[cudnn] implement different precission

undertherain / benchmarker

modular framework for [not only] deep learning performance benchmarking

http://blackbird.pw/performance

Mozilla Public License 2.0

9 stars 5 forks source link

Open vatai opened 4 years ago

vatai commented 4 years ago

Everything is hardwired now to be single precision. Half and mixed precision should also be supported.