A C++ library to compress and intersect sorted lists of integers using SIMD instructions
Switch branches/tags
Nothing to show
Clone or download
Latest commit 6ca7bc7 Dec 9, 2017



Build Status

As the name suggests, this is a C/C++ library for fast compression and intersection of lists of sorted integers using SIMD instructions. The library focuses on innovative techniques and very fast schemes, with particular attention to differential coding. It introduces new SIMD intersections schemes such as SIMD Galloping.

This library can decode at least 4 billions of compressed integers per second on most desktop or laptop processors. That is, it can decompress data at a rate of 15 GB/s. This is significantly faster than generic codecs like gzip, LZO, Snappy or LZ4.

Authors: Leonid Boystov, Nathan Kurz, Daniel Lemire, Owen Kaser, Andrew Consroe, Shlomi Vaknin, Christoph Rupp, Bradley Grainger, and others.


This work has also inspired other work such as...

Simple demo

Check out example.cpp

You can run it like so:

make example ./example


make ./unit

To run tests, you can do


(follow the instructions)

For a simple C library

This library is a C++ research library. For something simpler, written in C, see:


Comparison with the FastPFOR C++ library

The FastPFOR C++ Library available at https://github.com/lemire/FastPFor implements some of the same compression schemes except that it is not optimized for the compression of sorted lists of integers.

Other recommended libraries


Apache License, Version 2.0

As far as the authors know, this work is patent-free.


A CPU (AMD or Intel) with support for SSE2 (Pentium 4 or better) is required while a CPU with SSE 4.1* (Penryn [2007] processors or better) is recommended.

A recent GCC (4.7 or better), Clang, Intel or Visual C++ compiler.

A processor support AVX (Intel or AMD).

Tested on Linux, MacOS and Windows. It should be portable to other platforms.

*- The default makefile might assume AVX support, but AVX is not required. For GCC compilers you might need the -msse2 flag, but you will not need the -mavx flag.

For advanced benchmarking, please see


where there is additional information as well as links to real data sets.


Thanks to Kelly Sommers for useful feedback.

This work was supported by NSERC grant number 26143.