PyTorch on RISC-V: From Cross-Compilation To Native CI
summary
Porting PyTorch to RISC‑V moves from cross‑compilation to native CI by resolving concrete gaps: GCC/LLVM, Ubuntu 24.04 builds, OpenBLAS RVV support and upstream oneDNN work. A native build produces a 258 MB native wheel that runs, but tests lack scale and CI is immature — current native runs take about 10 hour native build on available machines. Performance lacks micro‑arch tuning (RVV vector kernels missing microarch tuning). RISC‑V runners exist as native GitHub runners (bare metal) and roadmap targets faster hardware, green CI labels, then full PR testing.
AI-written summary · about
chapters (11)
- 0:00PyTorch and RISC-V intro
- 3:17What is RISC-V
- 6:47PyTorch dependency chain
- 8:33RISC-V in Linux distros
- 11:23Native PyTorch builds
- 12:04CI setup for RISC-V
- 13:51Community engagement
- 15:40Hardware acceleration gaps
- 17:30RISC-V for edge AI
- 18:52Help port libraries
- 20:01Q&A on toolchains
related talks
19mBuilding Tool Chains for RISC-V AI Accelerators
21mWhat's New for PyTorch Developer Infrastructure
17mEnabling RISC-V Success: From Design to Deployment
26mRISCV patchwork CI
30mGetting Linux Distros to New Architectures
34mScheduling in CI/CD systems
01h 45mRun Frontier AI at Home
04mRecognizing the PyTorch Foundation CICD Infrastructure Community