[Bug target/105980] [11/12/13 Regression] ICE in final_scan_insn_1, at final.cc:2811

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105980

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
If we follow what other backends do in *_output_mi_thunk, it would be
--- gcc/config/i386/i386.cc.jj  2022-07-26 10:32:23.373275765 +0200
+++ gcc/config/i386/i386.cc     2022-07-29 15:25:26.023555541 +0200
@@ -21396,6 +21396,7 @@ x86_output_mi_thunk (FILE *file, tree th
        tmp_regno = CX_REG;
     }

+  reload_completed = 1;
   emit_note (NOTE_INSN_PROLOGUE_END);

   /* CET is enabled, insert EB instruction.  */
@@ -21561,6 +21562,9 @@ x86_output_mi_thunk (FILE *file, tree th
   final (insn, file, 1);
   final_end_function ();
   assemble_end_function (thunk_fndecl, fnname);
+
+  /* Stop pretending to be a post-reload pass.  */
+  reload_completed = 0;
 }

 static void
Unfortunately, that doesn't help much, the next problem is that it needs
pic_offset_table_rtx as a hard register, so we probably before
          tmp = gen_rtx_REG (word_mode, tmp_regno);
          if (GET_MODE (fnaddr) != word_mode)
            fnaddr = gen_rtx_ZERO_EXTEND (word_mode, fnaddr);
          emit_move_insn (tmp, fnaddr);
          fnaddr = tmp;
in the above mentioned snippet need to do something like:
          if (flag_pic && !TARGET_64BIT && !TARGET_MACHO && SYMBOLIC_CONST
(fnaddr))
            {
              pic_offset_table_rtx = gen_reg_RTX (Pmode, some_suitable_reg);
              ix86_init_pic_reg ();
            }
or something similar.  But unsure about the exact conditions,
insert_insn_on_edge will likely not work either etc.
So it likely needs emit_insn (gen_set_got (reg)) etc. directly.
            }