[Bug fortran/89282] New: Garbage arithmetics results in fortran with -O3 and overloaded operators

[perini at wisc dot edu]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89282

            Bug ID: 89282
           Summary: Garbage arithmetics results in fortran with -O3 and
                    overloaded operators
           Product: gcc
           Version: 6.2.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: perini at wisc dot edu
  Target Milestone: ---

Created attachment 45654
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=45654&action=edit
Small test program reproducing the error

Hi, 

I'm getting garbage results out of simple arithmetics (i.e., A minus B) when
called in from a derived type, and defined using array-based notation. The
problem only appears when compiling using -O3 optimization flag

I am not sure whether this issue comes from wrong array bounds checking or its
related with the processor's cacheline. It seems that the error appears only
for some specific array sizes, i.e., n=5. Anyway, the attached program
reproduces the error with gcc/gfortran 6.2.0 on Windows:

The program is trying to compute 3.d0 - [(1.d0, j=1,5)], where the data array
is contained in a derived type, and the "scalar minus array" operation is
performed using an overloaded operator. All array elements from the test should
return 2.d0. 
Compiled using:

gfortran -O3 test_gcc_overloaded.f90

it returns:

$ a.exe
type(t) object <r1>, size=5
  x(1) = 2.00000E+00
  x(2) = 2.00000E+00
  x(3) = 2.00000E+00
  x(4) = 2.00000E+00
  x(5) = 4.46848-317
type(t) object <r1>, size=6
  x(1) = 2.00000E+00
  x(2) = 2.00000E+00
  x(3) = 2.00000E+00
  x(4) = 2.00000E+00
  x(5) = 2.00000E+00
  x(6) = 2.00000E+00

Thanks,
Federico