Bug in MemoryData that does not allow using arrays with n_ * size_ > 2^31

[DmitryUlyanov]

So the problem is in here:

top[0]->set_cpu_data(data_ + pos_ * size_);

pos_ and size_ are int, so I had segmentation fault due to overflow in there, I think, when using huge array. To overcome this we need to change some variables types, so please comment how you think to do it better since I did the first thing that came to my mind that would fix the crashes I had.

[longjon]

Thanks for the fix; only issue is this should probably be size_t, and please squash your commits into one (git rebase -i and git push -f to this branch). Using int for size_ is still a limitation, but it's a known issue, see discussion at #1584.

[DmitryUlyanov]

Done

[longjon]

Thanks!