[llvm] [VectorCombine] Allow shuffling with bitcast for not multiple offset for loadsize (PR #119139)

[via llvm-commits]

ParkHanbum wrote:

@RKSimon 
Last time, I asked if the number of IRs increases, can we see if it improves the performance? I was able to see it through the number of Assembly. Thank you.

This time, it's a little difficult, so I'm asking you a question.
I understand that CostModel judge the costs internally, but, 
Is there any way to infer performance improvement through assembly?

https://alive2.llvm.org/ce/z/2gqJjB 

https://github.com/llvm/llvm-project/pull/119139