Reusing Auto-Schedules for Efficient DNN Compilation

Auto-scheduling is a process where a search algorithm automatically explores candidate schedules (program transformations) for a given tensor program on a given hardware platform to improve its performance. However this can be a very time consuming process, depending on the complexity of the tensor program, and capacity of the target device, with often many thousands of program variants being explored. To address this, in this paper we introduce and demonstrate the idea of \emph{tuning-reuse}, a novel approach to identify and re-use auto-schedules between tensor programs. We demonstrate this concept using Deep Neural Networks (DNNs), taking sets of auto-schedules from pre-tuned DNNs, and using them to reduce the inference time of a new DNN. Given a set of pre-tuned schedules, tuning-reuse provides its maximum speedup in less time than auto-scheduling using the state-of-the-art Ansor auto-scheduler. On a set of widely used DNN models, we apply tuning-reuse and achieve maximum speedups between $1.16\times$ and $4.76\times$, while outperforming Ansor when given limited tuning time.