[CV project debugging] CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT problem

Foreword

error

Copy after login

./src/convolutional_layer.c:153:13: error: ‘CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT’ undeclared (first use in this function)
 153
1.
2.

Reason

CUDNN_CONVOLUTION_FWD_SPECIFY_WORKSPACE_LIMIT is not defined in cudnn8.x, and cudnn7.x cannot be used with CUDA11.x, but the GPU of RTX30 sequence must use CUDA11.x to run normally, feels like a dead end.Later, after searching for a long time, I found that NVIDIA gave a solution code for cudnn8, which is to modify the code of the erroneous file src/convolutional_layer.c and increase the processing for CUDNN_MAJOR>=8.

Reference

1. ​ ​yolo_with_cudnn​​;

2. ​ ​How to solve the problem that the pjreddie version of darknet cannot be compiled with cudnn8​​;

End