Postări cu #codeoptimization
Postări publice de pe X care folosesc #codeoptimization.
Vixs ✨ @vixs101#100DaysOfCode Challenge Day 7: i wasn't able to do much, but i - learned a bit about conditional rendering in react. it's quite a fun concept. (check comments for thread) - Also worked on my personal project. i can't wait to show u guys when it's done.
exQUIZitely 🕹️ @exQUIZitely22 kilobytes. I often think about the technical limitations that game designers had to work with in the 80s - both in terms of software and hardware. To me, the game that stands at the very top is Elite (1984). Elite is smaller in data size than many modern text files or desktop icons, yet it lets you freely explore a multi-galaxy-spanning universe that felt gigantic and virtually limitless. Think about this for a second: The core game code on the BBC Micro version has roughly 22 KB of memory. The C64 version from 1985, which I played as a kid, isn't much bigger. Now think about what Braben and Bell turned those few kilobytes into: a massive universe with eight galaxies, each containing 256 star systems (for a total of 2,048 planets per system). Each system featured unique details: government type, economy, technology level, population, commodity prices, and even descriptive text. Oh, and by the way, the game also rendered 3D wireframe ships, stations, and planets in real time on a 2 MHz 6502 processor. This is no slight on today’s game designers. They work with what they have, and that's okay. But when I think about the worlds that some old school programmers created with the tools they were given, it sometimes breaks my brain trying to to even remotely understand how they did it. Not megabytes, not gigabytes... kilobytes!
Sebastian Aaltonen @SebAaltonenIn my experience LLMs are bad for code optimization. When I am using LLMs, I let the LLM write the boilerplate code and then I rewrite the perf critical parts such as inner loops. You have to understand performance to instruct LLM to write performant code for you. Super strict instructions about the algorithm and data structures. Then review and tell the LLM exactly what should be fixed and why. Most of the code in GitHub has performance issues. Idiomatic C++ and Rust code tends to be filled with short lived single object allocations and awful fine grained locking. The training material simply isn't good enough for high performance real-time apps such as games, rendering or physics simulation. You have to be an expert of your respective area and make the architectural/performance decisions yourself. AI is great for writing glue code, gameplay code, UI code and similar, but you should write perf critical inner loops and shaders yourself (AI doesn't really understand SIMD loop/branch coherence).
