Re: [patch] TARGET_MEM_REF

[Daniel Berlin <dberlin at dberlin dot org>]

On Tue, 2005-03-08 at 22:05 +0100, Zdenek Dvorak wrote:
> Hello,
> 
> > On Sat, Mar 05, 2005 at 10:38:11AM +0100, Zdenek Dvorak wrote:
> > > + @item TARGET_MEM_REF
> > > + These nodes represent memory accesses whose address directly map to
> > > + an addressing mode of the target architecture.  The first argument
> > > + is @code{TMR_SYMBOL} and must be a @code{VAR_DECL} of an object with
> > > + a fixed address.  The second argument is @code{TMR_BASE} and the
> > > + third one is @code{TMR_INDEX}.  The fourth argument is
> > > + @code{TMR_STEP} and must be an @code{INTEGER_CST}.  The fifth
> > > + argument is @code{TMR_OFFSET} and must be an @code{INTEGER_CST}.
> > > + Any of the arguments may be NULL if the appropriate component
> > > + does not appear in the address.  Address of the @code{TARGET_MEM_REF}
> > > + is determined in the following way.
> > > + 
> > > + @smallexample
> > > + &TMR_SYMBOL + TMR_BASE + TMR_INDEX * TMR_STEP + TMR_OFFSET
> > > + @end smallexample
> > 
> > Would you like to explain what these names are for?  In
> > particular, why are SYMBOL and BASE different?
> 
> symbol must be an object with fixed address.  The rtl MEM produced
> for full TARGET_MEM_REF with all components is
> 
> ((plus (plus (mult INDEX STEP)
>         BASE)
>        (const (plus (symbol_ref SYMBOL) OFFSET))))
> 
> > While I might could see how such a construct could be
> > useful, I wonder how it integrates with COMPONENT_REFs
> > and ARRAY_REFs.
> 
> It does not (i.e., COMPONENT_REF will never be argument of
> TARGET_MEM_REF and vice versa).  They just coexist.  In future, I would
> like to convert all memory references to TARGET_MEM_REF before expansion
> to rtl (in order to be able to take advantage of SSA form for addressing
> mode selection), but we are quite far from that.  Or do you ask for
> something different?

Right, i think he was thinking of what i've done with ARRAY_MEM_REF,
which is exactly like ARRAY_REF (minus the extra lower bound argument),
but for pointers.