The output of splitLargeGEPOffsets does not appear to be deterministic because of the way that we iterate over a DenseMap. I've changed it to a MapVector for consistent output.
The test case here isn't particularly great. It shows a difference in output, but only really a cosmetic one. I have a (much) larger case that has the splitgep elements placed in a different order, leading to different registry allocation on Arm. It's the kind of problem goes away if I start removing comment out lines though :-/