[Bug tree-optimization/51136] New: [4.7 Regression] ICE: SIGSEGV in unlink_virtual_phi (tree-flow-inline.h:427) with -O2 -fno-tree-dominator-opts (-ftree-tail-merge)

[Bug tree-optimization/51136] New: [4.7 Regression] ICE: SIGSEGV in unlink_virtual_phi (tree-flow-inline.h:427) with -O2 -fno-tree-dominator-opts (-ftree-tail-merge)

[zsojka at seznam dot cz]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51136

             Bug #: 51136
           Summary: [4.7 Regression] ICE: SIGSEGV in unlink_virtual_phi
                    (tree-flow-inline.h:427) with -O2
                    -fno-tree-dominator-opts (-ftree-tail-merge)
    Classification: Unclassified
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: zsojka@seznam.cz
                CC: vries@gcc.gnu.org
              Host: x86_64-pc-linux-gnu
            Target: x86_64-pc-linux-gnu


Created attachment 25824
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=25824
reduced testcase

Compiler output:
$ gcc -O -ftree-pre -ftree-tail-merge -fno-tree-dominator-opts testcase.c
==11289== Invalid read of size 8
==11289==    at 0xB29249: unlink_virtual_phi(gimple_statement_d*, tree_node*)
(tree-flow-inline.h:427)
==11289==    by 0xB2D65F: tail_merge_optimize(unsigned int)
(tree-ssa-tail-merge.c:1464)
==11289==    by 0xAF5C66: execute_pre(bool) (tree-ssa-pre.c:4919)
==11289==    by 0x8E03D7: execute_one_pass(opt_pass*) (passes.c:2074)
==11289==    by 0x8E0744: execute_pass_list(opt_pass*) (passes.c:2129)
==11289==    by 0x8E0756: execute_pass_list(opt_pass*) (passes.c:2130)
==11289==    by 0xA4230D: tree_rest_of_compilation(tree_node*)
(tree-optimize.c:420)
==11289==    by 0x695C29: cgraph_expand_function(cgraph_node*)
(cgraphunit.c:1819)
==11289==    by 0x697C0B: cgraph_optimize() (cgraphunit.c:1886)
==11289==    by 0x698139: cgraph_finalize_compilation_unit()
(cgraphunit.c:1327)
==11289==    by 0x57566A: c_write_global_declarations() (c-decl.c:10023)
==11289==    by 0x9D55F3: toplev_main(int, char**) (toplev.c:581)
==11289==    by 0x674ED2C: (below main) (in /lib64/libc-2.12.2.so)
==11289==  Address 0x0 is not stack'd, malloc'd or (recently) free'd
==11289== 
testcase.c: In function 'foo':
testcase.c:4:1: internal compiler error: Segmentation fault
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

or

$ /mnt/svn/gcc-trunk/binary-latest/bin/gcc -O2 -fno-tree-dominator-opts
testcase.c 
testcase.c: In function 'foo':
testcase.c:3:6: internal compiler error: Segmentation fault
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

Tested revisions:
r181360 - crash

[Bug tree-optimization/51136] [4.7 Regression] ICE: SIGSEGV in unlink_virtual_phi (tree-flow-inline.h:427) with -O2 -fno-tree-dominator-opts (-ftree-tail-merge)

[vries at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51136

--- Comment #1 from vries at gcc dot gnu.org 2011-11-15 14:00:44 UTC ---
Created attachment 25827
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=25827
tentative patch

The problem is that in update_vuses, we're removing the def of the
def-use-chain we're walking with
FOR_EACH_IMM_USE_STMT/FOR_EACH_IMM_USE_ON_STMT.

Postponing the removal till after we've walked all the uses fixes the problem.

[Bug tree-optimization/51136] [4.7 Regression] ICE: SIGSEGV in unlink_virtual_phi (tree-flow-inline.h:427) with -O2 -fno-tree-dominator-opts (-ftree-tail-merge)

[vries at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51136

vries at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |FIXED

--- Comment #2 from vries at gcc dot gnu.org 2011-11-15 14:09:38 UTC ---
The problem is in update_vuses, which is obsolete since the fix for PR51005.
Marking it resolved, fixed.