[HiPerCoPS] Multiplier Computation Process

Here's the process for setting up and calculating a result with a multiplier:

1. Create the multiplier
   
2. Specify the operands
   
3. Perform the computation
4. Return the result

Of course, this is hugely generalized. I'll write more about the details of each step in separate posts.

[HiPerCoPS] Cell Interconnect

I really want to find a simple and clean way to connect the cells of a functional unit together. One option is to implement each cell with eight different pointers and connect the cells directly with these pointers. Connecting these pointers to an intermediate class is another possibility. This could help if I ever need to model delay, pipelining, or connection faults, but adds complexity to the overall system. Maybe the cells could account for delay, pipelining, and faults just as easily. I'll have to think about it more.

[HiPerCoPS] SVN Repository Up

Today I figured out how to connect my NetBeans IDE to the Google Code SubVersion repository for the project. Now you can browse through the latest version of my code, which at the moment only includes the multiplier algorithm proof of concept.

[HiPerCoPS] Goals for 2/19/08

1. Implement multiplier cell types A-F at the cell level.
2. Establish inter-cell connections.
3. Perform multiplications.
   

[HiPerCoPS] Lowest Level of Abstraction

Today I met with Dr. Delgado, and we decided that the Cell will be the lowest level of abstraction for my simulator software. This is good because it will allow me to take a few shortcuts to speed up the development process. For example, if we included the Element level I would have to spend a lot of time implementing single bit operations. Of course, this might create some problems when we try to branch out of the math mode realm and implement some memory units. I think it will be alright, though, since my software architecture can differentiate between math cells and memory cells.

[HiPerCoPS] Multiplier Proof of Concept

Yesterday I finished a proof of concept for my multiplier algorithms. The program takes n, m, and the multiplier type as inputs and prints out the appropriate multiplier layout. I was able to use the program to produce a 5000x5000 bit multiplier in about 1min30s.