codaaaaaaaaa's profile picture. 24

🌱

@codaaaaaaaaa

24

🌱 heeft deze post opnieuw geplaatst

このCコードをpopcnt 1命令にコンパイルするのは難しくないか? と思うかもしれないけど、実は場合によってはできる(godbolt.org/z/6ncKaq1df)。推論ではなくハードコードされたパターンマッチで変換してる。 そういう単純化のルールをLLMで自動生成できたら嬉しいというのはわかる。


Game engines should be designed so that creating empty textures is never necessary. Solving the problem is much better than encouraging a workaround.

Incredibly cold take: every game engine should come with a 1×1 transparent png by default 🤷‍♂️



This is what goes wrong when your wrapper library assumes no one would make syscalls directly on a platform that guarantees doing so works correctly. glibc has too many design flaws. Operating systems & programs should be working to remove any dependency on it whatsoever.

While researching the getpid() syscall for my daily series, I ran into an interesting piece of history. This goes back directly to the discussion from yesterday: should the OS offer a raw syscall interface or offer libc as the only contract? Here's what can go wrong!

popovicu94's tweet image. While researching the getpid() syscall for my daily series, I ran into an interesting piece of history.

This goes back directly to the discussion from yesterday: should the OS offer a raw syscall interface or offer libc as the only contract?

Here's what can go wrong!


🌱 heeft deze post opnieuw geplaatst

CERVEZA CRISTAL were truly ahead of their times.

Halftime: Dynamically weaves AI-generated ads into the scenes you’re watching, so breaks feel like part of the story instead of interruptions. @krishgarg @yuviecodes @lohanipravin



🌱 heeft deze post opnieuw geplaatst

It's crazy how inviting the reactor pools are. Something about seeing that blue glow with your own eyes, it's like a nuclear siren. Seems very comfy.


🌱 heeft deze post opnieuw geplaatst

I was curious whether compilers would convert a naive loop implementation of "set range of bits in u64 to 1" into O(1) bitwise ops. Answer: MSVC compiles the loop into a loop. Clang's output is bonkers: 418 bytes long, branches everywhere, 256-bit SIMD operations get involved?!

despair's tweet image. I was curious whether compilers would convert a naive loop implementation of "set range of bits in u64 to 1" into O(1) bitwise ops.

Answer: MSVC compiles the loop into a loop. Clang's output is bonkers: 418 bytes long, branches everywhere, 256-bit SIMD operations get involved?!
despair's tweet image. I was curious whether compilers would convert a naive loop implementation of "set range of bits in u64 to 1" into O(1) bitwise ops.

Answer: MSVC compiles the loop into a loop. Clang's output is bonkers: 418 bytes long, branches everywhere, 256-bit SIMD operations get involved?!
despair's tweet image. I was curious whether compilers would convert a naive loop implementation of "set range of bits in u64 to 1" into O(1) bitwise ops.

Answer: MSVC compiles the loop into a loop. Clang's output is bonkers: 418 bytes long, branches everywhere, 256-bit SIMD operations get involved?!
despair's tweet image. I was curious whether compilers would convert a naive loop implementation of "set range of bits in u64 to 1" into O(1) bitwise ops.

Answer: MSVC compiles the loop into a loop. Clang's output is bonkers: 418 bytes long, branches everywhere, 256-bit SIMD operations get involved?!

🌱 heeft deze post opnieuw geplaatst

Good post. A underemphasized benefit of SIMD is that so much of the cost of running an instruction on a modern ISA is fixed (decode, rename, register access, speculative execution) whether we're adding two u8s or two 512-bit registers with 64 u8s each. The gap is so large it ...


🌱 heeft deze post opnieuw geplaatst


Android failing at trivial input handling on a 2200MHz processor + dedicated GPU. Comments believe this is because the HW is somehow not powerful enough

Proof that iPhone > Android.



🌱 heeft deze post opnieuw geplaatst

what I dislike about autocomplete is that i will want to use the arrow keys to navigate the text, but autocomplete will briefly make the arrows do something else, where I have to hit ESC, and the feeling is like doing parkour over random barriers

I think many software developers don't understand the concept of pipelining. If you're doing one thing that has Y steps, it makes sense to do one step at a time: go from step 0 to step Y and be done. If you're doing N things that each have Y steps, it makes sense to do the



🌱 heeft deze post opnieuw geplaatst

Loading...

Something went wrong.


Something went wrong.