neural_renderer是一个非常好的三维结构渲染器，[github链接](https://github.com/daniilidis-group/neural_renderer)。 我down下来使用以下命令安装时报了一个错误。 ```python python setup.py install ``` ```shell D:/Anaconda/envs/python37/lib/site-packages/torch/include\ATen/Context.h(119): warning: field of class type without a DLL interface used in a class with a DLL interface D:/Anaconda/envs/python37/lib/site-packages/torch/include\ATen/Context.h(120): warning: field of class type without a DLL interface used in a class with a DLL interface D:/Anaconda/envs/python37/lib/site-packages/torch/include\ATen/TensorGeometry.h(56): warning: field of class type without a DLL interface used in a class with a DLL interface D:/Anaconda/envs/python37/lib/site-packages/torch/include\ATen/TensorGeometry.h(57): warning: field of class type without a DLL interface used in a class with a DLL interface neural_renderer/cuda/rasterize_cuda_kernel.cu(9): error: function "atomicAdd(double *, double)" has already been defined 1 error detected in the compilation of "C:/Users/luxuff/AppData/Local/Temp/tmpxft_000007ec_00000000-8_rasterize_cuda_kernel.cpp4.ii". error: command 'C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v10.0\\bin\\nvcc.exe' failed with exit code 2 ``` 问题的关键其实就是上述描述的倒数第四行，"atomicAdd(double *, double)" has already been defined。 打开该文件查看，似乎没什么问题： ```c++ #include <iostream> #include <ATen/ATen.h> #include <cuda.h> #include <cuda_runtime.h> // for the older gpus atomicAdd with double arguments does not exist #if __CUDA_ARCH__ < 600 and defined(__CUDA_ARCH__) static __inline__ __device__ double atomicAdd(double* address, double val) { unsigned long long int* address_as_ull = (unsigned long long int*)address; unsigned long long int old = *address_as_ull, assumed; do { assumed = old; old = atomicCAS(address_as_ull, assumed, __double_as_longlong(val + __longlong_as_double(assumed))); // Note: uses integer comparison to avoid hang in case of NaN (since NaN != NaN) } while (assumed != old); } while (assumed != old); return __longlong_as_double(old); } #endif ``` 后来在CSDN [这篇文章](https://blog.csdn.net/AUTO1993/article/details/78753576)中找到了解决方案： ```c++ #if !defined(__CUDA_ARCH__) || __CUDA_ARCH__ >= 600 #else static __inline__ __device__ double atomicAdd(double *address, double val) { unsigned long long int* address_as_ull = (unsigned long long int*)address; unsigned long long int old = *address_as_ull, assumed; if (val==0.0) return __longlong_as_double(old); do { assumed = old; old = atomicCAS(address_as_ull, assumed, __double_as_longlong(val +__longlong_as_double(assumed))); } while (assumed != old); return __longlong_as_double(old); } #endif ``` 解决措施是：将原来代码的宏定义由 ```c++ #if __CUDA_ARCH__ < 600 and defined(__CUDA_ARCH__) //代码内容 #endif ``` 改为 ```c++ #if !defined(__CUDA_ARCH__) || __CUDA_ARCH__ >= 600 #else //代码内容 #endif ``` 改完之后确实能正常安装neural_renderer了，但我懵逼了，这俩结构翻译过来不是一个意思吗？？？ 我坚信我应该遗漏了什么关键要素，最后试出了另外两种方式。 我将原宏定义改成了如下内容 ```c++ #if defined(__CUDA_ARCH__) and __CUDA_ARCH__ < 600 //代码内容 #endif 或者 #if __CUDA_ARCH__ < 600 && defined(__CUDA_ARCH__) //代码内容 #endif ``` 结果都可以正常执行安装了，原因未知，待后期查证。