Reproduce “mold: A Massively Parallel Linker” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Rui Ueyama
The `mold` linker dramatically accelerates software build processes by implementing a massively parallel architecture across all major linking passes. This design achieves substantial speedups on large, real-world workloads, for instance, linking 16.1 times faster than `lld` for a TensorFlow debug build on x86-64, thereby enhancing developer productivity.
Key insights
- Pervasive data parallelism, when systematically applied to all major linking passes, can drastically reduce link times beyond what incremental parallelization in existing linkers has achieved.
- Decoupling input parsing from symbol resolution and other core phases is crucial for enabling widespread concurrent execution in a linker, mitigating issues like order-sensitive semantics.
- A combination of optimized parallel algorithms for each linker pass and system-level optimizations (e.g., memory management, I/O) are collectively essential for maximizing performance gains and scaling efficiently on multi-core hardware.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.