Building Tool Chains for RISC-V AI Accelerators

on YouTube ↗19m

" summary

A working open-source flow that runs PyTorch, TensorFlow and OpenXLA on an emulated RV64 accelerator using QEMU RISC‑V model of a few cores and a memory‑mapped tensor/command register interface, proving early software bring‑up before silicon. Key tools are private use one and execuTorch hooks to intercept ops, and the OneAPI/OpenCL/SYCL route (OneAPI Construction Kit + HAL) to generate host plus accelerator ELF code. Performance tuning around RVV remains hard but the prototype successfully executes and parallelizes large kernels across cores.

AI-written summary · about

" chapters (6)

1. 0:00AI system tooling overview
2. 3:12Prototyping RISC-V co-processors
3. 5:46Implementing the oneAPI framework
4. 11:38Construction kit resources
5. 12:27Developer community and support
6. 15:16Q&A and closing remarks

" related talks

• 23mEmpower Upstream ML Frameworks on RISC-VTiejun Chen · RISC-V Summit 2022
• 56mHow to: Bringing up ExecuTorch on a RISC-V MicrocontrollerJeremy Bennett · RISC-V Summit Europe 2026 Developer Workshops
• 25mPyTorch on RISC-V: From Cross-Compilation To Native CILudovic Henry · PyTorch Conference Europe 2026
• 10mLightning Talk: Software Development for ML and RISC-V Vector Acceler...Simon Davidmann, Lee Moore · RISC-V Forum: Vector & Machine Learning 2021
• 13mKeynote: Taking RISC-V Intelligence EverywhereCharlie Su · RISC-V Summit North America 2023
• 56mHow to: Bringing up ExecuTorch on a RISC-V Microcontroller | Corrected AudioJeremy Bennett · RISC-V Summit Europe 2026 Developer Workshops
• 19mDemocratizing Inference of Open-weight Models on RISC-V Manycore Acc...R. Shaposhnik, T. Dadasheva · RISC-V Summit North America 2025
• 25mDeep Dive: Accelerating Neural Networks using RVV and Open Standard SoftwareMehdi Goli · RISC-V Forum: Vector & Machine Learning 2021