Group Norm 2D

The group_norm_2d module implements the GroupNorm PyTorch layer with the assumption that affine=False.

With this assumption, this module is able to be specialized into LayerNorm and InstanceNorm modules simply by setting the number of GROUP_CHANNELS to the channel dimension C or 1 respectively.

Note that the group_norm_2d component is an internal hardware component and is exposed to the rest of the MASE software stack through the unified norm component.

Overview

The group_norm_2d is a fully pipelined module which follows the dataflow streaming protocol.