[Bug target/82221] internal compiler error: in print_reg, at config/i386/i386.c:17656

["hjl.tools at gmail dot com" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82221

--- Comment #11 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to Josh Poimboeuf from comment #9)
> I might be misunderstanding, but I don't think we need DRAP for our use
> case.  We just need assurance that the frame pointer (RBP) has been set up
> before the inline asm is inserted.
> 

The problem is

#define stack_realign_fp (crtl->stack_realign_needed && !crtl->need_drap)

Since function returns 64-bit integer and incoming stack is aligned to
32 bits, initially GCC assume stack realignment is needed.  If need_drap
is false, stack_realign_fp is true.  In this case, we won't eliminate
frame pointer with stack pointer

static bool
ix86_can_eliminate (const int from, const int to)
{
  if (stack_realign_fp)
    return ((from == ARG_POINTER_REGNUM
             && to == HARD_FRAME_POINTER_REGNUM)
            || (from == FRAME_POINTER_REGNUM
                && to == STACK_POINTER_REGNUM));
  else  
    return to == STACK_POINTER_REGNUM ? !frame_pointer_needed : true; 
}

when sp_is_unchanging is false.  Setting need_drap when sp_is_unchanging is
false allows it to work.  BTW, drap is really used since stack_realign_needed
is changed to false by ix86_finalize_stack_frame_flags.