Missing stride_ in vertical pointer arithmetic when advance_rank=0 (PredicatedTileAccessIterator)

[piaoyanglink]

I‘m researching cutlass！when advance_rank == 0 and tile_offset.strided() != 0, the pointer offset calculation is missing params_.stride_, which will lead to incorrect address computation.Code Location (Provide key snippets) is add_tile_offsetin predicated_tile_access_iterator.h
the code is below:

pointer_ += Shape::kStrided * tile_offset.strided() * sizeof_bits::value / 8;

maybe it's like this:

pointer_ += Shape::kStrided * tile_offset.strided() * params_.stride_ * sizeof_bits::value / 8;

This is my speculation; I’m not sure if it’s correct, and I hope to get a response.