This submission initially received positive but cautious reviews, with two reviewers recommending a borderline accept. After reading the author response, both of these reviewers were satisfied that their concerns were addressed (regarding correctness of GM-SM implementation + the technical writing), so they both increased their score to "accept." While more experiments would have made the paper stronger, all reviewers ultimately agreed that this is an important problem and the paper makes a good contribution that can spur future work in the area.