Partially parallelize randperm on CPU. (#21529)
Summary:
This commit parallelizes the variable initialization (from 1 to n) step on CPU.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/21529
Differential Revision: D15855402
Pulled By: VitalyFedyunin
fbshipit-source-id: f1ba54587451f9cb0eb5e542c3c5b458b48e1a3d