mirror of https://github.com/alibaba/MNN.git
56 lines
1.8 KiB
C++
56 lines
1.8 KiB
C++
|
//
|
||
|
|
// TRTCommonExecution.cpp
|
||
|
|
// MNN
|
||
|
|
//
|
||
|
|
// Created by MNN on 2019/02/28.
|
||
|
|
// Copyright © 2018, Alibaba Group Holding Limited
|
||
|
|
//
|
||
|
|
|
||
|
|
#include "TRTCommonExecution.hpp"
|
||
|
|
namespace MNN {
|
||
|
|
|
||
|
|
TRTCommonExecution::TRTCommonExecution(Backend *backend, const Op *op) : Execution(backend) {
|
||
|
|
mTrtBackend = (TRTBackend *)backend;
|
||
|
|
mOp = op;
|
||
|
|
}
|
||
|
|
|
||
|
|
ErrorCode TRTCommonExecution::onResize(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) {
|
||
|
|
mInputs = inputs;
|
||
|
|
mOutputs = outputs;
|
||
|
|
// MNN_PRINT("layer info: Type:%s name:%s \n", EnumNameOpType(mOp->type()), mOp->name()->c_str());
|
||
|
|
// MNN_PRINT("layer info: Type:%s \n", EnumNameOpType(mOp->type()));
|
||
|
|
std::vector<ITensor *> nvTensors(inputs.size());
|
||
|
|
for (int i = 0; i < inputs.size(); ++i) {
|
||
|
|
nvTensors[i] = mTrtBackend->getTensorOps(inputs[i]);
|
||
|
|
}
|
||
|
|
// printf("input : \n");
|
||
|
|
// auto dims = nvTensors[0]->getDimensions();
|
||
|
|
// for(int i = 0; i < dims.nbDims; i++){
|
||
|
|
// printf("%d ", dims.d[i]);
|
||
|
|
// }
|
||
|
|
// printf("\n");
|
||
|
|
// for(int i = 0; i < dims.nbDims; i++){
|
||
|
|
// printf("%d ", inputs[0]->shape()[i]);
|
||
|
|
// }
|
||
|
|
// printf("\n");
|
||
|
|
auto outputsTRT = this->onEncode(nvTensors);
|
||
|
|
// printf("output : \n");
|
||
|
|
// auto out_dims = outputsTRT[0]->getDimensions();
|
||
|
|
// for(int i = 0; i < out_dims.nbDims; i++){
|
||
|
|
// printf("%d ", out_dims.d[i]);
|
||
|
|
// }
|
||
|
|
// printf("\n");
|
||
|
|
// for(int i = 0; i < out_dims.nbDims; i++){
|
||
|
|
// printf("%d ", outputs[0]->shape()[i]);
|
||
|
|
// }
|
||
|
|
// printf("\n");
|
||
|
|
mTrtBackend->setTensorOps(outputs, std::move(outputsTRT));
|
||
|
|
return NO_ERROR;
|
||
|
|
}
|
||
|
|
|
||
|
|
ErrorCode TRTCommonExecution::onExecute(const std::vector<Tensor *> &inputs, const std::vector<Tensor *> &outputs) {
|
||
|
|
return NO_ERROR;
|
||
|
|
}
|
||
|
|
|
||
|
|
}; // namespace MNN
|