[Bug tree-optimization/107090] [aarch64] sequence logic should be combined with mul and umulh

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107090

--- Comment #24 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Philipp Tomsich <[email protected]>:

https://gcc.gnu.org/g:006150ddc40844105eb99bc6514267314715f6dc

commit r17-3441-g006150ddc40844105eb99bc6514267314715f6dc
Author: Konstantinos Eleftheriou <[email protected]>
Date:   Wed Jul 8 02:43:36 2026 -0700

    forwprop: Add long-multiply low-plus variant

    Extend the long-multiply fold to a LOW_PART variant recovering the
    lower 2N bits as

      xl*yl + (cross_sum << N)
      cross_sum = xh*yl + xl*yh

    This form is also the low half of a two-carry long-multiply, where an
    unsigned overflow compare against one of the PLUS operands feeds the
    matching high-part fold.  Folding here would destroy (cross_sum << N)
    while the compare still needs it, so the fold is deferred while any
    GT/LT/GE/LE use shares an operand with the PLUS; once the high-part
    fold has consumed the compare, the next forwprop instance folds the
    rest.

            PR tree-optimization/107090

    gcc/ChangeLog:

            * tree-ssa-forwprop.cc (long_mul_check_low_plus_defer): New
            helper, defers the fold when any GT/LT/GE/LE use shares an
            operand with the PLUS; gates the LOW_PART PLUS_EXPR row added
            to long_mul_table.

    gcc/testsuite/ChangeLog:

            * gcc.dg/tree-ssa/long-mul-two-carry.c: Add a forwprop4 LOW
            fold expectation for full_mul_two_carry, deferred from
            forwprop3 by long_mul_check_low_plus_defer.
            * gcc.dg/tree-ssa/long-mul-low-plus.c: New test.