xref: /dpdk/lib/table/rte_table_hash_key32.c (revision daa02b5cddbb8e11b31d41e2bf7bb1ae64dcae2f)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2017 Intel Corporation
3  */
4 #include <string.h>
5 #include <stdio.h>
6 
7 #include <rte_common.h>
8 #include <rte_mbuf.h>
9 #include <rte_memory.h>
10 #include <rte_malloc.h>
11 #include <rte_log.h>
12 
13 #include "rte_table_hash.h"
14 #include "rte_lru.h"
15 
16 #define KEY_SIZE						32
17 
18 #define KEYS_PER_BUCKET					4
19 
20 #define RTE_BUCKET_ENTRY_VALID						0x1LLU
21 
22 #ifdef RTE_TABLE_STATS_COLLECT
23 
24 #define RTE_TABLE_HASH_KEY32_STATS_PKTS_IN_ADD(table, val) \
25 	table->stats.n_pkts_in += val
26 #define RTE_TABLE_HASH_KEY32_STATS_PKTS_LOOKUP_MISS(table, val) \
27 	table->stats.n_pkts_lookup_miss += val
28 
29 #else
30 
31 #define RTE_TABLE_HASH_KEY32_STATS_PKTS_IN_ADD(table, val)
32 #define RTE_TABLE_HASH_KEY32_STATS_PKTS_LOOKUP_MISS(table, val)
33 
34 #endif
35 
36 #ifdef RTE_ARCH_64
37 struct rte_bucket_4_32 {
38 	/* Cache line 0 */
39 	uint64_t signature[4 + 1];
40 	uint64_t lru_list;
41 	struct rte_bucket_4_32 *next;
42 	uint64_t next_valid;
43 
44 	/* Cache lines 1 and 2 */
45 	uint64_t key[4][4];
46 
47 	/* Cache line 3 */
48 	uint8_t data[0];
49 };
50 #else
51 struct rte_bucket_4_32 {
52 	/* Cache line 0 */
53 	uint64_t signature[4 + 1];
54 	uint64_t lru_list;
55 	struct rte_bucket_4_32 *next;
56 	uint32_t pad;
57 	uint64_t next_valid;
58 
59 	/* Cache lines 1 and 2 */
60 	uint64_t key[4][4];
61 
62 	/* Cache line 3 */
63 	uint8_t data[0];
64 };
65 #endif
66 
67 struct rte_table_hash {
68 	struct rte_table_stats stats;
69 
70 	/* Input parameters */
71 	uint32_t n_buckets;
72 	uint32_t key_size;
73 	uint32_t entry_size;
74 	uint32_t bucket_size;
75 	uint32_t key_offset;
76 	uint64_t key_mask[4];
77 	rte_table_hash_op_hash f_hash;
78 	uint64_t seed;
79 
80 	/* Extendible buckets */
81 	uint32_t n_buckets_ext;
82 	uint32_t stack_pos;
83 	uint32_t *stack;
84 
85 	/* Lookup table */
86 	uint8_t memory[0] __rte_cache_aligned;
87 };
88 
89 static int
90 keycmp(void *a, void *b, void *b_mask)
91 {
92 	uint64_t *a64 = a, *b64 = b, *b_mask64 = b_mask;
93 
94 	return (a64[0] != (b64[0] & b_mask64[0])) ||
95 		(a64[1] != (b64[1] & b_mask64[1])) ||
96 		(a64[2] != (b64[2] & b_mask64[2])) ||
97 		(a64[3] != (b64[3] & b_mask64[3]));
98 }
99 
100 static void
101 keycpy(void *dst, void *src, void *src_mask)
102 {
103 	uint64_t *dst64 = dst, *src64 = src, *src_mask64 = src_mask;
104 
105 	dst64[0] = src64[0] & src_mask64[0];
106 	dst64[1] = src64[1] & src_mask64[1];
107 	dst64[2] = src64[2] & src_mask64[2];
108 	dst64[3] = src64[3] & src_mask64[3];
109 }
110 
111 static int
112 check_params_create(struct rte_table_hash_params *params)
113 {
114 	/* name */
115 	if (params->name == NULL) {
116 		RTE_LOG(ERR, TABLE, "%s: name invalid value\n", __func__);
117 		return -EINVAL;
118 	}
119 
120 	/* key_size */
121 	if (params->key_size != KEY_SIZE) {
122 		RTE_LOG(ERR, TABLE, "%s: key_size invalid value\n", __func__);
123 		return -EINVAL;
124 	}
125 
126 	/* n_keys */
127 	if (params->n_keys == 0) {
128 		RTE_LOG(ERR, TABLE, "%s: n_keys is zero\n", __func__);
129 		return -EINVAL;
130 	}
131 
132 	/* n_buckets */
133 	if ((params->n_buckets == 0) ||
134 		(!rte_is_power_of_2(params->n_buckets))) {
135 		RTE_LOG(ERR, TABLE, "%s: n_buckets invalid value\n", __func__);
136 		return -EINVAL;
137 	}
138 
139 	/* f_hash */
140 	if (params->f_hash == NULL) {
141 		RTE_LOG(ERR, TABLE, "%s: f_hash function pointer is NULL\n",
142 			__func__);
143 		return -EINVAL;
144 	}
145 
146 	return 0;
147 }
148 
149 static void *
150 rte_table_hash_create_key32_lru(void *params,
151 		int socket_id,
152 		uint32_t entry_size)
153 {
154 	struct rte_table_hash_params *p = params;
155 	struct rte_table_hash *f;
156 	uint64_t bucket_size, total_size;
157 	uint32_t n_buckets, i;
158 
159 	/* Check input parameters */
160 	if ((check_params_create(p) != 0) ||
161 		((sizeof(struct rte_table_hash) % RTE_CACHE_LINE_SIZE) != 0) ||
162 		((sizeof(struct rte_bucket_4_32) % 64) != 0))
163 		return NULL;
164 
165 	/*
166 	 * Table dimensioning
167 	 *
168 	 * Objective: Pick the number of buckets (n_buckets) so that there a chance
169 	 * to store n_keys keys in the table.
170 	 *
171 	 * Note: Since the buckets do not get extended, it is not possible to
172 	 * guarantee that n_keys keys can be stored in the table at any time. In the
173 	 * worst case scenario when all the n_keys fall into the same bucket, only
174 	 * a maximum of KEYS_PER_BUCKET keys will be stored in the table. This case
175 	 * defeats the purpose of the hash table. It indicates unsuitable f_hash or
176 	 * n_keys to n_buckets ratio.
177 	 *
178 	 * MIN(n_buckets) = (n_keys + KEYS_PER_BUCKET - 1) / KEYS_PER_BUCKET
179 	 */
180 	n_buckets = rte_align32pow2(
181 		(p->n_keys + KEYS_PER_BUCKET - 1) / KEYS_PER_BUCKET);
182 	n_buckets = RTE_MAX(n_buckets, p->n_buckets);
183 
184 	/* Memory allocation */
185 	bucket_size = RTE_CACHE_LINE_ROUNDUP(sizeof(struct rte_bucket_4_32) +
186 		KEYS_PER_BUCKET * entry_size);
187 	total_size = sizeof(struct rte_table_hash) + n_buckets * bucket_size;
188 	if (total_size > SIZE_MAX) {
189 		RTE_LOG(ERR, TABLE, "%s: Cannot allocate %" PRIu64 " bytes "
190 			"for hash table %s\n",
191 			__func__, total_size, p->name);
192 		return NULL;
193 	}
194 
195 	f = rte_zmalloc_socket(p->name,
196 		(size_t)total_size,
197 		RTE_CACHE_LINE_SIZE,
198 		socket_id);
199 	if (f == NULL) {
200 		RTE_LOG(ERR, TABLE, "%s: Cannot allocate %" PRIu64 " bytes "
201 			"for hash table %s\n",
202 			__func__, total_size, p->name);
203 		return NULL;
204 	}
205 	RTE_LOG(INFO, TABLE,
206 		"%s: Hash table %s memory footprint "
207 		"is %" PRIu64 " bytes\n",
208 		__func__, p->name, total_size);
209 
210 	/* Memory initialization */
211 	f->n_buckets = n_buckets;
212 	f->key_size = KEY_SIZE;
213 	f->entry_size = entry_size;
214 	f->bucket_size = bucket_size;
215 	f->key_offset = p->key_offset;
216 	f->f_hash = p->f_hash;
217 	f->seed = p->seed;
218 
219 	if (p->key_mask != NULL) {
220 		f->key_mask[0] = ((uint64_t *)p->key_mask)[0];
221 		f->key_mask[1] = ((uint64_t *)p->key_mask)[1];
222 		f->key_mask[2] = ((uint64_t *)p->key_mask)[2];
223 		f->key_mask[3] = ((uint64_t *)p->key_mask)[3];
224 	} else {
225 		f->key_mask[0] = 0xFFFFFFFFFFFFFFFFLLU;
226 		f->key_mask[1] = 0xFFFFFFFFFFFFFFFFLLU;
227 		f->key_mask[2] = 0xFFFFFFFFFFFFFFFFLLU;
228 		f->key_mask[3] = 0xFFFFFFFFFFFFFFFFLLU;
229 	}
230 
231 	for (i = 0; i < n_buckets; i++) {
232 		struct rte_bucket_4_32 *bucket;
233 
234 		bucket = (struct rte_bucket_4_32 *) &f->memory[i *
235 			f->bucket_size];
236 		bucket->lru_list = 0x0000000100020003LLU;
237 	}
238 
239 	return f;
240 }
241 
242 static int
243 rte_table_hash_free_key32_lru(void *table)
244 {
245 	struct rte_table_hash *f = table;
246 
247 	/* Check input parameters */
248 	if (f == NULL) {
249 		RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__);
250 		return -EINVAL;
251 	}
252 
253 	rte_free(f);
254 	return 0;
255 }
256 
257 static int
258 rte_table_hash_entry_add_key32_lru(
259 	void *table,
260 	void *key,
261 	void *entry,
262 	int *key_found,
263 	void **entry_ptr)
264 {
265 	struct rte_table_hash *f = table;
266 	struct rte_bucket_4_32 *bucket;
267 	uint64_t signature, pos;
268 	uint32_t bucket_index, i;
269 
270 	signature = f->f_hash(key, f->key_mask, f->key_size, f->seed);
271 	bucket_index = signature & (f->n_buckets - 1);
272 	bucket = (struct rte_bucket_4_32 *)
273 		&f->memory[bucket_index * f->bucket_size];
274 	signature |= RTE_BUCKET_ENTRY_VALID;
275 
276 	/* Key is present in the bucket */
277 	for (i = 0; i < 4; i++) {
278 		uint64_t bucket_signature = bucket->signature[i];
279 		uint8_t *bucket_key = (uint8_t *) &bucket->key[i];
280 
281 		if ((bucket_signature == signature) &&
282 			(keycmp(bucket_key, key, f->key_mask) == 0)) {
283 			uint8_t *bucket_data = &bucket->data[i * f->entry_size];
284 
285 			memcpy(bucket_data, entry, f->entry_size);
286 			lru_update(bucket, i);
287 			*key_found = 1;
288 			*entry_ptr = (void *) bucket_data;
289 			return 0;
290 		}
291 	}
292 
293 	/* Key is not present in the bucket */
294 	for (i = 0; i < 4; i++) {
295 		uint64_t bucket_signature = bucket->signature[i];
296 		uint8_t *bucket_key = (uint8_t *) &bucket->key[i];
297 
298 		if (bucket_signature == 0) {
299 			uint8_t *bucket_data = &bucket->data[i * f->entry_size];
300 
301 			bucket->signature[i] = signature;
302 			keycpy(bucket_key, key, f->key_mask);
303 			memcpy(bucket_data, entry, f->entry_size);
304 			lru_update(bucket, i);
305 			*key_found = 0;
306 			*entry_ptr = (void *) bucket_data;
307 
308 			return 0;
309 		}
310 	}
311 
312 	/* Bucket full: replace LRU entry */
313 	pos = lru_pos(bucket);
314 	bucket->signature[pos] = signature;
315 	keycpy(&bucket->key[pos], key, f->key_mask);
316 	memcpy(&bucket->data[pos * f->entry_size], entry, f->entry_size);
317 	lru_update(bucket, pos);
318 	*key_found = 0;
319 	*entry_ptr = (void *) &bucket->data[pos * f->entry_size];
320 
321 	return 0;
322 }
323 
324 static int
325 rte_table_hash_entry_delete_key32_lru(
326 	void *table,
327 	void *key,
328 	int *key_found,
329 	void *entry)
330 {
331 	struct rte_table_hash *f = table;
332 	struct rte_bucket_4_32 *bucket;
333 	uint64_t signature;
334 	uint32_t bucket_index, i;
335 
336 	signature = f->f_hash(key, f->key_mask, f->key_size, f->seed);
337 	bucket_index = signature & (f->n_buckets - 1);
338 	bucket = (struct rte_bucket_4_32 *)
339 		&f->memory[bucket_index * f->bucket_size];
340 	signature |= RTE_BUCKET_ENTRY_VALID;
341 
342 	/* Key is present in the bucket */
343 	for (i = 0; i < 4; i++) {
344 		uint64_t bucket_signature = bucket->signature[i];
345 		uint8_t *bucket_key = (uint8_t *) &bucket->key[i];
346 
347 		if ((bucket_signature == signature) &&
348 			(keycmp(bucket_key, key, f->key_mask) == 0)) {
349 			uint8_t *bucket_data = &bucket->data[i * f->entry_size];
350 
351 			bucket->signature[i] = 0;
352 			*key_found = 1;
353 			if (entry)
354 				memcpy(entry, bucket_data, f->entry_size);
355 
356 			return 0;
357 		}
358 	}
359 
360 	/* Key is not present in the bucket */
361 	*key_found = 0;
362 	return 0;
363 }
364 
365 static void *
366 rte_table_hash_create_key32_ext(void *params,
367 	int socket_id,
368 	uint32_t entry_size)
369 {
370 	struct rte_table_hash_params *p = params;
371 	struct rte_table_hash *f;
372 	uint64_t bucket_size, stack_size, total_size;
373 	uint32_t n_buckets_ext, i;
374 
375 	/* Check input parameters */
376 	if ((check_params_create(p) != 0) ||
377 		((sizeof(struct rte_table_hash) % RTE_CACHE_LINE_SIZE) != 0) ||
378 		((sizeof(struct rte_bucket_4_32) % 64) != 0))
379 		return NULL;
380 
381 	/*
382 	 * Table dimensioning
383 	 *
384 	 * Objective: Pick the number of bucket extensions (n_buckets_ext) so that
385 	 * it is guaranteed that n_keys keys can be stored in the table at any time.
386 	 *
387 	 * The worst case scenario takes place when all the n_keys keys fall into
388 	 * the same bucket. Actually, due to the KEYS_PER_BUCKET scheme, the worst
389 	 * case takes place when (n_keys - KEYS_PER_BUCKET + 1) keys fall into the
390 	 * same bucket, while the remaining (KEYS_PER_BUCKET - 1) keys each fall
391 	 * into a different bucket. This case defeats the purpose of the hash table.
392 	 * It indicates unsuitable f_hash or n_keys to n_buckets ratio.
393 	 *
394 	 * n_buckets_ext = n_keys / KEYS_PER_BUCKET + KEYS_PER_BUCKET - 1
395 	 */
396 	n_buckets_ext = p->n_keys / KEYS_PER_BUCKET + KEYS_PER_BUCKET - 1;
397 
398 	/* Memory allocation */
399 	bucket_size = RTE_CACHE_LINE_ROUNDUP(sizeof(struct rte_bucket_4_32) +
400 		KEYS_PER_BUCKET * entry_size);
401 	stack_size = RTE_CACHE_LINE_ROUNDUP(n_buckets_ext * sizeof(uint32_t));
402 	total_size = sizeof(struct rte_table_hash) +
403 		(p->n_buckets + n_buckets_ext) * bucket_size + stack_size;
404 	if (total_size > SIZE_MAX) {
405 		RTE_LOG(ERR, TABLE, "%s: Cannot allocate %" PRIu64 " bytes "
406 			"for hash table %s\n",
407 			__func__, total_size, p->name);
408 		return NULL;
409 	}
410 
411 	f = rte_zmalloc_socket(p->name,
412 		(size_t)total_size,
413 		RTE_CACHE_LINE_SIZE,
414 		socket_id);
415 	if (f == NULL) {
416 		RTE_LOG(ERR, TABLE, "%s: Cannot allocate %" PRIu64 " bytes "
417 			"for hash table %s\n",
418 			__func__, total_size, p->name);
419 		return NULL;
420 	}
421 	RTE_LOG(INFO, TABLE,
422 		"%s: Hash table %s memory footprint "
423 		"is %" PRIu64" bytes\n",
424 		__func__, p->name, total_size);
425 
426 	/* Memory initialization */
427 	f->n_buckets = p->n_buckets;
428 	f->key_size = KEY_SIZE;
429 	f->entry_size = entry_size;
430 	f->bucket_size = bucket_size;
431 	f->key_offset = p->key_offset;
432 	f->f_hash = p->f_hash;
433 	f->seed = p->seed;
434 
435 	f->n_buckets_ext = n_buckets_ext;
436 	f->stack_pos = n_buckets_ext;
437 	f->stack = (uint32_t *)
438 		&f->memory[(p->n_buckets + n_buckets_ext) * f->bucket_size];
439 
440 	if (p->key_mask != NULL) {
441 		f->key_mask[0] = (((uint64_t *)p->key_mask)[0]);
442 		f->key_mask[1] = (((uint64_t *)p->key_mask)[1]);
443 		f->key_mask[2] = (((uint64_t *)p->key_mask)[2]);
444 		f->key_mask[3] = (((uint64_t *)p->key_mask)[3]);
445 	} else {
446 		f->key_mask[0] = 0xFFFFFFFFFFFFFFFFLLU;
447 		f->key_mask[1] = 0xFFFFFFFFFFFFFFFFLLU;
448 		f->key_mask[2] = 0xFFFFFFFFFFFFFFFFLLU;
449 		f->key_mask[3] = 0xFFFFFFFFFFFFFFFFLLU;
450 	}
451 
452 	for (i = 0; i < n_buckets_ext; i++)
453 		f->stack[i] = i;
454 
455 	return f;
456 }
457 
458 static int
459 rte_table_hash_free_key32_ext(void *table)
460 {
461 	struct rte_table_hash *f = table;
462 
463 	/* Check input parameters */
464 	if (f == NULL) {
465 		RTE_LOG(ERR, TABLE, "%s: table parameter is NULL\n", __func__);
466 		return -EINVAL;
467 	}
468 
469 	rte_free(f);
470 	return 0;
471 }
472 
473 static int
474 rte_table_hash_entry_add_key32_ext(
475 	void *table,
476 	void *key,
477 	void *entry,
478 	int *key_found,
479 	void **entry_ptr)
480 {
481 	struct rte_table_hash *f = table;
482 	struct rte_bucket_4_32 *bucket0, *bucket, *bucket_prev;
483 	uint64_t signature;
484 	uint32_t bucket_index, i;
485 
486 	signature = f->f_hash(key, f->key_mask, f->key_size, f->seed);
487 	bucket_index = signature & (f->n_buckets - 1);
488 	bucket0 = (struct rte_bucket_4_32 *)
489 			&f->memory[bucket_index * f->bucket_size];
490 	signature |= RTE_BUCKET_ENTRY_VALID;
491 
492 	/* Key is present in the bucket */
493 	for (bucket = bucket0; bucket != NULL; bucket = bucket->next) {
494 		for (i = 0; i < 4; i++) {
495 			uint64_t bucket_signature = bucket->signature[i];
496 			uint8_t *bucket_key = (uint8_t *) &bucket->key[i];
497 
498 			if ((bucket_signature == signature) &&
499 				(keycmp(bucket_key, key, f->key_mask) == 0)) {
500 				uint8_t *bucket_data = &bucket->data[i *
501 					f->entry_size];
502 
503 				memcpy(bucket_data, entry, f->entry_size);
504 				*key_found = 1;
505 				*entry_ptr = (void *) bucket_data;
506 
507 				return 0;
508 			}
509 		}
510 	}
511 
512 	/* Key is not present in the bucket */
513 	for (bucket_prev = NULL, bucket = bucket0; bucket != NULL;
514 		bucket_prev = bucket, bucket = bucket->next)
515 		for (i = 0; i < 4; i++) {
516 			uint64_t bucket_signature = bucket->signature[i];
517 			uint8_t *bucket_key = (uint8_t *) &bucket->key[i];
518 
519 			if (bucket_signature == 0) {
520 				uint8_t *bucket_data = &bucket->data[i *
521 					f->entry_size];
522 
523 				bucket->signature[i] = signature;
524 				keycpy(bucket_key, key, f->key_mask);
525 				memcpy(bucket_data, entry, f->entry_size);
526 				*key_found = 0;
527 				*entry_ptr = (void *) bucket_data;
528 
529 				return 0;
530 			}
531 		}
532 
533 	/* Bucket full: extend bucket */
534 	if (f->stack_pos > 0) {
535 		bucket_index = f->stack[--f->stack_pos];
536 
537 		bucket = (struct rte_bucket_4_32 *)
538 			&f->memory[(f->n_buckets + bucket_index) *
539 			f->bucket_size];
540 		bucket_prev->next = bucket;
541 		bucket_prev->next_valid = 1;
542 
543 		bucket->signature[0] = signature;
544 		keycpy(&bucket->key[0], key, f->key_mask);
545 		memcpy(&bucket->data[0], entry, f->entry_size);
546 		*key_found = 0;
547 		*entry_ptr = (void *) &bucket->data[0];
548 		return 0;
549 	}
550 
551 	return -ENOSPC;
552 }
553 
554 static int
555 rte_table_hash_entry_delete_key32_ext(
556 	void *table,
557 	void *key,
558 	int *key_found,
559 	void *entry)
560 {
561 	struct rte_table_hash *f = table;
562 	struct rte_bucket_4_32 *bucket0, *bucket, *bucket_prev;
563 	uint64_t signature;
564 	uint32_t bucket_index, i;
565 
566 	signature = f->f_hash(key, f->key_mask, f->key_size, f->seed);
567 	bucket_index = signature & (f->n_buckets - 1);
568 	bucket0 = (struct rte_bucket_4_32 *)
569 		&f->memory[bucket_index * f->bucket_size];
570 	signature |= RTE_BUCKET_ENTRY_VALID;
571 
572 	/* Key is present in the bucket */
573 	for (bucket_prev = NULL, bucket = bucket0; bucket != NULL;
574 		bucket_prev = bucket, bucket = bucket->next)
575 		for (i = 0; i < 4; i++) {
576 			uint64_t bucket_signature = bucket->signature[i];
577 			uint8_t *bucket_key = (uint8_t *) &bucket->key[i];
578 
579 			if ((bucket_signature == signature) &&
580 				(keycmp(bucket_key, key, f->key_mask) == 0)) {
581 				uint8_t *bucket_data = &bucket->data[i *
582 					f->entry_size];
583 
584 				bucket->signature[i] = 0;
585 				*key_found = 1;
586 				if (entry)
587 					memcpy(entry, bucket_data, f->entry_size);
588 
589 				if ((bucket->signature[0] == 0) &&
590 					(bucket->signature[1] == 0) &&
591 					(bucket->signature[2] == 0) &&
592 					(bucket->signature[3] == 0) &&
593 					(bucket_prev != NULL)) {
594 					bucket_prev->next = bucket->next;
595 					bucket_prev->next_valid =
596 						bucket->next_valid;
597 
598 					memset(bucket, 0,
599 						sizeof(struct rte_bucket_4_32));
600 					bucket_index = (((uint8_t *)bucket -
601 						(uint8_t *)f->memory)/f->bucket_size) - f->n_buckets;
602 					f->stack[f->stack_pos++] = bucket_index;
603 				}
604 
605 				return 0;
606 			}
607 		}
608 
609 	/* Key is not present in the bucket */
610 	*key_found = 0;
611 	return 0;
612 }
613 
614 #define lookup_key32_cmp(key_in, bucket, pos, f)			\
615 {								\
616 	uint64_t xor[4][4], or[4], signature[4], k[4];		\
617 								\
618 	k[0] = key_in[0] & f->key_mask[0];				\
619 	k[1] = key_in[1] & f->key_mask[1];				\
620 	k[2] = key_in[2] & f->key_mask[2];				\
621 	k[3] = key_in[3] & f->key_mask[3];				\
622 								\
623 	signature[0] = ((~bucket->signature[0]) & 1);		\
624 	signature[1] = ((~bucket->signature[1]) & 1);		\
625 	signature[2] = ((~bucket->signature[2]) & 1);		\
626 	signature[3] = ((~bucket->signature[3]) & 1);		\
627 								\
628 	xor[0][0] = k[0] ^ bucket->key[0][0];			\
629 	xor[0][1] = k[1] ^ bucket->key[0][1];			\
630 	xor[0][2] = k[2] ^ bucket->key[0][2];			\
631 	xor[0][3] = k[3] ^ bucket->key[0][3];			\
632 								\
633 	xor[1][0] = k[0] ^ bucket->key[1][0];			\
634 	xor[1][1] = k[1] ^ bucket->key[1][1];			\
635 	xor[1][2] = k[2] ^ bucket->key[1][2];			\
636 	xor[1][3] = k[3] ^ bucket->key[1][3];			\
637 								\
638 	xor[2][0] = k[0] ^ bucket->key[2][0];			\
639 	xor[2][1] = k[1] ^ bucket->key[2][1];			\
640 	xor[2][2] = k[2] ^ bucket->key[2][2];			\
641 	xor[2][3] = k[3] ^ bucket->key[2][3];			\
642 								\
643 	xor[3][0] = k[0] ^ bucket->key[3][0];			\
644 	xor[3][1] = k[1] ^ bucket->key[3][1];			\
645 	xor[3][2] = k[2] ^ bucket->key[3][2];			\
646 	xor[3][3] = k[3] ^ bucket->key[3][3];			\
647 								\
648 	or[0] = xor[0][0] | xor[0][1] | xor[0][2] | xor[0][3] | signature[0];\
649 	or[1] = xor[1][0] | xor[1][1] | xor[1][2] | xor[1][3] | signature[1];\
650 	or[2] = xor[2][0] | xor[2][1] | xor[2][2] | xor[2][3] | signature[2];\
651 	or[3] = xor[3][0] | xor[3][1] | xor[3][2] | xor[3][3] | signature[3];\
652 								\
653 	pos = 4;						\
654 	if (or[0] == 0)						\
655 		pos = 0;					\
656 	if (or[1] == 0)						\
657 		pos = 1;					\
658 	if (or[2] == 0)						\
659 		pos = 2;					\
660 	if (or[3] == 0)						\
661 		pos = 3;					\
662 }
663 
664 #define lookup1_stage0(pkt0_index, mbuf0, pkts, pkts_mask, f)	\
665 {								\
666 	uint64_t pkt_mask;					\
667 	uint32_t key_offset = f->key_offset;	\
668 								\
669 	pkt0_index = __builtin_ctzll(pkts_mask);		\
670 	pkt_mask = 1LLU << pkt0_index;				\
671 	pkts_mask &= ~pkt_mask;					\
672 								\
673 	mbuf0 = pkts[pkt0_index];				\
674 	rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf0, key_offset));\
675 }
676 
677 #define lookup1_stage1(mbuf1, bucket1, f)				\
678 {								\
679 	uint64_t *key;						\
680 	uint64_t signature;					\
681 	uint32_t bucket_index;					\
682 								\
683 	key = RTE_MBUF_METADATA_UINT64_PTR(mbuf1, f->key_offset);	\
684 	signature = f->f_hash(key, f->key_mask, KEY_SIZE, f->seed);	\
685 								\
686 	bucket_index = signature & (f->n_buckets - 1);		\
687 	bucket1 = (struct rte_bucket_4_32 *)			\
688 		&f->memory[bucket_index * f->bucket_size];	\
689 	rte_prefetch0(bucket1);					\
690 	rte_prefetch0((void *)(((uintptr_t) bucket1) + RTE_CACHE_LINE_SIZE));\
691 	rte_prefetch0((void *)(((uintptr_t) bucket1) + 2 * RTE_CACHE_LINE_SIZE));\
692 }
693 
694 #define lookup1_stage2_lru(pkt2_index, mbuf2, bucket2,		\
695 	pkts_mask_out, entries, f)				\
696 {								\
697 	void *a;						\
698 	uint64_t pkt_mask;					\
699 	uint64_t *key;						\
700 	uint32_t pos;						\
701 								\
702 	key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\
703 	lookup_key32_cmp(key, bucket2, pos, f);			\
704 								\
705 	pkt_mask = (bucket2->signature[pos] & 1LLU) << pkt2_index;\
706 	pkts_mask_out |= pkt_mask;				\
707 								\
708 	a = (void *) &bucket2->data[pos * f->entry_size];	\
709 	rte_prefetch0(a);					\
710 	entries[pkt2_index] = a;				\
711 	lru_update(bucket2, pos);				\
712 }
713 
714 #define lookup1_stage2_ext(pkt2_index, mbuf2, bucket2, pkts_mask_out,\
715 	entries, buckets_mask, buckets, keys, f)		\
716 {								\
717 	struct rte_bucket_4_32 *bucket_next;			\
718 	void *a;						\
719 	uint64_t pkt_mask, bucket_mask;				\
720 	uint64_t *key;						\
721 	uint32_t pos;						\
722 								\
723 	key = RTE_MBUF_METADATA_UINT64_PTR(mbuf2, f->key_offset);\
724 	lookup_key32_cmp(key, bucket2, pos, f);			\
725 								\
726 	pkt_mask = (bucket2->signature[pos] & 1LLU) << pkt2_index;\
727 	pkts_mask_out |= pkt_mask;				\
728 								\
729 	a = (void *) &bucket2->data[pos * f->entry_size];	\
730 	rte_prefetch0(a);					\
731 	entries[pkt2_index] = a;				\
732 								\
733 	bucket_mask = (~pkt_mask) & (bucket2->next_valid << pkt2_index);\
734 	buckets_mask |= bucket_mask;				\
735 	bucket_next = bucket2->next;				\
736 	buckets[pkt2_index] = bucket_next;			\
737 	keys[pkt2_index] = key;					\
738 }
739 
740 #define lookup_grinder(pkt_index, buckets, keys, pkts_mask_out,	\
741 	entries, buckets_mask, f)				\
742 {								\
743 	struct rte_bucket_4_32 *bucket, *bucket_next;		\
744 	void *a;						\
745 	uint64_t pkt_mask, bucket_mask;				\
746 	uint64_t *key;						\
747 	uint32_t pos;						\
748 								\
749 	bucket = buckets[pkt_index];				\
750 	key = keys[pkt_index];					\
751 								\
752 	lookup_key32_cmp(key, bucket, pos, f);			\
753 								\
754 	pkt_mask = (bucket->signature[pos] & 1LLU) << pkt_index;\
755 	pkts_mask_out |= pkt_mask;				\
756 								\
757 	a = (void *) &bucket->data[pos * f->entry_size];	\
758 	rte_prefetch0(a);					\
759 	entries[pkt_index] = a;					\
760 								\
761 	bucket_mask = (~pkt_mask) & (bucket->next_valid << pkt_index);\
762 	buckets_mask |= bucket_mask;				\
763 	bucket_next = bucket->next;				\
764 	rte_prefetch0(bucket_next);				\
765 	rte_prefetch0((void *)(((uintptr_t) bucket_next) + RTE_CACHE_LINE_SIZE));\
766 	rte_prefetch0((void *)(((uintptr_t) bucket_next) +	\
767 		2 * RTE_CACHE_LINE_SIZE));				\
768 	buckets[pkt_index] = bucket_next;			\
769 	keys[pkt_index] = key;					\
770 }
771 
772 #define lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01,\
773 	pkts, pkts_mask, f)					\
774 {								\
775 	uint64_t pkt00_mask, pkt01_mask;			\
776 	uint32_t key_offset = f->key_offset;		\
777 								\
778 	pkt00_index = __builtin_ctzll(pkts_mask);		\
779 	pkt00_mask = 1LLU << pkt00_index;			\
780 	pkts_mask &= ~pkt00_mask;				\
781 								\
782 	mbuf00 = pkts[pkt00_index];				\
783 	rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf00, key_offset));\
784 								\
785 	pkt01_index = __builtin_ctzll(pkts_mask);		\
786 	pkt01_mask = 1LLU << pkt01_index;			\
787 	pkts_mask &= ~pkt01_mask;				\
788 								\
789 	mbuf01 = pkts[pkt01_index];				\
790 	rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, key_offset));\
791 }
792 
793 #define lookup2_stage0_with_odd_support(pkt00_index, pkt01_index,\
794 	mbuf00, mbuf01, pkts, pkts_mask, f)			\
795 {								\
796 	uint64_t pkt00_mask, pkt01_mask;			\
797 	uint32_t key_offset = f->key_offset;		\
798 								\
799 	pkt00_index = __builtin_ctzll(pkts_mask);		\
800 	pkt00_mask = 1LLU << pkt00_index;			\
801 	pkts_mask &= ~pkt00_mask;				\
802 								\
803 	mbuf00 = pkts[pkt00_index];				\
804 	rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf00, key_offset));	\
805 								\
806 	pkt01_index = __builtin_ctzll(pkts_mask);		\
807 	if (pkts_mask == 0)					\
808 		pkt01_index = pkt00_index;			\
809 								\
810 	pkt01_mask = 1LLU << pkt01_index;			\
811 	pkts_mask &= ~pkt01_mask;				\
812 								\
813 	mbuf01 = pkts[pkt01_index];				\
814 	rte_prefetch0(RTE_MBUF_METADATA_UINT8_PTR(mbuf01, key_offset));	\
815 }
816 
817 #define lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f)	\
818 {								\
819 	uint64_t *key10, *key11;					\
820 	uint64_t signature10, signature11;				\
821 	uint32_t bucket10_index, bucket11_index;			\
822 								\
823 	key10 = RTE_MBUF_METADATA_UINT64_PTR(mbuf10, f->key_offset);	\
824 	signature10 = f->f_hash(key10, f->key_mask,	 KEY_SIZE, f->seed); \
825 								\
826 	bucket10_index = signature10 & (f->n_buckets - 1);		\
827 	bucket10 = (struct rte_bucket_4_32 *)			\
828 		&f->memory[bucket10_index * f->bucket_size];	\
829 	rte_prefetch0(bucket10);					\
830 	rte_prefetch0((void *)(((uintptr_t) bucket10) + RTE_CACHE_LINE_SIZE));\
831 	rte_prefetch0((void *)(((uintptr_t) bucket10) + 2 * RTE_CACHE_LINE_SIZE));\
832 								\
833 	key11 = RTE_MBUF_METADATA_UINT64_PTR(mbuf11, f->key_offset);	\
834 	signature11 = f->f_hash(key11, f->key_mask, KEY_SIZE, f->seed);\
835 								\
836 	bucket11_index = signature11 & (f->n_buckets - 1);		\
837 	bucket11 = (struct rte_bucket_4_32 *)			\
838 		&f->memory[bucket11_index * f->bucket_size];	\
839 	rte_prefetch0(bucket11);					\
840 	rte_prefetch0((void *)(((uintptr_t) bucket11) + RTE_CACHE_LINE_SIZE));\
841 	rte_prefetch0((void *)(((uintptr_t) bucket11) + 2 * RTE_CACHE_LINE_SIZE));\
842 }
843 
844 #define lookup2_stage2_lru(pkt20_index, pkt21_index, mbuf20, mbuf21,\
845 	bucket20, bucket21, pkts_mask_out, entries, f)		\
846 {								\
847 	void *a20, *a21;					\
848 	uint64_t pkt20_mask, pkt21_mask;			\
849 	uint64_t *key20, *key21;				\
850 	uint32_t pos20, pos21;					\
851 								\
852 	key20 = RTE_MBUF_METADATA_UINT64_PTR(mbuf20, f->key_offset);\
853 	key21 = RTE_MBUF_METADATA_UINT64_PTR(mbuf21, f->key_offset);\
854 								\
855 	lookup_key32_cmp(key20, bucket20, pos20, f);		\
856 	lookup_key32_cmp(key21, bucket21, pos21, f);		\
857 								\
858 	pkt20_mask = (bucket20->signature[pos20] & 1LLU) << pkt20_index;\
859 	pkt21_mask = (bucket21->signature[pos21] & 1LLU) << pkt21_index;\
860 	pkts_mask_out |= pkt20_mask | pkt21_mask;		\
861 								\
862 	a20 = (void *) &bucket20->data[pos20 * f->entry_size];	\
863 	a21 = (void *) &bucket21->data[pos21 * f->entry_size];	\
864 	rte_prefetch0(a20);					\
865 	rte_prefetch0(a21);					\
866 	entries[pkt20_index] = a20;				\
867 	entries[pkt21_index] = a21;				\
868 	lru_update(bucket20, pos20);				\
869 	lru_update(bucket21, pos21);				\
870 }
871 
872 #define lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21, bucket20, \
873 	bucket21, pkts_mask_out, entries, buckets_mask, buckets, keys, f)\
874 {								\
875 	struct rte_bucket_4_32 *bucket20_next, *bucket21_next;	\
876 	void *a20, *a21;					\
877 	uint64_t pkt20_mask, pkt21_mask, bucket20_mask, bucket21_mask;\
878 	uint64_t *key20, *key21;				\
879 	uint32_t pos20, pos21;					\
880 								\
881 	key20 = RTE_MBUF_METADATA_UINT64_PTR(mbuf20, f->key_offset);\
882 	key21 = RTE_MBUF_METADATA_UINT64_PTR(mbuf21, f->key_offset);\
883 								\
884 	lookup_key32_cmp(key20, bucket20, pos20, f);		\
885 	lookup_key32_cmp(key21, bucket21, pos21, f);		\
886 								\
887 	pkt20_mask = (bucket20->signature[pos20] & 1LLU) << pkt20_index;\
888 	pkt21_mask = (bucket21->signature[pos21] & 1LLU) << pkt21_index;\
889 	pkts_mask_out |= pkt20_mask | pkt21_mask;		\
890 								\
891 	a20 = (void *) &bucket20->data[pos20 * f->entry_size];	\
892 	a21 = (void *) &bucket21->data[pos21 * f->entry_size];	\
893 	rte_prefetch0(a20);					\
894 	rte_prefetch0(a21);					\
895 	entries[pkt20_index] = a20;				\
896 	entries[pkt21_index] = a21;				\
897 								\
898 	bucket20_mask = (~pkt20_mask) & (bucket20->next_valid << pkt20_index);\
899 	bucket21_mask = (~pkt21_mask) & (bucket21->next_valid << pkt21_index);\
900 	buckets_mask |= bucket20_mask | bucket21_mask;		\
901 	bucket20_next = bucket20->next;				\
902 	bucket21_next = bucket21->next;				\
903 	buckets[pkt20_index] = bucket20_next;			\
904 	buckets[pkt21_index] = bucket21_next;			\
905 	keys[pkt20_index] = key20;				\
906 	keys[pkt21_index] = key21;				\
907 }
908 
909 static int
910 rte_table_hash_lookup_key32_lru(
911 	void *table,
912 	struct rte_mbuf **pkts,
913 	uint64_t pkts_mask,
914 	uint64_t *lookup_hit_mask,
915 	void **entries)
916 {
917 	struct rte_table_hash *f = (struct rte_table_hash *) table;
918 	struct rte_bucket_4_32 *bucket10, *bucket11, *bucket20, *bucket21;
919 	struct rte_mbuf *mbuf00, *mbuf01, *mbuf10, *mbuf11, *mbuf20, *mbuf21;
920 	uint32_t pkt00_index, pkt01_index, pkt10_index;
921 	uint32_t pkt11_index, pkt20_index, pkt21_index;
922 	uint64_t pkts_mask_out = 0;
923 
924 	__rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask);
925 	RTE_TABLE_HASH_KEY32_STATS_PKTS_IN_ADD(f, n_pkts_in);
926 
927 	/* Cannot run the pipeline with less than 5 packets */
928 	if (__builtin_popcountll(pkts_mask) < 5) {
929 		for ( ; pkts_mask; ) {
930 			struct rte_bucket_4_32 *bucket;
931 			struct rte_mbuf *mbuf;
932 			uint32_t pkt_index;
933 
934 			lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask, f);
935 			lookup1_stage1(mbuf, bucket, f);
936 			lookup1_stage2_lru(pkt_index, mbuf, bucket,
937 					pkts_mask_out, entries, f);
938 		}
939 
940 		*lookup_hit_mask = pkts_mask_out;
941 		RTE_TABLE_HASH_KEY32_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(pkts_mask_out));
942 		return 0;
943 	}
944 
945 	/*
946 	 * Pipeline fill
947 	 *
948 	 */
949 	/* Pipeline stage 0 */
950 	lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts,
951 		pkts_mask, f);
952 
953 	/* Pipeline feed */
954 	mbuf10 = mbuf00;
955 	mbuf11 = mbuf01;
956 	pkt10_index = pkt00_index;
957 	pkt11_index = pkt01_index;
958 
959 	/* Pipeline stage 0 */
960 	lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts,
961 		pkts_mask, f);
962 
963 	/* Pipeline stage 1 */
964 	lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f);
965 
966 	/*
967 	 * Pipeline run
968 	 *
969 	 */
970 	for ( ; pkts_mask; ) {
971 		/* Pipeline feed */
972 		bucket20 = bucket10;
973 		bucket21 = bucket11;
974 		mbuf20 = mbuf10;
975 		mbuf21 = mbuf11;
976 		mbuf10 = mbuf00;
977 		mbuf11 = mbuf01;
978 		pkt20_index = pkt10_index;
979 		pkt21_index = pkt11_index;
980 		pkt10_index = pkt00_index;
981 		pkt11_index = pkt01_index;
982 
983 		/* Pipeline stage 0 */
984 		lookup2_stage0_with_odd_support(pkt00_index, pkt01_index,
985 			mbuf00, mbuf01, pkts, pkts_mask, f);
986 
987 		/* Pipeline stage 1 */
988 		lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f);
989 
990 		/* Pipeline stage 2 */
991 		lookup2_stage2_lru(pkt20_index, pkt21_index,
992 			mbuf20, mbuf21, bucket20, bucket21, pkts_mask_out,
993 			entries, f);
994 	}
995 
996 	/*
997 	 * Pipeline flush
998 	 *
999 	 */
1000 	/* Pipeline feed */
1001 	bucket20 = bucket10;
1002 	bucket21 = bucket11;
1003 	mbuf20 = mbuf10;
1004 	mbuf21 = mbuf11;
1005 	mbuf10 = mbuf00;
1006 	mbuf11 = mbuf01;
1007 	pkt20_index = pkt10_index;
1008 	pkt21_index = pkt11_index;
1009 	pkt10_index = pkt00_index;
1010 	pkt11_index = pkt01_index;
1011 
1012 	/* Pipeline stage 1 */
1013 	lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f);
1014 
1015 	/* Pipeline stage 2 */
1016 	lookup2_stage2_lru(pkt20_index, pkt21_index,
1017 		mbuf20, mbuf21, bucket20, bucket21, pkts_mask_out, entries, f);
1018 
1019 	/* Pipeline feed */
1020 	bucket20 = bucket10;
1021 	bucket21 = bucket11;
1022 	mbuf20 = mbuf10;
1023 	mbuf21 = mbuf11;
1024 	pkt20_index = pkt10_index;
1025 	pkt21_index = pkt11_index;
1026 
1027 	/* Pipeline stage 2 */
1028 	lookup2_stage2_lru(pkt20_index, pkt21_index,
1029 		mbuf20, mbuf21, bucket20, bucket21, pkts_mask_out, entries, f);
1030 
1031 	*lookup_hit_mask = pkts_mask_out;
1032 	RTE_TABLE_HASH_KEY32_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(pkts_mask_out));
1033 	return 0;
1034 } /* rte_table_hash_lookup_key32_lru() */
1035 
1036 static int
1037 rte_table_hash_lookup_key32_ext(
1038 	void *table,
1039 	struct rte_mbuf **pkts,
1040 	uint64_t pkts_mask,
1041 	uint64_t *lookup_hit_mask,
1042 	void **entries)
1043 {
1044 	struct rte_table_hash *f = (struct rte_table_hash *) table;
1045 	struct rte_bucket_4_32 *bucket10, *bucket11, *bucket20, *bucket21;
1046 	struct rte_mbuf *mbuf00, *mbuf01, *mbuf10, *mbuf11, *mbuf20, *mbuf21;
1047 	uint32_t pkt00_index, pkt01_index, pkt10_index;
1048 	uint32_t pkt11_index, pkt20_index, pkt21_index;
1049 	uint64_t pkts_mask_out = 0, buckets_mask = 0;
1050 	struct rte_bucket_4_32 *buckets[RTE_PORT_IN_BURST_SIZE_MAX];
1051 	uint64_t *keys[RTE_PORT_IN_BURST_SIZE_MAX];
1052 
1053 	__rte_unused uint32_t n_pkts_in = __builtin_popcountll(pkts_mask);
1054 	RTE_TABLE_HASH_KEY32_STATS_PKTS_IN_ADD(f, n_pkts_in);
1055 
1056 	/* Cannot run the pipeline with less than 5 packets */
1057 	if (__builtin_popcountll(pkts_mask) < 5) {
1058 		for ( ; pkts_mask; ) {
1059 			struct rte_bucket_4_32 *bucket;
1060 			struct rte_mbuf *mbuf;
1061 			uint32_t pkt_index;
1062 
1063 			lookup1_stage0(pkt_index, mbuf, pkts, pkts_mask, f);
1064 			lookup1_stage1(mbuf, bucket, f);
1065 			lookup1_stage2_ext(pkt_index, mbuf, bucket,
1066 				pkts_mask_out, entries, buckets_mask, buckets,
1067 				keys, f);
1068 		}
1069 
1070 		goto grind_next_buckets;
1071 	}
1072 
1073 	/*
1074 	 * Pipeline fill
1075 	 *
1076 	 */
1077 	/* Pipeline stage 0 */
1078 	lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts,
1079 		pkts_mask, f);
1080 
1081 	/* Pipeline feed */
1082 	mbuf10 = mbuf00;
1083 	mbuf11 = mbuf01;
1084 	pkt10_index = pkt00_index;
1085 	pkt11_index = pkt01_index;
1086 
1087 	/* Pipeline stage 0 */
1088 	lookup2_stage0(pkt00_index, pkt01_index, mbuf00, mbuf01, pkts,
1089 		pkts_mask, f);
1090 
1091 	/* Pipeline stage 1 */
1092 	lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f);
1093 
1094 	/*
1095 	 * Pipeline run
1096 	 *
1097 	 */
1098 	for ( ; pkts_mask; ) {
1099 		/* Pipeline feed */
1100 		bucket20 = bucket10;
1101 		bucket21 = bucket11;
1102 		mbuf20 = mbuf10;
1103 		mbuf21 = mbuf11;
1104 		mbuf10 = mbuf00;
1105 		mbuf11 = mbuf01;
1106 		pkt20_index = pkt10_index;
1107 		pkt21_index = pkt11_index;
1108 		pkt10_index = pkt00_index;
1109 		pkt11_index = pkt01_index;
1110 
1111 		/* Pipeline stage 0 */
1112 		lookup2_stage0_with_odd_support(pkt00_index, pkt01_index,
1113 			mbuf00, mbuf01, pkts, pkts_mask, f);
1114 
1115 		/* Pipeline stage 1 */
1116 		lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f);
1117 
1118 		/* Pipeline stage 2 */
1119 		lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21,
1120 			bucket20, bucket21, pkts_mask_out, entries,
1121 			buckets_mask, buckets, keys, f);
1122 	}
1123 
1124 	/*
1125 	 * Pipeline flush
1126 	 *
1127 	 */
1128 	/* Pipeline feed */
1129 	bucket20 = bucket10;
1130 	bucket21 = bucket11;
1131 	mbuf20 = mbuf10;
1132 	mbuf21 = mbuf11;
1133 	mbuf10 = mbuf00;
1134 	mbuf11 = mbuf01;
1135 	pkt20_index = pkt10_index;
1136 	pkt21_index = pkt11_index;
1137 	pkt10_index = pkt00_index;
1138 	pkt11_index = pkt01_index;
1139 
1140 	/* Pipeline stage 1 */
1141 	lookup2_stage1(mbuf10, mbuf11, bucket10, bucket11, f);
1142 
1143 	/* Pipeline stage 2 */
1144 	lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21,
1145 		bucket20, bucket21, pkts_mask_out, entries,
1146 		buckets_mask, buckets, keys, f);
1147 
1148 	/* Pipeline feed */
1149 	bucket20 = bucket10;
1150 	bucket21 = bucket11;
1151 	mbuf20 = mbuf10;
1152 	mbuf21 = mbuf11;
1153 	pkt20_index = pkt10_index;
1154 	pkt21_index = pkt11_index;
1155 
1156 	/* Pipeline stage 2 */
1157 	lookup2_stage2_ext(pkt20_index, pkt21_index, mbuf20, mbuf21,
1158 		bucket20, bucket21, pkts_mask_out, entries,
1159 		buckets_mask, buckets, keys, f);
1160 
1161 grind_next_buckets:
1162 	/* Grind next buckets */
1163 	for ( ; buckets_mask; ) {
1164 		uint64_t buckets_mask_next = 0;
1165 
1166 		for ( ; buckets_mask; ) {
1167 			uint64_t pkt_mask;
1168 			uint32_t pkt_index;
1169 
1170 			pkt_index = __builtin_ctzll(buckets_mask);
1171 			pkt_mask = 1LLU << pkt_index;
1172 			buckets_mask &= ~pkt_mask;
1173 
1174 			lookup_grinder(pkt_index, buckets, keys, pkts_mask_out,
1175 				entries, buckets_mask_next, f);
1176 		}
1177 
1178 		buckets_mask = buckets_mask_next;
1179 	}
1180 
1181 	*lookup_hit_mask = pkts_mask_out;
1182 	RTE_TABLE_HASH_KEY32_STATS_PKTS_LOOKUP_MISS(f, n_pkts_in - __builtin_popcountll(pkts_mask_out));
1183 	return 0;
1184 } /* rte_table_hash_lookup_key32_ext() */
1185 
1186 static int
1187 rte_table_hash_key32_stats_read(void *table, struct rte_table_stats *stats, int clear)
1188 {
1189 	struct rte_table_hash *t = table;
1190 
1191 	if (stats != NULL)
1192 		memcpy(stats, &t->stats, sizeof(t->stats));
1193 
1194 	if (clear)
1195 		memset(&t->stats, 0, sizeof(t->stats));
1196 
1197 	return 0;
1198 }
1199 
1200 struct rte_table_ops rte_table_hash_key32_lru_ops = {
1201 	.f_create = rte_table_hash_create_key32_lru,
1202 	.f_free = rte_table_hash_free_key32_lru,
1203 	.f_add = rte_table_hash_entry_add_key32_lru,
1204 	.f_delete = rte_table_hash_entry_delete_key32_lru,
1205 	.f_add_bulk = NULL,
1206 	.f_delete_bulk = NULL,
1207 	.f_lookup = rte_table_hash_lookup_key32_lru,
1208 	.f_stats = rte_table_hash_key32_stats_read,
1209 };
1210 
1211 struct rte_table_ops rte_table_hash_key32_ext_ops = {
1212 	.f_create = rte_table_hash_create_key32_ext,
1213 	.f_free = rte_table_hash_free_key32_ext,
1214 	.f_add = rte_table_hash_entry_add_key32_ext,
1215 	.f_delete = rte_table_hash_entry_delete_key32_ext,
1216 	.f_add_bulk = NULL,
1217 	.f_delete_bulk = NULL,
1218 	.f_lookup = rte_table_hash_lookup_key32_ext,
1219 	.f_stats = rte_table_hash_key32_stats_read,
1220 };
1221