Google Git
Sign in
android / platform / external / FP16 / refs/heads/android12L-tests-dev
commitbbf0513965aa9c53af60f61f708f2597dfeb04f8[log] [tgz]
authorIan Hua <ianhua@google.com>Thu Aug 12 11:16:47 2021 +0100
committerMiao Wang <miaowang@google.com>Mon May 23 18:07:37 2022 +0000
treedf61340909aacfee61ba9b551e63df7ae3d3eceb
parent015fd0f2a06d43d5274e6d9a9675f7d575df0316 [diff]
Add host_supported, min_sdk_version and apex_available configs to Android.bp for FP16.

This would solve build erros such as
"missing variant:os:linux_glibc,arch:x86_64" when adding
fp16_headers to external/tensorflow repo.

Cherrypicked from Ia8ac144df6a18f17cc9c181ed8618a5d365abaa6

Bug: 206089870
Test: mma
Change-Id: Ia8ac144df6a18f17cc9c181ed8618a5d365abaa6
(cherry picked from commit af3911c8e40bf2383767bd32506488fd85ae84d1)
1 file changed
tree: df61340909aacfee61ba9b551e63df7ae3d3eceb
  1. bench/
  2. cmake/
  3. include/
  4. test/
  5. .gitignore
  6. Android.bp
  7. CMakeLists.txt
  8. configure.py
  9. confu.yaml
  10. LICENSE
  11. METADATA
  12. MODULE_LICENSE_MIT
  13. OWNERS
  14. README.md
  15. TEST_MAPPING
README.md

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.