https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128324

--- Comment #3 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The trunk branch has been updated by Richard Earnshaw <[email protected]>:

https://gcc.gnu.org/g:f3de20480c782e6f7f13ec0da7c4c26ff71f83ed

commit r17-5977-gf3de20480c782e6f7f13ec0da7c4c26ff71f83ed
Author: Richard Earnshaw <[email protected]>
Date:   Wed Nov 4 16:23:33 2026 +0000

    arm: fix neon 16-bit lane duplication for big-endian [PR128324]

    There was an issue with the ordering of statements in my patch r17-2930
    when compiling for big-endian.  The call to subreg_regno_offset to pick
    the core register for the extraction needs to be made before we adjust
    the element number to map the big-endian model on the the Arm lane
    ordering.

            PR target/128324

    gcc/ChangeLog:

            * config/arm/neon.md
            (neon_vdup_lan<VDQW:mode>_internal): Calculate the register
            before adjusting the element.
            (neon_vdup_lane<VHFBF:mode>_internal): Likewise.

    gcc/testsuite/ChangeLog:

            * gcc.target/arm/pr128324.c: New test.

Reply via email to