Zhongchun Zheng

Zhongchun Zheng

Publications

Conference Papers

VecTrans: LLM Transformation Framework for Better Auto-vectorization on High-performance CPU

Published in , 2025

VecTrans leverages LLMs to enhance compiler-based code vectorization. And it first employs compiler analysis to identify potentially vectorizable code regions. It then utilizes an LLM to refactor these regions into patterns that are more amenable to the compiler’s auto-vectorization.

Recommended citation: Zheng, Z., Cheng, L., Li, L., Rocha, R., Liu, T., Wei, W., Zhang, X. \& Gao, Y. VecTrans: LLM Transformation Framework for Better Auto-vectorization on High-performance CPU. (2025), https://arxiv.org/abs/2503.19449
Download Paper

mLOOP: Optimize Loop Unrolling in Compilation with a ML-based Approach

Published in NAS 2024, 2024

mLOOP employs the XGBoost model to predict loop unroll factors. And it is implemented as an LLVM optimization pass, enabling seamless deployment within existing compiler pipelines.

Recommended citation: Z. Zheng, Y. Wu and X. Zhang, "mLOOP: Optimize Loop Unrolling in Compilation with a ML-based Approach," 2024 International Conference on Networking, Architecture and Storage (NAS), Zhuhai, China, 2024, pp. 1-8, doi: 10.1109/NAS63802.2024.10781373.
Download Paper