diff options
| author |  A. Unique TensorFlower <nobody@tensorflow.org> | 2016-04-27 14:14:18 -0800 |
|---|---|---|
| committer |  TensorFlower Gardener <gardener@tensorflow.org> | 2016-04-27 15:21:22 -0700 |
| commit | afcbe82da42b433a06a19c17260a59282b25cc61 (patch) | |
| tree | 99d1153d5839afb28ee0aa1aeec40051da859cd6 /WORKSPACE | |
| parent | cb78f99de569c13dcc0fc66c555cd77c8884f79d (diff) | |
Switch CuDNN avg pooling to not use padding when computing the average.
This brings the CuDNN kernel in line with the Eigen-based CPU/GPU avg pooling implementation. This kernel is currently only used if the input tensor is in the NCHW format.
Change: 120959680
Diffstat (limited to 'WORKSPACE')
0 files changed, 0 insertions, 0 deletions