This takes 75% less time per operation per some benchmarks on my mac. Signed-off-by: Andrew Dunham <andrew@du.nham.ca>