Auto Re-vectorization Methodology and Performance Analysis

This whitepaper discusses a novel approach to auto re-vectorization of code for RISC-V Vector architecture. By leveraging existing hand-optimized kernels and LLVM IR transformations, the methodology aims to optimize performance metrics while minimizing manual effort. The paper details the conversion process from x86 AVX and ARM Neon to RISC-V Vector assembly, compares performance results with alternative methods, and provides insights into the effectiveness of the proposed tool.