Overview
SimSIMD is an open‑source, SIMD‑accelerated library for fast similarity and distance computations on vectors. It targets use cases such as vector search, vector databases, and high‑performance numerical workloads, providing a faster alternative to libraries like scipy.spatial.distance and numpy.inner.
Key Capabilities
- Implements dot products and similarity metrics
- Supports multiple numeric types:
f64, f32, f16, real & complex, i8, and bit vectors
- SIMD acceleration across multiple CPU instruction sets: AVX2, AVX‑512, NEON, SVE, and SVE2
- Achieves large speedups (reported up to ~200x) over naïve or non‑SIMD implementations for inner products and distances
- Designed for integration into vector search and vector database pipelines
- Core implementation in C / C++ (via
include and c directories)
- Bindings / integrations for multiple languages:
- Python
- Rust
- JavaScript
- Swift
- Go (Golang)
Features
-
SIMD‑Accelerated Operations
- Vector inner products (dot products)
- Similarity and distance metrics for numerical vectors
- Optimized kernels per CPU architecture (AVX2, AVX‑512, NEON, SVE, SVE2)
-
Data Type Support
- Floating‑point:
f64, f32, f16
- Real and complex numbers
- Integer:
i8
- Bit vectors
-
Multi‑Language Bindings
- Python package for numerical workloads and vector search pipelines
- Rust crate for systems and performance‑critical applications
- JavaScript bindings for web or Node.js environments
- Swift bindings for Apple platforms
- Go bindings for Go‑based services
-
Cross‑Platform / Cross‑Architecture Design
- Portable SIMD abstractions targeting multiple CPU instruction sets
- Suitable for x86 and ARM ecosystems (via AVX*, NEON, SVE, SVE2)
-
Open‑Source & Integration‑Friendly
- Source available on GitHub
- Modular layout with language‑specific folders for easier integration
- Apache‑2.0 license for permissive use in open‑source and commercial projects
License
Pricing
- SimSIMD is an open‑source library under the Apache‑2.0 license and is free to use. No pricing plans are listed.
