app/test-pmd/macswap_sse.h

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2018 Intel Corporation
   3  */
   4
   5 #ifndef _MACSWAP_SSE_H_
   6 #define _MACSWAP_SSE_H_
   7
   8 #include "macswap_common.h"
   9
  10 static inline void
  11 do_macswap(struct rte_mbuf *pkts[], uint16_t nb,
  12                 struct rte_port *txp)
  13 {
  14         struct rte_ether_hdr *eth_hdr[4];
  15         struct rte_mbuf *mb[4];
  16         uint64_t ol_flags;
  17         int i;
  18         int r;
  19         __m128i addr0, addr1, addr2, addr3;
  20         /**
  21          * shuffle mask be used to shuffle the 16 bytes.
  22          * byte 0-5 wills be swapped with byte 6-11.
  23          * byte 12-15 will keep unchanged.
  24          */
  25         __m128i shfl_msk = _mm_set_epi8(15, 14, 13, 12,
  26                                         5, 4, 3, 2,
  27                                         1, 0, 11, 10,
  28                                         9, 8, 7, 6);
  29
  30         ol_flags = ol_flags_init(txp->dev_conf.txmode.offloads);
  31         vlan_qinq_set(pkts, nb, ol_flags,
  32                         txp->tx_vlan_id, txp->tx_vlan_id_outer);
  33
  34         i = 0;
  35         r = nb;
  36
  37         while (r >= 4) {
  38                 if (r >= 8) {
  39                         rte_prefetch0(rte_pktmbuf_mtod(pkts[i + 4], void *));
  40                         rte_prefetch0(rte_pktmbuf_mtod(pkts[i + 5], void *));
  41                         rte_prefetch0(rte_pktmbuf_mtod(pkts[i + 6], void *));
  42                         rte_prefetch0(rte_pktmbuf_mtod(pkts[i + 7], void *));
  43                 }
  44
  45                 mb[0] = pkts[i++];
  46                 eth_hdr[0] = rte_pktmbuf_mtod(mb[0], struct rte_ether_hdr *);
  47                 addr0 = _mm_loadu_si128((__m128i *)eth_hdr[0]);
  48
  49                 mb[1] = pkts[i++];
  50                 eth_hdr[1] = rte_pktmbuf_mtod(mb[1], struct rte_ether_hdr *);
  51                 addr1 = _mm_loadu_si128((__m128i *)eth_hdr[1]);
  52
  53
  54                 mb[2] = pkts[i++];
  55                 eth_hdr[2] = rte_pktmbuf_mtod(mb[2], struct rte_ether_hdr *);
  56                 addr2 = _mm_loadu_si128((__m128i *)eth_hdr[2]);
  57
  58                 mb[3] = pkts[i++];
  59                 eth_hdr[3] = rte_pktmbuf_mtod(mb[3], struct rte_ether_hdr *);
  60                 addr3 = _mm_loadu_si128((__m128i *)eth_hdr[3]);
  61
  62                 addr0 = _mm_shuffle_epi8(addr0, shfl_msk);
  63                 addr1 = _mm_shuffle_epi8(addr1, shfl_msk);
  64                 addr2 = _mm_shuffle_epi8(addr2, shfl_msk);
  65                 addr3 = _mm_shuffle_epi8(addr3, shfl_msk);
  66
  67                 _mm_storeu_si128((__m128i *)eth_hdr[0], addr0);
  68                 _mm_storeu_si128((__m128i *)eth_hdr[1], addr1);
  69                 _mm_storeu_si128((__m128i *)eth_hdr[2], addr2);
  70                 _mm_storeu_si128((__m128i *)eth_hdr[3], addr3);
  71
  72                 mbuf_field_set(mb[0], ol_flags);
  73                 mbuf_field_set(mb[1], ol_flags);
  74                 mbuf_field_set(mb[2], ol_flags);
  75                 mbuf_field_set(mb[3], ol_flags);
  76                 r -= 4;
  77         }
  78
  79         for ( ; i < nb; i++) {
  80                 if (i < nb - 1)
  81                         rte_prefetch0(rte_pktmbuf_mtod(pkts[i+1], void *));
  82                 mb[0] = pkts[i];
  83                 eth_hdr[0] = rte_pktmbuf_mtod(mb[0], struct rte_ether_hdr *);
  84
  85                 /* Swap dest and src mac addresses. */
  86                 addr0 = _mm_loadu_si128((__m128i *)eth_hdr[0]);
  87                 addr0 = _mm_shuffle_epi8(addr0, shfl_msk);
  88                 _mm_storeu_si128((__m128i *)eth_hdr[0], addr0);
  89
  90                 mbuf_field_set(mb[0], ol_flags);
  91         }
  92 }
  93
  94 #endif /* _MACSWAP_SSE_H_ */