Meta-compilation of Baseline JIT Compilers with Druid

2025-02-15

The Art, Science, and Engineering of Programming

AI Summary

The coexistence of interpreters and JIT compilers in virtual machines incurs high development/maintenance costs and hampers evolutionary agility. Method: This paper proposes Druid—the first pre-compilation-based meta-compilation approach targeting baseline JIT frontends. It introduces lightweight annotations for built-in primitives within interpreter code, enabling automatic derivation and generation of high-performance baseline JIT frontends while reusing existing JIT infrastructure. Contribution/Results: Druid requires modifying only 60 call sites to produce a deployable implementation. Evaluated on the Pharo VM, the auto-generated JIT frontend achieves, on average, 2× the performance of the interpreter and 70% of hand-written JIT performance. Crucially, it significantly improves long-term VM maintainability and iteration efficiency, demonstrating for the first time the feasibility of low-intrusion, high-fidelity automated baseline JIT frontend generation.

Technology Category

• Search and Optimization: Metareasoning and Metaheuristics
• Machine Learning: Hardware-aware ML
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Virtualization and resource management in Web systems and infrastructures
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Virtual Machines (VMs) combine interpreters and just-in-time (JIT) compiled code to achieve good performance. However, implementing different execution engines increases the cost of developing and maintaining such solutions. JIT compilers based on meta-compilation cope with these issues by automatically generating optimizing JIT compilers. This leaves open the question of how meta-compilation applies to baseline JIT compilers, which improve warmup times by trading off optimizations. In this paper, we present Druid, an ahead-of-time automatic approach to generate baseline JIT compiler frontends from interpreters. Language developers guide meta-compilation by annotating interpreter code and using Druid's intrinsics. Druid targets the meta-compilation to an existing JIT compiler infrastructure to achieve good warm-up performance. We applied Druid in the context of the Pharo programming language and evaluated it by comparing an autogenerated JIT compiler frontend against the one in production for more than 10 years. Our generated JIT compiler frontend is 2x faster on average than the interpreter and achieves on average 0.7x the performance of the handwritten JIT compiler. Our experiment only required changes in 60 call sites in the interpreter, showing that our solution makes language VMs **easier to maintain and evolve in the long run**.

Problem

• Automatically generate baseline JIT compiler frontends from interpreters.
• Improve warm-up performance using meta-compilation with Druid.
• Reduce maintenance costs by simplifying VM development and evolution.

Innovation

• Ahead-of-time automatic baseline JIT compiler generation
• Meta-compilation guided by interpreter code annotations
• Integration with existing JIT compiler infrastructure