The paper considers a mathematical model of the behavior of an assembly of N stars. The 'Kepler' Microsoft Windows demo application based of this model enables to perform real-time simulation of star clusters dynamics for N~=2500. Such performance rate is possible through the use of the Intel Integrated Performance Primitives (IPP) library. The paper also estimates the efficiency of the IPP application and provides an example of C-code with the IPP functions calls. Computer-simulated images of the spiral galaxy forming process, as well as the real galaxies photos, are presented.
Post a Comment
All is good, except the use of Winblows for scientific and serious applications.
The images are fuzzy to read, they hurt my eyes and brain. :-(
This makes for a nice change of pace in the types of articles posted to osnews.com I welcome more of the same.
It is not the compression of the images that makes them harder to read. It is because I had to scale them down from 670pix to 600, because they wouldn't fit on our design.
Are you sure your eyes hurt because of the image and not maybe because of the advanced math? ;-D
... thumbnails wouldn't look nice for such an article btw.
> I welcome more of the same.
Tomorrow we have one more such article. 
It's (for me at least) a little difficult to fully understand the advantages of Intel's IPP, but the article does a nice job, very interesting.
I now understand better why some 3D software (3d Studio ! maybe maya 4.5) started to appear more independent from the VGA card performance. But, since it's closed source, I can be wrong and the cause of this independence might be something else.
Eugenia, you seem to have applied a suboptimal scaling algorithm to the text that results in aliasing -- jaggies -- making it hard to read. Next time, try bicubic interpolation.
Ironic that this is mentioned in an article partly about IPP (http://www.intel.com/software/products/ipp/ipp30/), which among other things contains the implementation of some actual interpolated scaling algorithms (afaik only bilinear interpolation, at least when I last used it).
IPP is an extremely helpful toolkit. It covers anything from alpha blending to MPEG encoding, everything making use of Intel's optimized, specialized instruction sets (MMX, SIMD and what not).
It's unfortunate that Intel has not open-sourced what is arguably a potential sales-booster for Intel processors, but at $199 for a commercial license it's not exactly expensive.
>try bicubic interpolation
I use Bilinear Resample + light sharp filter. It is the best with Paint Shop Pro.
Paint Shop Yuk, in my book.
Photoshop does bicubic, with style. As does, I'm sure, The Gimp and probably other imaging apps, like ThumbsPlus and ACDSee. Only with less panache. I would be happy to rescale stuff for you for future articles, should you not have access to these programs.
Are you sure your eyes hurt because of the image and not maybe because of the advanced math? ;-D
Haha, nice catch.. :-P Problem solved by save the images on HD and view in the gThumb by view as 'zoom in' (120% - 150%)..
Anyway, it's nice article, I understand ok around 3/4, but not the math part. ;-)
For all who is really interested, I can send images with good resolution. Connect with me via e-mail.
> It's (for me at least) a little difficult to fully
> understand the advantages of Intel's IPP...
2.5 times performance increase - isn't it impressive? And it is not the limit.
Hi
Nice article - makes a refreshing change! Two questions though...
I can see how the IPP may be useful for multimedia stuff but how much difference does it make for the array operations, over and above using either BLAS/LAPACK or Fortran 90 functions?
2.5 times performance increase - isn't it impressive? And it is not the limit.
That is very impressive. A few quick questions. Is te IPP an exclusively C library only? By that I mean that it has only C functions, with no C++ stuff there.
Also, does it work on other compilers, apart from hte Intel C++?
1. Reference 4 should be replaced to the 2nd position. References 2 and 3 - to 3 and 4 respectively.
2. The symbol '(R)' (registered trademark) is shown as '?'.
Try to answer some questions. IPP library has been written in C and Assembler. It contains a number of static and dynamic modules (.lib and .dll) which can be used with ANY C/C++ compiler without problems. Of course, you may use IPP with Fortran, if you write interfaces to necessary C-functions, as usual. But I'd like to suggest another Intel product - MKL (Math Kernel Library) http://www.intel.com/software/products/mkl/mkl52/ . It already contains Fortran interfaces for all functions.
2.5 times performance increase - isn't it impressive? And it is not the limit.
__________________________________
The IPP library is very effective for processing data arrays, such as vectors, matrices, images, etc.
...
the IPP library is to simplify code writing. IPP includes functions that are often used in different calculation regions, such as vector and matrix algebra, image development, etc. To avoid bulky code writing, for example, multiplication of two matrixes, a developer may write only one line of code, which is a call to the corresponding library function.
Yes, thanks. (I am only an "afficionado", not a real hard core programer like you Mister Abrosimov).
I understood that (in relation to my old and trusty dual PIII 1Ghz).
The difficult part for me was the implications in real life of IPP (MMX, SIMD are proprietary - AMD has 3DNow proprietary instructions). 2.5 increase is very good (beats the competition), the cpu chip seems to already have this lib half (if not the full instruction without floating point calc) computed on die, or so it seems, you just have to call it to calculate the rest. This is the difficult part for me to understand.
It contains a number of static and dynamic modules (.lib and .dll) which can be used with ANY C/C++ compiler without problems.
It would be intersting to know if it's a library for windows only (.dll .lib) or if one can use it on Linux too ... with the C, C++ Intel compiler for Linux (seems like you say it can be used with gcc++ too).
There are a number of applications on Linux that could be compiled optimized to use IPP on Pentium processors (i586). The first that come to mind are Blender and Moonlight and some audio/video Linux applictions (mpeg2 ...), maybe even GIMP could benefit on some "GIMP extras", mostly written in Perl, as far as i know 
If you want to write Russian, send me e-mail. It seems this site doesn't support cyrillic.
If you read IPP web page attentively, you'll find that two IPP packages exist - for Windows and for Linux. Functionality is the same. Any compiler as well.
Wonder how my neural network will perform with this library.
Guess I'll download the trial and spend the next few weeks tuning and see.
Dima, It was a decent joke about the first page with math/phys. equations, and how it would be accepted on this site :-). Euginia has moderated my post down anyway, I think it did not match the design of the site =)
mmmm,, advanced math,, *thinks* *smoke comes out of ears* AGHHHHHHHHHHGHH!!!! BBBBBOOOOOOOOOMMMMM! *head explodes*
What is the point of a "real time" simulation of a star cluster? Who has time to wait billions of years for the results?
>> What is the point of a "real time" simulation of a star
>> cluster? Who has time to wait billions of years for the
>> results?
'Real time' means more than 20 frames (or integration steps in our case) per second. So we may see the galaxy evolution as a movie (many times accelerated, of course). I thought it's clear.
So....when do we get a Stargate?
7 points, 6 points zero in on where you want to go, the 7th is where you are from.
At least...that's how Jackson explained it.
Jack O'Neill.