Re: rtl semantics

[Russell Shaw <rjshaw at netspace dot net dot au>]

Ian Lance Taylor wrote:
> Russell Shaw <rjshaw@netspace.net.au> writes:
>
>
> > I just read all the gcc internals manual.
> > I have trouble with the contextual polymorphic
> > nature of rtl objects.
> >
> > In h8300.md, there is:
> >
> >   (define_attr "cpu" "h8300,h8300h"
> >     (const (symbol_ref "cpu_type")))
> >
> > yet in gccint.info, const should have a machine mode: (const:M EXP)
> >
> > Do rtl keywords have a variable meaning, depending on the context
> > where they're used? Is there other languages similar to rtl i could
> > learn from? (scheme?)
>
>
> The define_attr construct is not quite RTL.  It does not actually
> describe a machine instruction, and thus does not require a mode.  It
> is described here:
>     http://gcc.gnu.org/onlinedocs/gccint/Defining-Attributes.html#Defining-Attributes
>
> The const attribute in define_attr is described here:
>     http://gcc.gnu.org/onlinedocs/gccint/Constant-Attributes.html#Constant-Attributes

Are the templates in <machine>.md real RTL, or only templates used to generate RTL?
Is "RTL" a textual language, or just some C struct?