I was confused at first, but from what I can tell the high costs for ordered reductions cause us to run into a different vectorizer "give-up path"? That explains the -mmax-vectorization behavior and adjusting the expectation the way you did is appropriate.
I'm seeing one more fail in the CI: https://github.com/ewlu/gcc-precommit-ci/issues/4573#issuecomment-4475975522 FAIL: gcc.target/riscv/rvv/autovec/cond/pr111401.c -ftree-vectorize -O2 -mrvv-max-lmul=m1 scan-tree-dump-times vect "vectorized 1 loop" 5 You'll need to check if -mmax-vectorization results in similar code as before but I suppose it will. Once this is done, I think we're good. -- Regards Robin
