Bug 53019 - [4.7 regression] ICE (segfault) with -flto=jobserver -fwhole-program
Summary: [4.7 regression] ICE (segfault) with -flto=jobserver -fwhole-program
Status: RESOLVED FIXED
Alias: None
Product: gcc
Classification: Unclassified
Component: lto (show other bugs)
Version: 4.7.0
: P2 normal
Target Milestone: 4.7.1
Assignee: Not yet assigned to anyone
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2012-04-17 12:57 UTC by Matthias Klose
Modified: 2012-06-01 09:12 UTC (History)
1 user (show)

See Also:
Host:
Target:
Build:
Known to work: 4.6.3
Known to fail: 4.7.0
Last reconfirmed: 2012-04-26 00:00:00

Attachments
test case (5.71 KB, application/x-gzip)
2012-04-17 12:57 UTC, Matthias Klose 		Details
View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description Matthias Klose 2012-04-17 12:57:37 UTC 
Created attachment 27177 [details]
test case

works with 4.6.3, fails with 4.7 (at least on x86_64-linux-gnu)

$ make
g++-4.7 -O2 -flto=jobserver -fwhole-program -c clua.cc -o clua.o
g++-4.7 -O2 -flto=jobserver -fwhole-program -c l_global.cc -o l_global.o
g++-4.7 -O2 -flto=jobserver -fwhole-program clua.o l_global.o -o /dev/null
In file included from :22:0:
l_global.cc: In function '_ZNSt8_Rb_treeISsSt4pairIKSs18basic_text_patternIXadL_Z15compile_patternPKcbEEXadL_Z21free_compiled_patternPvEEXadL_Z13pattern_matchS5_S4_iEEEESt10_Select1stIS7_ESt4lessISsEN9__gnu_cxx13new_allocatorIS7_EEE14_M_lower_boundEPSt13_Rb_tree_nodeIS7_ESI_RS1_.isra.12.3024.constprop.51':
l_global.cc:310:169: internal compiler error: Segmentation fault
Please submit a full bug report,
with preprocessed source if appropriate.
Comment 1 Andrew Pinski 2012-04-24 23:03:11 UTC 
This works on the trunk, we get undefined references.

Are you using gold or BFD ld?
Comment 2 Richard Biener 2012-04-26 14:20:19 UTC 
Confirmed.  It is an ICE in set_is_used, the var annotation is NULL.  Called
from remove_unused_locals.

This sounds somewhat familiar ...
Comment 3 Adam Borowski 2012-05-08 20:20:48 UTC 
Trunk fails for me, although not on this reduced testcase.
The message says the problem is the same:

In file included from :8272:0:
/usr/lib/gcc-snapshot/include/c++/4.8.0/bits/stl_tree.h: In function '_ZNSt8_Rb_treeISsSt4pairIKSs18basic_text_patternIXadL_Z15compile_patternPKcbEEXadL_Z21free_compiled_patternPvEEXadL_Z13pattern_matchS5_S4_iEEEESt10_Select1stIS7_ESt4lessISsESaIS7_EE14_M_lower_boundEPSt13_Rb_tree_nodeIS7_ESG_RS1_.isra.17.794313.constprop.6959':
/usr/lib/gcc-snapshot/include/c++/4.8.0/bits/stl_tree.h:1093:0: internal compiler error: Segmentation fault
     _Rb_tree<_Key, _Val, _KeyOfValue, _Compare, _Alloc>::
 ^

Both gold and BFD.

Should I delta it again to get something that breaks on both trunk and 4.7?
Comment 4 Richard Biener 2012-05-30 14:38:37 UTC 
Fixed on trunk, no longer reproduces on the 4.7 branch for me (rev. 188004),
reproduces with branch rev. 187228.  Can you re-check please?
Comment 5 Adam Borowski 2012-05-30 21:56:36 UTC 
The changelog for Debian upload which included the fix is:

  * Update to SVN 20120509 (r187339) from the gcc-4_7-branch.
    - Fix PR libstdc++/53193, PR target/53272, PR tree-optimization/53239,
      PR tree-optimization/53195, PR target/52999, PR target/53228,
      PR tree-optimization/52633, PR tree-optimization/52870, PR target/48496,
      PR target/53199, PR target/52684, PR lto/52605, PR plugins/53126,
      PR debug/53174, PR target/53187, PR tree-optimization/53144,
      PR c++/53186, PR fortran/53255, PR fortran/53111, PR fortran/52864.
Previous one was SVN 20120502 (r187039).

I'll try trunk, although with its complex build system and me being quite busy, I sadly don't promise to finish it by tonight.
Comment 6 Adam Borowski 2012-06-01 08:04:43 UTC 
Ok, works correctly on current trunk as well.

Since the testcase is messy and vulnerable to small alterations, and I guess the underlying bug fixes already include a test for it, it's probably pointless to try to introduce a variant of this one.

Thus, nothing more to do here.  Thanks guys for the fix, let's close this one.
Comment 7 Richard Biener 2012-06-01 09:12:08 UTC 
Fixed.