Re: RFA: Use .opt files for the m68k port

[Gunther Nikl <gni at gecko dot de>]

On Fri, Apr 15, 2005 at 02:34:31PM +0100, Richard Sandiford wrote:
> Gunther Nikl <gni@gecko.de> writes:
> >   But I have question to the new option handling. In this message
> >
> >       http://gcc.gnu.org/ml/gcc-patches/2004-06/msg02053.html
> >
> >   you can find a proposal for a target option -mregparm for m68k. Its an
> >   option *and* a switch. Is that still possible? Having it an option and
> >   switch had problems with the option before the no- switch. since the
> >   option was handled after the switch.
> 
> Yes, this should work.  You should also be able to avoid the problem
> you mention.

  Yes, it looks like the problem is gone.

> First initialise m68k_regparm like this:
> 
>     int m68k_regparm = M68K_DEFAULT_REGPARM;

  Seems I need some code in override_options to clear this variable because
  later code checks for zero to decide whether arguments should be passed
  through registers. If I want to keep it that way I added this code to
  override_options:

    if (!TARGET_REGPARM)
      m68k_regparm = 0;

> then add two target option entries like this:
> 
>     mregparm
>     Target Mask(REGPARM)
>     Help string...

  That was the easy one.

>     mregparm=
>     Target RejectNegative Joined UInteger
>     Help string...

  Thats the one I would have had problems with. Thank you!

> and then add something like this to handle_option:
> 
>     case OPT_mregparm_:
>       target_flags |= MASK_REGPARM;
>       if (value < 1 || value > M68K_MAX_REGPARM)
>         error ("-mregparm=%d is not between 1 and %d",
>                value, M68K_MAX_REGPARM);
>       else
>         m68k_regparm = value;
>       return true;
> 
> (All untested of course. ;)

  Its almost correct (see above). Many, many thanks for your help!

  I just had another look at the m68k change and I noticed that
  TARGET_DEFAULT_TARGET_FLAGS contains MASK_STRICT_ALIGNMENT. I must
  admit that I never noticed the strange old defintions for it. Do
  I read that correctly that this is the old default value?

  BTW, there seems to be a bug in handle_option/OPT_mshared_library_id_:
  the error message passes arg but should pass value.

  And I have further question: Is it possible to have an option (switch?)
  that clears several other flags at once? And how is a subtarget supposed
  to handle own options? With something like a SUBTARGET_OPTIONS call in
  the target option handler?

  Gunther