cant install neural_renderer

[BussanQ]

error info: D:/dev/app/Anaconda3/envs/pytorch/lib/site-packages/torch/include\c10/util/flat_hash_map.h(1490): error C3203: 'templated_iterator': unspecialized class template can't be used as a template argument for template parameter '_Ty1', expected a real type error: command 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\bin\nvcc.exe' failed with exit status 2

[piaozhx]

Maybe there are some different steps between windows and linux when installing neural_renderer, you can try it on linux environment (recommend ubuntu16.04 )

[GoooIce]

I installed successfully with win10, only modified rasterize_cuda_kernel.cu, but the error is not the same, I don't understand but hope to help.

rasterize_cuda_kernel.cu

---

// 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

#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

[BussanQ]

@GoooIce the same error after modified , but thank u .

[AndroXD]

@BussanQ

error: command 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\bin\nvcc.exe' failed with exit status 2 v10.1

Try with CUDA 10.0 , I successfully installed it on Windows 10 with that version, to my knowledge Pytorch isn't supposed to work with CUDA 10.1

[us]

@BussanQ

error: command 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1\bin\nvcc.exe' failed with exit status 2 v10.1

Try with CUDA 10.0 , I successfully installed it on Windows 10 with that version, to my knowledge Pytorch isn't supposed to work with CUDA 10.1

I have CUDA 10.0 but I got same error on Windows10.

[us]

I install the Visual Studio 2017 then I fixed the code as @GoooIce suggested. Then problem is fixing.

[Aspen233]

I tried it on two separate computers with, respectively, cuda 10.0 and cuda 9.2. However different errors occurred but neither of them was like the ones mentioned above.

D:\written_programs\python\machine_learning\Exercise\venv\lib\site-packages\torch\utils\cpp_extension.py:184: UserWarning: Error checking compiler version for cl: [WinError 2] 系统找不到指定的文件。 warnings.warn('Error checking compiler version for {}: {}'.format(compiler, error)) C:\Apps\Microsoft\VisualStudio\2017\Community\VC\Tools\MSVC\14.16.27023\bin\HostX86\x64\cl.exe /c /nologo /Ox /W3 /GL /DNDEBUG /MD -ID:\written_programs\python\machine_learning\Exercise\venv\lib\site-packages\torch\lib\include -ID:\written_programs\python\machine_learning\Exercise\venv\lib\site-packages\torch\lib\include\torch\csrc\api\include -ID:\written_programs\python\machine_learning\Exercise\venv\lib\site-packages\torch\lib\include\TH -ID:\written_programs\python\machine_learning\Exercise\venv\lib\site-packages\torch\lib\include\THC "-IC:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\include" -ID:\written_programs\python\machine_learning\Exercise\venv\include -IC:\Apps\Microsoft\VisualStudio\Shared\Anaconda3_64\include -IC:\Apps\Microsoft\VisualStudio\Shared\Anaconda3_64\include -IC:\Apps\Microsoft\VisualStudio\2017\Community\VC\Tools\MSVC\14.16.27023\ATLMFC\include -IC:\Apps\Microsoft\VisualStudio\2017\Community\VC\Tools\MSVC\14.16.27023\include /EHsc /Tpneural_renderer/cuda/load_textures_cuda.cpp /Fobuild\temp.win-amd64-3.6\Release\neural_renderer/cuda/load_textures_cuda.obj -DTORCH_API_INCLUDE_EXTENSION_H -DTORCH_EXTENSION_NAME=load_textures -D_GLIBCXX_USE_CXX11_ABI=0 load_textures_cuda.cpp C:\Apps\Microsoft\VisualStudio\2017\Community\VC\Tools\MSVC\14.16.27023\include\cstddef(7): fatal error C1083: 无法打开包括文件: “stddef.h”: No such file or directory error: command 'C:\Apps\Microsoft\VisualStudio\2017\Community\VC\Tools\MSVC\14.16.27023\bin\HostX86\x64\cl.exe' failed with exit status 2 This one comes from the one with cuda 10.0

[tab-1]

Where is the file rasterize_cuda_kernel.cu? Why didn't I find it?@GoooIce

[taotaoyuhust]

i also got some error while installing :

Python/Python36/lib/site-packages/torch/include\c10/util/flat_hash_map.h(1379): error C3203: “templated_iterator”: 未专用化的 类 模板 不能用 作 模板 变量，该变量属于 模板 参数“_Ty1”，应为 real 类型

C:/Users/Administrator/AppData/Local/Programs/Python/Python36/lib/site-packages/torch/include\c10/util/flat_hash_map.h(1437): note: 参见对正在编译的类 模板 实例化“ska::flat_hash_map<K,V,H,E,A>”的引用 C:/Users/Administrator/AppData/Local/Programs/Python/Python36/lib/site-packages/torch/include\c10/util/flat_hash_map.h(1383): error C3203: “templated_iterator”: 未专用化的 类 模板 不能用 作 模板 变量，该变量属于 模板 参数“_Ty1”，应为 real 类型 C:/Users/Administrator/AppData/Local/Programs/Python/Python36/lib/site-packages/torch/include\c10/util/flat_hash_map.h(1391): error C3203: “templated_iterator”: 未专用化的 类 模板 不能用 作 模板 变量，该变量属于 模板 参数“_Ty1”，应为 real 类型 C:/Users/Administrator/AppData/Local/Programs/Python/Python36/lib/site-packages/torch/include\c10/util/flat_hash_map.h(1473): error C3203: “templated_iterator”: 未专用化的 类 模板 不能用 作 模板 变量，该变量属于 模板 参数“_Ty1”，应为 real 类型 C:/Users/Administrator/AppData/Local/Programs/Python/Python36/lib/site-packages/torch/include\c10/util/flat_hash_map.h(1510): note: 参见对正在编译的类 模板 实例化“ska::flat_hash_set<T,H,E,A>”的引用 C:/Users/Administrator/AppData/Local/Programs/Python/Python36/lib/site-packages/torch/include\c10/util/flat_hash_map.h(1478): error C3203: “templated_iterator”: 未专用化的 类 模板 不能用 作 模板 变量，该变量属于 模板 参数“_Ty1”，应为 real 类型 C:/Users/Administrator/AppData/Local/Programs/Python/Python36/lib/site-packages/torch/include\c10/util/flat_hash_map.h(1482): error C3203: “templated_iterator”: 未专用化的 类 模板 不能用 作 模板 变量，该变量属于 模板 参数“_Ty1”，应为 real 类型 C:/Users/Administrator/AppData/Local/Programs/Python/Python36/lib/site-packages/torch/include\c10/util/flat_hash_map.h(1486): error C3203: “templated_iterator”: 未专用化的 类 模板 不能用 作 模板 变量，该变量属于 模板 参数“_Ty1”，应为 real 类型 C:/Users/Administrator/AppData/Local/Programs/Python/Python36/lib/site-packages/torch/include\c10/util/flat_hash_map.h(1490): error C3203: “templated_iterator”: 未专用化的 类 模板 不能用 作 模板 变量，该变量属于 模板 参数“_Ty1”，应为 real 类型 error: command 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.2\bin\nvcc.exe' failed with exit status 2

windows10 ,cuda9.2 compiling this file nvcc.exe -c neural_renderer/cuda/load_textures_cuda_kernel.cu not the file "rasterize_cuda_kernel.cu" mentiond above

[taotaoyuhust]

solved this error and finished the installing using the method mentioned in https://github.com/pytorch/pytorch/issues/19156

warp the error lines using #ifndef CUDACC and #endif

win10, cuda9.2, torch1.2.0

[Leo9113]

solved this error and finished the installing using the method mentioned in pytorch/pytorch#19156

warp the error lines using #ifndef CUDACC and #endif

win10, cuda9.2, torch1.2.0

请问具体怎么修改

[zhangdanfeng888]

solved this error and finished the installing using the method mentioned in pytorch/pytorch#19156

warp the error lines using #ifndef CUDACC and #endif

win10, cuda9.2, torch1.2.0

请问可以分享一下具体的修改方法吗？

[shahbuland]

I installed successfully with win10, only modified rasterize_cuda_kernel.cu, but the error is not the same, I don't understand but hope to help.

rasterize_cuda_kernel.cu

// 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

#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

This fixed for me as well on Win10, cuda10.0, torch 1.4.0

[iterhating]

I installed successfully with win10

Any hints? I have gotten a number of errors sorted but am stuck on: ModuleNotFoundError: No module named 'neural_renderer.cuda'

Though as far as I can tell the folders are set up fine.

Sorry to necro, just frustrated with this repo

[shahbuland]

I installed successfully with win10

Any hints? I have gotten a number of errors sorted but am stuck on: ModuleNotFoundError: No module named 'neural_renderer.cuda'

Though as far as I can tell the folders are set up fine.

Sorry to necro, just frustrated with this repo

Module not found errors can have a ton of reasons. This is kind of a shot in the dark, but I had the same type of error in a completely unrelated situation: try adding an empty __init__.py file in the same folder as neural_renderer.cuda

Otherwise, version mismatches could be an issue. Some questions would be: what's your python version, your CUDA version, your GPU, are you running with -m or not?

[iterhating]

Thanks for following up, sorry for the delay in response. I tried for several hours after this and rebuilt my environment and worked out a number of errors but always more persist. I'm giving up. I already have IPERCore working, which is apparently a successor project from the same crew, but the demos of this looked like it actually generated better results. Possibly it's just cherry-picking of examples. Anyway, it generates no results for me, so I'll try to hone IPERCore.

Anyway, thanks for being wiling to help!