[Bug tree-optimization/112104] loop of ^1 should just be reduced to ^(n&1)

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112104

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

https://gcc.gnu.org/g:62993cc9d2200ef7a05f317cdef3262996c0ac10

commit r17-3460-g62993cc9d2200ef7a05f317cdef3262996c0ac10
Author: Eikansh Gupta <eikansh.gupta@oss.qualcomm.com>
Date:   Thu Jul 23 17:04:17 2026 +0530

    tree-optimization: Handle variable trip count XOR in final value
replacement [PR112104]

    analyze_and_compute_bitop_with_inv_effect replaces a loop that repeatedly
    applies a bitwise op with a loop-invariant operand by its closed form.  The
    BIT_XOR case only handled a constant trip count; for a variable count it
    bailed out, leaving loops like "for (i=0;i<n;i++) j ^= 1;" in place.

    Mask the invariant with that runtime parity to handle a variable count.
    A constant count folds as before.

            PR tree-optimization/112104

    gcc/ChangeLog:

            * tree-scalar-evolution.cc
(analyze_and_compute_bitop_with_inv_effect):
            Handle BIT_XOR_EXPR with a variable trip count.

    gcc/testsuite/ChangeLog:

            * gcc.target/i386/pr105735-1.c: Bump final value replacement count
            from 8 to 9.
            * gcc.target/i386/pr105735-3.c: Likewise.
            * gcc.dg/tree-ssa/pr112104-1.c: New test.
            * gcc.dg/tree-ssa/pr112104-2.c: New test.

    Signed-off-by: Eikansh Gupta <eikansh.gupta@oss.qualcomm.com>