使用设备链接和强制同步PDB写入时的Cuda错误

Question

我正在尝试通过使用__device__ __host__函数限定符来创建在cuda和cuda上均可使用的代码。为此，我必须使用设备链接选项-dc。

执行此操作时，出现错误：fatal error C1041: cannot open program database ... ; if multiple CL.EXE write to the same .PDB file, please use /FS。

因此，我尝试在nvcc选项中添加/FS。但是当我这样做时，我得到了错误：nvcc fatal : A single input file is required for a non-link phase when an outputfile is specified

例如，这是我要编译的Vector3类的头文件

#include <cuda.h>
#include <cuda_runtime.h>
#include <device_launch_parameters.h>

class Vector3
{
public:
    __device__ __host__ Vector3();
    __device__ __host__ Vector3(double const& x, double const& y, double const& z);

    __device__ __host__ double get_x() const;
    __device__ __host__ double get_y() const;
    __device__ __host__ double get_z() const;

    __device__ __host__ double get_magnitude() const;

    __device__ __host__ void set(double const& x, double const& y, double const& z);

    __device__ __host__ void set_x(double const& x);
    __device__ __host__ void set_y(double const& y);
    __device__ __host__ void set_z(double const& z);

    __device__ __host__ Vector3 get_normalized() const;

    __device__ __host__ void rotate_x(double const& angle);
    __device__ __host__ void rotate_y(double const& angle);
    __device__ __host__ void rotate_z(double const& angle);

    __device__ __host__ Vector3 operator+(Vector3 const& v) const;
    __device__ __host__ Vector3 operator*(double const& v) const;

    __device__ __host__ void operator+=(Vector3 const& v);
    __device__ __host__ void operator*=(double const& v);

    __device__ __host__ double dot(Vector3 const& v) const;

protected:
    double x_, y_, z_;
};