https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123343

--- Comment #12 from Zhongyao Chen <chenzhongyao.hit at gmail dot com> ---
(In reply to Richard Biener from comment #10)

Dump shows multi-lane SLP failed, but size-1 subgroup analysis "succeeded" with
poor code generation.

Shouldn't subgroups also require size > 1? I added a check and now size-1 cases
fall back to single-lane reduction, which gives the expected asm.

Reply via email to