[clang] [llvm] [RISCV] Assembler support for XRivosVizip (PR #127694)

[Philip Reames via cfe-commits]

preames wrote:

> > > I am kind of confused now. So the situation here is that RVIOS has already implemented these vendor extensions in cores and RVIOS is also trying to make these extensions official RVI standards, right?
> > 
> > 
> > You're confusing two things. We (Rivos) have defined a set of custom vendor extensions - this patch is the first in that sub-series. We are _also and separately_ proposing a subset of instructions for consideration at RVI for eventual standardization. As an example of how these are different, see XRivosVisni in the linked repo. That's a vendor extension which relates to gaps identified in RVV, but for which we're not currently proposing an extension at RVI.
> 
> Thanks, I'd like to know if these vendor extensions have been implemented in some type-out-ed cores. I have this question because I saw that these vendor extensions are of version 0.1. If they are going to be changeable (and experimental), then why are we supporting XRivosVizip extension in a hurry since XRivosVizip and Zvzip are overlapped.

In general, vendor extensions remain in draft (i.e. experimental), until hardware is announced.  I can not comment on any hardware plans, or timelines.  

As the lead of the Zvzip work, I can say that I believe having a vendor extension with in tree support will make it easier for others to evaluate the zvzip proposal.  In particular, the current assignment of the zvzip opcodes in custom_2 is about to be removed as that's really not appropriate for an RVI proposal.  Given that, landing anything for zvzip is pretty much a non-starter.  However, this is only vaguely related to this patch.  

Also, I disagree with your characterization of anything being done here as "in a hurry".  All normal review and discussion timelines have been, and will continue to be, followed.  


https://github.com/llvm/llvm-project/pull/127694