Google Git
Sign in
android / platform / external / FP16 / 2cc626c378b7a58cc8ca25d3686fa963e765f6ee
commit2cc626c378b7a58cc8ca25d3686fa963e765f6ee[log] [tgz]
authorIan Hua <ianhua@google.com>Thu Aug 12 11:16:47 2021 +0100
committerIan Hua <ianhua@google.com>Wed Nov 10 22:02:05 2021 +0000
treedf61340909aacfee61ba9b551e63df7ae3d3eceb
parent692d39568e39535f3772b5ffaa9be64e605ef24a [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 AOSP Ia8ac144df6a18f17cc9c181ed8618a5d365abaa6

Bug: 196532440
Test: mma
Change-Id: Ia8ac144df6a18f17cc9c181ed8618a5d365abaa6
Merged-In: Ia8ac144df6a18f17cc9c181ed8618a5d365abaa6
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.