wehub-resource-sync
80 lines
2.7 KiB
C++
80 lines
2.7 KiB
C++
//
|
|
// ReductionTflite.cpp
|
|
// MNNConverter
|
|
//
|
|
// Created by MNN on 2019/12/09.
|
|
// Copyright © 2018, Alibaba Group Holding Limited
|
|
//
|
|
|
|
#include <stdio.h>
|
|
#include "liteOpConverter.hpp"
|
|
|
|
DECLARE_OP_COVERTER(ReductionTflite);
|
|
MNN::OpType ReductionTflite::opType(int quantizedModel) {
|
|
return MNN::OpType_Reduction;
|
|
}
|
|
MNN::OpParameter ReductionTflite::type(int quantizedModel) {
|
|
return MNN::OpParameter_ReductionParam;
|
|
}
|
|
|
|
void ReductionTflite::run(MNN::OpT* dstOp, const std::unique_ptr<tflite::OperatorT>& tfliteOp,
|
|
const std::vector<std::unique_ptr<tflite::TensorT>>& tfliteTensors,
|
|
const std::vector<std::unique_ptr<tflite::BufferT>>& tfliteModelBuffer,
|
|
const std::vector<std::unique_ptr<tflite::OperatorCodeT>>& tfliteOpSet, int quantizedModel){
|
|
auto param = new MNN::ReductionParamT;
|
|
auto opt = tfliteOp->builtin_options.AsReducerOptions();
|
|
param->keepDims=opt->keep_dims;
|
|
#ifdef TF_CONVERT_ORIGIN
|
|
const int input1Idx = tfliteOp->inputs[1];
|
|
const auto& input1Tensor = tfliteTensors[input1Idx];
|
|
if(input1Tensor.is_variable == false){
|
|
auto buffer1Idx=input1Tensor.buffer;
|
|
auto buffer1=tfliteModelBuffer[buffer1Idx];
|
|
auto shape=input1Tensor.shape;
|
|
param->dim.resize(shape.size());
|
|
for(decltype(shape.size()) x=0;x<shape.size();x++){
|
|
param->dim[x]=shape[x];
|
|
}
|
|
}
|
|
#endif
|
|
switch(liteOpConverter::getOpCode(tfliteOpSet[tfliteOp->opcode_index].get())){
|
|
case tflite::BuiltinOperator_SUM:{
|
|
param->operation=MNN::ReductionType_SUM;
|
|
break;
|
|
}
|
|
case tflite::BuiltinOperator_REDUCE_MAX:{
|
|
param->operation=MNN::ReductionType_MAXIMUM;
|
|
break;
|
|
}
|
|
case tflite::BuiltinOperator_REDUCE_MIN:{
|
|
param->operation=MNN::ReductionType_MINIMUM;
|
|
break;
|
|
}
|
|
case tflite::BuiltinOperator_REDUCE_ANY:{
|
|
param->operation=MNN::ReductionType_ANY;
|
|
break;
|
|
}
|
|
case tflite::BuiltinOperator_REDUCE_PROD:{
|
|
param->operation=MNN::ReductionType_PROD;
|
|
break;
|
|
}
|
|
case tflite::BuiltinOperator_MEAN:{
|
|
param->operation=MNN::ReductionType_MEAN;
|
|
break;
|
|
}
|
|
default:{
|
|
LOG(ERROR) << "MNN Converter Not "
|
|
"Supported!!! Reduction Op: "
|
|
<< tfliteOpSet[tfliteOp->opcode_index]->custom_code;
|
|
}
|
|
}
|
|
dstOp->main.value = param;
|
|
}
|
|
using namespace tflite;
|
|
REGISTER_CONVERTER(ReductionTflite,BuiltinOperator_SUM);
|
|
REGISTER_CONVERTER(ReductionTflite,BuiltinOperator_REDUCE_MAX);
|
|
REGISTER_CONVERTER(ReductionTflite,BuiltinOperator_REDUCE_MIN);
|
|
REGISTER_CONVERTER(ReductionTflite,BuiltinOperator_REDUCE_ANY);
|
|
REGISTER_CONVERTER(ReductionTflite,BuiltinOperator_REDUCE_PROD);
|
|
REGISTER_CONVERTER(ReductionTflite,BuiltinOperator_MEAN);
|