Dear all, the simplification of the TRANSFER intrinsic produces a redundant representation of the result, one in expr->value and another in expr->representation.string. This is done to ensure a safe "round-trip" for nested TRANSFER.
In a subsequent use of this result we either need to make sure that both parts stay consistent, or drop the latter part. The simplifications of IBCLR and IBSET do a gfc_copy_expr of their argument x and modify only the former part. Depending on context, either value.integer or representation.string (if non-NULL) are used later, leading to surprising results. A conservative approach to fix this PR is to simply drop the unneeded representation.string in the simplification of the intrinsics IBCLR and IBSET, see attached patch. (A quick glance did not turn up other intrinsics affected the same way.) Regtested on x86_64-pc-linux-gnu. OK for mainline? Thanks, Harald
From a540a806cf02d739a408f129738252e73f03e60c Mon Sep 17 00:00:00 2001 From: Harald Anlauf <anl...@gmx.de> Date: Sat, 20 Aug 2022 20:36:28 +0200 Subject: [PATCH] Fortran: fix simplification of intrinsics IBCLR and IBSET [PR106557] gcc/fortran/ChangeLog: PR fortran/106557 * simplify.cc (gfc_simplify_ibclr): Ensure consistent results of the simplification by dropping a redundant memory representation of argument x. (gfc_simplify_ibset): Likewise. gcc/testsuite/ChangeLog: PR fortran/106557 * gfortran.dg/pr106557.f90: New test. --- gcc/fortran/simplify.cc | 14 ++++++++++++++ gcc/testsuite/gfortran.dg/pr106557.f90 | 19 +++++++++++++++++++ 2 files changed, 33 insertions(+) create mode 100644 gcc/testsuite/gfortran.dg/pr106557.f90 diff --git a/gcc/fortran/simplify.cc b/gcc/fortran/simplify.cc index fb725994653..f992c31e5d7 100644 --- a/gcc/fortran/simplify.cc +++ b/gcc/fortran/simplify.cc @@ -3380,6 +3380,13 @@ gfc_simplify_ibclr (gfc_expr *x, gfc_expr *y) k = gfc_validate_kind (x->ts.type, x->ts.kind, false); result = gfc_copy_expr (x); + /* Drop any separate memory representation of x to avoid potential + inconsistencies in result. */ + if (result->representation.string) + { + free (result->representation.string); + result->representation.string = NULL; + } convert_mpz_to_unsigned (result->value.integer, gfc_integer_kinds[k].bit_size); @@ -3471,6 +3478,13 @@ gfc_simplify_ibset (gfc_expr *x, gfc_expr *y) k = gfc_validate_kind (x->ts.type, x->ts.kind, false); result = gfc_copy_expr (x); + /* Drop any separate memory representation of x to avoid potential + inconsistencies in result. */ + if (result->representation.string) + { + free (result->representation.string); + result->representation.string = NULL; + } convert_mpz_to_unsigned (result->value.integer, gfc_integer_kinds[k].bit_size); diff --git a/gcc/testsuite/gfortran.dg/pr106557.f90 b/gcc/testsuite/gfortran.dg/pr106557.f90 new file mode 100644 index 00000000000..d073f3e7186 --- /dev/null +++ b/gcc/testsuite/gfortran.dg/pr106557.f90 @@ -0,0 +1,19 @@ +! { dg-do run } +! { dg-additional-options "-fdump-tree-original" } +! PR fortran/106557 - nesting intrinsics ibset and transfer gives wrong result + +program p + implicit none + character(1) :: s + + write(s,'(i1)') ibset (transfer (0, 0), 0) + if (s /= '1') stop 1 + write(s,'(i1)') ibclr (transfer (1, 0), 0) + if (s /= '0') stop 2 + + ! These shall be fully resolved at compile time: + if (transfer (ibset (transfer (0, 0), 0), 0) /= 1) stop 3 + if (transfer (ibclr (transfer (1, 0), 0), 0) /= 0) stop 4 +end + +! { dg-final { scan-tree-dump-times "_gfortran_stop_numeric" 2 "original" } } -- 2.35.3