[go: up one dir, main page]
Google Git
Sign in
chromium / external / github.com / Maratyszcza / FP16
Clone this repo:

Branches

  1. 3d2de18 Add missing standard lib includes to tests by Guy Hershenbaum · 3 months ago master
  2. 84faa71 Fix compiling intrinsics with MSVC. by Brendan Dahl · 3 months ago
  3. a7e499d Fix bitcasts with clang on x86 by Austin Sullivan · 3 months ago
  4. 98b0a46 Support native conversions without __fp16/_Float16 types by Marat Dukhan · 1 year, 4 months ago
  5. 0854aa1 Use Intel bitcast intrinsics with recent MSVC by Marat Dukhan · 1 year, 5 months ago

FP16

Header-only library for conversion to/from half-precision floating point formats

Features

  • Supports IEEE and ARM alternative half-precision floating-point format
    • Property converts infinities and NaNs
    • Properly converts denormal numbers, even on systems without denormal support
  • Header-only library, no installation or build required
  • Compatible with C99 and C++11
  • Fully covered with unit tests and microbenchmarks

Acknowledgements

HPC Garage logo Georgia Tech College of Computing logo

The library is developed by Marat Dukhan of Georgia Tech. FP16 is a research project at Richard Vuduc's HPC Garage lab in the Georgia Institute of Technology, College of Computing, School of Computational Science and Engineering.

This material is based upon work supported by the U.S. National Science Foundation (NSF) Award Number 1339745. Any opinions, findings and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect those of NSF.