SpscLinkedQueue#fill#soNext(temp) may be replaced with spNext(next)

[hl845740757]

Just like an object factory, the oldPNode.soNext(head) can ensure correct construction.

   @Override
    public int fill(Supplier<E> s, int limit)
    {
        if (null == s)
            throw new IllegalArgumentException("supplier is null");
        if (limit < 0)
            throw new IllegalArgumentException("limit is negative:" + limit);
        if (limit == 0)
            return 0;

        LinkedQueueNode<E> tail = newNode(s.get());
        final LinkedQueueNode<E> head = tail;
        for (int i = 1; i < limit; i++)
        {
            final LinkedQueueNode<E> temp = newNode(s.get());
            // The fence here may be redundant
            tail.soNext(temp);
            tail = temp;
        }

        final LinkedQueueNode<E> oldPNode = lpProducerNode();
        oldPNode.soNext(head);
        spProducerNode(tail);
        return limit;
    }

[hl845740757]

MpscLinkedQueue#fill

    @Override
    public int fill(Supplier<E> s, int limit)
    {
        if (null == s)
            throw new IllegalArgumentException("supplier is null");
        if (limit < 0)
            throw new IllegalArgumentException("limit is negative:" + limit);
        if (limit == 0)
            return 0;

        LinkedQueueNode<E> tail = newNode(s.get());
        final LinkedQueueNode<E> head = tail;
        for (int i = 1; i < limit; i++)
        {
            final LinkedQueueNode<E> temp = newNode(s.get());
            // The fence here may be redundant
            tail.soNext(temp);
            tail = temp;
        }

        final LinkedQueueNode<E> oldPNode = xchgProducerNode(tail);
        oldPNode.soNext(head);
        return limit;
    }

[nitsanw]

The redundant fences are potentially costly on ARM processors. PR is welcome :-)

[hl845740757]

#299 pull request

[hl845740757]

#307 pull request