Simon Felix
Sebastian SchönerSomeone wrote a CUDA compiler. I like the spirit of this: https://github.com/Zaneham/BarraCUDA
the tower fairy@sschoener "no CMake" yet they rely on an Unix only buildsystem, not different than using autotools from the outset
manx@crystalmoon @sschoener While I agree with the sentiment that "POSIX-only" is really not "portable", in this particular case, the Makefile does not really do anything. It merely lists the files that need to be compiled. Trivial to "port" to any non-POSIX compiler. Frankly, "cl.exe src\*.c" does the trick.
Arseny Kapoulkine@crystalmoon @sschoener There’s five circles of hell separating autotools from manually written Makefiles. And what are you going to do with gfx11 elf binaries on Windows anyway?
@zeux @sschoener cross build and/or examining them. I do that a lot in my daily job, usually prefer Win32 GUIs over GNOME mouse quirks, and macOS (pre 26) over either
Bernd Petrovitsch🔴🔴🔴♂️🏳️🌈🇦🇹🇪🇺@crystalmoon @sschoener 'make.exe' for Windows exists a d Windows has POSIX Subsystem and is POSIX-compatible according to M$FT ...
Sharlock93@sschoener I checked and it's not unity builds which I'm very happy about
Micha Mettke@sschoener Feel sorry for the developer. The amount of backlash/harassment he got for generating some of the tests and documentation using AI for a project (while being open about it) while he never public released to hackernews is just toxic. Having the world judge your project/work alone is already a lot to handle but this is on a different level.
@vurtun I haven't actually followed that.
If it is any consolation: My takeaway was actually that I this is a super neat thing to try and in the right spirit. My main thought actually was "why has nobody else ever thought about doing this?" and I give that guy a lot of credit for coming up with this gap and doing something about it.
(FWIW, I would go full on AI on such a project as well)
Per Vognsen@sschoener @vurtun There's ZLUDA as well in the "CUDA on AMD" category and it's much further along: https://vosen.github.io/ZLUDA/blog/zluda-update-q4-2025/. If you write stuff like "LLVM is NOT required. BarraCUDA does its own instruction encoding like an adult" (which the README is full of) any criticism of your effort seems fair game. It would definitely be nice to have something standalone, though.
@pervognsen @sschoener I interpreted "LLVM is NOT required. BarraCUDA does its own instruction encoding like an adult" and the other quips more as a new zealand kind of humor or inside joke and less the sadly prevalent toxic arrogance. But who knows.
@vurtun @sschoener Also, it's not "just" documentation and tests he's writing with coding agents. Just look at his GH output. It's hundreds of thousands of lines per month for dozens of new projects. Here's just one initial commit for one of his random projects that has 300,000 added lines.
Ell 🏳️⚧️@sschoener The acknowelgements section is also a delight, in a different way, particularly the outro:
> He aha te mea nui o te ao. He tāngata, he tāngata, he tāngata.
> What is the most important thing in the world? It is people, it is people, it is people.