Fix overflow issue in quantized instance_norm/layer_norm/group_norm (#54872)
Summary:
Fixes https://github.com/pytorch/pytorch/issues/54837
`hsum_sq` has the overflow issue when the input image size is large such as (H,W,D) as (224,224,160). `hsum_sq` is used in the quantized instance_norm/layer_norm/group_norm.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/54872
Reviewed By: dskhudia
Differential Revision: D27690767
Pulled By: vkuzo
fbshipit-source-id: 9b9ac3e76220d42a3b48f8bf4e20823f775789a2
