Re: New branch created for "available features" support

[Daniel Jacobowitz <drow at false dot org>]

On Fri, Mar 03, 2006 at 02:36:22PM -0800, Jim Blandy wrote:
> So a "feature" would be something like "SSE", or "MMX" with the
> additional SSE registers, and then a "feature set" would be something
> like "Pentium MMX" or "Pentium 3", referring to the appropriate
> individual feature sets?

s/feature sets?/features?/ at the end.  Otherwise, yes.  At the moment
at least (before I get back to the questions of nesting) a feature set
represents the entire target.

> I understand why you want to supply base register numbers in feature
> sets, rather than features; it makes it possible to share features. 
> But if we're going to have other sorts of features as of yet
> unimagined, they may or may not expect a base register number as a
> parameter.  That is, you've tied something specific to register banks
> to your <feature> entities.
> 
> I'd say, rename <feature> to <register-set>.  Introduce new,
> appropriately named kinds of entities for new kinds of features.  That
> way, you can look at the tag and know what sorts of parameters it
> expects.

My mental model was that a single feature might contain registers, or
not, and something else, or not.  That's why base-regnum is optional.
So a feature might contain, say, registers and MMIO ports and have a
base for both; but if it contains no registers, base-regnum might be
omitted.

However, you may be right that I've got the separation at the wrong
level - I'll have to think about that one for a little.  If there's
any advantage to the separation it would be easy to change.

-- 
Daniel Jacobowitz
CodeSourcery