DPDK  16.04.0
rte_lpm_neon.h
1 /*-
2  * BSD LICENSE
3  *
4  * Copyright(c) 2015 Cavium Networks. All rights reserved.
5  * All rights reserved.
6  *
7  * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
8  * All rights reserved.
9  *
10  * Derived rte_lpm_lookupx4 implementation from lib/librte_lpm/rte_lpm_sse.h
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  *
16  * * Redistributions of source code must retain the above copyright
17  * notice, this list of conditions and the following disclaimer.
18  * * Redistributions in binary form must reproduce the above copyright
19  * notice, this list of conditions and the following disclaimer in
20  * the documentation and/or other materials provided with the
21  * distribution.
22  * * Neither the name of Cavium Networks nor the names of its
23  * contributors may be used to endorse or promote products derived
24  * from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
27  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
28  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
29  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
30  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
31  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
32  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
33  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
34  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
35  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
36  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 #ifndef _RTE_LPM_NEON_H_
40 #define _RTE_LPM_NEON_H_
41 
42 #include <rte_branch_prediction.h>
43 #include <rte_byteorder.h>
44 #include <rte_common.h>
45 #include <rte_vect.h>
46 
47 #ifdef __cplusplus
48 extern "C" {
49 #endif
50 
51 static inline void
52 rte_lpm_lookupx4(const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4],
53  uint32_t defv)
54 {
55  uint32x4_t i24;
56  rte_xmm_t i8;
57  uint32_t tbl[4];
58  uint64_t idx, pt, pt2;
59  const uint32_t *ptbl;
60 
61  const uint32_t mask = UINT8_MAX;
62  const int32x4_t mask8 = vdupq_n_s32(mask);
63 
64  /*
65  * RTE_LPM_VALID_EXT_ENTRY_BITMASK for 2 LPM entries
66  * as one 64-bit value (0x0300000003000000).
67  */
68  const uint64_t mask_xv =
69  ((uint64_t)RTE_LPM_VALID_EXT_ENTRY_BITMASK |
70  (uint64_t)RTE_LPM_VALID_EXT_ENTRY_BITMASK << 32);
71 
72  /*
73  * RTE_LPM_LOOKUP_SUCCESS for 2 LPM entries
74  * as one 64-bit value (0x0100000001000000).
75  */
76  const uint64_t mask_v =
77  ((uint64_t)RTE_LPM_LOOKUP_SUCCESS |
78  (uint64_t)RTE_LPM_LOOKUP_SUCCESS << 32);
79 
80  /* get 4 indexes for tbl24[]. */
81  i24 = vshrq_n_u32((uint32x4_t)ip, CHAR_BIT);
82 
83  /* extract values from tbl24[] */
84  idx = vgetq_lane_u64((uint64x2_t)i24, 0);
85 
86  ptbl = (const uint32_t *)&lpm->tbl24[(uint32_t)idx];
87  tbl[0] = *ptbl;
88  ptbl = (const uint32_t *)&lpm->tbl24[idx >> 32];
89  tbl[1] = *ptbl;
90 
91  idx = vgetq_lane_u64((uint64x2_t)i24, 1);
92 
93  ptbl = (const uint32_t *)&lpm->tbl24[(uint32_t)idx];
94  tbl[2] = *ptbl;
95  ptbl = (const uint32_t *)&lpm->tbl24[idx >> 32];
96  tbl[3] = *ptbl;
97 
98  /* get 4 indexes for tbl8[]. */
99  i8.x = vandq_s32(ip, mask8);
100 
101  pt = (uint64_t)tbl[0] |
102  (uint64_t)tbl[1] << 32;
103  pt2 = (uint64_t)tbl[2] |
104  (uint64_t)tbl[3] << 32;
105 
106  /* search successfully finished for all 4 IP addresses. */
107  if (likely((pt & mask_xv) == mask_v) &&
108  likely((pt2 & mask_xv) == mask_v)) {
109  *(uint64_t *)hop = pt & RTE_LPM_MASKX4_RES;
110  *(uint64_t *)(hop + 2) = pt2 & RTE_LPM_MASKX4_RES;
111  return;
112  }
113 
114  if (unlikely((pt & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
115  RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
116  i8.u32[0] = i8.u32[0] +
117  (uint8_t)tbl[0] * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
118  ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[0]];
119  tbl[0] = *ptbl;
120  }
121  if (unlikely((pt >> 32 & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
122  RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
123  i8.u32[1] = i8.u32[1] +
124  (uint8_t)tbl[1] * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
125  ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[1]];
126  tbl[1] = *ptbl;
127  }
128  if (unlikely((pt2 & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
129  RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
130  i8.u32[2] = i8.u32[2] +
131  (uint8_t)tbl[2] * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
132  ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[2]];
133  tbl[2] = *ptbl;
134  }
135  if (unlikely((pt2 >> 32 & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
136  RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
137  i8.u32[3] = i8.u32[3] +
138  (uint8_t)tbl[3] * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
139  ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[3]];
140  tbl[3] = *ptbl;
141  }
142 
143  hop[0] = (tbl[0] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[0] & 0x00FFFFFF : defv;
144  hop[1] = (tbl[1] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[1] & 0x00FFFFFF : defv;
145  hop[2] = (tbl[2] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[2] & 0x00FFFFFF : defv;
146  hop[3] = (tbl[3] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[3] & 0x00FFFFFF : defv;
147 }
148 
149 #ifdef __cplusplus
150 }
151 #endif
152 
153 #endif /* _RTE_LPM_NEON_H_ */