Re: RFR: 8148937: (str) Adapt StringJoiner for Compact Strings [v3]
On Mon, 15 Mar 2021 21:47:28 GMT, Сергей Цыпанов
wrote:
>> Hello,
>>
>> as of now `java.util.StringJoiner` still uses `char[]` as a storage for
>> joined Strings.
>>
>> This applies for the cases when all joined Strings as well as delimiter,
>> prefix and suffix contain only ASCII symbols.
>>
>> As a result when `StringJoiner.toString()` is called `byte[]` stored in
>> Strings is inflated in order to fill in `char[]` and after that `char[]` is
>> compressed when constructor of String is called:
>> String delimiter = this.delimiter;
>> char[] chars = new char[this.len + addLen];
>> int k = getChars(this.prefix, chars, 0);
>> if (size > 0) {
>> k += getChars(elts[0], chars, k);// inflate byte[]
>>
>> for(int i = 1; i < size; ++i) {
>> k += getChars(delimiter, chars, k);
>> k += getChars(elts[i], chars, k);
>> }
>> }
>>
>> k += getChars(this.suffix, chars, k);
>> return new String(chars);// compress char[] -> byte[]
>> This can be improved by utilizing new method `String.getBytes(byte[], int,
>> int, byte, int)` [introduced](https://github.com/openjdk/jdk/pull/402) in
>> [JDK-8224986](https://bugs.openjdk.java.net/browse/JDK-8254082)
>> covering both cases when resulting String is Latin1 or UTF-16
>>
>> I've prepared a patch along with benchmark proving that this change is
>> correct and brings improvement.
>>
>> @BenchmarkMode(Mode.AverageTime)
>> @OutputTimeUnit(TimeUnit.NANOSECONDS)
>> @Fork(jvmArgsAppend = {"-Xms2g", "-Xmx2g"})
>> public class StringJoinerBenchmark {
>>
>> @Benchmark
>> public String stringJoiner(Data data) {
>> String[] stringArray = data.stringArray;
>> return Joiner.joinWithStringJoiner(stringArray);
>> }
>>
>> @State(Scope.Thread)
>> public static class Data {
>>
>> @Param({"latin", "cyrillic", "mixed"})
>> private String mode;
>>
>> @Param({"8", "32", "64"})
>> private int length;
>>
>> @Param({"5", "10", "100"})
>> private int count;
>>
>> private String[] stringArray;
>>
>> @Setup
>> public void setup() {
>> RandomStringGenerator generator = new RandomStringGenerator();
>>
>> stringArray = new String[count];
>>
>> for (int i = 0; i < count; i++) {
>> String alphabet = getAlphabet(i, mode);
>> stringArray[i] = generator.randomString(alphabet, length);
>> }
>> }
>>
>> private static String getAlphabet(int index, String mode) {
>> var latin = "abcdefghijklmnopqrstuvwxyz"; //English
>> var cyrillic = "абвгдеёжзиклмнопрстуфхцчшщьыъэюя"; // Russian
>>
>> String alphabet;
>> switch (mode) {
>> case "mixed" -> alphabet = index % 2 == 0 ? cyrillic : latin;
>> case "latin" -> alphabet = latin;
>> case "cyrillic" -> alphabet = cyrillic;
>> default -> throw new RuntimeException("Illegal mode " + mode);
>> }
>> return alphabet;
>> }
>> }
>> }
>>
>> public class Joiner {
>>
>> public static String joinWithStringJoiner(String[] stringArray) {
>> StringJoiner joiner = new StringJoiner(",", "[", "]");
>> for (String str : stringArray) {
>> joiner.add(str);
>> }
>> return joiner.toString();
>> }
>> }
>>
>>
>> (count) (length)(mode)
>> Java 14 patched Units
>> stringJoiner 5 8 latin 78.836
>> ± 0.20867.546 ± 0.500 ns/op
>> stringJoiner 532 latin 92.877
>> ± 0.42266.760 ± 0.498 ns/op
>> stringJoiner 564 latin 115.423
>> ± 0.88373.224 ± 0.289 ns/op
>> stringJoiner 10 8 latin 152.587
>> ± 0.429 161.427 ± 0.635 ns/op
>> stringJoiner 1032 latin 189.998
>> ± 0.478 164.099 ± 0.963 ns/op
>> stringJoiner 1064 latin 238.679
>> ± 1.419 176.825 ± 0.533 ns/op
>> stringJoiner100 8 latin 1215.612
>> ± 17.413 1541.802 ± 126.166 ns/op
>> stringJoiner10032 latin 1699.998
>> ± 28.407 1563.341 ± 4.439 ns/op
>> stringJoiner10064 latin 2289.388
>> ± 45.319 2215.931 ± 137.583 ns/op
>> stringJoiner 5 8 cyrillic 96.692
>> ± 0.94780.946 ± 0.371 ns/op
>> stringJoiner 532 cyrillic 107.806
>> ± 0.42984.717 ± 0.541 ns/op
>> stringJoiner 564 cyrillic 150.762
>> ± 2.26796.214 ± 1.251 ns/op
>> stringJoiner 10 8 cyrillic
Re: RFR: 8148937: (str) Adapt StringJoiner for Compact Strings [v3]
On Mon, 15 Mar 2021 21:47:28 GMT, Сергей Цыпанов
wrote:
>> Hello,
>>
>> as of now `java.util.StringJoiner` still uses `char[]` as a storage for
>> joined Strings.
>>
>> This applies for the cases when all joined Strings as well as delimiter,
>> prefix and suffix contain only ASCII symbols.
>>
>> As a result when `StringJoiner.toString()` is called `byte[]` stored in
>> Strings is inflated in order to fill in `char[]` and after that `char[]` is
>> compressed when constructor of String is called:
>> String delimiter = this.delimiter;
>> char[] chars = new char[this.len + addLen];
>> int k = getChars(this.prefix, chars, 0);
>> if (size > 0) {
>> k += getChars(elts[0], chars, k);// inflate byte[]
>>
>> for(int i = 1; i < size; ++i) {
>> k += getChars(delimiter, chars, k);
>> k += getChars(elts[i], chars, k);
>> }
>> }
>>
>> k += getChars(this.suffix, chars, k);
>> return new String(chars);// compress char[] -> byte[]
>> This can be improved by utilizing new method `String.getBytes(byte[], int,
>> int, byte, int)` [introduced](https://github.com/openjdk/jdk/pull/402) in
>> [JDK-8224986](https://bugs.openjdk.java.net/browse/JDK-8254082)
>> covering both cases when resulting String is Latin1 or UTF-16
>>
>> I've prepared a patch along with benchmark proving that this change is
>> correct and brings improvement.
>>
>> @BenchmarkMode(Mode.AverageTime)
>> @OutputTimeUnit(TimeUnit.NANOSECONDS)
>> @Fork(jvmArgsAppend = {"-Xms2g", "-Xmx2g"})
>> public class StringJoinerBenchmark {
>>
>> @Benchmark
>> public String stringJoiner(Data data) {
>> String[] stringArray = data.stringArray;
>> return Joiner.joinWithStringJoiner(stringArray);
>> }
>>
>> @State(Scope.Thread)
>> public static class Data {
>>
>> @Param({"latin", "cyrillic", "mixed"})
>> private String mode;
>>
>> @Param({"8", "32", "64"})
>> private int length;
>>
>> @Param({"5", "10", "100"})
>> private int count;
>>
>> private String[] stringArray;
>>
>> @Setup
>> public void setup() {
>> RandomStringGenerator generator = new RandomStringGenerator();
>>
>> stringArray = new String[count];
>>
>> for (int i = 0; i < count; i++) {
>> String alphabet = getAlphabet(i, mode);
>> stringArray[i] = generator.randomString(alphabet, length);
>> }
>> }
>>
>> private static String getAlphabet(int index, String mode) {
>> var latin = "abcdefghijklmnopqrstuvwxyz"; //English
>> var cyrillic = "абвгдеёжзиклмнопрстуфхцчшщьыъэюя"; // Russian
>>
>> String alphabet;
>> switch (mode) {
>> case "mixed" -> alphabet = index % 2 == 0 ? cyrillic : latin;
>> case "latin" -> alphabet = latin;
>> case "cyrillic" -> alphabet = cyrillic;
>> default -> throw new RuntimeException("Illegal mode " + mode);
>> }
>> return alphabet;
>> }
>> }
>> }
>>
>> public class Joiner {
>>
>> public static String joinWithStringJoiner(String[] stringArray) {
>> StringJoiner joiner = new StringJoiner(",", "[", "]");
>> for (String str : stringArray) {
>> joiner.add(str);
>> }
>> return joiner.toString();
>> }
>> }
>>
>>
>> (count) (length)(mode)
>> Java 14 patched Units
>> stringJoiner 5 8 latin 78.836
>> ± 0.20867.546 ± 0.500 ns/op
>> stringJoiner 532 latin 92.877
>> ± 0.42266.760 ± 0.498 ns/op
>> stringJoiner 564 latin 115.423
>> ± 0.88373.224 ± 0.289 ns/op
>> stringJoiner 10 8 latin 152.587
>> ± 0.429 161.427 ± 0.635 ns/op
>> stringJoiner 1032 latin 189.998
>> ± 0.478 164.099 ± 0.963 ns/op
>> stringJoiner 1064 latin 238.679
>> ± 1.419 176.825 ± 0.533 ns/op
>> stringJoiner100 8 latin 1215.612
>> ± 17.413 1541.802 ± 126.166 ns/op
>> stringJoiner10032 latin 1699.998
>> ± 28.407 1563.341 ± 4.439 ns/op
>> stringJoiner10064 latin 2289.388
>> ± 45.319 2215.931 ± 137.583 ns/op
>> stringJoiner 5 8 cyrillic 96.692
>> ± 0.94780.946 ± 0.371 ns/op
>> stringJoiner 532 cyrillic 107.806
>> ± 0.42984.717 ± 0.541 ns/op
>> stringJoiner 564 cyrillic 150.762
>> ± 2.26796.214 ± 1.251 ns/op
>> stringJoiner 10 8 cyrillic
Re: RFR: 8148937: (str) Adapt StringJoiner for Compact Strings [v3]
> Hello,
>
> as of now `java.util.StringJoiner` still uses `char[]` as a storage for
> joined Strings.
>
> This applies for the cases when all joined Strings as well as delimiter,
> prefix and suffix contain only ASCII symbols.
>
> As a result when `StringJoiner.toString()` is called `byte[]` stored in
> Strings is inflated in order to fill in `char[]` and after that `char[]` is
> compressed when constructor of String is called:
> String delimiter = this.delimiter;
> char[] chars = new char[this.len + addLen];
> int k = getChars(this.prefix, chars, 0);
> if (size > 0) {
> k += getChars(elts[0], chars, k);// inflate byte[]
>
> for(int i = 1; i < size; ++i) {
> k += getChars(delimiter, chars, k);
> k += getChars(elts[i], chars, k);
> }
> }
>
> k += getChars(this.suffix, chars, k);
> return new String(chars);// compress char[] -> byte[]
> This can be improved by utilizing new method `String.getBytes(byte[], int,
> int, byte, int)` [introduced](https://github.com/openjdk/jdk/pull/402) in
> [JDK-8224986](https://bugs.openjdk.java.net/browse/JDK-8254082)
> covering both cases when resulting String is Latin1 or UTF-16
>
> I've prepared a patch along with benchmark proving that this change is
> correct and brings improvement.
>
> @BenchmarkMode(Mode.AverageTime)
> @OutputTimeUnit(TimeUnit.NANOSECONDS)
> @Fork(jvmArgsAppend = {"-Xms2g", "-Xmx2g"})
> public class StringJoinerBenchmark {
>
> @Benchmark
> public String stringJoiner(Data data) {
> String[] stringArray = data.stringArray;
> return Joiner.joinWithStringJoiner(stringArray);
> }
>
> @State(Scope.Thread)
> public static class Data {
>
> @Param({"latin", "cyrillic", "mixed"})
> private String mode;
>
> @Param({"8", "32", "64"})
> private int length;
>
> @Param({"5", "10", "100"})
> private int count;
>
> private String[] stringArray;
>
> @Setup
> public void setup() {
> RandomStringGenerator generator = new RandomStringGenerator();
>
> stringArray = new String[count];
>
> for (int i = 0; i < count; i++) {
> String alphabet = getAlphabet(i, mode);
> stringArray[i] = generator.randomString(alphabet, length);
> }
> }
>
> private static String getAlphabet(int index, String mode) {
> var latin = "abcdefghijklmnopqrstuvwxyz"; //English
> var cyrillic = "абвгдеёжзиклмнопрстуфхцчшщьыъэюя"; // Russian
>
> String alphabet;
> switch (mode) {
> case "mixed" -> alphabet = index % 2 == 0 ? cyrillic : latin;
> case "latin" -> alphabet = latin;
> case "cyrillic" -> alphabet = cyrillic;
> default -> throw new RuntimeException("Illegal mode " + mode);
> }
> return alphabet;
> }
> }
> }
>
> public class Joiner {
>
> public static String joinWithStringJoiner(String[] stringArray) {
> StringJoiner joiner = new StringJoiner(",", "[", "]");
> for (String str : stringArray) {
> joiner.add(str);
> }
> return joiner.toString();
> }
> }
>
>
> (count) (length)(mode)
> Java 14 patched Units
> stringJoiner 5 8 latin 78.836
> ± 0.20867.546 ± 0.500 ns/op
> stringJoiner 532 latin 92.877
> ± 0.42266.760 ± 0.498 ns/op
> stringJoiner 564 latin 115.423
> ± 0.88373.224 ± 0.289 ns/op
> stringJoiner 10 8 latin 152.587
> ± 0.429 161.427 ± 0.635 ns/op
> stringJoiner 1032 latin 189.998
> ± 0.478 164.099 ± 0.963 ns/op
> stringJoiner 1064 latin 238.679
> ± 1.419 176.825 ± 0.533 ns/op
> stringJoiner100 8 latin 1215.612
> ± 17.413 1541.802 ± 126.166 ns/op
> stringJoiner10032 latin 1699.998
> ± 28.407 1563.341 ± 4.439 ns/op
> stringJoiner10064 latin 2289.388
> ± 45.319 2215.931 ± 137.583 ns/op
> stringJoiner 5 8 cyrillic 96.692
> ± 0.94780.946 ± 0.371 ns/op
> stringJoiner 532 cyrillic 107.806
> ± 0.42984.717 ± 0.541 ns/op
> stringJoiner 564 cyrillic 150.762
> ± 2.26796.214 ± 1.251 ns/op
> stringJoiner 10 8 cyrillic 190.570
> ± 0.381 182.754 ± 0.678 ns/op
> stringJoiner 1032 cyrillic 240.239
> ± 1.110 187.991 ± 1.575 ns/op
>
