Add decomposition for aten.native_batch_norm_backward op

This commit adds decomposition for the `aten.native_batch_norm_backward`
op.

Signed-Off-by: Gaurav Shukla <gaurav@nod-labs.com>

Author: Gaurav Shukla

functorch/_src/decompositions.py

@@ -497,6 +497,61 @@ def native_layer_norm_backward(grad_out: Tensor, input: Tensor, normalized_shape
     return (d_input, d_weight, d_bias)
 
 
+@register_decomposition(aten.native_batch_norm_backward)
+def native_batch_norm_backward(grad_out: Tensor, input: Tensor, weight: Optional[Tensor], running_mean: Optional[Tensor], running_var: Optional[Tensor], save_mean: Optional[Tensor], save_invstd: Optional[Tensor], train: bool, eps: float, output_mask: List[bool]) -> Tuple[Tensor, Optional[Tensor], Optional[Tensor]]:
+    input_shape = input.shape
+    input_rank = input.dim()
+    assert input_rank >= 2, "rank of the input must be at least 2"
+
+    axis = 1
+    num_features = prod(input_shape) / input_shape[axis]
+    mean = save_mean
+    invstd = save_invstd
+    if train:
+        assert save_mean is not None and save_invstd is not None, "when train=True, save_mean and save_invstd are required"
+    else:
+        assert running_mean is not None and running_var is not None
+        mean = running_mean
+        invstd = torch.rsqrt(running_var + eps)
+
+    broadcast_mask = [1] * input_rank
+    broadcast_mask[axis] = input_shape[axis]
+
+    reduction_axes = []
+    for i in range(input_rank):
+        if i != axis:
+            reduction_axes.append(i)
+
+    mean = torch.reshape(mean, broadcast_mask)
+    norm = 1.0 / num_features
+    grad_output_sum = torch.sum(grad_out, reduction_axes)
+    dot_p = torch.sum(grad_out * (input - mean), reduction_axes)
+
+    grad_mean = torch.reshape(grad_output_sum * norm, broadcast_mask)
+    proj_scale = torch.reshape(torch.mul(dot_p * norm, invstd * invstd), broadcast_mask)
+
+    grad_scale = None
+    if weight is None:
+        grad_scale = torch.reshape(invstd, broadcast_mask) * 1.0
+    else:
+        grad_scale = torch.reshape(invstd * weight, broadcast_mask)
+    grad_input = None
+    if train:
+        proj = (input - mean) * proj_scale
+        grad_input = ((grad_out - proj) - grad_mean) * grad_scale
+    else:
+        grad_input = grad_out * grad_scale
+
+    grad_weight = None
+    if output_mask[1]:
+        grad_weight = dot_p * invstd
+
+    grad_bias = None
+    if output_mask[2]:
+        grad_bias = grad_output_sum
+    return (grad_input, grad_weight, grad_bias)
+
+
 @register_decomposition(aten.clamp_min)
 def clamp_min(self: Tensor, min: float):
     return torch.clamp(self, min=min)