> tried to use a new -mrvv-max-lmul=m1-dynamic for such cases. As Robin > suggested using a cost model property, I composed this one.
Just FYI : check this proposal, if approved, we may use Olvt in -march to guide if LMUL latency is dependent of VL. https://riscv.atlassian.net/wiki/spaces/TOXX/pages/585400366/Optimization+Guidance+Options+Fast+Track+Approval+Request -- Regards, Zhongyao
