Dear all, this patch fixes a 15/16/17 regression when using the LOC() intrinsic on SELECT TYPE temporaries, where gfc_is_simply_contiguous could return false and gfc_conv_array_parameter would create unnecessarily messy code that lead to an ICE.
I looked at how C_LOC solved this, and adjusted gfc_conv_intrinsic_loc accordingly. Regtested on x86_64-pc-linux-gnu. OK for mainline / 16 / 15? Thanks, Harald
From 2dcdb2463f89930b8cdd2b15f962bd3df5dd83f7 Mon Sep 17 00:00:00 2001 From: Harald Anlauf <[email protected]> Date: Sat, 6 Jun 2026 21:59:49 +0200 Subject: [PATCH] Fortran: fix ICE on LOC intrinsic with polymorphic argument [PR125606] The fix for PR122977 improved the gfc_is_simply_contiguous check for associate variables, but had a side effect when using the LOC() intrinsic on SELECT TYPE temporaries where gfc_is_simply_contiguous could return false. Fall back to simply taking the address of the variable's address. PR fortran/125606 gcc/fortran/ChangeLog: * trans-intrinsic.cc (gfc_conv_intrinsic_loc): When the argument of LOC() is not scalar or known to be simply contiguous, simply take the address of the argument array's first element. gcc/testsuite/ChangeLog: * gfortran.dg/loc_3.f90: New test. --- gcc/fortran/trans-intrinsic.cc | 8 ++++++- gcc/testsuite/gfortran.dg/loc_3.f90 | 36 +++++++++++++++++++++++++++++ 2 files changed, 43 insertions(+), 1 deletion(-) create mode 100644 gcc/testsuite/gfortran.dg/loc_3.f90 diff --git a/gcc/fortran/trans-intrinsic.cc b/gcc/fortran/trans-intrinsic.cc index fdb9ddb52ea..a2a6c6979d9 100644 --- a/gcc/fortran/trans-intrinsic.cc +++ b/gcc/fortran/trans-intrinsic.cc @@ -9829,9 +9829,15 @@ gfc_conv_intrinsic_loc (gfc_se * se, gfc_expr * expr) gfc_add_data_component (arg_expr); gfc_conv_expr_reference (se, arg_expr); } - else + else if (gfc_is_simply_contiguous (arg_expr, false, false)) gfc_conv_array_parameter (se, arg_expr, true, NULL, NULL, NULL); + else + { + gfc_conv_expr_descriptor (se, arg_expr); + se->expr = gfc_conv_descriptor_data_get (se->expr); + } se->expr = convert (gfc_get_int_type (gfc_index_integer_kind), se->expr); + se->expr = gfc_evaluate_now (se->expr, &se->pre); /* Create a temporary variable for loc return value. Without this, we get an error an ICE in gcc/expr.cc(expand_expr_addr_expr_1). */ diff --git a/gcc/testsuite/gfortran.dg/loc_3.f90 b/gcc/testsuite/gfortran.dg/loc_3.f90 new file mode 100644 index 00000000000..efc7d0e56e7 --- /dev/null +++ b/gcc/testsuite/gfortran.dg/loc_3.f90 @@ -0,0 +1,36 @@ +! { dg-do compile } +! { dg-additional-options "-O1" } +! +! PR fortran/125606 +! +! Derived from Fujitsu testsuite +! Reported by David Binderman + +module m + implicit none + type ty + class(*), allocatable :: c1(:) + end type ty + type tt + class(*), allocatable :: node1(:) + end type tt + type,extends(tt)::tte + class(*), allocatable :: c2e(:) + end type tte +contains + subroutine put_addr + class(*), allocatable :: t(:) + integer :: unit + select type (t) + class is (tt) + select type (p=>t(2)%node1) + class is (ty) + write (unit) loc(p(2)%c1) + end select + select type (t) + type is (tte) + write (7) loc(t(2)%c2e) + end select + end select + end subroutine put_addr +end module m -- 2.51.0
