Jeandle: How Ant Group’s LLVM‑Based JVM JIT Aims to Supercharge Java Performance

Jeandle Introduction

Jeandle is a brand‑new Just‑In‑Time (JIT) compiler for the OpenJDK HotSpot JVM, open‑sourced by Ant Group. It leverages LLVM for compilation optimization and code generation, bringing LLVM’s performance and ecosystem benefits into the JVM.

What is a JIT compiler

Java achieves “write once, run anywhere” through an interpreter that reads bytecode on any CPU. Interpreters are inefficient, akin to translating each word of a sentence without context. A JIT compiler acts like an experienced translator: when a code segment is hot, the JVM compiles it, analyzes and optimizes it, and emits efficient machine instructions.

What is LLVM

LLVM is a widely‑used open‑source compiler infrastructure offering modular design, powerful optimizations, and extensive backend support. It can be viewed as a ready‑made toolbox for building new compilers.

Write a language front‑end that converts source code to LLVM IR ( Intermediate Representation).

Use LLVM’s optimizer to analyze and improve the IR.

Generate real CPU instructions from the optimized IR via LLVM’s code generator.

LLVM’s additional strengths include:

Extreme performance optimization: industry‑leading code analysis and optimization.

High modularity: selective use and flexible extension, making it the compiler of choice for modern languages such as Rust and Swift.

Vibrant open‑source ecosystem: maintained by top companies, research institutions, and developers, with strong relevance in AI and emerging languages.

Future Plan

Integrating the complex JVM and LLVM systems poses major challenges. The Jeandle team aims to address several technical problems, including perfect support for JVM garbage collection, custom LLVM features for Java’s dynamic constructs, and Java‑specific optimization algorithms.

Perfect support for JVM garbage‑collection (GC) mechanisms.

Custom LLVM support for Java dynamic features such as synchronized.

Design a set of LLVM‑based optimizations tailored for Java.

The roadmap is clear:

End of 2025: Full bytecode support, covering exceptions, GC, synchronization, and all basic JVM features.

2026: Focus on performance‑boosting “black‑tech” including:

Java‑specific optimization suite (lock optimization, escape analysis, advanced inlining).

Intrinsic support: hand‑written assembly for hot paths and common library functions.

On‑Stack Replacement (OSR) for dynamic hot‑swap of running code.

Support for the modern G1 garbage collector.

Project Address

https://github.com/jeandle/jeandle-jdk https://github.com/jeandle/jeandle-l