The Linux kernel build for RISC-V doesn't currently make use of the CPU ISA's Vector extension. but leveraging it with compiler auto vectorization is showing the ability to provide nice performance results for enhancing RISC-V Linux performance.

Andy Chiu presented on optimizing the Linux kernel build with RISC-V Vector this past week at Linux Plumbers Conference 2026. For kernel use-cases like crypto, RAID, XOR, and user-copy, the RISC-V Vector capabilities can help increase performance. Albeit the kernel then depends upon the Vector unit and is a less portable build. Really tailoring the performance for the best auto vectorization performance also requires per-CPU compiler targeting.

But in experiments conducted, RISC-V Vector optimizing the Linux kernel can pay nice dividends:

There were up to 15% gains in LZ4, Btrfs saw up to 8.4% gains, XFS up to 5%, and the impact on Nginx performance for the RISC-V Linux system ended being up to 2%. The downside is that enabling Vector use requires managing extra context and in turn time and increased memory needs. Plus FPU/vector use can be taboo in kernel space. And again allowing the kernel build to perform the best requires per-CPU tuning by the compiler.

Those wishing to learn more around the prospects of RISC-V Vector usage for kernel mode on Linux can be found via the [1]LPC presentation .



[1] https://lpc.events/event/20/contributions/2455/