rok commented on issue #22806: URL: https://github.com/apache/arrow/issues/22806#issuecomment-1510507313
I've looked at some libraries to see what is usually used: * [TensorFlow](https://github.com/tensorflow/tensorflow/blob/master/third_party/FP16/workspace.bzl) - uses FP16 * [Numpy](https://github.com/numpy/numpy/blob/7e86c2aadfc6dcee12274db698407275cede1b63/numpy/core/src/common/half.hpp#L2) seems to implement it's own arithmetic. * [PyTorch](https://github.com/pytorch/pytorch/tree/be0b12ece576c86c5f059d15a64dcba0eb886ddd/third_party) - uses FP16 * [FP16](https://github.com/Maratyszcza/FP16/) - seems to be [composed](https://github.com/Maratyszcza/FP16/tree/master/third-party) mostly of [half](https://half.sourceforge.net/) and few functions taken from other places. Half [seems pretty complete](https://half.sourceforge.net/namespacehalf__float.html) so I think it or [FP16](https://github.com/Maratyszcza/FP16) would make good candidates. cc @pitrou @bkietz -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
