Ósc*_*pez 153 java string iteration performance char
在Java中,迭代字符串中所有字符的最快方法是什么,这:
String str = "a really, really long string";
for (int i = 0, n = str.length(); i < n; i++) {
char c = str.charAt(i);
}
或这个:
char[] chars = str.toCharArray();
for (int i = 0, n = chars.length; i < n; i++) {
char c = chars[i];
}
编辑:
如果重复调用的成本是什么,我想知道的是charAt长迭代期间方法最终是要么不是比执行一个调用的成本小于或大于toCharArray开头,然后直接在迭代期间访问阵列.
如果有人能够为不同的字符串长度提供强大的基准测试,考虑到JIT预热时间,JVM启动时间等等,而不仅仅是两次调用之间的差异,那就太棒了System.currentTimeMillis().
The*_*tor 342
FIRST UPDATE: Before you try this ever in a production environment (not advised), read this first: http://www.javaspecialists.eu/archive/Issue237.html Starting from Java 9, the solution as described won't work anymore, because now Java will store strings as byte[] by default.
SECOND UPDATE: As of 2016-10-25, on my AMDx64 8core and source 1.8, there is no difference between using 'charAt' and field access. It appears that the jvm is sufficiently optimized to inline and streamline any 'string.charAt(n)' calls.
It all depends on the length of the String being inspected. If, as the question says, it is for long strings, the fastest way to inspect the string is to use reflection to access the backing char[] of the string.
A fully randomized benchmark with JDK 8 (win32 and win64) on an 64 AMD Phenom II 4 core 955 @ 3.2 GHZ (in both client mode and server mode) with 9 different techniques (see below!) shows that using String.charAt(n) is the fastest for small strings and that using reflection to access the String backing array is almost twice as fast for large strings.
9 different optimization techniques are tried.
All string contents are randomized
The test are done for string sizes in multiples of two starting with 0,1,2,4,8,16 etc.
The tests are done 1,000 times per string size
The tests are shuffled into random order each time. In other words, the tests are done in random order every time they are done, over 1000 times over.
The entire test suite is done forwards, and backwards, to show the effect of JVM warmup on optimization and times.
整个套件完成两次,一次处于-client模式,另一种处于-server模式.
对于长度为1到256个字符的字符串,调用string.charAt(i)wins的平均处理速度为每秒1340万到5.88亿个字符.
此外,它总体上提高了5.5%(客户端)和13.9%(服务器),如下所示:
for (int i = 0; i < data.length(); i++) {
if (data.charAt(i) <= ' ') {
doThrow();
}
}
比这个本地最终长度变量:
final int len = data.length();
for (int i = 0; i < len; i++) {
if (data.charAt(i) <= ' ') {
doThrow();
}
}
对于长字符串,长度为512到256K,使用反射访问String的后备阵列是最快的.这种技术几乎是String.charAt(i)的两倍(快 178%).该范围内的平均速度为每秒11.11亿个字符.
必须提前获取Field,然后可以在库中的不同字符串上重复使用.有趣的是,与上面的代码不同,使用Field访问,拥有本地最终长度变量比在循环检查中使用'chars.length'快9%.以下是Field访问可以设置为最快的方式:
final Field field = String.class.getDeclaredField("value");
field.setAccessible(true);
try {
final char[] chars = (char[]) field.get(data);
final int len = chars.length;
for (int i = 0; i < len; i++) {
if (chars[i] <= ' ') {
doThrow();
}
}
return len;
} catch (Exception ex) {
throw new RuntimeException(ex);
}
在我的AMD 64机器上的64位Java机器上,在服务器模式下,字段访问开始在32个字符长度字符串之后获胜.在客户端模式下,直到512个字符长度才能看到.
另外值得注意的是,当我在服务器模式下运行JDK 8(32位版本)时,大字符串和小字符串的整体性能降低了7%.这是2013年12月121日JDK 8早期版本的构建.所以,就目前而言,似乎32位服务器模式比32位客户端模式慢.
话虽如此......似乎唯一值得调用的服务器模式是在64位机器上.否则它实际上会妨碍性能.
对于在-server modeAMD64上运行的32位构建,我可以这样说:
另外值得一提的是,String.chars()(Stream和并行版)是一个半身像.比任何其他方式慢.该StreamsAPI是执行一般的字符串操作一个相当缓慢的方式.
Java String could have predicate accepting optimized methods such as contains(predicate), forEach(consumer), forEachWithIndex(consumer). Thus, without the need for the user to know the length or repeat calls to String methods, these could help parsing libraries beep-beep beep speedup.
Keep dreaming :)
Happy Strings!
~SH
"charAt1" -- CHECK THE STRING CONTENTS THE USUAL WAY:
int charAtMethod1(final String data) {
final int len = data.length();
for (int i = 0; i < len; i++) {
if (data.charAt(i) <= ' ') {
doThrow();
}
}
return len;
}
"charAt2" -- SAME AS ABOVE BUT USE String.length() INSTEAD OF MAKING A FINAL LOCAL int FOR THE LENGTh
int charAtMethod2(final String data) {
for (int i = 0; i < data.length(); i++) {
if (data.charAt(i) <= ' ') {
doThrow();
}
}
return data.length();
}
"stream" -- USE THE NEW JAVA-8 String's IntStream AND PASS IT A PREDICATE TO DO THE CHECKING
int streamMethod(final String data, final IntPredicate predicate) {
if (data.chars().anyMatch(predicate)) {
doThrow();
}
return data.length();
}
"streamPara" -- SAME AS ABOVE, BUT OH-LA-LA - GO PARALLEL!!!
// avoid this at all costs
int streamParallelMethod(final String data, IntPredicate predicate) {
if (data.chars().parallel().anyMatch(predicate)) {
doThrow();
}
return data.length();
}
"reuse" -- REFILL A REUSABLE char[] WITH THE STRINGS CONTENTS
int reuseBuffMethod(final char[] reusable, final String data) {
final int len = data.length();
data.getChars(0, len, reusable, 0);
for (int i = 0; i < len; i++) {
if (reusable[i] <= ' ') {
doThrow();
}
}
return len;
}
"new1" -- OBTAIN A NEW COPY OF THE char[] FROM THE STRING
int newMethod1(final String data) {
final int len = data.length();
final char[] copy = data.toCharArray();
for (int i = 0; i < len; i++) {
if (copy[i] <= ' ') {
doThrow();
}
}
return len;
}
"new2" -- SAME AS ABOVE, BUT USE "FOR-EACH"
int newMethod2(final String data) {
for (final char c : data.toCharArray()) {
if (c <= ' ') {
doThrow();
}
}
return data.length();
}
"field1" -- FANCY!! OBTAIN FIELD FOR ACCESS TO THE STRING'S INTERNAL char[]
int fieldMethod1(final Field field, final String data) {
try {
final char[] chars = (char[]) field.get(data);
final int len = chars.length;
for (int i = 0; i < len; i++) {
if (chars[i] <= ' ') {
doThrow();
}
}
return len;
} catch (Exception ex) {
throw new RuntimeException(ex);
}
}
"field2" -- SAME AS ABOVE, BUT USE "FOR-EACH"
int fieldMethod2(final Field field, final String data) {
final char[] chars;
try {
chars = (char[]) field.get(data);
} catch (Exception ex) {
throw new RuntimeException(ex);
}
for (final char c : chars) {
if (c <= ' ') {
doThrow();
}
}
return chars.length;
}
-client MODE (forwards and backwards tests combined)Note: that the -client mode with Java 32 bit and -server mode with Java 64 bit are the same as below on my AMD64 machine.
Size WINNER charAt1 charAt2 stream streamPar reuse new1 new2 field1 field2
1 charAt 77.0 72.0 462.0 584.0 127.5 89.5 86.0 159.5 165.0
2 charAt 38.0 36.5 284.0 32712.5 57.5 48.3 50.3 89.0 91.5
4 charAt 19.5 18.5 458.6 3169.0 33.0 26.8 27.5 54.1 52.6
8 charAt 9.8 9.9 100.5 1370.9 17.3 14.4 15.0 26.9 26.4
16 charAt 6.1 6.5 73.4 857.0 8.4 8.2 8.3 13.6 13.5
32 charAt 3.9 3.7 54.8 428.9 5.0 4.9 4.7 7.0 7.2
64 charAt 2.7 2.6 48.2 232.9 3.0 3.2 3.3 3.9 4.0
128 charAt 2.1 1.9 43.7 138.8 2.1 2.6 2.6 2.4 2.6
256 charAt 1.9 1.6 42.4 90.6 1.7 2.1 2.1 1.7 1.8
512 field1 1.7 1.4 40.6 60.5 1.4 1.9 1.9 1.3 1.4
1,024 field1 1.6 1.4 40.0 45.6 1.2 1.9 2.1 1.0 1.2
2,048 field1 1.6 1.3 40.0 36.2 1.2 1.8 1.7 0.9 1.1
4,096 field1 1.6 1.3 39.7 32.6 1.2 1.8 1.7 0.9 1.0
8,192 field1 1.6 1.3 39.6 30.5 1.2 1.8 1.7 0.9 1.0
16,384 field1 1.6 1.3 39.8 28.4 1.2 1.8 1.7 0.8 1.0
32,768 field1 1.6 1.3 40.0 26.7 1.3 1.8 1.7 0.8 1.0
65,536 field1 1.6 1.3 39.8 26.3 1.3 1.8 1.7 0.8 1.0
131,072 field1 1.6 1.3 40.1 25.4 1.4 1.9 1.8 0.8 1.0
262,144 field1 1.6 1.3 39.6 25.2 1.5 1.9 1.9 0.8 1.0
-server MODE (forwards and backwards tests combined)Note: this is the test for Java 32 bit running in server mode on an AMD64. The server mode for Java 64 bit was the same as Java 32 bit in client mode except that Field access starting winning after 32 characters size.
Size WINNER charAt1 charAt2 stream streamPar reuse new1 new2 field1 field2
1 charAt 74.5 95.5 524.5 783.0 90.5 102.5 90.5 135.0 151.5
2 charAt 48.5 53.0 305.0 30851.3 59.3 57.5 52.0 88.5 91.8
4 charAt 28.8 32.1 132.8 2465.1 37.6 33.9 32.3 49.0 47.0
8 new2 18.0 18.6 63.4 1541.3 18.5 17.9 17.6 25.4 25.8
16 new2 14.0 14.7 129.4 1034.7 12.5 16.2 12.0 16.0 16.6
32 new2 7.8 9.1 19.3 431.5 8.1 7.0 6.7 7.9 8.7
64 reuse 6.1 7.5 11.7 204.7 3.5 3.9 4.3 4.2 4.1
128 reuse 6.8 6.8 9.0 101.0 2.6 3.0 3.0 2.6 2.7
256 field2 6.2 6.5 6.9 57.2 2.4 2.7 2.9 2.3 2.3
512 reuse 4.3 4.9 5.8 28.2 2.0 2.6 2.6 2.1 2.1
1,024 charAt 2.0 1.8 5.3 17.6 2.1 2.5 3.5 2.0 2.0
2,048 charAt 1.9 1.7 5.2 11.9 2.2 3.0 2.6 2.0 2.0
4,096 charAt 1.9 1.7 5.1 8.7 2.1 2.6 2.6 1.9 1.9
8,192 charAt 1.9 1.7 5.1 7.6 2.2 2.5 2.6 1.9 1.9
16,384 charAt 1.9 1.7 5.1 6.9 2.2 2.5 2.5 1.9 1.9
32,768 charAt 1.9 1.7 5.1 6.1 2.2 2.5 2.5 1.9 1.9
65,536 charAt 1.9 1.7 5.1 5.5 2.2 2.4 2.4 1.9 1.9
131,072 charAt 1.9 1.7 5.1 5.4 2.3 2.5 2.5 1.9 1.9
262,144 charAt 1.9 1.7 5.1 5.1 2.3 2.5 2.5 1.9 1.9
(to test on Java 7 and earlier, remove the two streams tests)
import java.lang.reflect.Field;
import java.util.ArrayList;
import java.util.Collections;
import java.util.List;
import java.util.Random;
import java.util.function.IntPredicate;
/**
* @author Saint Hill <http://stackoverflow.com/users/1584255/saint-hill>
*/
public final class TestStrings {
// we will not test strings longer than 512KM
final int MAX_STRING_SIZE = 1024 * 256;
// for each string size, we will do all the tests
// this many times
final int TRIES_PER_STRING_SIZE = 1000;
public static void main(String[] args) throws Exception {
new TestStrings().run();
}
void run() throws Exception {
// double the length of the data until it reaches MAX chars long
// 0,1,2,4,8,16,32,64,128,256 ...
final List<Integer> sizes = new ArrayList<>();
for (int n = 0; n <= MAX_STRING_SIZE; n = (n == 0 ? 1 : n * 2)) {
sizes.add(n);
}
// CREATE RANDOM (FOR SHUFFLING ORDER OF TESTS)
final Random random = new Random();
System.out.println("Rate in nanoseconds per character inspected.");
System.out.printf("==== FORWARDS (tries per size: %s) ==== \n", TRIES_PER_STRING_SIZE);
printHeadings(TRIES_PER_STRING_SIZE, random);
for (int size : sizes) {
reportResults(size, test(size, TRIES_PER_STRING_SIZE, random));
}
// reverse order or string sizes
Collections.reverse(sizes);
System.out.println("");
System.out.println("Rate in nanoseconds per character inspected.");
System.out.printf("==== BACKWARDS (tries per size: %s) ==== \n", TRIES_PER_STRING_SIZE);
printHeadings(TRIES_PER_STRING_SIZE, random);
for (int size : sizes) {
reportResults(size, test(size, TRIES_PER_STRING_SIZE, random));
}
}
///
///
/// METHODS OF CHECKING THE CONTENTS
/// OF A STRING. ALWAYS CHECKING FOR
/// WHITESPACE (CHAR <=' ')
///
///
// CHECK THE STRING CONTENTS
int charAtMethod1(final String data) {
final int len = data.length();
for (int i = 0; i < len; i++) {
if (data.charAt(i) <= ' ') {
doThrow();
}
}
return len;
}
// SAME AS ABOVE BUT USE String.length()
// instead of making a new final local int
int charAtMethod2(final String data) {
for (int i = 0; i < data.length(); i++) {
if (data.charAt(i) <= ' ') {
doThrow();
}
}
return data.length();
}
// USE new Java-8 String's IntStream
// pass it a PREDICATE to do the checking
int streamMethod(final String data, final IntPredicate predicate) {
if (data.chars().anyMatch(predicate)) {
doThrow();
}
return data.length();
}
// OH LA LA - GO PARALLEL!!!
int streamParallelMethod(final String data, IntPredicate predicate) {
if (data.chars().parallel().anyMatch(predicate)) {
doThrow();
}
return data.length();
}
// Re-fill a resuable char[] with the contents
// of the String's char[]
int reuseBuffMethod(final char[] reusable, final String data) {
final int len = data.length();
data.getChars(0, len, reusable, 0);
for (int i = 0; i < len; i++) {
if (reusable[i] <= ' ') {
doThrow();
}
}
return len;
}
// Obtain a new copy of char[] from String
int newMethod1(final String data) {
final int len = data.length();
final char[] copy = data.toCharArray();
for (int i = 0; i < len; i++) {
if (copy[i] <= ' ') {
doThrow();
}
}
return len;
}
// Obtain a new copy of char[] from String
// but use FOR-EACH
int newMethod2(final String data) {
for (final char c : data.toCharArray()) {
if (c <= ' ') {
doThrow();
}
}
return data.length();
}
// FANCY!
// OBTAIN FIELD FOR ACCESS TO THE STRING'S
// INTERNAL CHAR[]
int fieldMethod1(final Field field, final String data) {
try {
final char[] chars = (char[]) field.get(data);
final int len = chars.length;
for (int i = 0; i < len; i++) {
if (chars[i] <= ' ') {
doThrow();
}
}
return len;
} catch (Exception ex) {
throw new RuntimeException(ex);
}
}
// same as above but use FOR-EACH
int fieldMethod2(final Field field, final String data) {
final char[] chars;
try {
chars = (char[]) field.get(data);
} catch (Exception ex) {
throw new RuntimeException(ex);
}
for (final char c : chars) {
if (c <= ' ') {
doThrow();
}
}
return chars.length;
}
/**
*
* Make a list of tests. We will shuffle a copy of this list repeatedly
* while we repeat this test.
*
* @param data
* @return
*/
List<Jobber> makeTests(String data) throws Exception {
// make a list of tests
final List<Jobber> tests = new ArrayList<Jobber>();
tests.add(new Jobber("charAt1") {
int check() {
return charAtMethod1(data);
}
});
tests.add(new Jobber("charAt2") {
int check() {
return charAtMethod2(data);
}
});
tests.add(new Jobber("stream") {
final IntPredicate predicate = new IntPredicate() {
public boolean test(int value) {
return value <= ' ';
}
};
int check() {
return streamMethod(data, predicate);
}
});
tests.add(new Jobber("streamPar") {
final IntPredicate predicate = new IntPredicate() {
public boolean test(int value) {
return value <= ' ';
}
};
int check() {
return streamParallelMethod(data, predicate);
}
});
// Reusable char[] method
tests.add(new Jobber("reuse") {
final char[] cbuff = new char[MAX_STRING_SIZE];
int check() {
return reuseBuffMethod(cbuff, data);
}
});
// New char[] from String
tests.add(new Jobber("new1") {
int check() {
return newMethod1(data);
}
});
// New char[] from String
tests.add(new Jobber("new2") {
int check() {
return newMethod2(data);
}
});
// Use reflection for field access
tests.add(new Jobber("field1") {
final Field field;
{
field = String.class.getDeclaredField("value");
field.setAccessible(true);
}
int check() {
return fieldMethod1(field, data);
}
});
// Use reflection for field access
tests.add(new Jobber("field2") {
final Field field;
{
field = String.class.getDeclaredField("value");
field.setAccessible(true);
}
int check() {
return fieldMethod2(field, data);
}
});
return tests;
}
/**
* We use this class to keep track of test results
*/
abstract class Jobber {
final String name;
long nanos;
long chars;
long runs;
Jobber(String name) {
this.name = name;
}
abstract int check();
final double nanosPerChar() {
double charsPerRun = chars / runs;
long nanosPerRun = nanos / runs;
return charsPerRun == 0 ? nanosPerRun : nanosPerRun / charsPerRun;
}
final void run() {
runs++;
long time = System.nanoTime();
chars += check();
nanos += System.nanoTime() - time;
}
}
// MAKE A TEST STRING OF RANDOM CHARACTERS A-Z
private String makeTestString(int testSize, char start, char end) {
Random r = new Random();
char[] data = new char[testSize];
for (int i = 0; i < data.length; i++) {
data[i] = (char) (start + r.nextInt(end));
}
return new String(data);
}
// WE DO THIS IF WE FIND AN ILLEGAL CHARACTER IN THE STRING
public void doThrow() {
throw new RuntimeException("Bzzzt -- Illegal Character!!");
}
/**
* 1. get random string of correct length 2. get tests (List<Jobber>) 3.
* perform tests repeatedly, shuffling each time
*/
List<Jobber> test(int size, int tries, Random random) throws Exception {
String data = makeTestString(size, 'A', 'Z');
List<Jobber> tests = makeTests(data);
List<Jobber> copy = new ArrayList<>(tests);
while (tries-- > 0) {
Collections.shuffle(copy, random);
for (Jobber ti : copy) {
ti.run();
}
}
// check to make sure all char counts the same
long runs = tests.get(0).runs;
long count = tests.get(0).chars;
for (Jobber ti : tests) {
if (ti.runs != runs && ti.chars != count) {
throw new Exception("Char counts should match if all correct algorithms");
}
}
return tests;
}
private void printHeadings(final int TRIES_PER_STRING_SIZE, final Random random) throws Exception {
System.out.print(" Size");
for (Jobber ti : test(0, TRIES_PER_STRING_SIZE, random)) {
System.out.printf("%9s", ti.name);
}
System.out.println("");
}
private void reportResults(int size, List<Jobber> tests) {
System.out.printf("%6d", size);
for (Jobber ti : tests) {
System.out.printf("%,9.2f", ti.nanosPerChar());
}
System.out.println("");
}
}
Buh*_*ndi 13
这只是微观优化,您不必担心.
char[] chars = str.toCharArray();
返回一个str字符数组的副本(在JDK中,它通过调用返回一个字符副本System.arrayCopy).
除此之外,str.charAt()仅检查索引是否确实在边界内并返回数组索引中的字符.
第一个不会在JVM中创建额外的内存.
只是为了好奇,并与圣希尔的答案进行比较.
如果需要处理大量数据,则不应在客户端模式下使用JVM.客户端模式不适用于优化.
让我们在客户端模式和服务器模式下使用JVM比较@Saint Hill基准测试的结果.
Core2Quad Q6600 G0 @ 2.4GHz
JavaSE 1.7.0_40
另请参阅:"java -server"和"java -client"之间的真正区别？
客户模式:
len = 2: 111k charAt(i), 105k cbuff[i], 62k new[i], 17k field access. (chars/ms)
len = 4: 285k charAt(i), 166k cbuff[i], 114k new[i], 43k field access. (chars/ms)
len = 6: 315k charAt(i), 230k cbuff[i], 162k new[i], 69k field access. (chars/ms)
len = 8: 333k charAt(i), 275k cbuff[i], 181k new[i], 85k field access. (chars/ms)
len = 12: 342k charAt(i), 342k cbuff[i], 222k new[i], 117k field access. (chars/ms)
len = 16: 363k charAt(i), 347k cbuff[i], 275k new[i], 152k field access. (chars/ms)
len = 20: 363k charAt(i), 392k cbuff[i], 289k new[i], 180k field access. (chars/ms)
len = 24: 375k charAt(i), 428k cbuff[i], 311k new[i], 205k field access. (chars/ms)
len = 28: 378k charAt(i), 474k cbuff[i], 341k new[i], 233k field access. (chars/ms)
len = 32: 376k charAt(i), 492k cbuff[i], 340k new[i], 251k field access. (chars/ms)
len = 64: 374k charAt(i), 551k cbuff[i], 374k new[i], 367k field access. (chars/ms)
len = 128: 385k charAt(i), 624k cbuff[i], 415k new[i], 509k field access. (chars/ms)
len = 256: 390k charAt(i), 675k cbuff[i], 436k new[i], 619k field access. (chars/ms)
len = 512: 394k charAt(i), 703k cbuff[i], 439k new[i], 695k field access. (chars/ms)
len = 1024: 395k charAt(i), 718k cbuff[i], 462k new[i], 742k field access. (chars/ms)
len = 2048: 396k charAt(i), 725k cbuff[i], 471k new[i], 767k field access. (chars/ms)
len = 4096: 396k charAt(i), 727k cbuff[i], 459k new[i], 780k field access. (chars/ms)
len = 8192: 397k charAt(i), 712k cbuff[i], 446k new[i], 772k field access. (chars/ms)
服务器模式:
len = 2: 86k charAt(i), 41k cbuff[i], 46k new[i], 80k field access. (chars/ms)
len = 4: 571k charAt(i), 250k cbuff[i], 97k new[i], 222k field access. (chars/ms)
len = 6: 666k charAt(i), 333k cbuff[i], 125k new[i], 315k field access. (chars/ms)
len = 8: 800k charAt(i), 400k cbuff[i], 181k new[i], 380k field access. (chars/ms)
len = 12: 800k charAt(i), 521k cbuff[i], 260k new[i], 545k field access. (chars/ms)
len = 16: 800k charAt(i), 592k cbuff[i], 296k new[i], 640k field access. (chars/ms)
len = 20: 800k charAt(i), 666k cbuff[i], 408k new[i], 800k field access. (chars/ms)
len = 24: 800k charAt(i), 705k cbuff[i], 452k new[i], 800k field access. (chars/ms)
len = 28: 777k charAt(i), 736k cbuff[i], 368k new[i], 933k field access. (chars/ms)
len = 32: 800k charAt(i), 780k cbuff[i], 571k new[i], 969k field access. (chars/ms)
len = 64: 800k charAt(i), 901k cbuff[i], 800k new[i], 1306k field access. (chars/ms)
len = 128: 1084k charAt(i), 888k cbuff[i], 633k new[i], 1620k field access. (chars/ms)
len = 256: 1122k charAt(i), 966k cbuff[i], 729k new[i], 1790k field access. (chars/ms)
len = 512: 1163k charAt(i), 1007k cbuff[i], 676k new[i], 1910k field access. (chars/ms)
len = 1024: 1179k charAt(i), 1027k cbuff[i], 698k new[i], 1954k field access. (chars/ms)
len = 2048: 1184k charAt(i), 1043k cbuff[i], 732k new[i], 2007k field access. (chars/ms)
len = 4096: 1188k charAt(i), 1049k cbuff[i], 742k new[i], 2031k field access. (chars/ms)
len = 8192: 1157k charAt(i), 1032k cbuff[i], 723k new[i], 2048k field access. (chars/ms)
结论:
如您所见,服务器模式要快得多.
第一个使用str.charAt应该更快.
如果你深入挖掘String类的源代码,我们可以看到charAt实现如下:
public char charAt(int index) {
if ((index < 0) || (index >= count)) {
throw new StringIndexOutOfBoundsException(index);
}
return value[index + offset];
}
在这里,它所做的只是索引数组并返回值.
现在,如果我们看到实施toCharArray,我们将在下面找到:
public char[] toCharArray() {
char result[] = new char[count];
getChars(0, count, result, 0);
return result;
}
public void getChars(int srcBegin, int srcEnd, char dst[], int dstBegin) {
if (srcBegin < 0) {
throw new StringIndexOutOfBoundsException(srcBegin);
}
if (srcEnd > count) {
throw new StringIndexOutOfBoundsException(srcEnd);
}
if (srcBegin > srcEnd) {
throw new StringIndexOutOfBoundsException(srcEnd - srcBegin);
}
System.arraycopy(value, offset + srcBegin, dst, dstBegin,
srcEnd - srcBegin);
}
正如你所看到的那样,它正在做System.arraycopy一个肯定会比不这样做慢一点的速度.
| 归档时间： |
|
| 查看次数： |
225559 次 |
| 最近记录： |