This package implements integer compression using the Stream VByte algorithm.

"Stream VByte: Faster Byte-Oriented Integer Compression" Daniel Lemire, Nathan Kurz, Christoph Rupp Information Processing Letters Volume 130, February 2018, Pages 1-6 https://arxiv.org/abs/1709.08990

The focus of this method is fast decoding speed on processors with SIMD instructions. As such, this package contains fast assembly implementations for decoding on amd64 processors with SSE3 instructions. Other processors will use the slower pure go implementation.

Assembly implementations were generated using the excellent avo

Reference benchmarks on 1,000,000 Zipfian-distributed 32-bit integers.

Intel(R) Core(TM) i3-4010U CPU @ 1.70GHz go version go1.14.2 linux/amd64

For reference, a pure memory copy of the uncompressed data runs at 6.58GB/s ± 2%.