This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: Printing decimal128 types out of registers
> From: Luis Machado <luisgpm@linux.vnet.ibm.com>
> Reply-To: luisgpm@linux.vnet.ibm.com
>
> > But they're pseudo-registers; no target "supports" them because
> > they're not real registers. Are you trying to say that some targets
> > have hardware support for decimal floating point, and you want to make
> > sure that these pseudo registers only show up on targets that have
> > such hardware support? If so I'd appreciate it if you could change
> > the comments to reflect that.
>
> The idea is to have a "decimal128" feature meaning that we're supposed
> to show these pseudo-registers for ppc32 and ppc64, similar to the way
> the EV registers work on the e500, except that we have (in the e500) 32
> of those being real registers and the other 32 being pseudo-registers.
> All those 64 registers sum up to create a single set of 32 64-bit
> registers (power-spe feature).
But that's a real hardware feature. It also controls the
pseudo-registers because it doesn't make sense to show those without
the real registers.
> > There are no software implementations that use the floating-point
> > registers in the same way as the hardware implementation does?
>
> Yes, there is. This would work the same way with software
> implementations, that is, we would have the same parameter passing
> scheme using pairs of FP registers.
In that case, I think you should just add the decimal floating-point
pseudo-registers unconditionally.