Clément Grégoire @lectem

Mostly moved to the place where the sky is blue. Performance, GPU APIs, multithreading and game engines specialist @siliceum. Does reverse engineering for fun. lectem.github.io Clermont-Ferrand, France Joined August 2014

Tweets

4K
Followers

394
Following

532
Likes

2K

Clément Grégoire @lectem

a week ago

@NOTimothyLottes @joseph_h_garvin Ah that's worst case, cause with tracy we generally have <=2ns average iirc.

0 0 1 25 0

View Details

Clément Grégoire @lectem

2 weeks ago

@noop_dev @matiasgoldberg @NOTimothyLottes I'm not sure I follow exactly but if you can limit the number of things you send with traditional APIs, why can't you do the same with iouring if you need to limit oversubscription? Having bounds for the submitted work is always desirable, be it iouring or not IMHO.

1 0 0 20 0

View Details

Clément Grégoire @lectem

2 weeks ago

@noop_dev @matiasgoldberg @NOTimothyLottes I guess the real question ends up being, what do you let the kernel know upfront or discover.

2 0 0 28 0

View Details

Clément Grégoire @lectem

2 weeks ago

@noop_dev @matiasgoldberg @NOTimothyLottes But it also gives the kernel/driver less control (or at least not for the same price). The kernel has fastpaths and can usually do things better than us depending on the hardware (notably I/O scheduling). It can still do it without io rings obviously but it's more expensive.

1 0 1 27 0

View Details

Clément Grégoire @lectem

2 weeks ago

@noop_dev @matiasgoldberg @NOTimothyLottes Except the posix APIs have always been either blocking or fully shit to use async. And async is required if you want proper perf, spinning more threads is not the solution. There's a reason (as mentioned by the others) drivers and hardware (not just GPUs) use command buffers.

1 0 0 31 0

View Details

Clément Grégoire @lectem

2 weeks ago

I have no words.

0 0 0 37 0

View Details

Clément Grégoire @lectem

a month ago

@Ced_haurus On peut meme pas opt-out préventivement...

0 0 0 211 0

View Details

Clément Grégoire @lectem

a month ago

What the actual fuck @claudeai

0 0 0 59 0

View Details

Clément Grégoire @lectem

2 months ago

To the fucktard designer at @SamsungMobile or @Android who decided that opening an update dialog under my fingers AT 15% battery WITH A BUTTON AT THE KEYBOARD LOCATION WITHOUT ANY FUCKING CONFIRMATION DIALOG OR GRACE PERIOD: YOU SHOULD BE FIRED AND EAT MY PHONE

Clément Grégoire @lectem

2 years ago

Fuck you very much @Samsung. I've been postponing the oneui6.1 update for a while, and the dialog asking if I want to do it just popped WHILE I WAS TYPING TEXT. Result: I pressed "update" unknowingly because the button appeared right under my finger. I could have lost much

1 0 0 183 0

0 0 0 97 0

View Details

Clément Grégoire @lectem

2 months ago

@vkrajacic To be honest dynamic libraries have their uses, but they're hell on Linux. At least on windows it's easy to use/replace 😂

0 0 1 89 0

View Details

Clément Grégoire @lectem

2 months ago

@blelbach So doesn't std::simd. Or half of the std library additions since c++17. Yet std::hive would be more useful than most of those.

0 0 2 573 0

View Details

yoh @yohhoy

2 months ago

C++26 Shipped a SIMD Library Nobody Asked For lucisqr.substack.com/p/c26-shipped-…

4 36 138 28K 55

View Details

Clément Grégoire @lectem

2 months ago

@rfleury @matiasgoldberg Did you start the kernel part? Cause you'll have to parse (barely documented) text based files instead of proper APIs, for which the content may change whenever. At least you can read the kernel source to figure stuff out but that's not better than what the tool chains have.

0 0 1 1K 0

View Details

Clément Grégoire @lectem

3 months ago

Yearly report you should not be using GCP. It's not the first mishap with governments, ask Australia.

Joel 🇦🇺 @ptr_to_joel

3 months ago

is gcp actually cooked

23 42 948 100K 116

0 0 0 95 0

View Details

Clément Grégoire @lectem

3 months ago

@ForrestTheWoods I ended up using zigcc which is the only easily accessed toolchain targeting an old glibc. The fact we still need these kind of things (or go musl, which has its own issue) is sad.

0 0 2 107 0

View Details

Clément Grégoire @lectem

3 months ago

@jedisct1 Removing the perl requirement when? Insane we still need this to build in 2026

0 0 0 36 0

View Details

Clément Grégoire @lectem

3 months ago

@pandadiestro Circa 2022 😂 x.com/lectem/status/…

Clément Grégoire @lectem

5 years ago

I mean, this is only one of the most used libraries in the world, no worries. And it still relies on fucking Perl.

0 0 0 0 0

1 0 1 27 0

View Details

Clément Grégoire @lectem

3 months ago

@molecularmusing @SheriefFYI @jbevain Yeah it's something I do a lot too, it could be cached but I guess that some rules could be easily invalidated. Not surprising they didn't optimize it but yeah.