The following source is compiled into wrong code: interface elemental integer function foo(x) integer, intent(in), value :: x end function end interface print *, foo(42) print *, foo([0,1]) end For all versions of gfortran starting from 4.3 (and up to current trunk), the second call to FOO is miscompiled. Looking at the generated code (-fdump-tree-original), the first call is, as expected, by value: D.1882 = foo (42); _gfortran_transfer_integer_write (&dt_parm.0, &D.1882, 4); while the second call is by-reference: D.1887 = A.2[S.3]; D.1888 = foo (&D.1887); _gfortran_transfer_integer_write (&dt_parm.1, &D.1888, 4); This can be confirmed by adding an actual FOO function such as: elemental integer function foo(x) integer, intent(in), value :: x foo = x - 1 end function Running the complete program then yields: 41 1370438451 1370438451
Created attachment 31176 [details] Fixes the PR This is regtesting right now - I'll prepare the testcase and submit, as soon as it is done. Best regards Paul
Author: pault Date: Sun Feb 9 19:45:06 2014 New Revision: 207645 URL: http://gcc.gnu.org/viewcvs?rev=207645&root=gcc&view=rev Log: 2014-02-09 Paul Thomas <pault@gcc.gnu.org> PR fortran/59026 * trans-expr.c (gfc_conv_procedure_call): Pass the value of the actual argument to a formal argument with the value attribute in an elemental procedure. 2014-02-09 Paul Thomas <pault@gcc.gnu.org> PR fortran/59026 * gfortran.dg/elemental_by_value_1.f90 : New test Added: trunk/gcc/testsuite/gfortran.dg/elemental_by_value_1.f90 Modified: trunk/gcc/fortran/ChangeLog trunk/gcc/fortran/trans-expr.c trunk/gcc/testsuite/ChangeLog
Was fixed on 4.9, not a regression: closing.