(资料图)

//查询设备数目intdeviceCount;cudaGetDeviceCount(&deviceCount);//遍历设备编号信息intdevice;for(device=0;device

// 设置当前设备为GPU 0cudaSetDevice(0);// 初始化TensorRTthis->runtime=createInferRuntime(gLogger);assert(this->runtime!=nullptr);this->engine=runtime->deserializeCudaEngine(trtModelStream,size);assert(this->engine!=nullptr);this->context=engine->createExecutionContext();assert(this->context!=nullptr);delete[]trtModelStream;// do more thing here// insert query input and output layers information//创建GPU显存输入/输出缓冲区std::cout<<"input/outpu:"<getNbBindings()<input_h*this->input_w*3*sizeof(float));cudaMalloc(&buffers[2],this->output_h*this->output_w*sizeof(float));cudaMalloc(&buffers[1],32*25600*sizeof(float));//创建临时缓存输出prob.resize(output_h*output_w);mprob.resize(32*25600);//创建cuda流cudaStreamCreate(&stream);

//初始化时间标记cudaEvent_tstart,stop;cudaEventCreate(&start);cudaEventCreate(&stop);cudaEventRecord(start,0);//查询设备数目intdeviceCount;cudaGetDeviceCount(&deviceCount);//遍历设备编号信息intdevice;for(device=0;device

参考资料：

扫码查看OpenCV+OpenVIO+Pytorch系统化学习路线图