Notices by William D. Jones (cr1901@mastodon.social)

@dalias > Is this an honest shitpost or a troll?

I don't know actually. The subtext is: "I think it's great we critique (putting it mildly) AI as a way to abdicate ourselves from caring about the lower levels. I wish we were more considerate like that with lots of other software tools we used, and I'm picking on compilers b/c I'm bitter that I could spend the rest of my life trying to understand all of e.g. LLVM/GCC but I physically couldn't/can't."

@dalias To be clear, I think Servo accepting AI contributions is a bad idea and the slippery slope ("it's only minor contributions") is not a fallacy here.

The increase in contributors is likely not going to be from a pool of people who care about their work or want to understand what's going on. I've seen the drive-by AI PRs for ppl looking to pad their GH with contributions.

What makes compilers different from AI? They're both abstracting away the need to understand what's going on; even further than the lies macroarch (via microarch), and transistor models (via quantum mechanics) are already telling you.

@atax1a *Looks up rlwinm semantics*

Wow, that's a mouthful :o... what an interesting insn...

@pkhuong @mcc What's special about 16-bit partial deps that doesn't apply to 8-bit ones?

@pkhuong @mcc What makes 8 bit ops faster, even tho both will have partial deps?

@joe @steve @pkhuong Ya know, if an extra xor/movsx/movsxd is less performance penalty than eating a partial load, I'm beginning to realize why RISC-V zero/sign-extends even byte loads...

Today I watch a friend discover the horrors of gas and AT&T syntax... :o

I have released Sentinel v0.1.0-beta. For those not following along (this is a new thread for reach), Sentinel is my microcoded RISC-V CPU core in ~1000 LUTs on an ice40 FPGA. It implements RV32I_Zicsr and the Machine Mode privileged spec, and passes the RISCOF and RISC-V Formal test suites (as of end of 2023- needs updating).

I rewrote the README.md/quickstart. So if you have git and Python 3 installed, you can have a generated Verilog RISC-V core in 5-10 mins.

https://github.com/cr1901/sentinel/tree/v0.1.0-beta

@steve Hi, I need my memory jogged... a while back, we had a conversation where you mentioned someone did a PhD thesis on "a rounding/compliance test suite for floating point arithmetic". And it's a decent read for anyone who might ever want to make a software FP library. Does this ring a bell?

I could look on Twitter, but... your posts are gone :P.

@gsuberland Trumpet Winsock is magic. DOS TSR that somehow hooks into 3.1 to provide a stack. I have to see if I can find my copy. I think the author's dead, so if I want to know how it works, I have to RE it :(...

@pinskia What the f***? :o

#lazyweb Is a "Mailbox" in SoC parlance a cute term for two spsc queues (in both directions)?

@dalias @ariadne Is "Bunday" like Caturday, 'cept it's Sunday and for bunnies? :D

Wonder what happens if you forcefully add forward-slash to the on-disk repr of a POSIX filename (via hex edit).

Probably "nothing good"...

Sentinel is my microcoded RISC-V CPU written in Amaranth. A release (v0.1.0-alpha.2) is likely soon. I have written a lot of docs for this release, including a Quick Start.

The Quick Start is meant to be doable with only Python and pip(x) installed initially. Would anybody be willing to test out the quick start on their own machines, and report back any issues?

Use the "pdm install -G examples -G yowasp; pdm run use-yowasp" code snippet if you don't have yosys:

https://sentinel-cpu.readthedocs.io/en/latest/usage/quickstart.html#quick-start

@Polychrome @mcc Isn't this the episode with a cat who ate a bomb (or something like that), and the Dirty Pair back the wrong cat at the end?

@dalias @kenshirriff Well, @steve deleted his hellsite acct, so context is missing.

https://x.com/cr1901/status/1601969347899133954

But according to my FF History, the correct way to do FP trig functions is arg reduce to +/-pi/4:

https://userpages.cs.umbc.edu/phatak/645/supl/Ng-ArgReduction.pdf

Then do a polynomial approx. Except I remember Steve saying the polynomial shouldn't be a Taylor series for some reason (slow convergence?).

According to @kenshirriff, the original 8087 uses CORDIC. With all the rounding bullshit that FP arith has to do it surprises me that it's a good fit.

But I also don't understand it well.

At some point, I'll learn CORDIC. But part of me doesn't want to because I don't want the magic ruined. Kinda like Rubik's cubes.

It feels like like a cheat code to calculate complex functions. Like, it shouldn't work at all.