关于volatile和UNSAFE.getLong的取舍和区别？

Question

最近在看JCTools的源码，在查看SPSC队列的时候，有个疑问，


既然producerIndex属性已经被volatile修饰了，为什么还有一个UNSAFE.getLong方法来获取，性能更高吗，为什么要这样取舍？

Answer 1

因为你这是 Spsc 呀。

反正只有一个线程，没必要读的时候也 volatile，能省一点儿 LoadLoad Memory Barrier 是一点儿。

Answer 2

直接读会快啊，然后另一方面看作者的注释，作者应该是为了避开JIT，虽然volatile不会被JIT优化，但是毕竟JIT这种基本黑盒的东西，谁知道下一个版本又会多出来什么骚操作，直接unsafe读取，反而更简单可靠一些。

/**
 * Why should we resort to using Unsafe?<br>
 * <ol>
 * <li>To construct class fields which allow volatile/ordered/plain access: This requirement is covered by
 * {@link AtomicReferenceFieldUpdater} and similar but their performance is arguably worse than the DIY approach
 * (depending on JVM version) while Unsafe intrinsification is a far lesser challenge for JIT compilers.
 * <li>To construct flavors of {@link AtomicReferenceArray}.
 * <li>Other use cases exist but are not present in this library yet.
 * </ol>
 *
 * @author nitsanw
 */