Escaping the AST: A Data-Oriented, Lock-Free Parallel Compiler Architecture

Note : It is mandatory to RSVP on this page to receive the meeting link.

Abstract :

The architecture of legacy compilers presents limitations for modern software development. As codebases scale, developers face increased compilation times. While language complexity is often cited, key architectural bottlenecks include pointer-chasing across deeply nested ASTs (cache misses), single-threaded type resolution, and significant thread-lock contention (RwLock) during parallel semantic analysis. What happens when we discard the Abstract Syntax Tree entirely and apply strict Data-Oriented Design to the compiler itself?

In this session, we will explore the internal architecture of a compiler frontend co-designed with the programming language to safely maximize utilization of available CPU cores. We will dissect how to translate complex, tree-like program semantics into flat, contiguous arrays of 256-bit bit-packed Global Identifiers (GIDs).

By stepping away from traditional recursive tree-walking and object-oriented compiler design, attendees will learn how to implement high-throughput parallel pipelines.

This talk is not just for language designers. The architectural patterns used to build the compiler: flattening graphs into arrays, deferred identity, and lock-free synchronization boundaries—are directly applicable to any C++ developer building high-performance, multithreaded systems.