ATen-derived composite ops

Composite ops recognized automatically from the ATen nodes (fx.Nodes) in your ExportedProgram during conversion. These have no corresponding nn.Module wrapper — use the standard PyTorch APIs (e.g., torch.nn.BatchNorm2d, torch.nn.functional.pixel_shuffle) and Core AI preserves them as composite ops, as long as get_decomp_table() keeps them from being decomposed.

• batch_norm
• group_norm
• hard_sigmoid
• instance_norm
• layer_norm
• linalg_vector_norm
• log_softmax
• pixel_shuffle

For the ATen ops these are derived from, see Supported ATen ops.