[throwaway81523]

I thought the cool way to do this was reverse each sub-block in place, and then reverse the whole vector. If you draw a few diagrams you can see that accomplishes the rotation. Yes it moves everything twice, but it has near perfect cache locality, so at least on large blocks tends to be faster than cycle decomposition or other juggling tricks. It's discussed in one of Bentley's "Programming Pearls" books.