Thrust, CUDA Runtime 13.0 throws CUDACOMPILE : nvcc error : 'cudafe++' died with status 0xC0000409 in Visual Studio 2022 Version 17.14.21

[venkatkasirajan]

Inclusion of thrust headers
#include <thrust/host_vector.h>
#include <thrust/device_vector.h>
Causes Visual Studio 2022 Version 17.14.21 to throw an exception: CUDACOMPILE : nvcc error : 'cudafe++' died with status 0xC0000409

The same code works well with CUDA Runtime 12.9 on the same compiler.

Build dependencies are set well.

What could be wrong?

[gevtushenko]

Hello @venkatkasirajan!

I'm not able to reproduce this with Visual Studio 17.14 (latest patch version) and CTK 13.0:

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.0\bin\nvcc.exe" -gencode=arch=compute_75,code=\"sm_75,compute_75\" --use-local-env -ccbin "C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.44.35207\bin\HostX64\x64" -x cu   -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.0\include" -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.0\include"     --keep-dir x64\Release  -maxrregcount=0    --machine 64 --compile -cudart static -std=c++17 -Xcompiler "/std:c++17"   -DWIN32 -DWIN64 -DNDEBUG -D_CONSOLE -D_MBCS -Xcompiler "/EHsc /W3 /nologo /O2 /FS   /MD " -Xcompiler "/Fdx64\Release\vc143.pdb" -o C:\Users\gevtushenko\source\repos\CudaRuntime130\CudaRuntime130\x64\Release\kernel.cu.obj "C:\Users\gevtushenko\source\repos\CudaRuntime130\CudaRuntime130\kernel.cu"
========== Build: 1 succeeded, 0 failed, 0 up-to-date, 0 skipped ==========
========== Build completed at 10:32 AM and took 09.151 seconds ========== 

The program I've built was simply:

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

int main() {}

Could you please provide a standalone reproducer (command line options, source code etc.)? Also, do you see this error without thrust includes?

[venkatkasirajan]

You can close the ticket. I couldnt find time to create an equivalent code that I can share with you due to varying priorities here. I rather replaced thrust with my own version now and I am good. However, I'm of the opinion that the problem is still there & if I am able to spend some time on this, I will try to open this ticket again. tan(q) ⋅ lim Regards(r) r → ∞ Venkat Kasirajan *Author: Fundamentals of Quantum Computing: Theory and Practice <https://www.springer.com/us/book/9783030636883>* “From the ashes, a fire shall be woken, A light from the shadows shall spring. Renewed shall be the blade that was broken; the crownless again shall be king.” ― J.R.R. Tolkien, The Fellowship of the Ring

…

On Wed, Dec 31, 2025 at 11:40 AM Georgii Evtushenko < ***@***.***> wrote: Hello @venkatkasirajan <https://github.com/venkatkasirajan>! I'm not able to reproduce this with Visual Studio 17.14 (latest patch version) and CTK 13.0: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.0\bin\nvcc.exe" -gencode=arch=compute_75,code=\"sm_75,compute_75\" --use-local-env -ccbin "C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.44.35207\bin\HostX64\x64" -x cu -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.0\include" -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v13.0\include" --keep-dir x64\Release -maxrregcount=0 --machine 64 --compile -cudart static -std=c++17 -Xcompiler "/std:c++17" -DWIN32 -DWIN64 -DNDEBUG -D_CONSOLE -D_MBCS -Xcompiler "/EHsc /W3 /nologo /O2 /FS /MD " -Xcompiler "/Fdx64\Release\vc143.pdb" -o C:\Users\gevtushenko\source\repos\CudaRuntime130\CudaRuntime130\x64\Release\kernel.cu.obj "C:\Users\gevtushenko\source\repos\CudaRuntime130\CudaRuntime130\kernel.cu" ========== Build: 1 succeeded, 0 failed, 0 up-to-date, 0 skipped ========== ========== Build completed at 10:32 AM and took 09.151 seconds ========== The program I've built was simply: #include <thrust/host_vector.h> #include <thrust/device_vector.h> int main() {} Could you please provide a standalone reproducer (command line options, source code etc.)? — Reply to this email directly, view it on GitHub <#6724 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BI3Q4LHOZNPPMAIBDIRQGVL4EQKCHAVCNFSM6AAAAACMY6XT4WVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTKMZYGI4TQNY> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[nnnunnn]

I get the same problem using CUDA 13.3 and Visual Studio 2026 (18.9.0) on Windows 11.
Two issues appear when including a cccl header, e.g. <thrust/device_vector.h>

(1) a problem with the definition of _CCCL_RESTRICT
(2) '__asm' blocks are not supported by the MSVC host compiler

nvcc details:

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2026 NVIDIA Corporation
Built on Tue_Jun__9_14:30:19_Pacific_Daylight_Time_2026
Cuda compilation tools, release 13.3, V13.3.73
Build cuda_13.3.r13.3/compiler.38244171_0

Compiling the following file fails, either inside Visual Studio 2026 or from the command line,

// file: basic_vector.cu
#include <thrust/device_vector.h>
#include <thrust/host_vector.h>
int main() {
  thrust::host_vector<int> H{1, 2, 3, 4};
  thrust::device_vector<int> D = H;
  return 0;
}

Output from the command line:

D:\NN\thrust_example>nvcc basic_vector.cu -Xcompiler "/std:c++17 /Zc:preprocessor" -o basic_vector.exe
test_vector.cu
Compilation aborted.
nvcc error   : 'cudafe++' died with status 0xC0000409

Some output from Visual Studio:

Rebuild started at 5:19 AM...
1>------ Rebuild All started: Project: basic_vector, Configuration: Debug x64 ------
1>  Compiling CUDA source file basic_vector.cu...
1>
1>  D:\NN\thrust_example>"C:\VS\CUDA\bin\nvcc.exe" -gencode=arch=compute_86,code=\"sm_86,compute_86\" --use-local-env -ccbin "C:\VS\VC\Tools\MSVC\14.51.36231\bin\HostX64\x64" -x cu   -IC:\VS\CUDA\include\cccl -IC:\VS\CUDA\include     --keep-dir basic_vector\x64\Debug  -maxrregcount=0     --machine 64 --compile -cudart static -std=c++17 -Xcompiler="-Zc:preprocessor" -g  -D_WINDOWS -D_DEBUG -D_MBCS -D"CMAKE_INTDIR=\"Debug\"" -Xcompiler "/EHsc /W4 /nologo /Od /FS /Zi /RTC1 /MDd /GR" -Xcompiler "/Fdbasic_vector.dir\Debug\vc145.pdb" -o basic_vector.dir\Debug\basic_vector.obj "D:\NN\thrust_example\basic_vector.cu"
1>  basic_vector.cu
1>  tmpxft_00006150_00000000-7_basic_vector.cudafe1.cpp
1>C:\VS\CUDA\include\cccl\cuda/std/__string/constexpr_c_functions.h(53): error C2146: syntax error: missing ')' before identifier '__dst'
1>C:\VS\CUDA\include\cccl\cuda/std/__string/constexpr_c_functions.h(53): error C3646: '__dst': unknown override specifier
1>C:\VS\CUDA\include\cccl\cuda/std/__string/constexpr_c_functions.h(53): error C2988: unrecognizable template declaration/definition
...
...

The above errors refer to line 53 in file constexpr_c_functions.h, and appear to be caused by a wrong definition of "_CCCL_RESTRICT",

#if !_CCCL_COMPILER(NVRTC)
template <class _CharT>
_CCCL_HIDE_FROM_ABI _CCCL_HOST _CharT*
__cccl_strcpy_impl_host(_CharT* _CCCL_RESTRICT __dst, const _CharT* _CCCL_RESTRICT __src) noexcept
{
  if constexpr (sizeof(_CharT) == 1)
  {
    return reinterpret_cast<_CharT*>(::strcpy(reinterpret_cast<char*>(__dst), reinterpret_cast<const char*>(__src)));
  }
  else
  {
    return ::cuda::std::__cccl_strcpy_impl_constexpr(__dst, __src);
  }
}
#endif // !_CCCL_COMPILER(NVRTC)

Here are examples of the 2nd issue (unsupported '__asm' blocks),

1>C:\VS\CUDA\include\cccl\cub/util_ptx.cuh(431): error C4235: nonstandard extension used: '__asm' keyword not supported on this architecture
1>C:\VS\CUDA\include\cccl\cub/warp/specializations/warp_reduce_shfl.cuh(170): error C4235: nonstandard extension used: '__asm' keyword not supported on this architecture
1>C:\VS\CUDA\include\cccl\cub/warp/specializations/warp_reduce_shfl.cuh(206): error C4235: nonstandard extension used: '__asm' keyword not supported on this architecture
...
...

Is anyone able to use the current thrust library using Visual Studio 2026?
thanks for any pointers,
Nigel