https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127846
Bug ID: 127846
Summary: MINLOC/MAXLOC with MASK and BACK=.true. ignore the
mask
Product: gcc
Version: 16.2.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: fortran
Assignee: unassigned at gcc dot gnu.org
Reporter: federico.perini at gmail dot com
Target Milestone: ---
Possibly related to #127845 that I also just filed.
minloc(y, mask=m, back=.true.) on a rank-2 array returns a position also where
the mask is .false. In the test case, row 2 of the mask is all .false., yet
minloc returns (2,4) and maxloc returns (2,3). The calls go to
_gfortran_mminloc0_4_r8 and _gfortran_mmaxloc0_4_r8, they do not seem to
advance the mask properly.
program minloc_mask_back
implicit none
real(8) :: y(3,4)
logical :: m(3,4)
integer :: i, j, l
l = 0
do j = 1, 4; do i = 1, 3
y(i,j) = real(mod(l*7,5)+1, 8); m(i,j) = mod(l,3) /= 1; l = l + 1
end do; end do
print '(a)', 'y m'
do i = 1, 3
print '(4f4.1,6x,4l2)', y(i,:), m(i,:)
end do
print '(a,2i3,a)', 'minloc(y, mask=m, back=.true.) =', minloc(y, mask=m,
back=.true.), &
' expected 3 2 (m(2,4) is F)'
print '(a,2i3,a)', 'maxloc(y, mask=m, back=.true.) =', maxloc(y, mask=m,
back=.true.), &
' expected 3 1 (m(2,3) is F)'
end program minloc_mask_back
Check on Compiler Explorer: https://godbolt.org/z/Mcf4n4bTT
The defect: In libgfortran/m4/minloc0.m4, the masked routine's back branch,
lines 124-135 on trunk, advances base but not mbase: changing to `mbase +=
mstride[0];` might fix this (untested).
Thank you!
Federico
Federico
