Canonical tag clearing pseudo

[arichardson]

ISAv9 had a dedicated CClearTag instruction that writes back the input register without a tag. This is not hugely common. One place I've seen it is in the CheriBSD kernel does for a tag-stripping memcpy since it predates tag-clearing semantics for loads without permit_load_cap.

With #82 we could treat cbuildcap cd, c0, cs2 as the canonical ccleartag cd, cs1 (previously I believe the best we could do was the two-instruction cgethigh cd, cs1; csethigh cd, cs1, cd;.

1. Should we add a note to buildcap that it might be a good idea for a high-performance CHERI core to special-case cbuildcap cd, c0, cs2 in the decoder to be a tag-clearing move?

2. Should there be an assembly pseudo for ccleartag that maps to cbuildcap cd, c0, cs2?

[andresag01]

I like the idea of making cbuildcap cd, c0, cs2 an assembly pseudo for ccleartag. The function is already there and this encoding is probably never going to get used as a hint or something like that.

I am less keen about the informative note because this ccleartag is hardly ever used, so I doubt optimising that special cbuildcap case as a move will make any difference for performance.

[andresag01]

@francislaus pointed out that cbuildcap cd, c0, cs2 does not behave like ccleartag when otypes are implemented. Given that otypes might be an extension in future, it would be best to add a note to the spec reserving cbuildcap cd, c0, cs2 instead of declaring it a ccleartag pseudo.

[tariqkurd-repo]

I've marked this as a bug as this is a spec change.

[andresag01]

Closed via #103