Works on my computer: RTX 3090, CUDA 12.6

Interesting project! I haven't really worked with Vulkan myself yet. Hence my question: how is the code compiled and then loaded into the cores?

Or is the entire code always compiled in the REPL and then uploaded, with only the existing data addresses being updated?

I'm curious how a gpu language's syntax design can be different from CUDA kernel?

Because I think there is no way to avoid concepts like thread_id.

I'm curious how GPU programming can be made (a lot) simpler than CUDA.