[PATCH] MIR Serialization: Serialize the machine function's liveins.

[Duncan P. N. Exon Smith]

> On 2015-Jul-24, at 11:50, Alex L <arphaman at gmail.com> wrote:
> 
> I forgot to attach a patch, here it is.
> 
> 2015-07-24 11:49 GMT-07:00 Alex L <arphaman at gmail.com>:
> Hi,
> 
> This patch serializes the machine function's live ins. They are serialized using a sequence of YAML inline mappings, e.g:
> 
> liveins:
>   - { reg: '%edi', virtual-reg: '%0' }
> 
> There's a slight problem with the printing right now, caused by the fact that the virtual registers in function's liveins aren't cleared when we clear the virtual register definitions. Thus, after post RA passes, MIR printer will print out MIR that won't be parsable (it will print out the virtual register reference but not the definition). I intend to send a follow up patch that modifies the clear method in MachineRegisterInfo so that virtual register references in liveins are cleared as well, and it should come after this one as I would like to test it using the live in printing functionality that's introduced in this patch.

This is a little weird, but as long as the follow-up goes in
quickly it makes sense.

> 
> Thanks,
> Alex
> 
> <0001-MIR-Serialization-Serialize-machine-the-function-s-l.patch>

LGTM.