Hacker News new | past | comments | ask | show | jobs | submit login

I love the idea of this project, but as even Mike Pall has mentioned the "proper" way to do this is to target Lua code directly. LuaJIT is designed to speedup Lua, not LJ bytecode.



Could I see a reference? I've been reading a lot of Mike Pall lately and this is relevant to my interests.


I looked on http://www.freelists.org/archive/luajit/ and http://lua-users.org/lists/lua-l/ and couldn't find a smoking gun.

The closest thing is this http://www.reddit.com/r/programming/comments/1r2s82/lua_fun_...

Where someone else http://rtsisyk.github.io/luafun/intro.html attempts to generate LuaJIT bytecode directly and tracing JIT doesn't speedup their code because LuaJIT isn't designed to speedup arbitrary bytecode but bytecode generated by the LuaJIT compiler.

There was still a more specific post that I am unable to locate where Mike specifically warned against targeting LuaJIT bytecode. I'd probably just post a message on the luajit list if you want clarification.


> There was still a more specific post that I am unable to locate

This one? http://www.reddit.com/r/programming/comments/badl2/luajit_2_...


Close.


Just checked out your github page, you are like my brother from another mother. If I am in Berkeley I will drop you a line and we can code on some Forth JITs in Lua.


You read my mind!

Ok, you read my github ;-) But yes! Lua has multiple returns, we can perform 'register allocation' and spill over onto locals. Stack? Hah.




Consider applying for YC's Spring batch! Applications are open till Feb 11.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: