Re: [RFA] W.I.P. AltiVec ppc registers support.

[Daniel Jacobowitz <drow@mvista.com>]

On Thu, Nov 29, 2001 at 09:20:52PM -0500, Andrew Cagney wrote:
> >>>
> >>>
> >>>How does it work?  Badly.
> >>>
> >>>The types look like this right now:
> >>>
> >>>+typedef unsigned char elf_greg_t[4];
> >>>+typedef unsigned char elf_fpreg_t[8];
> >>>+typedef struct {
> >>>+  unsigned char u[16];
> >>>+} __attribute((aligned(16))) __uint128_t;
> >>>+/* Altivec registers */
> >>>+typedef __uint128_t elf_vrreg_t;
> >
> >>
> >>Hmm, Kevin did have good reason to be concerned.  That is wrong. 
> >>Mechanisms similar to shlib et.al. should be used.
> >
> >
> >What is wrong with it?
> >
> >I agree that it is messy, but it is completely correct.  I know of no
> >platform with 8-bit bytes stupid enough to do this any way other than
> >the way I intend.
> 
> Several things come to mind: the __attribute() - there are alignment 
> independant functions for packing/unpacking/copying stuff; the 
> __unint128_t is is a name in system space.
> 
> The shlib code does this by describing the actual layout using a very 
> thin interface (like how remote.c unpacks the g-packet) so host issues 
> aren't a problem (but yes, it would be a suprise to find one that didn't 
> have 8 bit chars).
> 
> The other way of doing this is more like C++ where it creates a ``struct 
> type'' to describe the structure exactly how the target sees it.

The __attribute__ and __uint128_t are both, as I said further down the
message, are both unused :)  The remainder of it is fine, AFAICT.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer