diff options
| author |  Xiaoqiang Zheng <zhengxq@google.com> | 2017-03-21 15:53:52 -0800 |
|---|---|---|
| committer |  TensorFlower Gardener <gardener@tensorflow.org> | 2017-03-21 17:11:32 -0700 |
| commit | be81281b3c9aba3749719e7b3b08cfb51ed55b42 (patch) | |
| tree | 711475ee6227ed24f4aea9887454046434942100 /tensorflow/core/kernels/conv_ops_3d.cc | |
| parent | bc4bed8d08183dbb8232ee89f944e9a44e421782 (diff) | |
Add a multi-pass mode for autotuning. It tests the autotune multiple times,
and only if the winning margin exceeds a certain threshold, it is accepted.
Change: 150822045
Diffstat (limited to 'tensorflow/core/kernels/conv_ops_3d.cc')
| -rw-r--r-- | tensorflow/core/kernels/conv_ops_3d.cc | 4 |
1 files changed, 3 insertions, 1 deletions
diff --git a/tensorflow/core/kernels/conv_ops_3d.cc b/tensorflow/core/kernels/conv_ops_3d.cc index 74b71551a2..dfcb2cfbe2 100644 --- a/tensorflow/core/kernels/conv_ops_3d.cc +++ b/tensorflow/core/kernels/conv_ops_3d.cc @@ -152,7 +152,9 @@ TF_CALL_double(REGISTER_CPU_KERNEL); #if GOOGLE_CUDA // A dummy type to group forward convolution autotune results together. -struct Conv3dAutoTuneGroup {}; +struct Conv3dAutoTuneGroup { + static string name() { return "Conv3d"; } +}; typedef AutoTuneSingleton<Conv3dAutoTuneGroup, ConvParameters, perftools::gputools::dnn::AlgorithmConfig> AutoTuneConv3d; |