[Bug fortran/71615] New: wrong float point result with {-Ofast, -march=native, and valgrind}

["wangmianzhi1 at linuxmail dot org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71615

            Bug ID: 71615
           Summary: wrong float point result with {-Ofast, -march=native,
                    and valgrind}
           Product: gcc
           Version: 5.3.1
            Status: UNCONFIRMED
          Severity: major
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: wangmianzhi1 at linuxmail dot org
  Target Milestone: ---

Created attachment 38742
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=38742&action=edit
reduced test

Problem:
float point calculation gives wrong result

Hardware:
AMD A4-5000 CPU

Software:
Ubuntu 16.04 x86_64
gcc and valgrind from apt

Steps to Repeat the Bug:
1. compile the attached REDUCED test code with
$ gfortran -Wall -Ofast -march=native test.f90 -o test 
2. execute the program via valgrind
$ valgriand ./test

Result:
The calculation gives
   0.0000000000000000       -2.6901205975860183       0.95110109754591643
While the correct result should be
   0.0000000000000000      -0.94280905929685954       0.33333328322831940

Note:
Would give the correct result WITHOUT ANY of {-Ofast, -march=native, or
valgrind}.