Hello,

The architecture vector contains the methods:

	REGISTER_CONVERT_TO_VIRTUAL
	REGISTER_CONVERT_TO_RAW
	REGISTER_CONVERTIBLE
	REGISTER_VIRTUAL_TYPE
	REGISTER_RAW_SIZE
	REGISTER_VIRTUAL_SIZE

GDB's core uses them in two different ways:

- to convert between a registers raw and virtual representations

   For instance, the MIPS uses this in an attempt to handle an ABI with 
32 bit registers on a 64 bit ISA. (I'm not saying it works mind).

- to convert between a data values memory (normal) and register 
representation

   For instance, the Alpha might store a 32 bit int in a floating point 
register.

The attached patch adds, documents, and uses, a new set of architecture 
methods:

	CONVERT_REGISTER_P
	REGISTER_TO_VALUE
	VALUE_TO_REGISTER

that handle just the second case.  To keep existing architectures 
working they default to corresponding *CONVERT* macros (which means the 
patch really doesn't change anything).

Reviewing the various architectures:

A number could define just the new methods vis:

	alpha

A number [possibly] don't need to define either (but do):

	ia64-tdep.c (should use floatformat)
	rs6000-tdep.c
	x86-64-tdep.c (?)
	i386-tdep.c (?)

and one (*&*!^$(*&^(!@$) still needs the old and new methods.

	MIPS


Thoughts?  I'll look to commit it in a week.

Andrew