cuda.h not found (trying to use cudNN with theano)

Hello,

I can’t make cuDNN work with theano. (I use OSX 10.9)

Here is the error:

Theano can not compile with cuDNN. We got this error:
clang: warning: treating ‘c’ input as ‘c++’ when in C++ mode, this behavior is deprecated
/var/folders/4s/rbfh6g_94jl3ztjyw8d1wb4h0000gn/T/try_flags_d0Ur73.c:4:10:
fatal error: ‘cuda.h’ file not found
#include <cuda.h>

Actually, Theano is looking for cuda.h in

/lib/Frameworks/Cuda.Framework/Versions/A/Headers/. (it finds it if I

include <cuda/cuda.h>).

How can I make it look for it in /usr/local/cuda/include ?

I’ve installed the Developer version of Theano. I’ve installed CUDA 6.5 and

copied the cudnn files in the directory /usr/local/cuda/lib

(and/usr/local/cuda/include). A friend did the same thing and it works, I don’t what happens.

Here is my .zshrc

export PYLEARN2_DATA_PATH=/Users/Adrien/Documents/projet2a/friesnsteaks/data/

export PYLEARN2_VIEWER_COMMAND=“open -Wn”

export

PATH=“/Users/Adrien/Documents/framework/pylearn2/pylearn2/scripts:$PATH”

export PYTHONPATH=$PYTHONPATH:/Users/Adrien/Documents/projet2a/

GPU

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib

#export PATH=“/usr/local/cuda:$PATH”

export CUDA_ROOT=“/usr/local/cuda/bin:$PATH”

export PATH=“/Developer/NVIDIA/CUDA-6.5/bin:$PATH”

export

THEANO_FLAGS=mode=FAST_RUN,device=gpu,floatX=float32,optimizer_including=cudnn,exception_verbosity=high

Any ideas please ?

I met the same problem as yours when I replace tensor RT version 2 with that version 3.x

I met the same problem as yours when I replace tensor RT version 2 with that version 3.x