Projects

Well I now work on SCADA projects for my company, as a software engineer. SCADA is Supervisory Control Data Acquisition. Its a system involving RTUs (Remote Terminal Units), in the "field"..in this case, an oil well, and they collect data and store in a real time database which is accessed by software in the form of a Human Machine Interface (HMI). More on SCADA can be found here. So I write extensions for our SCADA software, and also custom programming, as a Software Engineer.

Now following some conversations I have had, I have come up with come crazy stuff...which I plan on implementing on my own time. By crazy stuff, I mean crazy applications. Not your conventional web thingies...ewww. I have always had a soft spot for Artificial Intelligence(AI). I created a checkers game with the minimax algorithm which played against a user using a search tree of depth 1....so it was basically a step ahead..so from watching sci fi it would be cool to do something with voice activated search or programs....then we could go to voice recognition..so, instead of typing in instructions, you speak to the PC...and it handles the rest. Imagine being able to say stuff like..."Hey, google this word..." and all of a sudden, it happens...so anyways, a lot of this is in my head..

1. The GENIE project.....its an implementation of what I described above. I call it GENIE after aladdin..all he had to do was wish, and the genie took care of the rest..lol...my vision for this is like some platform...and yeah one can take off from there.

Lets see, what tools do I need?

1. A voice API (preferably for JAVA)

2. An arbiter, which decodes the voice instructions, maps them to some opcode and then executes the instruction corresponding to the particular opcode.

3. This is going to involve some hard core research (groan) into this field....

Deadline: A beta version (October next year). At least implement part of the system.
Now I have this written down, and gotten it out of me. Lets see how it works, let me sleep over it...lol

7 months.

I have not updated my geek blog in 7 months. So whats been happening?
I finally finished my thesis titled "Hardware Acceleration of the x264 H264 Encoder". I ported the x264 encoder to run on a reconfigurable platform (FPGA). I had to determine which parts of the code took up the most CPU cycles, and then replace those parts with calls to the FPGA. The calls to the FPGA were by means of the MOLEN computing paradigm which is part of the Delft WorkBench Toolchain, (A set of tools for preparing applications for running on FPGA's). I ended up getting speeddown instead of speedup because the calls to the FPGA were done using by means of OS calls. Running the application without OS calls did get speedup. Unfortunately, I had issues with the tools and ish, so I ended up with a grade 7/10... *sigh...no distinction, but at least I am done.
I played around with ruby on rails, using InstantRails, got bored, now I am delving into the cracking world using SoftICE...using this tutorial.
I guess I will still play around with this some more before I start my new job.
Later!

X264

My head aches,and I feel weak. Why? Its a little (Yeah right) H.264 Encoder written in c, called the x264. Now whats my bizness with it? Its part of my MSc thesis project on Hardware Acceleration.
I have to pick the computationally intensive parts of the code and send them to an FPGA, and investigate if theres speed up.I gotta make use of a C to VHDL compiler written in house.
Sounds easy right?
Yawwwnnnnn. My first assignment was to see if I could serialize the application, by removing the pthreads. Then I had to do some profiling. The call graphs generated by this thing were so damn huge.I mean this is an encoder software, and its opensource....gosh some people really have time.
Now I gotta pick the functions which only operate on macroblocks, and use them to estimate area and delay on an FPGA. Some of my friends think I am losing it...lets just see.
Try stepping through a large piece of software like that, and see how you feel...but its interesting anyways, Thank God for doxygen, which allowed me to generate documentation for the whole encoder.
For the estimates, I am considering the hardware arithmetic units which will be used by the functions. So if theres a multiplication operation, followed by addition, I know that I will need an multiplier and an adder. I will have to generate some data flow graphs, and maybe perform ASAP scheduling for resources, and add an extra 20% for routing and logic.
Now let me get back to picking out my functions....

Maven Maven and Maven

I have been battling with installing maven for my new project....now this thing has been showing me hell. I downloaded all the stuff, and the atlassian pdk.
Now I followed all the instructions they had on their page, for some reason, they prefered to use jdk 1.4, but for me, that didnt matter anyways.
So I did all the exporting environmental variables, got the thing to work, created the maven archetype(Directory skeleton for a particular kind of project).
Now I had a full screen(Oh did I mention I got a new phillips flat screen yesterday?), filled with weird looking linux commmands...then since I Wasnt going to kill myself, I exported the whole thing into my eclipse IDE with the command eclipse:eclipse.......this frigging download took 29 mins approximately. When I was rejoicing that it had finally worked, came the hard part, getting eclipse to compile and build this thing.
Maven uses a test/development life cycle method. The first stage is compile, then some tests, and then an install. These stages are called "goals" in maven terminology.
I wasnt making any progress, so I installed the maven plug in for eclipse. Now I began suffering because for maven to compile, it needs to download all the dependencies from the local maven repository(29 minute download...remember?), and if it doesn't see what it wants, it begins to bitch like a starving cat.
I got so many errors. imagine getting an error called mojo:ActionMojo:execute, or some crap like that...reminds me of a friend in school who saw a line of code commented as "suicide".
Now I have been trying to get this to work all day with no avail. Now its hinting that one of the POMS for the dependencies its trying to download is faulty. Mavens proponents say its better than ant, but for some reason, I find that doubtful.
Anyways, back to the grind...this thing must and will work, if I have to beat it with a stick...

In case anyone sees this, this is an example of some of the crap eclipse feeds me:

2/19/08 4:27:46 PM CET: [WARN] POM for 'seraph:seraph:pom:0.7.17:provided' is invalid. It will be ignored for artifact resolution. Reason: Not a v4.0.0 POM. for project seraph:seraph at /home/ade/.m2/repository/seraph/seraph/0.7.17/seraph-0.7.17.pom
2/19/08 4:27:50 PM CET: [WARN] POM for 'oscore:oscore:pom:2.1.0:provided' is invalid. It will be ignored for artifact resolution. Reason: Not a v4.0.0 POM. for project oscore:oscore at /home/ade/.m2/repository/oscore/oscore/2.1.0/oscore-2.1.0.pom
2/19/08 4:27:50 PM CET: [WARN] POM for 'cargo-test-runner:cargo-test-runner:pom:8Aug07:provided' is invalid. It will be ignored for artifact resolution. Reason: Not a v4.0.0 POM. for project cargo-test-runner:cargo-test-runner at /home/ade/.m2/repository/cargo-test-runner/cargo-test-runner/8Aug07/cargo-test-runner-8Aug07.pom
2/19/08 4:28:14 PM CET: [WARN] POM for 'seraph:seraph:pom:0.7.17:provided' is invalid. It will be ignored for artifact resolution. Reason: Not a v4.0.0 POM. for project seraph:seraph at /home/ade/.m2/repository/seraph/seraph/0.7.17/seraph-0.7.17.pom
2/19/08 4:28:25 PM CET: [WARN] POM for 'oscore:oscore:pom:2.1.0:provided' is invalid. It will be ignored for artifact resolution. Reason: Not a v4.0.0 POM. for project oscore:oscore at /home/ade/.m2/repository/oscore/oscore/2.1.0/oscore-2.1.0.pom
2/19/08 4:28:27 PM CET: [WARN] POM for 'cargo-test-runner:cargo-test-runner:pom:8Aug07:provided' is invalid. It will be ignored for artifact resolution. Reason: Not a v4.0.0 POM. for project cargo-test-runner:cargo-test-runner at /home/ade/.m2/repository/cargo-test-runner/cargo-test-runner/8Aug07/cargo-test-runner-8Aug07.pom
2/19/08 4:28:30 PM CET: Parsing error /home/ade/.m2/repository/seraph/seraph/0.7.17/seraph-0.7.17.pom; org.codehaus.plexus.util.xml.pull.XmlPullParserException: Unrecognised tag: 'logo' (position: START_TAG seen ...\n ... @9:11)
2/19/08 4:28:30 PM CET: Unable to read search-portlet/pom.xml; org.eclipse.core.runtime.CoreException: Parsing error /home/ade/.m2/repository/seraph/seraph/0.7.17/seraph-0.7.17.pom; org.codehaus.plexus.util.xml.pull.XmlPullParserException: Unrecognised tag: 'logo' (position: START_TAG seen ...\n ... @9:11)

Do I sound pissed off? You bet I am.... :-)

so far...since last year

Havent posted in a long while. So this is gonna be a bit messy.
I am working on my thesis in Hardware Acceleration. Basically you have an application, and you are looking for ways in which it can run faster in hardware. So you profile the code using tools like gnuprof, and then you can detect which functions in the code are the most computationally intensive. You then have to merge the functions, because you want to reduce the amount of times you reconfigure your FPGA when you run the code, and also by merging, you should achieve some speedup.
Now, the thing is that merging isnt so easy. You have to consider some criteria, i.e the number of times the function is called, and also, the data dependencies which might exist between the functions. We have a tool that can generate control flow graphs (CFGs), which show the rest.
The application I am using is the h264 decoder, its open source, written in c, and I am going to write about it as a chapter in my thesis. I also had to remove the pthreads it had, to make it serial, and do some profiling. That included some call graph generation by the way.
On the other side, my web development skills are improving. I finished the tag search extension to the LDAS application (a web based Java/LDAP application) given to me, and now I am going to work on developing a JIRA portlet, as an interface to opensearch.
I wonder if its possible to run two different coding interests at the same time, but It should be.
And for the assembly assignment, I got a grade of 8.5, and I got a 9 as a final mark for compiler construction. Yup......all that reading of the grammar files, and generating assembly code paid off......we got the highest mark in the lab...68/70.
Cheers, and later!

I have just spent most of today (sunday) working on my assembly programming assignment..its finished by the way.....lol, I just have to write another version of the code that runs faster. It was just a matrix multiplication code in C that we had to rewrite in MIPS 64 Assembly. The hardest part for me was knowing how to calculate the index in a 2-d array in assembly. I finally figured it out, so check this out:

consider a simple 2-d array a[i][j]. To get the element at this location, we simply use the formula:

(i * max_index_of_i +j)<<3.

We need to shift by 3, because in assembly, we are just stepping through an array, and each element occupies 8 bits.
So this principle is shown in MIPS64 assembly code to reference p[i][k] in a 3x3 array.

daddi $t7, $r0, 3 ; $t7=3

dmul $t6, $t0, $t7 ; $t6=i*3

dadd $t6 , $t6, $t2 ; i*3 +k


dsll $t4, $t6, 3 ; $t4=(i*3+k)<<3 multiply by 3

daddi $t8, $r0, p ; get base address of p

dadd $t8, $t8,$t4 ; $t8 has the address of p[i][k]

lw $t9, 0($t8) ; $t9 has the value of p[i][k]

Now I am getting back to debugging. See you later!

We finally finished the first assignment of the compiler construction project, which was to make the reference compiler recognize records. We were able to compile the test file, record.ast, and it generated the right c-code. Today, we simply ported the code from the old compiler to the new one. The new one had some new assert statements, to check if the right subtype was being used, and all, but it didn't take so long.
We now began the testing phase. We have this Java applet called diagnostix which is able to evaluate the quality of our test suite, and it showed us that our test cases detected 3/5 bugs the compiler should guard against...not bad for our first day of testing, and we are way ahead of schedule.
An interesting case we had was checking that a record name could not be used as a variable name, and also checking that it is illegal in our compiler to check for equality of two records.
Later, thats all for today.