Use atomic subtract in shared mutex.

src/core/SkSharedMutex.cpp

 /*
  * Copyright 2015 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #include "SkSharedMutex.h"
 
 #include "SkAtomics.h"
@@ skipping 166 matching lines @@
         fSharedQueue.wait();
     }
     ANNOTATE_RWLOCK_ACQUIRED(this, 0);
    
 }
 
 void SkSharedMutex::releaseShared() {
     ANNOTATE_RWLOCK_RELEASED(this, 0);
 
     // Decrement the shared count.
-    int32_t oldQueueCounts = fQueueCounts.fetch_add(~0U << kSharedOffset,
+    int32_t oldQueueCounts = fQueueCounts.fetch_sub(1 << kSharedOffset,
                                                     sk_memory_order_release);
 
     // If shared count is going to zero (because the old count == 1) and there are exclusive
     // waiters, then run a single exclusive waiter.
     if (((oldQueueCounts & kSharedMask) >> kSharedOffset) == 1
         && (oldQueueCounts & kWaitingExclusiveMask) > 0) {
         fExclusiveQueue.signal();
     }
 }
 
 #ifdef SK_DEBUG
 void SkSharedMutex::assertHeldShared() const {
     int32_t queueCounts = fQueueCounts.load(sk_memory_order_relaxed);
     // A very loose assert about the mutex being shared.
     SkASSERTF((queueCounts & kSharedMask) > 0,
               "running shared: %d, exclusive: %d, waiting shared: %d",
               (queueCounts & kSharedMask) >> kSharedOffset,
               (queueCounts & kWaitingExclusiveMask) >> kWaitingExlusiveOffset,
               (queueCounts & kWaitingSharedMask) >> kWaitingSharedOffset);
 }
 
 #endif