My goals this week are
- “Finish” the Glenside arxiv draft. Tentatively I think this means re-drafting the intro and figuring out whether I need an overview. There’ll still be much to be done, but I want to get the paper out.
- Begin my LATTE draft.
- Read MAESTRO paper. Related to multiple systolic arrays.
I have a bunch of threads running right now.
- Writing a paper about Glenside to put on arxiv.
- Related: looking into Glenside as a tool for tensor layout exploration.
- Planning a submission to LATTE.
- Trying to understand how multiple systolic arrays work together, and when they make sense.
- Eyeriss, MAESTRO, PhD thesis from the Stanford
Some of my time (probably a good amount) will be taken up by internship interview practice, too.
I also had a great meeting with Jonathan Balkind, Deeksha Dangwal, and Chris Zhang from UCSB’s ArchLab. Jon had seen a tweet of mine from a few months ago and reached out, as we have similar research interests.
We specifically talked about (note that this is through my lens!) ways to evaluate designs without expensive things like full RTL simulation. Perhaps more generally, we talked about architectural modeling, and what options exist beyond expensive cycle-accurate simulation. Deeksha’s paper, Trace Wringing for Program Trace Privacy, had specifically been on his mind; he thinks that heatmaps are a good example of ways to boil down workloads into a “lower dimensional” space.
Reflection on Goals
Glenside draft is not finished, though I did get a bunch of work done on it. I don’t know how to pace myself when writing a paper; in fact, I’m not sure how to set goals with most things. There’s the Eisenhower quote that Zach loves:
Plans are worthless, but planning is everything.
In some sense, the act of planning out my week is the good thing; hitting the goals at the end of the week shouldn’t be seen as a necessity. Still, it’d be useful to be able to predict when I might be done with something, working at a normal pace.
I did begin the LATTE draft, and have some good feelings about it. Tangentially related, I’ve been having great discussions with Luis the past few meetings. It took me a while to feel comfortable having research conversations with him, or with anyone; perhaps it just took me a while to feel like I knew enough to speak, or to feel like I knew enough to know I’ll never know enough, but never will anyone else.
Did not read the MAESTRO paper. That task got swallowed up when I needed to study for an internship interview. Will do that next week. It’s lower priority anyway.