Community Packages

Please read this before reporting a bug:
https://wiki.archlinux.org/title/Bug_reporting_guidelines

Do NOT report bugs when a package is just outdated, or it is in the AUR. Use the 'flag out of date' link on the package page, or the Mailing List.

REPEAT: Do NOT report bugs for outdated packages!
Tasklist

FS#70932 - [cuda] Segfault when using compiler

Attached to Project: Community Packages
Opened by Alex Dewar (billypilgrim) - Tuesday, 18 May 2021, 09:58 GMT
Last edited by Sven-Hendrik Haase (Svenstaro) - Saturday, 16 October 2021, 13:22 GMT
Task Type Bug Report
Category Upstream Bugs
Status Closed
Assigned To Sven-Hendrik Haase (Svenstaro)
Architecture x86_64
Severity High
Priority Normal
Reported Version
Due in Version Undecided
Due Date Undecided
Percent Complete 100%
Votes 2
Private No

Details

I'm currently getting a compiler segfault when building with cuda. See comments here:
https://aur.archlinux.org/packages/opencv-cuda/

This appears to be the same issue as reported here: https://github.com/NVIDIA/nccl/issues/494

Additional info:
- package: cuda v11.3.0-1
This task depends upon

Closed by  Sven-Hendrik Haase (Svenstaro)
Saturday, 16 October 2021, 13:22 GMT
Reason for closing:  Upstream
Additional comments about closing:  Can't do much here I reckon.
Comment by bartus (bartus) - Tuesday, 18 May 2021, 15:19 GMT
Same issue with `AUR/popsfit` looks like a gcc:11 bug to me.

I've dumped `nvcc` intermediate file that produces `gcc:11` segfoult.

To reproduce segfoult:

1. download `features.cudafe1.cpp` `features.cudafe1.stub.c` `features.fatbin.c`

2. install cuda

3. run `gcc features.cudafe1.cpp -I /opt/cuda/targets/x86_64-linux/include`

/usr/include/c++/11.1.0/chrono: In substitution of 'template<class _Rep, class _Period> template<class _Period2> using __is_harmonic = std::__bool_constant<(std::ratio<((_Period2::num / std::chrono::duration<_Rep, _Period>::_S_gcd(_Period2::num, _Period::num)) * (_Period::den / std::chrono::duration<_Rep, _Period>::_S_gcd(_Period2::den, _Period::den))), ((_Period2::den / std::chrono::duration<_Rep, _Period>::_S_gcd(_Period2::den, _Period::den)) * (_Period::num / std::chrono::duration<_Rep, _Period>::_S_gcd(_Period2::num, _Period::num)))>::den == 1)> [with _Period2 = _Period2; _Rep = _Rep; _Period = _Period]':
/usr/include/c++/11.1.0/chrono:512:154: required from here
/usr/include/c++/11.1.0/chrono:454:34: internal compiler error: Segmentation fault
454 | _S_gcd(intmax_t __m, intmax_t __n) noexcept
| ^~~~~~
0x1797368 internal_error(char const*, ...)
???:0
0x7e73b1 tsubst(tree_node*, tree_node*, int, tree_node*)
???:0
0x7fa400 tsubst_template_args(tree_node*, tree_node*, int, tree_node*)
???:0
0x7fa400 tsubst_template_args(tree_node*, tree_node*, int, tree_node*)
???:0
0x7fa400 tsubst_template_args(tree_node*, tree_node*, int, tree_node*)
???:0
0x7e7bfa tsubst(tree_node*, tree_node*, int, tree_node*)
???:0
0x7e7d16 tsubst(tree_node*, tree_node*, int, tree_node*)
???:0
0x7f227c lookup_template_class(tree_node*, tree_node*, tree_node*, tree_node*, int, int)
???:0
0x816d4d finish_template_type(tree_node*, tree_node*, int)
???:0
0x7ce2fe c_parse_file()
???:0
0x8a252d c_common_parse_file()
???:0
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <https://bugs.archlinux.org/> for instructions.

   features.cudafe1.cpp (1.7 MiB)
   features.cudafe1.stub.c (4.4 KiB)
   features.fatbin.c (63.8 KiB)
Comment by Sven-Hendrik Haase (Svenstaro) - Tuesday, 18 May 2021, 21:32 GMT
Fixed by cuda 11.3.0-2.
Comment by Mathias Ciliberto (ThiasTux) - Thursday, 14 October 2021, 21:42 GMT
  • Field changed: Percent Complete (100% → 0%)
The bug is still present in cuda 11.3.1-1 with the same modalities as described originally.
Comment by Sven-Hendrik Haase (Svenstaro) - Thursday, 14 October 2021, 21:44 GMT
The problem is that there's realistically nothing we can do here as downstream packagers considering nvcc is a closed source blob. I'd strongly recommend reporting this to nvidia in some fashion. nvcc should never just crash, that's always a bug on their side.
Comment by Alex Dewar (billypilgrim) - Saturday, 16 October 2021, 12:17 GMT
At least for my code I'm no longer getting segfaults (cuda v11.4.0).

Loading...