https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118580
Bug ID: 118580
Summary: Incorrect complex (sp) - real (dp) operation within
maxval
Product: gcc
Version: 15.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: fortran
Assignee: unassigned at gcc dot gnu.org
Reporter: federico.perini at gmail dot com
Target Milestone: ---
Sample program:
```
program complex_eye
implicit none
integer, parameter :: dp = kind(0.d0)
integer, parameter :: sp = kind(0.0)
integer, parameter :: k = 2
integer :: i,j
complex(sp), dimension(k,k) :: A, B
real(dp) :: eye(k, k), mx, mxB
do j = 1,k; do i=1, k
eye(i,j) = merge(1,0,i==j)
A(i,j) = merge(1,0,i==j)
enddo
B = A - eye
mx = maxval(abs(A-eye))
mxB = maxval(abs(B))
if (mx>1.0e-6) then
write(*, *) "maxval(abs(A - eye)) = ", mx ! Should be 0, it is 1.0
write(*, *) "maxval(abs(B)) = ", mxB
stop 1
else
stop 0
endif
end program complex_eye
```
when `maxval` is run on a temporary expression, it returns a wrong value (often
1.0, but sometimes junk values). when the operation is saved in variable B
first, it returns the correct value.
In Compiler explorer, see https://godbolt.org/z/E6dWcEc9s :
- fails on all versions of gfortran from 4.9.4 to 15.0
- fails with any optimization levels, also -O0
We found this issue after an update to the Fortran Standard Library, but the
issue seems not related to the library itself.
Thank you,
Federico
