Per Vognsen / is on Nostr

Performance, compilers, hardware, mathematics, computer science.

I've worked in or adjacent to the video game industry for most of my career.

Last Notes

2024-07-05 19:09:13

- reply

@npub1c9m…9xjd I think that's the book I read (I went to AU).

2024-04-16 03:14:07

- reply

@npub1y0e…mm9n The whole paperclip thing to swap SIM cards has always struck me as a usability disaster. It's crazy we're stuck with that.

2024-03-19 05:05:04

- reply

@npub1pu3…8r76 On that general note, I remember looking up a while ago whether the whole thing about children suffocating to death from plastic bags was exaggerated (it always sounded implausible to me since you start to get the CO2 build-up suffocation reflex long before oxygen deprivation leads to loss of consciousness or brain damage). Turns out it's a thing: https://skeptics.stackexchange.com/a/6530

2024-03-19 04:58:20

- reply

@npub1pu3…8r76 Counterpoint: I always remember being told "don't run with scissors" but never heard anyone say "don't run with knives".

2023-12-16 23:24:50

- reply

@npub1a4c…t2kf Doyle's Sherlock Holmes stories are really bad about this but in a different direction than the Jack Reacher example. It's all about Holmes seizing on some hyper-specific, easily missed detail and pretending that you can reconstruct everything from that. I actually hate that way more and it's prevalent in a lot of criminal forensics in pop culture, e.g. blood spatter analysis in Dexter.

2023-11-02 03:12:59

- reply

@npub1u3f…nne2 It looks like archive.org has a pretty good set of episodes from the first season but obviously not DVD quality.

2023-10-08 02:11:51

- reply

@npub1u3f…nne2 Technics SL-1200 or an ancient gramophone with a brass horn?

2023-09-18 09:12:08

- reply

@npub1y0e…mm9n Three orders of magnitude is such an incredible gap that the null hypothesis is that he screwed up the benchmark (or the workload is truly pathological).

2023-09-09 06:31:58

- reply

@npub1y0e…mm9n Tfw someone submits a bug for AIX.

2023-09-08 14:17:27

- reply

@npub1src…fms4 It seems like a certainty that there's going to be open-and-shut cases of copyright violation. But unless some sort of massive class action succeeds, yeah, it's probably going to be more than worth it for Microsoft to just take it on the chin in those cases.

2023-08-31 03:52:37

- reply

@npub1y0e…mm9n The 18 MB is referring to the generated Rust code, not the size in the static data segments when compiled/linked. I agree that's a large blow-up but you do see a significant blow-up when trying to convert from a raw byte array to an equivalent C static initializer using hex literals or whatever.

2023-08-31 03:45:39

- reply

@npub1y0e…mm9n Just to be clear, I mean the representation where you can resolve an lvalue relptr with (uintptr_t)&relptr + (uintptr_t)relptr which just turns into
mov rbx, [rax]
add rax, rbx
It shouldn't have issues with deeply nested fields.

2023-08-31 03:35:35

- reply

@npub1y0e…mm9n Yeah, I've used the fixed-address trick before although usually only as a fast path, i.e. I would still have the relocation information, I just don't need to apply if I can mmap at the preferred base address. But I usually just use position-independent representations these days if this is something I care about.

2023-08-31 03:00:52

While this has nice ergonomics, I take exception to the idea that it's zero-copy: https://manishearth.github.io/blog/2022/08/03/zero-copy-3-so-zero-its-dot-dot-dot-negative/
As soon as your types have references (e.g. the Person struct example), the operating system loader has to do pointer fix-ups on load. And because this usually happens with the additional overhead of copy-on-write soft page faults, I could see this being significantly slower than equivalent non-copy-on-write pointer fix-up code. I'd have to do proper testing.