sklears-simd
Latest release:
0.1.0(March 20, 2026). See the workspace release notes for highlights and upgrade guidance.
Overview
sklears-simd exposes low-level SIMD, GPU, and hardware acceleration utilities used across the sklears ecosystem. While primarily an internal crate, it is documented for contributors building new high-performance components.
Key Features
- Vector Abstractions: Portable SIMD types (f32x4, f32x8, f32x16) with architecture-specific intrinsics.
- Alignment & Memory: Alignment helpers, prefetching hints, cache-aware allocation strategies.
- GPU Bridges: CUDA/WebGPU adapters, Tensor Core pathways, and multi-GPU orchestration helpers.
- Benchmark Harnesses: Criterion-based benchmarks and profiling utilities for micro-optimizations.
Quick Peek
use F32x4;
let a = new;
let b = splat;
let result = a.mul;
assert_eq!;
Status
- Core building block validated by the 11,292 passing workspace tests for
0.1.0. - Powers SIMD/GPU acceleration in linear models, neighbors, metrics, and more.
- Contributor roadmap (new architectures, auto-vectorization tooling) maintained in
TODO.md.