segmentation fault when returning from a function

[Blade6570]

❓ Questions and Help

Please note that this issue tracker is not a help form and this issue will be closed.

We have a set of listed resources available on the website. Our primary means of support is our discussion forum:

• Discussion Forum
  Ubuntu: 18.04
  Pytorch:0.3.0
  cuda:9.0
  python3
  ............................
  I have run the code with gdb and the stack-trace is as follows

##0  0x00007ffff7a7b956 in __GI___libc_free (mem=0x8f8a0030) at malloc.c:3086
#1  0x00007fffbd456863 in __gnu_cxx::new_allocator<long>::deallocate (this=<optimized out>, 
    __p=<optimized out>) at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/ext/new_allocator.h:110
#2  std::_Vector_base<long, std::allocator<long> >::_M_deallocate (this=<optimized out>, 
    __n=<optimized out>, __p=<optimized out>)
    at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/bits/stl_vector.h:174
#3  std::_Vector_base<long, std::allocator<long> >::~_Vector_base (this=0x8f8a8038, 
    __in_chrg=<optimized out>) at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/bits/stl_vector.h:160
#4  std::vector<long, std::allocator<long> >::~vector (this=0x8f8a8038, __in_chrg=<optimized out>)
    at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/bits/stl_vector.h:416
#5  torch::autograd::generated::CatBackward::~CatBackward (this=0x8f8a7fb8, __in_chrg=<optimized out>)
    at torch/csrc/autograd/generated/Functions.h:220
#6  0x00007fffbd389c68 in std::_Sp_counted_base<(__gnu_cxx::_Lock_policy)2>::_M_release (this=0x8f8a7fa0)
    at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/bits/shared_ptr_base.h:144
#7  std::__shared_count<(__gnu_cxx::_Lock_policy)2>::~__shared_count (this=0x8f89c3b8, 
    __in_chrg=<optimized out>)
    at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/bits/shared_ptr_base.h:546
#8  std::__shared_ptr<torch::autograd::Function, (__gnu_cxx::_Lock_policy)2>::~__shared_ptr (
    this=0x8f89c3b0, __in_chrg=<optimized out>)
    at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/bits/shared_ptr_base.h:781
#9  std::shared_ptr<torch::autograd::Function>::~shared_ptr (this=0x8f89c3b0, __in_chrg=<optimized out>)
    at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/bits/shared_ptr.h:93
#10 std::pair<std::shared_ptr<torch::autograd::Function>, int>::~pair (this=0x8f89c3b0, 
    __in_chrg=<optimized out>) at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/bits/stl_pair.h:96
#11 std::_Destroy<std::pair<std::shared_ptr<torch::autograd::Function>, int> > (__pointer=0x8f89c3b0)
    at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/bits/stl_construct.h:93
#12 std::_Destroy_aux<false>::__destroy<std::pair<std::shared_ptr<torch::autograd::Function>, int>*> (
    __last=<optimized out>, __first=0x8f89c3b0)
    at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/bits/stl_construct.h:103
#13 std::_Destroy<std::pair<std::shared_ptr<torch::autograd::Function>, int>*> (__last=<optimized out>, 
    __first=<optimized out>) at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/bits/stl_construct.h:126
#14 std::_Destroy<std::pair<std::shared_ptr<torch::autograd::Function>, int>*, std::pair<std::shared_ptr<torch::autograd::Function>, int> > (__last=0x8f89c3e0, __first=0x8f89c3b0)
    at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/bits/stl_construct.h:151
#15 std::vector<std::pair<std::shared_ptr<torch::autograd::Function>, int>, std::allocator<std::pair<std::shared_ptr<torch::autograd::Function>, int> > >::~vector (this=0x8f8a1308, __in_chrg=<optimized out>)
    at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/bits/stl_vector.h:415
#16 torch::autograd::Function::~Function (this=0x8f8a12e8, __in_chrg=<optimized out>)
    at /pytorch/torch/csrc/autograd/function.h:77
#17 0x00007fffbd389c68 in std::_Sp_counted_base<(__gnu_cxx::_Lock_policy)2>::_M_release (this=0x8f8a12d0)
    at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/bits/shared_ptr_base.h:144
#18 std::__shared_count<(__gnu_cxx::_Lock_policy)2>::~__shared_count (this=0x8f8a37a8, 
    __in_chrg=<optimized out>)
    at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/bits/shared_ptr_base.h:546
#19 std::__shared_ptr<torch::autograd::Function, (__gnu_cxx::_Lock_policy)2>::~__shared_ptr (
    this=0x8f8a37a0, __in_chrg=<optimized out>)
    at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/bits/shared_ptr_base.h:781
#20 std::shared_ptr<torch::autograd::Function>::~shared_ptr (this=0x8f8a37a0, __in_chrg=<optimized out>)
    at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/bits/shared_ptr.h:93
#21 std::pair<std::shared_ptr<torch::autograd::Function>, int>::~pair (this=0x8f8a37a0, 
    __in_chrg=<optimized out>) at /opt/rh/devtoolset-2/root/usr/include/c++/4.8.2/bits/stl_pair.h:96
#22 std::_Destroy<std::pair<std::shared_ptr<torch::autograd::Function>, int> > (__pointer=0x8f8a37a0)
#

[zou3519]

How deep does the stack trace go?

[Blade6570]

@zou3519 I used "where" in gdb when segfault occurred.

[zou3519]

I think the problem is that you are running into a known and fixed bug of crashing on deletion of a deep computation graph. We fixed that issue in v0.4.1. Could you please upgrade your PyTorch and see if that solves the problem for you?

[zhangguanheng66]

@zou3519 feel free to close the issue after the user confirm pytorch version. Thanks

[Blade6570]

@zou3519, I think you are right and v0.4.1 solves the problem. Thanks a lot. closing the issue.