[llvm] [LV] Convert scatter stores with constant stride into strided stores (PR #206896)

[Luke Lau via llvm-commits]

lukel97 wrote:

After investigating a bit, both the interleaved access and strided accesses are created in VPlan. But for factor=5 the VF range is split to VF=[vscale x 1] because for i64 a vscale x 2 NF=5 vsseg is illegal. 

Subsequently, the 5 strided stores at VF=vscale x 2 ends up being cheaper than a single NF=5 segmented store at VF=vscale x 1. So it's an issue somewhere in the RISC-V cost model issue. I'll take a look, but I don't think it should be blocking for this PR.

https://github.com/llvm/llvm-project/pull/206896