diff --git a/builtins-test/tests/lse.rs b/builtins-test/tests/lse.rs
index 53167d98f..0d85228d7 100644
--- a/builtins-test/tests/lse.rs
+++ b/builtins-test/tests/lse.rs
@@ -1,4 +1,5 @@
 #![feature(decl_macro)] // so we can use pub(super)
+#![feature(macro_metavar_expr_concat)]
 #![cfg(all(target_arch = "aarch64", target_os = "linux", not(feature = "no-asm")))]
 
 /// Translate a byte size to a Rust type.
@@ -87,7 +88,7 @@ test_op!(add, |left, right| left.wrapping_add(right));
 test_op!(clr, |left, right| left & !right);
 test_op!(xor, std::ops::BitXor::bitxor);
 test_op!(or, std::ops::BitOr::bitor);
-
+use compiler_builtins::{foreach_bytes, foreach_ordering};
 compiler_builtins::foreach_cas!(cas::test);
 compiler_builtins::foreach_cas16!(test_cas16);
 compiler_builtins::foreach_swp!(swap::test);
diff --git a/compiler-builtins/build.rs b/compiler-builtins/build.rs
index d37fdc5df..e909a0dcb 100644
--- a/compiler-builtins/build.rs
+++ b/compiler-builtins/build.rs
@@ -1,9 +1,6 @@
 mod configure;
 
-use std::collections::BTreeMap;
 use std::env;
-use std::path::PathBuf;
-use std::sync::atomic::Ordering;
 
 use configure::{Target, configure_aliases, configure_f16_f128};
 
@@ -85,10 +82,6 @@ fn main() {
     {
         println!("cargo:rustc-cfg=kernel_user_helpers")
     }
-
-    if llvm_target[0].starts_with("aarch64") {
-        generate_aarch64_outlined_atomics();
-    }
 }
 
 /// Run configuration for `libm` since it is included directly.
@@ -131,61 +124,6 @@ fn configure_libm(target: &Target) {
     println!("cargo:rustc-cfg=feature=\"unstable-intrinsics\"");
 }
 
-fn aarch64_symbol(ordering: Ordering) -> &'static str {
-    match ordering {
-        Ordering::Relaxed => "relax",
-        Ordering::Acquire => "acq",
-        Ordering::Release => "rel",
-        Ordering::AcqRel => "acq_rel",
-        _ => panic!("unknown symbol for {ordering:?}"),
-    }
-}
-
-/// The `concat_idents` macro is extremely annoying and doesn't allow us to define new items.
-/// Define them from the build script instead.
-/// Note that the majority of the code is still defined in `aarch64.rs` through inline macros.
-fn generate_aarch64_outlined_atomics() {
-    use std::fmt::Write;
-    // #[macro_export] so that we can use this in tests
-    let gen_macro =
-        |name| format!("#[macro_export] macro_rules! foreach_{name} {{ ($macro:path) => {{\n");
-
-    // Generate different macros for add/clr/eor/set so that we can test them separately.
-    let sym_names = ["cas", "ldadd", "ldclr", "ldeor", "ldset", "swp"];
-    let mut macros = BTreeMap::new();
-    for sym in sym_names {
-        macros.insert(sym, gen_macro(sym));
-    }
-
-    // Only CAS supports 16 bytes, and it has a different implementation that uses a different macro.
-    let mut cas16 = gen_macro("cas16");
-
-    for ordering in [
-        Ordering::Relaxed,
-        Ordering::Acquire,
-        Ordering::Release,
-        Ordering::AcqRel,
-    ] {
-        let sym_ordering = aarch64_symbol(ordering);
-        for size in [1, 2, 4, 8] {
-            for (sym, macro_) in &mut macros {
-                let name = format!("__aarch64_{sym}{size}_{sym_ordering}");
-                writeln!(macro_, "$macro!( {ordering:?}, {size}, {name} );").unwrap();
-            }
-        }
-        let name = format!("__aarch64_cas16_{sym_ordering}");
-        writeln!(cas16, "$macro!( {ordering:?}, {name} );").unwrap();
-    }
-
-    let mut buf = String::new();
-    for macro_def in macros.values().chain(std::iter::once(&cas16)) {
-        buf += macro_def;
-        buf += "}; }\n";
-    }
-    let out_dir = PathBuf::from(std::env::var("OUT_DIR").unwrap());
-    std::fs::write(out_dir.join("outlined_atomics.rs"), buf).unwrap();
-}
-
 /// Emit directives for features we expect to support that aren't in `Cargo.toml`.
 ///
 /// These are mostly cfg elements emitted by this `build.rs`.
diff --git a/compiler-builtins/src/aarch64_linux.rs b/compiler-builtins/src/aarch64_linux.rs
index e238d0237..2402a3fe1 100644
--- a/compiler-builtins/src/aarch64_linux.rs
+++ b/compiler-builtins/src/aarch64_linux.rs
@@ -262,8 +262,78 @@ macro_rules! or {
     };
 }
 
-// See `generate_aarch64_outlined_atomics` in build.rs.
-include!(concat!(env!("OUT_DIR"), "/outlined_atomics.rs"));
+#[macro_export]
+macro_rules! foreach_ordering {
+    ($macro:path, $bytes:tt, $name:ident) => {
+        $macro!( Relaxed, $bytes, ${concat($name, _relax)} );
+        $macro!( Acquire, $bytes, ${concat($name, _acq)} );
+        $macro!( Release, $bytes, ${concat($name, _rel)} );
+        $macro!( AcqRel, $bytes, ${concat($name, _acq_rel)} );
+    };
+    ($macro:path, $name:ident) => {
+        $macro!( Relaxed, ${concat($name, _relax)} );
+        $macro!( Acquire, ${concat($name, _acq)} );
+        $macro!( Release, ${concat($name, _rel)} );
+        $macro!( AcqRel, ${concat($name, _acq_rel)} );
+    };
+}
+
+#[macro_export]
+macro_rules! foreach_bytes {
+    ($macro:path, $name:ident) => {
+        foreach_ordering!( $macro, 1, ${concat(__aarch64_, $name, "1")} );
+        foreach_ordering!( $macro, 2, ${concat(__aarch64_, $name, "2")} );
+        foreach_ordering!( $macro, 4, ${concat(__aarch64_, $name, "4")} );
+        foreach_ordering!( $macro, 8, ${concat(__aarch64_, $name, "8")} );
+    };
+}
+
+/// Generate different macros for cas/swp/add/clr/eor/set so that we can test them separately.
+#[macro_export]
+macro_rules! foreach_cas {
+    ($macro:path) => {
+        foreach_bytes!($macro, cas);
+    };
+}
+
+/// Only CAS supports 16 bytes, and it has a different implementation that uses a different macro.
+#[macro_export]
+macro_rules! foreach_cas16 {
+    ($macro:path) => {
+        foreach_ordering!($macro, __aarch64_cas16);
+    };
+}
+#[macro_export]
+macro_rules! foreach_swp {
+    ($macro:path) => {
+        foreach_bytes!($macro, swp);
+    };
+}
+#[macro_export]
+macro_rules! foreach_ldadd {
+    ($macro:path) => {
+        foreach_bytes!($macro, ldadd);
+    };
+}
+#[macro_export]
+macro_rules! foreach_ldclr {
+    ($macro:path) => {
+        foreach_bytes!($macro, ldclr);
+    };
+}
+#[macro_export]
+macro_rules! foreach_ldeor {
+    ($macro:path) => {
+        foreach_bytes!($macro, ldeor);
+    };
+}
+#[macro_export]
+macro_rules! foreach_ldset {
+    ($macro:path) => {
+        foreach_bytes!($macro, ldset);
+    };
+}
+
 foreach_cas!(compare_and_swap);
 foreach_cas16!(compare_and_swap_i128);
 foreach_swp!(swap);
diff --git a/compiler-builtins/src/lib.rs b/compiler-builtins/src/lib.rs
index 6a6b28067..ef3299d69 100644
--- a/compiler-builtins/src/lib.rs
+++ b/compiler-builtins/src/lib.rs
@@ -8,6 +8,7 @@
 #![feature(linkage)]
 #![feature(naked_functions)]
 #![feature(repr_simd)]
+#![feature(macro_metavar_expr_concat)]
 #![cfg_attr(f16_enabled, feature(f16))]
 #![cfg_attr(f128_enabled, feature(f128))]
 #![no_builtins]