Spent too long chasing virtual register chains to see that the underlying problem is an ordering issue... phis happen simultaneously, and I can make the emulator cheat less with its upsilon/phi implementation to avoid failing here, but will have to sort out unraveling this for actual code generation. Time to read a bit more, as I'm sure this is a well explored problem.

10 = phi 2, 11
8 = phi 1, 10
5 = phi 3, 15
4 = phi 0, 14

#Projects #Compiler

Ohm's Peg-to-WASM Compiler

https://ohmjs.org/blog/2026/03/12/peg-to-wasm

#HackerNews #Peg-to-WASM #Compiler #OhmJS #WASM #Compiler #Programming #News #Tech #Innovation

Ohm's Peg-to-WASM Compiler

https://ohmjs.org/blog/2026/03/12/peg-to-wasm

#compiler #wasm

So I'm writing a #compiler. No LLMs just my squishy brain and textbooks for the language #Oberon and docs for #llvm

This is a path well travelled but my trick is I'm keeping Professor Wirth's freely licensed original source and replacing only the code generation parts with calls to the llvm-c api.

So I guess I'm not even writing the compiler, I'm adapting it.

Huge thanks to obnc which translates oberon to c. When the translated compiler can compile itself I'll just stop using the translator.

[CPython 3.15의 JIT, 다시 궤도에 오르다

CPython 3.15의 JIT가 macOS AArch64에서 테일 콜링 인터프리터 대비 11~12%, x86_64 Linux에서 5~6% 빨라지며 성능 목표를 조기 달성했다. 주요 스폰서 철수 이후에도 커뮤니티 주도 방식으로 전환해 신규 기여자를 늘리고 버스 팩터를 낮췄으며, 세부 작업 분해와 측정 인프라(doesjitgobrrr.com)가 개발 속도와 품질 개선에 크게 기여했다. 또한 트레이스 레코딩, 참조 카운트 제거, 듀얼 디스패치 같은 설계 선택이 JIT 성능 향상의 핵심 요인으로 작용했고, 프리 스레딩은 3.15/3.16을 목표로 계속 진행 중이다.

https://news.hada.io/topic?id=27971

#cpython #python #jit #performance #compiler

Portable BASIC cross compiler: BASSM

Jens Henschel writes: "What began as an 'agentic development' experiment is now a nearly usable product. BASSM is an IDE for Windows and Linux that allows you to build games for an Amiga with an OCS/ECS chipset and 512 KB of RAM. The IDE and the transpiler are built in JavaScript and the Electron framework and are therefore fully portable. Further details:

https://www.amiga-news.de/en/news/AN-2026-03-00127-EN.html

#Amiga #Basic #compiler #Windows #Linux #retrocomputing

test/1900-md5.spl is paying off nicely. What a glorious mess of virtual registers to unsnarl in the IR for md5_step()

#Projects #Compiler

Whee. Compiler stage3 now passes all tests that stage1 and stage2 do, when running the tests via the IR emulator.

Time to start improving the IR, allocating registers, and implementing a backend to generate "real" code from the IR.

I am now a ways past where this project was when I restarted it because the original approach was becoming too much of a mess. Very happy with the state of things now.

#Projects #Compiler