ndesaulniers

Many great memories of KOTOR, Jade Empire, Mass Effect. RIP

This is a great high level overview of some the techniques and problems encountered with developing a register allocator!

LLVM 's FastRegAlloc also uses linear scan.

> Grok 4.5 solved some issues setting up a buildroot system that Fable 5 couldn't do.

Lol, probably because Tesla's software stack is buildroot based. I'll bet that was in the training data.

> The important thing is really who's on the girl's soccer team

I've not heard this expression before; can someone explain it to me?

Definitely. I wasn't aware about constant time constraints at the time.

I was working on reimplementing the WebRTC stack in JavaScript a long time ago, never finished. The crypto folks at Mozilla (ekr@) suggested "maybe don't do that for dTLS."

I encourage folks with a PlayStation 5 (or 4) to participate in the blackout by unplugging their console for the week of August 23 at 7:00 PM your local time and run to August 31 at 7:00 PM your local time. Hopefully that helps send a message. We should irrevocably own our games, not rent a revokable license. And we should have price competition from multiple store fronts.

Ah, I'm behind the times.

Yeah, guess I should have used the past tense there.

Neat. XLA predates MLIR. Interesting stories there. You'd have to stop by the Bay Area LLVM monthly meetup to hear them. :-X

> So if XLA already uses LLVM, why is our approach different?

Uses MLIR, XLA does not.

> So… what is the point?

> Honestly? We aren’t entirely sure yet.

> Let me be perfectly clear: this is not going to beat XLA for standard deep learning workloads. XLA has years of hyper-specific optimizations for linear algebra on GPUs and TPUs. If you are training a massive transformer, stick to standard JAX.

> But what we do think is cool is what happens when you connect JAX directly to the broader LLVM ecosystem and drop the heavy XLA runtime. (Plus, no need to build XLA using Bazel either! You’re welcome.)