Chris Leary

200 Followers
95 Following
266 Posts
computers/compilers/accelerators, mostly · projects! XLS http://bit.ly/xlsynth TLBHit http://tlbh.it · prev: XLA ML compiler, JAX, accelerator co-design, dynlang JITs
@regehr @lenary got it, IIRC the peephole work was more “find existence of an improvement and generalize as much as possible” which I imagine is a totally different angle, thanks for the correction, I remember thinking it was super cool!
@regehr I’m not an expert so if this is too ignorant feel free to just ignore, but I thought with CEGIS kind of techniques LLMs just fill in a “previously missing piece” of how to reasonably pick something to inject to make progress — template based injection was very bespoke and domain specific.
☕️ LATTE, our little workshop on hardware design languages/compilers/etc., has 24 (!) rad-looking position papers this year. It’s on Monday, and you can attend on Zoom or in Pittsburgh: https://capra.cs.cornell.edu/latte26/
LATTE ’26

@sayrer great game!

One of the cool things about our line of work is sometimes you learn a technique and really feel like you’ve “leveled up” in the video game sense.

I remember one time a colleague/friend came in to work and excitedly said “I can see in the SWAR dimension now”. Stuff like that rocks. 🤘

@andrewrk @pervognsen Yeah it was a close race, In Silico being a bit less polished makes it feel more "punk rock ethos" but "Encoder" sealed the deal for me. Thanks for the replies and glad everybody jumped on this thread, cool to see a bunch of folks from the community in the same boat. 😂
@andrewrk @pervognsen "Out of curiosity" -- this is your brain on AOL Instant Messenger
@andrewrk @pervognsen OOC favorite pendulum album? I think I settled on Immersion. I like how different Knife Party is too…

One of my favorite things about Rust, after properly using it for about two years now, is that it puts what I think is “the correct amount” of annoyance on using polymorphism (in its various forms).

Balling up function pointers in a struct in C days was maybe a bit too high, nice middle ground vs the languages that leaned into OOP fanaticism, and monomorphization being clean and easy is a great default people will naturally lean towards for systems programming.

De gustibus non est disputandum tho

Abstract org dynamics thought experiment: as people become arbitrarily careful things get missed that are /intuitively/ quite clear but nobody has the time to spend on the analysis to confirm beyond “background-radiation-level of doubt” that is applied to any proposition.

“Background doubt radiation” is interesting. You don’t want it to be zero, but it being high is stifling. You want it to be roughly proportional to how reversible/costly a decision is, with bias towards action so you collect empirical data. Cost is not absolute but evaluated vs an alternative (which could be the result of inaction). Mostly I’ve seen people mitigate its impact by having clear notions of being responsible for what you put in place, but often ownership and responsibility is more inherently diffuse. For example, teams often own code bases collectively and there is the “haunted graveyard” effect for a component where nobody actively champions the functionality/purpose.

Must be a theory of “satisficing for orgs” this ties to…