Auto-vectorizing the RISC-V Linux kernel shows measurable gains, with portability costs
Experiments presented at Linux Plumbers Conference report gains from enabling compiler auto-vectorization for RISC-V kernel workloads: up to 15% for LZ4, 8.4% for Btrfs, 5% for XFS, and 2% for Nginx in the tested setups. This is experimental work, not a general kernel configuration change.
The tradeoffs are real: vector use needs extra context handling and memory, can be contentious in kernel space, and performance tuning may require per-CPU compiler targets. Forum commenters debated vector-register context-switch costs; one pushed back that the scalability tradeoff is not so black-and-white.