[compiler-rt] [sanitizer_common][AIX] Use scoped pragma to suppress atomic alignment warnings (PR #139272)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-compiler-rt-sanitizer

Author: Jake Egan (jakeegan)

<details>
<summary>Changes</summary>

Have the warning suppression apply only to the code that is currently affected. The suppression is guarded via preprocessor conditions to cases where it is tested and known to be needed.

Issue: https://github.com/llvm/llvm-project/issues/138916

---
Full diff: https://github.com/llvm/llvm-project/pull/139272.diff


1 Files Affected:

- (modified) compiler-rt/lib/sanitizer_common/sanitizer_atomic_clang.h (+15) 


``````````diff
diff --git a/compiler-rt/lib/sanitizer_common/sanitizer_atomic_clang.h b/compiler-rt/lib/sanitizer_common/sanitizer_atomic_clang.h
index 1414092e38d7e..fded66546f810 100644
--- a/compiler-rt/lib/sanitizer_common/sanitizer_atomic_clang.h
+++ b/compiler-rt/lib/sanitizer_common/sanitizer_atomic_clang.h
@@ -14,6 +14,18 @@
 #ifndef SANITIZER_ATOMIC_CLANG_H
 #define SANITIZER_ATOMIC_CLANG_H
 
+// Helper to suppress warnings related to 8-byte atomic accesses when the target
+// is 32-bit AIX (where such accesses use libatomic).
+#if defined(_AIX) && !defined(__powerpc64__) && defined(__clang__)
+#  define SANITIZER_IGNORE_ATOMIC_ALIGNMENT_BEGIN \
+    _Pragma("clang diagnostic push")              \
+        _Pragma("clang diagnostic ignored \"-Watomic-alignment\"")
+#  define SANITIZER_IGNORE_ATOMIC_ALIGNMENT_END _Pragma("clang diagnostic pop")
+#else
+#  define SANITIZER_IGNORE_ATOMIC_ALIGNMENT_BEGIN
+#  define SANITIZER_IGNORE_ATOMIC_ALIGNMENT_END
+#endif
+
 namespace __sanitizer {
 
 // We use the compiler builtin atomic operations for loads and stores, which
@@ -35,6 +47,7 @@ inline void proc_yield(int cnt) {
 #endif
 }
 
+SANITIZER_IGNORE_ATOMIC_ALIGNMENT_BEGIN
 template <typename T>
 inline typename T::Type atomic_load(const volatile T *a, memory_order mo) {
   DCHECK(mo == memory_order_relaxed || mo == memory_order_consume ||
@@ -92,6 +105,8 @@ inline bool atomic_compare_exchange_weak(volatile T *a, typename T::Type *cmp,
   return atomic_compare_exchange_strong(a, cmp, xchg, mo);
 }
 
+SANITIZER_IGNORE_ATOMIC_ALIGNMENT_END
+
 }  // namespace __sanitizer
 
 #undef ATOMIC_ORDER

``````````

</details>


https://github.com/llvm/llvm-project/pull/139272