1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2015 Cavium, Inc.
3 * Copyright(c) 2010-2014 Intel Corporation.
4 */
5
6 #ifndef _RTE_LPM_NEON_H_
7 #define _RTE_LPM_NEON_H_
8
9 #include <rte_branch_prediction.h>
10 #include <rte_byteorder.h>
11 #include <rte_common.h>
12 #include <rte_vect.h>
13
14 #ifdef __cplusplus
15 extern "C" {
16 #endif
17
18 static inline void
rte_lpm_lookupx4(const struct rte_lpm * lpm,xmm_t ip,uint32_t hop[4],uint32_t defv)19 rte_lpm_lookupx4(const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4],
20 uint32_t defv)
21 {
22 uint32x4_t i24;
23 rte_xmm_t i8;
24 uint32_t tbl[4];
25 uint64_t idx, pt, pt2;
26 const uint32_t *ptbl;
27
28 const uint32_t mask = UINT8_MAX;
29 const int32x4_t mask8 = vdupq_n_s32(mask);
30
31 /*
32 * RTE_LPM_VALID_EXT_ENTRY_BITMASK for 2 LPM entries
33 * as one 64-bit value (0x0300000003000000).
34 */
35 const uint64_t mask_xv =
36 ((uint64_t)RTE_LPM_VALID_EXT_ENTRY_BITMASK |
37 (uint64_t)RTE_LPM_VALID_EXT_ENTRY_BITMASK << 32);
38
39 /*
40 * RTE_LPM_LOOKUP_SUCCESS for 2 LPM entries
41 * as one 64-bit value (0x0100000001000000).
42 */
43 const uint64_t mask_v =
44 ((uint64_t)RTE_LPM_LOOKUP_SUCCESS |
45 (uint64_t)RTE_LPM_LOOKUP_SUCCESS << 32);
46
47 /* get 4 indexes for tbl24[]. */
48 i24 = vshrq_n_u32((uint32x4_t)ip, CHAR_BIT);
49
50 /* extract values from tbl24[] */
51 idx = vgetq_lane_u64((uint64x2_t)i24, 0);
52
53 ptbl = (const uint32_t *)&lpm->tbl24[(uint32_t)idx];
54 tbl[0] = *ptbl;
55 ptbl = (const uint32_t *)&lpm->tbl24[idx >> 32];
56 tbl[1] = *ptbl;
57
58 idx = vgetq_lane_u64((uint64x2_t)i24, 1);
59
60 ptbl = (const uint32_t *)&lpm->tbl24[(uint32_t)idx];
61 tbl[2] = *ptbl;
62 ptbl = (const uint32_t *)&lpm->tbl24[idx >> 32];
63 tbl[3] = *ptbl;
64
65 /* get 4 indexes for tbl8[]. */
66 i8.x = vandq_s32(ip, mask8);
67
68 pt = (uint64_t)tbl[0] |
69 (uint64_t)tbl[1] << 32;
70 pt2 = (uint64_t)tbl[2] |
71 (uint64_t)tbl[3] << 32;
72
73 /* search successfully finished for all 4 IP addresses. */
74 if (likely((pt & mask_xv) == mask_v) &&
75 likely((pt2 & mask_xv) == mask_v)) {
76 *(uint64_t *)hop = pt & RTE_LPM_MASKX4_RES;
77 *(uint64_t *)(hop + 2) = pt2 & RTE_LPM_MASKX4_RES;
78 return;
79 }
80
81 if (unlikely((pt & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
82 RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
83 i8.u32[0] = i8.u32[0] +
84 (tbl[0] & 0x00FFFFFF) * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
85 ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[0]];
86 tbl[0] = *ptbl;
87 }
88 if (unlikely((pt >> 32 & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
89 RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
90 i8.u32[1] = i8.u32[1] +
91 (tbl[1] & 0x00FFFFFF) * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
92 ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[1]];
93 tbl[1] = *ptbl;
94 }
95 if (unlikely((pt2 & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
96 RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
97 i8.u32[2] = i8.u32[2] +
98 (tbl[2] & 0x00FFFFFF) * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
99 ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[2]];
100 tbl[2] = *ptbl;
101 }
102 if (unlikely((pt2 >> 32 & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
103 RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
104 i8.u32[3] = i8.u32[3] +
105 (tbl[3] & 0x00FFFFFF) * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
106 ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[3]];
107 tbl[3] = *ptbl;
108 }
109
110 hop[0] = (tbl[0] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[0] & 0x00FFFFFF : defv;
111 hop[1] = (tbl[1] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[1] & 0x00FFFFFF : defv;
112 hop[2] = (tbl[2] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[2] & 0x00FFFFFF : defv;
113 hop[3] = (tbl[3] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[3] & 0x00FFFFFF : defv;
114 }
115
116 #ifdef __cplusplus
117 }
118 #endif
119
120 #endif /* _RTE_LPM_NEON_H_ */
121