We introduce Mirage, the first multi-level superoptimizer for tensor programs. A key idea in Mirage is Graphs, a uniform representation of tensor programs at the kernel, thread block, and thread levels of the GPU compute hierarchy. Graphs enable Mirage to discover novel optimizations that combine algebraic transformations, schedule transformations, and generation of new custom kernels. To navigate the large search space, Mirage introduces a pruning technique based on abstraction that significantly reduces the search space and provides a certain optimality guarantee. To ensure that the optimized Graph is equivalent to the input program, Mirage introduces a probabilistic equivalence verification procedure with strong theoretical guarantees. Our evaluation shows that Mirage outperforms existing approaches by up to 3.5 even for DNNs that are widely used and heavily optimized. Mirage is publicly available at https://github.com/mirage-project/mirage.
View on arXiv@article{wu2025_2405.05751,
title={ Mirage: A Multi-Level Superoptimizer for Tensor Programs },
author={ Mengdi Wu and Xinhao Cheng and Shengyu Liu and Chunan Shi and Jianan Ji and Kit Ao and Praveen Velliengiri and Xupeng Miao and Oded Padon and Zhihao Jia },
journal={arXiv preprint arXiv:2405.05751},
year={ 2025 }
}