[Bug cobol/119456] gcobol: huge codegen for use of reference-modification (and missing optimization)

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119456

--- Comment #2 from GCC Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Robert Dubner <rdubner@gcc.gnu.org>:

https://gcc.gnu.org/g:435346eafadece9d045874ba000cb4a53f625f0c

commit r16-7853-g435346eafadece9d045874ba000cb4a53f625f0c
Author: Robert Dubner <rdubner@symas.com>
Date:   Mon Mar 2 15:36:40 2026 -0500

    cobol: Improved efficiency of code generated for MOVE "A" TO VAR(1:1).
[119456]

    This PR rightly noted that COBOL source code which obviously could
    result in simple machine language did not.  These changes take advantage
    of the compiler knowing, at compile time, the values of literal offsets
    and lengths, and uses that knowledge to generate much more efficient
    GENERIC for those cases.

    gcc/cobol/ChangeLog:

            PR cobol/119456

            * genapi.cc (mh_source_is_literalA): Don't set refmod_e attribute
            unless it is necessary.
            (have_common_parent): Helper routine that determines whether two
            COBOL variables are members of the same data description.
            (mh_alpha_to_alpha): Modified for greater efficiency when table
            subscripts and reference modification parameters are numeric
            literals.
            * genutil.cc (get_data_offset): Recognizes when table subscripts
            and refmod offsets are numeric literals.
            (refer_size): Recognizes when refmod offsets are numeric literals.
            (refer_size_source): Recognizes when table subscripts are numeric
            literals.