State of the Cell

Originally posted at http://cs.ucsb.edu/~cgb/stateOfTheCell.html, and thus, looks much better there.

The Cell Broadband Engine Architecture (which we shall refer to as simply the Cell architecture) was designed as a compromise between the general-purpose but slower CPU and the specific-purpose and faster GPU. It is a heterogeneous architecture: it contains processing units that specialize in different tasks. However, critics (and even some fans) of the Cell architecture claim that it is incredibly difficult to produce good, fast code on it. Having spent the last quarter working with the Cell architecture, we agree with this sentiment. But why?

Multigrid Methods with CUDA

Originally posted at http://cs.ucsb.edu/~cgb/multigridCUDA.html, and thus, looks much better there.

Over the last quarter I have been investigating how to implement a multigrid algorithm with CUDA, a new language aimed at making General Purpose Computation on GPUs (GPGPU) easier. This web page documents this experience.

My Favorite Error Message Ever

While running a Poisson Equation Finite Difference approximation algorithm (whew, a mouthful there!) today on Cell, I got the greatest error message ever:

[cgb@cell pfd]$ ./withRapidMind
terminate called after throwing an instance of ‘rapidmind::Exception’
what(): The type ‘double’ is not supported by the Cell backend.
Aborted

What the hell indeed. Much thanks to the RapidMind guys for an awesome function called “what” and not supporting the double type (although I’m sure there’s a long explanation about that I haven’t found yet).

Setting up Linux on the Cell (Part 2)

So…Yellow Dog Linux doesn’t like SDK 2.1 very much. Turns out I must have done something wrong while trying to install it because it only wants to work on Fedora and I fucked up my ability to compile code for the SPEs and PPE while doing so. So I reinstalled Yellow Dog and this time, followed the very simple explanation at the RapidMind website which said just to install the libspe2 rpm and not mess around with anything else. It worked great, but let’s step back a second here.

Setting up Linux on the Cell (Part 1)

So we got a PS3 for research in the lab the other day and we got to set up Linux on it. It’s actually pretty painless, although if you can, get a monitor that supports DVI and supports HDCP so you can get the digital connection. We only have analog in the lab so it looks like crap. Seriously though, it looks like feces and I have to use SSH to get anything done. That being said, here’s how to do it:

“Portable” Package Management is Bullshit {Mostly}

I don’t always want to code in Java. In fact, I’ve recently wanted to not program in Java and get a feel for everything else that’s neat and nifty and blah blah blah. But Java spoiled me. I got used to having my libraries distributed with my package, and having it work on other systems pretty easily. So imagine the fun times I had trying to get MySQL and SMTP support working in other languages…

Bloom Filters

I’ve been getting more into my programming roots lately, and after I read this CodeKata, I decided to give it a try. The idea is to write a Bloom Filter, fiddle with the parameters that make it what it is, and test its performance. That being said, let’s look at it.

HOWTO: Converting Simple OpenGL Code from C++ to Perl

Over the last few weeks I’ve been trying to learn OpenGL. However, I’ve had a miserable time trying to get various implementations of it to work on my Mac, notably with C++ and Java. I was surprised to learn that installing the Perl OpenGL implementation (or POGL) is pretty simple, and that converting OpenGL code written in C++ to Perl is also reasonably easy. The following is a heuristic method you can use to do the same, using an example from the OpenGL SuperBible by Richard S. Wright, Jr. The code is freely available from his site at http://www.starstonesoftware.com/OpenGL/, and feel free to use my Perl conversion of his code.

HOWTO: Call Matlab from Java {Kinda}

Looking over some of the reports on my blog reveal that pretty much everyone to this point looking at this blog look at the Java / Matlab post and not much else. Therefore, here’s everything I’ve learned concretely on how to call Matlab from Java and what NOT to do to save you some grief:

What I did:
Make a file that contains all the commands you want to run when MATLAB starts. My script file looks like this (named scriptName):

cd /directoryWithMATLABScript/
myMATLABScript
quit

The Java code I have used is a modified version of the code here. Now, use the following Java code to call MATLAB with the commands from your script:

Calling Matlab from Java: Now 20% Less Painful!

UPDATE: The follow-up to this article can be found here.

Normally I’m not one to rant about programming or computer science stuff here (despite that degree I have in it), as I like to keep this more about philosophical crap (which I don’t have a degree in). However, I’ve spent quite a while on a research project and had a couple problems that all boil down to one thing: Matlab’s Java interface sucks. Bad.

If you don’t care for Computer Science stuff or when someone says “Java”, you think “Coffee”, this probably isn’t for you. In fact, it’s probably only really for me to bitch about randomly since I’ve wasted so much time on this.