[clang] [clang][CodeGen] Do not use atomicrmw in compound assignments with FP conversions (PR #209114)

[Antonio Frighetto via cfe-commits]

================
@@ -152,6 +152,25 @@ void testandeq(void)
   l &= 42;
   s &= 42;
 }
+// CHECK: testaddeq_fp
+void testaddeq_fp(void)
+{
+  // CHECK:      [[CONV:%.*]] = uitofp i32 [[TMP0:%.*]] to double
----------------
antoniofrighetto wrote:

This appears to be an orthogonal issue, as `_Atomic int` is signed and the this should be a sitofp. Will open a patch soon.

https://github.com/llvm/llvm-project/pull/209114