[Bug fortran/86945] [8/9 Regression] BUG with optimisation of select case statement in gfortran

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86945

--- Comment #9 from Richard Biener  ---
Author: rguenth
Date: Wed Aug 22 08:07:36 2018
New Revision: 263761

URL: https://gcc.gnu.org/viewcvs?rev=263761=gcc=rev
Log:
2018-08-22  Richard Biener  

PR tree-optimization/86945
* tree-cfg.c (generate_range_test): Use unsigned arithmetic.

* gcc.dg/torture/pr86945.c: New testcase.

Added:
trunk/gcc/testsuite/gcc.dg/torture/pr86945.c
Modified:
trunk/gcc/ChangeLog
trunk/gcc/testsuite/ChangeLog
trunk/gcc/tree-cfg.c

[Bug fortran/86945] [8/9 Regression] BUG with optimisation of select case statement in gfortran

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86945

--- Comment #8 from Richard Biener  ---
C testcase:

void __attribute__((noinline,noipa))
foo(int id)
{
  switch (id)
{
case (-__INT_MAX__ - 1)...-1:
  __builtin_abort ();
default:;
}
}

int main()
{
  foo(1);
  return 0;
}

[Bug fortran/86945] [8/9 Regression] BUG with optimisation of select case statement in gfortran

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86945

--- Comment #7 from Richard Biener  ---
With -fsanitize=undefined we get even at -O0

 id=   1
: runtime error: signed integer overflow: 1 - -2147483647 cannot be
represented in type 'integer(kind=4)'
 ierr1, OK =0 T
: runtime error: signed integer overflow: 1 - -2147483648 cannot be
represented in type 'integer(kind=4)'
 ierr2, OK =0 T

so both foo1 and foo2 do not "work".

[Bug fortran/86945] [8/9 Regression] BUG with optimisation of select case statement in gfortran

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86945

Richard Biener  changed:

   What|Removed |Added

   Priority|P4  |P2
 Status|NEW |ASSIGNED
   Assignee|unassigned at gcc dot gnu.org  |rguenth at gcc dot 
gnu.org

--- Comment #6 from Richard Biener  ---
Confirmed.  Likely caused by CFG cleanup convert_single_case_switch which
generates a range test which is done in a signed type:

   :
  id.8_1 = CHAIN.11->id;
  _3 = id.8_1 - -2147483648;
  _4 = (unsigned int) _3;
  if (_4 <= 2147483647)

when forwprop then combines this we fold (unsigned int) _3 <= 2147483647 to
_3 >= 2147483647 which then is folded as id.8_1 - -2147483648 >= 2147483647
which is then folded to id.8_1 >= 2147483647 + -2147483648.  But something
goes wrong end we end up simplifying it as true.

[Bug fortran/86945] [8/9 Regression] BUG with optimisation of select case statement in gfortran

[dominiq at lps dot ens.fr]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86945

Dominique d'Humieres  changed:

   What|Removed |Added

   Priority|P3  |P4
 Status|UNCONFIRMED |NEW
  Known to work||7.3.0
   Keywords||wrong-code
   Last reconfirmed||2018-08-16
 Ever confirmed|0   |1
Summary|BUG with optimisation of|[8/9 Regression] BUG with
   |select case statement in|optimisation of select case
   |gfortran v8.x   |statement in gfortran
   Target Milestone|--- |8.3
  Known to fail||8.2.0, 9.0

--- Comment #5 from Dominique d'Humieres  ---
The bug appeared between revisions r251624 (2017-09-03, OK) and r251946
(2017-09-10, wrong code).