System Information

OS: Ubuntu 22.04
CUDA: 12.5
CuDNN: 9.1.1, dynamic

Detailed description

Tests produce a lot of noisy warnings like this:

[ RUN      ] Test_Torch_layers.net_logsoftmax/0, where GetParam() = CUDA/CUDA
[ WARN:0@410.036] global init.hpp:32 checkVersions cuDNN reports version 90.1 which is not compatible with the version 9.1 with which OpenCV was built
[ WARN:0@410.041] global init.hpp:32 checkVersions cuDNN reports version 90.1 which is not compatible with the version 9.1 with which OpenCV was built
[       OK ] Test_Torch_layers.net_logsoftmax/0 (11 ms)

The test was built against the same CuDNN that is used in inference. The warning does not affect result, but disturbs users.
Looks like CuDNN changed version formula in cudnnGetVersion()

The warning location: modules/dnn/src/cuda4dnn/init.hpp:26

Steps to reproduce

1. Build with CuDNN 9.x (dynamic linkage)
2. Run dnn test:

export OPENCV_TEST_DATA_PATH=...
./bin/opencv_test_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)