[llvm] [VectorCombine] foldShuffleOfShuffles - fold shuffle(shuffle(x,y),poison) length changing masks (PR #144690)

[David Green via llvm-commits]

davemgreen wrote:

The cost model for length-changing shuffle vectors currently has some target-dependant assumptions build in, that I was hoping to try and remove. I've not had as much time to look into addressing them as I would have hoped. Do you mind if we try to address that first?

I was thinking it might be nice to push a lot of the logic for shuffle masks currently in getInstructionCost into the Target functions, so that getInstructionCost and getShuffleInstrCost are less different and more easily testable.

https://github.com/llvm/llvm-project/pull/144690