opencv: DNN module fails to compile against cuDNN 9.0

System Information

OpenCV version: 4.x (09/02/2024) OS: Windows 11 Compiler: VS 2022 CUDA: 12.3 cuDNN: 9.0

Detailed description

Switching from cuDNN 8.9.7 to 9.0 results in the following build error

D:\repos\opencv\opencv\modules\dnn\src\cuda4dnn\csl\cudnn/recurrent.hpp(122): error C3861: 'cudnnSetRNNDescriptor_v6': identifier not found

when compiling the DNN module.

Full error trace

[372/491] Building CXX object modules\dnn\CMakeFiles\opencv_dnn.dir\Release\src\layers\recurrent_layers.cpp.obj
FAILED: modules/dnn/CMakeFiles/opencv_dnn.dir/Release/src/layers/recurrent_layers.cpp.obj
C:\PROGRA~1\MICROS~2\2022\COMMUN~1\VC\Tools\MSVC\1436~1.325\bin\Hostx64\x64\cl.exe  /nologo /TP -DCVAPI_EXPORTS -DCV_CUDA4DNN=1 -DCV_OCL4DNN=1 -DENABLE_PLUGINS -DHAVE_FLATBUFFERS=1 -DHAVE_PROTOBUF=1 -D_CRT_SECURE_NO_WARNINGS=1 -D_USE_MATH_DEFINES -D_VARIADIC_MAX=10 -D_WIN32_WINNT=0x0601 -D__OPENCV_BUILD=1 -D__STDC_CONSTANT_MACROS -D__STDC_FORMAT_MACROS -D__STDC_LIMIT_MACROS -DCMAKE_INTDIR=\"Release\" -ID:\build\opencv\cuda_12_3_dnn\3rdparty\ippicv\ippicv_win\icv\include -ID:\build\opencv\cuda_12_3_dnn\3rdparty\ippicv\ippicv_win\iw\include -ID:\repos\opencv\opencv\modules\dnn\src -ID:\repos\opencv\opencv\modules\dnn\include -ID:\build\opencv\cuda_12_3_dnn\modules\dnn -ID:\repos\opencv\contrib\modules\cudev\include -ID:\repos\opencv\opencv\modules\core\include -ID:\repos\opencv\opencv\modules\imgproc\include -ID:\repos\opencv\opencv\modules\dnn\misc\caffe -ID:\repos\opencv\opencv\modules\dnn\misc\tensorflow -ID:\repos\opencv\opencv\modules\dnn\misc\onnx -ID:\repos\opencv\opencv\modules\dnn\misc\tflite -ID:\repos\opencv\opencv\3rdparty\include\opencl\1.2 -ID:\repos\opencv\opencv\modules\ts\include -ID:\repos\opencv\opencv\modules\imgcodecs\include -ID:\repos\opencv\opencv\modules\videoio\include -ID:\repos\opencv\opencv\modules\highgui\include -external:ID:\build\opencv\cuda_12_3_dnn -external:I"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.3\include" -external:ID:\repos\opencv\opencv\3rdparty\flatbuffers\include -external:ID:\repos\opencv\opencv\3rdparty\protobuf\src -external:W0 /DWIN32 /D_WINDOWS /W4 /GR  /D _CRT_SECURE_NO_DEPRECATE /D _CRT_NONSTDC_NO_DEPRECATE /D _SCL_SECURE_NO_WARNINGS /Gy /bigobj /Oi  /fp:precise /FS     /EHa /wd4127 /wd4251 /wd4324 /wd4275 /wd4512 /wd4589 /wd4819 /wd4244 /wd4267 /wd4018 /wd4355 /wd4800 /wd4251 /wd4996 /wd4146 /wd4305 /wd4127 /wd4100 /wd4512 /wd4125 /wd4389 /wd4510 /wd4610 /wd4702 /wd4456 /wd4457 /wd4065 /wd4310 /wd4661 /wd4506 /wd4125 /wd4267 /wd4127 /wd4244 /wd4512 /wd4702 /wd4456 /wd4510 /wd4610 /wd4800 /wd4701 /wd4703 /wd4505 /wd4458  /O2 /Ob2 /DNDEBUG  /Zi -MD /showIncludes /Fomodules\dnn\CMakeFiles\opencv_dnn.dir\Release\src\layers\recurrent_layers.cpp.obj /Fdlib\Release\opencv_dnn490.pdb /FS -c D:\repos\opencv\opencv\modules\dnn\src\layers\recurrent_layers.cpp
D:\repos\opencv\opencv\modules\dnn\src\cuda4dnn\csl\cudnn/recurrent.hpp(122): error C3861: 'cudnnSetRNNDescriptor_v6': identifier not found
D:\repos\opencv\opencv\modules\dnn\src\cuda4dnn\csl\cudnn/recurrent.hpp(100): note: while compiling class template member function 'cv::dnn::cuda4dnn::csl::cudnn::RNNDescriptor<T>::RNNDescriptor(const cv::dnn::cuda4dnn::csl::cudnn::Handle &,cv::dnn::cuda4dnn::csl::cudnn::RNNDescriptor<T>::RNNMode,int,int,bool,const cv::dnn::cuda4dnn::csl::cudnn::DropoutDescriptor &)'
        with
        [
            T=float
        ]
D:\repos\opencv\opencv\modules\dnn\src\cuda4dnn\primitives\../csl/tensor_ops.hpp(541): note: see the first reference to 'cv::dnn::cuda4dnn::csl::cudnn::RNNDescriptor<T>::RNNDescriptor' in 'cv::dnn::cuda4dnn::csl::LSTM<T>::LSTM'
        with
        [
            T=float
        ]
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/recurrent_cells.hpp(48): note: see the first reference to 'cv::dnn::cuda4dnn::csl::LSTM<T>::LSTM' in 'cv::dnn::cuda4dnn::LSTMOp<float>::LSTMOp'
        with
        [
            T=float
        ]
D:\repos\opencv\opencv\modules\dnn\src\cuda4dnn\primitives\../../op_cuda.hpp(196): note: see the first reference to 'cv::dnn::cuda4dnn::LSTMOp<float>::LSTMOp' in 'cv::dnn::make_cuda_node'
D:\repos\opencv\opencv\modules\dnn\src\cuda4dnn\primitives\../csl/tensor_ops.hpp(511): note: see reference to class template instantiation 'cv::dnn::cuda4dnn::csl::cudnn::RNNDescriptor<T>' being compiled
        with
        [
            T=float
        ]
D:\repos\opencv\opencv\modules\dnn\src\layers\../cuda4dnn/primitives/recurrent_cells.hpp(88): note: see reference to class template instantiation 'cv::dnn::cuda4dnn::csl::LSTM<T>' being compiled
        with
        [
            T=float
        ]
D:\repos\opencv\opencv\modules\dnn\src\cuda4dnn\primitives\../../op_cuda.hpp(196): note: see reference to class template instantiation 'cv::dnn::cuda4dnn::LSTMOp<float>' being compiled
D:\repos\opencv\opencv\modules\dnn\src\layers\recurrent_layers.cpp(763): note: see reference to function template instantiation 'cv::Ptr<cv::dnn::dnn4_v20231225::BackendNode> cv::dnn::make_cuda_node<cv::dnn::cuda4dnn::LSTMOp,cv::dnn::cuda4dnn::csl::Stream,cv::dnn::cuda4dnn::csl::cudnn::Handle,cv::Mat&,cv::Mat&,cv::Mat&,cv::dnn::cuda4dnn::RNNConfiguration&>(int,cv::dnn::cuda4dnn::csl::Stream &&,cv::dnn::cuda4dnn::csl::cudnn::Handle &&,cv::Mat &,cv::Mat &,cv::Mat &,cv::dnn::cuda4dnn::RNNConfiguration &)' being compiled
[393/491] Building CXX object modules\dnn\CMakeFiles\opencv_dnn.dir\Release\src\layers\split_layer.cpp.obj
ninja: build stopped: subcommand failed.

Steps to reproduce

cmake --build . --target opencv_dnn

Issue submission checklist

I report the issue, it’s not a question
I checked the problem with documentation, FAQ, open issues, forum.opencv.org, Stack Overflow, etc and have not found any solution
I updated to the latest OpenCV version and the issue is still there
There is reproducer code and related data files (videos, images, onnx, etc)

About this issue

Original URL
State: open
Created 5 months ago
Comments: 34 (6 by maintainers)

Most upvoted comments

Out of boredom and curiosity I tried to update the code on my own to accommodate for the breaking changes. For now it seems to build with cudnn 9, I just need to update it to be compatible with CUDA 12.4 as well.

Will push a PR upwards when I’m done. I’m guessing this will need to be for OpenCV 5? @cudawarped This would be my first contribution to openCV.

ZelboK on Apr 13, 2024

I hit the same issue on Windows with Cuda 12,3 and cuDNN 9.0 with generator Visual Studio 16 2019

Have you solved the issue?? If you have please mention what you have done

I went down to cuDNN 8.9.7 and that fixed this issue for me.

kramamurthi on Mar 12, 2024

I have a PR up for this but it seems like one of hte builds broke. I don’t have time to fix this for a while but I am pretty sure cuda 12.4’s thrust toolkit has some problems when it comes to the tuples.

If you’re getting errors like

: error: incomplete type is not allowed
      static_assert((VecTraits<DstType>::cn == tuple_size<SrcPtrTuple>::value), "" " " "VecTraits<DstType>::cn == tuple_size<SrcPtrTuple>::value");

then be aware that there was a bug in calculating the fake tuple_size. That is fixed in CCCL’s main however.

ZelboK on May 1, 2024

What is the Opencv version used and can you please mention what flags have you enabled

Use the latest version of OpenCV with CUDA 12.3 and cuDNN 8.97 (currently neither CUDA 12.4 or cuDNN 9.0 are supported by OpenCV).

For flags see https://cudawarped.github.io/opencv-experiments/qmd/opencv_cuda_python_windows.html#building-opencv-with-cmake

cudawarped on Mar 18, 2024

I hit the same issue on Windows with Cuda 12,3 and cuDNN 9.0 with generator Visual Studio 16 2019

kramamurthi on Mar 4, 2024

@cudawarped Thanks; cuDNN 8.9.7 fixed the issue for me.

pfmephisto on Feb 20, 2024

I have what I think is the same issue on Ubuntu 22.04.3 LTS error: ‘cudnnSetRNNDescriptor_v6’ was not declared in this scope; did you mean ‘cudnnSetRNNDescriptor_v8’? Tested on OpenCV 4.9.0 and 4.x dev branch.

pfmephisto on Feb 19, 2024