endlos - A multithreaded fractal generator in Java

[Jestre]

@Jestre: thank you very much! Can you tell me what operating system you run endlos on? And perhaps other hardware configuration (CPU, RAM)? I'm always interested in the target systems of endlos.

Hey there,

Looking forward to the new release, though I've certainly not worn out the current version

I am running it on my laptop.  A Gateway m465e Centrino Duo thingy.  It has the "Genuine Intel(R) CPU T2500  @ 2.00GHz" processor according to cpuinfo, 2G RAM, and I installed the Java 6 JDK on the recommendation of the readme file.  The OS is Ubuntu Edgy (6.10).

Keep up the good work, and thanks for the welcome!

[dentaku2]

@Jestre: oh, that's interesting! You also run Ubuntu Edgy like me (I run the 64 Bit version and you?). But you have an Intel dual core 2GHz and I an AMD dual core 2 GHz (also 2 GB RAM). I expect the new Intel CPU being faster. Can you tell me the runtime of the demo fractal "Sun (deep)", please? It's in 0.14.0:

Code:

2007-01-14 : Version 0.14.0 - fast colors
---------------------------

CHANGED:
  1. Recoloring of an image doesn't recalculate fractal anymore! Now you can change color maps, smoothing,
     anti-aliasing on the fly - it's much more satisfying to play with colors.
  2. As some users complained about a "missing" GUI and told that they had to maximize it from the task panel,
     the workbench window's initial size is set to 1024x768. I hope this addresses the issue as I can't
     reproduce this on my computer.
  3. Several operands increased from 32 bit (int) to 64 bit (long). This might decrease performance on 32 bit
     systems but gives a much wider range of precision on 64 bit systems without slowing down.
  4. The demo fractals now use the updated ColorMaps version 1.1.0.

FIXED:
  1. "Open location" didn't work in high precision mode.
  2. If zoom clicks came too fast, the fractal could fall into endless sleep. :-(

NEW:
  1. Cycle iterations of fractal which is basically a colour cycling (via mouse wheel).
  2. The demo directory contains a new fractal 'Sun (deep)' with a precision of 34 (128 Bit). I will use this
     for benchmarking to measure the future performance increase of the iteration algorithm.
     Calculation time on my AMD X2 3800+ EE (2 GHz), Sun Java 1.6, Ubuntu Linux 64 bit: 03:14:35. Tell me yours ... ;-)
  3. Usage of sourceforge subversion repository for resource control.

http://endlos.sourceforge.net/

[dentaku2]

Code:

2007-01-20 : Version 0.15.0 - speed & misc
---------------------------

CHANGED:
  1. High precision speed improvement: demo fractal "Sun (deep)" now takes 02:23:28 compared to
     03:14:35 of former version.

NEW:
  1. Progress information (percentage, remaining) in fractal window title.
  2. New image popup menu item to save current color map. E.g. to save a random color map.
  3. New demo fractal: "Four Comets"
  4. Further speed optimization through Java parameters. I already found one Java parameter that
     reduced calculation time of demo fractal "Sun (deep)" by 5 minutes: -XX:+UseBiasedLocking
     
     java -server -XX:+UseBiasedLocking -jar endlos.jar
  
  5. New feature: play ... MIDI files! If a subdirectory "MIDI/" is available in the startdirectory (e.g.
     a shortcut to your MIDI collection folder), a random file will be played in the background. Here's
     one suggested place to get MIDI files: http://www.vgmusic.com/

http://endlos.sourceforge.net/

[Jestre]

@Jestre: oh, that's interesting! You also run Ubuntu Edgy like me (I run the 64 Bit version and you?). But you have an Intel dual core 2GHz and I an AMD dual core 2 GHz (also 2 GB RAM). I expect the new Intel CPU being faster. Can you tell me the runtime of the demo fractal "Sun (deep)", please? It's in 0.14.0:

Well, this is a Centrino chip, so I'm sure it would run faster on the full-out version.  Regardless, it took 08:34:50.254 on the laptop.  I just let it run overnight last night.  For the record, I am currently starting endlos with:

 /opt/java6/bin/java -Dsun.java2d.opengl=true -cp ./endlos.jar de.dentaku.endlos.Endlos &

That's a very cool fractal.  I'd love to have that as my wallpaper, but I can't imagine how long it would take if the screen size were set to 1400x900 or whatever.  One thing I found interesting is that it started from the vertical borders and worked towards the center.  Somehow I assumed these things were generated from the center point outward, but again I've no clue in that regard

I am going to get 0.15 and I may let it run overnight again to see if that changes anything.

[dentaku2]

Are you running 64 Bit Ubuntu or 32 Bit Ubuntu?

You can run endlos via

Code:

java -jar endlos.jar

instead of explicit classpath and class. This is prefered as the file 'endlos.jar' won't change but the class/package names may change.

The demo "Four Comets" takes about 00:01:30 with 0.15.0 - and less than 30 seconds with 0.16.0 (in development) ... that's a 70% speed gain just by technical code optimization. No tricks or other shortcuts - still brute force! 

[dentaku2]

Code:

2007-01-26 : Version 0.16.0 - speed & effects
---------------------------

CHANGED:
  1. Improved speed of fast (float/double) precision: demo fractal "Four Comets" is 75% faster!
  2. Checkbox 'Anti-Aliasing' removed - see new tabs Rendering and Effects ('blur').
  3. Decreased precision of demo fractal "Sun" from 34 to 22. It's now 30% faster.

FIXED:
  1. Flickering of image during color cycling via mouse. Cycling is also a little faster.

NEW:
  1. Effects tabulator. Post process an image with several filters (Spin, Blur, Emboss, etc.).
  2. Rendering tabulator. Set misc rendering hints for image manipulation.
  3. Improve speed by using the Java parameter "-XX:MaxTenuringThreshold":
  
     java -server -XX:MaxTenuringThreshold=31 -jar endlos.jar
     
     With this parameter, the "Sun" demo fractal is 3% faster with my configuration.

http://endlos.sourceforge.net/

[dentaku2]

Code:

2007-02-02 : Version 0.17.0 - new fractals
---------------------------

CHANGED:
  1. Fractal definition XML file. To migrate your old XML files replace
          <type>Mandelbrot</type>
     by   <class>de.dentaku.endlos.fractals.Mandelbrot</class>
     and equivalent with the Julia type:
          <type>Julia</type>
     by   <class>de.dentaku.endlos.fractals.Julia</class>
  2. Popup menu items "Open Mandelbrot", "Open Julia" replaced by "Open new fractal".
  3. Removed fast float datatype as it just bloats the code without any real benefit.
     (The fast double datatype is good enough.)
  4. Several GUI changes as a preparation to generalize the iterator and introduce
     custom fractals.
     
NEW:
  1. 9 new fractal types available: Barnsley's, Peterson's, Magnetic, Manowar Mandelbrot
     and Julia sets; Spider.
  2. New demo fractal "Spider".

http://endlos.sourceforge.net/

[Nahee_Enterprises]

M. Hilpert (dentaku) wrote:
>
> CHANGED:
>  1. & 2. & 3. & 4.  .....
> NEW:
>  1.  9 new fractal types available:  Barnsley's, Peterson's, Magnetic,
>      Manowar Mandelbrot and Julia sets; Spider.

Some nice changes to your application, especially the addition of those formulae.  The Barnsley has always been one of my favorites.

[dentaku2]

Code:

2007-02-12 : Version 0.18.0
---------------------------

CHANGED: 
  1. Fractal definition XML file. Migration:
     <real>           => <x>
     <imaginary>      => <y>
     <real0>          => <x0>
     <imaginary0>     => <y0>
     <width>/<height> => <imageDimension>
  2. Introduced color formulas (alpha). Instead of just switching between classic and smoothed
     coloring, a list box offers several coloring methods. So far, only one additional test algorithm
     "Function" - look how the colors change when you cycle them with the mouse wheel ... 8-
     
FIXED:
  1. Rendering hints weren't used for post processing image effects.
  2. Progress update now at least after each block line to see progress for block/image draw modes.
  3. Real number of threads (= number of blocks) weren't set in text field which also could lead
     to a progress state of > 100%.

NEW:
  1. Negative exponents supported in fast precision mode (16). So far, negative exponents were treated
     as exponent 0. Now, negative exponents are correctly treated. Thanks to Paul Bourke's site
     http://local.wasp.uwa.edu.au/~pbourke/fractals/mandelpower/ that shows examples.
  2. Demo fractal 'Dubai'.
  3. Move location (up, down, left, right) via buttons (10% per click).
  4. Image dimension presets.

[dentaku2]

endlos 0.19.0

Code:

2007-02-22 : Version 0.19.0 - flexibel formulas
---------------------------

CHANGED:
  1. Fractal definition XML file. Migration:
     <x0> => <custom><cr>
     <y0> => <custom><ci>
  2. Tabular "Fractal" is full generic now, i.e. all fractals are decoupled from the main
     iterator.

FIXED:
  1. Open new fractal didn't use the mouse click location.

NEW:
  1. Parameters of formulas configurable in settings GUI.
  2. Demo fractals "Blossom", "Eye", "Protozoan".
  3. Fractal class "Newton".

[Nahee_Enterprises]

M. Hilpert (dentaku2) wrote:
>
>   CHANGED:  1. & 2.
>   FIXED:  1.
>   NEW:  1. & 2. &  3. Fractal class "Newton".

Glad to see the Newtons added in.   

Thanks for these recent updates!!

[quaestor]

I wasn't sure if this was the right place, since this is an announcement forum, but here goes;
Love endlos!! I always like well-made programs written in Java.
As long as I switch from Beryl to Metacity before I start endlos, I have no problems (so far ) using both -server and -Dsun.java2d.opengl=true

I have two small critiques:
1. Thread Count goes from 4 to 9 with no intermediates
2. On my "short" screen (800px tall), the default layout is a little tight. I couldn't access the child window's bottom border without some serious re-jiggering, which was a problem as nearly all of the property-sheet tabs were hidden. It would help a little if the window list at the bottom of the main window could be hidden (preference?).

Some numbers, all with "idle" system (single runs; i.e. not average):

Default Mandelbrot
priority	threads	time
5	4	:02.330
5	9	:02.648
9	4	:02.231
9	9	:02.205
Dubai Demo
priority	threads	time
5	4	:10.244
5	9	:09.995
9	4	:10.770
9	9	:10.583

Here're my specs, as dentaku seemed curious about people's rigs:

Sun Java 6 (from Ubuntu backports repo), started with
java -server -Xmx768m -XX:+UseBiasedLocking -XX:MaxTenuringThreshold=31 -XX:+UseParNewGC -Dsun.java2d.opengl=true -jar endlos.jar

Hardware
Compaq Presario R3000z CTO (laptop)
AMD Athlon 64 Processor 3400+ (2.2GHz), IM L2 cache, mmx sse sse2 mmxext 3dnowext 3dnow (32-bit OS, though)
1.5G PC2700 DDR RAM
nForce 3 mobo chipset
GeForce4 440 Go 64MB dedicated
15" WXGA LCD screen (1280x800)

Software
Ubuntu 6.10 "Edgy" (x32; mostly stock with latest updates)
Nvidia binary driver (latest available)

Thanks a bunch and keep up the great work!

q

[lycium]

....


I have two small critiques:
1. Thread Count goes from 4 to 9 with no intermediates

i'm guessing he spawns a thread per (regular) anti-aliasing sample; that really shouldn't be the case, particularly as N gets to be 4 or 5 - thread creation and management isn't free, and that N^2 will become a bottleneck very soon...

the app should spawn as many threads as there are (logical) cores, and then do load balancing.

[Nahee_Enterprises]

Ben Alkov (quaestor) wrote:
>
>    I wasn't sure if this was the right place, since this is an
>    announcement forum, but here goes;  Love endlos!!

Greetings, and welcome to this Forum!!   

And yes, Thomas Ludwig (lycium) has done some nice things with his application over the past few months.

Anyway, just thought I would say hello to a new member.
 
 

[dentaku2]

I have two small critiques:
1. Thread Count goes from 4 to 9 with no intermediates
2. On my "short" screen (800px tall), the default layout is a little tight. I couldn't access the child window's bottom border without some serious re-jiggering, which was a problem as nearly all of the property-sheet tabs were hidden. It would help a little if the window list at the bottom of the main window could be hidden (preference?).

Hi! I'm the author of endlos. For your point 1: that's okay. the number of threads need to be the same as the number of blocks the image is divided. It
's easier to tile the image every increase by 1 row and 1 column. Therefore, the number of threads is 1, 4, 9, 16, 25, 36, ...n*n = n^2.

o

oo
oo

ooo
ooo
ooo

...


My first versions had a linear increase (1,2 , 3, 4, 5, 6, ...) but I didn't like the image blocks as they just were horizontal or vertical bars. With the current tiling, the calculated places are better scattered over the screen. Anyway, it doesn't matter much if you use 8 instead of 9 threads. It's also not the best idea to use as many threads as there are cores because in most images, the first finishing thread wouldn't have to do anything as the remaining threads are calculating. It's better to have more threads than cores as thread switching is less expensive than idle cores at the end. It's no big difference to use 9 or 16 or 25 threads during calcualtion - but the more threads you have, the less remaining image part needs to be calculated by less threads than cores. So, having waiting cores at the end of an image is much more expensive than thread switching during the major part of the image. Just test for yourself. And that's the reason why it doesn't really hurt to have a non-linear thread increase. But it looks better! 

For your point 2: so, there are still people out there with screens smaller than 1024x768 ... hmmm. The "the window list at the bottom" is no special feature of mine - it's automatically done by Java's desktop feature. It also looks different depending on the Look&Feel you use. E.g. the Java (metal) look and feel has no such window icon border - it's just iconifies the windows and you can maximize a window that covers those minimized icons.

As you can see on your single core computer there's no real difference between 4 or 9 threads. Dubai takes about 5 seconds on my Athlon X2 3800+ with 4 threads and about 4.3 seconds with 9 threads. (Average on calculating several times).