Contradiction in comparison of summation algorithms? #28
Description
In the Haddocks for Numeric.Sum, under the Kahan sum, it says
This summation method is included for completeness. Its use is not recommended. In practice, KBNSum is both 30% faster and more accurate.
But under KBN sum, it says:
Kahan-Babuška-Neumaier summation. This is a little more computationally costly than plain Kahan summation, but is always at least as accurate.
These statements seem to contradict each other---the first says that KBN sum is 30% faster than Kahan, but the second says that KBN is more costly (presumably in time, but maybe it means something else) than Kahan.
Am I interpreting this correctly?