Rollup merge of rust-lang#83655 - sebpop:arm64-outline-atomics, r=joshtriplett

[aarch64] add target feature outline-atomics

Enable outline-atomics by default as enabled in clang by the following commit
https://reviews.llvm.org/rGc5e7e649d537067dec7111f3de1430d0fc8a4d11

Performance improves by several orders of magnitude when using the LSE instructions
instead of the ARMv8.0 compatible load/store exclusive instructions.

Tested on Graviton2 aarch64-linux with
x.py build && x.py install && x.py test

Author: Dylan-DPC

compiler/rustc_target/src/spec/aarch64_unknown_linux_gnu.rs

@@ -8,6 +8,7 @@ pub fn target() -> Target {
         | SanitizerSet::MEMORY
         | SanitizerSet::THREAD
         | SanitizerSet::HWADDRESS;
+    base.features = "+outline-atomics".to_string();
 
     Target {
         llvm_target: "aarch64-unknown-linux-gnu".to_string(),

compiler/rustc_target/src/spec/aarch64_unknown_linux_musl.rs

@@ -3,6 +3,7 @@ use crate::spec::{Target, TargetOptions};
 pub fn target() -> Target {
     let mut base = super::linux_musl_base::opts();
     base.max_atomic_width = Some(128);
+    base.features = "+outline-atomics".to_string();
 
     Target {
         llvm_target: "aarch64-unknown-linux-musl".to_string(),

src/test/assembly/asm/aarch64-outline-atomics.rs

@@ -0,0 +1,15 @@
+// min-llvm-version: 12.0
+// assembly-output: emit-asm
+// compile-flags: -O
+// compile-flags: --target aarch64-unknown-linux-gnu
+// needs-llvm-components: aarch64
+
+#![crate_type = "rlib"]
+
+use std::sync::atomic::{AtomicI32, Ordering::*};
+
+pub fn compare_exchange(a: &AtomicI32) {
+    // On AArch64 LLVM should outline atomic operations.
+    // CHECK: __aarch64_cas4_relax
+    let _ = a.compare_exchange(0, 10, Relaxed, Relaxed);
+}