Re: RFA: Support separate class for base regs in REG+REG addressing

[Mark Mitchell <mark at codesourcery dot com>]

Daniel Jacobowitz wrote:

> On Mon, Sep 20, 2004 at 05:06:28PM -0400, Daniel Jacobowitz wrote:
>  
>
> > On Mon, Sep 20, 2004 at 04:58:51PM -0400, Ian Lance Taylor wrote:
> >    
> >
> > > Daniel Jacobowitz <drow@false.org> writes:
> > >
> > >      
> > >
> > > > @@ -5234,7 +5226,8 @@ update_auto_inc_notes (rtx insn ATTRIBUT
> > > >    is strictly valid.)
> > > >
> > > >    CONTEXT = 1 means we are considering regs as index regs,
> > > > -   = 0 means we are considering them as base regs.
> > > > +   = 0 means we are considering them as base regs, = 2 means we
> > > > +   are considering them as base regs for REG + REG.
> > > >
> > > >    OPNUM and TYPE specify the purpose of any reloads made.
> > > >
> > > >        
> > > I'm sure I'm missing something, but where does this get called with
> > > CONTEXT == 2?
> > >      
> > Um, no, you're not missing anything.  I appear to have completely
> > fouled up a merge.  It's supposed to happen in the recursive call for
> > REG+REG.  I am going to go investigate this - a backported version of
> > the patch with that omission still fixed the original proprietay
> > testcase.
> >    
>
> The CONTEXT == 2 case only triggers if the address is of the form
> [register ok for index but not base + register not ok for base].
> ARM doesn't have any, so I didn't notice the omission.  Here's the
> revised patch
>  
This patch is OK.

Thanks,

--
Mark Mitchell
CodeSourcery, LLC
(916) 791-8304
mark@codesourcery.com