1 /* Copyright 2018 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 #include <map> 17 18 #include "tensorflow/lite/c/builtin_op_data.h" 19 #include "tensorflow/lite/c/c_api_internal.h" 20 #include "tensorflow/lite/kernels/internal/reference/reference_ops.h" 21 #include "tensorflow/lite/kernels/internal/tensor.h" 22 #include "tensorflow/lite/kernels/kernel_util.h" 23 24 namespace tflite { 25 namespace ops { 26 namespace builtin { 27 namespace unique { 28 29 void* Init(TfLiteContext* context, const char* buffer, size_t length) { 30 return nullptr; 31 } 32 33 void Free(TfLiteContext* context, void* buffer) {} 34 35 TfLiteStatus Prepare(TfLiteContext* context, TfLiteNode* node) { 36 static const int kOutputUniqueTensor = 0; 37 static const int kOutputIndexTensor = 1; 38 39 TF_LITE_ENSURE_EQ(context, NumInputs(node), 1); 40 TF_LITE_ENSURE_EQ(context, NumOutputs(node), 2); 41 const TfLiteTensor* input = GetInput(context, node, 0); 42 TfLiteTensor* output_unique_tensor = 43 GetOutput(context, node, kOutputUniqueTensor); 44 TfLiteTensor* output_index_tensor = 45 GetOutput(context, node, kOutputIndexTensor); 46 47 // The op only supports 1D input. 48 TF_LITE_ENSURE_EQ(context, NumDimensions(input), 1); 49 TfLiteIntArray* output_index_shape = TfLiteIntArrayCopy(input->dims); 50 // The unique values are determined during evaluation, so we don't know yet 51 // the size of the output tensor. 52 SetTensorToDynamic(output_unique_tensor); 53 return context->ResizeTensor(context, output_index_tensor, 54 output_index_shape); 55 } 56 57 namespace { 58 59 // Actual evaluation for the unique op. 60 template <typename T, typename I> 61 TfLiteStatus EvalImpl(TfLiteContext* context, const TfLiteTensor* input, 62 TfLiteNode* node) { 63 // Map from value, to index in the unique elements vector. 64 // Note that we prefer to use map than unordered_map as it showed less 65 // increase in the binary size. 66 std::map<T, int> unique_values; 67 TfLiteTensor* output_indexes = GetOutput(context, node, 1); 68 I* indexes = GetTensorData<I>(output_indexes); 69 const T* data = GetTensorData<T>(input); 70 const int num_elements = NumElements(input); 71 72 for (int i = 0; i < num_elements; ++i) { 73 const auto element_it = unique_values.find(data[i]); 74 if (element_it != unique_values.end()) { 75 indexes[i] = element_it->second; 76 } else { 77 const int unique_index = unique_values.size(); 78 unique_values[data[i]] = unique_index; 79 indexes[i] = unique_index; 80 } 81 } 82 // Allocate output tensor. 83 TfLiteTensor* unique_output = GetOutput(context, node, 0); 84 std::unique_ptr<TfLiteIntArray, void (*)(TfLiteIntArray*)> shape( 85 TfLiteIntArrayCreate(NumDimensions(input)), TfLiteIntArrayFree); 86 shape->data[0] = unique_values.size(); 87 TF_LITE_ENSURE_STATUS( 88 context->ResizeTensor(context, unique_output, shape.release())); 89 // Set the values in the output tensor. 90 T* output_unique_values = GetTensorData<T>(unique_output); 91 for (int i = 0; i < unique_values.size(); ++i) { 92 output_unique_values[i] = data[indexes[i]]; 93 } 94 return kTfLiteOk; 95 } 96 97 template <typename T> 98 TfLiteStatus EvalImpl(TfLiteContext* context, const TfLiteTensor* input, 99 TfLiteNode* node) { 100 auto* params = reinterpret_cast<TfLiteUniqueParams*>(node->builtin_data); 101 if (params == nullptr) { 102 context->ReportError(context, "Null params passed"); 103 return kTfLiteError; 104 } 105 switch (params->index_out_type) { 106 case kTfLiteInt32: 107 return EvalImpl<T, int32_t>(context, input, node); 108 case kTfLiteInt64: 109 return EvalImpl<T, int64_t>(context, input, node); 110 default: 111 context->ReportError( 112 context, 113 "Unique index output array can only be Int32 or In64, requested: ", 114 TfLiteTypeGetName(params->index_out_type)); 115 } 116 return kTfLiteError; 117 } 118 119 } // namespace 120 121 TfLiteStatus Eval(TfLiteContext* context, TfLiteNode* node) { 122 const TfLiteTensor* input = GetInput(context, node, 0); 123 TfLiteTensor* output_index_tensor = GetOutput(context, node, 1); 124 TF_LITE_ENSURE_EQ(context, NumElements(output_index_tensor), 125 NumElements(input)); 126 127 switch (input->type) { 128 case kTfLiteInt8: 129 TF_LITE_ENSURE_STATUS(EvalImpl<int8_t>(context, input, node)); 130 break; 131 case kTfLiteInt16: 132 TF_LITE_ENSURE_STATUS(EvalImpl<int16_t>(context, input, node)); 133 break; 134 case kTfLiteInt32: 135 TF_LITE_ENSURE_STATUS(EvalImpl<int32_t>(context, input, node)); 136 break; 137 case kTfLiteInt64: 138 TF_LITE_ENSURE_STATUS(EvalImpl<int64_t>(context, input, node)); 139 break; 140 case kTfLiteFloat32: 141 TF_LITE_ENSURE_STATUS(EvalImpl<float>(context, input, node)); 142 break; 143 case kTfLiteUInt8: 144 TF_LITE_ENSURE_STATUS(EvalImpl<uint8_t>(context, input, node)); 145 break; 146 default: 147 context->ReportError(context, "Currently Unique doesn't support type: %s", 148 TfLiteTypeGetName(input->type)); 149 return kTfLiteError; 150 } 151 return kTfLiteOk; 152 } 153 154 } // namespace unique 155 156 TfLiteRegistration* Register_UNIQUE() { 157 static TfLiteRegistration r = {unique::Init, unique::Free, unique::Prepare, 158 unique::Eval}; 159 return &r; 160 } 161 162 } // namespace builtin 163 } // namespace ops 164 } // namespace tflite 165