This pass is to handle computationally complex operations like tensor.pad which are not simply lowered to the exact same operation in the memref dialect. Differential Revision: https://reviews.llvm.org/D125384
