[clang] [llvm] [AArch64] Implement the atomic store with hint intrinsic (PR #198316)

[via cfe-commits]

================
@@ -0,0 +1,288 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6
+; RUN: llc -mtriple=aarch64-none-linux-gnu -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-none-linux-gnu -global-isel=1 -verify-machineinstrs < %s | FileCheck %s
+
+;
+; STSHH: Keep, Relaxed
+;
+
+define dso_local void @test_atomic_store_keep_relaxed_i8(ptr %ptr, i8 %val) nounwind {
+; CHECK-LABEL: test_atomic_store_keep_relaxed_i8:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    stshh keep
+; CHECK-NEXT:    strb w1, [x0]
+; CHECK-NEXT:    ret
+  store atomic i8 %val, ptr %ptr monotonic, align 8, !aarch64.atomic.hint !0
+  ret void
+}
+
+define dso_local void @test_atomic_store_keep_relaxed_i16(ptr %ptr, i16 %val) nounwind {
+; CHECK-LABEL: test_atomic_store_keep_relaxed_i16:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    stshh keep
+; CHECK-NEXT:    strh w1, [x0]
+; CHECK-NEXT:    ret
+  store atomic i16 %val, ptr %ptr monotonic, align 8, !aarch64.atomic.hint !0
+  ret void
+}
+
+define dso_local void @test_atomic_store_keep_relaxed_i32(ptr %ptr, i32 %val) nounwind {
+; CHECK-LABEL: test_atomic_store_keep_relaxed_i32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    stshh keep
+; CHECK-NEXT:    str w1, [x0]
+; CHECK-NEXT:    ret
+  store atomic i32 %val, ptr %ptr monotonic, align 8, !aarch64.atomic.hint !0
+  ret void
+}
+
+define dso_local void @test_atomic_store_keep_relaxed_i64(ptr %ptr, i64 %val) nounwind {
+; CHECK-LABEL: test_atomic_store_keep_relaxed_i64:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    stshh keep
+; CHECK-NEXT:    str x1, [x0]
+; CHECK-NEXT:    ret
+  store atomic i64 %val, ptr %ptr monotonic, align 8, !aarch64.atomic.hint !0
+  ret void
+}
+
+;
+; STSHH: Keep, Release
+;
+
+define dso_local void @test_atomic_store_keep_release_bfloat(ptr %ptr, bfloat %val) nounwind {
+; CHECK-LABEL: test_atomic_store_keep_release_bfloat:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    // kill: def $h0 killed $h0 def $s0
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    stshh keep
+; CHECK-NEXT:    stlrh w8, [x0]
+; CHECK-NEXT:    ret
+  store atomic bfloat %val, ptr %ptr release, align 8, !aarch64.atomic.hint !0
+  ret void
+}
+
+define dso_local void @test_atomic_store_keep_release_half(ptr %ptr, half %val) nounwind {
+; CHECK-LABEL: test_atomic_store_keep_release_half:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    // kill: def $h0 killed $h0 def $s0
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    stshh keep
+; CHECK-NEXT:    stlrh w8, [x0]
----------------
Lukacma wrote:

Just to add that the issue with rcpc3 exists for general builtins as well:
https://godbolt.org/z/bjGvKj5Yr



https://github.com/llvm/llvm-project/pull/198316