1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2014 Intel Corporation
5 #ifndef _RTE_LPM_SSE_H_
6 #define _RTE_LPM_SSE_H_
8 #include <rte_branch_prediction.h>
9 #include <rte_byteorder.h>
10 #include <rte_common.h>
19 rte_lpm_lookupx4(const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4],
25 uint64_t idx, pt, pt2;
29 _mm_set_epi32(UINT8_MAX, UINT8_MAX, UINT8_MAX, UINT8_MAX);
32 * RTE_LPM_VALID_EXT_ENTRY_BITMASK for 2 LPM entries
33 * as one 64-bit value (0x0300000003000000).
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);
40 * RTE_LPM_LOOKUP_SUCCESS for 2 LPM entries
41 * as one 64-bit value (0x0100000001000000).
43 const uint64_t mask_v =
44 ((uint64_t)RTE_LPM_LOOKUP_SUCCESS |
45 (uint64_t)RTE_LPM_LOOKUP_SUCCESS << 32);
47 /* get 4 indexes for tbl24[]. */
48 i24 = _mm_srli_epi32(ip, CHAR_BIT);
50 /* extract values from tbl24[] */
51 idx = _mm_cvtsi128_si64(i24);
52 /* With -O0 option, gcc 4.8 - 5.4 fails to fold sizeof() into a constant */
53 i24 = _mm_srli_si128(i24, /* sizeof(uint64_t) */ 8);
55 ptbl = (const uint32_t *)&lpm->tbl24[(uint32_t)idx];
57 ptbl = (const uint32_t *)&lpm->tbl24[idx >> 32];
60 idx = _mm_cvtsi128_si64(i24);
62 ptbl = (const uint32_t *)&lpm->tbl24[(uint32_t)idx];
64 ptbl = (const uint32_t *)&lpm->tbl24[idx >> 32];
67 /* get 4 indexes for tbl8[]. */
68 i8.x = _mm_and_si128(ip, mask8);
70 pt = (uint64_t)tbl[0] |
71 (uint64_t)tbl[1] << 32;
72 pt2 = (uint64_t)tbl[2] |
73 (uint64_t)tbl[3] << 32;
75 /* search successfully finished for all 4 IP addresses. */
76 if (likely((pt & mask_xv) == mask_v) &&
77 likely((pt2 & mask_xv) == mask_v)) {
78 *(uint64_t *)hop = pt & RTE_LPM_MASKX4_RES;
79 *(uint64_t *)(hop + 2) = pt2 & RTE_LPM_MASKX4_RES;
83 if (unlikely((pt & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
84 RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
85 i8.u32[0] = i8.u32[0] +
86 (uint8_t)tbl[0] * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
87 ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[0]];
90 if (unlikely((pt >> 32 & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
91 RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
92 i8.u32[1] = i8.u32[1] +
93 (uint8_t)tbl[1] * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
94 ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[1]];
97 if (unlikely((pt2 & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
98 RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
99 i8.u32[2] = i8.u32[2] +
100 (uint8_t)tbl[2] * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
101 ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[2]];
104 if (unlikely((pt2 >> 32 & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
105 RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
106 i8.u32[3] = i8.u32[3] +
107 (uint8_t)tbl[3] * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
108 ptbl = (const uint32_t *)&lpm->tbl8[i8.u32[3]];
112 hop[0] = (tbl[0] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[0] & 0x00FFFFFF : defv;
113 hop[1] = (tbl[1] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[1] & 0x00FFFFFF : defv;
114 hop[2] = (tbl[2] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[2] & 0x00FFFFFF : defv;
115 hop[3] = (tbl[3] & RTE_LPM_LOOKUP_SUCCESS) ? tbl[3] & 0x00FFFFFF : defv;
122 #endif /* _RTE_LPM_SSE_H_ */