可能由于错误的实现而导致带有推力::: complex的推力:: device_vector的错误

Question

I'm actually learning CUDA and thrust , and I'm trying to make a project with .cpp , .hpp files and .cu , .cuh files. Therefore, I have done a first little implementation (see code below), however I have a compile error. Here is the output of my compile error.

Honestly, I don't know exactly what this kind of error means, but I found out that this came from this line:

thrust::device_vector<thrust::complex<T>> deviceVec_;

because when I comment out this line there is no compile error anymore. Therefore, I assume that this is due to the content of the implementation of thrust::device_vector and the fact that I included the .cuh file in the .hpp file and since my main.cpp is handled by g++ the preprocessing will be performed by g++ instead of nvcc .

Content of my main.cpp file:

#include "QGPU.hpp"
int main()
{
     QGPU::GPU<double>  gpu;
     return (0);
}

Content of my .hpp file:

#pragma once

# include "QCUDA.cuh"

namespace QGPU {

 template<typename T>
 class GPU {
 private:
  QCUDA::CUDAGPU<T> cgpu_;
 public:
  GPU();
  virtual ~GPU();

};

template<typename T>
GPU<T>::GPU()
{};

template<typename T>
GPU<T>::~GPU()
{};

};

Content of my .cuh file:

#pragma once

# include <thrust/host_vector.h>
# include <thrust/device_vector.h>
# include <thrust/complex.h>

namespace QCUDA {

 template<typename T>
 class CUDAGPU {
 private:
  thrust::host_vector<thrust::complex<T>> hostVec_;
  thrust::device_vector<thrust::complex<T>> deviceVec_;
 public:
  CUDAGPU();
  virtual ~CUDAGPU();
 };

 template<typename T>
 CUDAGPU<T>::CUDAGPU()
 {};

 template<typename T>
 CUDAGPU<T>::~CUDAGPU()
 {};

};

Thus, my question is:

Is there a way to solve this compile error, and therefore, maintaining this implementation ?

Or I must change my vision on how I should implement a project where .cpp , .hpp files and .cu , .cuh files are mixed together ?

If I must change my implementation, there is a possibility to get a similar example of what would be a good implementation by following my attention ?

NB: I'm actually working with a GTX 1060 with the cuda version:

$nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2018 NVIDIA Corporation
Built on Wed_Apr_11_23:16:29_CDT_2018
Cuda compilation tools, release 9.2, V9.2.88

Answer 1

You need to rename main.cpp to main.cu to make this work. Otherwise you are importing CUDA code into a plain .cpp file, and the host C++ will fail to compile the code.