1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
|
/* SPDX-License-Identifier: BSD-3-Clause
* Copyright(c) 2010-2014 Intel Corporation
*/
#include "acl_run_sse.h"
static const rte_ymm_t ymm_match_mask = {
.u32 = {
RTE_ACL_NODE_MATCH,
RTE_ACL_NODE_MATCH,
RTE_ACL_NODE_MATCH,
RTE_ACL_NODE_MATCH,
RTE_ACL_NODE_MATCH,
RTE_ACL_NODE_MATCH,
RTE_ACL_NODE_MATCH,
RTE_ACL_NODE_MATCH,
},
};
static const rte_ymm_t ymm_index_mask = {
.u32 = {
RTE_ACL_NODE_INDEX,
RTE_ACL_NODE_INDEX,
RTE_ACL_NODE_INDEX,
RTE_ACL_NODE_INDEX,
RTE_ACL_NODE_INDEX,
RTE_ACL_NODE_INDEX,
RTE_ACL_NODE_INDEX,
RTE_ACL_NODE_INDEX,
},
};
static const rte_ymm_t ymm_shuffle_input = {
.u32 = {
0x00000000, 0x04040404, 0x08080808, 0x0c0c0c0c,
0x00000000, 0x04040404, 0x08080808, 0x0c0c0c0c,
},
};
static const rte_ymm_t ymm_ones_16 = {
.u16 = {
1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1,
},
};
static const rte_ymm_t ymm_range_base = {
.u32 = {
0xffffff00, 0xffffff04, 0xffffff08, 0xffffff0c,
0xffffff00, 0xffffff04, 0xffffff08, 0xffffff0c,
},
};
/*
* Process 8 transitions in parallel.
* tr_lo contains low 32 bits for 8 transition.
* tr_hi contains high 32 bits for 8 transition.
* next_input contains up to 4 input bytes for 8 flows.
*/
static __rte_always_inline ymm_t
transition8(ymm_t next_input, const uint64_t *trans, ymm_t *tr_lo, ymm_t *tr_hi)
{
const int32_t *tr;
ymm_t addr;
tr = (const int32_t *)(uintptr_t)trans;
/* Calculate the address (array index) for all 8 transitions. */
ACL_TR_CALC_ADDR(mm256, 256, addr, ymm_index_mask.y, next_input,
ymm_shuffle_input.y, ymm_ones_16.y, ymm_range_base.y,
*tr_lo, *tr_hi);
/* load lower 32 bits of 8 transactions at once. */
*tr_lo = _mm256_i32gather_epi32(tr, addr, sizeof(trans[0]));
next_input = _mm256_srli_epi32(next_input, CHAR_BIT);
/* load high 32 bits of 8 transactions at once. */
*tr_hi = _mm256_i32gather_epi32(tr + 1, addr, sizeof(trans[0]));
return next_input;
}
/*
* Process matches for 8 flows.
* tr_lo contains low 32 bits for 8 transition.
* tr_hi contains high 32 bits for 8 transition.
*/
static inline void
acl_process_matches_avx2x8(const struct rte_acl_ctx *ctx,
struct parms *parms, struct acl_flow_data *flows, uint32_t slot,
ymm_t matches, ymm_t *tr_lo, ymm_t *tr_hi)
{
ymm_t t0, t1;
ymm_t lo, hi;
xmm_t l0, l1;
uint32_t i;
uint64_t tr[MAX_SEARCHES_SSE8];
l1 = _mm256_extracti128_si256(*tr_lo, 1);
l0 = _mm256_castsi256_si128(*tr_lo);
for (i = 0; i != RTE_DIM(tr) / 2; i++) {
/*
* Extract low 32bits of each transition.
* That's enough to process the match.
*/
tr[i] = (uint32_t)_mm_cvtsi128_si32(l0);
tr[i + 4] = (uint32_t)_mm_cvtsi128_si32(l1);
l0 = _mm_srli_si128(l0, sizeof(uint32_t));
l1 = _mm_srli_si128(l1, sizeof(uint32_t));
tr[i] = acl_match_check(tr[i], slot + i,
ctx, parms, flows, resolve_priority_sse);
tr[i + 4] = acl_match_check(tr[i + 4], slot + i + 4,
ctx, parms, flows, resolve_priority_sse);
}
/* Collect new transitions into 2 YMM registers. */
t0 = _mm256_set_epi64x(tr[5], tr[4], tr[1], tr[0]);
t1 = _mm256_set_epi64x(tr[7], tr[6], tr[3], tr[2]);
/* For each transition: put low 32 into tr_lo and high 32 into tr_hi */
ACL_TR_HILO(mm256, __m256, t0, t1, lo, hi);
/* Keep transitions wth NOMATCH intact. */
*tr_lo = _mm256_blendv_epi8(*tr_lo, lo, matches);
*tr_hi = _mm256_blendv_epi8(*tr_hi, hi, matches);
}
static inline void
acl_match_check_avx2x8(const struct rte_acl_ctx *ctx, struct parms *parms,
struct acl_flow_data *flows, uint32_t slot,
ymm_t *tr_lo, ymm_t *tr_hi, ymm_t match_mask)
{
uint32_t msk;
ymm_t matches, temp;
/* test for match node */
temp = _mm256_and_si256(match_mask, *tr_lo);
matches = _mm256_cmpeq_epi32(temp, match_mask);
msk = _mm256_movemask_epi8(matches);
while (msk != 0) {
acl_process_matches_avx2x8(ctx, parms, flows, slot,
matches, tr_lo, tr_hi);
temp = _mm256_and_si256(match_mask, *tr_lo);
matches = _mm256_cmpeq_epi32(temp, match_mask);
msk = _mm256_movemask_epi8(matches);
}
}
/*
* Execute trie traversal for up to 16 flows in parallel.
*/
static inline int
search_avx2x16(const struct rte_acl_ctx *ctx, const uint8_t **data,
uint32_t *results, uint32_t total_packets, uint32_t categories)
{
uint32_t n;
struct acl_flow_data flows;
uint64_t index_array[MAX_SEARCHES_AVX16];
struct completion cmplt[MAX_SEARCHES_AVX16];
struct parms parms[MAX_SEARCHES_AVX16];
ymm_t input[2], tr_lo[2], tr_hi[2];
ymm_t t0, t1;
acl_set_flow(&flows, cmplt, RTE_DIM(cmplt), data, results,
total_packets, categories, ctx->trans_table);
for (n = 0; n < RTE_DIM(cmplt); n++) {
cmplt[n].count = 0;
index_array[n] = acl_start_next_trie(&flows, parms, n, ctx);
}
t0 = _mm256_set_epi64x(index_array[5], index_array[4],
index_array[1], index_array[0]);
t1 = _mm256_set_epi64x(index_array[7], index_array[6],
index_array[3], index_array[2]);
ACL_TR_HILO(mm256, __m256, t0, t1, tr_lo[0], tr_hi[0]);
t0 = _mm256_set_epi64x(index_array[13], index_array[12],
index_array[9], index_array[8]);
t1 = _mm256_set_epi64x(index_array[15], index_array[14],
index_array[11], index_array[10]);
ACL_TR_HILO(mm256, __m256, t0, t1, tr_lo[1], tr_hi[1]);
/* Check for any matches. */
acl_match_check_avx2x8(ctx, parms, &flows, 0, &tr_lo[0], &tr_hi[0],
ymm_match_mask.y);
acl_match_check_avx2x8(ctx, parms, &flows, 8, &tr_lo[1], &tr_hi[1],
ymm_match_mask.y);
while (flows.started > 0) {
uint32_t in[MAX_SEARCHES_SSE8];
/* Gather 4 bytes of input data for first 8 flows. */
in[0] = GET_NEXT_4BYTES(parms, 0);
in[4] = GET_NEXT_4BYTES(parms, 4);
in[1] = GET_NEXT_4BYTES(parms, 1);
in[5] = GET_NEXT_4BYTES(parms, 5);
in[2] = GET_NEXT_4BYTES(parms, 2);
in[6] = GET_NEXT_4BYTES(parms, 6);
in[3] = GET_NEXT_4BYTES(parms, 3);
in[7] = GET_NEXT_4BYTES(parms, 7);
input[0] = _mm256_set_epi32(in[7], in[6], in[5], in[4],
in[3], in[2], in[1], in[0]);
/* Gather 4 bytes of input data for last 8 flows. */
in[0] = GET_NEXT_4BYTES(parms, 8);
in[4] = GET_NEXT_4BYTES(parms, 12);
in[1] = GET_NEXT_4BYTES(parms, 9);
in[5] = GET_NEXT_4BYTES(parms, 13);
in[2] = GET_NEXT_4BYTES(parms, 10);
in[6] = GET_NEXT_4BYTES(parms, 14);
in[3] = GET_NEXT_4BYTES(parms, 11);
in[7] = GET_NEXT_4BYTES(parms, 15);
input[1] = _mm256_set_epi32(in[7], in[6], in[5], in[4],
in[3], in[2], in[1], in[0]);
input[0] = transition8(input[0], flows.trans,
&tr_lo[0], &tr_hi[0]);
input[1] = transition8(input[1], flows.trans,
&tr_lo[1], &tr_hi[1]);
input[0] = transition8(input[0], flows.trans,
&tr_lo[0], &tr_hi[0]);
input[1] = transition8(input[1], flows.trans,
&tr_lo[1], &tr_hi[1]);
input[0] = transition8(input[0], flows.trans,
&tr_lo[0], &tr_hi[0]);
input[1] = transition8(input[1], flows.trans,
&tr_lo[1], &tr_hi[1]);
input[0] = transition8(input[0], flows.trans,
&tr_lo[0], &tr_hi[0]);
input[1] = transition8(input[1], flows.trans,
&tr_lo[1], &tr_hi[1]);
/* Check for any matches. */
acl_match_check_avx2x8(ctx, parms, &flows, 0,
&tr_lo[0], &tr_hi[0], ymm_match_mask.y);
acl_match_check_avx2x8(ctx, parms, &flows, 8,
&tr_lo[1], &tr_hi[1], ymm_match_mask.y);
}
return 0;
}
|
