[Bug tree-optimization/102360] ICE in can_native_interpret_type_p at gcc/fold-const.c:8800

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102360

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
             Status|UNCONFIRMED                 |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |rguenth at gcc dot gnu.org
   Last reconfirmed|                            |2021-09-16

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.

Program received signal SIGSEGV, Segmentation fault.
0x00000000012f2e6b in can_native_interpret_type_p (type=<tree 0x0>)
    at ../../src/trunk/gcc/fold-const.c:8800
8800      switch (TREE_CODE (type))


3049              if (!can_native_interpret_type_p (var_type))
3050                {
3051                  alt_type
3052                    = lang_hooks.types.type_for_mode (TYPE_MODE (var_type),
3053                                                      TYPE_UNSIGNED
(var_type));
3054                  gcc_assert (can_native_interpret_type_p (alt_type));
3055                }

alt_type doesn't need to exist.  TYPE_MODE is OImode here.
Another issue will be that native_interpret_int refuses to interpret
types bigger than 128bit.  So using

              alt_type = build_nonstandard_integer_type 
                           (GET_MODE_BITSIZE (TYPE_MODE (var_type)), 1);

will likely not help.  But if we know we're using an integer type
we can natively interpret ourselves.

Note there's another hole with variable-size and SVE vector modes I think.

I'm testing something.