llvm.rs
@@ -1061,13 +1061,18 @@ pub fn intrinsic<'gcc, 'tcx>(name: &str, cx: &CodegenCx<'gcc, 'tcx>) -> Function
 
         "llvm.x86.xgetbv" => "__builtin_ia32_xgetbv",
         // NOTE: this doc specifies the equivalent GCC builtins: http://huonw.github.io/llvmint/llvmint/x86/index.html
+        // FIXME: Should handle other targets than `ia32`.
         "llvm.sqrt.v2f64" => "__builtin_ia32_sqrtpd",
+        // FIXME: Should handle other targets than `ia32`.
+        "llvm.sqrt.v4f32" => "__builtin_ia32_sqrtps",
         "llvm.sqrt.f32" => {
             let gcc_name = "__builtin_sqrtf";
             let func = cx.context.get_builtin_function(gcc_name);
             cx.functions.borrow_mut().insert(gcc_name.to_string(), func);
             return func;
         }
+        // FIXME: Should handle other targets than `ia32`.
+        "llvm.smax.v4i32" => "__builtin_ia32_pmaxsd128",
         "llvm.x86.avx512.pmul.dq.512" => "__builtin_ia32_pmuldq512_mask",
         "llvm.x86.avx512.pmulu.dq.512" => "__builtin_ia32_pmuludq512_mask",
         "llvm.x86.avx512.max.ps.512" => "__builtin_ia32_maxps512_mask",

From 436ae1e0a0d48967677bb6bf567d3d7c4697f498 Mon Sep 17 00:00:00 2001
From: Guillaume Gomez <guillaume1.gomez@gmail.com>
Date: Mon, 26 Jan 2026 18:25:25 +0100
Subject: [PATCH 2/2] Add regression test for `llvm.sqrt.v4f64` and
 `llvm.smax.v4i32` mapping

---
 tests/run/simd-ffi.rs | 102 ++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 102 insertions(+)
 create mode 100644 tests/run/simd-ffi.rs

diff --git a/tests/run/simd-ffi.rs b/tests/run/simd-ffi.rs
new file mode 100644
index 00000000000..67cc2e5b96e
--- /dev/null
+++ b/tests/run/simd-ffi.rs
@@ -0,0 +1,102 @@
+// Compiler:
+//
+// Run-time:
+//   status: 0
+
+// FIXME: Remove this test once <tests/run-make/simd-ffi/simd.rs> stops
+// ignoring GCC backend.
+
+#![allow(internal_features, non_camel_case_types)]
+// we can compile to a variety of platforms, because we don't need
+// cross-compiled standard libraries.
+#![feature(no_core, auto_traits)]
+#![no_core]
+#![feature(repr_simd, simd_ffi, link_llvm_intrinsics, lang_items, rustc_attrs)]
+
+#[derive(Copy)]
+#[repr(simd)]
+pub struct f32x4([f32; 4]);
+
+extern "C" {
+    #[link_name = "llvm.sqrt.v4f32"]
+    fn vsqrt(x: f32x4) -> f32x4;
+}
+
+pub fn foo(x: f32x4) -> f32x4 {
+    unsafe { vsqrt(x) }
+}
+
+#[derive(Copy)]
+#[repr(simd)]
+pub struct i32x4([i32; 4]);
+
+extern "C" {
+    // _mm_sll_epi32
+    #[cfg(all(any(target_arch = "x86", target_arch = "x86-64"), target_feature = "sse2"))]
+    #[link_name = "llvm.x86.sse2.psll.d"]
+    fn integer(a: i32x4, b: i32x4) -> i32x4;
+
+    // vmaxq_s32
+    #[cfg(target_arch = "arm")]
+    #[link_name = "llvm.arm.neon.vmaxs.v4i32"]
+    fn integer(a: i32x4, b: i32x4) -> i32x4;
+    // vmaxq_s32
+    #[cfg(target_arch = "aarch64")]
+    #[link_name = "llvm.aarch64.neon.maxs.v4i32"]
+    fn integer(a: i32x4, b: i32x4) -> i32x4;
+
+    // Use a generic LLVM intrinsic to do type checking on other platforms
+    #[cfg(not(any(
+        all(any(target_arch = "x86", target_arch = "x86-64"), target_feature = "sse2"),
+        target_arch = "arm",
+        target_arch = "aarch64"
+    )))]
+    #[link_name = "llvm.smax.v4i32"]
+    fn integer(a: i32x4, b: i32x4) -> i32x4;
+}
+
+pub fn bar(a: i32x4, b: i32x4) -> i32x4 {
+    unsafe { integer(a, b) }
+}
+
+#[lang = "pointee_sized"]
+pub trait PointeeSized {}
+
+#[lang = "meta_sized"]
+pub trait MetaSized: PointeeSized {}
+
+#[lang = "sized"]
+pub trait Sized: MetaSized {}
+
+#[lang = "copy"]
+pub trait Copy {}
+
+impl Copy for f32 {}
+impl Copy for i32 {}
+impl Copy for [f32; 4] {}
+impl Copy for [i32; 4] {}
+
+pub mod marker {
+    pub use Copy;
+}
+
+#[lang = "freeze"]
+auto trait Freeze {}
+
+#[macro_export]
+#[rustc_builtin_macro]
+macro_rules! Copy {
+    () => {};
+}
+#[macro_export]
+#[rustc_builtin_macro]
+macro_rules! derive {
+    () => {};
+}
+
+#[lang = "start"]
+fn start<T>(_main: fn() -> T, _argc: isize, _argv: *const *const u8, _sigpipe: u8) -> isize {
+    0
+}
+
+fn main() {}