quick temporary fix for a perf issue we just identified with
vectorization....
now the sum benchmark runs 3x faster with vectorization than without.
1 file changed