1 | /* Copyright 2019 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 | #include <stdint.h> |
16 | |
17 | #include "tensorflow/lite/c/c_api_types.h" |
18 | #include "tensorflow/lite/c/common.h" |
19 | #include "tensorflow/lite/kernels/internal/reference/reference_ops.h" |
20 | #include "tensorflow/lite/kernels/internal/tensor.h" |
21 | #include "tensorflow/lite/kernels/internal/tensor_ctypes.h" |
22 | #include "tensorflow/lite/kernels/internal/types.h" |
23 | #include "tensorflow/lite/kernels/kernel_util.h" |
24 | |
25 | namespace tflite { |
26 | namespace ops { |
27 | namespace builtin { |
28 | namespace where { |
29 | |
30 | constexpr int kInputConditionTensor = 0; |
31 | constexpr int kOutputTensor = 0; |
32 | |
33 | template <typename T> |
34 | TfLiteStatus ResizeOutputTensor(TfLiteContext* context, |
35 | const TfLiteTensor* cond_tensor, |
36 | TfLiteTensor* output_tensor) { |
37 | // Output tensor should have shape: |
38 | // (num_true, cond_rank), where num_true denotes the number of true values |
39 | // in condition. |
40 | const RuntimeShape& cond_shape = GetTensorShape(cond_tensor); |
41 | const int size = cond_shape.FlatSize(); |
42 | const int cond_rank = cond_shape.DimensionsCount(); |
43 | const T* cond_data = GetTensorData<T>(cond_tensor); |
44 | |
45 | int true_count = 0; |
46 | for (int i = 0; i < size; ++i) { |
47 | if (cond_data[i] != T(0)) { |
48 | true_count++; |
49 | } |
50 | } |
51 | TfLiteIntArray* output_dims = TfLiteIntArrayCreate(2); |
52 | output_dims->data[0] = true_count; |
53 | output_dims->data[1] = cond_rank; |
54 | return context->ResizeTensor(context, output_tensor, output_dims); |
55 | } |
56 | |
57 | template <typename T> |
58 | TfLiteStatus PrepareOutput(TfLiteContext* context, |
59 | const TfLiteTensor* cond_tensor, |
60 | TfLiteTensor* output) { |
61 | // As output will be a 2D tensor of indices, use int64 to be consistent with |
62 | // tensorflow. |
63 | output->type = kTfLiteInt64; |
64 | |
65 | // Exit early if cond is a non-const tensor. Set output tensor to dynamic so |
66 | // output size can be determined in Eval. |
67 | if (!IsConstantTensor(cond_tensor)) { |
68 | SetTensorToDynamic(output); |
69 | return kTfLiteOk; |
70 | } |
71 | return ResizeOutputTensor<T>(context, cond_tensor, output); |
72 | } |
73 | |
74 | TfLiteStatus Prepare(TfLiteContext* context, TfLiteNode* node) { |
75 | TF_LITE_ENSURE_EQ(context, NumInputs(node), 1); |
76 | TF_LITE_ENSURE_EQ(context, NumOutputs(node), 1); |
77 | |
78 | const TfLiteTensor* cond_tensor; |
79 | TF_LITE_ENSURE_OK(context, GetInputSafe(context, node, kInputConditionTensor, |
80 | &cond_tensor)); |
81 | TfLiteTensor* output; |
82 | TF_LITE_ENSURE_OK(context, |
83 | GetOutputSafe(context, node, kOutputTensor, &output)); |
84 | |
85 | switch (cond_tensor->type) { |
86 | case kTfLiteBool: |
87 | return PrepareOutput<bool>(context, cond_tensor, output); |
88 | case kTfLiteFloat32: |
89 | return PrepareOutput<float>(context, cond_tensor, output); |
90 | case kTfLiteInt64: |
91 | return PrepareOutput<int64_t>(context, cond_tensor, output); |
92 | case kTfLiteInt32: |
93 | return PrepareOutput<int32_t>(context, cond_tensor, output); |
94 | case kTfLiteInt8: |
95 | return PrepareOutput<int8_t>(context, cond_tensor, output); |
96 | case kTfLiteUInt8: |
97 | return PrepareOutput<uint8_t>(context, cond_tensor, output); |
98 | case kTfLiteUInt32: |
99 | return PrepareOutput<uint32_t>(context, cond_tensor, output); |
100 | default: |
101 | TF_LITE_KERNEL_LOG(context, |
102 | "Condition tensor has unsupported type: '%s'." , |
103 | TfLiteTypeGetName(cond_tensor->type)); |
104 | } |
105 | return kTfLiteOk; |
106 | } |
107 | |
108 | TfLiteStatus Eval(TfLiteContext* context, TfLiteNode* node) { |
109 | const TfLiteTensor* cond_tensor; |
110 | TF_LITE_ENSURE_OK(context, GetInputSafe(context, node, kInputConditionTensor, |
111 | &cond_tensor)); |
112 | TfLiteTensor* output; |
113 | TF_LITE_ENSURE_OK(context, |
114 | GetOutputSafe(context, node, kOutputTensor, &output)); |
115 | |
116 | if (IsDynamicTensor(output)) { |
117 | switch (cond_tensor->type) { |
118 | case kTfLiteBool: |
119 | TF_LITE_ENSURE_OK( |
120 | context, ResizeOutputTensor<bool>(context, cond_tensor, output)); |
121 | break; |
122 | case kTfLiteFloat32: |
123 | TF_LITE_ENSURE_OK( |
124 | context, ResizeOutputTensor<float>(context, cond_tensor, output)); |
125 | break; |
126 | case kTfLiteInt64: |
127 | TF_LITE_ENSURE_OK( |
128 | context, ResizeOutputTensor<int64_t>(context, cond_tensor, output)); |
129 | break; |
130 | case kTfLiteInt32: |
131 | TF_LITE_ENSURE_OK( |
132 | context, ResizeOutputTensor<int32_t>(context, cond_tensor, output)); |
133 | break; |
134 | case kTfLiteInt8: |
135 | TF_LITE_ENSURE_OK( |
136 | context, ResizeOutputTensor<int8_t>(context, cond_tensor, output)); |
137 | break; |
138 | case kTfLiteUInt8: |
139 | TF_LITE_ENSURE_OK( |
140 | context, ResizeOutputTensor<uint8_t>(context, cond_tensor, output)); |
141 | break; |
142 | case kTfLiteUInt32: |
143 | TF_LITE_ENSURE_OK(context, ResizeOutputTensor<uint32_t>( |
144 | context, cond_tensor, output)); |
145 | break; |
146 | default: |
147 | TF_LITE_KERNEL_LOG(context, |
148 | "Condition tensor has unsupported type: '%s'." , |
149 | TfLiteTypeGetName(cond_tensor->type)); |
150 | } |
151 | } |
152 | |
153 | TfLiteIntArray* dims = cond_tensor->dims; |
154 | if (dims->size == 0) { |
155 | // Scalar tensors are not supported. |
156 | TF_LITE_KERNEL_LOG(context, "Where op requires condition w/ rank > 0" ); |
157 | return kTfLiteError; |
158 | } |
159 | |
160 | switch (cond_tensor->type) { |
161 | case kTfLiteBool: |
162 | reference_ops::SelectTrueCoords(GetTensorShape(cond_tensor), |
163 | GetTensorData<bool>(cond_tensor), |
164 | GetTensorData<int64_t>(output)); |
165 | break; |
166 | case kTfLiteFloat32: |
167 | reference_ops::SelectTrueCoords(GetTensorShape(cond_tensor), |
168 | GetTensorData<float>(cond_tensor), |
169 | GetTensorData<int64_t>(output)); |
170 | break; |
171 | case kTfLiteInt64: |
172 | reference_ops::SelectTrueCoords(GetTensorShape(cond_tensor), |
173 | GetTensorData<int64_t>(cond_tensor), |
174 | GetTensorData<int64_t>(output)); |
175 | break; |
176 | case kTfLiteInt32: |
177 | reference_ops::SelectTrueCoords(GetTensorShape(cond_tensor), |
178 | GetTensorData<int32_t>(cond_tensor), |
179 | GetTensorData<int64_t>(output)); |
180 | break; |
181 | case kTfLiteInt8: |
182 | reference_ops::SelectTrueCoords(GetTensorShape(cond_tensor), |
183 | GetTensorData<int8_t>(cond_tensor), |
184 | GetTensorData<int64_t>(output)); |
185 | break; |
186 | case kTfLiteUInt8: |
187 | reference_ops::SelectTrueCoords(GetTensorShape(cond_tensor), |
188 | GetTensorData<uint8_t>(cond_tensor), |
189 | GetTensorData<int64_t>(output)); |
190 | break; |
191 | case kTfLiteUInt32: |
192 | reference_ops::SelectTrueCoords(GetTensorShape(cond_tensor), |
193 | GetTensorData<uint32_t>(cond_tensor), |
194 | GetTensorData<int64_t>(output)); |
195 | break; |
196 | default: |
197 | TF_LITE_KERNEL_LOG(context, |
198 | "Condition tensor has unsupported type: '%s'." , |
199 | TfLiteTypeGetName(cond_tensor->type)); |
200 | } |
201 | return kTfLiteOk; |
202 | } |
203 | } // namespace where |
204 | |
205 | TfLiteRegistration* Register_WHERE() { |
206 | static TfLiteRegistration r = {/*init*/ nullptr, /*free*/ nullptr, |
207 | where::Prepare, where::Eval}; |
208 | return &r; |
209 | } |
210 | |
211 | } // namespace builtin |
212 | } // namespace ops |
213 | } // namespace tflite |
214 | |