Viewing profile — rurban

More probably worst use of NixOS. In embedded we care for every single kb, and don't have tolerance for this kind of space waste management system

62s for a single instruction! Wonder if an compilers cost tables knows that. But it's data dependent, and cost functions probably don't do that.

I had to write a gnome shell extension to see my various LLM token budgets. Green, orange, red for Claude, OpenAI, GitHub Business, Kimi, DeepSeek, and such. Very useful.

I had similar problems with my free movie festival ratings database, so eventually I had to get rid of the dynamic lookup, and I only dump now static sites, hosted on GitHub Pages.…

This was the first time for me. All tests failed, for the whole day. For 20m, ok, I can tolerate that. But for a whole day? Microsoft sucks big time.

I'd really need to marry this with qemu and osxcross, the ability to cross compile from x86_64-linux to darwin. So far I could only test apple compilation, but not runtime yet.

Thanks for the depth-anything.cpp port. Saves us a couple of kernel OOM's, causing connection disruptions.

USA above average? That cannot be true

In gdb? It's a shame gdb picked python

Certainly not. Every hand-optimized tight loop or lookup beats every compiler still. Compilers are extremely stupid, and esp. the language designers.

Because the Intel syntax is limited, and pretty broken by binutils. I had to change my C compiler rcc to emit AT&T GAS, instead of Intel syntax, even if Intel syntax looks easier o…

Well, the middle ages. Thanks we got around this mess

Not really fine tuned. But the base was Qwen. We are analyzing images

For LLM inference in production we use vLLM (with our own model).

Did they train on sama? Highly unlikely

Happy with oh-my-pi (omp)

No openmp, no openacc, though that would all be trivial. For linking, my own linker still has a lot of regressions, but is at least 10x faster than shelling out to gcc/ld. Similar …

No way. No open prompts. But I use omp /collab, so I prompt it from my phone. Constantly

It compiles the kernel and my big projects, but I haven't released it yet. Feature wise I don't do bitint, decimal and FloatX. No auto-vectorization, but better than most simd/neon…

Of course not. I don't care for gcc's broken optimizer to remove dead code without any warning. gcc has too many bugs for my taste, sorry. And I don't want to wait 20m on gcc compi…

The proof is gcc's torture test suite. Have a look

I had to reverse engineer Chinese SDK's for various embedded chips. Their libs were full of this shit. It works in some cases, but rarely in all. In embedded you can thankfully thr…

If you manage to click through all >200 partners to say no, the TV startup will be significantly faster. Like 10x faster. Cutting them off the wire and view through another dongle …

You've seen Australia. Every kid who wanted had access to forbidden social media. The control bit is the dangerous one, sure.