This is to reduce PCTRIE lookups.
Given swap_pager_haspage() is called frequently during heavy swap in and out activities, avoiding each individual lookup helps a lot.
For example, on i386, SWB_NPAGES is 32 and each swap_pager_haspage may call
PCTRIE_LOOKUP up to 65 times. Instead, new code will call 9 lookups for
the same size giving speed up of 700%.
On amd64, it is 64 and even twice faster as SWB_NPAGES is 64.