On 11/17/26 11:57 AM, Thomas Koenig wrote:
This patch also uses memset for zeroing contiguous array slices. A
regression led to the discovery that, for
real, dimension (1,2) :: b
b(1,:) = 0.
end
gfc_is_simply_contiguous returned false, whereas gfc_full_array_ref_p
returned true. This is also fixed.
Regression-tested. OK for trunk?
Looks OK Thomas.
Thanks,
Jerry
Best regards
Thomas
gcc/fortran/ChangeLog:
PR fortran/40598
PR fortran/128776
* expr.cc (gfc_is_simply_contiguous): An extent of one can
also be contiguous.
* trans-expr.cc (gfc_trans_zero_assign): Modify to also deal with
array slices.
(copyable_slice_p): Reallocatable expressions can be fine; check
is in gfc_trans_assignment. Also return false for expressions
which have a corank.
(gfc_trans_assignment): Use copyable_slice_p instead of
copyable_array_p for gfc_trans_zero_assign.
gcc/testsuite/ChangeLog:
PR fortran/40598
PR fortran/128776
* gfortran.dg/vect/fast-math-rnflow-trs2a2.f90: Adjust vect count;
zeroing is now done by memset.
* gfortran.dg/array_memset_4.f90: New test.