[llvm] [RISCV] Move vnclip patterns into DAGCombiner. (PR #93728)

[Craig Topper via llvm-commits]

topperc wrote:

> LGTM
> 
> Aside - This is starting to look a bit like isSaturatingMinMax in DAGCombine. This version handles the _VL variants, and that one handles the select variants, so neither is a subset of the other. Do we have any room to optimize the saturating fp_to_sint patterns the DAG is working on? I don't see anything but figured it's worth asking.. Alternatively, is there anything we can do to optimize/recognize the clamp idiom on it's own?

Looks like there are some vnclips in test/CodeGen/RISCV/rvv/fpclamptosat_vec.ll but also some min/max/vnsrl.

https://github.com/llvm/llvm-project/pull/93728