On Sat, Jan 05, 2019 at 11:45:42AM -0800, Steve Kargl wrote:
> Execution timeout is: 300
> spawn [open ...]
> STOP 2
> FAIL: gfortran.dg/class_alias.f90 -Os execution test
>
> STOP 2 occurs at the end of testcase, so I suspect finalization is messed up.
This is bizarre. Changing the code where STOP 2 occurs to
if (var_p%x .ne. 2) then
print *, var_p%x
STOP 2
end if
Compiling with gfortran gives
% gfcx -o z -Os class_alias.f90 && ./z
2
STOP 2
Something is broken with -Os and your patch. The original dump
contains
if (var_p._data->x != 2)
{
{
struct __st_parameter_dt dt_parm.5;
dt_parm.5.common.filename = &"class_alias.f90"[1]{lb: 1 sz: 1};
dt_parm.5.common.line = 92;
dt_parm.5.common.flags = 128;
dt_parm.5.common.unit = 6;
_gfortran_st_write (&dt_parm.5);
_gfortran_transfer_integer_write (&dt_parm.5, &var_p._data->x, 4);
_gfortran_st_write_done (&dt_parm.5);
}
_gfortran_stop_numeric (2, 0);
}
which looks correct. The -fdump-tree-optimized looks like var_p has been
sort of skipped.
_11 = var_a._data;
_12 = _11->x;
if (_12 != 2)
goto <bb 6>; [0.04%]
else
goto <bb 7>; [99.96%]
<bb 6> [local count: 429153]:
_gfortran_stop_numeric (1, 0);
This is the 'if (var_a%x /= 2) STOP 1' chunk. We jump to <bb 7>,
which is the body of the if-endif block that contains STOP 2.
<bb 7> [local count: 428982]:
dt_parm.5.common.filename = &"class_alias.f90"[1]{lb: 1 sz: 1};
dt_parm.5.common.line = 92;
MEM[(integer(kind=4) *)&dt_parm.5] = 25769803904;
_gfortran_st_write (&dt_parm.5);
_14 = &MEM[(struct test *)_6].x;
_gfortran_transfer_integer_write (&dt_parm.5, _14, 4);
_gfortran_st_write_done (&dt_parm.5);
dt_parm.5 ={v} {CLOBBER};
_gfortran_stop_numeric (2, 0);
I would expect the basic block <bb 7> to look like
<bb 7> [local count: xxxxx]
_13 = var_p._data;
_14 = _13->x;
if (_14 != 2)
goto <bb 8>
else
goto <bb 9>
Now, what is <bb 7> above would be the contents of <bb 8> and
<bb 9> would then jump to the deallocation statements.
--
Steve
