Show newer
glaebhoerl boosted

@pervognsen @dougall The best public attempt at benchmarking I could find is this:
martin.ankerl.com/2022/08/27/h
Generally speaking, the best RH implementations do better than the SIMD ones. Strangely, the only benchmark where the SIMD ones dominate is the 500k integers test.

glaebhoerl boosted

@glaebhoerl @regehr @comex Also see Compilation On The GPU? A Feasibility Study[1], which is in some ways more conventional than co-dfns (it's a C-like language) and also in some ways more ambitious (the parsing handles arbitrarily nested depth without performance compromise).

[1]: dl.acm.org/doi/pdf/10.1145/352

glaebhoerl boosted

".text" sections don't contain text. ".data" contains some but not all data. ".bss" stands for block started by symbol, a directive in one particular assembler for the IBM 704 (1954!) that just.. stuck around.

Object files are files but neither are nor contain objects (in the modern sense). Core dumps are memory dump but they've not been dumps of (magnetic) core memory since the 70s.

There are segments and sections, which mean slightly different things in every format.

Show thread
glaebhoerl boosted
glaebhoerl boosted

relevant to my post last night that I'm going back and fixing up my 23 years of blog archives, and restoring broken links, it is *astounding* how much better a job the personal web does at keeping links alive and content online. Most personal blogs I linked to are either still around or redirect to a place where the content is easy to find. The vast majority of corporate content (including news) has been erased, with only imperfect Internet Archive copies available.

glaebhoerl boosted

It's so odd that the concept of "true or false" is named after some guy, and there's no other word for it

According to my investigations, it seems there is no mathematical concept called a "doodad"

glaebhoerl boosted

Great read about the tradeoffs of Mesa building its own GPU-focused compiler vs. using LLVM jlekstrand.net/jason/blog/2022

There is a github issue about it: github.com/mastodon/mastodon/i and apparently the behavior has been changed as of Mastodon 4.0, which is presumably what @kha had in mind in their earlier reply to me.

Show thread

Update: I discovered that if you *left-click* on a thing in the timeline, it opens a view of it that's still on the same server, so interaction works. It's only middle click to open a new tab that goes to the originating server. The internet has thoroughly trained me out of ever left clicking anything on a website with infinite scroll.

Show thread
glaebhoerl boosted

LLVM's design for pattern matching in C++ is way more reasonable than the usual failed attempts at emulating functional programming languages. In my below examples I use explicit pointers for binders, which I think is clearer than the implicit references which LLVM uses.

Value x, y;
if (match(value, Add(&x, &y)) {
// x and y are bound
}

And it is nestable:

Value x;
int y;
if (match(value, Add(&x, Const(&y))) {
// x and y are bound
}

Apparently so.

Being not a security person or authentication-knower, couldn't they use... (OAuth? or another of these things I've vaguely heard of?)... to let you sign in on one server using your account from another? Even if I had to do this for 20 different servers, *once*, it'd still be a great improvement.

Obviously since they haven't, I'd guess they probably couldn't, but.

Show thread

So um. Sorry, I'm sure every new person asks this. But do I have it right that if I open some toots in tabs (and they're not coincidentally on my "home server"), and, upon reading, want to interact with them, then the only way is to copy the URL, go back to my home server, and paste the URL into the search bar? I'm not missing anything?

Mastodon

a Schelling point for those who seek one