spark 'join with condition' versus 'join with filter', which is better?

Question

I want to join two dataframes based on some condition. Which of the two approaches has better performance characteristics?

first approach