15-740 Project Milestone

An-Cheng Huang <pach@cs.cmu.edu> Leejay Wu <lw2j@cs.cmu.edu>

November 10, MCMXCIX

Major Changes

There have been no major changes since the proposal.

Accomplishments So Far

We have implemented quasi-sequential single stream-buffer next-N-line instruction prefetching, as a patch to the out-of-order SimpleScalar simulator.

We have built part of the OOCSB suite of C++ benchmarks for SimpleScalar.

We have spent an inordinate amount of CPU time running benchmarks, as can be seen on our web page.

Meeting Our Milestone

We have essentially achieved our milestone.

Surprises

We did note that several of the OOCSB benchmarks appear to be incompatible with SimpleScalar 3.0 due to unimplemented system calls.

We did not expect applu to require approximately 1.5 days per run. It's still going.

Revised Schedule

We may shift our focus to optimizing the prefetcher, perhaps not requesting purely linearly...

Resources Needed

Arguably, our benchmarks would go much more quickly with a few quad-Xeons at our disposable. We regret to say that this is rather unrealistic, and that in reality we don't really need that amount of CPU power.