[Bug tree-optimization/127130] [14/15/16/17 Regression] wrong code at -O{2,3} on x86_64-linux-gnu

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127130

--- Comment #7 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Richard Biener <rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:7686bcbb2b7b08688008a6a0e29e6561af1a9a98

commit r17-3818-g7686bcbb2b7b08688008a6a0e29e6561af1a9a98
Author: Richard Biener <rguenther@suse.de>
Date:   Tue Sep 1 15:09:42 2026 +0200

    tree-optimization/127130 - avoid peeled CHREC with UB

    The following properly guards the peeled CHREC optimization
    when we are analyzing a CHREC with non-wrapping overflow.
    Otherwise we can introduce UB which can manifest via
    infer_loop_bounds_from_signedness and thus wrong niter analysis.

    This regresses gcc.dg/gomp/static-chunk-size-one.c because we
    can no longer replace a signed IV via a peeled chrec and lack
    other means.

            PR tree-optimization/127130
            * tree-scalar-evolution.cc (simplify_peeled_chrec):
            Guard against signed integer overflow UB.

            * gcc.dg/torture/pr127130-1.c: New testcase.
            * gcc.dg/torture/pr127130-2.c: Likewise.
            * gcc.dg/gomp/static-chunk-size-one.c: Adjust.