Skip to content

Latest commit

 

History

History
52 lines (36 loc) · 1.2 KB

README.md

File metadata and controls

52 lines (36 loc) · 1.2 KB
Raw

hamming_weight

C library to compute the Hamming weight of arrays. The Hamming weight is the number of ones in a stream of bits. Computing this count quickly has important applications in indexing, machine learning, cryptography and so forth.

Library has several highly optimized implementations, which use popcnt, SSE, AVX2, and AVX512 instructions.

Please note that an AVX512F variant is under development, AVX512BW is just a proof of concept (no hardware support yet), and AVX512CD was written to show it's feasible.

Paper

Faster Population Counts using AVX2 Instructions

Related library

Usage

make
./unit
./basic_benchmark

Building

It's assumed that target CPU supports both AVX2 and popcnt.

  • If a CPU doesn't support AVX2 define SSE.
  • If a CPU doesn't support popcnt instruction then define NOPOPCNT.
  • If you want to build AVX512 variants, define AVX512F or AVX512BW or AVX512CD.
  • If you want to build for AMD XOP, define XOP.

For example:

export SSE=1 NOPOPCNT=1; make
./basic_benchmark

export AVX512F=1; make
./basic_benchmark