Re: SH: use register names instead of numbers

[Joern Rennecke <amylaar at redhat dot com>]

> This looks good to me.  I'm wondering why you parsed the expression by
> hand instead of creating a new DEFINE_CONSTANTS RTL code, as is done
> for all other define_* expressions.

This is not supposed to be returned from read_rtx, but rather immediately
affect all following rtx expressions that are read.
Moreover, proper checking for duplicate definitions is only possible if
you switch off constant expansion during constant definition.

> As a separate patch, I think it'd be a good idea to split the md-file
> reader out of rtl.c.  It could go either in gensupport.c or a new
> read-rtl.c.  The compiler proper doesn't use it, so it's just wasting
> space there.  Not much space at the moment, but when we do a complete
> macro preprocessor with support for include files and conditional
> blocks, it'll get bigger.

I don't see a need for that.  Well, maybe you can provide one, along with
a patch to fill it ;-)

> ... reduceS legibility ...

Thanks.

> > + To overcome these problems, the @code{define_constants} built-in
> > + is provided.  It contains a vector of name-value pairs.
> 
> This paragraph isn't very detailed and doesn't match the style of the
> rest of the document.  I suggest instead
> 
>   To overcome this problem, you may use the @code{define_constants}
>   expression.  It contains a vector of name-value pairs.  From that
>   point on, wherever any of the names appears in the MD file, it is as
>   if the corresponding value had been written instead.  You may use
>   @code{define_constants} multiple times; each appearance adds more
>   constants to the table.  It is an error to redefine a constant with
>   a different value.

That's fine with me.
I'm waiting for someone with approval powers to voice a verdict...