首先为自己没有经过严格测试得出的错误结论感到抱歉，原博文，测试完感觉自己发现了一个新bug，后来思前想后觉得不应该是这样的，如果效率差的这么多，jdk的开发人员会不去优化它吗，但是怎么重复测试任然得到一样的结果，非常疑惑。
我觉得应该是测试方法出问题了，可是怎么也想不到原因，后来了解到jmh，深入研究了一番，觉得jmh的测试值得借鉴，jmh在测试的时候都会先经过预热几遍要测试的代码，示例如下：
public static void main(String[] args) throws RunnerException { Options opt = new OptionsBuilder() .include(JMHSample_02_BenchmarkModes.class.getSimpleName()) .warmupIterations(5) .measurementIterations(5) .forks(1) .build(); new Runner(opt).run(); }
我觉得这是有必要的，而我的测试恰恰缺少了这个步骤，随后我在测试中应用了预热处理（均是3次预热，5次运行取平均值），得出的以下结论。
首先是上个测试中被冤枉的java8的foreach循环，测试代码：
package yiwangzhibujian.jmh.test; import org.openjdk.jmh.annotations.*; import org.openjdk.jmh.runner.Runner; import org.openjdk.jmh.runner.RunnerException; import org.openjdk.jmh.runner.options.Options; import org.openjdk.jmh.runner.options.OptionsBuilder; import java.util.ArrayList; import java.util.HashMap; import java.util.List; import java.util.Map; import java.util.concurrent.TimeUnit; @State(Scope.Thread) @BenchmarkMode(Mode.SingleShotTime) @OutputTimeUnit(TimeUnit.MILLISECONDS) public class J8Loop { public static Map<Integer,List<Dog>> dogMap=new HashMap<>(); static { dogMap.put(10, getDogs(10)); dogMap.put(100, getDogs(100)); dogMap.put(1000, getDogs(1000)); dogMap.put(10000, getDogs(10000)); dogMap.put(100000, getDogs(100000)); dogMap.put(1000000, getDogs(1000000)); } private static List<Dog> getDogs(int num) { List<Dog> dogs=new ArrayList<>(); for(int i=0;i<num;i++){ dogs.add(new Dog(i,"dog"+i)); } return dogs; } private void loop(List<Dog> list) { list.forEach(dog->{ dog.hashCode(); }); } @Benchmark @OperationsPerInvocation(10) public void measureWrong_10() { List<Dog> list = dogMap.get(10); loop(list); } @Benchmark @OperationsPerInvocation(100) public void measureWrong_100() { List<Dog> list = dogMap.get(100); loop(list); } @Benchmark @OperationsPerInvocation(1000) public void measureWrong_1000() { List<Dog> list = dogMap.get(1000); loop(list); } @Benchmark @OperationsPerInvocation(10000) public void measureWrong_10000() { List<Dog> list = dogMap.get(10000); loop(list); } @Benchmark @OperationsPerInvocation(100000) public void measureWrong_100000() { List<Dog> list = dogMap.get(100000); loop(list); } @Benchmark @OperationsPerInvocation(1000000) public void measureWrong_1000000() { List<Dog> list = dogMap.get(1000000); loop(list); } /* * ============================== HOW TO RUN THIS TEST: * ==================================== * * You might notice the larger the repetitions count, the lower the * "perceived" cost of the operation being measured. Up to the point we do * each addition with 1/20 ns, well beyond what hardware can actually do. * * This happens because the loop is heavily unrolled/pipelined, and the * operation to be measured is hoisted from the loop. Morale: don't overuse * loops, rely on JMH to get the measurement right. * * You can run this test: * * a) Via the command line: $ mvn clean install $ java -jar * target/benchmarks.jar JMHSample_11 -wi 5 -i 5 -f 1 (we requested 5 * warmup/measurement iterations, single fork) * * b) Via the Java API: (see the JMH homepage for possible caveats when * running from IDE: http://openjdk.java.net/projects/code-tools/jmh/) */ public static void main(String[] args) throws RunnerException { Options opt = new OptionsBuilder() .include(J8Loop.class.getSimpleName()) .warmupIterations(2) .measurementIterations(2).forks(1).build(); new Runner(opt).run(); } } class Dog{ private int age; private String name; public Dog(int age, String name) { super(); this.age = age; this.name = name; } public int getAge() { return age; } public void setAge(int age) { this.age = age; } public String getName() { return name; } public void setName(String name) { this.name = name; } @Override public String toString() { return "Dog [age=" + age + ", name=" + name + "]"; } }
测试结果：
# JMH version: 1.19 # VM version: JDK 1.8.0_51, VM 25.51-b03 # VM invoker: D:softjavajava8jreinjava.exe # VM options: -Dfile.encoding=UTF-8 # Warmup: 3 iterations, single-shot each # Measurement: 5 iterations, single-shot each # Timeout: 10 min per iteration # Threads: 1 thread # Benchmark mode: Single shot invocation time # Benchmark: yiwangzhibujian.jmh.test.J8Loop.measureWrong_10 # Run progress: 0.00% complete, ETA 00:00:00 # Fork: 1 of 1 # Warmup Iteration 1: 85.205 ms/op # Warmup Iteration 2: 0.013 ms/op # Warmup Iteration 3: 0.012 ms/op Iteration 1: 0.010 ms/op Iteration 2: 0.013 ms/op Iteration 3: 0.009 ms/op Iteration 4: 0.010 ms/op Iteration 5: 0.011 ms/op Result "yiwangzhibujian.jmh.test.J8Loop.measureWrong_10": N = 5 mean = 0.010 ±(99.9%) 0.006 ms/op Histogram, ms/op: [0.008, 0.009) = 0 [0.009, 0.009) = 1 [0.009, 0.010) = 0 [0.010, 0.010) = 2 [0.010, 0.011) = 0 [0.011, 0.011) = 1 [0.011, 0.012) = 0 [0.012, 0.012) = 0 [0.012, 0.013) = 0 [0.013, 0.013) = 0 [0.013, 0.014) = 1 Percentiles, ms/op: p(0.0000) = 0.009 ms/op p(50.0000) = 0.010 ms/op p(90.0000) = 0.013 ms/op p(95.0000) = 0.013 ms/op p(99.0000) = 0.013 ms/op p(99.9000) = 0.013 ms/op p(99.9900) = 0.013 ms/op p(99.9990) = 0.013 ms/op p(99.9999) = 0.013 ms/op p(100.0000) = 0.013 ms/op # JMH version: 1.19 # VM version: JDK 1.8.0_51, VM 25.51-b03 # VM invoker: D:softjavajava8jreinjava.exe # VM options: -Dfile.encoding=UTF-8 # Warmup: 3 iterations, single-shot each # Measurement: 5 iterations, single-shot each # Timeout: 10 min per iteration # Threads: 1 thread # Benchmark mode: Single shot invocation time # Benchmark: yiwangzhibujian.jmh.test.J8Loop.measureWrong_100 # Run progress: 16.67% complete, ETA 00:00:10 # Fork: 1 of 1 # Warmup Iteration 1: 81.395 ms/op # Warmup Iteration 2: 0.046 ms/op # Warmup Iteration 3: 0.070 ms/op Iteration 1: 0.040 ms/op Iteration 2: 0.065 ms/op Iteration 3: 0.023 ms/op Iteration 4: 0.025 ms/op Iteration 5: 0.024 ms/op Result "yiwangzhibujian.jmh.test.J8Loop.measureWrong_100": N = 5 mean = 0.035 ±(99.9%) 0.069 ms/op Histogram, ms/op: [0.020, 0.025) = 3 [0.025, 0.030) = 0 [0.030, 0.035) = 0 [0.035, 0.040) = 1 [0.040, 0.045) = 0 [0.045, 0.050) = 0 [0.050, 0.055) = 0 [0.055, 0.060) = 0 [0.060, 0.065) = 1 Percentiles, ms/op: p(0.0000) = 0.023 ms/op p(50.0000) = 0.025 ms/op p(90.0000) = 0.065 ms/op p(95.0000) = 0.065 ms/op p(99.0000) = 0.065 ms/op p(99.9000) = 0.065 ms/op p(99.9900) = 0.065 ms/op p(99.9990) = 0.065 ms/op p(99.9999) = 0.065 ms/op p(100.0000) = 0.065 ms/op # JMH version: 1.19 # VM version: JDK 1.8.0_51, VM 25.51-b03 # VM invoker: D:softjavajava8jreinjava.exe # VM options: -Dfile.encoding=UTF-8 # Warmup: 3 iterations, single-shot each # Measurement: 5 iterations, single-shot each # Timeout: 10 min per iteration # Threads: 1 thread # Benchmark mode: Single shot invocation time # Benchmark: yiwangzhibujian.jmh.test.J8Loop.measureWrong_1000 # Run progress: 33.33% complete, ETA 00:00:08 # Fork: 1 of 1 # Warmup Iteration 1: 85.176 ms/op # Warmup Iteration 2: 0.201 ms/op # Warmup Iteration 3: 0.183 ms/op Iteration 1: 0.195 ms/op Iteration 2: 0.184 ms/op Iteration 3: 0.566 ms/op Iteration 4: 0.101 ms/op Iteration 5: 0.091 ms/op Result "yiwangzhibujian.jmh.test.J8Loop.measureWrong_1000": N = 5 mean = 0.227 ±(99.9%) 0.751 ms/op Histogram, ms/op: [0.000, 0.050) = 0 [0.050, 0.100) = 1 [0.100, 0.150) = 1 [0.150, 0.200) = 2 [0.200, 0.250) = 0 [0.250, 0.300) = 0 [0.300, 0.350) = 0 [0.350, 0.400) = 0 [0.400, 0.450) = 0 [0.450, 0.500) = 0 [0.500, 0.550) = 0 [0.550, 0.600) = 1 Percentiles, ms/op: p(0.0000) = 0.091 ms/op p(50.0000) = 0.184 ms/op p(90.0000) = 0.566 ms/op p(95.0000) = 0.566 ms/op p(99.0000) = 0.566 ms/op p(99.9000) = 0.566 ms/op p(99.9900) = 0.566 ms/op p(99.9990) = 0.566 ms/op p(99.9999) = 0.566 ms/op p(100.0000) = 0.566 ms/op # JMH version: 1.19 # VM version: JDK 1.8.0_51, VM 25.51-b03 # VM invoker: D:softjavajava8jreinjava.exe # VM options: -Dfile.encoding=UTF-8 # Warmup: 3 iterations, single-shot each # Measurement: 5 iterations, single-shot each # Timeout: 10 min per iteration # Threads: 1 thread # Benchmark mode: Single shot invocation time # Benchmark: yiwangzhibujian.jmh.test.J8Loop.measureWrong_10000 # Run progress: 50.00% complete, ETA 00:00:06 # Fork: 1 of 1 # Warmup Iteration 1: 85.345 ms/op # Warmup Iteration 2: 0.698 ms/op # Warmup Iteration 3: 0.857 ms/op Iteration 1: 0.866 ms/op Iteration 2: 0.778 ms/op Iteration 3: 0.871 ms/op Iteration 4: 1.047 ms/op Iteration 5: 0.306 ms/op Result "yiwangzhibujian.jmh.test.J8Loop.measureWrong_10000": N = 5 mean = 0.774 ±(99.9%) 1.074 ms/op Histogram, ms/op: [0.300, 0.350) = 1 [0.350, 0.400) = 0 [0.400, 0.450) = 0 [0.450, 0.500) = 0 [0.500, 0.550) = 0 [0.550, 0.600) = 0 [0.600, 0.650) = 0 [0.650, 0.700) = 0 [0.700, 0.750) = 0 [0.750, 0.800) = 1 [0.800, 0.850) = 0 [0.850, 0.900) = 2 [0.900, 0.950) = 0 [0.950, 1.000) = 0 [1.000, 1.050) = 1 Percentiles, ms/op: p(0.0000) = 0.306 ms/op p(50.0000) = 0.866 ms/op p(90.0000) = 1.047 ms/op p(95.0000) = 1.047 ms/op p(99.0000) = 1.047 ms/op p(99.9000) = 1.047 ms/op p(99.9900) = 1.047 ms/op p(99.9990) = 1.047 ms/op p(99.9999) = 1.047 ms/op p(100.0000) = 1.047 ms/op # JMH version: 1.19 # VM version: JDK 1.8.0_51, VM 25.51-b03 # VM invoker: D:softjavajava8jreinjava.exe # VM options: -Dfile.encoding=UTF-8 # Warmup: 3 iterations, single-shot each # Measurement: 5 iterations, single-shot each # Timeout: 10 min per iteration # Threads: 1 thread # Benchmark mode: Single shot invocation time # Benchmark: yiwangzhibujian.jmh.test.J8Loop.measureWrong_100000 # Run progress: 66.67% complete, ETA 00:00:04 # Fork: 1 of 1 # Warmup Iteration 1: 96.430 ms/op # Warmup Iteration 2: 6.303 ms/op # Warmup Iteration 3: 1.515 ms/op Iteration 1: 1.337 ms/op Iteration 2: 1.260 ms/op Iteration 3: 1.279 ms/op Iteration 4: 1.210 ms/op Iteration 5: 1.107 ms/op Result "yiwangzhibujian.jmh.test.J8Loop.measureWrong_100000": N = 5 mean = 1.238 ±(99.9%) 0.334 ms/op Histogram, ms/op: [1.100, 1.125) = 1 [1.125, 1.150) = 0 [1.150, 1.175) = 0 [1.175, 1.200) = 0 [1.200, 1.225) = 1 [1.225, 1.250) = 0 [1.250, 1.275) = 1 [1.275, 1.300) = 1 [1.300, 1.325) = 0 [1.325, 1.350) = 1 [1.350, 1.375) = 0 [1.375, 1.400) = 0 Percentiles, ms/op: p(0.0000) = 1.107 ms/op p(50.0000) = 1.260 ms/op p(90.0000) = 1.337 ms/op p(95.0000) = 1.337 ms/op p(99.0000) = 1.337 ms/op p(99.9000) = 1.337 ms/op p(99.9900) = 1.337 ms/op p(99.9990) = 1.337 ms/op p(99.9999) = 1.337 ms/op p(100.0000) = 1.337 ms/op # JMH version: 1.19 # VM version: JDK 1.8.0_51, VM 25.51-b03 # VM invoker: D:softjavajava8jreinjava.exe # VM options: -Dfile.encoding=UTF-8 # Warmup: 3 iterations, single-shot each # Measurement: 5 iterations, single-shot each # Timeout: 10 min per iteration # Threads: 1 thread # Benchmark mode: Single shot invocation time # Benchmark: yiwangzhibujian.jmh.test.J8Loop.measureWrong_1000000 # Run progress: 83.33% complete, ETA 00:00:02 # Fork: 1 of 1 # Warmup Iteration 1: 155.436 ms/op # Warmup Iteration 2: 16.059 ms/op # Warmup Iteration 3: 10.600 ms/op Iteration 1: 10.347 ms/op Iteration 2: 10.207 ms/op Iteration 3: 11.790 ms/op Iteration 4: 10.717 ms/op Iteration 5: 10.128 ms/op Result "yiwangzhibujian.jmh.test.J8Loop.measureWrong_1000000": N = 5 mean = 10.638 ±(99.9%) 2.628 ms/op Histogram, ms/op: [10.000, 10.125) = 0 [10.125, 10.250) = 2 [10.250, 10.375) = 1 [10.375, 10.500) = 0 [10.500, 10.625) = 0 [10.625, 10.750) = 1 [10.750, 10.875) = 0 [10.875, 11.000) = 0 [11.000, 11.125) = 0 [11.125, 11.250) = 0 [11.250, 11.375) = 0 [11.375, 11.500) = 0 [11.500, 11.625) = 0 [11.625, 11.750) = 0 [11.750, 11.875) = 1 Percentiles, ms/op: p(0.0000) = 10.128 ms/op p(50.0000) = 10.347 ms/op p(90.0000) = 11.790 ms/op p(95.0000) = 11.790 ms/op p(99.0000) = 11.790 ms/op p(99.9000) = 11.790 ms/op p(99.9900) = 11.790 ms/op p(99.9990) = 11.790 ms/op p(99.9999) = 11.790 ms/op p(100.0000) = 11.790 ms/op # Run complete. Total time: 00:00:13 Benchmark Mode Cnt Score Error Units J8Loop.measureWrong_10 ss 5 0.010 ± 0.006 ms/op J8Loop.measureWrong_100 ss 5 0.035 ± 0.069 ms/op J8Loop.measureWrong_1000 ss 5 0.227 ± 0.751 ms/op J8Loop.measureWrong_10000 ss 5 0.774 ± 1.074 ms/op J8Loop.measureWrong_100000 ss 5 1.238 ± 0.334 ms/op J8Loop.measureWrong_1000000 ss 5 10.638 ± 2.628 ms/op
其中最终的结论是：
Benchmark Mode Cnt Score Error Units J8Loop.measureWrong_10 ss 5 0.010 ± 0.006 ms/op J8Loop.measureWrong_100 ss 5 0.035 ± 0.069 ms/op J8Loop.measureWrong_1000 ss 5 0.227 ± 0.751 ms/op J8Loop.measureWrong_10000 ss 5 0.774 ± 1.074 ms/op J8Loop.measureWrong_100000 ss 5 1.238 ± 0.334 ms/op J8Loop.measureWrong_1000000 ss 5 10.638 ± 2.628 ms/op
然后是表现最好的增强for循环，测试代码：
package yiwangzhibujian.jmh.test; import org.openjdk.jmh.annotations.*; import org.openjdk.jmh.runner.Runner; import org.openjdk.jmh.runner.RunnerException; import org.openjdk.jmh.runner.options.Options; import org.openjdk.jmh.runner.options.OptionsBuilder; import java.util.ArrayList; import java.util.HashMap; import java.util.List; import java.util.Map; import java.util.concurrent.TimeUnit; @State(Scope.Thread) @BenchmarkMode(Mode.SingleShotTime) @OutputTimeUnit(TimeUnit.MILLISECONDS) public class J8Loop { public static Map<Integer,List<Dog>> dogMap=new HashMap<>(); static { dogMap.put(10, getDogs(10)); dogMap.put(100, getDogs(100)); dogMap.put(1000, getDogs(1000)); dogMap.put(10000, getDogs(10000)); dogMap.put(100000, getDogs(100000)); dogMap.put(1000000, getDogs(1000000)); } private static List<Dog> getDogs(int num) { List<Dog> dogs=new ArrayList<>(); for(int i=0;i<num;i++){ dogs.add(new Dog(i,"dog"+i)); } return dogs; } private void loop(List<Dog> list) { for(Dog dog:list){ dog.hashCode(); } } @Benchmark @OperationsPerInvocation(10) public void measureWrong_10() { List<Dog> list = dogMap.get(10); loop(list); } @Benchmark @OperationsPerInvocation(100) public void measureWrong_100() { List<Dog> list = dogMap.get(100); loop(list); } @Benchmark @OperationsPerInvocation(1000) public void measureWrong_1000() { List<Dog> list = dogMap.get(1000); loop(list); } @Benchmark @OperationsPerInvocation(10000) public void measureWrong_10000() { List<Dog> list = dogMap.get(10000); loop(list); } @Benchmark @OperationsPerInvocation(100000) public void measureWrong_100000() { List<Dog> list = dogMap.get(100000); loop(list); } @Benchmark @OperationsPerInvocation(1000000) public void measureWrong_1000000() { List<Dog> list = dogMap.get(1000000); loop(list); } /* * ============================== HOW TO RUN THIS TEST: * ==================================== * * You might notice the larger the repetitions count, the lower the * "perceived" cost of the operation being measured. Up to the point we do * each addition with 1/20 ns, well beyond what hardware can actually do. * * This happens because the loop is heavily unrolled/pipelined, and the * operation to be measured is hoisted from the loop. Morale: don't overuse * loops, rely on JMH to get the measurement right. * * You can run this test: * * a) Via the command line: $ mvn clean install $ java -jar * target/benchmarks.jar JMHSample_11 -wi 5 -i 5 -f 1 (we requested 5 * warmup/measurement iterations, single fork) * * b) Via the Java API: (see the JMH homepage for possible caveats when * running from IDE: http://openjdk.java.net/projects/code-tools/jmh/) */ public static void main(String[] args) throws RunnerException { Options opt = new OptionsBuilder() .include(J8Loop.class.getSimpleName()) .warmupIterations(3) .measurementIterations(5).forks(1).build(); new Runner(opt).run(); } } class Dog{ private int age; private String name; public Dog(int age, String name) { super(); this.age = age; this.name = name; } public int getAge() { return age; } public void setAge(int age) { this.age = age; } public String getName() { return name; } public void setName(String name) { this.name = name; } @Override public String toString() { return "Dog [age=" + age + ", name=" + name + "]"; } }
测试结果：
# JMH version: 1.19 # VM version: JDK 1.8.0_51, VM 25.51-b03 # VM invoker: D:softjavajava8jreinjava.exe # VM options: -Dfile.encoding=UTF-8 # Warmup: 3 iterations, single-shot each # Measurement: 5 iterations, single-shot each # Timeout: 10 min per iteration # Threads: 1 thread # Benchmark mode: Single shot invocation time # Benchmark: yiwangzhibujian.jmh.test.J8Loop.measureWrong_10 # Run progress: 0.00% complete, ETA 00:00:00 # Fork: 1 of 1 # Warmup Iteration 1: 0.047 ms/op # Warmup Iteration 2: 0.010 ms/op # Warmup Iteration 3: 0.009 ms/op Iteration 1: 0.009 ms/op Iteration 2: 0.011 ms/op Iteration 3: 0.009 ms/op Iteration 4: 0.009 ms/op Iteration 5: 0.010 ms/op Result "yiwangzhibujian.jmh.test.J8Loop.measureWrong_10": N = 5 mean = 0.010 ±(99.9%) 0.003 ms/op Histogram, ms/op: [0.008, 0.008) = 0 [0.008, 0.009) = 0 [0.009, 0.009) = 0 [0.009, 0.009) = 2 [0.009, 0.009) = 0 [0.009, 0.010) = 1 [0.010, 0.010) = 0 [0.010, 0.010) = 1 [0.010, 0.010) = 0 [0.010, 0.011) = 0 [0.011, 0.011) = 1 Percentiles, ms/op: p(0.0000) = 0.009 ms/op p(50.0000) = 0.009 ms/op p(90.0000) = 0.011 ms/op p(95.0000) = 0.011 ms/op p(99.0000) = 0.011 ms/op p(99.9000) = 0.011 ms/op p(99.9900) = 0.011 ms/op p(99.9990) = 0.011 ms/op p(99.9999) = 0.011 ms/op p(100.0000) = 0.011 ms/op # JMH version: 1.19 # VM version: JDK 1.8.0_51, VM 25.51-b03 # VM invoker: D:softjavajava8jreinjava.exe # VM options: -Dfile.encoding=UTF-8 # Warmup: 3 iterations, single-shot each # Measurement: 5 iterations, single-shot each # Timeout: 10 min per iteration # Threads: 1 thread # Benchmark mode: Single shot invocation time # Benchmark: yiwangzhibujian.jmh.test.J8Loop.measureWrong_100 # Run progress: 16.67% complete, ETA 00:00:10 # Fork: 1 of 1 # Warmup Iteration 1: 0.068 ms/op # Warmup Iteration 2: 0.038 ms/op # Warmup Iteration 3: 0.038 ms/op Iteration 1: 0.037 ms/op Iteration 2: 0.025 ms/op Iteration 3: 0.125 ms/op Iteration 4: 0.038 ms/op Iteration 5: 0.038 ms/op Result "yiwangzhibujian.jmh.test.J8Loop.measureWrong_100": N = 5 mean = 0.053 ±(99.9%) 0.158 ms/op Histogram, ms/op: [0.000, 0.013) = 0 [0.013, 0.025) = 0 [0.025, 0.038) = 2 [0.038, 0.050) = 2 [0.050, 0.063) = 0 [0.063, 0.075) = 0 [0.075, 0.088) = 0 [0.088, 0.100) = 0 [0.100, 0.113) = 0 [0.113, 0.125) = 0 [0.125, 0.138) = 1 [0.138, 0.150) = 0 [0.150, 0.163) = 0 [0.163, 0.175) = 0 [0.175, 0.188) = 0 Percentiles, ms/op: p(0.0000) = 0.025 ms/op p(50.0000) = 0.038 ms/op p(90.0000) = 0.125 ms/op p(95.0000) = 0.125 ms/op p(99.0000) = 0.125 ms/op p(99.9000) = 0.125 ms/op p(99.9900) = 0.125 ms/op p(99.9990) = 0.125 ms/op p(99.9999) = 0.125 ms/op p(100.0000) = 0.125 ms/op # JMH version: 1.19 # VM version: JDK 1.8.0_51, VM 25.51-b03 # VM invoker: D:softjavajava8jreinjava.exe # VM options: -Dfile.encoding=UTF-8 # Warmup: 3 iterations, single-shot each # Measurement: 5 iterations, single-shot each # Timeout: 10 min per iteration # Threads: 1 thread # Benchmark mode: Single shot invocation time # Benchmark: yiwangzhibujian.jmh.test.J8Loop.measureWrong_1000 # Run progress: 33.33% complete, ETA 00:00:08 # Fork: 1 of 1 # Warmup Iteration 1: 0.266 ms/op # Warmup Iteration 2: 0.310 ms/op # Warmup Iteration 3: 0.635 ms/op Iteration 1: 0.297 ms/op Iteration 2: 0.255 ms/op Iteration 3: 0.425 ms/op Iteration 4: 0.413 ms/op Iteration 5: 0.317 ms/op Result "yiwangzhibujian.jmh.test.J8Loop.measureWrong_1000": N = 5 mean = 0.341 ±(99.9%) 0.287 ms/op Histogram, ms/op: [0.200, 0.225) = 0 [0.225, 0.250) = 0 [0.250, 0.275) = 1 [0.275, 0.300) = 1 [0.300, 0.325) = 1 [0.325, 0.350) = 0 [0.350, 0.375) = 0 [0.375, 0.400) = 0 [0.400, 0.425) = 2 [0.425, 0.450) = 0 [0.450, 0.475) = 0 Percentiles, ms/op: p(0.0000) = 0.255 ms/op p(50.0000) = 0.317 ms/op p(90.0000) = 0.425 ms/op p(95.0000) = 0.425 ms/op p(99.0000) = 0.425 ms/op p(99.9000) = 0.425 ms/op p(99.9900) = 0.425 ms/op p(99.9990) = 0.425 ms/op p(99.9999) = 0.425 ms/op p(100.0000) = 0.425 ms/op # JMH version: 1.19 # VM version: JDK 1.8.0_51, VM 25.51-b03 # VM invoker: D:softjavajava8jreinjava.exe # VM options: -Dfile.encoding=UTF-8 # Warmup: 3 iterations, single-shot each # Measurement: 5 iterations, single-shot each # Timeout: 10 min per iteration # Threads: 1 thread # Benchmark mode: Single shot invocation time # Benchmark: yiwangzhibujian.jmh.test.J8Loop.measureWrong_10000 # Run progress: 50.00% complete, ETA 00:00:06 # Fork: 1 of 1 # Warmup Iteration 1: 2.482 ms/op # Warmup Iteration 2: 2.617 ms/op # Warmup Iteration 3: 2.814 ms/op Iteration 1: 2.656 ms/op Iteration 2: 2.848 ms/op Iteration 3: 2.636 ms/op Iteration 4: 1.800 ms/op Iteration 5: 0.927 ms/op Result "yiwangzhibujian.jmh.test.J8Loop.measureWrong_10000": N = 5 mean = 2.173 ±(99.9%) 3.101 ms/op Histogram, ms/op: [0.000, 0.250) = 0 [0.250, 0.500) = 0 [0.500, 0.750) = 0 [0.750, 1.000) = 1 [1.000, 1.250) = 0 [1.250, 1.500) = 0 [1.500, 1.750) = 0 [1.750, 2.000) = 1 [2.000, 2.250) = 0 [2.250, 2.500) = 0 [2.500, 2.750) = 2 Percentiles, ms/op: p(0.0000) = 0.927 ms/op p(50.0000) = 2.636 ms/op p(90.0000) = 2.848 ms/op p(95.0000) = 2.848 ms/op p(99.0000) = 2.848 ms/op p(99.9000) = 2.848 ms/op p(99.9900) = 2.848 ms/op p(99.9990) = 2.848 ms/op p(99.9999) = 2.848 ms/op p(100.0000) = 2.848 ms/op # JMH version: 1.19 # VM version: JDK 1.8.0_51, VM 25.51-b03 # VM invoker: D:softjavajava8jreinjava.exe # VM options: -Dfile.encoding=UTF-8 # Warmup: 3 iterations, single-shot each # Measurement: 5 iterations, single-shot each # Timeout: 10 min per iteration # Threads: 1 thread # Benchmark mode: Single shot invocation time # Benchmark: yiwangzhibujian.jmh.test.J8Loop.measureWrong_100000 # Run progress: 66.67% complete, ETA 00:00:04 # Fork: 1 of 1 # Warmup Iteration 1: 17.857 ms/op # Warmup Iteration 2: 7.360 ms/op # Warmup Iteration 3: 2.186 ms/op Iteration 1: 1.542 ms/op Iteration 2: 1.419 ms/op Iteration 3: 2.023 ms/op Iteration 4: 1.382 ms/op Iteration 5: 1.448 ms/op Result "yiwangzhibujian.jmh.test.J8Loop.measureWrong_100000": N = 5 mean = 1.563 ±(99.9%) 1.017 ms/op Histogram, ms/op: [1.300, 1.350) = 0 [1.350, 1.400) = 1 [1.400, 1.450) = 2 [1.450, 1.500) = 0 [1.500, 1.550) = 1 [1.550, 1.600) = 0 [1.600, 1.650) = 0 [1.650, 1.700) = 0 [1.700, 1.750) = 0 [1.750, 1.800) = 0 [1.800, 1.850) = 0 [1.850, 1.900) = 0 [1.900, 1.950) = 0 [1.950, 2.000) = 0 [2.000, 2.050) = 1 Percentiles, ms/op: p(0.0000) = 1.382 ms/op p(50.0000) = 1.448 ms/op p(90.0000) = 2.023 ms/op p(95.0000) = 2.023 ms/op p(99.0000) = 2.023 ms/op p(99.9000) = 2.023 ms/op p(99.9900) = 2.023 ms/op p(99.9990) = 2.023 ms/op p(99.9999) = 2.023 ms/op p(100.0000) = 2.023 ms/op # JMH version: 1.19 # VM version: JDK 1.8.0_51, VM 25.51-b03 # VM invoker: D:softjavajava8jreinjava.exe # VM options: -Dfile.encoding=UTF-8 # Warmup: 3 iterations, single-shot each # Measurement: 5 iterations, single-shot each # Timeout: 10 min per iteration # Threads: 1 thread # Benchmark mode: Single shot invocation time # Benchmark: yiwangzhibujian.jmh.test.J8Loop.measureWrong_1000000 # Run progress: 83.33% complete, ETA 00:00:02 # Fork: 1 of 1 # Warmup Iteration 1: 78.570 ms/op # Warmup Iteration 2: 21.032 ms/op # Warmup Iteration 3: 10.005 ms/op Iteration 1: 12.503 ms/op Iteration 2: 10.849 ms/op Iteration 3: 12.988 ms/op Iteration 4: 12.030 ms/op Iteration 5: 11.531 ms/op Result "yiwangzhibujian.jmh.test.J8Loop.measureWrong_1000000": N = 5 mean = 11.980 ±(99.9%) 3.206 ms/op Histogram, ms/op: [10.000, 10.250) = 0 [10.250, 10.500) = 0 [10.500, 10.750) = 0 [10.750, 11.000) = 1 [11.000, 11.250) = 0 [11.250, 11.500) = 0 [11.500, 11.750) = 1 [11.750, 12.000) = 0 [12.000, 12.250) = 1 [12.250, 12.500) = 0 [12.500, 12.750) = 1 Percentiles, ms/op: p(0.0000) = 10.849 ms/op p(50.0000) = 12.030 ms/op p(90.0000) = 12.988 ms/op p(95.0000) = 12.988 ms/op p(99.0000) = 12.988 ms/op p(99.9000) = 12.988 ms/op p(99.9900) = 12.988 ms/op p(99.9990) = 12.988 ms/op p(99.9999) = 12.988 ms/op p(100.0000) = 12.988 ms/op # Run complete. Total time: 00:00:12 Benchmark Mode Cnt Score Error Units J8Loop.measureWrong_10 ss 5 0.010 ± 0.003 ms/op J8Loop.measureWrong_100 ss 5 0.053 ± 0.158 ms/op J8Loop.measureWrong_1000 ss 5 0.341 ± 0.287 ms/op J8Loop.measureWrong_10000 ss 5 2.173 ± 3.101 ms/op J8Loop.measureWrong_100000 ss 5 1.563 ± 1.017 ms/op J8Loop.measureWrong_1000000 ss 5 11.980 ± 3.206 ms/op
其中最终的结论是：
Benchmark Mode Cnt Score Error Units J8Loop.measureWrong_10 ss 5 0.010 ± 0.003 ms/op J8Loop.measureWrong_100 ss 5 0.053 ± 0.158 ms/op J8Loop.measureWrong_1000 ss 5 0.341 ± 0.287 ms/op J8Loop.measureWrong_10000 ss 5 2.173 ± 3.101 ms/op J8Loop.measureWrong_100000 ss 5 1.563 ± 1.017 ms/op J8Loop.measureWrong_1000000 ss 5 11.980 ± 3.206 ms/op
如果用折线图对比的话：
会发现java8的foreach循环效率会更高一筹，真是冤枉了它。
其他几种循环就没必要测试了，用这两种我觉得就差不多了，如果你写代码思维不够清晰，需要经常debug，建议使用增强for循环，方便调试。
然后说一下有可能让人困惑的预热，我自己也做过测试，这个java8的foreach预热是jvm级别的，只要预热一次，以后速度都是这样的，至于深层次的原因我后续会做研究的。