Rust wgpu now has a
-
@amarioguy @mcc I mean I feel about modern GPUs and their intentionally prohibitive feature span the way I do about web standards: "that's a nice miracle you destroyed the ecosystem with. Undoing the damage is probably a multi-decade job and none of the normal folk will understand why feature reductions are needed. Won't stop me from patiently pushing in the right direction, though."
@MaddieM4 @amarioguy @mcc i would also consider that this is exactly how things start. a lot of open source pillars started as someone's cool project
-
@mcc @decay @amarioguy @MaddieM4 we'd keep hyperthreading, too
we just don't think management engines have any place heh
@ireneista @mcc @decay @MaddieM4 one thing i appreciate about Apple's SEP (despite yes, being from Apple and a closed, black-box so still not gonna say that anyone should just blindly trust it) is that the threat model they went for it was mutual distrust, where both the main CPU and the SEP core distrust each other, so both isolate their memory from each other
for context, both ME/PSP are modeled such that they own the platform entirely afaict, so if you get code exec in the ME or PSP, you own the whole platform, while in Apple's case, owning the SEP these days does not let you own the AP entirely, you still have to overcome most the other normal hurdles and can't just trivially DMA into host memory
that said, imo security processors including the SEP should not be unauditable black boxes at all, there's too much at stake for that to imo be a truly trustworthy system...
-
@mcc @decay @amarioguy @MaddieM4 vector units are one of those "sigh .... yes" things, as well. frustrating complexity, but important, and at least it's self-contained.
@ireneista @decay @amarioguy @MaddieM4 i suspect vector units are widely underutilized, and thus skippable, especially if you have access to a coprocessor that can run compute shaders. maybe as compilers have got better this (vector units are underutilized) has become less true.
-
RE: https://mastodon.social/@mcc/117338324202161763
@eniko thanks.
and yeah, we need a kernel. one person can't write a kernel. it's not a matter of scope. it's a matter of you have to test more hardware configurations than one person can get in a private residence. we need that kernel handled by an *institution*.
the problem with freedos is if we want to make *games* then probably at some point we want access to sound and video. and if you want your game to render by any method other than CPU-on-framebuffer sound and video become Very Hard.
-
@ireneista @decay @amarioguy @MaddieM4 i suspect vector units are widely underutilized, and thus skippable, especially if you have access to a coprocessor that can run compute shaders. maybe as compilers have got better this (vector units are underutilized) has become less true.
@mcc @ireneista @decay @amarioguy
It has, and for a lot of software with hot loops that are meaningful to accelerate with SIMD, people have manually dictated the vector instructions as necessary. That is a particular feature where a lot of people would not willingly go back. I'd probably be one of those people tbh.
-
@ireneista @decay @amarioguy @MaddieM4 i suspect vector units are widely underutilized, and thus skippable, especially if you have access to a coprocessor that can run compute shaders. maybe as compilers have got better this (vector units are underutilized) has become less true.
@mcc @decay @amarioguy @MaddieM4 Fourier and wavelet transforms are really important though, even though most programs don't do them. but you raise a good point.
-
@ireneista @mcc @decay @MaddieM4 one thing i appreciate about Apple's SEP (despite yes, being from Apple and a closed, black-box so still not gonna say that anyone should just blindly trust it) is that the threat model they went for it was mutual distrust, where both the main CPU and the SEP core distrust each other, so both isolate their memory from each other
for context, both ME/PSP are modeled such that they own the platform entirely afaict, so if you get code exec in the ME or PSP, you own the whole platform, while in Apple's case, owning the SEP these days does not let you own the AP entirely, you still have to overcome most the other normal hurdles and can't just trivially DMA into host memory
that said, imo security processors including the SEP should not be unauditable black boxes at all, there's too much at stake for that to imo be a truly trustworthy system...
@amarioguy @mcc @decay @MaddieM4 yes, we do appreciate that choice, it's a good choice
-
@mcc @decay @amarioguy @MaddieM4 we'd keep hyperthreading, too
we just don't think management engines have any place heh
@ireneista @mcc @decay @amarioguy @MaddieM4
FWIW I *would* keep most of modern CPUs, but strongly support on security grounds Apple’s decision to not have hyperthreading.
Spectre v4 is just far too unnerving.
-
@mcc @decay @amarioguy @MaddieM4 Fourier and wavelet transforms are really important though, even though most programs don't do them. but you raise a good point.
@ireneista @mcc @decay @amarioguy That's a good particular example. A part of computing freedom that I think is pretty important is "being able to make music with computers, ideally in realtime." I have that on my list of vital user stories for my someday project microkernel OS. The arts are the kind of thing that make computers worth doing.
A lot of realtime music plugin stuff depends of Fourier transforms and SIMD in general.
-
The fourth backend is this guy named Mutabah.
https://github.com/thepowersgang/mrustc
Now, looking, I now see that he may be using Claude. What I remember is the last time I looked at this was three or four months ago and there weren't Claude signals, but also at the time it had been dormant for a while. I don't know why people think that resurrecting a dead project by starting to use claude is better than leaving it dead, but they seem to think this.
-
The fourth backend is this guy named Mutabah.
https://github.com/thepowersgang/mrustc
Now, looking, I now see that he may be using Claude. What I remember is the last time I looked at this was three or four months ago and there weren't Claude signals, but also at the time it had been dormant for a while. I don't know why people think that resurrecting a dead project by starting to use claude is better than leaving it dead, but they seem to think this.
@argv_minus_one @lispi314 Nevertheless, the fact it *did* have a dormant period means that it is easier to find a fork point than it would be with, for example, LLVM, especially because this implementation is all-in-one. The downside: C++. The entire goal here as far as I'm concerned is to avoid writing C++.
-
@ireneista @mcc @decay @amarioguy That's a good particular example. A part of computing freedom that I think is pretty important is "being able to make music with computers, ideally in realtime." I have that on my list of vital user stories for my someday project microkernel OS. The arts are the kind of thing that make computers worth doing.
A lot of realtime music plugin stuff depends of Fourier transforms and SIMD in general.
@MaddieM4 @mcc @decay @amarioguy yeah, absolutely. those are our reasons for bringing it up, as well. thank you.
we have given this topic thought, even if we don't have those thoughts organized in a way that makes them easy to present
-
@amarioguy @mcc @MaddieM4 one hopeful note
in our experience when a company is very big on having NDAs around its tooling, it's because there's nothing hard to replicate about the tooling itself, once you know the basic idea of it
we suspect people who do chip design can already basically say, yeah you need these specific types of simulator to eliminate whatever specific classes of problem
that seems like a thing the community could take on, in principle
@ireneista @amarioguy @mcc @MaddieM4 there are *so many* random missing pieces that you might not even have thought about yet that are possible "in principle", but we're struggling to remain optimistic that herding the cats is actually possible
there's "flagship-ish" missing tooling for things like "optical proximity correction" that people at least know about, but did you know that we basically don't even have a f/oss general-purpose optics simulator yet?
-
The fourth backend is this guy named Mutabah.
https://github.com/thepowersgang/mrustc
Now, looking, I now see that he may be using Claude. What I remember is the last time I looked at this was three or four months ago and there weren't Claude signals, but also at the time it had been dormant for a while. I don't know why people think that resurrecting a dead project by starting to use claude is better than leaving it dead, but they seem to think this.
-
@ireneista @amarioguy @mcc @MaddieM4 there are *so many* random missing pieces that you might not even have thought about yet that are possible "in principle", but we're struggling to remain optimistic that herding the cats is actually possible
there's "flagship-ish" missing tooling for things like "optical proximity correction" that people at least know about, but did you know that we basically don't even have a f/oss general-purpose optics simulator yet?
@r @amarioguy @mcc @MaddieM4 we did know that.
it's a big task. we think the immediate next step is to try to list out all the important types of simulation and what they're used for.
-
@mcc anyway yeah you're right we do need a kernel. Something that goes "here's file system, video, sound, and input" and then from there things get a lot more manageable. At least for games
@eniko have you heard of arcan
https://www.youtube.com/watch?v=07nqZIFRDJg -
@r @amarioguy @mcc @MaddieM4 we did know that.
it's a big task. we think the immediate next step is to try to list out all the important types of simulation and what they're used for.
@r @amarioguy @mcc @MaddieM4 speaking as someone with some modest success in tech labor organizing, the trick is to dangle a shiny thing just out of reach :3
-
@argv_minus_one @lispi314 I have blocked the user "Claude" on github. This causes GitHub to display a warning banner whenever there are commits advertised as being done with Claude's agent thing. The banner is glitchy and sometimes lies. However looking, I find these commits from August. It is possible the banner stems from this single third party PR.
-
@eniko have you heard of arcan
https://www.youtube.com/watch?v=07nqZIFRDJg@kasimir I haven't. I can't watch a video right now, what is it?
-
@argv_minus_one @lispi314 I have blocked the user "Claude" on github. This causes GitHub to display a warning banner whenever there are commits advertised as being done with Claude's agent thing. The banner is glitchy and sometimes lies. However looking, I find these commits from August. It is possible the banner stems from this single third party PR.
@argv_minus_one @lispi314 However, merging a PR that is specifically marked as Claude-coauthored (it's very much the case not all Claude code has this marking) is a strong signal a maintainer doesn't care very much either way about the embrace of AI coding assistants by the open source community. So this may indicate other things to come in the near future.
