saker-rs 0.1.1

Fast SIMD-accelerated linear algebra
Documentation
  • Coverage
  • 0%
    0 out of 41 items documented0 out of 20 items with examples
  • Size
  • Source code size: 51.65 kB This is the summed size of all the files inside the crates.io package for this release.
  • Documentation size: 1.53 MB This is the summed size of all files generated by rustdoc for all configured targets
  • Ø build duration
  • this release: 21s Average build duration of successful builds.
  • all releases: 22s Average build duration of successful builds in releases after 2024-10-23.
  • Links
  • MihneaDarie/saker-rs
    1 0 0
  • crates.io
  • Dependencies
  • Versions
  • Owners
  • MihneaDarie

saker-rs

Fast SIMD-accelerated linear algebra for neural network inference, written in Rust.

Named after the Saker Falcon — a fast bird of prey native to the Carpathian region.

Features

  • Blocked GEMM with cache-aware tiling
  • AVX-512 and AVX2 micro-kernels with FMA
  • Scalar fallback for tail tiles
  • Fused bias + activation (SiLU, Sigmoid)
  • Parallel execution via Rayon

Usage

[dependencies]

saker-rs = "0.1.0"


use saker_rs::linarg::operations::sgemm_bias_parallel;
use saker_rs::activations::Activation;

sgemm_bias_parallel(
    m, n, k,
    &weights,
    &input,
    Some(&bias),
    &mut output,
    Activation::Silu,
);

Requirements

  • x86_64 with AVX-512F + FMA for the fast path

License

MIT