Skip to main content

Module integer

simd_kernels::structs::variants

Module integer 

Source
Expand description

§IntegerArray Module- Mid-Level, Inner Typed Integer Array

Arrow-compatible, SIMD-aligned integer array optimised for analytical workloads.

§Overview

  • Logical type: fixed-width signed/unsigned integers (T: Integer).
  • Physical storage: Buffer<T> (backed by Vec64<T> for 64-byte alignment) plus optional bit-packed validity mask (Bitmask).
  • Usable standalone or as the numeric arm of higher-level enums (NumericArray, Array).
  • Zero-copy friendly and integrates with Arrow FFI.

§Features

  • Construction from slices, Vec, or Vec64, with optional null mask.
  • Mutation: push/set, bulk null insertion, resize.
  • Iteration: safe and (optionally) parallel, via the MaskedArray trait surface.
  • Null handling: Arrow-style bitmask (1 = valid, 0 = null) with length validation.

§Usage Tips

Prefer function signatures with &[T] or generic T: Integer where possible to keep callsites simple while remaining zero-copy compatible with IntegerArray.

§Performance Benchmarks

Benchmarks under examples/hotloop_benchmark_std compare construction + sum over 1000 elems, averaged across 1000 runs on a 2024 Intel(R) Core(TM) Ultra 7 155H (22 logical CPUs, 32 GB RAM).

Averaged Results (1000 runs, size = 1000)

  • Vec<i64> (raw): avg = 0.131 µs
  • IntegerArray (minarrow, direct): avg = 0.130 µs
  • Int64Array (arrow-rs, struct): avg = 0.327 µs
  • IntegerArray (minarrow, enum): avg = 2.494 µs
  • Int64Array (arrow-rs, dynamic trait): avg = 3.283 µs

Build note: use --release for representative numbers; debug builds distort results.

Takeaways

  • Direct IntegerArray matches raw Vec<i64> while adding SIMD alignment + Arrow interop.
  • Enum dispatch adds modest overhead but remains faster than dynamic dispatch (arrow-rs).
  • For (extreme) latency-sensitive paths, prefer direct IntegerArray; otherwise differences are minor vs. real compute/system overhead.

Structs§

IntegerArray
IntegerArray