Getting the Most out of the LLVM Auto Vectorizer for RISC-V Vectors (RVV)

on YouTube ↗19m

" summary

LLVM 14 enables VLA autovectorization for RISC‑V but SiFive adds an internal RVVLA removes remainder loop implementation that computes VL each vector iteration, yielding mixed SPEC2006 results (noticeable wins overall, regressions in hmmer and libquantum) and code‑size shifts. LLVM’s type system cannot express fractional LMUL encoding problem, loop vectorizer heuristics and codegen need fixes, and SLP still forces power‑of‑two trees and straight‑line only. Practical advice: use LLVM vectorizer reports, avoid pragma vector_width for RVV, apply IVDEP or loop FP‑reassoc pragmas, or disable vectorization/SLP when needed.

AI-written summary · about

" chapters (7)

1. 0:00Introduction to Vectorization
2. 1:08Loop Vectorization History
3. 4:47Technical Challenges
4. 7:19User and Terminology Issues
5. 9:48Best Practices and Pragmas
6. 14:15SLP Vectorization Limitations
7. 17:37Summary and Conclusion

" related talks

• 16mImprovements to RISC-V Vector code generation in LLVMAlex Bradbury, Luke Lau · RISC-V Summit Europe 2025
• 17mMaking LLVM RISC-V Compiler More Performant for EveryoneIvan Baev · RISC-V Summit North America 2023
• 18mPreview RISC-V GCC Features and Optimization Improvements for GCC-14Jeff Law · RISC-V Summit North America 2023
• 16mAccelerate HPC and AI Applications with RVV Auto-VectorizationHualin Wu · RISC-V Summit North America 2023
• 10mAn Introduction to RISC-V Vector Programming with C IntrinsicsChih-Mao Chen · RISC-V Global Forum 2020
• 17mRecent Developments in Optimizing Compilers for RISC-VJeff Law · RISC-V Summit North America 2025
• 19mAggregation Optimization for SIMD Everywhere from ARM Neon to RISC-V Vector and Crypto ExtensionsJenq-Kuen Lee, Hung-Ming Lai · RISC-V Summit North America 2024
• 28mLLVM for RISCVAlex Bradbury · 8th RISC-V Workshop Barcelona 2018