[llvm] [RISCV] Implement cross basic block VXRM write insertion. (PR #70382)

[Craig Topper via llvm-commits]

topperc wrote:

> LGTM. I admit dataflow isn't always my forte and I had to reach my compiler books for the anticipation part :)
> 
> The approach looks sensible to me within limits already acknowledged in the notes and fixmes.
> 

Thanks for the review.

> I fail to see a test in `vxrm-insert.ll` that shows the redundant write case caused by a critical edge. If there isn't one, would it make sense to add it?

I will add one.

https://github.com/llvm/llvm-project/pull/70382