C++ prefetch函数代码示例

static char *url_buff_gets(URL url, char *buff, int maxsiz)
{
    URL_buff *urlp = (URL_buff *)url;
    int c, r, w;
    long len, maxlen;
    int newline = url_newline_code;
    unsigned char *bp;

    if(urlp->eof)
	return NULL;

    maxlen = maxsiz - 1;
    if(maxlen == 0)
	*buff = '\0';
    if(maxlen <= 0)
	return buff;
    len = 0;
    r = urlp->rp;
    w = urlp->wp;
    bp = urlp->buffer;

    do
    {
	if(r == w)
	{
	    urlp->wp = w;
	    prefetch(urlp);
	    w = urlp->wp;
	    if(r == w)
	    {
		urlp->eof = 1;
		if(len == 0)
		    return NULL;
		buff[len] = '\0';
		urlp->pos += len;
		urlp->rp = r;
		return buff;
	    }
	}
	c = bp[r];
	buff[len++] = c;
	r = ((r + 1) & BASEMASK);
    } while(c != newline && len < maxlen);
    buff[len] = '\0';
    urlp->pos += len;
    urlp->rp = r;
    return buff;
}

/*
 * Invoke the RCU callbacks on the specified rcu_ctrlkblk structure
 * whose grace period has elapsed.
 */
static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp)
{
	const char *rn = NULL;
	struct rcu_head *next, *list;
	unsigned long flags;
	RCU_TRACE(int cb_count = 0);

	/* If no RCU callbacks ready to invoke, just return. */
	if (&rcp->rcucblist == rcp->donetail) {
		RCU_TRACE(trace_rcu_batch_start(rcp->name, 0, 0, -1));
		RCU_TRACE(trace_rcu_batch_end(rcp->name, 0,
!!ACCESS_ONCE(rcp->rcucblist),
need_resched(),
is_idle_task(current),
false));
		return;
	}

	/* Move the ready-to-invoke callbacks to a local list. */
	local_irq_save(flags);
	RCU_TRACE(trace_rcu_batch_start(rcp->name, 0, rcp->qlen, -1));
	list = rcp->rcucblist;
	rcp->rcucblist = *rcp->donetail;
	*rcp->donetail = NULL;
	if (rcp->curtail == rcp->donetail) {
		rcp->curtail = &rcp->rcucblist;
	}
	rcp->donetail = &rcp->rcucblist;
	local_irq_restore(flags);

	/* Invoke the callbacks on the local list. */
	RCU_TRACE(rn = rcp->name);
	while (list) {
		next = list->next;
		prefetch(next);
		debug_rcu_head_unqueue(list);
		local_bh_disable();
		__rcu_reclaim(rn, list);
		local_bh_enable();
		list = next;
		RCU_TRACE(cb_count++);
	}
	RCU_TRACE(rcu_trace_sub_qlen(rcp, cb_count));
	RCU_TRACE(trace_rcu_batch_end(rcp->name,
cb_count, 0, need_resched(),
is_idle_task(current),
false));
}

static long url_buff_read(URL url, void *buff, long n)
{
    URL_buff *urlp = (URL_buff *)url;
    char *s = (char *)buff;
    int r, i, j;

    if(urlp->eof)
	return 0;

    r = urlp->rp;
    if(r == urlp->wp)
    {
	prefetch(urlp);
	if(r == urlp->wp)
	{
	    urlp->eof = 1;
	    return EOF;
	}
    }

    /* first fragment */
    i = urlp->wp - r;
    if(i < 0)
	i = BASESIZE - r;
    if(i > n)
	i = n;
    memcpy(s, urlp->buffer + r, i);
    r = ((r + i) & BASEMASK);

    if(i == n || r == urlp->wp || r != 0)
    {
	urlp->rp = r;
	urlp->pos += i;
	return i;
    }

    /* second fragment */
    j = urlp->wp;
    n -= i;
    s += i;
    if(j > n)
	j = n;
    memcpy(s, urlp->buffer, j);
    urlp->rp = j;
    urlp->pos += i + j;

    return i + j;
}

static int write_packet(struct imx_ep_struct *imx_ep, struct imx_request *req)
{
	u8	*buf;
	int	length, count, temp;

	if (unlikely(__raw_readl(imx_ep->imx_usb->base +
				 USB_EP_STAT(EP_NO(imx_ep))) & EPSTAT_ZLPS)) {
		D_TRX(imx_ep->imx_usb->dev, "<%s> zlp still queued in EP %s\n",
			__func__, imx_ep->ep.name);
		return -1;
	}

	buf = req->req.buf + req->req.actual;
	prefetch(buf);

	length = min(req->req.length - req->req.actual, (u32)imx_ep->fifosize);

	if (imx_fifo_bcount(imx_ep) + length > imx_ep->fifosize) {
		D_TRX(imx_ep->imx_usb->dev, "<%s> packet overfill %s fifo\n",
			__func__, imx_ep->ep.name);
		return -1;
	}

	req->req.actual += length;
	count = length;

	if (!count && req->req.zero) {	/* zlp */
		temp = __raw_readl(imx_ep->imx_usb->base
			+ USB_EP_STAT(EP_NO(imx_ep)));
		__raw_writel(temp | EPSTAT_ZLPS, imx_ep->imx_usb->base
			+ USB_EP_STAT(EP_NO(imx_ep)));
		D_TRX(imx_ep->imx_usb->dev, "<%s> zero packet\n", __func__);
		return 0;
	}

	while (count--) {
		if (count == 0) {	/* last byte */
			temp = __raw_readl(imx_ep->imx_usb->base
				+ USB_EP_FCTRL(EP_NO(imx_ep)));
			__raw_writel(temp | FCTRL_WFR, imx_ep->imx_usb->base
				+ USB_EP_FCTRL(EP_NO(imx_ep)));
		}
		__raw_writeb(*buf++,
			imx_ep->imx_usb->base + USB_EP_FDAT0(EP_NO(imx_ep)));
	}

	return length;
}

void AudioPrefetch::seek(unsigned seekTo)
{
    // printf("seek %d\n", seekTo);
#ifdef AUDIOPREFETCH_DEBUG
    printf("AudioPrefetch::seek to:%u seekCount:%d\n", seekTo, seekCount);
#endif

    // Speedup: More than one seek message pending?
    // Eat up seek messages until we get to the very LATEST one,
    //  because all the rest which came before it are irrelevant now,
    //  and processing them all was taking extreme time, especially with
    //  resampling enabled.
    // In particular, when the user 'slides' the play cursor back and forth
    //  there are MANY seek messages in the pipe, and with resampling enabled
    //  it was taking minutes to finish seeking. If the user hit play during that time,
    //  things were messed up (FIFO underruns, choppy intermittent sound etc).
    // Added by Tim. p3.3.20
    if (seekCount > 1)
    {
        --seekCount;
        return;
    }

    writePos = seekTo;

    bool isFirstPrefetch = true;
    for (unsigned int i = 0; i < (fifoLength) - 1; ++i)//prevent compiler warning: comparison of signed/unsigned
    {
        // Indicate do a seek command before read, but only on the first pass.
        // Changed by Tim. p3.3.17
        //prefetch();
        prefetch(isFirstPrefetch);

        isFirstPrefetch = false;

        // To help speed things up even more, check the count again. Return if more seek messages are pending.
        // Added by Tim. p3.3.20
        if (seekCount > 1)
        {
            --seekCount;
            return;
        }
    }

    seekPos = seekTo;
    //seekDone = true;
    --seekCount;
}

static irqreturn_t bnxt_qplib_creq_irq(int irq, void *dev_instance)
{
	struct bnxt_qplib_rcfw *rcfw = dev_instance;
	struct bnxt_qplib_hwq *creq = &rcfw->creq;
	struct creq_base **creq_ptr;
	u32 sw_cons;

	/* Prefetch the CREQ element */
	sw_cons = HWQ_CMP(creq->cons, creq);
	creq_ptr = (struct creq_base **)rcfw->creq.pbl_ptr;
	prefetch(&creq_ptr[get_creq_pg(sw_cons)][get_creq_idx(sw_cons)]);

	tasklet_schedule(&rcfw->worker);

	return IRQ_HANDLED;
}

void GenomeSequence::setup(const char *referenceFilename)
{
    setReferenceName(referenceFilename);

    if (_progressStream) *_progressStream << "open and prefetch reference genome " << referenceFilename << ": " << std::flush;

    if (open(false))
    {
        std::cerr << "Failed to open reference genome " << referenceFilename << std::endl;
        std::cerr << errorStr << std::endl;
        exit(1);
    }

    prefetch();
    if (_progressStream) *_progressStream << "done." << std::endl << std::flush;
}

/* Handle a received packet.  Second half: Touches packet payload. */
void __efx_rx_packet(struct efx_channel *channel,
		     struct efx_rx_buffer *rx_buf, bool checksummed)
{
	struct efx_nic *efx = channel->efx;
	struct sk_buff *skb;

	/* If we're in loopback test, then pass the packet directly to the
	 * loopback layer, and free the rx_buf here
	 */
	if (unlikely(efx->loopback_selftest)) {
		efx_loopback_rx_packet(efx, rx_buf->data, rx_buf->len);
		efx_free_rx_buffer(efx, rx_buf);
		return;
	}

	if (rx_buf->skb) {
		prefetch(skb_shinfo(rx_buf->skb));

		skb_put(rx_buf->skb, rx_buf->len);

		/* Move past the ethernet header. rx_buf->data still points
		 * at the ethernet header */
		rx_buf->skb->protocol = eth_type_trans(rx_buf->skb,
						       efx->net_dev);

		skb_record_rx_queue(rx_buf->skb, channel->channel);
	}

	if (likely(checksummed || rx_buf->page)) {
		efx_rx_packet_lro(channel, rx_buf, checksummed);
		return;
	}

	/* We now own the SKB */
	skb = rx_buf->skb;
	rx_buf->skb = NULL;
	EFX_BUG_ON_PARANOID(!skb);

	/* Set the SKB flags */
	skb->ip_summed = CHECKSUM_NONE;

	/* Pass the packet up */
	netif_receive_skb(skb);

	/* Update allocation strategy method */
	channel->rx_alloc_level += RX_ALLOC_FACTOR_SKB;
}

/*
 * sfe_cm_recv()
 *	Handle packet receives.
 *
 * Returns 1 if the packet is forwarded or 0 if it isn't.
 */
int sfe_cm_recv(struct sk_buff *skb)
{
	struct net_device *dev;
#if (SFE_HOOK_ABOVE_BRIDGE)
	struct in_device *in_dev;
#endif

	/*
	 * We know that for the vast majority of packets we need the transport
	 * layer header so we may as well start to fetch it now!
	 */
	prefetch(skb->data + 32);
	barrier();

	dev = skb->dev;

#if (SFE_HOOK_ABOVE_BRIDGE)
	/*
	 * Does our input device support IP processing?
	 */
	in_dev = (struct in_device *)dev->ip_ptr;
	if (unlikely(!in_dev)) {
		DEBUG_TRACE("no IP processing for device: %s\n", dev->name);
		return 0;
	}

	/*
	 * Does it have an IP address?  If it doesn't then we can't do anything
	 * interesting here!
	 */
	if (unlikely(!in_dev->ifa_list)) {
		DEBUG_TRACE("no IP address for device: %s\n", dev->name);
		return 0;
	}
#endif

	/*
	 * We're only interested in IP packets.
	 */	
	if (likely(htons(ETH_P_IP) == skb->protocol)) {
		return sfe_ipv4_recv(dev, skb);
	}

	DEBUG_TRACE("not IP packet\n");
	return 0;
}

static void isp1362_write_ptd(struct isp1362_hcd *isp1362_hcd, struct isp1362_ep *ep,
			      struct isp1362_ep_queue *epq)
{
	struct ptd *ptd = &ep->ptd;
	int len = PTD_GET_DIR(ptd) == PTD_DIR_IN ? 0 : ep->length;

	_BUG_ON(ep->ptd_offset < 0);

	prefetch(ptd);
	isp1362_write_buffer(isp1362_hcd, ptd, ep->ptd_offset, PTD_HEADER_SIZE);
	if (len)
		isp1362_write_buffer(isp1362_hcd, ep->data,
				     ep->ptd_offset + PTD_HEADER_SIZE, len);

	dump_ptd(ptd);
	dump_ptd_out_data(ptd, ep->data);
}

int wait_packet_function_ptr(void *data, int mode) {
  struct e1000_adapter *adapter = (struct e1000_adapter*)data;

  if(unlikely(enable_debug)) printk("[wait_packet_function_ptr] called [mode=%d]\n", mode);

  if(mode == 1) {
    struct e1000_ring *rx_ring = adapter->rx_ring;
    union e1000_rx_desc_extended *rx_desc;

    u16 i = E1000_READ_REG(&adapter->hw, E1000_RDT(0));
    
    /* Very important: update the value from the register set from userland.
     * Here i is the last I've read (zero-copy implementation) */
    if(++i == rx_ring->count) i = 0;
    /* Here i is the next I have to read */
    
    rx_ring->next_to_clean = i;

    rx_desc = E1000_RX_DESC_EXT(*rx_ring, rx_ring->next_to_clean);
    if(unlikely(enable_debug)) printk("[wait_packet_function_ptr] Check if a packet is arrived\n");

    prefetch(rx_desc);

    if(!(le32_to_cpu(rx_desc->wb.upper.status_error) & E1000_RXD_STAT_DD)) {
      adapter->dna.interrupt_received = 0;

#if 0
      if(!adapter->dna.interrupt_enabled) {
	e1000_irq_enable(adapter), adapter->dna.interrupt_enabled = 1;
	if(unlikely(enable_debug)) printk("[wait_packet_function_ptr] Packet not arrived yet: enabling interrupts\n");
      }
#endif
    } else
      adapter->dna.interrupt_received = 1;

    return(le32_to_cpu(rx_desc->wb.upper.status_error) & E1000_RXD_STAT_DD);
  } else {
    if(adapter->dna.interrupt_enabled) {
      e1000_irq_disable(adapter);
      adapter->dna.interrupt_enabled = 0;
      if(unlikely(enable_debug)) printk("[wait_packet_function_ptr] Disabled interrupts\n");
    }
    return(0);
  }
}

/*
 *	s3c2410_udc_write_packet
 */
static inline int s3c2410_udc_write_packet(int fifo,
		struct s3c2410_request *req,
		unsigned max)
{
	unsigned len = min(req->req.length - req->req.actual, max);
	u8 *buf = req->req.buf + req->req.actual;

	prefetch(buf);

	dprintk(DEBUG_VERBOSE, "%s %d %d %d %d\n", __func__,
		req->req.actual, req->req.length, len, req->req.actual + len);

	req->req.actual += len;

	udelay(5);
	writesb(base_addr + fifo, buf, len);
	return len;
}

        void siftDownSingleStep(ssize_t const end, ssize_t const root) {
            ssize_t const left = root * 2;
            ssize_t const right = left + 1;

            if (right <= end) {
                ssize_t const maxChild = root * 2
                        + compOp(a[left], Below, a[right]);
                if (compOp(a[root], Below, a[maxChild])) {
                    std::swap(a[root], a[maxChild]);
                    queue[queueStoreIndex] = maxChild;
                    queueStoreIndex++;
                    prefetch(a + std::min(maxChild * 2, end));
                }
            } else {
                if (left == end && compOp(a[root], Below, a[left])) {
                    std::swap(a[root], a[left]);
                }
            }
        }

int
pcap_dispatch(pcap_t *p, int cnt, pcap_handler callback, u_char *user)
{
	struct pcap_ring *pme = p;
	struct my_ring *me = &pme->me;
	int got = 0;
	u_int si;

	ND("cnt %d", cnt);
	if (cnt == 0)
		cnt = -1;
	/* scan all rings */
	for (si = me->begin; si < me->end; si++) {
		struct netmap_ring *ring = NETMAP_RXRING(me->nifp, si);
		ND("ring has %d pkts", ring->avail);
		if (ring->avail == 0)
			continue;
		pme->hdr.ts = ring->ts;
		/*
		 * XXX a proper prefetch should be done as
		 *	prefetch(i); callback(i-1); ...
		 */
		while ((cnt == -1 || cnt != got) && ring->avail > 0) {
			u_int i = ring->cur;
			u_int idx = ring->slot[i].buf_idx;
			if (idx < 2) {
				D("%s bogus RX index %d at offset %d",
					me->nifp->ni_name, idx, i);
				sleep(2);
			}
			u_char *buf = (u_char *)NETMAP_BUF(ring, idx);
			prefetch(buf);
			pme->hdr.len = pme->hdr.caplen = ring->slot[i].len;
			// D("call %p len %d", p, me->hdr.len);
			callback(user, &pme->hdr, buf);
			ring->cur = NETMAP_RING_NEXT(ring, i);
			ring->avail--;
			got++;
		}
	}
	pme->st.ps_recv += got;
	return got;
}

/**
 * this function polls the CQ, and extracts the needed fields
 * upon CQE error state it will return -1
 * if a bad checksum packet or a filler bit it will return VMA_MP_RQ_BAD_PACKET
 */
int cq_mgr_mp::poll_mp_cq(uint16_t &size, uint32_t &strides_used,
			  uint32_t &flags, struct mlx5_cqe64 *&out_cqe64)
{
	struct mlx5_cqe64 *cqe= check_cqe();
	if (likely(cqe)) {
		if (unlikely(MLX5_CQE_OPCODE(cqe->op_own) != MLX5_CQE_RESP_SEND)) {
			cq_logdbg("Warning op_own is %x", MLX5_CQE_OPCODE(cqe->op_own));
			// optimize checks in ring by setting size non zero
			if (MLX5_CQE_OPCODE(cqe->op_own) == MLX5_CQE_RESP_ERR) {
				cq_logdbg("poll_length, CQE response error, "
					 "syndrome=0x%x, vendor syndrome error=0x%x, "
					 "HW syndrome 0x%x, HW syndrome type 0x%x\n",
					 ((struct mlx5_err_cqe *)cqe)->syndrome,
					 ((struct mlx5_err_cqe *)cqe)->vendor_err_synd,
					 ((struct mlx5_err_cqe *)cqe)->hw_err_synd,
					 ((struct mlx5_err_cqe *)cqe)->hw_synd_type);
			}
			size = 1;
			m_p_cq_stat->n_rx_pkt_drop++;
			return -1;
		}
		m_p_cq_stat->n_rx_pkt_drop += cqe->sop_qpn.sop;
		out_cqe64 = cqe;
		uint32_t stride_byte_cnt = ntohl(cqe->byte_cnt);
		strides_used = (stride_byte_cnt & MP_RQ_NUM_STRIDES_FIELD_MASK) >>
				MP_RQ_NUM_STRIDES_FIELD_SHIFT;
		flags = (!!(cqe->hds_ip_ext & MLX5_CQE_L4_OK) * IBV_EXP_CQ_RX_TCP_UDP_CSUM_OK) |
			(!!(cqe->hds_ip_ext & MLX5_CQE_L3_OK) * IBV_EXP_CQ_RX_IP_CSUM_OK);
		if (likely(flags == UDP_OK_FLAGS)) {
			size = stride_byte_cnt & MP_RQ_BYTE_CNT_FIELD_MASK;
		} else {
			// if CSUM is bad it can be either filler or bad packet
			flags = VMA_MP_RQ_BAD_PACKET;
			size = 1;
			if (stride_byte_cnt & MP_RQ_FILLER_FIELD_MASK) {
				m_p_cq_stat->n_rx_pkt_drop++;
			}
		}
		++m_mlx5_cq.cq_ci;
		prefetch((uint8_t*)m_mlx5_cq.cq_buf + ((m_mlx5_cq.cq_ci & (m_mlx5_cq.cqe_count - 1)) << m_mlx5_cq.cqe_size_log));
	} else {