tree dfda2b0a0297f90158aeffdfe045d31c13eea717
parent e75c33f0f32989f467fa88f0bf789ce1acf19d01
author Vincent Rabaud <vrabaud@google.com> 1669308673 +0100
committer Vincent Rabaud <vrabaud@google.com> 1670341063 +0000

K-means: use L1 norm when dim == 1.

In 1 dimension, finding the nearest neighbor using L2 or L1 is
equivalent. Only the best distance needs to be squared.

The speed-up is negligible: the goal is to make the code easier
to transition to 16-bit inputs.

Change-Id: Ibbff387aec1055655b4d96f3b28d0dcfddd80e6e