Just merged a CPU↔GPU language parity PR into Kalandra. Most language features now also run on Vulkan, CUDA, and Apple Metal: loops, local mutation, recursion, and closures. With memory safety, checked arithmetic, and cooperative cancellation.
@mikulasflorek I ported my language from JS to LLVM/MLIR with Codex. CPU performance matches Zig, Rust, and C++; the GPU target is 10-12x faster than the CPU. GPT allowed me to revive an 8-year-old project: https://t.co/0SN0Xy98Q7
@peduarte Unironically, I'm building my own for the same reason, cause I know something someone else built won't last more than 3 months before some other shiny thing grabs their attention.
Wanted to see how my language compares with this little benchmark, so I added a few experimental features and ran it. Basically, it's a tie: we use LLVM like the others, so there isn't much of a surprise there.
Sauce: https://t.co/G6Yv1QOS6b
@TutaPrivacy Cool, but Google is one account to manage, one ecosystem. DeGoogling means switching to dozens of other providers, risking data leaks and security problems with each one, plus the hassle of managing more subscriptions and accounts. That doesn't work.