1 | /* Copyright 2015 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 | #ifndef TENSORFLOW_CORE_KERNELS_STRIDED_SLICE_OP_H_ |
17 | #define TENSORFLOW_CORE_KERNELS_STRIDED_SLICE_OP_H_ |
18 | |
19 | // Functor definition for StridedSliceOp, must be compilable by nvcc. |
20 | |
21 | #include "third_party/eigen3/unsupported/Eigen/CXX11/Tensor" |
22 | #include "tensorflow/core/framework/resource_handle.h" |
23 | #include "tensorflow/core/framework/tensor_types.h" |
24 | #include "tensorflow/core/framework/variant_encode_decode.h" |
25 | #include "tensorflow/core/platform/types.h" |
26 | #include "tensorflow/core/util/strided_slice_op.h" |
27 | |
28 | namespace tensorflow { |
29 | namespace functor { |
30 | |
31 | template <typename Device, typename T, int NDIMS> |
32 | struct StridedSlice { |
33 | void operator()(const Device& d, typename TTypes<T, NDIMS>::Tensor output, |
34 | typename TTypes<T, NDIMS>::ConstTensor input, |
35 | const Eigen::DSizes<Eigen::DenseIndex, NDIMS>& start_indices, |
36 | const Eigen::DSizes<Eigen::DenseIndex, NDIMS>& stop_indices, |
37 | const Eigen::DSizes<Eigen::DenseIndex, NDIMS>& strides) { |
38 | MaybeWith32BitIndexing<Device>( |
39 | [&](auto output32, auto input32, const auto& start_indices32, |
40 | const auto& stop_indices32, const auto& strides32) { |
41 | output32.device(d) = |
42 | input32.stridedSlice(start_indices32, stop_indices32, strides32); |
43 | }, |
44 | output, input, start_indices, stop_indices, strides); |
45 | } |
46 | }; |
47 | |
48 | template <typename T, int NDIMS, typename Device> |
49 | struct InitOutput { |
50 | static void run(const Device& d, typename TTypes<T, NDIMS>::Tensor output) { |
51 | output.device(d) = output.constant(T(0)); |
52 | } |
53 | }; |
54 | |
55 | template <int NDIMS, typename Device> |
56 | struct InitOutput<ResourceHandle, NDIMS, Device> { |
57 | static void run(const Device& d, |
58 | typename TTypes<ResourceHandle, NDIMS>::Tensor output) { |
59 | output.device(d) = output.constant(ResourceHandle()); |
60 | } |
61 | }; |
62 | |
63 | template <int NDIMS, typename Device> |
64 | struct InitOutput<tstring, NDIMS, Device> { |
65 | static void run(const Device& d, |
66 | typename TTypes<tstring, NDIMS>::Tensor output) { |
67 | output.device(d) = output.constant(tstring()); |
68 | } |
69 | }; |
70 | |
71 | template <typename Device, typename T, int NDIMS> |
72 | struct StridedSliceGrad { |
73 | void operator()(const Device& d, typename TTypes<T, NDIMS>::Tensor output, |
74 | typename TTypes<T, NDIMS>::ConstTensor input, |
75 | const Eigen::DSizes<Eigen::DenseIndex, NDIMS>& start_indices, |
76 | const Eigen::DSizes<Eigen::DenseIndex, NDIMS>& stop_indices, |
77 | const Eigen::DSizes<Eigen::DenseIndex, NDIMS>& strides) { |
78 | InitOutput<T, NDIMS, Device>::run(d, output); |
79 | MaybeWith32BitIndexing<Device>( |
80 | [&](auto output32, const auto& start_indices32, |
81 | const auto& stop_indices32, const auto& strides32) { |
82 | output32.stridedSlice(start_indices32, stop_indices32, strides32) |
83 | .device(d) = input; |
84 | }, |
85 | output, start_indices, stop_indices, strides); |
86 | } |
87 | }; |
88 | |
89 | template <typename Device, typename T, int NDIMS> |
90 | struct StridedSliceAssign { |
91 | void operator()(const Device& d, typename TTypes<T, NDIMS>::Tensor output, |
92 | typename TTypes<T, NDIMS>::ConstTensor input, |
93 | const Eigen::DSizes<Eigen::DenseIndex, NDIMS>& start_indices, |
94 | const Eigen::DSizes<Eigen::DenseIndex, NDIMS>& stop_indices, |
95 | const Eigen::DSizes<Eigen::DenseIndex, NDIMS>& strides, |
96 | const StridedSliceAssignBCast& bcast) { |
97 | MaybeWith32BitIndexing<Device>( |
98 | [&](auto output32, auto input32, const auto& start_indices32, |
99 | const auto& stop_indices32, const auto& strides32) { |
100 | if (bcast.IsBroadcastingRequired()) { |
101 | output32.stridedSlice(start_indices32, stop_indices32, strides32) |
102 | .device(d) = input32.broadcast(bcast.bcast()); |
103 | } else { |
104 | output32.stridedSlice(start_indices32, stop_indices32, strides32) |
105 | .device(d) = input32; |
106 | } |
107 | }, |
108 | output, input, start_indices, stop_indices, strides); |
109 | } |
110 | }; |
111 | |
112 | template <typename Device, typename T> |
113 | struct StridedSliceAssignScalar { |
114 | void operator()(const Device& d, typename TTypes<T, 1>::Tensor output, |
115 | typename TTypes<T, 1>::ConstTensor input) { |
116 | output.device(d) = input; |
117 | } |
118 | }; |
119 | |
120 | } // namespace functor |
121 | } // namespace tensorflow |
122 | |
123 | #endif // TENSORFLOW_CORE_KERNELS_STRIDED_SLICE_OP_H_ |
124 | |