Wild Performance Tricks in Rust
a day ago
- #Rust
- #Parallel Processing
- #Performance Optimization
- Attended RustForge in Wellington, New Zealand, and gave a talk titled 'Wild performance tricks'.
- Discussed mutable slicing for sharing between threads in the Wild linker, optimizing for cache locality and parallel processing.
- Explored parallel initialization of a Vec using the sharded-vec-writer crate to avoid sequential placeholder filling.
- Introduced atomic to non-atomic in-place conversion techniques for Vec<SymbolId> to Vec<AtomicSymbolId> without runtime cost.
- Highlighted buffer reuse techniques to minimize heap allocations, especially with non-static lifetimes.
- Suggested deallocating large memory buffers on a separate thread to improve performance.
- Provided a bonus tip on stripping lifetimes with non-trivial Drop by converting to MaybeUninit.
- Acknowledged GitHub sponsors for their support in enabling this work.