Re: "Segmentation fault (core dumped)" as re: segmentation fault
The array I was using is indeed from a pickle. I ran the c example, and got the following:
$ LD_PRELOAD=/usr/lib/sse2/libcblas.so.3.0 ./sse2-crash-test good_align: ok bad_align: Segmentation fault (core dumped)
$ LD_PRELOAD=/usr/lib/libcblas.so.3.0 ./sse2-crash-test good_align: ok bad_align: ok
So it looks as if this is indeed the problem. At least it explains why my code ran well originally, I used either fresh arrays or ran on Windows, without ATLAS. This is one obscure bug!
Thanks for the help. For the moment is there anything I can do to keep it from reoccurring in other contexts? I am copying the arrays now before use, but odds are I will forget that one time. Is using scipy.io instead of cPickle going to improve the situation?