[Bug debug/108573] [13 Regression] '-fcompare-debug' failure (length) at -O2

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108573

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

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

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
So, I believe this has been caused by the r7-4871-g4da41abf84cc6796aa8b
PR78437 change.
      /* On RISC machines we must make sure that changing the mode of SRC_REG
         as destination register will not affect its reaching uses, which may
         read its value in a larger mode because DEF_INSN implicitly sets it
         in word mode.  */
      poly_int64 prec
        = GET_MODE_PRECISION (GET_MODE (SET_DEST (*dest_sub_rtx)));
      if (WORD_REGISTER_OPERATIONS && known_lt (prec, BITS_PER_WORD))
        {
          struct df_link *uses = get_uses (def_insn, src_reg);
          if (!uses)
            return false;

          for (df_link *use = uses; use; use = use->next)
            if (paradoxical_subreg_p (GET_MODE (*DF_REF_LOC (use->ref)),
                                      GET_MODE (SET_DEST (*dest_sub_rtx))))
              return false;
        }
The problem is that without -g, there are just 2 uses:
(insn 10029 10027 10052 2 (set (reg:SI 10 a0 [167])
        (minus:SI (reg:SI 10 a0 [160])
            (reg:SI 9 s1 [165]))) "pr108573.c":12:5 12 {subsi3}
     (nil))
and
(insn 10023 9 10027 2 (set (reg:DI 15 a5 [orig:139 _6 ] [139])
        (sign_extend:DI (reg:SI 10 a0 [160]))) "pr108573.c":11:5 116
{extendsidi2}
     (nil))
but with -g there another one:
(debug_insn 8 7 9 2 (var_location:HI s (minus:HI (subreg:HI (and:DI (reg:DI 10
a0 [160])
                (const_int 1 [0x1])) 0)
        (subreg:HI (ashiftrt:DI (reg/v:DI 9 s1 [orig:151 l ] [151])
                (debug_expr:SI D#1)) 0))) "pr108573.c":12:5 -1
     (nil))
Now, because it uses DImode for a0 rather than SImode like the others, without
-g
we don't return false but with -g we do.

Code generation decisions shouldn't be based on the DEBUG_INSN uses.
So, at least in this decision we should ignore debug uses, we can then do
something with the DEBUG_INSNs, reset them, adjust, whatever (I think at least
the above one clearly doesn't care about the upper bits).