[PATCH] PR fortran/103610 - ICE in gfc_convert_mpz_to_signed, at fortran/simplify.c:193

[PATCH] PR fortran/103610 - ICE in gfc_convert_mpz_to_signed, at fortran/simplify.c:193

[Harald Anlauf]

[-- Attachment #1: Type: text/plain, Size: 557 bytes --]

Dear all,

when simplifying SHAPE(), we rely on spec_dimen_size returning
non-negative dimensions.  That actually did not happen in all
cases, as the testcase by Gerhard shows.  (The testcase is
rather fallout from the wrong simplification).

The solution is trivial: simply check the dimensions against 0
and fix if needed.

Regtested on x86_64-pc-linux-gnu.  OK for mainline.

As it turns out that we generated "wrong code" by the simplification,
I'd backport at least to 11-branch after some waiting time if no
issues show up elsewhere.

Thanks,
Harald


[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: 0001-Fortran-dimensions-of-an-array-have-to-be-non-negati.patch --]
[-- Type: text/x-patch, Size: 2245 bytes --]

From f8d6aae66b1cb3f75b47d50695ef1be7770abc30 Mon Sep 17 00:00:00 2001
From: Harald Anlauf <anlauf@gmx.de>
Date: Tue, 7 Dec 2021 23:06:41 +0100
Subject: [PATCH] Fortran: dimensions of an array have to be non-negative

gcc/fortran/ChangeLog:

	PR fortran/103610
	* array.c (spec_dimen_size): Fix simplification of SHAPE:
	dimensions must be non-negative.

gcc/testsuite/ChangeLog:

	PR fortran/103610
	* gfortran.dg/shape_11.f90: New test.
---
 gcc/fortran/array.c                    |  6 ++++--
 gcc/testsuite/gfortran.dg/shape_11.f90 | 16 ++++++++++++++++
 2 files changed, 20 insertions(+), 2 deletions(-)
 create mode 100644 gcc/testsuite/gfortran.dg/shape_11.f90

diff --git a/gcc/fortran/array.c b/gcc/fortran/array.c
index e5e22099405..e4891dfef77 100644
--- a/gcc/fortran/array.c
+++ b/gcc/fortran/array.c
@@ -2296,8 +2296,7 @@ gfc_copy_iterator (gfc_iterator *src)
 /********* Subroutines for determining the size of an array *********/

 /* These are needed just to accommodate RESHAPE().  There are no
-   diagnostics here, we just return a negative number if something
-   goes wrong.  */
+   diagnostics here, we just return false if something goes wrong.  */


 /* Get the size of single dimension of an array specification.  The
@@ -2330,6 +2329,9 @@ spec_dimen_size (gfc_array_spec *as, int dimen, mpz_t *result)

   mpz_add_ui (*result, *result, 1);

+  if (mpz_cmp_si (*result, 0) < 0)
+    mpz_set_si (*result, 0);
+
   return true;
 }

diff --git a/gcc/testsuite/gfortran.dg/shape_11.f90 b/gcc/testsuite/gfortran.dg/shape_11.f90
new file mode 100644
index 00000000000..127d221e710
--- /dev/null
+++ b/gcc/testsuite/gfortran.dg/shape_11.f90
@@ -0,0 +1,16 @@
+! { dg-do compile }
+! { dg-options "-fdump-tree-original" }
+! PR fortran/103610 - ICE while simplifying SHAPE
+! Contributed by G.Steinmetz
+
+program p
+  integer, parameter :: a(-1) = 1
+  integer, parameter :: b(1)  = maskl(shape(a))
+  integer, parameter :: c(1)  = shape(a)
+  integer, parameter :: d(1)  = maskr(shape(a))
+  if (b(1) /= 0) stop 1
+  if (c(1) /= 0) stop 2
+  if (d(1) /= 0) stop 3
+end
+
+! { dg-final { scan-tree-dump-not "_gfortran_stop_numeric" "original" } }
--
2.26.2

Re: [PATCH] PR fortran/103610 - ICE in gfc_convert_mpz_to_signed, at fortran/simplify.c:193

[Thomas Koenig]

Hi Harald,

> when simplifying SHAPE(), we rely on spec_dimen_size returning
> non-negative dimensions.  That actually did not happen in all
> cases, as the testcase by Gerhard shows.  (The testcase is
> rather fallout from the wrong simplification).
> 
> The solution is trivial: simply check the dimensions against 0
> and fix if needed.
> 
> Regtested on x86_64-pc-linux-gnu.  OK for mainline.

Yes, and

> As it turns out that we generated "wrong code" by the simplification,
> I'd backport at least to 11-branch after some waiting time if no
> issues show up elsewhere.

... yes again.

Thanks for the patch!

Regards

	Thomas