david-arm wrote: I wonder if the DataWithoutLaneMask style is even functionally correct in all cases because get.active.lane.mask correctly deals with any potential overflow of the induction variable mid-vector? https://github.com/llvm/llvm-project/pull/93303