Re: Unit at a time 2/3 - backward inlining, local functions

[Jan Hubicka <jh at suse dot cz>]

> On Thu, Feb 13, 2003 at 12:13:42AM +0100, Jan Hubicka wrote:
> > There are two bits I am nervous about - first is fixup_calls thing.  Is that
> > safe?  It appears to work everywhere :)
> 
> I havn't figured out why you need it.  Oh for the attribute
> frobbing?  I think it's a bad idea.
> 
> > Other is the scheme of target hook for local function - it is most
> > generic and target hook may use some analysis to decide what changes
> > to do on the function.
> 
> I don't think I like this.
> 
> > On the other hand, it perhaps may be easier to add simple local flag
> > to generic trees
> 
> I think it's sufficient to have
> 
> 	!TREE_PUBLIC && !TREE_ADDRESSABLE
> 
> mean what you want.  This can then be checked from the backends
> when placing arguments, and when generating prologues, etc.
OK,
I've added cgraph_local_function_p as "shortcut" for this, but now I
need to hook this into i386 backend into argument passing.
init_cumulative_args is taking only function type as an operand (not
declaration), so it can't test it.
What would be best interface for this?  Adding extra argument into
init_cumulative_args with declaration?

Honza
> 
> 
> r~