On 25 May 2016 at 20:29, Jerin Jacob <jerin.jacob at caviumnetworks.com> wrote:
> On Fri, May 06, 2016 at 11:55:46AM +0530, Jianbo Liu wrote:
>> use ARM NEON intrinsic to implement ixgbe vPMD
>>
>> Signed-off-by: Jianbo Liu <jianbo.liu at linaro.org>
>> ---
>> drivers/net/ixgbe/Makefile | 4 +
>> drivers/net/ixgbe/ixgbe_rxtx_vec_neon.c | 561
>> ++++++++++++++++++++++++++++++++
>> 2 files changed, 565 insertions(+)
>> create mode 100644 drivers/net/ixgbe/ixgbe_rxtx_vec_neon.c
>> + /* Read desc statuses backwards to avoid race condition */
>> + /* A.1 load 4 pkts desc */
>> + descs[3] = vld1q_u64((uint64_t *)(rxdp + 3));
>> + rte_rmb();
>
> Any specific reason to add rte_rmb() here, If there is no performance
> drop then it makes sense to add before descs[3] uses it.i.e
> at rte_compiler_barrier() place in x86 code.
>
To avoid desc statuses inconsistent since they are read backwards.
>> +
>> + /* B.2 copy 2 mbuf point into rx_pkts */
>> + vst1q_u64((uint64_t *)&rx_pkts[pos], mbp1);
>> +
>> + /* B.1 load 1 mbuf point */
>> + mbp2 = vld1q_u64((uint64_t *)&sw_ring[pos + 2]);
>> +
>> + descs[2] = vld1q_u64((uint64_t *)(rxdp + 2));
>> + /* B.1 load 2 mbuf point */
>> + descs[1] = vld1q_u64((uint64_t *)(rxdp + 1));
>> + descs[0] = vld1q_u64((uint64_t *)(rxdp));
>> +
>> + /* B.2 copy 2 mbuf point into rx_pkts */
>> + vst1q_u64((uint64_t *)&rx_pkts[pos + 2], mbp2);
>> +
>> + if (split_packet) {
>> + rte_prefetch_non_temporal(&rx_pkts[pos]->cacheline1);
>> + rte_prefetch_non_temporal(&rx_pkts[pos+1]->cacheline1);
>> + rte_prefetch_non_temporal(&rx_pkts[pos+2]->cacheline1);
>> + rte_prefetch_non_temporal(&rx_pkts[pos+3]->cacheline1);
>
> replace with rte_mbuf_prefetch_part2 or equivalent
>
rte_mbuf_prefetch_part2 is new functions after this patchset, so it's
better to submit a new patch as Bruce said.
