https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128818

--- Comment #2 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:59412e46c04fc5a6142283983ae7f8ff3d520c88

commit r17-6407-g59412e46c04fc5a6142283983ae7f8ff3d520c88
Author: Richard Biener <[email protected]>
Date:   Tue Nov 17 15:51:51 2026 +0100

    tree-optimization/128818 - wrong-code with VN skipping same-val store

    Similar issues as in PR128637 exist in the same-val store skipping
    code as noted by Sashiko.  Two revolve around redundant store
    elimination so the correct action is to avoid same-val store
    skipping when searching for an earlier redundant store.
    Another is inserting into the hashtable with an earlier vuse
    when we skipped defs in vn_walk_cb_data::finish.

    This makes us no longer optimize the apparent redundant store
    in gcc.dg/tree-ssa/pr91091-2.c and rightfully so because removing
    the later store changes the effective type of the storage if the
    objects are at the same address.

            PR tree-optimization/128818
            * tree-ssa-sccvn.cc (vn_reference_lookup_3): Avoid skipping
            same-val may-defs when doing redundant store elimination.
            (vn_walk_cb_data::finish): Also skip hashtable insertion when
            same-val stores were skipped.

            * gcc.dg/torture/pr128818-1.c: New testcase.
            * gcc.dg/torture/pr128818-2.c: Likewise.
            * gcc.dg/torture/pr128818-3.c: Likewise.
            * gcc.dg/tree-ssa/pr91091-2.c: Adjust.

Reply via email to