Boolean valueOf instead of new Boolean
The Boolean class has cache for true and false and using it, will save memory and will faster than using create new instance of boolean. Using JMH[1] with a code test[2] the result was: Benchmark Mode Samples Mean Mean error Units m.BooleanBenchmark.newInstanceBoolean thrpt 20 49801.326 369.897 ops/s m.BooleanBenchmark.newInstanceString thrpt 20 365.080 27.537 ops/s m.BooleanBenchmark.valueOfBoolean thrpt 20 764906233.316 9623009.653 ops/s m.BooleanBenchmark.valueOfString thrpt 20 371.174 28.216 ops/s The diff is on attachment or can is downloading the webdrev here: https://dl.dropboxusercontent.com/u/16109193/open_jdk/boolean_instance_of.zi... [1] http://openjdk.java.net/projects/code-tools/jmh/ [2] @State(Scope.Thread) @OutputTimeUnit(TimeUnit.SECONDS) public class BooleanBenchmark { private static final int SIZE = 1_000_000; private List<String> booleanString; private boolean[] booleans; { booleans = new boolean[SIZE]; booleanString = new ArrayList<>(SIZE); for (int index = 0; index < SIZE; index++) { if (index % 2 == 0) { booleans[index] = true; booleanString.add(Boolean.TRUE.toString()); } else { booleans[index] = false; booleanString.add(Boolean.FALSE.toString()); } } } @GenerateMicroBenchmark public void valueOfBoolean() { for(boolean b: booleans) { Boolean result = b; } } @GenerateMicroBenchmark public void valueOfString() { for(String b: booleanString) { Boolean result = Boolean.valueOf(b); } } @GenerateMicroBenchmark public void newInstanceBoolean() { for(boolean b: booleans) { Boolean result = new Boolean(b); } } @GenerateMicroBenchmark public void newInstanceString() { for(String b: booleanString) { Boolean result = new Boolean(b); } } } -- Atenciosamente. Otávio Gonçalves de Santana blog: http://otaviosantana.blogspot.com.br/ twitter: http://twitter.com/otaviojava site: http://www.otaviojava.com.br (11) 98255-3513
On 05/24/2014 06:34 PM, Otávio Gonçalves de Santana wrote:
The Boolean class has cache for true and false and using it, will save memory and will faster than using create new instance of boolean. Using JMH[1] with a code test[2] the result was:
I agree Boolean.valueOf (whether explicit or implicit) should be used if identity is not required. Do you really need explicit Boolean.valueOf in, say, GSSManagerImpl, instead of relying on autoboxing? That said, your benchmark is not correct. At very least, you have to use explicit BlackHoles to avoid DCE [1][2]. This is how you do it: @State(Scope.Thread) @OutputTimeUnit(TimeUnit.SECONDS) public class BooleanBenchmark { @Param("1000000") private int size; private List<String> booleanString; private boolean[] booleans; @Setup public void s() { booleans = new boolean[size]; booleanString = new ArrayList<String>(size); for (int index = 0; index < size; index++) { if (index % 2 == 0) { booleans[index] = true; booleanString.add(Boolean.TRUE.toString()); } else { booleans[index] = false; booleanString.add(Boolean.FALSE.toString()); } } } @GenerateMicroBenchmark public void valueOfBoolean(BlackHole bh) { for (boolean b : booleans) { Boolean result = b; bh.consume(result); } } @GenerateMicroBenchmark public void valueOfString(BlackHole bh) { for (String b : booleanString) { Boolean result = Boolean.valueOf(b); bh.consume(result); } } @GenerateMicroBenchmark public void newInstanceBoolean(BlackHole bh) { for (boolean b : booleans) { Boolean result = new Boolean(b); bh.consume(result); } } @GenerateMicroBenchmark public void newInstanceString(BlackHole bh) { for (String b : booleanString) { Boolean result = new Boolean(b); bh.consume(result); } } } Thanks, -Aleksey. [1] http://hg.openjdk.java.net/code-tools/jmh/file/75f8b23444f6/jmh-samples/src/... [2] http://hg.openjdk.java.net/code-tools/jmh/file/75f8b23444f6/jmh-samples/src/...
Hi Alexis. Thank you for fixes the benchMarck. About the class GSSManagerImpl, yes it is necessary because the System.getProperty(USE_NATIVE_PROP) returns a String. On Sat, May 24, 2014 at 12:19 PM, Aleksey Shipilev < aleksey.shipilev@oracle.com> wrote:
On 05/24/2014 06:34 PM, Otávio Gonçalves de Santana wrote:
The Boolean class has cache for true and false and using it, will save memory and will faster than using create new instance of boolean. Using JMH[1] with a code test[2] the result was:
I agree Boolean.valueOf (whether explicit or implicit) should be used if identity is not required. Do you really need explicit Boolean.valueOf in, say, GSSManagerImpl, instead of relying on autoboxing?
That said, your benchmark is not correct. At very least, you have to use explicit BlackHoles to avoid DCE [1][2]. This is how you do it:
@State(Scope.Thread) @OutputTimeUnit(TimeUnit.SECONDS) public class BooleanBenchmark {
@Param("1000000") private int size;
private List<String> booleanString; private boolean[] booleans;
@Setup public void s() { booleans = new boolean[size]; booleanString = new ArrayList<String>(size);
for (int index = 0; index < size; index++) { if (index % 2 == 0) { booleans[index] = true; booleanString.add(Boolean.TRUE.toString()); } else { booleans[index] = false; booleanString.add(Boolean.FALSE.toString()); } } }
@GenerateMicroBenchmark public void valueOfBoolean(BlackHole bh) { for (boolean b : booleans) { Boolean result = b; bh.consume(result); } }
@GenerateMicroBenchmark public void valueOfString(BlackHole bh) { for (String b : booleanString) { Boolean result = Boolean.valueOf(b); bh.consume(result); } }
@GenerateMicroBenchmark public void newInstanceBoolean(BlackHole bh) { for (boolean b : booleans) { Boolean result = new Boolean(b); bh.consume(result); } }
@GenerateMicroBenchmark public void newInstanceString(BlackHole bh) { for (String b : booleanString) { Boolean result = new Boolean(b); bh.consume(result); } } }
Thanks, -Aleksey.
[1]
http://hg.openjdk.java.net/code-tools/jmh/file/75f8b23444f6/jmh-samples/src/... [2]
http://hg.openjdk.java.net/code-tools/jmh/file/75f8b23444f6/jmh-samples/src/...
-- Atenciosamente. Otávio Gonçalves de Santana blog: http://otaviosantana.blogspot.com.br/ twitter: http://twitter.com/otaviojava site: http://www.otaviojava.com.br (11) 98255-3513
Hi Otávio, it would be nice, if you would not modify the classes sun.reflect.UnsafeXXXFieldAccessorImpl. This classes should be changed as a part of the fix for the issue JDK-5043030. The patch for this issue is already in work. Best regards, Andrej Golovnin On 24.05.2014, at 16:34, Otávio Gonçalves de Santana <otaviojava@java.net> wrote:
The Boolean class has cache for true and false and using it, will save memory and will faster than using create new instance of boolean. Using JMH[1] with a code test[2] the result was: Benchmark Mode Samples Mean Mean error Units m.BooleanBenchmark.newInstanceBoolean thrpt 20 49801.326 369.897 ops/s m.BooleanBenchmark.newInstanceString thrpt 20 365.080 27.537 ops/s m.BooleanBenchmark.valueOfBoolean thrpt 20 764906233.316 9623009.653 ops/s m.BooleanBenchmark.valueOfString thrpt 20 371.174 28.216 ops/s
The diff is on attachment or can is downloading the webdrev here: https://dl.dropboxusercontent.com/u/16109193/open_jdk/boolean_instance_of.zi...
[1] http://openjdk.java.net/projects/code-tools/jmh/
[2]
@State(Scope.Thread)
@OutputTimeUnit(TimeUnit.SECONDS)
public class BooleanBenchmark {
private static final int SIZE = 1_000_000;
private List<String> booleanString;
private boolean[] booleans;
{
booleans = new boolean[SIZE];
booleanString = new ArrayList<>(SIZE);
for (int index = 0; index < SIZE; index++) {
if (index % 2 == 0) {
booleans[index] = true;
booleanString.add(Boolean.TRUE.toString());
} else {
booleans[index] = false;
booleanString.add(Boolean.FALSE.toString());
}
}
}
@GenerateMicroBenchmark
public void valueOfBoolean() {
for(boolean b: booleans) {
Boolean result = b;
}
}
@GenerateMicroBenchmark
public void valueOfString() {
for(String b: booleanString) {
Boolean result = Boolean.valueOf(b);
}
}
@GenerateMicroBenchmark
public void newInstanceBoolean() {
for(boolean b: booleans) {
Boolean result = new Boolean(b);
}
}
@GenerateMicroBenchmark
public void newInstanceString() {
for(String b: booleanString) {
Boolean result = new Boolean(b);
}
}
}
-- Atenciosamente.
Otávio Gonçalves de Santana
blog: http://otaviosantana.blogspot.com.br/ twitter: http://twitter.com/otaviojava site: http://www.otaviojava.com.br (11) 98255-3513 <boolean.diff>
Really Happy to hear that. Done. On Sat, May 24, 2014 at 5:10 PM, Andrej Golovnin <andrej.golovnin@gmail.com>wrote:
Hi Otávio,
it would be nice, if you would not modify the classes sun.reflect.UnsafeXXXFieldAccessorImpl. This classes should be changed as a part of the fix for the issue JDK-5043030. The patch for this issue is already in work.
Best regards, Andrej Golovnin
On 24.05.2014, at 16:34, Otávio Gonçalves de Santana <otaviojava@java.net> wrote:
The Boolean class has cache for true and false and using it, will save memory and will faster than using create new instance of boolean. Using JMH[1] with a code test[2] the result was: Benchmark Mode Samples Mean Mean error Units m.BooleanBenchmark.newInstanceBoolean thrpt 20 49801.326 369.897 ops/s m.BooleanBenchmark.newInstanceString thrpt 20 365.080 27.537 ops/s m.BooleanBenchmark.valueOfBoolean thrpt 20 764906233.316 9623009.653 ops/s m.BooleanBenchmark.valueOfString thrpt 20 371.174 28.216 ops/s
The diff is on attachment or can is downloading the webdrev here:
https://dl.dropboxusercontent.com/u/16109193/open_jdk/boolean_instance_of.zi...
[1] http://openjdk.java.net/projects/code-tools/jmh/
[2]
@State(Scope.Thread)
@OutputTimeUnit(TimeUnit.SECONDS)
public class BooleanBenchmark {
private static final int SIZE = 1_000_000;
private List<String> booleanString;
private boolean[] booleans;
{
booleans = new boolean[SIZE];
booleanString = new ArrayList<>(SIZE);
for (int index = 0; index < SIZE; index++) {
if (index % 2 == 0) {
booleans[index] = true;
booleanString.add(Boolean.TRUE.toString());
} else {
booleans[index] = false;
booleanString.add(Boolean.FALSE.toString());
}
}
}
@GenerateMicroBenchmark
public void valueOfBoolean() {
for(boolean b: booleans) {
Boolean result = b;
}
}
@GenerateMicroBenchmark
public void valueOfString() {
for(String b: booleanString) {
Boolean result = Boolean.valueOf(b);
}
}
@GenerateMicroBenchmark
public void newInstanceBoolean() {
for(boolean b: booleans) {
Boolean result = new Boolean(b);
}
}
@GenerateMicroBenchmark
public void newInstanceString() {
for(String b: booleanString) {
Boolean result = new Boolean(b);
}
}
}
-- Atenciosamente.
Otávio Gonçalves de Santana
blog: http://otaviosantana.blogspot.com.br/ twitter: http://twitter.com/otaviojava site: http://www.otaviojava.com.br (11) 98255-3513 <boolean.diff>
-- Atenciosamente. Otávio Gonçalves de Santana blog: http://otaviosantana.blogspot.com.br/ twitter: http://twitter.com/otaviojava site: http://www.otaviojava.com.br (11) 98255-3513
Can anyone help me as sponsor? On May 25, 2014 2:08 AM, "Otávio Gonçalves de Santana" <otaviojava@java.net> wrote:
Really Happy to hear that. Done.
On Sat, May 24, 2014 at 5:10 PM, Andrej Golovnin < andrej.golovnin@gmail.com> wrote:
Hi Otávio,
it would be nice, if you would not modify the classes sun.reflect.UnsafeXXXFieldAccessorImpl. This classes should be changed as a part of the fix for the issue JDK-5043030. The patch for this issue is already in work.
Best regards, Andrej Golovnin
On 24.05.2014, at 16:34, Otávio Gonçalves de Santana <otaviojava@java.net> wrote:
The Boolean class has cache for true and false and using it, will save memory and will faster than using create new instance of boolean. Using JMH[1] with a code test[2] the result was: Benchmark Mode Samples Mean Mean error Units m.BooleanBenchmark.newInstanceBoolean thrpt 20 49801.326 369.897 ops/s m.BooleanBenchmark.newInstanceString thrpt 20 365.080 27.537 ops/s m.BooleanBenchmark.valueOfBoolean thrpt 20 764906233.316 9623009.653 ops/s m.BooleanBenchmark.valueOfString thrpt 20 371.174 28.216 ops/s
The diff is on attachment or can is downloading the webdrev here:
https://dl.dropboxusercontent.com/u/16109193/open_jdk/boolean_instance_of.zi...
[1] http://openjdk.java.net/projects/code-tools/jmh/
[2]
@State(Scope.Thread)
@OutputTimeUnit(TimeUnit.SECONDS)
public class BooleanBenchmark {
private static final int SIZE = 1_000_000;
private List<String> booleanString;
private boolean[] booleans;
{
booleans = new boolean[SIZE];
booleanString = new ArrayList<>(SIZE);
for (int index = 0; index < SIZE; index++) {
if (index % 2 == 0) {
booleans[index] = true;
booleanString.add(Boolean.TRUE.toString());
} else {
booleans[index] = false;
booleanString.add(Boolean.FALSE.toString());
}
}
}
@GenerateMicroBenchmark
public void valueOfBoolean() {
for(boolean b: booleans) {
Boolean result = b;
}
}
@GenerateMicroBenchmark
public void valueOfString() {
for(String b: booleanString) {
Boolean result = Boolean.valueOf(b);
}
}
@GenerateMicroBenchmark
public void newInstanceBoolean() {
for(boolean b: booleans) {
Boolean result = new Boolean(b);
}
}
@GenerateMicroBenchmark
public void newInstanceString() {
for(String b: booleanString) {
Boolean result = new Boolean(b);
}
}
}
-- Atenciosamente.
Otávio Gonçalves de Santana
blog: http://otaviosantana.blogspot.com.br/ twitter: http://twitter.com/otaviojava site: http://www.otaviojava.com.br (11) 98255-3513 <boolean.diff>
-- Atenciosamente.
Otávio Gonçalves de Santana
blog: http://otaviosantana.blogspot.com.br/ twitter: http://twitter.com/otaviojava site: http://www.otaviojava.com.br (11) 98255-3513
Hi Otávio, I can sponsor these two (Boolean and single char strings) for you. Because they cross over different repositories and require different reviews it would be more convenient to process each of them in two batches (client vs core) Please can you break out the 'client' changes into a separate webrev? The client packages are: (java and javax...) java2d, awt, swing, font, print, beans, media, imageio, applet, sound and accessibility. The client webrevs should be posted to awt-dev and the other two to core-libs, net-dev and servicability-dev. Thanks, Roger On 5/27/2014 1:16 PM, Otávio Gonçalves de Santana wrote:
Can anyone help me as sponsor? On May 25, 2014 2:08 AM, "Otávio Gonçalves de Santana" <otaviojava@java.net> wrote:
---------- Forwarded message ---------- From: roger riggs <roger.riggs@oracle.com> Date: Tue, May 27, 2014 at 3:02 PM Subject: Re: Boolean valueOf instead of new Boolean To: core-libs-dev@openjdk.java.net Hi Otávio, I can sponsor these two (Boolean and single char strings) for you. Because they cross over different repositories and require different reviews it would be more convenient to process each of them in two batches (client vs core) Please can you break out the 'client' changes into a separate webrev? The client packages are: (java and javax...) java2d, awt, swing, font, print, beans, media, imageio, applet, sound and accessibility. The client webrevs should be posted to awt-dev and the other two to core-libs, net-dev and servicability-dev. Thanks, Roger On 5/27/2014 1:16 PM, Otávio Gonçalves de Santana wrote:
Can anyone help me as sponsor? On May 25, 2014 2:08 AM, "Otávio Gonçalves de Santana" < otaviojava@java.net> wrote:
-- Atenciosamente. Otávio Gonçalves de Santana blog: http://otaviosantana.blogspot.com.br/ twitter: http://twitter.com/otaviojava site: http://www.otaviojava.com.br (11) 98255-3513
participants (5)
-
Aleksey Shipilev
-
Andrej Golovnin
-
Otávio Gonçalves de Santana
-
Otávio Gonçalves de Santana
-
roger riggs