1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2020 Arm Limited
3 */
4
5 #ifndef _RTE_LPM_SVE_H_
6 #define _RTE_LPM_SVE_H_
7
8 #include <rte_vect.h>
9
10 #ifdef __cplusplus
11 extern "C" {
12 #endif
13
14 __rte_internal
15 static void
__rte_lpm_lookup_vec(const struct rte_lpm * lpm,const uint32_t * ips,uint32_t * __rte_restrict next_hops,const uint32_t n)16 __rte_lpm_lookup_vec(const struct rte_lpm *lpm, const uint32_t *ips,
17 uint32_t *__rte_restrict next_hops, const uint32_t n)
18 {
19 uint32_t i = 0;
20 svuint32_t v_ip, v_idx, v_tbl24, v_tbl8, v_hop;
21 svuint32_t v_mask_xv, v_mask_v, v_mask_hop;
22 svbool_t pg = svwhilelt_b32(i, n);
23 svbool_t pv;
24
25 do {
26 v_ip = svld1(pg, &ips[i]);
27 /* Get indices for tbl24[] */
28 v_idx = svlsr_x(pg, v_ip, 8);
29 /* Extract values from tbl24[] */
30 v_tbl24 = svld1_gather_index(pg, (const uint32_t *)lpm->tbl24,
31 v_idx);
32
33 /* Create mask with valid set */
34 v_mask_v = svdup_u32_z(pg, RTE_LPM_LOOKUP_SUCCESS);
35 /* Create mask with valid and valid_group set */
36 v_mask_xv = svdup_u32_z(pg, RTE_LPM_VALID_EXT_ENTRY_BITMASK);
37 /* Create predicate for tbl24 entries: (valid && !valid_group) */
38 pv = svcmpeq(pg, svand_z(pg, v_tbl24, v_mask_xv), v_mask_v);
39 /* Create mask for next_hop in table entry */
40 v_mask_hop = svdup_u32_z(pg, 0x00ffffff);
41 /* Extract next_hop and write back */
42 v_hop = svand_x(pv, v_tbl24, v_mask_hop);
43 svst1(pv, &next_hops[i], v_hop);
44
45 /* Update predicate for tbl24 entries: (valid && valid_group) */
46 pv = svcmpeq(pg, svand_z(pg, v_tbl24, v_mask_xv), v_mask_xv);
47 /* Compute tbl8 index */
48 v_idx = svand_x(pv, v_tbl24, svdup_u32_z(pv, 0xffffff));
49 v_idx = svmul_x(pv, v_idx, RTE_LPM_TBL8_GROUP_NUM_ENTRIES);
50 v_idx = svadd_x(pv, svand_x(pv, v_ip, svdup_u32_z(pv, 0xff)),
51 v_idx);
52 /* Extract values from tbl8[] */
53 v_tbl8 = svld1_gather_index(pv, (const uint32_t *)lpm->tbl8,
54 v_idx);
55 /* Update predicate for tbl8 entries: (valid) */
56 pv = svcmpeq(pv, svand_z(pv, v_tbl8, v_mask_v), v_mask_v);
57 /* Extract next_hop and write back */
58 v_hop = svand_x(pv, v_tbl8, v_mask_hop);
59 svst1(pv, &next_hops[i], v_hop);
60
61 i += svlen(v_ip);
62 pg = svwhilelt_b32(i, n);
63 } while (svptest_any(svptrue_b32(), pg));
64 }
65
66 static inline void
rte_lpm_lookupx4(const struct rte_lpm * lpm,xmm_t ip,uint32_t hop[4],uint32_t defv)67 rte_lpm_lookupx4(const struct rte_lpm *lpm, xmm_t ip, uint32_t hop[4],
68 uint32_t defv)
69 {
70 uint32_t i, ips[4];
71
72 vst1q_s32((int32_t *)ips, ip);
73 for (i = 0; i < 4; i++)
74 hop[i] = defv;
75
76 __rte_lpm_lookup_vec(lpm, ips, hop, 4);
77 }
78
79 #ifdef __cplusplus
80 }
81 #endif
82
83 #endif /* _RTE_LPM_SVE_H_ */
84