[llvm] [RISCV][CodeGen] Add initial vzip codegen support (PR #194548)

[Luke Lau via llvm-commits]

lukel97 wrote:

> > > > @lukel97 do we need to do anything to disable this instruction the VLOptimizer? This instruction reads VL elements from the sources but writes 2*VL results.
> 
> > > 
> 
> > > 
> 
> > > Nevermind, I think the default value of ElementsDependOn being EltDepsVLMask protects this.
> 
> > 
> 
> > Yeah it should be disabled by default, I switched over the TSFlags to be opt-in a while ago
> 
> 
> 
> I know that prevents the instruction from having its VL reduced. Does it also prevent propagating the VL from the instruction to its inputs?

It should also prevent propagation to its inputs but its users will still have their VL reduced. I can double check later today

https://github.com/llvm/llvm-project/pull/194548