Move loglinks into separate datastructure

[Jan Hubicka]

> On Jan 23, 2004, Jan Hubicka <jh@suse.cz> wrote:
> 
> > The patch also needs the consumers of LOG_LINKS in regmove and sh backend
> > removed as they are dead now.
> 
> Would you please remove the uses there as well?  I see that at least
> SH is going to take a compile-time performance hit when doing
> relaxation because of this change, since every function call is
> indirect, and we'll have to resort to backward-scanning the
> instruction stream looking for the insn that sets the register,
> whereas previously we'd most often go straight to it by means of
> LOG_LINKS.  I guess we can live with that...

I've sent separate patches for that as I expected it would be easier to
discuss this separately.  I think SH can be quite easilly turned to use
df.c analyzer if the perofmrance problem is important, but I don't have
any numbers.

Honaz
> 
> -- 
> Alexandre Oliva   Enjoy Guarana', see http://www.ic.unicamp.br/~oliva/
> Happy GNU Year!                     oliva@{lsd.ic.unicamp.br, gnu.org}
> Red Hat GCC Developer                 aoliva@{redhat.com, gcc.gnu.org}
> Free Software Evangelist                Professional serial bug killer