Comment 21 for bug 2030515

The release/2.39/master branch has been updated by Arjun Shankar <email address hidden>:

https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=aa4249266e9906c4bc833e4847f4d8feef59504f

commit aa4249266e9906c4bc833e4847f4d8feef59504f
Author: Adhemerval Zanella <email address hidden>
Date: Thu Feb 8 10:08:38 2024 -0300

    x86: Fix Zen3/Zen4 ERMS selection (BZ 30994)

    The REP MOVSB usage on memcpy/memmove does not show much performance
    improvement on Zen3/Zen4 cores compared to the vectorized loops. Also,
    as from BZ 30994, if the source is aligned and the destination is not
    the performance can be 20x slower.

    The performance difference is noticeable with small buffer sizes, closer
    to the lower bounds limits when memcpy/memmove starts to use ERMS. The
    performance of REP MOVSB is similar to vectorized instruction on the
    size limit (the L2 cache). Also, there is no drawback to multiple cores
    sharing the cache.

    Checked on x86_64-linux-gnu on Zen3.
    Reviewed-by: H.J. Lu <email address hidden>

    (cherry picked from commit 0c0d39fe4aeb0f69b26e76337c5dfd5530d5d44e)