Simulate, trace, and evaluate a RISC-V system leveraging very long vectors

on YouTube ↗11m

" summary

A QEMU-based RAV tracing plugin runs RISCV vector binaries on a laptop and produces instruction counts, per-phase scalar/vector breakdowns, and time-series of vector length so programmers can see actual behavior. Targets the EPAC accelerator with 256 double-precision elements (≈16,000 bits) per vector. Example results show average VL 32 of 256 and phase vectorization as low as 7% vector instructions, guiding loop-collapsing and blocking changes that raise VL toward the hardware maximum and produce measured speedups on real EPAC silicon.

AI-written summary · about

" chapters (10)

1. 0:00Introduction to RAB
2. 1:26EPAC chip and vector units
3. 2:29Demo of vectorized code
4. 4:17Tracing with RAB plugin
5. 5:23Instrumenting code regions
6. 6:29Analyzing instruction traces
7. 7:12Compiler vectorization insights
8. 8:17Vector length evolution
9. 9:20Optimizing loop structures
10. 10:03Quantifying plugin benefits

" related talks

• 13mUnleashing ML Processing Power Through RISC-V Vectors: Applications, Implementation and...B. Barker · RISC-V Summit North America 2025
• 01h 45mPorting SIMD to RISC-V Vectors | Corrected AudioDaniel Thompson · RISC-V Summit Europe 2026 Developer Workshops
• 34mRISC V Vector Extension ProposalRoger Espasa · 7th RISC-V Workshop 2017
• 13mKeynote: Taking RISC-V Intelligence EverywhereCharlie Su · RISC-V Summit North America 2023
• 10mAvoiding Amdahl's Law RISC-V Architecture Exploration for AI & ML ComputeSimon Davidmann · RISC-V Summit 2019
• 36mVector ISARoger Espasa · 8th RISC-V Workshop Barcelona 2018
• 11mGetting Started with RISC-V Custom InstructionsJon Taylor · embedded world 2023
• 24mMoving to RISC-V Vector: A Practical Journey of AI Operator OptimizationGuodong Xu · RISC-V Summit North America 2025