I spent all day helping the AI figure out the semantics of BRX instruction (branch to relative address) on Nvidia assembly, which allowed me to write a debug macro I can use to log values per thread anywhere within my kernel.Originally I was trying to figure out how to use CALL instructions but I couldn't make it work.
Has anyone ever tried vibe coding in assembly?
>>108351800There was someone a few days ago who talked a little bit about it. He said AI seems to struggle with writing asm directly since it's not very good at tracking register state, and that tracks with how I understand how AI works. I suspect you could probably get a pretty long way by hand rolling some inline asm "primitives" as naked C functions so the compiler handles all the register allocation and stuff and the AI could leverage that.
Reminder to just fucking use opencode and abandon vendor lock in
>>108351841I just realized assembly would suck anyways because it's very hardware dependent and also there are no libraries for it. C/C++ has always been the answer idk why I keep trying to overcomplicate shit.
>>108351800https://github.com/project-everest/valeI'm thinking this would be better than trying to vibe assembly directly
>>108351521>vibetrooning general
>>108351937Can you imagine a more indian thread title?
>>108351867What about kilo? That okay?
has anyone used this shit? looks interesting but I only do little projects right now so I don't think I care to use it https://github.com/obra/superpowers
>>108351521funny to see how your project hit the AI-slop wall already
the last thread got jannied for some reasonseemed like a fine thread, but /dpt/ was complaining that their thread was dyinganyhow, I’m trying to get Codex to fix my MP3s because Claude tried but couldn’t — the same MP3s are still syncing over and over to my iPhone
