In this Pull Request #3802 , around 0.5s out of 1.2s is spent in check_pairwise_arrays which is huge, because I call pairwise_distances_argmin_min repeatedly.

1. There is dtype casting two times in check_pairwiseto make sure X and Y are float, first is to convert it into a ndarray, and the second is to cast it into float. Should there be a dtype argument provided externally so that lines from https://github.com/scikit-learn/scikit-learn/blob/master/sklearn/metrics/pairwise.py#L98 to L104 can be removed.
2. Should there be a check_input=False argument in the various metrics to avoid multiple checking of the input arrays? so that the user can override check_pairwise_arrays