LLVM optimisation levels land in Cheeku
Cheeku now exposes O0 through O3, a generic CPU target and opt-in diagnostic artifacts. The runtime checks stay exactly where they were.
IN THIS POST
- LLVM
- Optimisation
- Cheeku
Blog · Notes from the team · 7 posts
Compiler internals, performance measurements, type-system decisions and the honest edges of an early language. Everything here is written by the people building Chuchu and Cheeku.
01 · LATEST
Cheeku now exposes O0 through O3, a generic CPU target and opt-in diagnostic artifacts. The runtime checks stay exactly where they were.
IN THIS POST
02 · ALL POSTS
Short, specific notes rather than announcements. Pick a topic and read how the implementation actually works.
Inference should remove noise, not contracts. Locals take their type from the value; function signatures stay written down because callers depend on them.
A tour of the Cheeku pipeline: lex, parse, resolve, check, lower, optimise and link. Nothing reaches LLVM that has not already passed the type checker.
Eight workloads, four optimisation levels and three reference languages. The report is useful because it also says what the numbers cannot prove.
Highlighting, snippets, completions and scope-aware navigation work without a compiler. Point the extension at Cheeku when you are ready to build.
A young language is defined as much by its boundaries as its features. Here is the current edge of the implementation, written down plainly.
Fixed size, homogeneous and checked. Chuchu arrays keep element-wise arithmetic simple and make an out-of-range index a real error.
Install Cheeku, compile a .chuchu file and run the native executable yourself.