Michael Bien's Weblog

Java 11 adds with JEP330 the ability to launch a Java source file directly from the command line without requiring to explicitly compile it. I can see this feature being convenient for some simple scripting or quick tests. Although Java isn't the most concise language it still has the benefit of being quite readable and having powerful utility APIs.

The following example reads the bash shell history from a file into a LinkedHashSet to retain ordering and keeps the most recently used line if duplicates exist, essentially cleaning up the history.

import java.io.IOException;
import java.nio.file.Files;
import java.nio.file.Path;
import java.util.LinkedHashSet;
import java.util.HashSet;
import java.util.stream.Stream;

public class HistoryCleaner {

    public static void main(String[] args) throws IOException {

        Path path = Path.of(System.getProperty("user.home")+"/.bash_history");

        HashSet<String> deduplicated = new LinkedHashSet<>(1024);
        try (Stream<String> lines = Files.lines(path)) {
            lines.forEach(line -> {
                deduplicated.remove(line);
                deduplicated.add(line);
            });
        }

        Files.write(path, deduplicated);
    }

}

duke@virtual1:~$ java HistoryCleaner.java

will compile and run the "script".

Shebangs are supported too

#!/usr/bin/java --source 11 -Xmx32m -XX:+UseSerialGC

import java.io....

However, since this is not part of the java language specification, single-file java programs (SFJP) with a #! (shebang) in the first line are technically no regular java files and will not compile when used with javac. This also means that the common file naming conventions are not required for pure SFJPs (i.e. file name does not have to be the same as the class name).

Using a different file extension is advised. We can't call it .js - so lets call it .sfjp :).

duke@virtual1:~$ ./HistoryCleaner.sfjp

This will compile and run the HistoryCleaner SFJP containing the shebang declaration in the first line. Keeping the .java file extension would result in a compiler error. As bonus we can also set JVM flags using the shebang mechanism. Keep in mind that -source [version] must be the first flag to tell the java launcher that it should expect a source file instead of bytecode.

Since I am doing a lot with OpenCL lately I decided to try to improve the tooling around OpenCL a bit. A weekend later the NetBeans OpenCL Pack was born :).

Features Including:

• OpenCL Editor with syntax highlighting code completion and CL reference pages integration
• OpenCL compiler integration
• In-editor annotations of compiler warnings and errors updated as you type
• JOCL project template

Technical Details:

The editor uses ANTLR as parser and lexer. This allows such simple things like keyword highlighting and also more complex features like semantic highlighting, formatting and auto completion (formatting is not yet implemented). It can also detect and report syntax errors, however this feature is automatically disabled if an OpenCL compiler is present on the host system. All with help of JOCL detected OpenCL implementations can be used as compiler backend.

Instead of using the old OpenGL Pack as template I decided to write it from scratch using latest NetBeans 7 and Java 7 APIs. So you will have to start NB with JDK7 to be able to use it.

Download

you can download it from the NetBeans plugin portal [mirror], sourcecode is on github

feedback and/or contributions/bugreports are as always appreciated

Screenshots:

have fun!

Ok some of them are big, but I will only cover the little things with this blog entry :).

CLKernel

I added multiple utility methods to CLKernel and related classes. It is for example now possible to create a kernel and set its arguments in one line.

CLKernel sha512 = program.createCLKernel("sha512", padBuffer, digestBuffer, rangeBuffer);

Thanks to feedback in the jocl forums I also added methods to set vector typed arguments directly. In past you could do this only by setting them via a java.util.Buffer.

kernel.setArg(index, x, y, z, w);

Another small feature of CLKernel is to enforce 32bit arguments. You may want to switch between single and double floatingpoint precision at runtime or mix between both to improve performance you will have to compile the program with the double FP extension enabled. By setting kernel.setForce32bitArgs(true) all java doubles used as kernel arguments will be automatically cast down to 32bit CL floats (see MultiDeviceFractal demo for a example). This is nothing special but might safe you several if(single){setArg((float)foo)}else{setArg(foo)} constructs.

CLWork

CLKernel still only represents the function in the OpenCL program you want to call - nothing more. The new CLWork object contains everything required for kernel execution, like the NDRange and the kernel itself.

    int size = buffer.getNIOCapacity();
    CLWork1D work = CLWork.create1D(program.createCLKernel("sum", buffer, size));
    work.setWorkSize(size, 1).optimizeFor(device);

    // execute
    queue.putWriteBuffer(buffer, false)
         .putWork(work)
         .putReadBuffer(buffer, true);

optimizeFor(device) adjusts the workgroup size to meet device specific recommended values. This should make sure that all computing units of your GPU are used by dividing the work into groups (however this only works if your task does not care about the workgroup size, see javadoc).

CLSubDevice

Sometimes you don't want to put your CLDevice under 100% load. This might be the case for example if your device is the CPU your application is running on or if you have to share the GPU with an OpenGL context for rendering. One easy way of controlling device load is to limit the amount of compute units used for a task.

    CLPlatform platform = CLPlatform.getDefault(version(CL_1_1), type(CPU));

    CLDevice devices = platform.getMaxFLOPSDevice(type(CPU));
    CLSubDevice[] subs = device.createSubDevicesByCount(4, 4);
    // array contains now two virtual devices containing four CPU cores each

    CLContext context = CLContext.create(subs);
    CLCommandQueue queue = subs[0].createCommandQueue();
    ...

CLSubDevices extends CLDevice and can be used for context creation, queue creation and everywhere you would use the CLDevice. Prior to creating subdevices you should check if device.isFissionSupported() returns true.

CLProgram builder

Ok, this utility is not that new but I haven't blogged about it yet. If program.build() isn't enough you should take a look at the program builder. CLBuildConfiguration stores everything which is needed for program compilation and is easily configurable via the builder pattern :).

        // reusable builder
        CLBuildConfiguration builder = CLProgramBuilder.createConfiguration()
                                     .withOption(ENABLE_MAD)
                                     .forDevices(context.getDevices())
                                     .withDefine("RADIUS", 5)
                                     .withDefine("ENABLE_FOOBAR");
        builder.build(programA);
        builder.build(programB);
        ...

CLBuildConfiguration is fully reusable and can be upgraded to CLProgramConfiguration if you combine it with a CLProgram. Both can be serialised which allows to store the build configuration or the entire prebuild program on disc or send it over the network. (caching binaries on disc can safe startup time for example)

        // program configuration
        ois = new ObjectInputStream(new FileInputStream(file));
        CLProgramConfiguration programConfig = CLProgramBuilder.loadConfiguration(ois, context);
        assertNotNull(programConfig.getProgram());
        ois.close();
        program = programConfig.build(); // builds from source or loads binaries if possible
        assertTrue(program.isExecutable());

Note: loading binaries and associating them with the right driver/device is currently not trivial with OpenCL. Even if everything works as intended it is still possible that the driver refuses the binaries for some reason (driver update...etc). Thats why its recommended to add the program source to the configuration before calling build() to allow a automatic rebuild as fallback.

        // another entry point for complex builds (prepare() returns CLProgramConfiguration)
        program.prepare().withOption(ENABLE_MAD).forDevice(context.getMaxFlopsDevice()).build();

(all snippets have been stolen from the junit tests)
I am sure I forgot something... but this should cover at least some of the incremental improvements. Expect a few more blog entries for the larger features soon.

- - - - - -
In other news: Nvidia released OpenCL 1.1 drivers, some of us thought this would never happen -> all major vendors (AMD, Intel, NV, IBM, ZiiLABS ..) support now OpenCL 1.1 (screenshot)

have fun!

One nice feature of OpenCL is that the platform abstraction was handled in the spec from the first day on. You can install all OpenCL drivers side by side and let the application choose at runtime, on which device and on which platform it should execute the kernels.

As of today there are three four vendors which provide OpenCL implementations for the desktop. AMD and Intel support the OpenCL 1.1 specification where Nvidia apparently tries to stick with 1.0 to encourage their customers to stick with CUDA ;-). [edit] And of course there is also Apple providing out-of-the box OpenCL 1.0 support in MacOS 10.6.

JOCL contains a small CLInfo utility which can be used to quickly verify OpenCL installations. Here is the output on my system (ubuntu was booted) having all three SDKs installed:

The CLInfo utility is part of the jocl-demos project and is also available via webstart. For a plain text version of the above output you can run:

 java -jar jocl.jar:gluegen-rt.jar\
    -Djava.library.path="path/to/jocl/libs:path/to/gluegen/libs" com.jogamp.opencl.util.CLInfo

(btw to install the intel sdk on debian based systems follow this link)

happy coding!

I am currently working on Java Binding for the OpenCL API using GlueGen (as used in JOGL, JOAL). The project started as part of my bachelor of CS thesis short after the release of the first OpenCL specification draft and is now fully feature complete with OpenCL 1.1. JOCL is currently in the stabilization phase, a beta release shouldn't be far away.

Overview - How does it work?

The low level bindings (LLB) are automatically generated using the official OpenCL headers as input and provide a high performance, JNI based, 1:1 mapping to the C functions.

This has the following advantages:

• reduces maintenance overhead and ensures spec conformance
• compiletime JNI bindings are the fastest way to access native libs from the JVM
• makes translating OpenCL C code into Java + JOCL very easy (e.g. from books or tutorials)
• flexibility and stability: OpenCL libs are loaded dynamically and accessed via function pointers

The hand written high level bindings (HLB) is build on top of LLB and hides most boilerplate code (like object IDs, pointers and resource management) behind easy to use java objects. HLB use direct NIO buffers internally for fast memory transfers between the JVM and the OpenCL implementation and is very GC friendly. Most of the API is designed for method chaining but of course you don't have to use it this way if you don't want to. JOCL also seamlessly integrates with JOGL 2 (both are built and tested together). Just pass the JOGL context as parameter to the JOCL context factory and you will receive a shared context. If you already know OpenCL and Java, HLB should be very intuitive for you.

The project is available on jogamp.org. Please use the mailinglist / forum for feedback or questions and the bugtracker if you experience any issues. The JOCL root repository is located on github, you may also want to take a look at the jocl-demos project. (If the demos are not enough you might also want to take a look at the junit tests)

Screenshots (sourcecode in jocl-demos project):

More regarding OpenGL interoperability and other features in upcoming blog entries.

The following sample shows basic setup, computation and cleanup using the high level APIs.

Hello World or parallel a+b=c

/**
 * Hello Java OpenCL example. Adds all elements of buffer A to buffer B
 * and stores the result in buffer C.
 * Sample was inspired by the Nvidia VectorAdd example written in C/C++
 * which is bundled in the Nvidia OpenCL SDK.
 * @author Michael Bien
 */
public class HelloJOCL {

    public static void main(String[] args) throws IOException {
        // Length of arrays to process (arbitrary number)
        int elementCount = 11444777;
        // Local work size dimensions
        int localWorkSize = 256;
        // rounded up to the nearest multiple of the localWorkSize
        int globalWorkSize = roundUp(localWorkSize, elementCount);

        // setup
        CLContext context = CLContext.create();

        CLProgram program = context.createProgram(
                       HelloJOCL.class.getResourceAsStream("VectorAdd.cl")
                                 ).build();

        CLBuffer<FloatBuffer> clBufferA =
                       context.createFloatBuffer(globalWorkSize, READ_ONLY);
        CLBuffer<FloatBuffer> clBufferB =
                       context.createFloatBuffer(globalWorkSize, READ_ONLY);
        CLBuffer<FloatBuffer> clBufferC =
                       context.createFloatBuffer(globalWorkSize, WRITE_ONLY);

        out.println("used device memory: "
            + (clBufferA.getSize()+clBufferB.getSize()+clBufferC.getSize())/1000000 +"MB");

        // fill read buffers with random numbers (just to have test data).
        fillBuffer(clBufferA.getBuffer(), 12345);
        fillBuffer(clBufferB.getBuffer(), 67890);

        // get a reference to the kernel functon with the name 'VectorAdd'
        // and map the buffers to its input parameters.
        CLKernel kernel = program.createCLKernel("VectorAdd");
        kernel.putArgs(clBufferA, clBufferB, clBufferC).putArg(elementCount);

        // create command queue on fastest device.
        CLCommandQueue queue = context.getMaxFlopsDevice().createCommandQueue();

        // asynchronous write to GPU device,
        // blocking read later to get the computed results back.
        long time = nanoTime();
        queue.putWriteBuffer(clBufferA, false)
             .putWriteBuffer(clBufferB, false)
             .put1DRangeKernel(kernel, 0, globalWorkSize, localWorkSize)
             .putReadBuffer(clBufferC, true);
        time = nanoTime() - time;

        // cleanup all resources associated with this context.
        context.release();

        // print first few elements of the resulting buffer to the console.
        out.println("a+b=c results snapshot: ");
        for(int i = 0; i < 10; i++)
            out.print(clBufferC.getBuffer().get() + ", ");
        out.println("...; " + clBufferC.getBuffer().remaining() + " more");

        out.println("computation took: "+(time/1000000)+"ms");

    }

    private static final void fillBuffer(FloatBuffer buffer, int seed) {
        Random rnd = new Random(seed);
        while(buffer.remaining() != 0)
            buffer.put(rnd.nextFloat()*100);
        buffer.rewind();
    }

    private static final int roundUp(int groupSize, int globalSize) {
        int r = globalSize % groupSize;
        if (r == 0) {
            return globalSize;
        } else {
            return globalSize + groupSize - r;
        }
    }

}

VectorAdd.cl

    // OpenCL Kernel Function for element by element vector addition
    kernel void VectorAdd(global const float* a,
                          global const float* b,
                          global float* c, int numElements) {

        // get index into global data array
        int iGID = get_global_id(0);

        // bound check (equivalent to the limit on a 'for' loop)
        if (iGID >= numElements)  {
            return;
        }

        // add the vector elements
        c[iGID] = a[iGID] + b[iGID];
    }

Thanks to Justin Stoecker, computer science graduate student at the University of Miami, JOGL gets a new set of getting started tutorials:

Tutorials:

• index
• Tutorial 1 - Environment Setup
• Tutorial 2 - Creating a Window
• Tutorial 3 - Creating a Render Loop

• performance metrics
• logging, debugging or diagnostics
• to ignore specific function calls

    public void init(GLAutoDrawable drawable) {
        // wrap composeable pipeline in a Debug utility, all OpenGL error codes are automatically
        // converted to GLExceptions as soon as they appear
        drawable.setGL(new DebugGL3(drawable.getGL().getGL3()));
        //..
    }

        drawable.setGL(new TraceGL3(drawable.getGL().getGL3(), System.out));

This post is a short description how to convert a CVS repository into GIT. I used the tool cvs2svn from Tigris.org for this task. As an example I will use the project jake2.

0. Install cvs2svn and cvs

sudo apt-get install cvs
sudo apt-get install cvs2svn

1. Download the CVS repository

Sourceforge for example supports repository synchronization via rsync to allow backups. Thats what I used to get a copy of the jake2 repository.

rsync -av rsync://jake2.cvs.sourceforge.net/cvsroot/jake2/* .

2. Convert into GIT fast-import format

cvs2git --options=cvs2git.options

This will produce a dump and blob file for step 3.

3. Create GIT repository and import CVS history

mkdir gitrepo
cd gitrepo
git init
cat ../repo-tmp/git-blob.dat ../repo-tmp/git-dump.dat | git fast-import

4. Manual cleanup

btw you can find the jake2 git repo here.

You might remember the project called Jackpot which came out of SunLabs and had James Gosling involved with. It was basically a way to declaratively define refactoring rules, allowing for example, to migrate client code between incompatible third party libraries. The project has been integrated into NetBeans as the IDE's refactoring engine since then. NetBeans 6.9 uses Jackpot for most of the in-code hints and code inspections for instance.

There were various ways to specify the transformation rules, e.g. via a special declarative language or even in Annotations directly in the library-code which would cause incompatibilities (or e.g in conjunction with @Deprecated).

Jan Lahoda recently started with the efforts to make the project usable as standalone tool again. Jackpot 3.0 is available via bitbucket for early adopters.

Back to the Future

I used this opportunity to test jackpotc (the jackpot compiler) with JOGL. What I tired is to provide transformations which transform old JOGL 1.1.1 code into latest JOGL 2 compatible client code. So firstly thanks to Jan for fixing all the bugs we ran into while testing the experimental command line compiler.

The first thing I did was to transform the code to properly use OpenGL profiles. As testcode i will use the famous Gears OpenGL demo as example (but those kind of automatic transformations will only pay of if you use them on large codebases). Since it was written against JOGL 1.1.1 it can only use OpenGL up to version 2.x, which means we can simply use the GL2 profile.

Transformation Rules

'JOGL2 API change: javax.media.opengl.GL -> javax.media.opengl.GL2':
javax.media.opengl.GL=>javax.media.opengl.GL2;;

'JOGL2 API change: new javax.media.opengl.GLCapabilities(javax.media.opengl.GLProfile)':
new javax.media.opengl.GLCapabilities()=>
new javax.media.opengl.GLCapabilities(javax.media.opengl.GLProfile.get(javax.media.opengl.GLProfile.GL2));;

'JOGL2 API change: GL gl = drawable.getGL() -> GL2 gl = drawable.getGL().getGL2()':
$d.getGL() :: $d instanceof javax.media.opengl.GLAutoDrawable=>
$d.getGL().getGL2();; 

Just by looking at the transformation rules you can easily see that it is far more powerful than any simple text replacement could be. Jackpot uses javac and can therefore work with full qualified names, instanceof and more. It will also correctly fix imports for you (there is currently a open bug in this area). The quotes are used as description string which will be printed when jackpotc runs on every code occurrence which applies to that rule.

Invoking Jackpot

jackpotc -sourcepath $SRC -cp $LIBS -d $OUTPUT\
         -Ajackpot30_extra_hints=./jogl1Tojogl2.hint $FILESET

$LIBS must contain both library versions, JOGL 1.1.1 and JOGL 2. This is not optimal but it will probably work in most situations to just use both without thinking about an particular ordering or the need to do multiple iterations.

Results

./test/oldgears/src/jogl111/gears/Gears.java:54: warning: JOGL2 API change: GL gl = drawable.getGL() -> GL2 gl = drawable.getGL().getGL2()
    GL gl = drawable.getGL();
...

@@ -51,7 +51,7 @@
     // Use debug pipeline
     // drawable.setGL(new DebugGL(drawable.getGL()));
 
-    GL gl = drawable.getGL();
+    GL2 gl = drawable.getGL().getGL2();
...

You can find the complete demo and all ready-to-run shellscripts in the tools/jackpotc folder inside JOGL's git repository. The classic JOGL 2 Gears demo can be found in form of an applet here (uses latest hudson builds... can be unstable).

happy coding!

3D & Multimedia Across Platforms and Devices Using JOGL
Tuesday, 27 July | 4:00 PM - 6:00 PM

This session discusses the features, contributions, and future of OpenGL, OpenCL, and OpenMax
across devices and OS exposed on top of Java using the JogAmp open-source libraries.

If you are using GIT as SCM and NetBeans as IDE you should probably check out NBGit. The plugin integrates GIT in NetBeans in the same way as the out of the box Mercurial support does it. In fact both modules have the same origin since nbgit is a fork of the mercurial integration project and incrementally adds features to catch up.

NBGit Version 0.3 is already fairly stable and provides the basic set of features you would expect from distributed versioning system IDE integration.

Features

• Graph visualization of parallel branches (Browser similar to giggle)
• Versioning History (git log)
• Show changes (git status)
• update/commit/reset
• clone/clone other/git init
• custom actions (custom git commands)
• diff
• in-editor annotation of code changes
• ignore files (parsing '.gitignore' files)
• git properties (username, email etc via options)

The project is developed by volunteers outside Sun, if you like to see GIT integration as out-of-the-box feature in a future version of NetBeans please vote for this RFE.

I use the plugin for most of my open source projects and haven't experience any serious issues so far. I would say its already safe to use since you can't do anything wrong if you do a 'git status' -> 'git push' via command line as last step anyway.

June 16 2010, updated blogpost: OpenGL 4

JOGL 2 supports several OpenGL Profiles. In this blog entry I try to explain what profiles are and why they are needed.

History

SGI released the first OpenGL specification 1992. Since this point OpenGL 1.x constantly evolved (under the ARB and later Khronos Group) by adding new functions to the core API. This went well until programmable graphics hardware became mainstream and shaders became suddenly more flexible and efficient as the generic fixed function pipeline.

OpenGL 2.x was the last version in which you could freely mix the fixed function pipeline with the programmable pipeline (as a core feature).

With the release of OpenGL 3.0 the whole fixed function pipeline has been deprecated but you could still use it if you haven't requested a forward compatible context.

OpenGL 3.1 and 3.2 removed most deprecated functionality from core specification, however some implementations (e.g. Nvidia drivers) still allow to get them back via an optional compatibility extension. Since 3.1 was the first release which broke compatibility, it is often seen as major OpenGL 3 release.

JOGL 2 (JSR 231)

JOGL 1.1.1 lived in the timeframe up to OpenGL 3.0 which made it easy to stay in sync with the spec. To be able to solve the issue with the deprecation of functionality, JOGL 2 (JSR maintenance release) introduces an abstraction of the original OpenGL versioning called Profile. Profiles allow Java applications to be written in a way which allows compatibility with multiple OpenGL versions at the same time. Since OpenGL ES (GL for embedded systems) has overlapping functionality with OpenGL itself it opened the opportunity to add even Profiles which bridge desktop and embedded implementations. The class diagram below shows the dependencies between all available Profiles.

Before you start writing a JOGL application you will have to decide first which GLProfile you want to use. The code snippet below lists all currently supported profiles (extracted from GLProfile).

Current list of supported profiles and their mapping to the implementation versions

    /** The desktop OpenGL compatibility profile 4.x, with x >= 0, ie GL2 plus GL4.

        bc stands for backward compatibility. */
    public static final String GL4bc = "GL4bc";

    /** The desktop OpenGL core profile 4.x, with x >= 0 */
    public static final String GL4   = "GL4";

    /** The desktop OpenGL compatibility profile 3.x, with x >= 1, ie GL2 plus GL3.

        bc stands for backward compatibility. */
    public static final String GL3bc = "GL3bc";

    /** The desktop OpenGL core profile 3.x, with x >= 1 */
    public static final String GL3   = "GL3";

    /** The desktop OpenGL profile 1.x up to 3.0 */
    public static final String GL2   = "GL2";

    /** The embedded OpenGL profile ES 1.x, with x >= 0 */
    public static final String GLES1 = "GLES1";

    /** The embedded OpenGL profile ES 2.x, with x >= 0 */
    public static final String GLES2 = "GLES2";

    /** The intersection of the desktop GL2 and embedded ES1 profile */
    public static final String GL2ES1 = "GL2ES1";

    /** The intersection of the desktop GL3, GL2 and embedded ES2 profile */
    public static final String GL2ES2 = "GL2ES2";

    /** The intersection of the desktop GL3 and GL2 profile */
    public static final String GL2GL3 = "GL2GL3";

Note: GL2 Profile supports OpenGL up to version 3.0 (included) - this is not a bug: OpenGL 3.1 was the big game changer

The next two code snippets show the basic steps how to set up OpenGL with JOGL 2.

Context creation

        //create a profile, in this case OpenGL 3.1 or later
        GLProfile profile = GLProfile.get(GLProfile.GL3);
        
        //configure context
        GLCapabilities capabilities = new GLCapabilities(profile);
        capabilities.setNumSamples(2); // enable anti aliasing - just as a example
        capabilities.setSampleBuffers(true);
        
        //initialize a GLDrawable of your choice
        GLCanvas canvas = new GLCanvas(capabilities);

        //register GLEventListener
        canvas.addGLEventListener(...);
        //... (start rendering thread -> start rendering...)

Rendering

    public void display(GLAutoDrawable drawable) {
        GL3 gl = drawable.getGL().getGL3();
        gl.glClear(GL.GL_COLOR_BUFFER_BIT | GL.GL_DEPTH_BUFFER_BIT);
        //.. render something
    }

Summary

Profiles make JOGL 2 very flexible and allow it to build modular and portable applications. For instance part A of an application can be written against the GL2ES2 interface and part B (which is more hardware specific) against the GL3 interface. This would in theory allow to reuse A in an embedded application and B could e.g. disable itself on old desktop hardware which runs only OpenGL 2.x or fall back to a GL2 implementation.

More information can be found on JogAmp.org (direct link to javadoc)

The next release of the OpenGL Pack for NetBeans will fully support JOGL 2. Beta builds can be found here (builds contain JOGL2 beta5):

The XPath Utility I submitted to the NetBeans Plugin Portal over two months ago has been recently verified against NetBeans 6.7. This makes the plugin now directly available from within the IDE over the Plugin manager (Tools -> Plugins).

One .nbm less to carry with me ;)

Object pooling in java is often seen as an anti pattern and/or wasted effort - but there are still valid reasons to think about pooling for certain kind of applications.

The JVM allocates objects much faster from managed heap (young generation; contiguous and defragmented) as you could ever recycle objects from a self written pool running on top of a VM. A good configured garbage collector is also able to delete unused objects fast. GCs in fact don't delete objects explicitly, they rather evacuate all surviving objects and sweep whole memory regions in a very efficient manner and only when its necessary to reduce runtime overhead.

Object allocation (of small objects) on modern JVMs is even so fast that making a copy of immutable objects sometimes outperforms modification of mutable (and often old) objects. JVM languages like scala or clojure make heavy use of this observation. One of the reasons for that anomaly is that generational JVMs are designed to be able to deal with loads of short living objects which makes them inexpensive compared to long living objects in old generations.

Performance does not always mean Throughput

Rendering a game with 60fps might be optimal throughput for a renderer but the performance might be still unacceptable when all frames are rendered in the first half of the second with the second half spent on GC ;). Even if Object Pools may not increase system throughput they can still increase determinism of your application. Here are some observations and tips which might help:

When should I consider Object Pools?

• GC tuning did not help - you want to try something else
• The application creates a lot of objects which die in the old generation
• Your Objects are expensive to create but easy to recycle
• Determinism, e.g response time (soft real time requirements) is more important for you than throughput

Pro Pooling:

• pools reduce GC activity in peak times (worst case scenarios)
• are easy to implement and test (its basically an array ;))
• are easy to disable (inject a fake pool which returns only new Objects)

Con Pooling:

• more (old) objects are referenced when a GC kicks in (increases gc overhead)
• memory leaks (don't forget to reclaim your objects!)
• cause additional problems in a multi-threaded scenario (new Object() is thread safe!)
• may decrease throughput
• cumbersome, repetitive client code

When you decided to use pools you have to make sure to reclaim all objects as soon they are no longer used. One way of doing this is by applying the static factory method pattern for object allocation and a per object dispose method for deallocation.

/**not Thread safe!**/
public class Vector3f {
    
    private static final ObjectPool<Vector3f> pool;
    public float x, y, z;
    private boolean disposed;
    
    static{
        pool = new ObjectPool<Vector3f>(1024);
        for(int i = 0; i  < 1024; i++) {
            pool.reclaim(new Vector3f());
        }
    }

    private Vector3f() {}

    public static Vector3f create(float x, float y, float z) {
        Vector v = pool.isEmpty() ? new Vector() : pool.get();
        v.x = x;
        v.y = y;
        v.z = z;
        v.disposed = false;
        return v;
    }
    
    public void dispose() {
        if(!disposed) {
            disposed = true;
            pool.reclaim(this);
        }
    }
}

To demonstrate the perceived performance difference I captured two flyovers of my old 3d engine. The second flyover was captured with disabled object pools. The terrain engine triangulates the ground dependent on the position and view direction of the observer which makes object allocation hard to predict. The triangulation runs in parallel to the rendering thread which made the pool implementations a bit more complex as the example above.

Every vertex, normal, triangle and quad-tree node is a pooled object (wireframe on mouse over)

on the left: flyover with pre allocated object pools; right: dynamic object allocation (new Object())

Notice the pauses at 7, 17 and 26s on the flyover with disabled pools (right video).

Note on the videos: The quality is very bad since the tool I used created 700MB large files for the 30s videos a lot of frames got skipped. I even sampled them down from 1600x1200 to 1024x768 and limited the fps to 30 but the bottleneck was still the hard disk. This is the main reason why even the left video does not look smooth. (I even had to boot windows the first time in 2 years to use the tool!). I'll try to capture better vids next time.

Conclusion

Using pools requires discipline, is error prone, not good for system throughput and does not play very well with threads. However there are some attempts to make them more usable in case you think you need them. The physics engine JBullet for example uses JStackAlloc to prevent repetitive and cumbersome code by using automatic bytecode instrumentation in the build process. Type Annotations (JSR 308 targeted for OpenJDK 7) in combination with project lombok and/or the automatic resource management proposal might provide further possibilities for simplifying the usage of object pools in java and reduce the risk for memory leaks.