Building the fastest Lua interpreter automatically (sillycross.github.io)

Article summary

The article discusses a research project to automatically generate high-performance interpreters for dynamic languages, specifically Lua. The project, called LuaJIT Remake, uses a meta-compiler called Deegen to generate a Lua interpreter from a semantical description of the bytecode. The resulting interpreter outperforms LuaJIT's interpreter by 28% and the official Lua interpreter by 171% on average. The project aims to make writing high-performance virtual machines easier by automating the process.

Main themes

• automatic interpreter generation
• high-performance virtual machines
• Lua and LuaJIT
• compiler optimization
• calling conventions
• register usage
• garbage collection
• language implementation
• Haskell and Lua bindings
• dependency management

What commenters say

• The article's performance comparisons are misleading because they only compare the interpreter, not the JIT compiler.
• The use of inline caching is a significant optimization that contributes to the performance gain.
• The project's approach is not entirely new, as there are prior art and existing tools that achieve similar goals.
• The comparison between LuaJIT and the new interpreter may be affected by differences in garbage collection implementation.
• The project's dependency on LLVM may be a limitation, and alternative approaches using other compilers or languages could be explored.
• The article's discussion of calling conventions and register usage is interesting and relevant to the project's goals.
• The performance gain is not solely due to the automatic generation of assembly code, but also due to other optimizations and improvements.
• The project has implications for other languages and use cases, such as JavaScript and Haskell.
• The use of a meta-compiler like Deegen can abstract away low-level details and make it easier to write high-performance interpreters.