Design and Implementation of A Multi-Core CPU Compilation Optimization Framework Based on MLIR
Han Gao, Yixuan Zhao, Tanbao Yan, Yuhong Zhao · 2023
Most on-board software is developed and adapted around traditional hardware resources. When high-performance hardware is introduced to the on-board heterogeneous computing platform, on-board software cannot fully leverage the processing capabilities of the high-performance hardware, making it difficult to meet the real-time requirements of multiple application tasks. For signal processing algorithms and other non-matrix complex scientific computing tasks to be arranged and adapted on multi-core CPUs, traditional algorithms are mainly developed based on DSP processors, making it difficult to directly deploy them on multi-core CPU processors. This article proposes and constructs a MLIR-based multi-core CPU compilation optimization framework, designs an automatic software code execution optimization process, achieves full adaptation of on-board software on multi-core CPU resources, improves the processing efficiency of complex scientific computing tasks, and solves the problem of poor real-time performance of complex scientific computing tasks. Compared with the two types of compilation optimization systems currently in use, the average speedup ratio was 7.54 and 3.26 times, respectively.