Weird EIGEN Error While Installing PyCnn V2.0 about dynet

[neubig]

I wasn't able to reproduce this with the latest version of Eigen and cnn on El Capitan. Would you mind providing the version numbers of eigen, cnn, clang, and cnn v2?

[talbaumel]

I'm using Sierra, it might have something to do with that
All other libraries are also the latest

[talbaumel]

I tried installing pycnn V1 from scratch, the CPU went smoothly
After executing:

cmake .. -DBACKEND=cuda -DCUDA_TOOLKIT_ROOT_DIR=/Developer/NVIDIA/CUDA-7.5/
make -j 4

I got the following errors:

make[2]: *** [cnn/CMakeFiles/cnncuda_shared.dir/cnncuda_shared_generated_gpu-ops.cu.o] Error 1
make[2]: *** Waiting for unfinished jobs....
CMake Error at cnncuda_generated_gpu-ops.cu.o.cmake:263 (message):
Error generating file
/Users/talbaumel/cnn/build/cnn/CMakeFiles/cnncuda.dir//./cnncuda_generated_gpu-ops.cu.o

make[2]: *** [cnn/CMakeFiles/cnncuda.dir/cnncuda_generated_gpu-ops.cu.o] Error 1
make[1]: *** [cnn/CMakeFiles/cnncuda.dir/all] Error 2
/usr/include/stdio.h(133): error: expected a ")"

/usr/include/stdio.h(134): error: expected a ")"

/usr/include/stdio.h(134): error: "_Nullable" has already been declared in the current scope

/usr/include/stdio.h(135): error: expected a ")"

/usr/include/stdio.h(135): error: "_Nullable" has already been declared in the current scope

/usr/include/stdio.h(136): error: expected a ")"

/usr/include/stdio.h(136): error: "_Nullable" has already been declared in the current scope

/usr/include/stdio.h(477): error: duplicate parameter name

/usr/include/stdio.h(478): error: duplicate parameter name

/usr/include/stdio.h(479): error: duplicate parameter name

/usr/include/stdlib.h(139): error: expected a ")"

/usr/include/stdlib.h(158): error: expected a ")"

/usr/include/stdlib.h(294): error: expected a ")"

/usr/include/stdlib.h(300): error: expected a ")"

/usr/include/stdlib.h(306): error: expected a ")"

/usr/include/stdlib.h(312): error: expected a ")"

/usr/include/stdlib.h(318): error: expected a ")"

17 errors detected in the compilation of "/var/folders/fr/6yxg1k8j1xsfd5v3b4z20c480000gn/T//tmpxft_000040c3_00000000-22_gpu-ops.compute_52.cpp1.ii".
CMake Error at cnncuda_generated_gpu-ops.cu.o.cmake:263 (message):
Error generating file
/Users/talbaumel/cnn/build/cnn/CMakeFiles/cnncuda.dir//./cnncuda_generated_gpu-ops.cu.o

[neubig]

v1 will be deprecated very soon so I'll focus on v2. I don't have sierra at hand, but once I get my hands on a sierra machine I can try to debug. Of course, if you fix the problem yourself a patch would be appreciated.

[neubig]

@talbaumel Is this issue resolved now? If so, can you close it?

[talbaumel]

I installed the latest DyNet version using this manual (http://dynet.readthedocs.io/en/latest/python.html) on the os x 10.12

The CPU version worked great
The GPU version didn't (CUDA installed and using GeForce GT 750M)
Same machine is running Theano with GPU support

I can recreate the error but I don't think it's readable

Btw, on my mac while following the instructions I had to change the eigen path when compiling

[neubig]

Hmm, OK. Thanks for checking anyway. This'll have to wait until someone else can replicate, but my Mac doesn't have an nvidia GPU so we're stuck for now.

[neubig]

CUDA seems to be working on OSX since this: #305

Are things working now?

[talbaumel]

I updated CUDA and it didn't work

btw, the cmake line in the documantation should be:
cmake .. -DEIGEN3_INCLUDE_DIR=../../eigen -DPYTHON=which python -DBACKEND=cuda
intead of:
cmake .. -DEIGEN3_INCLUDE_DIR=../eigen -DPYTHON=which python -DBACKEND=cuda

[neubig]

Sorry, not updating CUDA, updating DyNet. Are you using the most recent version from master? And thanks for pointing out the error with the Eigen path.

[talbaumel]

Yeah, I tried to install the latest version

Thank you for the support and the great library!

[neubig]

Just checking regularly, has this been resolved with the latest version? I've seen this bug reported one other time but never have been able to reproduce it myself.