win11编译paddle2.4，CUDA缺少device.h

[hibean]

编译环境：
win11
cuda12.1
vs2019
python3.10.11

编译报错：

[684/3243] Building CUDA object paddle\phi\kernels\CMakeFiles\phi_gpu.dir\gpu\graph_send_recv_kernel.cu.obj
FAILED: paddle/phi/kernels/CMakeFiles/phi_gpu.dir/gpu/graph_send_recv_kernel.cu.obj
C:\PROGRA~1\NVIDIA~2\CUDA\v12.1\bin\nvcc.exe -forward-unknown-to-host-compiler -DCUDA_TOOLKIT_ROOT_DIR="\"C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.1\"" -DCUDA_VERSION_MAJOR=\"12\" -DCUDA_VERSION_MINOR=\"1\" -DCUDNN_MAJOR_VERSION=\"8\" -DEIGEN_STRONG_INLINE=inline -DEIGEN_USE_GPU -DGOOGLE_GLOG_DLL_DECL="" -DLAPACK_FOUND -DNOMINMAX -DPADDLE_DISABLE_PROFILER -DPADDLE_DLL_EXPORT -DPADDLE_DLL_INFERENCE -DPADDLE_ON_INFERENCE -DPADDLE_VERSION=0.0.0 -DPADDLE_VERSION_INTEGER=0 -DPADDLE_WITH_AVX -DPADDLE_WITH_CRYPTO -DPADDLE_WITH_CUDA -DPADDLE_WITH_MKLDNN -DPADDLE_WITH_MKLML -DPADDLE_WITH_POCKETFFT -DPADDLE_WITH_SSE3 -DTRT_PLUGIN_FP16_AVALIABLE -DUTF8PROC_STATIC -DWIN32_LEAN_AND_MEAN -D_XKEYCHECK_H -ID:\pythonWorkspace\Paddle\build -ID:\pythonWorkspace\Paddle\paddle\fluid\framework\io -ID:\pythonWorkspace\Paddle\build\third_party\install\zlib\include -ID:\pythonWorkspace\Paddle\build\third_party\install -ID:\pythonWorkspace\Paddle\build\third_party\install\gflags\include -ID:\pythonWorkspace\Paddle\build\third_party\install\glog\include -ID:\pythonWorkspace\Paddle\build\third_party\eigen3\src\extern_eigen3 -ID:\pythonWorkspace\Paddle\build\third_party\threadpool\src\extern_threadpool -ID:\pythonWorkspace\Paddle\build\third_party\dlpack\src\extern_dlpack\include -ID:\pythonWorkspace\Paddle\build\third_party\install\xxhash\include -ID:\pythonWorkspace\Paddle\build\third_party\install\warpctc\include -ID:\pythonWorkspace\Paddle\build\third_party\install\utf8proc\include -ID:\pythonWorkspace\Paddle\build\third_party\install\mklml\include -ID:\pythonWorkspace\Paddle\build\third_party\install\mkldnn\include -ID:\pythonWorkspace\Paddle\build\third_party\install\protobuf\include -I"C:\Program Files\Python310\include" -I"C:\Program Files\Python310\Lib\site-packages\numpy\core\include" -ID:\pythonWorkspace\Paddle\build\third_party\pybind\src\extern_pybind\include -ID:\pythonWorkspace\Paddle\build\third_party\install\cryptopp\include -ID:\pythonWorkspace\Paddle\build\third_party\pocketfft\src -ID:\pythonWorkspace\Paddle\build\third_party\dirent\src\extern_dirent\include -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1\include" -ID:\pythonWorkspace\Paddle -ID:\pythonWorkspace\Paddle\build\..\paddle\fluid\framework\io -D_WINDOWS -Xcompiler=" /W0  /GR /EHsc"  -gencode arch=compute_75,code=sm_75 -w --expt-relaxed-constexpr --expt-extended-lambda -Xcompiler "/wd4244 /wd4267 /wd4819 " -Xcompiler /bigobj  -Xcompiler="/arch:AVX" -Xcompiler="-MT -O2 -Ob2" -DNDEBUG -std=c++14 --generate-code=arch=compute_50,code=[compute_50,sm_50] --generate-code=arch=compute_52,code=[compute_52,sm_52] --generate-code=arch=compute_60,code=[compute_60,sm_60] --generate-code=arch=compute_61,code=[compute_61,sm_61] --generate-code=arch=compute_70,code=[compute_70,sm_70] --generate-code=arch=compute_75,code=[compute_75,sm_75] --generate-code=arch=compute_80,code=[compute_80,sm_80] --generate-code=arch=compute_86,code=[compute_86,sm_86] -MD -MT paddle\phi\kernels\CMakeFiles\phi_gpu.dir\gpu\graph_send_recv_kernel.cu.obj -MF paddle\phi\kernels\CMakeFiles\phi_gpu.dir\gpu\graph_send_recv_kernel.cu.obj.d -x cu -c D:\pythonWorkspace\Paddle\paddle\phi\kernels\gpu\graph_send_recv_kernel.cu -o paddle\phi\kernels\CMakeFiles\phi_gpu.dir\gpu\graph_send_recv_kernel.cu.obj -Xcompiler=-Fdpaddle\phi\kernels\CMakeFiles\phi_gpu.dir\phi_gpu.pdb,-FS
D:\pythonWorkspace\Paddle\paddle\phi\kernels\gpu\graph_send_recv_kernel.cu(70): error: namespace "thrust" has no member "device"
      thrust::fill(thrust::device,
                           ^

D:\pythonWorkspace\Paddle\paddle\phi\kernels\gpu\**graph_send_recv_kernel.cu**(76): error: namespace "thrust" has no member "device"
      thrust::fill(thrust::device,
                           ^

2 errors detected in the compilation of "D:/pythonWorkspace/Paddle/paddle/phi/kernels/gpu/graph_send_recv_kernel.cu".
graph_send_recv_kernel.cu
[685/3243] Building CUDA object paddle\phi\kernels\CMakeFiles\phi_gpu.dir\gpu\graph_send_ue_recv_kernel.cu.obj
FAILED: paddle/phi/kernels/CMakeFiles/phi_gpu.dir/gpu/graph_send_ue_recv_kernel.cu.obj
C:\PROGRA~1\NVIDIA~2\CUDA\v12.1\bin\nvcc.exe -forward-unknown-to-host-compiler -DCUDA_TOOLKIT_ROOT_DIR="\"C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.1\"" -DCUDA_VERSION_MAJOR=\"12\" -DCUDA_VERSION_MINOR=\"1\" -DCUDNN_MAJOR_VERSION=\"8\" -DEIGEN_STRONG_INLINE=inline -DEIGEN_USE_GPU -DGOOGLE_GLOG_DLL_DECL="" -DLAPACK_FOUND -DNOMINMAX -DPADDLE_DISABLE_PROFILER -DPADDLE_DLL_EXPORT -DPADDLE_DLL_INFERENCE -DPADDLE_ON_INFERENCE -DPADDLE_VERSION=0.0.0 -DPADDLE_VERSION_INTEGER=0 -DPADDLE_WITH_AVX -DPADDLE_WITH_CRYPTO -DPADDLE_WITH_CUDA -DPADDLE_WITH_MKLDNN -DPADDLE_WITH_MKLML -DPADDLE_WITH_POCKETFFT -DPADDLE_WITH_SSE3 -DTRT_PLUGIN_FP16_AVALIABLE -DUTF8PROC_STATIC -DWIN32_LEAN_AND_MEAN -D_XKEYCHECK_H -ID:\pythonWorkspace\Paddle\build -ID:\pythonWorkspace\Paddle\paddle\fluid\framework\io -ID:\pythonWorkspace\Paddle\build\third_party\install\zlib\include -ID:\pythonWorkspace\Paddle\build\third_party\install -ID:\pythonWorkspace\Paddle\build\third_party\install\gflags\include -ID:\pythonWorkspace\Paddle\build\third_party\install\glog\include -ID:\pythonWorkspace\Paddle\build\third_party\eigen3\src\extern_eigen3 -ID:\pythonWorkspace\Paddle\build\third_party\threadpool\src\extern_threadpool -ID:\pythonWorkspace\Paddle\build\third_party\dlpack\src\extern_dlpack\include -ID:\pythonWorkspace\Paddle\build\third_party\install\xxhash\include -ID:\pythonWorkspace\Paddle\build\third_party\install\warpctc\include -ID:\pythonWorkspace\Paddle\build\third_party\install\utf8proc\include -ID:\pythonWorkspace\Paddle\build\third_party\install\mklml\include -ID:\pythonWorkspace\Paddle\build\third_party\install\mkldnn\include -ID:\pythonWorkspace\Paddle\build\third_party\install\protobuf\include -I"C:\Program Files\Python310\include" -I"C:\Program Files\Python310\Lib\site-packages\numpy\core\include" -ID:\pythonWorkspace\Paddle\build\third_party\pybind\src\extern_pybind\include -ID:\pythonWorkspace\Paddle\build\third_party\install\cryptopp\include -ID:\pythonWorkspace\Paddle\build\third_party\pocketfft\src -ID:\pythonWorkspace\Paddle\build\third_party\dirent\src\extern_dirent\include -I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1\include" -ID:\pythonWorkspace\Paddle -ID:\pythonWorkspace\Paddle\build\..\paddle\fluid\framework\io -D_WINDOWS -Xcompiler=" /W0  /GR /EHsc"  -gencode arch=compute_75,code=sm_75 -w --expt-relaxed-constexpr --expt-extended-lambda -Xcompiler "/wd4244 /wd4267 /wd4819 " -Xcompiler /bigobj  -Xcompiler="/arch:AVX" -Xcompiler="-MT -O2 -Ob2" -DNDEBUG -std=c++14 --generate-code=arch=compute_50,code=[compute_50,sm_50] --generate-code=arch=compute_52,code=[compute_52,sm_52] --generate-code=arch=compute_60,code=[compute_60,sm_60] --generate-code=arch=compute_61,code=[compute_61,sm_61] --generate-code=arch=compute_70,code=[compute_70,sm_70] --generate-code=arch=compute_75,code=[compute_75,sm_75] --generate-code=arch=compute_80,code=[compute_80,sm_80] --generate-code=arch=compute_86,code=[compute_86,sm_86] -MD -MT paddle\phi\kernels\CMakeFiles\phi_gpu.dir\gpu\graph_send_ue_recv_kernel.cu.obj -MF paddle\phi\kernels\CMakeFiles\phi_gpu.dir\gpu\graph_send_ue_recv_kernel.cu.obj.d -x cu -c D:\pythonWorkspace\Paddle\paddle\phi\kernels\gpu\graph_send_ue_recv_kernel.cu -o paddle\phi\kernels\CMakeFiles\phi_gpu.dir\gpu\graph_send_ue_recv_kernel.cu.obj -Xcompiler=-Fdpaddle\phi\kernels\CMakeFiles\phi_gpu.dir\phi_gpu.pdb,-FS
D:\pythonWorkspace\Paddle\paddle\phi\kernels\gpu\**graph_send_ue_recv_kernel.cu(**68): error: namespace "thrust" has no member "device"
      thrust::fill(thrust::device,
                           ^

D:\pythonWorkspace\Paddle\paddle\phi\kernels\gpu\graph_send_ue_recv_kernel.cu(75): error: namespace "thrust" has no member "device"
      thrust::fill(thrust::device,
                           ^
2 errors detected in the compilation of "D:/pythonWorkspace/Paddle/paddle/phi/kernels/gpu/graph_send_ue_recv_kernel.cu".

错误提示
error: namespace "thrust" has no member "device"

目前的处理步骤
分别在以下这两个文件，添加 #include <thrust/device.h>
graph_send_ue_recv_kernel.cu
graph_send_recv_kernel.cu

类似如下：

#include <thrust/device.h>
#include <thrust/device_vector.h>
#include <thrust/fill.h>

但是再次编译依然报错

而我发现在CUDA12.1的目录：C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.1\include\thrust下根本没有device.h文件

请问各位，现在该怎么办？

[hibean]

参考 #49539 解决问题