mb_ereg_search & subpatterns
use loop:
<?php
    $str = "中国abc + abc ?!？！字符＃　china string";

    $reg = "\w+";

    mb_regex_encoding("UTF-8");

    mb_ereg_search_init($str, $reg);
    $r = mb_ereg_search();

    if(!$r)
    {
        echo "null\n";
    }
    else
    {
        $r = mb_ereg_search_getregs(); //get first result
        do
        {
            var_dump($r[0]);
            $r = mb_ereg_search_regs();//get next result
        }
        while($r);
    }
?>

A 'match_all' helper function based on dulao's answer. Someone might find it useful...

<?php
function mb_ereg_match_all($pattern, $subject, &$matches, $options = '', $setOrder = false, $offset = 0) {
    if (!mb_ereg_search_init($subject, $pattern, $options)) {
        return false;
    }
    if ($offset != 0 && !mb_ereg_search_setpos($offset)) {
        return false;
    }
    $matches = [];
    if (!mb_ereg_search()) {
        return 0;
    }
    $regs = mb_ereg_search_getregs();
    $count = 0;
    do {
        $count++;
        if ($setOrder) {
            foreach ($regs as $key => $val) {
                $matches[$key][] = $val;
            }
        } else {
            $matches[] = $regs;
        }
        $regs = mb_ereg_search_regs();
    }
    while($regs);
    return $count;
}
?>