为什么volatile比非易失性更快？

Question

阅读问题后为什么处理排序数组的速度比未排序数组快？ 我们曾尝试将变量设置为volatile(我预计,当我使用volatile时,它必须工作得更慢,但工作速度更快)这是我的代码没有volatile :(工作时间约为11秒)

import java.util.Arrays;
import java.util.Random;

public class GGGG {

public static void main(String[] args) {
    int arraySize = 32768;
    int data[];
    data = new int[arraySize];

    Random rnd = new Random(0);
    for (int c = 0; c < arraySize; ++c) {
        data[c] = rnd.nextInt() % 256;
    }

    Arrays.sort(data);

    long start = System.nanoTime();
    long sum = 0;

    for (int i = 0; i < 200000; ++i) {
        for (int c = 0; c < arraySize; ++c) {
            if (data[c] >= 128) {
                sum += data[c];
            }
        }
    }

    System.out.println((System.nanoTime() - start) / 1000000000.0);
    System.out.println("sum = " + sum);

    System.out.println("=========================");
}

输出是:

10.876173341
sum = 310368400000
=========================

这是当我使用arraySize和数据变量作为volatile时,它工作大约7秒:

import java.util.Arrays;
import java.util.Random;

public class GGGG {

static volatile int arraySize = 32768;
static volatile int data[];

public static void main(String[] args) {
    data = new int[arraySize];

    Random rnd = new Random(0);
    for (int c = 0; c < arraySize; ++c) {
        data[c] = rnd.nextInt() % 256;
    }

    Arrays.sort(data);

    long start = System.nanoTime();
    long sum = 0;

    for (int i = 0; i < 200000; ++i) {
        for (int c = 0; c < arraySize; ++c) {
            if (data[c] >= 128) {
                sum += data[c];
            }
        }
    }

    System.out.println((System.nanoTime() - start) / 1000000000.0);
    System.out.println("sum = " + sum);

    System.out.println("=========================");
}

而volatile的输出是:

6.776267265
sum = 310368400000
=========================

所有我都期望用volatile来减缓这个过程,但它的工作速度更快.发生了什么？

Answer 1

我将仅列出您的代码的两个主要问题:

1. 没有热身;
2. 一切都发生在main方法中,因此JIT编译的代码只能通过On-Stack Replacement运行.

使用该jmh工具重做您的情况,我得到的时间与预期一样.

@OutputTimeUnit(TimeUnit.MICROSECONDS)
@BenchmarkMode(Mode.AverageTime)
@Warmup(iterations = 3, time = 2)
@Measurement(iterations = 5, time = 3)
@State(Scope.Thread)
@Threads(1)
@Fork(2)
public class Writing
{
  static final int ARRAY_SIZE = 32768;

  int data[] = new int[ARRAY_SIZE];
  volatile int volatileData[] = new int[ARRAY_SIZE];

  @Setup public void setup() {
    Random rnd = new Random(0);
    for (int c = 0; c < ARRAY_SIZE; ++c) {
      data[c] = rnd.nextInt() % 256;
      volatileData[c] = rnd.nextInt() % 256;
    }
    Arrays.sort(data);
    System.arraycopy(data, 0, volatileData, 0, ARRAY_SIZE);
  }

  @GenerateMicroBenchmark
  public long sum() {
    long sum = 0;
    for (int c = 0; c < ARRAY_SIZE; ++c) if (data[c] >= 128) sum += data[c];
    return sum;
  }

  @GenerateMicroBenchmark
  public long volatileSum() {
    long sum = 0;
    for (int c = 0; c < ARRAY_SIZE; ++c) if (volatileData[c] >= 128) sum += volatileData[c];
    return sum;
  }
}

这些是结果:

Benchmark       Mode   Samples         Mean   Mean error    Units
sum             avgt        10       21.956        0.221    us/op
volatileSum     avgt        10       40.561        0.264    us/op