[Build] Failed when build with cuda

[ZhangWei125521]

Describe the issue

It always occur error even though I have copy the cudnn lib and include files into the cuda path("/usr/local/cuda/").
the failed file: "onnxruntime/contrib_ops/cuda/bert/cudnn_fmha/cudnn_flash_attention.cu.o".

Urgency

No response

Target platform

Ubuntu

Build script

./build.sh --use_cuda --config Debug --cuda_home /usr/local/cuda/ --cudnn_home /usr/local/cuda/

Error / output

the detail error in the attached image

Visual Studio Version

No response

GCC / Compiler Version

No response

[xiaoyu-work]

You could download cudnn separately and specify cudnn path in the build command:

wget  https://developer.download.nvidia.com/compute/cudnn/redist/cudnn/linux-x86_64/cudnn-linux-x86_64-9.6.0.74_cuda12-archive.tar.xz 
mkdir $HOME/cudnn9.6 
tar -Jxvf cudnn-linux-x86_64-9.6.0.74_cuda12-archive.tar.xz  -C $HOME/cudnn9.6 --strip=1  --no-overwrite-dir 

[ZhangWei125521]

You could download cudnn separately and specify cudnn path in the build command:

wget  https://developer.download.nvidia.com/compute/cudnn/redist/cudnn/linux-x86_64/cudnn-linux-x86_64-9.6.0.74_cuda12-archive.tar.xz 
mkdir $HOME/cudnn9.6 
tar -Jxvf cudnn-linux-x86_64-9.6.0.74_cuda12-archive.tar.xz  -C $HOME/cudnn9.6 --strip=1  --no-overwrite-dir 

Many thanks for your repy, but it seems not cudnn issue. In the same env, I can compile version 1.18.1 successful, Now I use this version production

[xiaoyu-work]

1.18 doesn't need cudnn as a mandatory requirement. Glad it works for you now. Feel free to reopen this issue if you have more questions.