// getDeclaredField to throw a NoSuchFieldException when the field is definitely there.
      // For these users fallback to a suboptimal implementation, based on synchronized. This will
      // be a definite performance hit to those users.
      thrownReflectionFailure = reflectionFailure;
      helper = new SynchronizedAtomicHelper();
    }
    ATOMIC_HELPER = helper;

 * {@link Multiset#entrySet()}, as do many methods acting on the set returned by {@link
 * #elementSet()}. Override those methods for better performance.
 *
 * @author Louis Wasserman
 */
@GwtCompatible(emulated = true)
@ElementTypesAreNonnullByDefault
abstract class AbstractSortedMultiset<E extends @Nullable Object> extends AbstractMultiset<E>

 *   <li>equalsImpl: which implementation of array equality to use
 * </ul>
 *
 * <p><b>Important note:</b> the primary goal of this benchmark is to ensure that varying {@code
 * whereToDiffer} produces no observable change in performance. We want to make sure that the array
 * equals implementation is *not* short-circuiting to prevent timing-based attacks. Being fast is
 * only a secondary goal.
 *
 * @author Kurt Alfred Kluever
 */

 *   <li>equalsImpl: which implementation of array equality to use
 * </ul>
 *
 * <p><b>Important note:</b> the primary goal of this benchmark is to ensure that varying {@code
 * whereToDiffer} produces no observable change in performance. We want to make sure that the array
 * equals implementation is *not* short-circuiting to prevent timing-based attacks. Being fast is
 * only a secondary goal.
 *
 * @author Kurt Alfred Kluever
 */

 *
 * <p><b>Note:</b> {@code ComparisonChain} instances are <b>immutable</b>. For this utility to work
 * correctly, calls must be chained as illustrated above.
 *
 * <p>Performance note: Even though the {@code ComparisonChain} caller always invokes its {@code
 * compare} methods unconditionally, the {@code ComparisonChain} implementation stops calling its

    long result = 0;
    // Due to the way we shift, we can stop iterating once we've run out of data, the rest
    // of the result already being filled with zeros.

    // This loop is critical to performance, so please check HashBenchmark if altering it.
    int limit = Math.min(length, 8);
    for (int i = 0; i < limit; i++) {
      // Shift value left while iterating logically through the array.

 *
 * <p>In order to allow proper garbage collection of unused locks, the edges of the lock graph are
 * weak references.
 *
 * <p><strong>Performance</strong>
 *
 * <p>The extra bookkeeping done by cycle detecting locks comes at some cost to performance.
 * Benchmarks (as of December 2011) show that:
 *
 * <ul>
 *   <li>for an unnested {@code lock()} and {@code unlock()}, a cycle detecting lock takes 38ns as

 * or even removal, in a future release. An API bearing this annotation is exempt from any
 * compatibility guarantees made by its containing library. Note that the presence of this
 * annotation implies nothing about the quality or performance of the API in question, only the fact
 * that it is not "API-frozen."
 *
 * <p>It is generally safe for <i>applications</i> to depend on beta APIs, at the cost of some extra

    } else {
      numberRemoved = oldCount;
      backingMap.remove(element);
    }

    frequency.add(-numberRemoved);
    size -= numberRemoved;
    return oldCount;
  }

  // Roughly a 33% performance improvement over AbstractMultiset.setCount().
  @CanIgnoreReturnValue
  @Override
  public int setCount(@ParametricNullness E element, int count) {
    checkNonnegative(count, "count");

    Count existingCounter;

   *
   * <p>If you know that no node in {@code graph} is reachable by more than one path from the start
   * node(s), consider using {@link #forTree(SuccessorsFunction)} instead.
   *
   * <p><b>Performance notes</b>
   *
   * <ul>
   *   <li>Traversals require <i>O(n)</i> time (where <i>n</i> is the number of nodes reachable from
   *       the start node), assuming that the node objects have <i>O(1)</i> {@code equals()} and