Both the implementations of CUDA merge & the set operations assume that the
number of splitters picked from both ranges will be the same. If the ranges
are of even modestly different size, this cannot work. We need to do book
keeping on the (different) number of splitters sampled from both ranges.
Original issue reported on code.google.com by jaredhoberock on 20 Jul 2011 at 9:19
Original issue reported on code.google.com by
jaredhoberock
on 20 Jul 2011 at 9:19