Following discussion in Code review comment of ⚙ D148498 [mlir] tosa.reshape - Add InferTensorType interface , this is a suggestion to remove ‘new_shape’ from tosa.reshape as this adds no information to the operation and was the most direct implementation of reshape from the TOSA specification to the dialect.
Any objections/thoughts regarding ‘new_shape’ removal?
@eric-k @jpienaar @mgehre-amd @maya_amrami
Sorry for the delay in getting back to you. I’ve been trying to think of cases where the new_shape would give us information that wouldn’t already be carried elsewhere in the dialect. My main concern is making sure that shape inference continues to work in all cases. I’ll also tag @rsuderman in case he can think of any problems.