This document provides performance benchmarks for key functions in the library, comparing optimized versions with their standard counterparts.
Note
- All benchmarks were run on an Apple M4 Pro MacBook Pro with 24GB of RAM using JMH (Java Microbenchmark Harness).
- Results may vary depending on the hardware and software environment.
- To run the benchmark on your device, use the
./gradlew :benchmark:maincommand and it will update the graphs in this file when it finishes - To read the exact numbers check the /processed folder in the benchmark module.
- The benchmarks below are not exhaustive and do not cover all functions in the library. They are intended to provide a general idea of the performance improvements offered by the optimized versions.
- Contributions or suggestions to improve the accuracy of the benchmarks and their results are welcome.
BitArray packs bits and applies word‑level bitwise ops, so bulk ops (count, all, xor ...etc) are dramatically faster;
random per‑index set can be slower.
BitArray stores 8 bits per byte (~8x+ smaller than BooleanArray), minimizing allocations—ideal for huge boolean sets or memory‑bound workloads.
Primitive Lists uses contiguous, unboxed primitives with pre‑sized results and index‑based loops (no boxing/iterators), making ops like map/forEach/chunked much faster for numeric pipelines.
Primitive Lists stores primitives contiguously with far fewer allocations and memory usage than generic List.
In‑place variants reuse the source list and index ranges, avoiding intermediate collections. best for large sliding windows, chunking, and scans.
Near‑zero allocations by eliminating intermediate lists; ideal for streaming/windowed processing under tight memory.
Primitive array ops are index‑based over primitives with pre‑sized outputs and direct copyInto, avoiding boxing, iterators, and extra bounds checks—ideal for large map/filter/take.
Preallocates exact‑size primitive buffers and writes directly (no intermediate lists), significantly lowering allocation and GC overhead.
