Finished the final touches on my MSE paper and uploaded it to their server. I already faxed over the copyright form, so I should be good to go to head out to Anaheim in June. Pretty sweet. I’ve got postscript and pdf versions of the paper ready to go up on my work page whenever I decide it’s appropriate to be publicly available.
The conference should prove interesting seeing as it falls smack between the abstract and paper deadlines for MICRO… Hopefully I’ll be in a position to worry about it at that point instead of still debugging the local data caches and getting more benchmarks up and running. I should have two undergrads reporting exclusively to me as of next Monday, and in theory Jeff’s undergrad will write a benchmark or two as well. Guess I should figure out what benchmarks they’re all going to write.
Oh, that reminds me. I have 1, 2, 4, and 8 programmable cluster versions of DNA up and running from compiled C code. The 2 and the 4 actually see a significant performance improvement over the 1 after I bumped up the input data to a reasonable size. Haven’t quite figured out why the 8 isn’t performing as well. I suppose it could be the limited memory bandwidth of the system (similar to the old assembly version of dither, where going from 4 to 8 main cache banks resulted in significant performance improvement). Will have to check that out.