Re: RFA: New builtin SIMD types

[Jim Blandy <jimb at cygnus dot com>]

> > Not sure what you mean.  Does REGISTER_NAME_ALIAS_HOOK do this?
> 
> Que? It isn't in the documenation ;-)

Oh!  Well, then it mustn't really exist.  I should lay off those
Italian sausages.

> >From memory, without looking at the code, no.  This just maps a register
> name onto one of the existing registers.

Yeah, I think that's right.

> Long ago GDB started out with the assumption that there were only raw
> register values.  From there it evolved into an assumption that there
> was a 1:1 mapping between virtual (user) and raw (target buffer)
> registers. You end up going through hoops trying to convince gdb that a
> raw register is made up of several virtual registers and vice versa.
> 
> I'm really wondering if there should be a clear separation between
> virtual registers and the raw register buffer. The user sees virtual
> registers, the arch code knows how to map between virtual registers and
> the raw register buffer. The target knows how to stuff bytes into the
> raw register buffer.

Ah hah.  If debug info could also refer to virtual registers, which
have some possibly complex mapping onto raw registers, then we are on
to something.  I need some hair of this sort to handle MMX registers
correctly.

We should ponder this further.