GAS: Handling option parsing for ARM co-processor extensions

[Richard Earnshaw <rearnsha at arm dot com>]

I've been thinking a bit about how we should handle the option parsing for 
various ARM co-processor extensions.  Currently we have the highly bogus 
situation where -marm9e is used to mean "add the Cirrus Maverick 
instructions": it's bogus because arm9e is a real cpu variant from ARM.

One way I can think of doing this is to allow syntax along the following 
lines

	-marmX+ext1+...

Then the Maverick extensions could be added by writing

	-marm9+maverick

If maverick were then to be implemented on a future arm chip, then the 
extension would also be available there, eg:

	-marm10+maverick

I think it should be fairly easy to implement this approach, but before I 
do I'd like to solicit any comments.

The same trick can also be applied to architecture variants, so the above 
would become

	-marmv4t+maverick

Incidentally, using this approach, a valid description of XScale would 
become

	-marmv5e+xscale

since technically, the extra xscale instructions are all in the 
co-processor extension space.  Though I'm not suggesting -mxscale should 
be removed in this case...

R.