https://developer.nvidia.com/cudnn
以下、リリースノートから引用。
UPDATES SINCE RELEASE CANDIDATE
- The API of cudnnBatchNormalizationBackward has been changed to include an additional set of scaling parameters (alphaParamsDiff and betaParamsDiff) applied to the dBnScaleResult and dBnBiasResult outputs of the function.
- The prior restriction of batch size 512 in all Batch Normalization routines has been removed.
- Numerical stability and performance of cudnnBatchNormalizationBackward in some cases has been improved.
- Performance of cudnnConvolutionBackwardFilter when using Algo 1 has been improved for some cases. This code path now also requires a workspace.