[Bug tree-optimization/80109] ICE in get_range_info, at tree-ssanames.c:375

["msebor at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80109

Martin Sebor <msebor at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|ice-on-valid-code           |ice-on-invalid-code
                 CC|                            |aldyh at gcc dot gnu.org

--- Comment #4 from Martin Sebor <msebor at gcc dot gnu.org> ---
It also ICEs with a native compiler.  The test case is invalid: it calls
alloca() with a pointer argument.  A smaller test case is:

$ cat z.c && gcc -S -Wall -Wextra -O2 -Walloca-larger-than=126812070 z.c
void g (int *p)
{
  extern void f (void*);

  void *q = __builtin_alloca (p);
  f (q);
}

z.c: In function ‘g’:
z.c:5:31: warning: passing argument 1 of ‘__builtin_alloca’ makes integer from
pointer without a cast [-Wint-conversion]
   void *q = __builtin_alloca (p);
                               ^
z.c:5:31: note: expected ‘long unsigned int’ but argument is of type ‘int *’
z.c:1:6: internal compiler error: in get_range_info, at tree-ssanames.c:375
 void g (int *p)
      ^

0x11ac2f9 get_range_info(tree_node const*, generic_wide_int<wide_int_storage>*,
generic_wide_int<wide_int_storage>*)
        /src/gcc/svn/gcc/tree-ssanames.c:375
0x190908e alloca_call_type
        /src/gcc/svn/gcc/gimple-ssa-warn-alloca.c:334
0x1909637 pass_walloca::execute(function*)
        /src/gcc/svn/gcc/gimple-ssa-warn-alloca.c:466
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.


The bug seems to be in the following block in the alloca_call_type function
that passes the pointer (len_casted) to get_range_info:

              gimple *def = SSA_NAME_DEF_STMT (len);
              if (gimple_assign_cast_p (def)
                  && TYPE_UNSIGNED (TREE_TYPE (gimple_assign_rhs1 (def))))
                {
                  len_casted = gimple_assign_rhs1 (def);
                  range_type = get_range_info (len_casted, &min, &max);
                }

The following otherwise untested patch avoids the ICE.  I CC'd Aldy for hits
thoughts.

diff --git a/gcc/gimple-ssa-warn-alloca.c b/gcc/gimple-ssa-warn-alloca.c
index b940efa..50b35f6 100644
--- a/gcc/gimple-ssa-warn-alloca.c
+++ b/gcc/gimple-ssa-warn-alloca.c
@@ -327,11 +327,20 @@ alloca_call_type (gimple *stmt, bool is_vla, tree
*invalid_casted_type)
              // away with better range information.  But it gets
              // most of the cases.
              gimple *def = SSA_NAME_DEF_STMT (len);
-             if (gimple_assign_cast_p (def)
-                 && TYPE_UNSIGNED (TREE_TYPE (gimple_assign_rhs1 (def))))
+             if (gimple_assign_cast_p (def))
                {
-                 len_casted = gimple_assign_rhs1 (def);
-                 range_type = get_range_info (len_casted, &min, &max);
+                 tree rhs = gimple_assign_rhs1 (def);
+                 tree rhstype = TREE_TYPE (rhs);
+
+                 // Bail if the argument type is not valid.
+                 if (!INTEGRAL_TYPE_P (rhstype))
+                   return alloca_type_and_limit (ALLOCA_OK);
+
+                 if (TYPE_UNSIGNED (rhstype))
+                   {
+                     len_casted = rhs;
+                     range_type = get_range_info (len_casted, &min, &max);
+                   }
                }
              // An unknown range or a range of the entire domain is
              // really no range at all.