/awesome-simd

A curated list of awesome SIMD frameworks, libraries and software

awesome-simd

A curated list of awesome SIMD frameworks, libraries and software.

This list showcases projects that have achieved 10x performance improvements using SIMD (Single Instruction Multiple Data) instructions. In general this should lead to execution speeds of GBs per second on modern CPUs for that task at hand.

Database / Structures

  • sneller - Fast SQL for JSON in Go with AVX-512: fast, simple, schemaless
  • bitmap - Go: Dense, zero-allocation, SIMD-enabled bitmap/bitset

Parsing

  • simdjson - C++: Parsing gigabytes of JSON per second
  • dictionary - C++: High-performance dictionary coding
  • simdjson-go - Go: Parsing gigabytes of JSON per second
  • simdcomp - C: A simple library for compressing lists of integers using binary packing
  • SIMDCompressionAndIntersection - C++: A library to compress and intersect sorted lists of integers using SIMD instructions
  • simdutf - C++: Unicode routines (UTF8, UTF16, UTF32)
  • Ada - C++: WHATWG-compliant and fast URL parser
  • StringZilla - C: Substring search, edit-distances, sorting, fuzzy matching, etc.
  • Hyperscan - C++: High-performance regular expression matching library
  • Various string algo's - C: Repository for string algorithms, snippets, toy programs, etc.
  • sse-popcount - SIMD (SSE) population count

Erasure Coding and Hashing

  • xxHash - C: Extremely fast Hash algorithm, processing at RAM speed limits.
  • Reed-Solomon - Go: Erasure Coding in Go
  • highwayhash - Go: Optimized HighwayHash implementation for Intel (over 10 GB/sec), ARM and Power9
  • sha256-simd - Go: Optimized SHA256 computations for Intel, ARM and Power9

Neural Network

  • ncnn - C++: High-performance NN inference framework optimized for mobile
  • mkl-dnn - C++: Math Kernel Library for Deep Neural Networks
  • nnpack - C/c++: Acceleration package for neural networks on multi-core CPUs
  • SimSIMD - C: Similarity measures for high-dimensional vectors

Image processing

  • Simd - C++: image processing library making use of SIMD
  • Pillow-SIMD - Python: SIMD version of PIL (Python Imaging Library)
  • ComputeLibrary - C++: Library for Computer Vision and Machine Learning (ARM only)

Classes

  • HPC-Class - High Performance Computing (HPC) class taught at FSU Jena by the Scalable Data- and Compute-intensive Analyses lab

Cool

Blogs

Links

Tools

  • avo - Go: Generate x86 Assembly with Go
  • PeachPy - Python: x86-64 assembler embedded in Python
  • c2goasm - Go: C to Go Assembly
  • LLVM MCA - LLVM Machine Code Analyzer
  • Highway - C++: Performance-portable, length-agnostic SIMD with runtime dispatch
  • Eve - C++: Expressive Vector Engine
  • SIMDe - C++: Header-only implementations of SIMD instruction sets (SSE*, AVX{,2,512}, Neon, and more) for systems which don't natively support them.
  • xsimd - C++: Wrappers for SIMD intrinsics and math implementations (SSE, AVX, NEON, AVX512)
  • Intel SDE debugging - Debugging with AVX-512
  • Asm-Dude - VS extension for assembly syntax highlighting and code completion
  • Intrinsics-Dude - VS extension for compiler instrinsics in C/C++
  • Intel® Implicit SPMD Program Compiler - An LLVM compiler for a C like language, with C linkage, that generates very good SIMD instructions for a wide range of platforms and ISAs. (Windows, iOS, Linux, ARM, PS5, Xbox, SSE, AVX, AVX2, AVX-512, ARM NEON.)

Online tools

AVX-512

ARM64 NEON

ARM64 SVE

License

Creative Commons License