Segmentation fault on Mac OS X Lion involving shared variable about theano

[nouiz]

People with knowledge of Mac and Theano don't look at issue. I forward that to theano-user. I don't use a Mac, so I can't help as this seam to be related to your mac installation.

[nouiz]

Just in case you didn't see on the theano-user mailing list. Olivier Delalleau told:

"""
This is probably the same problem already reported on theano-users: http://groups.google.com/group/theano-users/browse_thread/thread/d705d2727eefb033#

I might be able to upgrade my Mac to Lion next month, if that's the case I'll have a look at it.
"""

[maddanio]

Hi,

I am having this problem after upgrading to lion. Just wanted to play with RBM code and this is getting in the way badly. In my case (running rbm.py from the deep learning tutorials) I managed to catch something fishy in one of the c-files. I posted it here. On line 158 and around you find accesses to strides[0-1], which is also where it crashes her.
I also just ran the theano test suite, it also segfaults after a while, which completely stops the tests (I suppose for it to be more informative the tests somehow would have to catch the segfaults), output:
...................................E...............Segmentation fault: 11
Is there anything else I can do to help? Would really like to play with theano :).

[nouiz]

Hi,

thanks for this file. I have found the problem. As you told, it is the
"strides[0-1]" that should not be there. In fact it is all the if that
should not be there. The bad code still run fine on other computers.
That is why it didn't got found earlier.

I'm doing a patch for this. I will try to add a test, but the test
will only cash on your computer when the op is not fixed.

When the fix is merged in the trunk, I let you know.

thanks

Fred

On Mon, Oct 10, 2011 at 4:13 AM, maddanio
[email protected]
wrote:

Hi,

I am having this problem after upgrading to lion. Just wanted to play with RBM code and this is getting in the way badly. In my case (running rbm.py from the deep learning tutorials) I managed to catch something fishy in one of the c-files. I posted it  here. On line 158 and around you find accesses to strides[0-1], which is also where it crashes her.
I also just ran the theano test suite, it also segfaults after a while, which completely stops the tests (I suppose for it to be more informative the tests somehow would have to catch the segfaults), output:
...................................E...............Segmentation fault: 11
Is there anything else I can do to help? Would really like to play with theano :).

Reply to this email directly or view it on GitHub:
#24 (comment)

[nouiz]

The fix is now in the trunk.

Can you try again to confirm that it is fixed for you?

Can you run the DimShuffle test just in case there is other problem in
your config? This is the command

nosetests theano/tensor/tests/test_elemwise.py:test_DimShuffle

thanks

Fred

2011/10/11 Frédéric Bastien [email protected]:

Hi,

thanks for this file. I have found the problem. As you told, it is the
"strides[0-1]" that should not be there. In fact it is all the if that
should not be there. The bad code still run fine on other computers.
That is why it didn't got found earlier.

I'm doing a patch for this. I will try to add a test, but the test
will only cash on your computer when the op is not fixed.

When the fix is merged in the trunk, I let you know.

thanks

Fred

On Mon, Oct 10, 2011 at 4:13 AM, maddanio
[email protected]
wrote:

Hi,

I am having this problem after upgrading to lion. Just wanted to play with RBM code and this is getting in the way badly. In my case (running rbm.py from the deep learning tutorials) I managed to catch something fishy in one of the c-files. I posted it  here. On line 158 and around you find accesses to strides[0-1], which is also where it crashes her.
I also just ran the theano test suite, it also segfaults after a while, which completely stops the tests (I suppose for it to be more informative the tests somehow would have to catch the segfaults), output:
...................................E...............Segmentation fault: 11
Is there anything else I can do to help? Would really like to play with theano :).

Reply to this email directly or view it on GitHub:
#24 (comment)

[bayerj]

Hi Fred,

I just tested the test and the two snippets form above which I posted. Everything works on my setup now, great work!

[maddanio]

Here too, thanks!

Am 11.10.2011 um 20:18 schrieb Justin Bayer:

Hi Fred,

I just tested the test and the two snippets form above which I posted. Everything works on my setup now, great work!

Reply to this email directly or view it on GitHub:
#24 (comment)

[odelalleau]

Everyone seems happy, so I'll close this issue :)