[llvm] Update the base and index value for masked gather (PR #130920)

[Simon Pilgrim via llvm-commits]

https://github.com/RKSimon commented:

I'm thinking this is more complicated than it needs to be - we should be able to transfer some/all of the SHL into the gather/scatter scale operand (max scale = 8) - assuming we don't lose the sign bit. By doing that we should(?) be able to then recognize when its safe to remove the zext to v16i64 (illegal type so should always be able to truncate to v16i32) - does that make sense?

https://github.com/llvm/llvm-project/pull/130920