https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126998

--- Comment #5 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Richard Biener <[email protected]>:

https://gcc.gnu.org/g:600e25ecc075c82be09240255567180fc2849fb8

commit r17-3693-g600e25ecc075c82be09240255567180fc2849fb8
Author: Richard Biener <[email protected]>
Date:   Wed Aug 26 14:08:18 2026 +0200

    Refactor vect_reduction_update_partial_vector_usage

    vect_reduction_update_partial_vector_usage is called from two places,
    both time with vectype_in as vector type but once with the reduction
    slp_node and once with the slp_node for the input.  Where that
    makes a difference is in the nvectors argument that gets passed
    to vect_record_loop_{len,mask}.  This is a bit weird and it will
    stand in the way of PR126998.  So the following refactors this
    a bit, leaving the updating parts to the caller where it might
    be easier to reason this is (in-)correct.

            PR tree-optimization/126998
            * tree-vect-loop.cc (vect_reduction_update_partial_vector_usage):
            Return a bool, defer vect_record_loop_{len,mask} to the caller.
            (vectorizable_lane_reducing): Adjust.
            (vectorizable_reduction): Likewise.

Reply via email to