https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103571
--- Comment #28 from Uroš Bizjak <ubizjak at gmail dot com> --- (In reply to Hongtao.liu from comment #18) > codegen for foo1/foo2 is suboptimal under -mavx2, i guess we can have > vec_setv16hf_0 and with vpblendw. True, some opportunities are missing from expand_vec_perm* functions, someone should go through these expanders and add corresponding VxHFmode near VxHImode handling.