torch.fx.experimental.migrate_gradual_types.constraint_transformation.apply_padding#
- torch.fx.experimental.migrate_gradual_types.constraint_transformation.apply_padding(e1_var, e11, e2, e12, d2, d11, d12, counter)[source]#
We are considering the possibility where one input has less dimensions than another input, so we apply padding to the broadcasted results
- Parameters:
e1_var (TVar) – Variable representing the first input where padding will be
e11 (BinConstraintT) – constraint of the form e11 = Tensortype[d1, …, dn]
e2 (BinConstraintT) – constraint of the form e2 = Tensortype[d1, …, dn]
e12 (BinConstraintT) – constraint of the form e11 = Tensortype[d1, …, dn]
d2 (list[DVar]) – Tensor variables for the second input
d11 (list[DVar]) – Tensor variables for the broadcasted first input
d12 (list[DVar]) – Tensor variables for the broadcasted second input
counter (int) – variable tracking
Returns: A new constraint whose goal is to apply padding to the broadcasted result
