[Bug target/127114] i386: Bad convert-storm on _Float16 builtin_round vectorization

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127114

--- Comment #3 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Hongyu Wang <hongyuw@gcc.gnu.org>:

https://gcc.gnu.org/g:c8a9bc5cf5d788a3bc796341f8070d073076b91b

commit r17-3843-gc8a9bc5cf5d788a3bc796341f8070d073076b91b
Author: Hongyu Wang <hongyu.wang@intel.com>
Date:   Fri Aug 28 14:10:01 2026 +0800

    i386: Allow round for HFmode in optab hook [PR127114]

    The match.pd narrowing of (_Float16) round ((double) x) to a
    native HF round query direct_internal_fn_supported_p with
    OPTIMIZE_FOR_BOTH, so align the backend optab_supported_p hook
    like the ceil/floor/btrunc did to avoid enforced extend/truncate
    at vectorization.

    gcc/ChangeLog:

            PR target/127114
            * config/i386/i386.cc (ix86_optab_supported_p): Return true for
            round_optab in HFmode.

    gcc/testsuite/ChangeLog:

            PR target/127114
            * gcc.target/i386/pr127114.c: New test.