Java 6-what we really got?

[SluX]

Can any1 summarize performance and other benefits from java 6 release...especially the graphic part?
Maybe some links?

Ty vm.

[Eliwood]

I haven't really looked at graphical performance yet, but I did notice that ImageIO improved... a lot.

I don't have the link on me anymore, but I do have this graph.



In terms of real loading times, let's say that mine got slashed by 40% just because of this.

[kappa]

Applets are much nicer in Java 6 they have a nice loading progress thing instead of that horrible grey box, still could be better, more snappy and less freezing the browser but definitely an improvement.

Overall there an optimisations all over it.

Games published by our own members! Check 'em out!

[CommanderKeith]

www.javadesktop.org has had stuff on Java 6 performance lately.  Word is that the performance gain is 'double-digit'.  Here's 2 recent blogs about it:

http://blogs.sun.com/dagastine/entry/java_6_leads_out_of

http://jroller.com/page/dgilbert?entry=is_java_se_1_6

My experience with Java 6 is that its Direct3D (& also OpenGL) pipeline is now a lot faster (flag -Dsun.java2d.d3d=true from http://java.sun.com/javase/6/docs/technotes/guides/2d/flags.html).

Keith

PS: see Chris Campell's many blog entries for stuff on Graphics performance.  Eliwood's graph comes from there: http://weblogs.java.net/blog/campbell/archive/2006/01/400_horsepower.html

The best one: http://weblogs.java.net/blog/campbell/archive/2005/07/strcrazier_perf.html

[Eliwood]

I turned on the Java2D OpenGL flag and noticed that my game went just as smoothly as it did when using LWJGL, and to boot, loading times were faster, even versus the custom TGA loader. Makes me kinda feel bad about spending all that time working on something that Sun's effectively deprecated for me now, but I'm surprised at how much things have improved since Tiger.

Unfortunately, this was not the case with the D3D tag set to the true, which was about the same performance as plain vanilla Java2D.

[pepijnve]

A big boost in number crunching performance too. JSquish (pure java s3tc texture compressor) compresses images about twice as fast under jse6 compared to jse5.

[Linuxhippy]

For the server-jvm I did a benchmark some time ago which is almost exclusivly algorythmic code:

in our self-rolled framework we use a piece of encrypthion/decrption code a lot, I did some benchmarking of the server compiler just because of interrest - maybe you find the results interesting:

Time to decrypt 1Mib:

1.3.1_16: 208ms
1.4.2: 174ms
1.5.0: 165ms
6.0_b98: 158ms
6.0_b98: 148ms (with EscapeAnalysis enabled)
-------------------------------------------------
IBM 1.4.2: 174ms
JRockit-5: 187ms

I find these numbers really impressive especially when  taking into account that code mostly reflects the quality of jit generated code and does not use anything else.the jvm provides. Therefor I would expect even better numbers for real world code
lg Clemens

[Ken Russell]

the client-compiler produces now faster code at the expense of higher compilation times - the register allocator already used in the server-jit was moved to client too.

These statements are incorrect on two counts. First, based on measurements with several benchmarks, the new client compiler (linear scan + SSA) has the same or better compilation speed than the previous versions, in addition to producing higher-quality machine code. Second, a new linear scan register allocator was developed for the client compiler which is unrelated to the graph-coloring register allocator in the server compiler. You can find some more information about the new register allocator here and here.

[Linuxhippy]

sorry ... i guess i am a master in spreading misinformation.

[K.I.L.E.R]

Java 6 has EA?
I thought it was pushed back onto Dolphin?

Games published by our own members! Check 'em out!

[princec]

It has EA but it currently does not actually stack-allocate objects. Well, last time I looked. No time to squeeze that in presumably. And besides they want to be able to show off something big in Java 7 don't they

Cas

[Linuxhippy]

To be honest I really did not expect such huge wins for java-6 with the server-compiler, taking into account that all the "exciting" stuff was pushed back to dolphin.

I am really excited when we'll see:
* Stack based allocation
* Tired compilation
* JIT code cache (theres a JCP which speaks about bundling pre-compiled code ~-> quite like a JIT cache)

lg Clemens

[princec]

All the real fun happened on the client.

Cas

[kappa]

hopefully they'll manage to get SIMD's support in dolphin too.

[blahblahblahh]

I am really excited when we'll see:

...Structs!

* blahblahblahh ducks and runs for cover

[darkprophet]

Java 6 has EA to avoid deadlocks of threads i think, not for stack based allocation. Java 7 already has two teired compilation! Already some things to boast about...

But it seems like dolphin branched from mustang around b50ish, so alot of the bug fixes aren't in (like java_impl.exe cmd line thing that pops up). I wonder how they are going to combine the two...

DP

[fletchergames]

It's funny that you should mention bugs.  Some change in Java 6 causes the bug in http://www.java-gaming.org/forums/index.php?topic=15410.msg122714 .  I experience the same thing in Windows, though not with the same code the guy posted in the bug report.  It does however occur with some code I wrote.

I encountered a game-breaking bug with some other version several years ago.  I believe it was 1.4.2_02.  I didn't find out about the workaround until I bought Developing Games in Java.

Even so, the problem I'm having is a relatively minor one.  For the rest of my current project, I'll just use 1.5.0 update 10.  After it's done, I'll try to create a workaround if the good people at Sun haven't fixed it by then.

I just wanted to point out that Java 6 isn't perfect.  It's alot better than having to use C++ and an endless supply of 3rd party libaries though!

[Linuxhippy]

I just wanted to point out that Java 6 isn't perfect.  It's alot better than having to use C++ and an endless supply of 3rd party libaries though!

Software is never perfect - but it can be really good!

lg Clemens

[SluX]

Indeed. I guess that this edition we have now really pwns...
Now only to get users to download it...

Btw...since now jvm is opensourced, are there gonna be some "lighter" implementations? I mean flash player is kinda weightless compared to jre....maybe not a place to discuss here...

[oNyx]

>[...]are there gonna be some "lighter" implementations?

Yes, you can then bundle some stripped down (private) JRE with your application. If you avoid swing/awt (~if you use lwjgl heh), the overhead will be pretty small. With lzma/p200 voodoo it's only around 1.5-2.0mb* (depends on how much you actually need).

[* The latest flash installer is 1.4mb.]

A tiny lwjgl game fits nicely into 2.5mb for example and with a total size of 5mb you have *lots* of room.

[Linuxhippy]

1.) you would have to create bundles for all operating systems
2.) Your users could not benefit automatically from future updates made to java
3.) All the footprint-saving (shared classes, ...) will not work.

However the most important point:
4.) Your game will not help to spread! Users won't associate it with java, they won't download a JRE (helping other java games/software).

Depends on you wether you like this.

lg Clemens

[SluX]

You are right...

But, at least here in Serbia, when ppl have bad connections, it is harder to convince them to downloadm 5-15 megs of jre...if we had some official 1.5 megs version that would rock....maybe some day we will ..or not...

[princec]

Don't get too many customers from Slovenia yet

Cas

[oNyx]

1.) you would have to create bundles for all operating systems
2.) Your users could not benefit automatically from future updates made to java
3.) All the footprint-saving (shared classes, ...) will not work.
[...]
4.) Your game will not help to spread! Users won't associate it with java, they won't download a JRE (helping other java games/software).

1) No. I target 1.4. So there will be a Mac zip app bundle for example. No need to ship any Java there. And I also won't bundle one for Linux (for technical reasons).

2) So? The game will run fine on ancient hardware. That's enough for me.

3) In the short run... yes, it will. If the user downloads more than ~5 Java games, he/she will have to download a bit more. But that isn't really the reason for this anyways. It's for making it work out of the box on windows, too.

4) Yes, it won't increase the number of machines with Java installed.

Well, the user doesn't care which language you've used. They only want something which just works. Its already bad enough that I (will) ask for working opengl drivers.

I used to share your point of view. But at the end of the day I would rather have some hundred/thousand people more who try it instead of converting a dozen. I can propagate Java through other means more effectively anyways.

[kappa]

yup do agree, why bother players with more requirements like having them worry about having Java or even knowing what it is, something like bundling stripped jre would be more successful.

Until of course Java becomes as easy as flash to install, i largely consider its success is because its so easy to install, just look at Firefox if you don't have flash just a few clicks and its up and running on your computer all automated. Hoping future java will be as easy to install, good thing is theres already some working being done on that modular jre thing so could be ready for dolphin.

[Linuxhippy]

look at Firefox if you don't have flash just a few clicks and its up and running on your computer all automated. Hoping .

*lool* no get MS to do that for java 

lg Clemens

[SluX]

One more question: are bufferedimages accelerated by default now or we still need to create compatible ones etc...
Can any1 summarize the process in the 1.6 java?
Tnx

[oNyx]

The are supposed to be managed automatically from 1.5 on.

[whome]

hopefully they'll manage to get SIMD's support in dolphin too.

What is SIMD?

[darkprophet]

Single Instruction Multiple Data

Performing the same set of operations on multiple streams....thats how the GPUs these days can get these nice performance figures because rendering polygons is an embarrasingly parallel problem

Only SIMD is for CPU's and SSE2/3 only support 4 data streams at 1 point while modern GPUs support 32 (the G80 supports 64 doesn't it ?)

DP