[Bug fortran/106208] New: [12/13 Regression] ICE in branch_prob, at profile.cc:1459

[Bug fortran/106208] New: [12/13 Regression] ICE in branch_prob, at profile.cc:1459

[gscfq@t-online.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106208

            Bug ID: 106208
           Summary: [12/13 Regression] ICE in branch_prob, at
                    profile.cc:1459
           Product: gcc
           Version: 13.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: gscfq@t-online.de
  Target Milestone: ---

Started between 20211114 and 20211121,
with file gfortran.dg/coarray/registering_1.f90 :
(gcc configured with --enable-checking=yes)


$ gfortran-13-20220703 -c registering_1.f90 -ftest-coverage -fcoarray=lib
during IPA pass: profile
registering_1.f90:11:4:

   11 | use m
      |    ^
internal compiler error: in branch_prob, at profile.cc:1459
0xdca3c3 branch_prob(bool)
        ../../gcc/profile.cc:1459
0xf8f060 tree_profiling
        ../../gcc/tree-profile.cc:782
0xf8f060 execute
        ../../gcc/tree-profile.cc:888

[Bug fortran/106208] [12/13 Regression] ICE in branch_prob, at profile.cc:1459

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106208

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |12.2

[Bug fortran/106208] [12/13 Regression] ICE in branch_prob, at profile.cc:1459 since r12-5275-gbcebd05720540e25

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106208

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |marxin at gcc dot gnu.org,
                   |                            |tschwinge at gcc dot gnu.org
   Last reconfirmed|                            |2022-07-07
             Status|UNCONFIRMED                 |NEW
     Ever confirmed|0                           |1
            Summary|[12/13 Regression] ICE in   |[12/13 Regression] ICE in
                   |branch_prob, at             |branch_prob, at
                   |profile.cc:1459             |profile.cc:1459 since
                   |                            |r12-5275-gbcebd05720540e25

--- Comment #1 from Martin Liška <marxin at gcc dot gnu.org> ---
Reduced test-case:

$ cat reg.f90
module m
  integer a0[*]  
  end 

$ gcc reg.f90 -c -fcoarray=lib -ftest-coverage
during IPA pass: profile
f951: internal compiler error: in branch_prob, at profile.cc:1459
0x770d2d branch_prob(bool)
        /home/marxin/Programming/gcc/gcc/profile.cc:1459
0x10cde00 tree_profiling
        /home/marxin/Programming/gcc/gcc/tree-profile.cc:782
0x10cde00 execute
        /home/marxin/Programming/gcc/gcc/tree-profile.cc:888
Please submit a full bug report, with preprocessed source (by using
-freport-bug).
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.

Started with r12-5275-gbcebd05720540e25.

[Bug middle-end/106208] [12/13 Regression] ICE in branch_prob, at profile.cc:1459 since r12-5275-gbcebd05720540e25

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106208

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|fortran                     |middle-end
           Priority|P3                          |P2

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
#1  0x0000000001423960 in branch_prob (thunk=false)
    at /home/rguenther/src/trunk/gcc/profile.cc:1459
1459                  gcc_checking_assert (!RESERVED_LOCATION_P (loc));
(gdb) l
1454              gcov_position_t offset = 0;
1455
1456              if (bb == ENTRY_BLOCK_PTR_FOR_FN (cfun)->next_bb)
1457                {
1458                  location_t loc = DECL_SOURCE_LOCATION
(current_function_decl);
1459                  gcc_checking_assert (!RESERVED_LOCATION_P (loc));
1460                  seen_locations.add (loc);
1461                  expanded_location curr_location = expand_location (loc);
1462                  output_location (&streamed_locations, curr_location.file,
1463                                   MAX (1, curr_location.line), &offset,
bb);
(gdb) p loc
$1 = 0
(gdb) p current_function_decl
$2 = <function_decl 0x7ffff6317d00 _caf_init.0>
(gdb) p $2->decl_common.artificial_flag 
$3 = 0

I was to say that for artificial functions a reserved location could be
expected.  But _caf_init.0 isn't even artificial (which is probably a bug).

In the non-entry block case we do

              if (!RESERVED_LOCATION_P (loc))
                {
                  seen_locations.add (loc);

which seems an obvious thing to do here as well.  But it's really Thomas who
should comment since he put in the assert (uncovering a previous issue).

Not really a frontend bug (if it were DECL_ARTIFICIAL the ICE would occur
anyway).

[Bug middle-end/106208] [12/13 Regression] ICE in branch_prob, at profile.cc:1459 since r12-5275-gbcebd05720540e25

[tschwinge at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106208

Thomas Schwinge <tschwinge at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |tschwinge at gcc dot gnu.org

--- Comment #3 from Thomas Schwinge <tschwinge at gcc dot gnu.org> ---
Indeed I had added the 'assert' to prevent bogus behavior further down the line
-- and we've now found another offender.  I'm planning to look into that, but
it'll take a while (summer vacations/parental leave).


(In reply to Martin Liška from comment #1)
> Reduced test-case: [...]

Thanks!

[Bug middle-end/106208] [12/13 Regression] ICE in branch_prob, at profile.cc:1459 since r12-5275-gbcebd05720540e25

[tschwinge at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106208

Thomas Schwinge <tschwinge at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=101598
                 CC|                            |ebotcazou at gcc dot gnu.org,
                   |                            |edlinger at gcc dot gnu.org

--- Comment #5 from Thomas Schwinge <tschwinge at gcc dot gnu.org> ---
This ICE (when compiling with '-ftest-coverage -fcoarray=lib') reported here
originally: 'gfortran.dg/coarray/registering_1.f90' as well as Martin's reduced
test case 'reg.f90' disappears via Eric's PR101598-related commit
r13-3485-gcb21297f9259ceedd5f5dd7c3973535f14124d6d "Relax assertion in
profiler".

Should we still add these Fortran test cases -- or is in fact 'gfortran' still
generating "undesirable" location information?  Stated by Eric: "This assertion
in branch_prob [...] had been correct until the fix for PR debug/101598 was
installed", but that clearly doesn't apply to the Fortran case discussed in
this PR, as Bernd's commit r12-2846-ga45918f8a7444a40eb397a037683ba5900a2db74
"Ada: Remove debug line number for DECL_IGNORED_P functions" was relevant for
Ada only; 'gcc/ada/gcc-interface/trans.c'.

[Bug middle-end/106208] [12/13/14 Regression] ICE in branch_prob, at profile.cc:1459 since r12-5275-gbcebd05720540e25

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106208

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|12.3                        |12.4

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
GCC 12.3 is being released, retargeting bugs to GCC 12.4.

[Bug middle-end/106208] [12 Regression] ICE in branch_prob, at profile.cc:1459 since r12-5275-gbcebd05720540e25

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106208

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[12/13/14 Regression] ICE   |[12 Regression] ICE in
                   |in branch_prob, at          |branch_prob, at
                   |profile.cc:1459 since       |profile.cc:1459 since
                   |r12-5275-gbcebd05720540e25  |r12-5275-gbcebd05720540e25

--- Comment #7 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
>Should we still add these Fortran test cases -- or is in fact 'gfortran' still generating "undesirable" location information? 


The function (_caf_init.0) is a machine generated function and does not really
have any location information associated with it. it is an artifact of
fcoarray=lib registering the coarray at initialization time.

So I am going to say this just needs a testcase, removing the 13/14 regression
marker otherwise.