1 | /* Copyright 2016 The TensorFlow Authors. All Rights Reserved. |
2 | |
3 | Licensed under the Apache License, Version 2.0 (the "License"); |
4 | you may not use this file except in compliance with the License. |
5 | You may obtain a copy of the License at |
6 | |
7 | http://www.apache.org/licenses/LICENSE-2.0 |
8 | |
9 | Unless required by applicable law or agreed to in writing, software |
10 | distributed under the License is distributed on an "AS IS" BASIS, |
11 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
12 | See the License for the specific language governing permissions and |
13 | limitations under the License. |
14 | ==============================================================================*/ |
15 | |
16 | // Helper functions to run 3d pooling on GPU using CuDNN. |
17 | |
18 | #ifndef TENSORFLOW_CORE_KERNELS_CUDNN_POOLING_GPU_H_ |
19 | #define TENSORFLOW_CORE_KERNELS_CUDNN_POOLING_GPU_H_ |
20 | |
21 | #include <array> |
22 | |
23 | #include "tensorflow/core/framework/op_kernel.h" |
24 | |
25 | #if GOOGLE_CUDA || TENSORFLOW_USE_ROCM |
26 | #include "tensorflow/core/platform/stream_executor.h" |
27 | #endif |
28 | |
29 | #include "tensorflow/core/util/padding.h" |
30 | |
31 | namespace tensorflow { |
32 | |
33 | #if GOOGLE_CUDA || TENSORFLOW_USE_ROCM |
34 | |
35 | // Runs (avg/max)pooling on GPU. |
36 | // Dimension order for all array arguments is: x, y, z. |
37 | template <typename T> |
38 | class DnnPooling3dOp { |
39 | public: |
40 | static void Compute(OpKernelContext* context, |
41 | se::dnn::PoolingMode pooling_mode, |
42 | const std::array<int64, 3>& size, |
43 | const std::array<int64, 3>& stride, |
44 | const std::array<int64, 3>& padding, |
45 | TensorFormat data_format, const Tensor& tensor_in, |
46 | Tensor* output); |
47 | }; |
48 | |
49 | // Computes the gradient of (avg/max)pooling on GPU. |
50 | // Dimension order for all array arguments is: x, y, z. |
51 | template <typename T> |
52 | class DnnPooling3dGradOp { |
53 | public: |
54 | static void Compute(OpKernelContext* context, |
55 | se::dnn::PoolingMode pooling_mode, |
56 | const std::array<int64, 3>& window, |
57 | const std::array<int64, 3>& stride, |
58 | const std::array<int64, 3>& padding, |
59 | const std::array<int64, 3>& output_size, |
60 | TensorFormat data_format, const Tensor& out_backprop, |
61 | const TensorShape& tensor_in_shape, |
62 | const Tensor* tensor_in, const Tensor* tensor_out, |
63 | Tensor* input_backprop); |
64 | }; |
65 | |
66 | #endif |
67 | |
68 | } // namespace tensorflow |
69 | |
70 | #endif // TENSORFLOW_CORE_KERNELS_CUDNN_POOLING_GPU_H_ |
71 | |