El Capitan & Dynamic Libraries

[elbamos]

I get this when I try to run the test script:

CUDArray: Failed to load CUDA back-end. Traceback (most recent call last): File "./test.py", line 7, in import cudarray as ca File "redacted/lib/python2.7/site-packages/cudarray-0.1.dev0-py2.7-macosx-10.5-x86_64.egg/cudarray/init.py", line 20, in from .cudarray import * File "redacted/lib/python2.7/site-packages/cudarray-0.1.dev0-py2.7-macosx-10.5-x86_64.egg/cudarray/cudarray.py", line 2, in from .wrap.array_data import ArrayData ImportError: dlopen(redacted/lib/python2.7/site-packages/cudarray-0.1.dev0-py2.7-macosx-10.5-x86_64.egg/cudarray/wrap/array_data.so, 2): Library not loaded: @rpath/libcudart.7.5.dylib Referenced from: redacted/lib/python2.7/site-packages/cudarray-0.1.dev0-py2.7-macosx-10.5-x86_64.egg/cudarray/wrap/array_data.so Reason: image not found

I suspect the cause is, in OS X El Capitan, its no longer possible to set DYLD_FALLBACK_LIBRARY_PATH or its friends. (Well, if it is possible, it now required additional hoops.)

[elbamos]

With all of cuda/lib linked from /usr/local/lib, I get this from test.py, which one suspects would be related to the issue above:

True Traceback (most recent call last): File "examples/test.py", line 493, in run() File "examples/test.py", line 483, in run test_binary() File "examples/test.py", line 257, in test_binary c_ca = ca.divide(a_ca, b_ca) File "/Volumes/home500/anaconda/envs/amos27/lib/python2.7/site-packages/cudarray-0.1.dev0-py2.7-macosx-10.5-x86_64.egg/cudarray/elementwise.py", line 173, in divide return binary(elementwise.div_op, x1, x2, out) File "/Volumes/home500/anaconda/envs/amos27/lib/python2.7/site-packages/cudarray-0.1.dev0-py2.7-macosx-10.5-x86_64.egg/cudarray/elementwise.py", line 153, in binary 'Broadcast of non-commutative operations not supported' NotImplementedError: Broadcast of non-commutative operations not supported

[andersbll]

Better late than never! :)

Your installation is working. The last error is caused by CUDArray lacking broadcasting support for some operations in some rare cases. It should be easy to add - I just haven't needed that functionality yet.

[nresnick]

@elbamos Did you figure this out? Getting the original error you posted. Tried setting DYLD_FALLBACK_LIBRARY_PATH as well with no luck.

[elbamos]

No, I ended up forking a torch implementation of neural style.

On Apr 13, 2016, at 12:36 PM, nresnick [email protected] wrote:

@elbamos Did you figure this out? Getting the original error you posted. Tried setting DYLD_FALLBACK_LIBRARY_PATH as well with no luck.

— You are receiving this because you were mentioned. Reply to this email directly or view it on GitHub