Remove unnecessary synchronization when cloning shared representation (#404)

src/bytes.rs

@@ -928,7 +928,7 @@ const KIND_VEC: usize = 0b1;
 const KIND_MASK: usize = 0b1;
 
 unsafe fn shared_clone(data: &AtomicPtr<()>, ptr: *const u8, len: usize) -> Bytes {
-    let shared = data.load(Ordering::Acquire);
+    let shared = data.load(Ordering::Relaxed);
     shallow_clone_arc(shared as _, ptr, len)
 }
 

src/bytes_mut.rs

@@ -1495,7 +1495,7 @@ static SHARED_VTABLE: Vtable = Vtable {
 };
 
 unsafe fn shared_v_clone(data: &AtomicPtr<()>, ptr: *const u8, len: usize) -> Bytes {
-    let shared = data.load(Ordering::Acquire) as *mut Shared;
+    let shared = data.load(Ordering::Relaxed) as *mut Shared;
     increment_shared(shared);
 
     let data = AtomicPtr::new(shared as _);