This is the mail archive of the
mailing list for the GCC project.
Re: Patch for automaton based pipeline hazard recognizer (part #1).
Bernd Schmidt wrote:
> On Tue, 20 Feb 2001, Vladimir Makarov wrote:
> > > > + #if AUTOMATON_PIPELINE_INTERFACE
> > > > + if (USE_AUTOMATON_PIPELINE_INTERFACE)
> > >
> > > I believe the #if is unnecessary - if AUTOMATON_PIPELINE_INTERFACE is 0,
> > > then USE_AUTOMATON_PIPELINE_INTERFACE will be 0 as well, right?
> > You can have two description in md file (old and new). If there is a
> > new description, AUTOMATON_PIPELINE_INTERFACE will equal 1.
> > USE_AUTOMATON_PIPELINE_INTERFACE says about should we use the new
> > pipeline description or not.
> > So, You can describe one processor subtarget with the old model, and
> > another one with automaton based description and define
> > USE_AUTOMATON_PIPELINE_INTERFACE equal to zero for the first subtarget
> > and 1 for the second one.
> Yes, and that can still be achieved with just an if, not an #ifdef. Or is
> the problem that some functions or fields don't exist if only one of the
> two descriptions exist?
Yes. E.g. there are many functions for old pipeline description
interface which are absolutely unnecessary for new pipeline interface.
So if we use only new pipeline description the functions should not be
> > The old pipeline interface is also undocumented.
> Not entirely. There are bits of information in genattrtab.c (e.g. before
> and inside the function expand_units). Not the best place, but...
> > It is nice to see the automaton time generation (it is possibly to
> > write description for which automaton will be generated 10 or more
> > minutes).
> Oh. Possibly we'll need some kind of progress indicator so that people
> don't think their build has crashed.
Ok, I'll try write something. Actually, long time means that the
automaton will be big, so the user should decrease its size in any case
(by distributions of functional units to separate automata).