xref: /aosp_15_r20/external/tensorflow/tensorflow/lite/kernels/depth_to_space.cc (revision b6fb3261f9314811a0f4371741dbb8839866f948)
1 /* Copyright 2017 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/builtin_op_data.h"
18 #include "tensorflow/lite/c/common.h"
19 #include "tensorflow/lite/kernels/internal/optimized/optimized_ops.h"
20 #include "tensorflow/lite/kernels/internal/reference/reference_ops.h"
21 #include "tensorflow/lite/kernels/internal/tensor.h"
22 #include "tensorflow/lite/kernels/internal/tensor_ctypes.h"
23 #include "tensorflow/lite/kernels/internal/types.h"
24 #include "tensorflow/lite/kernels/kernel_util.h"
25 
26 namespace tflite {
27 namespace ops {
28 namespace builtin {
29 namespace depth_to_space {
30 
31 // This file has two implementation of DepthToSpace. Note that DepthToSpace only
32 // works on 4D tensors.
33 enum KernelType {
34   kReference,
35   kGenericOptimized,
36 };
37 
38 constexpr int kInputTensor = 0;
39 constexpr int kOutputTensor = 0;
40 
Prepare(TfLiteContext * context,TfLiteNode * node)41 TfLiteStatus Prepare(TfLiteContext* context, TfLiteNode* node) {
42   auto* params =
43       reinterpret_cast<TfLiteDepthToSpaceParams*>(node->builtin_data);
44 
45   TF_LITE_ENSURE_EQ(context, NumInputs(node), 1);
46   TF_LITE_ENSURE_EQ(context, NumOutputs(node), 1);
47 
48   const TfLiteTensor* input;
49   TF_LITE_ENSURE_OK(context, GetInputSafe(context, node, kInputTensor, &input));
50   TfLiteTensor* output;
51   TF_LITE_ENSURE_OK(context,
52                     GetOutputSafe(context, node, kOutputTensor, &output));
53 
54   TF_LITE_ENSURE_EQ(context, NumDimensions(input), 4);
55 
56   auto data_type = output->type;
57   TF_LITE_ENSURE(context,
58                  data_type == kTfLiteFloat32 || data_type == kTfLiteUInt8 ||
59                      data_type == kTfLiteInt8 || data_type == kTfLiteInt32 ||
60                      data_type == kTfLiteInt64);
61   TF_LITE_ENSURE_TYPES_EQ(context, input->type, output->type);
62 
63   const int block_size = params->block_size;
64   TF_LITE_ENSURE(context, block_size > 0);
65   const int input_height = input->dims->data[1];
66   const int input_width = input->dims->data[2];
67   const int input_channels = input->dims->data[3];
68   int output_height = input_height * block_size;
69   int output_width = input_width * block_size;
70   int output_channels = input_channels / block_size / block_size;
71 
72   TF_LITE_ENSURE_EQ(context, input_height, output_height / block_size);
73   TF_LITE_ENSURE_EQ(context, input_width, output_width / block_size);
74   TF_LITE_ENSURE_EQ(context, input_channels,
75                     output_channels * block_size * block_size);
76 
77   TfLiteIntArray* output_size = TfLiteIntArrayCreate(4);
78   output_size->data[0] = input->dims->data[0];
79   output_size->data[1] = output_height;
80   output_size->data[2] = output_width;
81   output_size->data[3] = output_channels;
82 
83   return context->ResizeTensor(context, output, output_size);
84 }
85 
86 template <KernelType kernel_type>
Eval(TfLiteContext * context,TfLiteNode * node)87 TfLiteStatus Eval(TfLiteContext* context, TfLiteNode* node) {
88   auto* params =
89       reinterpret_cast<TfLiteDepthToSpaceParams*>(node->builtin_data);
90 
91   const TfLiteTensor* input;
92   TF_LITE_ENSURE_OK(context, GetInputSafe(context, node, kInputTensor, &input));
93   TfLiteTensor* output;
94   TF_LITE_ENSURE_OK(context,
95                     GetOutputSafe(context, node, kOutputTensor, &output));
96 
97 #define TF_LITE_DEPTH_TO_SPACE(type, scalar)                               \
98   tflite::DepthToSpaceParams op_params;                                    \
99   op_params.block_size = params->block_size;                               \
100   type::DepthToSpace(op_params, GetTensorShape(input),                     \
101                      GetTensorData<scalar>(input), GetTensorShape(output), \
102                      GetTensorData<scalar>(output))
103   switch (input->type) {  // Already know in/out types are same.
104     case kTfLiteFloat32:
105       if (kernel_type == kReference) {
106         TF_LITE_DEPTH_TO_SPACE(reference_ops, float);
107       } else {
108         TF_LITE_DEPTH_TO_SPACE(optimized_ops, float);
109       }
110       break;
111     case kTfLiteUInt8:
112       if (kernel_type == kReference) {
113         TF_LITE_DEPTH_TO_SPACE(reference_ops, uint8_t);
114       } else {
115         TF_LITE_DEPTH_TO_SPACE(optimized_ops, uint8_t);
116       }
117       break;
118     case kTfLiteInt8:
119       if (kernel_type == kReference) {
120         TF_LITE_DEPTH_TO_SPACE(reference_ops, int8_t);
121       } else {
122         TF_LITE_DEPTH_TO_SPACE(optimized_ops, int8_t);
123       }
124       break;
125     case kTfLiteInt32:
126       if (kernel_type == kReference) {
127         TF_LITE_DEPTH_TO_SPACE(reference_ops, int32_t);
128       } else {
129         TF_LITE_DEPTH_TO_SPACE(optimized_ops, int32_t);
130       }
131       break;
132     case kTfLiteInt64:
133       if (kernel_type == kReference) {
134         TF_LITE_DEPTH_TO_SPACE(reference_ops, int64_t);
135       } else {
136         TF_LITE_DEPTH_TO_SPACE(optimized_ops, int64_t);
137       }
138       break;
139     default:
140       TF_LITE_KERNEL_LOG(context, "Type '%s' not currently supported.",
141                          TfLiteTypeGetName(input->type));
142       return kTfLiteError;
143   }
144 #undef TF_LITE_DEPTH_TO_SPACE
145 
146   return kTfLiteOk;
147 }
148 
149 }  // namespace depth_to_space
150 
Register_DEPTH_TO_SPACE_REF()151 TfLiteRegistration* Register_DEPTH_TO_SPACE_REF() {
152   static TfLiteRegistration r = {
153       nullptr, nullptr, depth_to_space::Prepare,
154       depth_to_space::Eval<depth_to_space::kReference>};
155   return &r;
156 }
157 
Register_DEPTH_TO_SPACE_GENERIC_OPT()158 TfLiteRegistration* Register_DEPTH_TO_SPACE_GENERIC_OPT() {
159   static TfLiteRegistration r = {
160       nullptr, nullptr, depth_to_space::Prepare,
161       depth_to_space::Eval<depth_to_space::kGenericOptimized>};
162   return &r;
163 }
164 
Register_DEPTH_TO_SPACE()165 TfLiteRegistration* Register_DEPTH_TO_SPACE() {
166   return Register_DEPTH_TO_SPACE_GENERIC_OPT();
167 }
168 
169 }  // namespace builtin
170 }  // namespace ops
171 }  // namespace tflite
172