Projects Galore
April 7, 2004
research
college
So I’ve been going nuts working on class projects this semester. I’m trying to wrap up the last of my class work and am currently taking a couple of ECE497 courses (special topics grad courses where professors largely discuss their own research interests).
Nick’s teaching one on Unconventional Computer Architecture that’s been pretty interesting so far. Project’s coming along pretty well for that and most of my responsibilities have been finished. We’re doing hyperblock formation (in SUIF) supplemented by path profiling (in amalsim) combined with a basic run-time reconfiguration simulator that should estimate performance benefits of swapping hyperblocks out to reconfigurable logic. I wrote up the path profiler stuff which has all been checked in and barring any more surprises in SUIF, should be good to go.
Matt Frank (of RAW fame) is teaching one on Multi-threaded Computer Architecture. It’s been really good. Sort of a good overview of how parallelizing compilers work, how the corresponding hardware works, without getting into really icky details (like CS426). The project is a big question mark at this point. We’re attempting to make the GSA-PDG actually work for real programs (lookin’ pretty scary), then somehow resequentializing it after we’ve got it, and of course do some simulations on it for some basic limit study sort of thing. I’ve just wrapped up my first pass at a simulator. It’s mostly there at this point (reads in Jeff’s GSA description, generates the internal graph representation, and then does some of the functional simulation of that graph in an event-driven logic simulator sense). I don’t have all the operations supported at this point… like I’m not sure what a CVT is, I have no idea how the memory operations should realistically work, and have absolutely no feel for what’s going on with the GSA-specific nodes. If I dig out the GSA papers again, I might stand a chance at those, but since I’m working with a moving target at this point, I’m not going to work too much more on it until I see what Brian and Jeff have been up to…