aboutsummaryrefslogtreecommitdiffstats
path: root/extras/deprecated/vppinfra
diff options
context:
space:
mode:
authorDave Barach <dave@barachs.net>2020-04-22 10:02:31 -0400
committerFlorin Coras <florin.coras@gmail.com>2020-04-22 19:58:20 +0000
commit053d0935244843199c433e92df666b4f79007727 (patch)
tree82fa449bd70f36a6d5ac75abcb3b89d8685bc868 /extras/deprecated/vppinfra
parentf932f97d7d8e55ab62aba0f18d59a9622c21efd8 (diff)
vppinfra: move unused code to extras/deprecated/vppinfra
Type: improvement Signed-off-by: Dave Barach <dave@barachs.net> Change-Id: Id28299a188feefa1899d835fd499f018af95d81b
Diffstat (limited to 'extras/deprecated/vppinfra')
-rw-r--r--extras/deprecated/vppinfra/anneal.c172
-rw-r--r--extras/deprecated/vppinfra/anneal.h89
-rw-r--r--extras/deprecated/vppinfra/cuckoo_16_8.h128
-rw-r--r--extras/deprecated/vppinfra/cuckoo_8_8.h120
-rw-r--r--extras/deprecated/vppinfra/cuckoo_common.h59
-rw-r--r--extras/deprecated/vppinfra/cuckoo_debug.h83
-rw-r--r--extras/deprecated/vppinfra/cuckoo_template.c1002
-rw-r--r--extras/deprecated/vppinfra/cuckoo_template.h460
-rw-r--r--extras/deprecated/vppinfra/fheap.c473
-rw-r--r--extras/deprecated/vppinfra/fheap.h140
-rw-r--r--extras/deprecated/vppinfra/flowhash_24_16.h77
-rw-r--r--extras/deprecated/vppinfra/flowhash_8_8.h67
-rw-r--r--extras/deprecated/vppinfra/flowhash_template.h608
-rw-r--r--extras/deprecated/vppinfra/pfhash.h276
-rw-r--r--extras/deprecated/vppinfra/phash.c1017
-rw-r--r--extras/deprecated/vppinfra/phash.h194
-rw-r--r--extras/deprecated/vppinfra/pipeline.h176
-rw-r--r--extras/deprecated/vppinfra/qhash.c859
-rw-r--r--extras/deprecated/vppinfra/qhash.h169
-rw-r--r--extras/deprecated/vppinfra/slist.c336
-rw-r--r--extras/deprecated/vppinfra/slist.h145
-rw-r--r--extras/deprecated/vppinfra/test_cuckoo_bihash.c451
-rw-r--r--extras/deprecated/vppinfra/test_cuckoo_template.c318
-rw-r--r--extras/deprecated/vppinfra/test_flowhash_template.c257
-rw-r--r--extras/deprecated/vppinfra/test_phash.c151
-rw-r--r--extras/deprecated/vppinfra/test_qhash.c335
-rw-r--r--extras/deprecated/vppinfra/test_slist.c228
-rw-r--r--extras/deprecated/vppinfra/test_timing_wheel.c391
-rw-r--r--extras/deprecated/vppinfra/test_vhash.c759
-rw-r--r--extras/deprecated/vppinfra/test_zvec.c119
-rw-r--r--extras/deprecated/vppinfra/unix-kelog.c415
-rw-r--r--extras/deprecated/vppinfra/vhash.c772
-rw-r--r--extras/deprecated/vppinfra/vhash.h850
-rw-r--r--extras/deprecated/vppinfra/xy.h56
-rw-r--r--extras/deprecated/vppinfra/zvec.c442
-rw-r--r--extras/deprecated/vppinfra/zvec.h166
36 files changed, 12360 insertions, 0 deletions
diff --git a/extras/deprecated/vppinfra/anneal.c b/extras/deprecated/vppinfra/anneal.c
new file mode 100644
index 00000000000..35d10946482
--- /dev/null
+++ b/extras/deprecated/vppinfra/anneal.c
@@ -0,0 +1,172 @@
+/*
+ Copyright (c) 2011 Cisco and/or its affiliates.
+
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+*/
+
+#include <vppinfra/anneal.h>
+
+/*
+ * Optimize an objective function by simulated annealing
+ *
+ * Here are a couple of short, easily-understood
+ * descriptions of simulated annealing:
+ *
+ * http://www.cs.sandia.gov/opt/survey/sa.html
+ * Numerical Recipes in C, 2nd ed., 444ff
+ *
+ * The description in the Wikipedia is not helpful.
+ *
+ * The algorithm tries to produce a decent answer to combinatorially
+ * explosive optimization problems by analogy to slow cooling
+ * of hot metal, aka annealing.
+ *
+ * There are (at least) three problem-dependent annealing parameters
+ * to consider:
+ *
+ * t0, the initial "temperature. Should be set so that the probability
+ * of accepting a transition to a higher cost configuration is
+ * initially about 0.8.
+ *
+ * ntemps, the number of temperatures to use. Each successive temperature
+ * is some fraction of the previous temperature.
+ *
+ * nmoves_per_temp, the number of configurations to try at each temperature
+ *
+ * It is a black art to set ntemps, nmoves_per_temp, and the rate
+ * at which the temperature drops. Go too fast with too few iterations,
+ * and the computation falls into a local minimum instead of the
+ * (desired) global minimum.
+ */
+
+void
+clib_anneal (clib_anneal_param_t * p)
+{
+ f64 t;
+ f64 cost, prev_cost, delta_cost, initial_cost, best_cost;
+ f64 random_accept, delta_cost_over_t;
+ f64 total_increase = 0.0, average_increase;
+ u32 i, j;
+ u32 number_of_increases = 0;
+ u32 accepted_this_temperature;
+ u32 best_saves_this_temperature;
+ int accept;
+
+ t = p->initial_temperature;
+ best_cost = initial_cost = prev_cost = p->anneal_metric (p->opaque);
+ p->anneal_save_best_configuration (p->opaque);
+
+ if (p->flags & CLIB_ANNEAL_VERBOSE)
+ fformat (stdout, "Initial cost %.2f\n", initial_cost);
+
+ for (i = 0; i < p->number_of_temperatures; i++)
+ {
+ accepted_this_temperature = 0;
+ best_saves_this_temperature = 0;
+
+ p->anneal_restore_best_configuration (p->opaque);
+ cost = best_cost;
+
+ for (j = 0; j < p->number_of_configurations_per_temperature; j++)
+ {
+ p->anneal_new_configuration (p->opaque);
+ cost = p->anneal_metric (p->opaque);
+
+ delta_cost = cost - prev_cost;
+
+ /* cost function looks better, accept this move */
+ if (p->flags & CLIB_ANNEAL_MINIMIZE)
+ accept = delta_cost < 0.0;
+ else
+ accept = delta_cost > 0.0;
+
+ if (accept)
+ {
+ if (p->flags & CLIB_ANNEAL_MINIMIZE)
+ if (cost < best_cost)
+ {
+ if (p->flags & CLIB_ANNEAL_VERBOSE)
+ fformat (stdout, "New best cost %.2f\n", cost);
+ best_cost = cost;
+ p->anneal_save_best_configuration (p->opaque);
+ best_saves_this_temperature++;
+ }
+
+ accepted_this_temperature++;
+ prev_cost = cost;
+ continue;
+ }
+
+ /* cost function worse, keep stats to suggest t0 */
+ total_increase += (p->flags & CLIB_ANNEAL_MINIMIZE) ?
+ delta_cost : -delta_cost;
+
+ number_of_increases++;
+
+ /*
+ * Accept a higher cost with Pr { e^(-(delta_cost / T)) },
+ * equivalent to rnd[0,1] < e^(-(delta_cost / T))
+ *
+ * AKA, the Boltzmann factor.
+ */
+ random_accept = random_f64 (&p->random_seed);
+
+ delta_cost_over_t = delta_cost / t;
+
+ if (random_accept < exp (-delta_cost_over_t))
+ {
+ accepted_this_temperature++;
+ prev_cost = cost;
+ continue;
+ }
+ p->anneal_restore_previous_configuration (p->opaque);
+ }
+
+ if (p->flags & CLIB_ANNEAL_VERBOSE)
+ {
+ fformat (stdout, "Temp %.2f, cost %.2f, accepted %d, bests %d\n", t,
+ prev_cost, accepted_this_temperature,
+ best_saves_this_temperature);
+ fformat (stdout, "Improvement %.2f\n", initial_cost - prev_cost);
+ fformat (stdout, "-------------\n");
+ }
+
+ t = t * p->temperature_step;
+ }
+
+ /*
+ * Empirically, one wants the probability of accepting a move
+ * at the initial temperature to be about 0.8.
+ */
+ average_increase = total_increase / (f64) number_of_increases;
+ p->suggested_initial_temperature = average_increase / 0.22; /* 0.22 = -ln (0.8) */
+
+ p->final_temperature = t;
+ p->final_metric = p->anneal_metric (p->opaque);
+
+ if (p->flags & CLIB_ANNEAL_VERBOSE)
+ {
+ fformat (stdout, "Average cost increase from a bad move: %.2f\n",
+ average_increase);
+ fformat (stdout, "Suggested t0 = %.2f\n",
+ p->suggested_initial_temperature);
+ }
+}
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/anneal.h b/extras/deprecated/vppinfra/anneal.h
new file mode 100644
index 00000000000..148d38ba551
--- /dev/null
+++ b/extras/deprecated/vppinfra/anneal.h
@@ -0,0 +1,89 @@
+/*
+ Copyright (c) 2011 Cisco and/or its affiliates.
+
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+*/
+
+#ifndef __included_anneal_h__
+#define __included_anneal_h__
+
+#include <vppinfra/clib.h>
+#include <vppinfra/format.h>
+#include <vppinfra/random.h>
+#include <math.h>
+
+typedef struct
+{
+ /* Initial temperature */
+ f64 initial_temperature;
+
+ /* Temperature fraction at each step, 0.95 is reasonable */
+ f64 temperature_step;
+
+ /* Number of temperatures used */
+ u32 number_of_temperatures;
+
+ /* Number of configurations tried at each temperature */
+ u32 number_of_configurations_per_temperature;
+
+ u32 flags;
+#define CLIB_ANNEAL_VERBOSE (1<<0)
+#define CLIB_ANNEAL_MINIMIZE (1<<1) /* mutually exclusive */
+#define CLIB_ANNEAL_MAXIMIZE (1<<2) /* mutually exclusive */
+
+ /* Random number seed, set to ensure repeatable results */
+ u32 random_seed;
+
+ /* Opaque data passed to callbacks */
+ void *opaque;
+
+ /* Final temperature (output) */
+ f64 final_temperature;
+
+ /* Final metric (output) */
+ f64 final_metric;
+
+ /* Suggested initial temperature (output) */
+ f64 suggested_initial_temperature;
+
+
+ /*--- Callbacks ---*/
+
+ /* objective function to minimize */
+ f64 (*anneal_metric) (void *opaque);
+
+ /* Generate a new configuration */
+ void (*anneal_new_configuration) (void *opaque);
+
+ /* Restore the previous configuration */
+ void (*anneal_restore_previous_configuration) (void *opaque);
+
+ /* Save best configuration found e.g at a certain temperature */
+ void (*anneal_save_best_configuration) (void *opaque);
+
+ /* restore best configuration found e.g at a certain temperature */
+ void (*anneal_restore_best_configuration) (void *opaque);
+
+} clib_anneal_param_t;
+
+void clib_anneal (clib_anneal_param_t * p);
+
+#endif /* __included_anneal_h__ */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/cuckoo_16_8.h b/extras/deprecated/vppinfra/cuckoo_16_8.h
new file mode 100644
index 00000000000..b667ff6cbb8
--- /dev/null
+++ b/extras/deprecated/vppinfra/cuckoo_16_8.h
@@ -0,0 +1,128 @@
+/*
+ * Copyright (c) 2017 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#undef CLIB_CUCKOO_TYPE
+
+#define CLIB_CUCKOO_TYPE _16_8
+#define CLIB_CUCKOO_KVP_PER_BUCKET (4)
+#define CLIB_CUCKOO_LOG2_KVP_PER_BUCKET (2)
+#define CLIB_CUCKOO_BFS_MAX_STEPS (2000)
+#define CLIB_CUCKOO_BFS_MAX_PATH_LENGTH (8)
+
+#ifndef __included_cuckoo_16_8_h__
+#define __included_cuckoo_16_8_h__
+
+#include <vppinfra/heap.h>
+#include <vppinfra/format.h>
+#include <vppinfra/pool.h>
+#include <vppinfra/xxhash.h>
+#include <vppinfra/cuckoo_debug.h>
+#include <vppinfra/cuckoo_common.h>
+
+#undef CLIB_CUCKOO_OPTIMIZE_PREFETCH
+#undef CLIB_CUCKOO_OPTIMIZE_UNROLL
+#undef CLIB_CUCKOO_OPTIMIZE_USE_COUNT_LIMITS_SEARCH
+#define CLIB_CUCKOO_OPTIMIZE_PREFETCH 1
+#define CLIB_CUCKOO_OPTIMIZE_UNROLL 1
+#define CLIB_CUCKOO_OPTIMIZE_USE_COUNT_LIMITS_SEARCH 1
+
+#if __SSE4_2__ && !defined (__i386__)
+#include <x86intrin.h>
+#endif
+
+/** 8 octet key, 8 octet key value pair */
+typedef struct
+{
+ u64 key[2]; /**< the key */
+ u64 value; /**< the value */
+} clib_cuckoo_kv_16_8_t;
+
+/** Decide if a clib_cuckoo_kv_16_8_t instance is free
+ @param v- pointer to the (key,value) pair
+*/
+always_inline int
+clib_cuckoo_kv_is_free_16_8 (const clib_cuckoo_kv_16_8_t * v)
+{
+ if (v->key[0] == ~0ULL && v->value == ~0ULL)
+ return 1;
+ return 0;
+}
+
+always_inline void
+clib_cuckoo_kv_set_free_16_8 (clib_cuckoo_kv_16_8_t * v)
+{
+ clib_memset (v, 0xff, sizeof (*v));
+}
+
+/** Format a clib_cuckoo_kv_16_8_t instance
+ @param s - u8 * vector under construction
+ @param args (vararg) - the (key,value) pair to format
+ @return s - the u8 * vector under construction
+*/
+always_inline u8 *
+format_cuckoo_kvp_16_8 (u8 * s, va_list * args)
+{
+ clib_cuckoo_kv_16_8_t *v = va_arg (*args, clib_cuckoo_kv_16_8_t *);
+
+ if (clib_cuckoo_kv_is_free_16_8 (v))
+ {
+ s = format (s, " -- empty -- ");
+ }
+ else
+ {
+ s =
+ format (s, "key %llu%llu value %llu", v->key[0], v->key[1], v->value);
+ }
+ return s;
+}
+
+always_inline u64
+clib_cuckoo_hash_16_8 (clib_cuckoo_kv_16_8_t * v)
+{
+#ifdef clib_crc32c_uses_intrinsics
+ return clib_crc32c ((u8 *) v->key, 16);
+#else
+ u64 tmp = v->key[0] ^ v->key[1];
+ return clib_xxhash (tmp);
+#endif
+}
+
+/** Compare two clib_cuckoo_kv_16_8_t instances
+ @param a - first key
+ @param b - second key
+*/
+always_inline int
+clib_cuckoo_key_compare_16_8 (u64 * a, u64 * b)
+{
+#if defined(CLIB_HAVE_VEC128) && defined(CLIB_HAVE_VEC128_UNALIGNED_LOAD_STORE)
+ u64x2 v;
+ v = u64x2_load_unaligned (a) ^ u64x2_load_unaligned (b);
+ return u64x2_is_all_zero (v);
+#else
+ return ((a[0] ^ b[0]) | (a[1] ^ b[1])) == 0;
+#endif
+}
+
+#undef __included_cuckoo_template_h__
+#include <vppinfra/cuckoo_template.h>
+
+#endif /* __included_cuckoo_16_8_h__ */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/cuckoo_8_8.h b/extras/deprecated/vppinfra/cuckoo_8_8.h
new file mode 100644
index 00000000000..67af79c0e50
--- /dev/null
+++ b/extras/deprecated/vppinfra/cuckoo_8_8.h
@@ -0,0 +1,120 @@
+/*
+ * Copyright (c) 2017 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#undef CLIB_CUCKOO_TYPE
+
+#define CLIB_CUCKOO_TYPE _8_8
+#define CLIB_CUCKOO_KVP_PER_BUCKET (4)
+#define CLIB_CUCKOO_LOG2_KVP_PER_BUCKET (2)
+#define CLIB_CUCKOO_BFS_MAX_STEPS (2000)
+#define CLIB_CUCKOO_BFS_MAX_PATH_LENGTH (8)
+
+#ifndef __included_cuckoo_8_8_h__
+#define __included_cuckoo_8_8_h__
+
+#include <vppinfra/heap.h>
+#include <vppinfra/format.h>
+#include <vppinfra/pool.h>
+#include <vppinfra/xxhash.h>
+#include <vppinfra/cuckoo_debug.h>
+#include <vppinfra/cuckoo_common.h>
+
+#undef CLIB_CUCKOO_OPTIMIZE_PREFETCH
+#undef CLIB_CUCKOO_OPTIMIZE_UNROLL
+#undef CLIB_CUCKOO_OPTIMIZE_USE_COUNT_LIMITS_SEARCH
+#define CLIB_CUCKOO_OPTIMIZE_PREFETCH 1
+#define CLIB_CUCKOO_OPTIMIZE_UNROLL 1
+#define CLIB_CUCKOO_OPTIMIZE_USE_COUNT_LIMITS_SEARCH 1
+
+#if __SSE4_2__ && !defined (__i386__)
+#include <x86intrin.h>
+#endif
+
+/** 8 octet key, 8 octet key value pair */
+typedef struct
+{
+ u64 key; /**< the key */
+ u64 value; /**< the value */
+} clib_cuckoo_kv_8_8_t;
+
+/** Decide if a clib_cuckoo_kv_8_8_t instance is free
+ @param v- pointer to the (key,value) pair
+*/
+always_inline int
+clib_cuckoo_kv_is_free_8_8 (const clib_cuckoo_kv_8_8_t * v)
+{
+ if (v->key == ~0ULL && v->value == ~0ULL)
+ return 1;
+ return 0;
+}
+
+always_inline void
+clib_cuckoo_kv_set_free_8_8 (clib_cuckoo_kv_8_8_t * v)
+{
+ clib_memset (v, 0xff, sizeof (*v));
+}
+
+/** Format a clib_cuckoo_kv_8_8_t instance
+ @param s - u8 * vector under construction
+ @param args (vararg) - the (key,value) pair to format
+ @return s - the u8 * vector under construction
+*/
+always_inline u8 *
+format_cuckoo_kvp_8_8 (u8 * s, va_list * args)
+{
+ clib_cuckoo_kv_8_8_t *v = va_arg (*args, clib_cuckoo_kv_8_8_t *);
+
+ if (clib_cuckoo_kv_is_free_8_8 (v))
+ {
+ s = format (s, " -- empty -- ");
+ }
+ else
+ {
+ s = format (s, "key %llu value %llu", v->key, v->value);
+ }
+ return s;
+}
+
+always_inline u64
+clib_cuckoo_hash_8_8 (clib_cuckoo_kv_8_8_t * v)
+{
+#if defined(clib_crc32c_uses_intrinsics) && !defined (__i386__)
+ return crc32_u64 (0, v->key);
+#else
+ return clib_xxhash (v->key);
+#endif
+}
+
+/** Compare two clib_cuckoo_kv_8_8_t instances
+ @param a - first key
+ @param b - second key
+*/
+always_inline int
+clib_cuckoo_key_compare_8_8 (u64 a, u64 b)
+{
+ return a == b;
+}
+
+#undef __included_cuckoo_template_h__
+#include <vppinfra/cuckoo_template.h>
+
+#endif /* __included_cuckoo_8_8_h__ */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/cuckoo_common.h b/extras/deprecated/vppinfra/cuckoo_common.h
new file mode 100644
index 00000000000..ea0fc30084b
--- /dev/null
+++ b/extras/deprecated/vppinfra/cuckoo_common.h
@@ -0,0 +1,59 @@
+/*
+ Copyright (c) 2017 Cisco and/or its affiliates.
+
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+*/
+
+/*
+ * Note: to instantiate the template multiple times in a single file,
+ * #undef __included_cuckoo_template_h__...
+ */
+#ifndef __included_cuckoo_common_h__
+#define __included_cuckoo_common_h__
+
+#include <vppinfra/types.h>
+
+#define CLIB_CUCKOO_OPTIMIZE_PREFETCH 1
+#define CLIB_CUCKOO_OPTIMIZE_UNROLL 1
+#define CLIB_CUCKOO_OPTIMIZE_USE_COUNT_LIMITS_SEARCH 1
+
+#define foreach_clib_cuckoo_error(F) \
+ F (CLIB_CUCKOO_ERROR_SUCCESS, 0, "success") \
+ F (CLIB_CUCKOO_ERROR_NOT_FOUND, -1, "object not found") \
+ F (CLIB_CUCKOO_ERROR_AGAIN, -2, "object busy")
+
+typedef enum
+{
+#define F(n, v, s) n = v,
+ foreach_clib_cuckoo_error (F)
+#undef F
+} clib_cuckoo_error_e;
+
+typedef struct
+{
+ uword bucket1;
+ uword bucket2;
+ u8 reduced_hash;
+} clib_cuckoo_lookup_info_t;
+
+#endif /* __included_cuckoo_common_h__ */
+
+/** @endcond */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/cuckoo_debug.h b/extras/deprecated/vppinfra/cuckoo_debug.h
new file mode 100644
index 00000000000..eb236509935
--- /dev/null
+++ b/extras/deprecated/vppinfra/cuckoo_debug.h
@@ -0,0 +1,83 @@
+/*
+ * Copyright (c) 2017 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/**
+ * @file
+ * @brief cuckoo debugs
+ */
+#ifndef __included_cuckoo_debug_h__
+#define __included_cuckoo_debug_h__
+
+/* controls debug counters */
+#define CLIB_CUCKOO_DEBUG_COUNTERS (0)
+
+/* controls debug prints */
+#define CLIB_CUCKOO_DEBUG (0)
+
+/* controls garbage collection related debug prints */
+#define CLIB_CUCKOO_DEBUG_GC (0)
+
+#if CLIB_CUCKOO_DEBUG
+#define CLIB_CUCKOO_DEBUG_FILE_DEF \
+ static const char *__file = NULL; \
+ { \
+ __file = strrchr (__FILE__, '/'); \
+ if (__file) \
+ { \
+ ++__file; \
+ } \
+ else \
+ { \
+ __file = __FILE__; \
+ } \
+ }
+
+#define CLIB_CUCKOO_DBG(fmt, ...) \
+ do \
+ { \
+ CLIB_CUCKOO_DEBUG_FILE_DEF \
+ static u8 *_s = NULL; \
+ _s = format (_s, "DBG:%s:%d:%s():" fmt, __file, __LINE__, __func__, \
+ ##__VA_ARGS__); \
+ printf ("%.*s\n", vec_len (_s), _s); \
+ vec_reset_length (_s); \
+ } \
+ while (0);
+
+#define CLIB_CUCKOO_ERR(fmt, ...) \
+ do \
+ { \
+ CLIB_CUCKOO_DEBUG_FILE_DEF \
+ static u8 *_s = NULL; \
+ _s = format (_s, "ERR:%s:%d:%s():" fmt, __file, __LINE__, __func__, \
+ ##__VA_ARGS__); \
+ printf ("%.*s\n", vec_len (_s), _s); \
+ vec_reset_length (_s); \
+ } \
+ while (0);
+
+#else
+#define CLIB_CUCKOO_DBG(...)
+#define CLIB_CUCKOO_ERR(...)
+#endif
+
+#endif /* __included_cuckoo_debug_h__ */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/cuckoo_template.c b/extras/deprecated/vppinfra/cuckoo_template.c
new file mode 100644
index 00000000000..8cd2a2be2b5
--- /dev/null
+++ b/extras/deprecated/vppinfra/cuckoo_template.c
@@ -0,0 +1,1002 @@
+/*
+ * Copyright (c) 2017 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/*
+ * cuckoo hash implementation based on paper
+ * 'Algorithmic Improvements for Fast Concurrent Cuckoo Hashing'
+ * by Xiaozhou Li, David G. Andersen, Michael Kaminsky, Michael J. Freedman
+ * and their libcuckoo implementation (https://github.com/efficient/libcuckoo)
+ */
+
+#include <vppinfra/vec.h>
+
+int CV (clib_cuckoo_search) (CVT (clib_cuckoo) * h,
+ CVT (clib_cuckoo_kv) * search_v,
+ CVT (clib_cuckoo_kv) * return_v)
+{
+ CVT (clib_cuckoo_kv) tmp = *search_v;
+ int rv = CV (clib_cuckoo_search_inline) (h, &tmp);
+ if (CLIB_CUCKOO_ERROR_SUCCESS == rv)
+ {
+ *return_v = tmp;
+ }
+ return rv;
+}
+
+static
+CVT (clib_cuckoo_bucket) *
+CV (clib_cuckoo_bucket_at_index) (CVT (clib_cuckoo) * h, uword bucket)
+{
+ return vec_elt_at_index (h->buckets, bucket);
+}
+
+static uword CV (clib_cuckoo_get_nbuckets) (CVT (clib_cuckoo) * h)
+{
+ return vec_len (h->buckets);
+}
+
+static inline uword
+CV (clib_cuckoo_elt_in_bucket_to_offset) (CVT (clib_cuckoo_bucket) * b,
+ CVT (clib_cuckoo_kv) * e)
+{
+ ASSERT (e >= b->elts);
+ ASSERT (e <= &b->elts[sizeof (b->elts) / sizeof (b->elts[0]) - 1]);
+ return e - b->elts;
+}
+
+u8 *CV (format_cuckoo_elt) (u8 * s, va_list * args)
+{
+ CVT (clib_cuckoo_kv) * elt = va_arg (*args, CVT (clib_cuckoo_kv) *);
+ unsigned reduced_hash = va_arg (*args, unsigned);
+ if (CV (clib_cuckoo_kv_is_free) (elt))
+ {
+ s = format (s, "[ -- empty -- ]");
+ }
+ else
+ {
+ s = format (s, "[%U, reduced hash: %u]", CV (format_cuckoo_kvp), elt,
+ reduced_hash);
+ }
+ return s;
+}
+
+u8 *CV (format_cuckoo_bucket) (u8 * s, va_list * args)
+{
+ CVT (clib_cuckoo_bucket) * bucket =
+ va_arg (*args, CVT (clib_cuckoo_bucket) *);
+ int i = 0;
+
+ /* *INDENT-OFF* */
+ clib_cuckoo_bucket_foreach_idx (i)
+ {
+ CVT (clib_cuckoo_kv) *elt = bucket->elts + i;
+ s = format (s, "bucket %p, offset %d: %U\n", bucket, i,
+ CV (format_cuckoo_elt), elt, bucket->reduced_hashes[i]);
+ }
+ /* *INDENT-ON* */
+ clib_cuckoo_bucket_aux_t aux = bucket->aux;
+ s = format (s, "version: %lld, use count: %d\n",
+ clib_cuckoo_bucket_aux_get_version (aux),
+ clib_cuckoo_bucket_aux_get_use_count (aux));
+ return s;
+}
+
+#if CLIB_CUCKOO_DEBUG
+static void CV (clib_cuckoo_deep_self_check) (CVT (clib_cuckoo) * h)
+{
+ CVT (clib_cuckoo_bucket) * bucket;
+ uword bucket_idx = 0;
+ /* *INDENT-OFF* */
+ clib_cuckoo_foreach_bucket (bucket, h, {
+ int i = 0;
+ int used = 0;
+ clib_cuckoo_bucket_foreach_idx (i)
+ {
+ CVT (clib_cuckoo_kv) *elt = bucket->elts + i;
+ if (!CV (clib_cuckoo_kv_is_free) (elt))
+ {
+ u64 hash = CV (clib_cuckoo_hash) (elt);
+ clib_cuckoo_lookup_info_t lookup =
+ CV (clib_cuckoo_calc_lookup) (h->buckets, hash);
+ CVT (clib_cuckoo_kv) kv = *elt;
+ int rv = CV (clib_cuckoo_search) (h, &kv, &kv);
+ if (CLIB_CUCKOO_ERROR_SUCCESS != rv)
+ {
+ CLIB_CUCKOO_DBG ("Search for elt `%U' failed!",
+ CV (format_cuckoo_elt), elt,
+ bucket->reduced_hashes[i]);
+ CLIB_CUCKOO_DBG ("%U", CV (format_cuckoo), h, 1);
+ }
+ ASSERT (lookup.bucket1 == bucket_idx || lookup.bucket2 == bucket_idx);
+ ASSERT (CLIB_CUCKOO_ERROR_SUCCESS == rv);
+ ++used;
+ }
+ }
+ clib_cuckoo_bucket_aux_t aux = bucket->aux;
+ ASSERT (used == clib_cuckoo_bucket_aux_get_use_count (aux));
+ ++bucket_idx;
+ });
+ /* *INDENT-ON* */
+ // CLIB_CUCKOO_DBG ("Deep self check passed: %U", CV (format_cuckoo), h);
+}
+
+#define CLIB_CUCKOO_DEEP_SELF_CHECK(h) CV (clib_cuckoo_deep_self_check) (h)
+#define CLIB_CUCKOO_ASSERT_BUCKET_SORTED(b) \
+ do \
+ { \
+ int i; \
+ int min_free = CLIB_CUCKOO_KVP_PER_BUCKET; \
+ int max_used = 0; \
+ clib_cuckoo_bucket_foreach_idx (i) \
+ { \
+ if (!CV (clib_cuckoo_kv_is_free) (b->elts + i)) \
+ { \
+ max_used = i; \
+ } \
+ if (CV (clib_cuckoo_kv_is_free) (b->elts + \
+ (CLIB_CUCKOO_KVP_PER_BUCKET - i))) \
+ { \
+ min_free = i; \
+ } \
+ } \
+ ASSERT (min_free > max_used); \
+ } \
+ while (0)
+
+#else
+#define CLIB_CUCKOO_DEEP_SELF_CHECK(h)
+#define CLIB_CUCKOO_ASSERT_BUCKET_SORTED(b)
+#endif
+
+void CV (clib_cuckoo_init) (CVT (clib_cuckoo) * h, const char *name,
+ uword nbuckets,
+ void (*garbage_callback) (CVT (clib_cuckoo) *,
+ void *),
+ void *garbage_ctx)
+{
+ uword log2_nbuckets = max_log2 (nbuckets);
+ nbuckets = 1ULL << (log2_nbuckets);
+ CLIB_CUCKOO_DBG ("New cuckoo, adjusted nbuckets %wu", nbuckets);
+ CVT (clib_cuckoo_bucket) * buckets = NULL;
+ vec_validate_aligned (buckets, nbuckets - 1, CLIB_CACHE_LINE_BYTES);
+ ASSERT (nbuckets == vec_len (buckets));
+ h->buckets = buckets;
+ clib_spinlock_init (&h->writer_lock);
+ /* mark all elements free ... */
+ CVT (clib_cuckoo_bucket) * bucket;
+ /* *INDENT-OFF* */
+ clib_cuckoo_foreach_bucket (
+ bucket, h, { clib_memset (bucket->elts, 0xff, sizeof (bucket->elts)); });
+ /* *INDENT-ON* */
+ h->name = name;
+ h->garbage_callback = garbage_callback;
+ h->garbage_ctx = garbage_ctx;
+}
+
+void CV (clib_cuckoo_free) (CVT (clib_cuckoo) * h)
+{
+ clib_memset (h, 0, sizeof (*h));
+}
+
+static clib_cuckoo_bucket_aux_t
+CV (clib_cuckoo_bucket_version_bump_and_lock) (CVT (clib_cuckoo_bucket) * b)
+{
+ clib_cuckoo_bucket_aux_t aux = b->aux;
+ u64 version = clib_cuckoo_bucket_aux_get_version (aux);
+ u8 use_count = clib_cuckoo_bucket_aux_get_use_count (aux);
+ u8 writer_flag = clib_cuckoo_bucket_aux_get_writer_flag (aux);
+ ASSERT (0 == writer_flag);
+ aux = clib_cuckoo_bucket_aux_pack (version + 1, use_count, 1);
+ b->aux = aux;
+ return aux;
+}
+
+static void CV (clib_cuckoo_bucket_unlock) (CVT (clib_cuckoo_bucket) * b,
+ clib_cuckoo_bucket_aux_t aux)
+{
+ u64 version = clib_cuckoo_bucket_aux_get_version (aux);
+ u8 use_count = clib_cuckoo_bucket_aux_get_use_count (aux);
+ u8 writer_flag = clib_cuckoo_bucket_aux_get_writer_flag (aux);
+ ASSERT (1 == writer_flag);
+ aux = clib_cuckoo_bucket_aux_pack (version, use_count, 0);
+ b->aux = aux;
+}
+
+#define CLIB_CUCKOO_DEBUG_PATH (1)
+#define CLIB_CUCKOO_DEBUG_PATH_DETAIL (0)
+
+#if CLIB_CUCKOO_DEBUG && CLIB_CUCKOO_DEBUG_PATH
+static u8 *CV (format_cuckoo_path) (u8 * s, va_list * args);
+#endif
+
+static clib_cuckoo_path_t *CV (clib_cuckoo_path_get) (CVT (clib_cuckoo) * h)
+{
+ clib_cuckoo_path_t *path;
+ pool_get (h->paths, path);
+ clib_memset (path, 0, sizeof (*path));
+#if CLIB_CUCKOO_DEBUG_PATH_DETAIL
+ CLIB_CUCKOO_DBG ("Get path @%lu", (long unsigned) (path - h->paths));
+#endif
+ return path;
+}
+
+static void CV (clib_cuckoo_path_put) (CVT (clib_cuckoo) * h, uword path_idx)
+{
+ clib_cuckoo_path_t *path = pool_elt_at_index (h->paths, path_idx);
+#if CLIB_CUCKOO_DEBUG_PATH_DETAIL
+ CLIB_CUCKOO_DBG ("Put path @%lu", (long unsigned) (path - h->paths));
+#endif
+ pool_put (h->paths, path);
+}
+
+static clib_cuckoo_path_t *CV (clib_cuckoo_path_begin) (CVT (clib_cuckoo) * h,
+ uword bucket,
+ uword next_offset)
+{
+ ASSERT (next_offset < CLIB_CUCKOO_KVP_PER_BUCKET);
+ clib_cuckoo_path_t *new_path = CV (clib_cuckoo_path_get) (h);
+ new_path->length = 1;
+ new_path->data = next_offset;
+ new_path->start = bucket;
+ new_path->bucket = bucket;
+#if CLIB_CUCKOO_DEBUG_PATH
+ CLIB_CUCKOO_DBG ("Create new path @%wu, length: %u data: %llu bucket: %wu "
+ "next-offset: %wu",
+ new_path - h->paths, new_path->length,
+ (long long unsigned) new_path->data, new_path->bucket,
+ next_offset);
+#endif
+ return new_path;
+}
+
+/**
+ * create a new path based on existing path extended by adding a bucket
+ * and offset
+ */
+static uword CV (clib_cuckoo_path_extend) (CVT (clib_cuckoo) * h,
+ uword path_idx, uword bucket,
+ unsigned offset)
+{
+ ASSERT (offset < CLIB_CUCKOO_KVP_PER_BUCKET);
+ clib_cuckoo_path_t *new_path = CV (clib_cuckoo_path_get) (h);
+ uword new_path_idx = new_path - h->paths;
+ clib_cuckoo_path_t *path = pool_elt_at_index (h->paths, path_idx);
+ new_path->start = path->start;
+ new_path->length = path->length + 1;
+ new_path->data = (path->data << CLIB_CUCKOO_LOG2_KVP_PER_BUCKET) + offset;
+ new_path->bucket = bucket;
+#if CLIB_CUCKOO_DEBUG_PATH
+ CLIB_CUCKOO_DBG ("Extend path @%wu, new path @%wu, %U", path_idx,
+ new_path_idx, CV (format_cuckoo_path), h, new_path_idx);
+#endif
+ return new_path_idx;
+}
+
+/** return the offset of the last element in the path */
+static uword CV (clib_cuckoo_path_peek_offset) (const clib_cuckoo_path_t *
+ path)
+{
+ ASSERT (path->length > 0);
+ uword mask = (1 << CLIB_CUCKOO_LOG2_KVP_PER_BUCKET) - 1;
+ uword offset = path->data & mask;
+ return offset;
+}
+
+static
+CVT (clib_cuckoo_kv) *
+CV (clib_cuckoo_bucket_find_empty) (CVT (clib_cuckoo_bucket) * bucket)
+{
+ clib_cuckoo_bucket_aux_t aux = bucket->aux;
+ u8 use_count = clib_cuckoo_bucket_aux_get_use_count (aux);
+ if (use_count < CLIB_CUCKOO_KVP_PER_BUCKET)
+ {
+ return bucket->elts + use_count;
+ }
+ return NULL;
+}
+
+/**
+ * walk the cuckoo path two ways,
+ * first backwards, extracting offsets,
+ * then forward, extracting buckets
+ *
+ * buckets and offsets are arrays filled with elements extracted from path
+ * the arrays must be able to contain CLIB_CUCKOO_BFS_MAX_PATH_LENGTH elements
+ */
+static void
+clib_cuckoo_path_walk (CVT (clib_cuckoo) * h, uword path_idx,
+ uword * buckets, uword * offsets)
+{
+ clib_cuckoo_path_t *path = pool_elt_at_index (h->paths, path_idx);
+ ASSERT (path->length > 0);
+ u64 data = path->data;
+ uword mask = (1 << CLIB_CUCKOO_LOG2_KVP_PER_BUCKET) - 1;
+ uword i;
+ for (i = path->length; i > 0; --i)
+ {
+ uword offset = data & mask;
+ offsets[i - 1] = offset;
+ data >>= CLIB_CUCKOO_LOG2_KVP_PER_BUCKET;
+ }
+ buckets[0] = path->start;
+ for (i = 1; i < path->length; ++i)
+ {
+ CVT (clib_cuckoo_bucket) * b =
+ CV (clib_cuckoo_bucket_at_index) (h, buckets[i - 1]);
+ buckets[i] =
+ clib_cuckoo_get_other_bucket (CV (clib_cuckoo_get_nbuckets) (h),
+ buckets[i - 1],
+ b->reduced_hashes[offsets[i - 1]]);
+ }
+}
+
+#if CLIB_CUCKOO_DEBUG && CLIB_CUCKOO_DEBUG_PATH
+static u8 *CV (format_cuckoo_path) (u8 * s, va_list * args)
+{
+ CVT (clib_cuckoo) * h = va_arg (*args, CVT (clib_cuckoo) *);
+ uword path_idx = va_arg (*args, uword);
+ clib_cuckoo_path_t *p = pool_elt_at_index (h->paths, path_idx);
+ uword buckets[CLIB_CUCKOO_BFS_MAX_PATH_LENGTH];
+ uword offsets[CLIB_CUCKOO_BFS_MAX_PATH_LENGTH];
+ clib_cuckoo_path_walk (h, path_idx, buckets, offsets);
+ s = format (s, "length %u: ", p->length);
+ for (uword i = p->length - 1; i > 0; --i)
+ {
+ s = format (s, "%wu[%wu]->", buckets[i], offsets[i]);
+ }
+ if (p->length)
+ {
+ s = format (s, "%wu[%wu]", buckets[0], offsets[0]);
+ }
+ return s;
+}
+#endif
+
+/*
+ * perform breadth-first search in the cuckoo hash, finding the closest
+ * empty slot, i.e. one which requires minimum swaps to move it
+ * to one of the buckets provided
+ */
+static int CV (clib_cuckoo_find_empty_slot_bfs) (CVT (clib_cuckoo) * h,
+ clib_cuckoo_lookup_info_t *
+ lookup, uword * path_idx_out,
+ uword * found_bucket,
+ CVT (clib_cuckoo_kv) *
+ *found_elt)
+{
+ uword *tail;
+ ASSERT (!vec_len (h->bfs_search_queue));
+ clib_cuckoo_path_t *tmp;
+ pool_flush (tmp, h->paths,);
+ int rv = CLIB_CUCKOO_ERROR_AGAIN;
+ int counter = 0;
+ /* start by creating paths starting in each of the buckets ... */
+ vec_add2 (h->bfs_search_queue, tail, CLIB_CUCKOO_KVP_PER_BUCKET);
+ int i;
+ for (i = 0; i < CLIB_CUCKOO_KVP_PER_BUCKET; ++i)
+ {
+ clib_cuckoo_path_t *path =
+ CV (clib_cuckoo_path_begin) (h, lookup->bucket1, i);
+ tail[i] = path - h->paths;
+ }
+ if (lookup->bucket1 != lookup->bucket2)
+ {
+ vec_add2 (h->bfs_search_queue, tail, CLIB_CUCKOO_KVP_PER_BUCKET);
+ for (i = 0; i < CLIB_CUCKOO_KVP_PER_BUCKET; ++i)
+ {
+ clib_cuckoo_path_t *path =
+ CV (clib_cuckoo_path_begin) (h, lookup->bucket2, i);
+ tail[i] = path - h->paths;
+ }
+ }
+ while (1)
+ {
+ if (counter >= CLIB_CUCKOO_BFS_MAX_STEPS)
+ {
+#if CLIB_CUCKOO_DEBUG_COUNTERS
+ ++h->steps_exceeded;
+#endif
+ break;
+ }
+ if (counter >= vec_len (h->bfs_search_queue))
+ {
+#if CLIB_CUCKOO_DEBUG_COUNTERS
+ ++h->bfs_queue_emptied;
+#endif
+ break;
+ }
+ const uword path_idx = vec_elt (h->bfs_search_queue, counter);
+ const clib_cuckoo_path_t *path = pool_elt_at_index (h->paths, path_idx);
+#if CLIB_CUCKOO_DEBUG_PATH
+ CLIB_CUCKOO_DBG ("Examine path @%wu: %U", path_idx,
+ CV (format_cuckoo_path), h, path_idx);
+#endif
+ /* TODO prefetch ? */
+ /* search the alternative bucket for free space */
+ int offset = CV (clib_cuckoo_path_peek_offset) (path);
+ CVT (clib_cuckoo_bucket) * bucket =
+ CV (clib_cuckoo_bucket_at_index) (h, path->bucket);
+ uword other_bucket =
+ clib_cuckoo_get_other_bucket (CV (clib_cuckoo_get_nbuckets) (h),
+ path->bucket,
+ bucket->reduced_hashes[offset]);
+ CLIB_CUCKOO_DBG
+ ("Path ends in bucket %wu, offset #%wu, other bucket is %wu",
+ path->bucket, CV (clib_cuckoo_path_peek_offset) (path),
+ other_bucket);
+ if (path->bucket != other_bucket)
+ {
+ if ((*found_elt =
+ CV (clib_cuckoo_bucket_find_empty) (CV
+ (clib_cuckoo_bucket_at_index)
+ (h, other_bucket))))
+ {
+ /* found empty element */
+ *found_bucket = other_bucket;
+ *path_idx_out = path_idx;
+ rv = CLIB_CUCKOO_ERROR_SUCCESS;
+#if CLIB_CUCKOO_DEBUG_PATH
+ CLIB_CUCKOO_DBG ("Bucket with empty slot:\n%U",
+ CV (format_cuckoo_bucket),
+ CV (clib_cuckoo_bucket_at_index) (h,
+ other_bucket));
+#endif
+ goto out;
+ }
+ /* extend the current path with possible next buckets and add to
+ * queue */
+ if (path->length < CLIB_CUCKOO_BFS_MAX_PATH_LENGTH &&
+ vec_len (h->bfs_search_queue) < CLIB_CUCKOO_BFS_MAX_STEPS)
+ {
+ uword *tail;
+ vec_add2 (h->bfs_search_queue, tail,
+ CLIB_CUCKOO_KVP_PER_BUCKET);
+ for (i = 0; i < CLIB_CUCKOO_KVP_PER_BUCKET; ++i)
+ {
+ uword new_path_idx =
+ CV (clib_cuckoo_path_extend) (h, path_idx, other_bucket,
+ i);
+ tail[i] = new_path_idx;
+ }
+ }
+ }
+ else
+ {
+ CLIB_CUCKOO_DBG ("Discard path @%wu, loop detected", path_idx);
+ }
+ /* done with this path - put back to pool for later reuse */
+ CV (clib_cuckoo_path_put) (h, path_idx);
+ ++counter;
+ }
+out:
+ vec_reset_length (h->bfs_search_queue);
+ return rv;
+}
+
+static void CV (clib_cuckoo_swap_elts_in_bucket) (CVT (clib_cuckoo_bucket) *
+ b, uword e1, uword e2)
+{
+ CVT (clib_cuckoo_kv) kv;
+ clib_memcpy (&kv, b->elts + e1, sizeof (kv));
+ clib_memcpy (b->elts + e1, b->elts + e2, sizeof (kv));
+ clib_memcpy (b->elts + e2, &kv, sizeof (kv));
+ u8 reduced_hash = b->reduced_hashes[e1];
+ b->reduced_hashes[e1] = b->reduced_hashes[e2];
+ b->reduced_hashes[e2] = reduced_hash;
+}
+
+static void CV (clib_cuckoo_bucket_tidy) (CVT (clib_cuckoo_bucket) * b)
+{
+ int i = 0;
+ int j = CLIB_CUCKOO_KVP_PER_BUCKET - 1;
+ while (i != j)
+ {
+ int min_free = i;
+ int max_used = j;
+ while (!CV (clib_cuckoo_kv_is_free) (&b->elts[min_free]))
+ {
+ ++min_free;
+ }
+ while (CV (clib_cuckoo_kv_is_free) (&b->elts[max_used]))
+ {
+ --max_used;
+ }
+ if (min_free < max_used)
+ {
+ CV (clib_cuckoo_swap_elts_in_bucket) (b, min_free, max_used);
+ i = min_free + 1;
+ j = max_used - 1;
+ }
+ else
+ {
+ break;
+ }
+ }
+}
+
+static void CV (clib_cuckoo_free_locked_elt) (CVT (clib_cuckoo_kv) * elt)
+{
+ /*
+ * FIXME - improve performance by getting rid of this clib_memset - make all
+ * functions in this file not rely on clib_cuckoo_kv_is_free but instead
+ * take use_count into account */
+ clib_memset (elt, 0xff, sizeof (*elt));
+}
+
+static void CV (clib_cuckoo_free_elt_in_bucket) (CVT (clib_cuckoo_bucket) * b,
+ CVT (clib_cuckoo_kv) * elt)
+{
+ clib_cuckoo_bucket_aux_t aux =
+ CV (clib_cuckoo_bucket_version_bump_and_lock) (b);
+ int use_count = clib_cuckoo_bucket_aux_get_use_count (aux);
+ int offset = elt - b->elts;
+ ASSERT (offset < use_count);
+ CV (clib_cuckoo_free_locked_elt) (elt);
+ if (offset != use_count - 1)
+ {
+ CV (clib_cuckoo_bucket_tidy) (b);
+ }
+ aux = clib_cuckoo_bucket_aux_set_use_count (aux, use_count - 1);
+ CV (clib_cuckoo_bucket_unlock) (b, aux);
+}
+
+static void CV (clib_cuckoo_set_locked_elt) (CVT (clib_cuckoo_bucket) * b,
+ CVT (clib_cuckoo_kv) * elt,
+ CVT (clib_cuckoo_kv) * kvp,
+ u8 reduced_hash)
+{
+ int offset = CV (clib_cuckoo_elt_in_bucket_to_offset) (b, elt);
+ clib_memcpy (elt, kvp, sizeof (*elt));
+ b->reduced_hashes[offset] = reduced_hash;
+ CLIB_CUCKOO_DBG ("Set bucket %p, offset %d, %U", b, offset,
+ CV (format_cuckoo_elt), elt, b->reduced_hashes[offset]);
+}
+
+static void CV (clib_cuckoo_set_elt) (CVT (clib_cuckoo_bucket) * b,
+ CVT (clib_cuckoo_kv) * elt,
+ CVT (clib_cuckoo_kv) * kvp,
+ u8 reduced_hash)
+{
+ clib_cuckoo_bucket_aux_t aux =
+ CV (clib_cuckoo_bucket_version_bump_and_lock) (b);
+ CV (clib_cuckoo_set_locked_elt) (b, elt, kvp, reduced_hash);
+ CV (clib_cuckoo_bucket_unlock) (b, aux);
+}
+
+static int CV (clib_cuckoo_add_slow) (CVT (clib_cuckoo) * h,
+ CVT (clib_cuckoo_kv) * kvp,
+ clib_cuckoo_lookup_info_t * lookup,
+ u8 reduced_hash)
+{
+ uword path_idx;
+ uword empty_bucket_idx;
+ CVT (clib_cuckoo_kv) * empty_elt;
+ int rv = CV (clib_cuckoo_find_empty_slot_bfs) (h, lookup, &path_idx,
+ &empty_bucket_idx,
+ &empty_elt);
+ if (CLIB_CUCKOO_ERROR_SUCCESS == rv)
+ {
+ uword buckets[CLIB_CUCKOO_BFS_MAX_PATH_LENGTH];
+ uword offsets[CLIB_CUCKOO_BFS_MAX_PATH_LENGTH];
+ clib_cuckoo_path_walk (h, path_idx, buckets, offsets);
+ /*
+ * walk back the path, moving the free element forward to one of our
+ * buckets ...
+ */
+ clib_cuckoo_path_t *path = pool_elt_at_index (h->paths, path_idx);
+ CVT (clib_cuckoo_bucket) * empty_bucket =
+ CV (clib_cuckoo_bucket_at_index) (h, empty_bucket_idx);
+ int i;
+ for (i = path->length - 1; i >= 0; --i)
+ {
+ /* copy the key-value in path to the bucket with empty element */
+ CVT (clib_cuckoo_bucket) * b =
+ CV (clib_cuckoo_bucket_at_index) (h, buckets[i]);
+ CVT (clib_cuckoo_kv) * elt = b->elts + offsets[i];
+ clib_cuckoo_bucket_aux_t empty_aux =
+ CV (clib_cuckoo_bucket_version_bump_and_lock) (empty_bucket);
+ CV (clib_cuckoo_set_locked_elt)
+ (empty_bucket, empty_elt, elt, b->reduced_hashes[elt - b->elts]);
+ if (i == path->length - 1)
+ {
+ /* we only need to increase the use count for the bucket with
+ * free element - all other buckets' use counts won't change */
+ empty_aux = clib_cuckoo_bucket_aux_set_use_count (empty_aux,
+ clib_cuckoo_bucket_aux_get_use_count
+ (empty_aux) +
+ 1);
+ }
+ CV (clib_cuckoo_bucket_unlock) (empty_bucket, empty_aux);
+ /*
+ * the element now exists in both places - in the previously empty
+ * element and in its original bucket - we can now safely overwrite
+ * the element in the original bucket with previous element in path
+ * without loosing data (and we don't need to modify the use count)
+ */
+ empty_bucket = b;
+ empty_elt = elt;
+ }
+ /* now we have a place to put the kvp in ... */
+ CV (clib_cuckoo_set_elt) (empty_bucket, empty_elt, kvp, reduced_hash);
+ CLIB_CUCKOO_DBG ("Slow insert success, bucket: %p\n%U", empty_bucket,
+ CV (format_cuckoo_bucket), empty_bucket);
+#if CLIB_CUCKOO_DEBUG_COUNTERS
+ ++h->slow_adds;
+#endif
+ }
+ return rv;
+}
+
+static int CV (clib_cuckoo_add_fast) (CVT (clib_cuckoo) * h,
+ clib_cuckoo_lookup_info_t * lookup,
+ CVT (clib_cuckoo_kv) * kvp,
+ u8 reduced_hash)
+{
+ CVT (clib_cuckoo_kv) * elt;
+ CVT (clib_cuckoo_bucket) * bucket1 =
+ CV (clib_cuckoo_bucket_at_index) (h, lookup->bucket1);
+ if ((elt = CV (clib_cuckoo_bucket_find_empty) (bucket1)))
+ {
+ clib_cuckoo_bucket_aux_t aux =
+ CV (clib_cuckoo_bucket_version_bump_and_lock) (bucket1);
+ CV (clib_cuckoo_set_locked_elt) (bucket1, elt, kvp, reduced_hash);
+ aux =
+ clib_cuckoo_bucket_aux_set_use_count (aux,
+ clib_cuckoo_bucket_aux_get_use_count
+ (aux) + 1);
+ CV (clib_cuckoo_bucket_unlock) (bucket1, aux);
+#if CLIB_CUCKOO_DEBUG_COUNTERS
+ ++h->fast_adds;
+#endif
+ return CLIB_CUCKOO_ERROR_SUCCESS;
+ }
+ CVT (clib_cuckoo_bucket) * bucket2 =
+ CV (clib_cuckoo_bucket_at_index) (h, lookup->bucket2);
+ if ((elt =
+ CV (clib_cuckoo_bucket_find_empty) (CV (clib_cuckoo_bucket_at_index)
+ (h, lookup->bucket2))))
+ {
+ clib_cuckoo_bucket_aux_t aux =
+ CV (clib_cuckoo_bucket_version_bump_and_lock) (bucket2);
+ CV (clib_cuckoo_set_locked_elt) (bucket2, elt, kvp, reduced_hash);
+ aux =
+ clib_cuckoo_bucket_aux_set_use_count (aux,
+ clib_cuckoo_bucket_aux_get_use_count
+ (aux) + 1);
+ CV (clib_cuckoo_bucket_unlock) (bucket2, aux);
+#if CLIB_CUCKOO_DEBUG_COUNTERS
+ ++h->fast_adds;
+#endif
+ return CLIB_CUCKOO_ERROR_SUCCESS;
+ }
+ return CLIB_CUCKOO_ERROR_AGAIN;
+}
+
+/**
+ * perform garbage collection
+ *
+ * this function assumes there is no other thread touching the cuckoo hash,
+ * not even a reader, it's meant to be called from main thread
+ * in a stop-the-world situation
+ */
+void CV (clib_cuckoo_garbage_collect) (CVT (clib_cuckoo) * h)
+{
+ CLIB_MEMORY_BARRIER ();
+ CVT (clib_cuckoo_bucket) * *b;
+ /* *INDENT-OFF* */
+ vec_foreach (b, h->to_be_freed)
+ {
+ if (*b == h->buckets)
+ {
+ continue;
+ }
+#if CLIB_CUCKOO_DEBUG_GC
+ fformat (stdout, "gc: free %p\n", *b);
+#endif
+ vec_free (*b);
+ }
+ /* *INDENT-ON* */
+ vec_free (h->to_be_freed);
+ CLIB_MEMORY_BARRIER ();
+}
+
+/**
+ * expand and rehash a cuckoo hash
+ *
+ * 1. double the size of the hash table
+ * 2. move items to new locations derived from the new size
+ */
+static void CV (clib_cuckoo_rehash) (CVT (clib_cuckoo) * h)
+{
+ CVT (clib_cuckoo_bucket) * old = h->buckets;
+ uword old_nbuckets = vec_len (old);
+ uword new_nbuckets = 2 * old_nbuckets;
+ CVT (clib_cuckoo_bucket) * new =
+ vec_dup_aligned (old, CLIB_CACHE_LINE_BYTES);
+ /* allocate space */
+ vec_validate_aligned (new, new_nbuckets - 1, CLIB_CACHE_LINE_BYTES);
+ ASSERT (new_nbuckets == vec_len (new));
+ /* store old pointer in to-be-freed list */
+ vec_add1 (h->to_be_freed, old);
+ /* mark new elements as free */
+ CVT (clib_cuckoo_bucket) * bucket;
+ for (bucket = new + old_nbuckets; bucket < vec_end (new); ++bucket)
+ {
+ clib_memset (bucket->elts, 0xff, sizeof (bucket->elts));
+ }
+ /*
+ * this for loop manipulates the new (unseen) memory, so no locks
+ * are required here
+ */
+ uword old_bucket_idx;
+ for (old_bucket_idx = 0; old_bucket_idx < old_nbuckets; ++old_bucket_idx)
+ {
+ /* items in old bucket might be moved to new bucket */
+ uword new_bucket_idx = old_bucket_idx + old_nbuckets;
+ CVT (clib_cuckoo_bucket) * old_bucket = new + old_bucket_idx;
+ CVT (clib_cuckoo_bucket) * new_bucket = new + new_bucket_idx;
+ int i = 0;
+ int moved = 0;
+ clib_cuckoo_bucket_aux_t aux = old_bucket->aux;
+ for (i = 0; i < clib_cuckoo_bucket_aux_get_use_count (aux); ++i)
+ {
+ CVT (clib_cuckoo_kv) * elt = old_bucket->elts + i;
+ u64 hash = CV (clib_cuckoo_hash) (elt);
+ clib_cuckoo_lookup_info_t old_lookup =
+ CV (clib_cuckoo_calc_lookup) (old, hash);
+ clib_cuckoo_lookup_info_t new_lookup =
+ CV (clib_cuckoo_calc_lookup) (new, hash);
+ if ((old_bucket_idx == old_lookup.bucket1 &&
+ new_bucket_idx == new_lookup.bucket1) ||
+ (old_bucket_idx == old_lookup.bucket2 &&
+ new_bucket_idx == new_lookup.bucket2))
+ {
+ /* move the item to new bucket */
+ CVT (clib_cuckoo_kv) * empty_elt = new_bucket->elts + moved;
+ ASSERT (empty_elt);
+ CV (clib_cuckoo_set_locked_elt)
+ (new_bucket, empty_elt, elt, old_bucket->reduced_hashes[i]);
+ CV (clib_cuckoo_free_locked_elt) (elt);
+ ++moved;
+ }
+ }
+ if (moved)
+ {
+ CV (clib_cuckoo_bucket_tidy) (old_bucket);
+ aux =
+ clib_cuckoo_bucket_aux_set_use_count (aux,
+ clib_cuckoo_bucket_aux_get_use_count
+ (aux) - moved);
+ old_bucket->aux = aux;
+ aux = new_bucket->aux;
+ aux =
+ clib_cuckoo_bucket_aux_set_use_count (aux,
+ clib_cuckoo_bucket_aux_get_use_count
+ (aux) + moved);
+ new_bucket->aux = aux;
+ }
+ }
+ h->buckets = new;
+#if CLIB_CUCKOO_DEBUG_COUNTERS
+ ++h->rehashes;
+#endif
+ h->garbage_callback (h, h->garbage_ctx);
+}
+
+static int CV (clib_cuckoo_bucket_search_internal) (CVT (clib_cuckoo) * h,
+ uword bucket,
+ CVT (clib_cuckoo_kv) *
+ kvp,
+ CVT (clib_cuckoo_kv) *
+ *found)
+{
+ CVT (clib_cuckoo_bucket) * b = CV (clib_cuckoo_bucket_at_index) (h, bucket);
+ int i;
+ /* *INDENT-OFF* */
+ clib_cuckoo_bucket_foreach_idx_unrolled (i, {
+ CVT (clib_cuckoo_kv) *elt = &b->elts[i];
+ if (CV (clib_cuckoo_key_compare) (elt->key, kvp->key))
+ {
+ *found = elt;
+ return CLIB_CUCKOO_ERROR_SUCCESS;
+ }
+ });
+ /* *INDENT-ON* */
+ return CLIB_CUCKOO_ERROR_NOT_FOUND;
+}
+
+int CV (clib_cuckoo_add_del) (CVT (clib_cuckoo) * h,
+ CVT (clib_cuckoo_kv) * kvp, int is_add,
+ int dont_overwrite)
+{
+ CLIB_CUCKOO_DBG ("%s %U", is_add ? "Add" : "Del", CV (format_cuckoo_kvp),
+ kvp);
+ clib_cuckoo_lookup_info_t lookup;
+ u64 hash = CV (clib_cuckoo_hash) (kvp);
+ u8 reduced_hash = clib_cuckoo_reduce_hash (hash);
+ clib_spinlock_lock (&h->writer_lock);
+restart:
+ lookup = CV (clib_cuckoo_calc_lookup) (h->buckets, hash);
+ CVT (clib_cuckoo_bucket) * b =
+ CV (clib_cuckoo_bucket_at_index) (h, lookup.bucket1);
+ CVT (clib_cuckoo_kv) * found;
+ int rv =
+ CV (clib_cuckoo_bucket_search_internal) (h, lookup.bucket1, kvp, &found);
+ if (CLIB_CUCKOO_ERROR_SUCCESS != rv)
+ {
+ ASSERT (CLIB_CUCKOO_ERROR_NOT_FOUND == rv);
+ b = CV (clib_cuckoo_bucket_at_index) (h, lookup.bucket2);
+ rv = CV (clib_cuckoo_bucket_search_internal) (h, lookup.bucket2, kvp,
+ &found);
+ }
+ if (CLIB_CUCKOO_ERROR_SUCCESS == rv)
+ {
+ if (is_add)
+ {
+ if (dont_overwrite)
+ {
+ CLIB_CUCKOO_DBG ("Refused replacing existing %U",
+ CV (format_cuckoo_elt), found,
+ b->reduced_hashes[found - b->elts]);
+ rv = CLIB_CUCKOO_ERROR_AGAIN;
+ }
+ else
+ {
+ /* prevent readers reading this bucket while we switch the values */
+ clib_cuckoo_bucket_aux_t aux =
+ CV (clib_cuckoo_bucket_version_bump_and_lock) (b);
+ clib_memcpy (&found->value, &kvp->value, sizeof (found->value));
+ CLIB_CUCKOO_DBG ("Replaced existing %U", CV (format_cuckoo_elt),
+ found, b->reduced_hashes[found - b->elts]);
+ CV (clib_cuckoo_bucket_unlock) (b, aux);
+ rv = CLIB_CUCKOO_ERROR_SUCCESS;
+ }
+ }
+ else
+ {
+ CV (clib_cuckoo_free_elt_in_bucket) (b, found);
+ rv = CLIB_CUCKOO_ERROR_SUCCESS;
+ }
+ CLIB_CUCKOO_DEEP_SELF_CHECK (h);
+ goto unlock;
+ }
+ if (!is_add)
+ {
+ CLIB_CUCKOO_DBG ("%U not present in table", CV (format_cuckoo_kvp),
+ kvp);
+ rv = CLIB_CUCKOO_ERROR_NOT_FOUND;
+ goto unlock;
+ }
+ /* from this point on, it's add code only */
+ ASSERT (CLIB_CUCKOO_ERROR_NOT_FOUND == rv);
+ /* fast path: try to search for unoccupied slot in one of the buckets */
+ rv = CV (clib_cuckoo_add_fast) (h, &lookup, kvp, reduced_hash);
+ CLIB_CUCKOO_DEEP_SELF_CHECK (h);
+ if (CLIB_CUCKOO_ERROR_SUCCESS != rv)
+ {
+ CLIB_CUCKOO_DBG
+ ("Fast insert failed, bucket 1: %wu, bucket 2: %wu\n%U%U",
+ lookup.bucket1, lookup.bucket2, CV (format_cuckoo_bucket),
+ CV (clib_cuckoo_bucket_at_index) (h, lookup.bucket1),
+ CV (format_cuckoo_bucket),
+ CV (clib_cuckoo_bucket_at_index) (h, lookup.bucket2));
+ /* slow path */
+ rv = CV (clib_cuckoo_add_slow) (h, kvp, &lookup, reduced_hash);
+ CLIB_CUCKOO_DEEP_SELF_CHECK (h);
+ if (CLIB_CUCKOO_ERROR_SUCCESS != rv)
+ {
+ CLIB_CUCKOO_DBG ("Slow insert failed, rehash required:\n%U",
+ CV (format_cuckoo), h, 1);
+ /* ultra slow path */
+ CV (clib_cuckoo_rehash) (h);
+ CLIB_CUCKOO_DEEP_SELF_CHECK (h);
+ CLIB_CUCKOO_DBG ("Restarting add after rehash...");
+ goto restart;
+ }
+ }
+unlock:
+ clib_spinlock_unlock (&h->writer_lock);
+ return rv;
+}
+
+u8 *CV (format_cuckoo) (u8 * s, va_list * args)
+{
+ CVT (clib_cuckoo) * h = va_arg (*args, CVT (clib_cuckoo) *);
+ int verbose = va_arg (*args, int);
+
+ s = format (s, "Hash table %s\n", h->name ? h->name : "(unnamed)");
+
+ uword free = 0;
+ uword used = 0;
+ uword use_count_total = 0;
+ float load_factor;
+ CVT (clib_cuckoo_bucket) * b;
+ /* *INDENT-OFF* */
+ clib_cuckoo_foreach_bucket (b, h, {
+ if (verbose)
+ {
+ s = format (s, "%U", CV (format_cuckoo_bucket), b);
+ }
+ int i;
+ clib_cuckoo_bucket_foreach_idx (i)
+ {
+ CVT (clib_cuckoo_kv) *elt = &b->elts[i];
+ if (CV (clib_cuckoo_kv_is_free) (elt))
+ {
+ ++free;
+ }
+ else
+ {
+ ++used;
+ }
+ }
+ clib_cuckoo_bucket_aux_t aux = b->aux;
+ use_count_total += clib_cuckoo_bucket_aux_get_use_count (aux);
+ });
+ /* *INDENT-ON* */
+ s = format (s, "Used slots: %wu\n", used);
+ s = format (s, "Use count total: %wu\n", use_count_total);
+ s = format (s, "Free slots: %wu\n", free);
+ if (free + used != 0)
+ load_factor = ((float) used) / ((float) (free + used));
+ else
+ load_factor = 0.0;
+ s = format (s, "Load factor: %.2f\n", load_factor);
+#if CLIB_CUCKOO_DEBUG_COUNTERS
+ s = format (s, "BFS attempts limited by max steps: %lld\n",
+ h->steps_exceeded);
+ s = format (s, "BFS cutoffs due to empty queue: %lld\n",
+ h->bfs_queue_emptied);
+ s = format (s, "Fast adds: %lld\n", h->fast_adds);
+ s = format (s, "Slow adds: %lld\n", h->slow_adds);
+ s = format (s, "Rehashes: %lld\n", h->rehashes);
+#endif
+ return s;
+}
+
+float CV (clib_cuckoo_calculate_load_factor) (CVT (clib_cuckoo) * h)
+{
+ uword nonfree = 0;
+ uword all = 0;
+ CVT (clib_cuckoo_bucket) * bucket;
+ /* *INDENT-OFF* */
+ clib_cuckoo_foreach_bucket (bucket, h, {
+ int i;
+ clib_cuckoo_bucket_foreach_idx (i)
+ {
+ CVT (clib_cuckoo_kv) *elt = bucket->elts + i;
+ ++all;
+ if (!CV (clib_cuckoo_kv_is_free) (elt))
+ {
+ ++nonfree;
+ }
+ }
+ });
+ /* *INDENT-ON* */
+ if (all)
+ return (float) nonfree / (float) all;
+ else
+ return 0.0;
+}
+
+/** @endcond */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/cuckoo_template.h b/extras/deprecated/vppinfra/cuckoo_template.h
new file mode 100644
index 00000000000..364c2919d96
--- /dev/null
+++ b/extras/deprecated/vppinfra/cuckoo_template.h
@@ -0,0 +1,460 @@
+/*
+ Copyright (c) 2017 Cisco and/or its affiliates.
+
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+*/
+
+/*
+ * cuckoo hash implementation based on paper
+ * 'Algorithmic Improvements for Fast Concurrent Cuckoo Hashing'
+ * by Xiaozhou Li, David G. Andersen, Michael Kaminsky, Michael J. Freedman
+ * and their libcuckoo implementation (https://github.com/efficient/libcuckoo)
+ */
+
+/*
+ * Note: to instantiate the template multiple times in a single file,
+ * #undef __included_cuckoo_template_h__...
+ */
+#ifndef __included_cuckoo_template_h__
+#define __included_cuckoo_template_h__
+
+#include <vppinfra/heap.h>
+#include <vppinfra/format.h>
+#include <vppinfra/pool.h>
+#include <vppinfra/lock.h>
+#include <vppinfra/error.h>
+#include <vppinfra/hash.h>
+#include <vppinfra/cache.h>
+
+#ifndef CLIB_CUCKOO_TYPE
+#error CLIB_CUCKOO_TYPE not defined
+#endif
+
+#ifndef CLIB_CUCKOO_BFS_MAX_STEPS
+#error CLIB_CUCKOO_BFS_MAX_STEPS not defined
+#endif
+
+#ifndef CLIB_CUCKOO_KVP_PER_BUCKET
+#error CLIB_CUCKOO_KVP_PER_BUCKET not defined
+#endif
+
+#ifndef CLIB_CUCKOO_LOG2_KVP_PER_BUCKET
+#error CLIB_CUCKOO_LOG2_KVP_PER_BUCKET not defined
+#endif
+
+#ifndef CLIB_CUCKOO_BFS_MAX_PATH_LENGTH
+#error CLIB_CUCKOO_BFS_MAX_PATH_LENGTH not defined
+#endif
+
+STATIC_ASSERT (CLIB_CUCKOO_KVP_PER_BUCKET ==
+ (1 << CLIB_CUCKOO_LOG2_KVP_PER_BUCKET),
+ "CLIB_CUCKOO_KVP_PER_BUCKET != (1 << CLIB_CUCKOO_LOG2_KVP_PER_BUCKET");
+
+#define _cv(a, b) a##b
+#define __cv(a, b) _cv (a, b)
+#define CV(a) __cv (a, CLIB_CUCKOO_TYPE)
+
+#define _cvt(a, b) a##b##_t
+#define __cvt(a, b) _cvt (a, b)
+#define CVT(a) __cvt (a, CLIB_CUCKOO_TYPE)
+
+typedef u64 clib_cuckoo_bucket_aux_t;
+
+#define CLIB_CUCKOO_USE_COUNT_BIT_WIDTH (1 + CLIB_CUCKOO_LOG2_KVP_PER_BUCKET)
+
+always_inline u64
+clib_cuckoo_bucket_aux_get_version (clib_cuckoo_bucket_aux_t aux)
+{
+ return aux >> (1 + CLIB_CUCKOO_USE_COUNT_BIT_WIDTH);
+}
+
+always_inline int
+clib_cuckoo_bucket_aux_get_use_count (clib_cuckoo_bucket_aux_t aux)
+{
+ u64 use_count_mask = (1 << CLIB_CUCKOO_USE_COUNT_BIT_WIDTH) - 1;
+ return (aux >> 1) & use_count_mask;
+}
+
+always_inline int
+clib_cuckoo_bucket_aux_get_writer_flag (clib_cuckoo_bucket_aux_t aux)
+{
+ return aux & 1;
+}
+
+always_inline clib_cuckoo_bucket_aux_t
+clib_cuckoo_bucket_aux_pack (u64 version, int use_count, int writer_flag)
+{
+ return (version << (1 + CLIB_CUCKOO_USE_COUNT_BIT_WIDTH)) +
+ (use_count << 1) + writer_flag;
+}
+
+always_inline clib_cuckoo_bucket_aux_t
+clib_cuckoo_bucket_aux_set_version (clib_cuckoo_bucket_aux_t aux, u64 version)
+{
+ int use_count = clib_cuckoo_bucket_aux_get_use_count (aux);
+ int writer_flag = clib_cuckoo_bucket_aux_get_writer_flag (aux);
+ return clib_cuckoo_bucket_aux_pack (version, use_count, writer_flag);
+}
+
+always_inline clib_cuckoo_bucket_aux_t
+clib_cuckoo_bucket_aux_set_use_count (clib_cuckoo_bucket_aux_t aux,
+ int use_count)
+{
+ u64 version = clib_cuckoo_bucket_aux_get_version (aux);
+ int writer_flag = clib_cuckoo_bucket_aux_get_writer_flag (aux);
+ return clib_cuckoo_bucket_aux_pack (version, use_count, writer_flag);
+}
+
+always_inline clib_cuckoo_bucket_aux_t
+clib_cuckoo_bucket_aux_set_writer_flag (clib_cuckoo_bucket_aux_t aux,
+ int writer_flag)
+{
+ u64 version = clib_cuckoo_bucket_aux_get_version (aux);
+ int use_count = clib_cuckoo_bucket_aux_get_use_count (aux);
+ return clib_cuckoo_bucket_aux_pack (version, use_count, writer_flag);
+}
+
+#define PATH_BITS_REQ \
+ (CLIB_CUCKOO_BFS_MAX_PATH_LENGTH * CLIB_CUCKOO_LOG2_KVP_PER_BUCKET)
+
+#if PATH_BITS_REQ <= 8
+typedef u8 path_data_t;
+#elif PATH_BITS_REQ <= 16
+typedef u16 path_data_t;
+#elif PATH_BITS_REQ <= 32
+typedef u32 path_data_t;
+#elif PATH_BITS_REQ <= 64
+typedef u64 path_data_t;
+#else
+#error no suitable datatype for path storage...
+#endif
+
+typedef struct
+{
+ /** bucket where this path begins */
+ u64 start;
+ /** bucket at end of path */
+ u64 bucket;
+ /** length of the path */
+ u8 length;
+ /** holds compressed offsets in buckets along path */
+ path_data_t data;
+} clib_cuckoo_path_t;
+
+typedef struct
+{
+ CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
+
+ /** reduced hashes corresponding to elements */
+ u8 reduced_hashes[CLIB_CUCKOO_KVP_PER_BUCKET];
+
+ /** auxiliary data - version, writer flag and used count */
+ volatile clib_cuckoo_bucket_aux_t aux;
+
+ /** cuckoo elements in this bucket */
+ CVT (clib_cuckoo_kv) elts[CLIB_CUCKOO_KVP_PER_BUCKET];
+} CVT (clib_cuckoo_bucket);
+
+#define clib_cuckoo_bucket_foreach_idx(var) \
+ for (var = 0; var < CLIB_CUCKOO_KVP_PER_BUCKET; var++)
+
+#if CLIB_CUCKOO_OPTIMIZE_UNROLL
+#if CLIB_CUCKOO_KVP_PER_BUCKET == 2
+#define clib_cuckoo_bucket_foreach_idx_unrolled(var, body) \
+ do \
+ { \
+ var = 0; \
+ body; \
+ var = 1; \
+ body; \
+ } \
+ while (0);
+#elif CLIB_CUCKOO_KVP_PER_BUCKET == 4
+#define clib_cuckoo_bucket_foreach_idx_unrolled(var, body) \
+ do \
+ { \
+ var = 0; \
+ body; \
+ var = 1; \
+ body; \
+ var = 2; \
+ body; \
+ var = 3; \
+ body; \
+ } \
+ while (0);
+#elif CLIB_CUCKOO_KVP_PER_BUCKET == 8
+#define clib_cuckoo_bucket_foreach_idx_unrolled(var, body) \
+ do \
+ { \
+ var = 0; \
+ body; \
+ var = 1; \
+ body; \
+ var = 2; \
+ body; \
+ var = 3; \
+ body; \
+ var = 4; \
+ body; \
+ var = 5; \
+ body; \
+ var = 6; \
+ body; \
+ var = 7; \
+ body; \
+ } \
+ while (0);
+#else
+#define clib_cuckoo_bucket_foreach_idx_unrolled(var, body) \
+ clib_cuckoo_bucket_foreach_idx (var) \
+ { \
+ body; \
+ }
+#endif
+#else /* CLIB_CUCKOO_OPTIMIZE_UNROLL */
+#define clib_cuckoo_bucket_foreach_idx_unrolled(var, body) \
+ clib_cuckoo_bucket_foreach_idx (var) \
+ { \
+ body; \
+ }
+#endif /* CLIB_CUCKOO_OPTIMIZE_UNROLL */
+
+#define clib_cuckoo_bucket_foreach_elt_index(var, bucket) \
+ for (var = 0; var < CLIB_CUCKOO_KVP_PER_BUCKET; ++i)
+
+#define clib_cuckoo_foreach_bucket(var, h, body) \
+ do \
+ { \
+ CVT (clib_cuckoo_bucket) *__buckets = h->buckets; \
+ vec_foreach (var, __buckets) \
+ { \
+ body; \
+ } \
+ } \
+ while (0)
+
+typedef struct CV (clib_cuckoo)
+{
+ /** vector of elements containing key-value pairs and auxiliary data */
+ CVT (clib_cuckoo_bucket) * volatile buckets;
+
+ /** garbage to be freed once its safe to do so .. */
+ CVT (clib_cuckoo_bucket) * *to_be_freed;
+
+ /** hash table name */
+ const char *name;
+
+ /** pool of cuckoo paths (reused when doing bfd search) */
+ clib_cuckoo_path_t *paths;
+
+ /**
+ * vector used as queue when doing cuckoo path searches - holds offsets
+ * in paths pool
+ */
+ uword *bfs_search_queue;
+
+ /**
+ * writer lock - whether this lock is taken or not has zero effect on
+ * readers
+ */
+ clib_spinlock_t writer_lock;
+
+ /** caller context passed to callback with garbage notification */
+ void *garbage_ctx;
+
+ /**
+ * garbage notify function - called when some garbage needs to be collected
+ * in main thread while other threads are stopped
+ */
+ void (*garbage_callback) (struct CV (clib_cuckoo) * h, void *garbage_ctx);
+
+#if CLIB_CUCKOO_DEBUG_COUNTERS
+ u64 steps_exceeded;
+ u64 bfs_queue_emptied;
+ u64 fast_adds;
+ u64 slow_adds;
+ u64 rehashes;
+#endif
+
+} CVT (clib_cuckoo);
+
+void CV (clib_cuckoo_init) (CVT (clib_cuckoo) * h, const char *name,
+ uword nbuckets,
+ void (*garbage_callback) (CVT (clib_cuckoo) *,
+ void *),
+ void *garbage_ctx);
+
+void CV (clib_cuckoo_garbage_collect) (CVT (clib_cuckoo) * h);
+
+void CV (clib_cuckoo_free) (CVT (clib_cuckoo) * h);
+
+int CV (clib_cuckoo_add_del) (CVT (clib_cuckoo) * h,
+ CVT (clib_cuckoo_kv) * add_v, int is_add,
+ int dont_overwrite);
+int CV (clib_cuckoo_search) (CVT (clib_cuckoo) * h,
+ CVT (clib_cuckoo_kv) * search_v,
+ CVT (clib_cuckoo_kv) * return_v);
+
+void CV (clib_cuckoo_foreach_key_value_pair) (CVT (clib_cuckoo) * h,
+ void *callback, void *arg);
+
+float CV (clib_cuckoo_calc_load) (CVT (clib_cuckoo) * h);
+
+format_function_t CV (format_cuckoo);
+format_function_t CV (format_cuckoo_kvp);
+
+always_inline u8
+clib_cuckoo_reduce_hash (u64 hash)
+{
+ u32 v32 = ((u32) hash) ^ ((u32) (hash >> 32));
+ u16 v16 = ((u16) v32) ^ ((u16) (v32 >> 16));
+ u8 v8 = ((u8) v16) ^ ((u8) (v16 >> 8));
+ return v8;
+}
+
+always_inline u64
+clib_cuckoo_get_other_bucket (u64 nbuckets, u64 bucket, u8 reduced_hash)
+{
+ u64 mask = (nbuckets - 1);
+ return (bucket ^ ((reduced_hash + 1) * 0xc6a4a7935bd1e995)) & mask;
+}
+
+always_inline clib_cuckoo_lookup_info_t
+CV (clib_cuckoo_calc_lookup) (CVT (clib_cuckoo_bucket) * buckets, u64 hash)
+{
+ clib_cuckoo_lookup_info_t lookup;
+ u64 nbuckets = vec_len (buckets);
+ u64 mask = (nbuckets - 1);
+ lookup.bucket1 = hash & mask;
+#if CLIB_CUCKOO_OPTIMIZE_PREFETCH
+ CLIB_PREFETCH (vec_elt_at_index (buckets, lookup.bucket1),
+ sizeof (*buckets), LOAD);
+#endif
+ u8 reduced_hash = clib_cuckoo_reduce_hash (hash);
+ lookup.bucket2 =
+ clib_cuckoo_get_other_bucket (nbuckets, lookup.bucket1, reduced_hash);
+#if CLIB_CUCKOO_OPTIMIZE_PREFETCH
+ CLIB_PREFETCH (vec_elt_at_index (buckets, lookup.bucket2),
+ sizeof (*buckets), LOAD);
+#endif
+ lookup.reduced_hash = reduced_hash;
+ ASSERT (lookup.bucket1 < nbuckets);
+ ASSERT (lookup.bucket2 < nbuckets);
+ return lookup;
+}
+
+/**
+ * search for key within bucket
+ */
+always_inline int CV (clib_cuckoo_bucket_search) (CVT (clib_cuckoo_bucket) *
+ b,
+ CVT (clib_cuckoo_kv) * kvp,
+ u8 reduced_hash)
+{
+ clib_cuckoo_bucket_aux_t bucket_aux;
+ u8 writer_flag;
+ do
+ {
+ bucket_aux = b->aux;
+ writer_flag = clib_cuckoo_bucket_aux_get_writer_flag (bucket_aux);
+ }
+ while (PREDICT_FALSE (writer_flag)); /* loop while writer flag is set */
+
+ int i;
+#if CLIB_CUCKOO_OPTIMIZE_USE_COUNT_LIMITS_SEARCH
+ const int use_count = clib_cuckoo_bucket_aux_get_use_count (bucket_aux);
+#endif
+ /* *INDENT-OFF* */
+ clib_cuckoo_bucket_foreach_idx_unrolled (i, {
+#if CLIB_CUCKOO_OPTIMIZE_USE_COUNT_LIMITS_SEARCH
+ if (i > use_count)
+ {
+ break;
+ }
+#endif
+ if (CV (clib_cuckoo_key_compare) (kvp->key, b->elts[i].key))
+ {
+ kvp->value = b->elts[i].value;
+ clib_cuckoo_bucket_aux_t bucket_aux2 = b->aux;
+ if (PREDICT_TRUE (clib_cuckoo_bucket_aux_get_version (bucket_aux) ==
+ clib_cuckoo_bucket_aux_get_version (bucket_aux2)))
+ {
+ /* yay, fresh data */
+ return CLIB_CUCKOO_ERROR_SUCCESS;
+ }
+ else
+ {
+ /* oops, modification detected */
+ return CLIB_CUCKOO_ERROR_AGAIN;
+ }
+ }
+ });
+ /* *INDENT-ON* */
+ return CLIB_CUCKOO_ERROR_NOT_FOUND;
+}
+
+always_inline int
+CV (clib_cuckoo_search_inline_with_hash) (CVT (clib_cuckoo) * h, u64 hash,
+ CVT (clib_cuckoo_kv) * kvp)
+{
+ CVT (clib_cuckoo_bucket) * buckets = h->buckets;
+ uword bucket1, bucket2;
+ u8 reduced_hash;
+ u64 nbuckets = vec_len (buckets);
+ u64 mask = nbuckets - 1;
+ int rv;
+
+ bucket1 = hash & mask;
+ reduced_hash = clib_cuckoo_reduce_hash (hash);
+
+again:
+ rv = CV (clib_cuckoo_bucket_search) (vec_elt_at_index (buckets, bucket1),
+ kvp, reduced_hash);
+
+ if (rv == CLIB_CUCKOO_ERROR_SUCCESS)
+ return CLIB_CUCKOO_ERROR_SUCCESS;
+
+ if (PREDICT_FALSE (rv == CLIB_CUCKOO_ERROR_AGAIN))
+ goto again;
+
+ bucket2 = clib_cuckoo_get_other_bucket (nbuckets, bucket1, reduced_hash);
+ rv = CV (clib_cuckoo_bucket_search) (vec_elt_at_index (buckets, bucket2),
+ kvp, reduced_hash);
+
+ /* change to 2nd bucket could bump the item to 1st bucket and the bucket
+ * indexes might not even be valid anymore - restart the search */
+ if (PREDICT_FALSE (rv == CLIB_CUCKOO_ERROR_AGAIN))
+ goto again;
+
+ return rv;
+}
+
+always_inline int CV (clib_cuckoo_search_inline) (CVT (clib_cuckoo) * h,
+ CVT (clib_cuckoo_kv) * kvp)
+{
+ u64 hash = CV (clib_cuckoo_hash) (kvp);
+ return CV (clib_cuckoo_search_inline_with_hash) (h, hash, kvp);
+}
+
+#endif /* __included_cuckoo_template_h__ */
+
+/** @endcond */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/fheap.c b/extras/deprecated/vppinfra/fheap.c
new file mode 100644
index 00000000000..034168e85ab
--- /dev/null
+++ b/extras/deprecated/vppinfra/fheap.c
@@ -0,0 +1,473 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include <vppinfra/fheap.h>
+
+/* Fibonacci heaps. */
+always_inline fheap_node_t *
+fheap_get_node (fheap_t * f, u32 i)
+{
+ return i != ~0 ? vec_elt_at_index (f->nodes, i) : 0;
+}
+
+always_inline fheap_node_t *
+fheap_get_root (fheap_t * f)
+{
+ return fheap_get_node (f, f->min_root);
+}
+
+static void
+fheap_validate (fheap_t * f)
+{
+ fheap_node_t *n, *m;
+ uword ni, si;
+
+ if (!CLIB_DEBUG || !f->enable_validate)
+ return;
+
+ vec_foreach_index (ni, f->nodes)
+ {
+ n = vec_elt_at_index (f->nodes, ni);
+
+ if (!n->is_valid)
+ continue;
+
+ /* Min root must have minimal key. */
+ m = vec_elt_at_index (f->nodes, f->min_root);
+ ASSERT (n->key >= m->key);
+
+ /* Min root must have no parent. */
+ if (ni == f->min_root)
+ ASSERT (n->parent == ~0);
+
+ /* Check sibling linkages. */
+ if (n->next_sibling == ~0)
+ ASSERT (n->prev_sibling == ~0);
+ else if (n->prev_sibling == ~0)
+ ASSERT (n->next_sibling == ~0);
+ else
+ {
+ fheap_node_t *prev, *next;
+ u32 si = n->next_sibling, si_start = si;
+ do
+ {
+ m = vec_elt_at_index (f->nodes, si);
+ prev = vec_elt_at_index (f->nodes, m->prev_sibling);
+ next = vec_elt_at_index (f->nodes, m->next_sibling);
+ ASSERT (prev->next_sibling == si);
+ ASSERT (next->prev_sibling == si);
+ si = m->next_sibling;
+ }
+ while (si != si_start);
+ }
+
+ /* Loop through all siblings. */
+ {
+ u32 n_siblings = 0;
+
+ foreach_fheap_node_sibling (f, si, n->next_sibling, (
+ {
+ m =
+ vec_elt_at_index
+ (f->nodes, si);
+ /* All siblings must have same parent. */
+ ASSERT (m->parent
+ ==
+ n->
+ parent);
+ n_siblings += 1;}
+ ));
+
+ /* Either parent is non-empty or there are siblings present. */
+ if (n->parent == ~0 && ni != f->min_root)
+ ASSERT (n_siblings > 0);
+ }
+
+ /* Loop through all children. */
+ {
+ u32 found_first_child = n->first_child == ~0;
+ u32 n_children = 0;
+
+ foreach_fheap_node_sibling (f, si, n->first_child, (
+ {
+ m =
+ vec_elt_at_index
+ (f->nodes, si);
+ /* Children must have larger keys than their parent. */
+ ASSERT (m->key >=
+ n->key);
+ if
+ (!found_first_child)
+ found_first_child =
+ si ==
+ n->first_child;
+ n_children += 1;}
+ ));
+
+ /* Check that first child is present on list. */
+ ASSERT (found_first_child);
+
+ /* Make sure rank is correct. */
+ ASSERT (n->rank == n_children);
+ }
+ }
+
+ /* Increment serial number for each successful validate.
+ Failure can be used as condition for gdb breakpoints. */
+ f->validate_serial++;
+}
+
+always_inline void
+fheap_node_add_sibling (fheap_t * f, u32 ni, u32 ni_to_add)
+{
+ fheap_node_t *n = vec_elt_at_index (f->nodes, ni);
+ fheap_node_t *n_to_add = vec_elt_at_index (f->nodes, ni_to_add);
+ fheap_node_t *n_next = fheap_get_node (f, n->next_sibling);
+ fheap_node_t *parent;
+
+ /* Empty list? */
+ if (n->next_sibling == ~0)
+ {
+ ASSERT (n->prev_sibling == ~0);
+ n->next_sibling = n->prev_sibling = ni_to_add;
+ n_to_add->next_sibling = n_to_add->prev_sibling = ni;
+ }
+ else
+ {
+ /* Add node after existing node. */
+ n_to_add->prev_sibling = ni;
+ n_to_add->next_sibling = n->next_sibling;
+
+ n->next_sibling = ni_to_add;
+ n_next->prev_sibling = ni_to_add;
+ }
+
+ n_to_add->parent = n->parent;
+ parent = fheap_get_node (f, n->parent);
+ if (parent)
+ parent->rank += 1;
+}
+
+void
+fheap_add (fheap_t * f, u32 ni, u32 key)
+{
+ fheap_node_t *r, *n;
+ u32 ri;
+
+ n = vec_elt_at_index (f->nodes, ni);
+
+ clib_memset (n, 0, sizeof (n[0]));
+ n->parent = n->first_child = n->next_sibling = n->prev_sibling = ~0;
+ n->key = key;
+
+ r = fheap_get_root (f);
+ ri = f->min_root;
+ if (!r)
+ {
+ /* No root? Add node as new root. */
+ f->min_root = ni;
+ }
+ else
+ {
+ /* Add node as sibling of current root. */
+ fheap_node_add_sibling (f, ri, ni);
+
+ /* New node may become new root. */
+ if (r->key > n->key)
+ f->min_root = ni;
+ }
+
+ fheap_validate (f);
+}
+
+always_inline u32
+fheap_node_remove_internal (fheap_t * f, u32 ni, u32 invalidate)
+{
+ fheap_node_t *n = vec_elt_at_index (f->nodes, ni);
+ u32 prev_ni = n->prev_sibling;
+ u32 next_ni = n->next_sibling;
+ u32 list_has_single_element = prev_ni == ni;
+ fheap_node_t *prev = fheap_get_node (f, prev_ni);
+ fheap_node_t *next = fheap_get_node (f, next_ni);
+ fheap_node_t *p = fheap_get_node (f, n->parent);
+
+ if (p)
+ {
+ ASSERT (p->rank > 0);
+ p->rank -= 1;
+ p->first_child = list_has_single_element ? ~0 : next_ni;
+ }
+
+ if (prev)
+ {
+ ASSERT (prev->next_sibling == ni);
+ prev->next_sibling = next_ni;
+ }
+ if (next)
+ {
+ ASSERT (next->prev_sibling == ni);
+ next->prev_sibling = prev_ni;
+ }
+
+ n->prev_sibling = n->next_sibling = ni;
+ n->parent = ~0;
+ n->is_valid = invalidate == 0;
+
+ return list_has_single_element ? ~0 : next_ni;
+}
+
+always_inline u32
+fheap_node_remove (fheap_t * f, u32 ni)
+{
+ return fheap_node_remove_internal (f, ni, /* invalidate */ 0);
+}
+
+always_inline u32
+fheap_node_remove_and_invalidate (fheap_t * f, u32 ni)
+{
+ return fheap_node_remove_internal (f, ni, /* invalidate */ 1);
+}
+
+static void
+fheap_link_root (fheap_t * f, u32 ni)
+{
+ fheap_node_t *n = vec_elt_at_index (f->nodes, ni);
+ fheap_node_t *r, *lo, *hi;
+ u32 ri, lo_i, hi_i, k;
+
+ while (1)
+ {
+ k = n->rank;
+ vec_validate_init_empty (f->root_list_by_rank, k, ~0);
+ ri = f->root_list_by_rank[k];
+ r = fheap_get_node (f, ri);
+ if (!r)
+ {
+ f->root_list_by_rank[k] = ni;
+ return;
+ }
+
+ f->root_list_by_rank[k] = ~0;
+
+ /* Sort n/r into lo/hi by their keys. */
+ lo = r, lo_i = ri;
+ hi = n, hi_i = ni;
+ if (hi->key < lo->key)
+ {
+ u32 ti;
+ fheap_node_t *tn;
+ ti = lo_i, tn = lo;
+ lo = hi, lo_i = hi_i;
+ hi = tn, hi_i = ti;
+ }
+
+ /* Remove larger key. */
+ fheap_node_remove (f, hi_i);
+
+ /* Add larger key as child of smaller one. */
+ if (lo->first_child == ~0)
+ {
+ hi->parent = lo_i;
+ lo->first_child = hi_i;
+ lo->rank = 1;
+ }
+ else
+ fheap_node_add_sibling (f, lo->first_child, hi_i);
+
+ /* Following Fredman & Trajan: "When making a root node X a child of another node in a linking step,
+ we unmark X". */
+ hi->is_marked = 0;
+
+ ni = lo_i;
+ n = lo;
+ }
+}
+
+u32
+fheap_del_min (fheap_t * f, u32 * min_key)
+{
+ fheap_node_t *r = fheap_get_root (f);
+ u32 to_delete_min_ri = f->min_root;
+ u32 ri, ni;
+
+ /* Empty heap? */
+ if (!r)
+ return ~0;
+
+ /* Root's children become siblings. Call this step a; see below. */
+ if (r->first_child != ~0)
+ {
+ u32 ci, cni, rni;
+ fheap_node_t *c, *cn, *rn;
+
+ /* Splice child & root circular lists together. */
+ ci = r->first_child;
+ c = vec_elt_at_index (f->nodes, ci);
+
+ cni = c->next_sibling;
+ rni = r->next_sibling;
+ cn = vec_elt_at_index (f->nodes, cni);
+ rn = vec_elt_at_index (f->nodes, rni);
+
+ r->next_sibling = cni;
+ c->next_sibling = rni;
+ cn->prev_sibling = to_delete_min_ri;
+ rn->prev_sibling = ci;
+ }
+
+ /* Remove min root. */
+ ri = fheap_node_remove_and_invalidate (f, to_delete_min_ri);
+
+ /* Find new min root from among siblings including the ones we've just added. */
+ f->min_root = ~0;
+ if (ri != ~0)
+ {
+ u32 ri_last, ri_next, i, min_ds;
+
+ r = fheap_get_node (f, ri);
+ ri_last = r->prev_sibling;
+ while (1)
+ {
+ /* Step a above can put children (with r->parent != ~0) on root list. */
+ r->parent = ~0;
+
+ ri_next = r->next_sibling;
+ fheap_link_root (f, ri);
+ if (ri == ri_last)
+ break;
+ ri = ri_next;
+ r = fheap_get_node (f, ri);
+ }
+
+ min_ds = ~0;
+ vec_foreach_index (i, f->root_list_by_rank)
+ {
+ ni = f->root_list_by_rank[i];
+ if (ni == ~0)
+ continue;
+ f->root_list_by_rank[i] = ~0;
+ r = fheap_get_node (f, ni);
+ if (r->key < min_ds)
+ {
+ f->min_root = ni;
+ min_ds = r->key;
+ ASSERT (r->parent == ~0);
+ }
+ }
+ }
+
+ /* Return deleted min root. */
+ r = vec_elt_at_index (f->nodes, to_delete_min_ri);
+ if (min_key)
+ *min_key = r->key;
+
+ fheap_validate (f);
+
+ return to_delete_min_ri;
+}
+
+static void
+fheap_mark_parent (fheap_t * f, u32 pi)
+{
+ fheap_node_t *p = vec_elt_at_index (f->nodes, pi);
+
+ /* Parent is a root: do nothing. */
+ if (p->parent == ~0)
+ return;
+
+ /* If not marked, mark it. */
+ if (!p->is_marked)
+ {
+ p->is_marked = 1;
+ return;
+ }
+
+ /* Its a previously marked, non-root parent.
+ Cut edge to its parent and add to root list. */
+ fheap_node_remove (f, pi);
+ fheap_node_add_sibling (f, f->min_root, pi);
+
+ /* Unmark it since its now a root node. */
+ p->is_marked = 0;
+
+ /* "Cascading cuts": check parent. */
+ if (p->parent != ~0)
+ fheap_mark_parent (f, p->parent);
+}
+
+/* Set key to new smaller value. */
+void
+fheap_decrease_key (fheap_t * f, u32 ni, u32 new_key)
+{
+ fheap_node_t *n = vec_elt_at_index (f->nodes, ni);
+ fheap_node_t *r = fheap_get_root (f);
+
+ n->key = new_key;
+
+ if (n->parent != ~0)
+ {
+ fheap_mark_parent (f, n->parent);
+
+ /* Remove node and add to root list. */
+ fheap_node_remove (f, ni);
+ fheap_node_add_sibling (f, f->min_root, ni);
+ }
+
+ if (n->key < r->key)
+ f->min_root = ni;
+
+ fheap_validate (f);
+}
+
+void
+fheap_del (fheap_t * f, u32 ni)
+{
+ fheap_node_t *n;
+
+ n = vec_elt_at_index (f->nodes, ni);
+
+ if (n->parent == ~0)
+ {
+ ASSERT (ni == f->min_root);
+ fheap_del_min (f, 0);
+ }
+ else
+ {
+ u32 ci;
+
+ fheap_mark_parent (f, n->parent);
+
+ /* Add children to root list. */
+ foreach_fheap_node_sibling (f, ci, n->first_child, (
+ {
+ fheap_node_remove
+ (f, ci);
+ fheap_node_add_sibling
+ (f, f->min_root,
+ ci);}
+ ));
+
+ fheap_node_remove_and_invalidate (f, ni);
+ }
+
+ fheap_validate (f);
+}
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/fheap.h b/extras/deprecated/vppinfra/fheap.h
new file mode 100644
index 00000000000..1dbd52bad76
--- /dev/null
+++ b/extras/deprecated/vppinfra/fheap.h
@@ -0,0 +1,140 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#ifndef included_clib_fheap_h
+#define included_clib_fheap_h
+
+/* Fibonacci Heaps Fredman, M. L.; Tarjan (1987).
+ "Fibonacci heaps and their uses in improved network optimization algorithms" */
+
+#include <vppinfra/vec.h>
+
+typedef struct
+{
+ /* Node index of parent. */
+ u32 parent;
+
+ /* Node index of first child. */
+ u32 first_child;
+
+ /* Next and previous nodes in doubly linked list of siblings. */
+ u32 next_sibling, prev_sibling;
+
+ /* Key (distance) for this node. Parent always has key
+ <= than keys of children. */
+ u32 key;
+
+ /* Number of children (as opposed to descendents). */
+ u32 rank;
+
+ u32 is_marked;
+
+ /* Set to one when node is inserted; zero when deleted. */
+ u32 is_valid;
+} fheap_node_t;
+
+#define foreach_fheap_node_sibling(f,ni,first_ni,body) \
+do { \
+ u32 __fheap_foreach_first_ni = (first_ni); \
+ u32 __fheap_foreach_ni = __fheap_foreach_first_ni; \
+ u32 __fheap_foreach_next_ni; \
+ fheap_node_t * __fheap_foreach_n; \
+ if (__fheap_foreach_ni != ~0) \
+ while (1) \
+ { \
+ __fheap_foreach_n = fheap_get_node ((f), __fheap_foreach_ni); \
+ __fheap_foreach_next_ni = __fheap_foreach_n -> next_sibling; \
+ (ni) = __fheap_foreach_ni; \
+ \
+ body; \
+ \
+ /* End of circular list? */ \
+ if (__fheap_foreach_next_ni == __fheap_foreach_first_ni) \
+ break; \
+ \
+ __fheap_foreach_ni = __fheap_foreach_next_ni; \
+ \
+ } \
+} while (0)
+
+typedef struct
+{
+ u32 min_root;
+
+ /* Vector of nodes. */
+ fheap_node_t *nodes;
+
+ u32 *root_list_by_rank;
+
+ u32 enable_validate;
+
+ u32 validate_serial;
+} fheap_t;
+
+/* Initialize empty heap. */
+always_inline void
+fheap_init (fheap_t * f, u32 n_nodes)
+{
+ fheap_node_t *save_nodes = f->nodes;
+ u32 *save_root_list = f->root_list_by_rank;
+
+ clib_memset (f, 0, sizeof (f[0]));
+
+ f->nodes = save_nodes;
+ f->root_list_by_rank = save_root_list;
+
+ vec_validate (f->nodes, n_nodes - 1);
+ vec_reset_length (f->root_list_by_rank);
+
+ f->min_root = ~0;
+}
+
+always_inline void
+fheap_free (fheap_t * f)
+{
+ vec_free (f->nodes);
+ vec_free (f->root_list_by_rank);
+}
+
+always_inline u32
+fheap_find_min (fheap_t * f)
+{
+ return f->min_root;
+}
+
+always_inline u32
+fheap_is_empty (fheap_t * f)
+{
+ return f->min_root == ~0;
+}
+
+/* Add/delete nodes. */
+void fheap_add (fheap_t * f, u32 ni, u32 key);
+void fheap_del (fheap_t * f, u32 ni);
+
+/* Delete and return minimum. */
+u32 fheap_del_min (fheap_t * f, u32 * min_key);
+
+/* Change key value. */
+void fheap_decrease_key (fheap_t * f, u32 ni, u32 new_key);
+
+#endif /* included_clib_fheap_h */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/flowhash_24_16.h b/extras/deprecated/vppinfra/flowhash_24_16.h
new file mode 100644
index 00000000000..64ee0796c7a
--- /dev/null
+++ b/extras/deprecated/vppinfra/flowhash_24_16.h
@@ -0,0 +1,77 @@
+/*
+ * Copyright (c) 2012 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef SRC_VPPINFRA_FLOWHASH_24_16_H_
+#define SRC_VPPINFRA_FLOWHASH_24_16_H_
+
+#ifdef __included_flowhash_template_h__
+#undef __included_flowhash_template_h__
+#endif
+
+#include <vppinfra/clib.h>
+#include <vppinfra/xxhash.h>
+#include <vppinfra/crc32.h>
+
+typedef struct {
+ u64 as_u64[3];
+} flowhash_skey_24_16_t;
+
+typedef struct {
+ u64 as_u64[3];
+} flowhash_lkey_24_16_t;
+
+typedef struct {
+ u64 as_u64[2];
+} flowhash_value_24_16_t;
+
+#define FLOWHASH_TYPE _24_16
+#include <vppinfra/flowhash_template.h>
+#undef FLOWHASH_TYPE
+
+static_always_inline
+u32 flowhash_hash_24_16(flowhash_lkey_24_16_t *k)
+{
+#ifdef clib_crc32c_uses_intrinsics
+ return clib_crc32c ((u8 *) &k->as_u64[0], 24);
+#else
+ u64 val = 0;
+ val ^= k->as_u64[0];
+ val ^= k->as_u64[1];
+ val ^= k->as_u64[2];
+ return (u32)clib_xxhash (val);
+#endif
+}
+
+static_always_inline
+u8 flowhash_cmp_key_24_16(flowhash_skey_24_16_t *a,
+ flowhash_lkey_24_16_t *b)
+{
+ u8 val = 0;
+ val |= (a->as_u64[0] != b->as_u64[0]);
+ val |= (a->as_u64[1] != b->as_u64[1]);
+ val |= (a->as_u64[2] != b->as_u64[2]);
+ return val;
+}
+
+static_always_inline
+void flowhash_cpy_key_24_16(flowhash_skey_24_16_t *dst,
+ flowhash_lkey_24_16_t *src)
+{
+ dst->as_u64[0] = src->as_u64[0];
+ dst->as_u64[1] = src->as_u64[1];
+ dst->as_u64[2] = src->as_u64[2];
+}
+
+#endif /* SRC_VPPINFRA_FLOWHASH_24_16_H_ */
diff --git a/extras/deprecated/vppinfra/flowhash_8_8.h b/extras/deprecated/vppinfra/flowhash_8_8.h
new file mode 100644
index 00000000000..4a5cfc0a0c6
--- /dev/null
+++ b/extras/deprecated/vppinfra/flowhash_8_8.h
@@ -0,0 +1,67 @@
+/*
+ * Copyright (c) 2012 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef SRC_VPPINFRA_FLOWHASH_8_8_H_
+#define SRC_VPPINFRA_FLOWHASH_8_8_H_
+
+#ifdef __included_flowhash_template_h__
+#undef __included_flowhash_template_h__
+#endif
+
+#include <vppinfra/clib.h>
+#include <vppinfra/xxhash.h>
+#include <vppinfra/crc32.h>
+
+typedef struct {
+ u64 as_u64[1];
+} flowhash_skey_8_8_t;
+
+typedef struct {
+ u64 as_u64[1];
+} flowhash_lkey_8_8_t;
+
+typedef struct {
+ u64 as_u64[1];
+} flowhash_value_8_8_t;
+
+#define FLOWHASH_TYPE _8_8
+#include <vppinfra/flowhash_template.h>
+#undef FLOWHASH_TYPE
+
+static_always_inline
+u32 flowhash_hash_8_8(flowhash_lkey_8_8_t *k)
+{
+#ifdef clib_crc32c_uses_intrinsics
+ return clib_crc32c ((u8 *) &k->as_u64[0], 8);
+#else
+ return clib_xxhash (k->as_u64[0]);
+#endif
+}
+
+static_always_inline
+u8 flowhash_cmp_key_8_8(flowhash_skey_8_8_t *a,
+ flowhash_lkey_8_8_t *b)
+{
+ return a->as_u64[0] != b->as_u64[0];
+}
+
+static_always_inline
+void flowhash_cpy_key_8_8(flowhash_skey_8_8_t *dst,
+ flowhash_lkey_8_8_t *src)
+{
+ dst->as_u64[0] = src->as_u64[0];
+}
+
+#endif /* SRC_VPPINFRA_FLOWHASH_8_8_H_ */
diff --git a/extras/deprecated/vppinfra/flowhash_template.h b/extras/deprecated/vppinfra/flowhash_template.h
new file mode 100644
index 00000000000..d7a621c1754
--- /dev/null
+++ b/extras/deprecated/vppinfra/flowhash_template.h
@@ -0,0 +1,608 @@
+/*
+ * Copyright (c) 2012 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+/*
+ * Author: Pierre Pfister <ppfister@cisco.com>
+ *
+ * DISCLAIMER !
+ *
+ * This most likely is not the hash table you are looking for !!
+ *
+ * This structure targets a very specific and quite narrow set of use-cases
+ * that are not covered by other hash tables.
+ *
+ * Read the following text carefully, or ask the author or one of VPP's
+ * committers to make sure this is what you are looking for.
+ *
+ *
+ * -- Abstract:
+ * This hash table intends to provide a very fast lookup and insertion of
+ * key-value pairs for flow tables (although it might be used for other
+ * purposes), with additional support for lazy-timeouts.
+ * In particular, it was designed to minimize blocking reads, register usage and
+ * cache-lines accesses during a typical lookup.
+ * This hash table therefore provides stateful packet processing
+ * without performance degradation even when every single lookup has to fetch
+ * memory from RAM.
+ * This hash table is not thread-safe and requires executing a garbage
+ * collection function to clean-up chained buckets.
+ *
+ * -- Overview:
+ *
+ * One first aspect of this hash table is that it is self-contained in a single
+ * bulk of memory. Each entry contains a key, a value, and a 32 bits timeout
+ * value; occupies a full and single cache line; and is identified by a unique
+ * 32 bits index. The entry index zero is reserved and used when an entry
+ * could not be found nor inserted. Which means it is not necessary to
+ * immediately check whether an insertion or lookup was successful before
+ * behaving accordingly. One can just keep doing business as usual and
+ * check for the error later.
+ *
+ * Each entry is associated with a timeout value (which unit or clock is up to
+ * the user of the hash table). An entry which timeout is strictly smaller
+ * than the current time is considered empty, whereas an entry which timeout is
+ * greater or equal to the current time contains a valid key-value pair.
+ *
+ * Hash table lookup and insertion are equivalent:
+ * - An entry index is always returned (possibly index 0 if no entry could be
+ * found nor created).
+ * - The returned entry always has its key set to the provided key.
+ * - Timeout value will be greater than the provided current time whenever a
+ * valid entry was found, strictly smaller otherwise. In particular, one can
+ * not differentiate between an entry which was just created, and an entry
+ * which already existed in the past but got timeouted in between.
+ *
+ * As mentioned earlier, entry index zero is used as an invalid entry which may
+ * be manipulated as a normal one. Entries which index go from 1 to
+ * N (where N is a power of 2) are used as direct buckets, each containing a
+ * single entry. In the absence of hash collision, a single entry which location
+ * can deterministically be determined from the key-hash and the hash table
+ * header is accessed (One single cache line, without indirection). This
+ * allows for efficient pre-fetching of the key-value for more than 95% of
+ * accesses.
+ *
+ * In order to handle hash collisions (i.e. when multiple keys
+ * end-up in the same bucket), entries which index are greater than N are
+ * grouped into M groups of 16 collision entries. Such groups are linked
+ * with regular entries whenever a collision needs to be handled.
+ * When looking up a key with a bucket where a collision occurred, unused bits
+ * from the key hash are used to select two entries (from the collision bucket)
+ * where the new entry might be inserted.
+ *
+ * Once an entry is inserted, it will never be moved as long as the entry
+ * timeout value remains greater or equal to the provided current time value.
+ * The entry index can therefore be stored in other data structure as a way
+ * to bypass the hash lookup. But when doing so, one should check if the
+ * present key is the actual looked-up key.
+ *
+ * -- Garbage Collection:
+ *
+ * Since there is no explicit element removal, a garbage collector mechanism
+ * is required in order to remove buckets used for hash collisions. This
+ * is done by calling the flowhash_gc function on a regular basis. Each call
+ * to this function examines a single fixed entry. It shall therefore be called
+ * as many times as there are fixed entries in the hash table in order to
+ * ensure a full inspection.
+ *
+ * -- Time and timeout mechanism:
+ *
+ * The hash table makes use of a time value between in [1, 2^32 - 1].
+ * The provided time value shall keep increasing, and looping is not handled.
+ * When seconds are used, the system should run for 136 years without any issue.
+ * If milliseconds are used, a shift should be operated on all timeout values
+ * on a regular basis (more than every 49 days).
+ */
+
+#ifndef __included_flowhash_template_h__
+#define __included_flowhash_template_h__
+
+#include <vppinfra/clib.h>
+#include <vppinfra/mem.h>
+#include <vppinfra/cache.h>
+
+#ifndef FLOWHASH_TYPE
+#error FLOWHASH_TYPE not defined
+#endif
+
+#define _fv(a,b) a##b
+#define __fv(a,b) _fv(a,b)
+#define FV(a) __fv(a,FLOWHASH_TYPE)
+
+#define _fvt(a,b) a##b##_t
+#define __fvt(a,b) _fvt(a,b)
+#define FVT(a) __fvt(a,FLOWHASH_TYPE)
+
+/* Same for all flowhash variants */
+#ifndef __included_flowhash_common__
+
+#define FLOWHASH_INVALID_ENTRY_INDEX 0
+
+#define FLOWHASH_ENTRIES_PER_BUCKETS_LOG 4
+#define FLOWHASH_ENTRIES_PER_BUCKETS (1 << FLOWHASH_ENTRIES_PER_BUCKETS_LOG)
+
+#endif /* ifndef __included_flowhash_common__ */
+
+ /**
+ * @brief Compare a stored key with a lookup key.
+ *
+ * This function must be defined to use this template. It must return 0
+ * when the two keys are identical, and a different value otherwise.
+ */
+static_always_inline
+u8 FV(flowhash_cmp_key)(FVT(flowhash_skey) *a, FVT(flowhash_lkey) *b);
+
+ /**
+ * @brief Hash a lookup key into a 32 bit integer.
+ *
+ * This function must be defined to use this template.
+ * It must provides close to 32 bits of entropy distributed amongst
+ * all 32 bits of the provided value.
+ * Keys that are equal must have the same hash.
+ */
+ static_always_inline
+ u32 FV(flowhash_hash)(FVT(flowhash_lkey) *k);
+
+/**
+ * @brief Copy a lookup key into a destination stored key.
+ *
+ * This function must be defined to use this template. It must modify the dst
+ * key such that a later call to flowhash_cmp_key with the same arguments
+ * would return 0.
+ */
+static_always_inline
+void FV(flowhash_cpy_key)(FVT(flowhash_skey) *dst, FVT(flowhash_lkey) *src);
+
+/**
+ * @brief One flow hash entry used for both direct buckets and collision
+ * buckets.
+ */
+typedef struct {
+ /* Each entry is cache-line aligned. */
+ CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
+
+ /* Key is first to take advantage of alignment. */
+ FVT(flowhash_skey) key;
+
+ /* Entry value. */
+ FVT(flowhash_value) value;
+
+ /* Timeout value */
+ u32 timeout;
+
+ /* Entry index to the chained bucket. */
+ u32 chained_entry_index;
+} FVT(flowhash_entry);
+
+typedef struct FVT(__flowhash_struct) {
+ /* Cache aligned to simplify allocation. */
+ CLIB_CACHE_LINE_ALIGN_MARK (cacheline0);
+
+ /* Array going downward containing free bucket indices */
+ u32 free_buckets_indices[0];
+
+ /* Negative index of the first free bucket */
+ i32 free_buckets_position;
+
+ /* Number of fixed buckets minus one */
+ u32 fixed_entries_mask;
+
+ /* Allocated pointer for this hash table */
+ void *mem;
+
+ u32 collision_buckets_mask;
+ u32 total_entries;
+
+ u64 not_enough_buckets_counter;
+ u64 collision_lookup_counter;
+ u64 garbage_collection_counter;
+
+ u32 gc_counter;
+
+ /* Entry array containing:
+ * - 1 Dummy entry for error return
+ * - (buckets_mask + 1) Fixed buckets
+ * - chained_buckets Chained Buckets
+ */
+ FVT(flowhash_entry) entries[0];
+} FVT(flowhash);
+
+/* Same for all flowhash variants */
+#ifndef __included_flowhash_common__
+#define __included_flowhash_common__
+
+/**
+ * @brief Test whether a returned entry index corresponds to an overflow event.
+ */
+#define flowhash_is_overflow(ei) \
+ ((ei) == FLOWHASH_INVALID_ENTRY_INDEX)
+
+/**
+ * @brief Iterate over all entries in the hash table.
+ *
+ * Iterate over all entries in the hash table, not including the first invalid
+ * entry (at index 0), but including all chained hash collision buckets.
+ *
+ */
+#define flowhash_foreach_entry(h, ei) \
+ for (ei = 1; \
+ ei < (h)->total_entries; \
+ ei++)
+
+/**
+ * @brief Iterate over all currently valid entries.
+ *
+ * Iterate over all entries in the hash table which timeout value is greater
+ * or equal to the current time.
+ */
+#define flowhash_foreach_valid_entry(h, ei, now) \
+ flowhash_foreach_entry(h, ei) \
+ if (((now) <= (h)->entries[ei].timeout))
+
+/**
+ * @brief Timeout variable from a given entry.
+ */
+#define flowhash_timeout(h, ei) (h)->entries[ei].timeout
+
+/**
+ * @brief Indicates whether the entry is being used.
+ */
+#define flowhash_is_timeouted(h, ei, time_now) \
+ ((time_now) > flowhash_timeout(h, ei))
+
+/**
+ * @brief Get the key from the entry index, casted to the provided type.
+ */
+#define flowhash_key(h, ei) (&(h)->entries[ei].key)
+
+/**
+ * @brief Get the value from the entry index, casted to the provided type.
+ */
+#define flowhash_value(h, ei) (&(h)->entries[ei].value)
+
+/**
+ * @brief Get the number of octets allocated to this structure.
+ */
+#define flowhash_memory_size(h) clib_mem_size((h)->mem)
+
+/**
+ * @brief Test whether the entry index is in hash table boundaries.
+ */
+#define flowhash_is_valid_entry_index(h, ei) (ei < (h)->total_entries)
+
+/**
+ * @brief Adjust, if necessary, provided parameters such as being valid flowhash
+ * sizes.
+ */
+static
+void flowhash_validate_sizes(u32 *fixed_entries, u32 *collision_buckets)
+{
+ /* Find power of two greater or equal to the provided value */
+ if (*fixed_entries < FLOWHASH_ENTRIES_PER_BUCKETS)
+ *fixed_entries = FLOWHASH_ENTRIES_PER_BUCKETS;
+ if (*fixed_entries > (1 << (32 - FLOWHASH_ENTRIES_PER_BUCKETS_LOG)))
+ *fixed_entries = (1 << (32 - FLOWHASH_ENTRIES_PER_BUCKETS_LOG));
+
+ *fixed_entries -= 1;
+ *fixed_entries |= *fixed_entries >> 16;
+ *fixed_entries |= *fixed_entries >> 8;
+ *fixed_entries |= *fixed_entries >> 4;
+ *fixed_entries |= *fixed_entries >> 2;
+ *fixed_entries |= *fixed_entries >> 1;
+ *fixed_entries += 1;
+
+ if (*collision_buckets != 0)
+ {
+ if (*collision_buckets < CLIB_CACHE_LINE_BYTES/sizeof(u32))
+ *collision_buckets = CLIB_CACHE_LINE_BYTES/sizeof(u32);
+
+ *collision_buckets -= 1;
+ *collision_buckets |= *collision_buckets >> 16;
+ *collision_buckets |= *collision_buckets >> 8;
+ *collision_buckets |= *collision_buckets >> 4;
+ *collision_buckets |= *collision_buckets >> 2;
+ *collision_buckets |= *collision_buckets >> 1;
+ *collision_buckets += 1;
+ }
+}
+
+/**
+ * @brief Prefetch the the hash entry bucket.
+ *
+ * This should be performed approximately 200-300 cycles before lookup
+ * if the table is located in RAM. Or 30-40 cycles before lookup
+ * in case the table is located in L3.
+ */
+#define flowhash_prefetch(h, hash) \
+ CLIB_PREFETCH (&(h)->entries[((hash) & (h)->fixed_entries_mask) + 1], \
+ sizeof((h)->entries[0]), LOAD)
+
+#endif /* ifndef __included_flowhash_common__ */
+
+/**
+ * @brief Allocate a flowhash structure.
+ *
+ * @param[in] fixed_entries The number of fixed entries in the hash table.
+ * @param[in] chained_buckets The number of chained buckets.
+ *
+ * fixed_entries and chained_buckets parameters may not be used as is but
+ * modified in order to fit requirements.
+ *
+ * Since the flowhash does not support dynamic resizing, it is fairly
+ * important to choose the parameters carefully. In particular the performance
+ * gain from using this structure comes from an efficient lookup in the
+ * absence of hash collision.
+ * As a rule of thumbs, if the number of active entries (flows) is M,
+ * there should be about 16*M fixed entries, and M/16 collision buckets.
+ * Which represents 17*M allocated entries.
+ *
+ * For example:
+ * M = 2^20 total_size ~= 1GiB collision ~= 3%
+ * M = 2^18 total_size ~= 250MiB collision ~= 3%
+ * M = 2^10 total_size ~= 1MiB collision ~= 6%
+ *
+ */
+static_always_inline
+FVT(flowhash) *FV(flowhash_alloc)(u32 fixed_entries, u32 collision_buckets)
+{
+ FVT(flowhash) *h;
+ uword size;
+ void *mem;
+ u32 entries;
+
+ flowhash_validate_sizes(&fixed_entries, &collision_buckets);
+
+ entries = 1 + fixed_entries +
+ collision_buckets * FLOWHASH_ENTRIES_PER_BUCKETS;
+ size = sizeof(*h) + sizeof(h->entries[0]) * entries +
+ sizeof(h->free_buckets_indices[0]) * collision_buckets;
+
+ mem = clib_mem_alloc_aligned(size, CLIB_CACHE_LINE_BYTES);
+ h = mem + collision_buckets * sizeof(h->free_buckets_indices[0]);
+ h->mem = mem;
+
+ /* Fill free elements list */
+ int i;
+ clib_memset(h->entries, 0, sizeof(h->entries[0]) * entries);
+ for (i = 1; i <= collision_buckets; i++)
+ {
+ h->free_buckets_indices[-i] =
+ entries - i * FLOWHASH_ENTRIES_PER_BUCKETS;
+ }
+
+ /* Init buckets */
+ for (i=0; i < entries; i++)
+ {
+ h->entries[i].chained_entry_index = FLOWHASH_INVALID_ENTRY_INDEX;
+ h->entries[i].timeout = 0;
+ }
+
+ h->free_buckets_position = -collision_buckets;
+ h->fixed_entries_mask = fixed_entries - 1;
+ h->collision_buckets_mask = collision_buckets - 1;
+ h->total_entries = entries;
+ h->not_enough_buckets_counter = 0;
+ h->collision_lookup_counter = 0;
+ h->garbage_collection_counter = 0;
+ h->gc_counter = 0;
+
+ return h;
+}
+
+/**
+ * @brief Free the flow hash memory.
+ */
+static_always_inline
+void FV(flowhash_free)(FVT(flowhash) *h)
+{
+ clib_mem_free(h->mem);
+}
+
+static void
+FV(__flowhash_get_chained) (FVT(flowhash) *h, FVT(flowhash_lkey) *k,
+ u32 hash, u32 time_now, u32 *ei);
+
+/**
+ * @brief Retrieves an entry index corresponding to a provided key and its hash.
+ *
+ * @param h The hash table pointer.
+ * @param k[in] A pointer to the key value.
+ * @param hash[in] The hash of the key.
+ * @param time_now[in] The current time.
+ * @param ei[out] A pointer set to the found entry index.
+ *
+ * This function always sets ei value to a valid entry index which can then be
+ * used to access the stored value as well as get or set its associated timeout.
+ * The key stored in the returned entry is always set to the provided key.
+ *
+ * In case the provided key is not found, and no entry could be created
+ * (either because there is no hash collision bucket available or
+ * the candidate entries in the collision bucket were already used), ei is
+ * set to the special value FLOWHASH_INVALID_ENTRY_INDEX (which can be tested
+ * with the flowhash_is_overflow macro).
+ *
+ * The timeout value is never modified during a lookup.
+ * - Use the flowhash_is_timeouted macro to test whether the returned entry
+ * was already valid, or is proposed for insertion.
+ * - Use the flowhash_timeout macro to get and set the entry timeout value.
+ *
+ */
+static_always_inline
+void FV(flowhash_get) (FVT(flowhash) *h, FVT(flowhash_lkey) *k,
+ u32 hash, u32 time_now, u32 *ei)
+{
+ *ei = (hash & h->fixed_entries_mask) + 1;
+
+ if (PREDICT_FALSE(FV(flowhash_cmp_key)(&h->entries[*ei].key, k) != 0))
+ {
+ if (PREDICT_TRUE(time_now > h->entries[*ei].timeout &&
+ (h->entries[*ei].chained_entry_index ==
+ FLOWHASH_INVALID_ENTRY_INDEX)))
+ {
+ FV(flowhash_cpy_key)(&h->entries[*ei].key, k);
+ }
+ else
+ {
+ FV(__flowhash_get_chained)(h, k, hash, time_now, ei);
+ }
+ }
+}
+
+static_always_inline void
+FV(__flowhash_get_chained) (FVT(flowhash) *h, FVT(flowhash_lkey) *k,
+ u32 hash, u32 time_now, u32 *ei)
+{
+ h->collision_lookup_counter++;
+
+ if (h->entries[*ei].chained_entry_index == FLOWHASH_INVALID_ENTRY_INDEX)
+ {
+ /* No chained entry yet. Let's chain one. */
+ if (h->free_buckets_position == 0)
+ {
+ /* Oops. No more buckets available. */
+ h->not_enough_buckets_counter++;
+ *ei = FLOWHASH_INVALID_ENTRY_INDEX;
+ h->entries[FLOWHASH_INVALID_ENTRY_INDEX].timeout =
+ time_now - 1;
+ FV(flowhash_cpy_key)(
+ &h->entries[FLOWHASH_INVALID_ENTRY_INDEX].key, k);
+ return;
+ }
+
+ /* Forward link */
+ h->entries[*ei].chained_entry_index =
+ h->free_buckets_indices[h->free_buckets_position];
+
+ /* Backward link (for garbage collection) */
+ h->entries[h->free_buckets_indices[h->free_buckets_position]].
+ chained_entry_index = *ei;
+
+ /* Move pointer */
+ h->free_buckets_position++;
+ }
+
+ /* Get the two indexes where to look at. */
+ u32 bi0 = h->entries[*ei].chained_entry_index +
+ (hash >> (32 - FLOWHASH_ENTRIES_PER_BUCKETS_LOG));
+ u32 bi1 = bi0 + 1;
+ bi1 = (bi0 & (FLOWHASH_ENTRIES_PER_BUCKETS - 1)) ? bi1 :
+ bi1 - FLOWHASH_ENTRIES_PER_BUCKETS;
+
+ /* It is possible that we wait while comparing bi0 key.
+ * It's better to prefetch bi1 so we don't wait twice. */
+ CLIB_PREFETCH(&h->entries[bi1], sizeof (h->entries[0]), READ);
+
+ if (FV(flowhash_cmp_key)(&h->entries[bi0].key, k) == 0)
+ {
+ *ei = bi0;
+ return;
+ }
+
+ if (FV(flowhash_cmp_key)(&h->entries[bi1].key, k) == 0)
+ {
+ *ei = bi1;
+ return;
+ }
+
+ if (h->entries[*ei].timeout >= time_now)
+ {
+ *ei = FLOWHASH_INVALID_ENTRY_INDEX;
+ *ei = (time_now > h->entries[bi0].timeout) ? bi0 : *ei;
+ *ei = (time_now > h->entries[bi1].timeout) ? bi1 : *ei;
+ }
+
+ FV(flowhash_cpy_key)(&h->entries[*ei].key, k);
+}
+
+static_always_inline void
+FV(flowhash_gc)(FVT(flowhash) *h, u32 time_now,
+ u32 *freed_index, u32 *freed_len)
+{
+ u32 ei;
+ if (freed_index)
+ *freed_len = 0;
+
+ if (PREDICT_FALSE(h->collision_buckets_mask == (((u32)0) - 1)))
+ return;
+
+ /* prefetch two rounds in advance */
+ ei = 2 + h->fixed_entries_mask +
+ ((h->gc_counter + 2) & h->collision_buckets_mask) *
+ FLOWHASH_ENTRIES_PER_BUCKETS;
+ CLIB_PREFETCH(&h->entries[ei], sizeof (h->entries[0]), READ);
+
+ /* prefetch one round in advance */
+ ei = 2 + h->fixed_entries_mask +
+ ((h->gc_counter + 1) & h->collision_buckets_mask) *
+ FLOWHASH_ENTRIES_PER_BUCKETS;
+ if (h->entries[ei].chained_entry_index != FLOWHASH_INVALID_ENTRY_INDEX)
+ {
+ CLIB_PREFETCH(&h->entries[ei], 4 * CLIB_CACHE_LINE_BYTES, READ);
+ }
+
+ /* do GC */
+ ei = 2 + h->fixed_entries_mask +
+ ((h->gc_counter) & h->collision_buckets_mask) *
+ FLOWHASH_ENTRIES_PER_BUCKETS;
+ if (h->entries[ei].chained_entry_index != FLOWHASH_INVALID_ENTRY_INDEX)
+ {
+ u8 found = 0;
+ int i;
+ for (i=0; i<FLOWHASH_ENTRIES_PER_BUCKETS; i++)
+ {
+ if (time_now <= h->entries[ei + i].timeout)
+ {
+ found = 1;
+ break;
+ }
+ }
+
+ if (!found)
+ {
+ /* Tell caller we freed this */
+ if (freed_index)
+ {
+ *freed_index = ei;
+ *freed_len = FLOWHASH_ENTRIES_PER_BUCKETS;
+ }
+ /* The bucket is not used. Let's free it. */
+ h->free_buckets_position--;
+ /* Reset forward link */
+ h->entries[h->entries[ei].chained_entry_index].chained_entry_index =
+ FLOWHASH_INVALID_ENTRY_INDEX;
+ /* Reset back link */
+ h->entries[ei].chained_entry_index = FLOWHASH_INVALID_ENTRY_INDEX;
+ /* Free element */
+ h->free_buckets_indices[h->free_buckets_position] = ei;
+ /* Count the garbage collection event */
+ h->garbage_collection_counter++;
+ }
+ }
+
+ h->gc_counter++;
+}
+
+static_always_inline
+u32 FV(flowhash_elts)(FVT(flowhash) *h, u32 time_now)
+{
+ u32 tot = 0;
+ u32 ei;
+
+ flowhash_foreach_valid_entry(h, ei, time_now)
+ tot++;
+
+ return tot;
+}
+
+#endif /* __included_flowhash_template_h__ */
diff --git a/extras/deprecated/vppinfra/pfhash.h b/extras/deprecated/vppinfra/pfhash.h
new file mode 100644
index 00000000000..2884fa81cf9
--- /dev/null
+++ b/extras/deprecated/vppinfra/pfhash.h
@@ -0,0 +1,276 @@
+/*
+ Copyright (c) 2013 Cisco and/or its affiliates.
+
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+*/
+
+#ifndef included_clib_pfhash_h
+#define included_clib_pfhash_h
+
+
+#include <vppinfra/clib.h>
+#include <vppinfra/hash.h>
+#include <vppinfra/pool.h>
+
+#if defined(CLIB_HAVE_VEC128) && ! defined (__ALTIVEC__)
+
+typedef struct
+{
+ /* 3 x 16 = 48 key bytes */
+ union
+ {
+ u32x4 k_u32x4[3];
+ u64 k_u64[6];
+ } kb;
+ /* 3 x 4 = 12 value bytes */
+ u32 values[3];
+ u32 pad;
+} pfhash_kv_16_t;
+
+typedef struct
+{
+ /* 5 x 8 = 40 key bytes */
+ union
+ {
+ u64 k_u64[5];
+ } kb;
+
+ /* 5 x 4 = 20 value bytes */
+ u32 values[5];
+ u32 pad;
+} pfhash_kv_8_t;
+
+typedef struct
+{
+ /* 4 x 8 = 32 key bytes */
+ union
+ {
+ u64 k_u64[4];
+ } kb;
+
+ /* 4 x 8 = 32 value bytes */
+ u64 values[4];
+} pfhash_kv_8v8_t;
+
+typedef struct
+{
+ /* 8 x 4 = 32 key bytes */
+ union
+ {
+ u32x4 k_u32x4[2];
+ u32 kb[8];
+ } kb;
+
+ /* 8 x 4 = 32 value bytes */
+ u32 values[8];
+} pfhash_kv_4_t;
+
+typedef union
+{
+ pfhash_kv_16_t kv16;
+ pfhash_kv_8_t kv8;
+ pfhash_kv_8v8_t kv8v8;
+ pfhash_kv_4_t kv4;
+} pfhash_kv_t;
+
+typedef struct
+{
+ /* Bucket vector */
+ u32 *buckets;
+#define PFHASH_BUCKET_OVERFLOW (u32)~0
+
+ /* Pool of key/value pairs */
+ pfhash_kv_t *kvp;
+
+ /* overflow plain-o-hash */
+ uword *overflow_hash;
+
+ /* Pretty-print name */
+ u8 *name;
+
+ u32 key_size;
+ u32 value_size;
+
+ u32 overflow_count;
+ u32 nitems;
+ u32 nitems_in_overflow;
+} pfhash_t;
+
+void pfhash_init (pfhash_t * p, char *name, u32 key_size, u32 value_size,
+ u32 nbuckets);
+void pfhash_free (pfhash_t * p);
+u64 pfhash_get (pfhash_t * p, u32 bucket, void *key);
+void pfhash_set (pfhash_t * p, u32 bucket, void *key, void *value);
+void pfhash_unset (pfhash_t * p, u32 bucket, void *key);
+
+format_function_t format_pfhash;
+
+static inline void
+pfhash_prefetch_bucket (pfhash_t * p, u32 bucket)
+{
+ CLIB_PREFETCH (&p->buckets[bucket], CLIB_CACHE_LINE_BYTES, LOAD);
+}
+
+static inline u32
+pfhash_read_bucket_prefetch_kv (pfhash_t * p, u32 bucket)
+{
+ u32 bucket_contents = p->buckets[bucket];
+ if (PREDICT_TRUE ((bucket_contents & PFHASH_BUCKET_OVERFLOW) == 0))
+ CLIB_PREFETCH (&p->kvp[bucket_contents], CLIB_CACHE_LINE_BYTES, LOAD);
+ return bucket_contents;
+}
+
+/*
+ * pfhash_search_kv_16
+ * See if the supplied 16-byte key matches one of three 16-byte (key,value) pairs.
+ * Return the indicated value, or ~0 if no match
+ *
+ * Note: including the overflow test, the fast path is 35 instrs
+ * on x86_64. Elves will steal your keyboard in the middle of the night if
+ * you "improve" it without checking the generated code!
+ */
+static inline u32
+pfhash_search_kv_16 (pfhash_t * p, u32 bucket_contents, u32x4 * key)
+{
+ u32x4 diff0, diff1, diff2;
+ u32 is_equal0, is_equal1, is_equal2;
+ u32 no_match;
+ pfhash_kv_16_t *kv;
+ u32 rv;
+
+ if (PREDICT_FALSE (bucket_contents == PFHASH_BUCKET_OVERFLOW))
+ {
+ uword *hp;
+ hp = hash_get_mem (p->overflow_hash, key);
+ if (hp)
+ return hp[0];
+ return (u32) ~ 0;
+ }
+
+ kv = &p->kvp[bucket_contents].kv16;
+
+ diff0 = u32x4_sub (kv->kb.k_u32x4[0], key[0]);
+ diff1 = u32x4_sub (kv->kb.k_u32x4[1], key[0]);
+ diff2 = u32x4_sub (kv->kb.k_u32x4[2], key[0]);
+
+ no_match = is_equal0 = (i16) u32x4_zero_byte_mask (diff0);
+ is_equal1 = (i16) u32x4_zero_byte_mask (diff1);
+ no_match |= is_equal1;
+ is_equal2 = (i16) u32x4_zero_byte_mask (diff2);
+ no_match |= is_equal2;
+ /* If any of the three items matched, no_match will be zero after this line */
+ no_match = ~no_match;
+
+ rv = (is_equal0 & kv->values[0])
+ | (is_equal1 & kv->values[1]) | (is_equal2 & kv->values[2]) | no_match;
+
+ return rv;
+}
+
+static inline u32
+pfhash_search_kv_8 (pfhash_t * p, u32 bucket_contents, u64 * key)
+{
+ pfhash_kv_8_t *kv;
+ u32 rv = (u32) ~ 0;
+
+ if (PREDICT_FALSE (bucket_contents == PFHASH_BUCKET_OVERFLOW))
+ {
+ uword *hp;
+ hp = hash_get_mem (p->overflow_hash, key);
+ if (hp)
+ return hp[0];
+ return (u32) ~ 0;
+ }
+
+ kv = &p->kvp[bucket_contents].kv8;
+
+ rv = (kv->kb.k_u64[0] == key[0]) ? kv->values[0] : rv;
+ rv = (kv->kb.k_u64[1] == key[0]) ? kv->values[1] : rv;
+ rv = (kv->kb.k_u64[2] == key[0]) ? kv->values[2] : rv;
+ rv = (kv->kb.k_u64[3] == key[0]) ? kv->values[3] : rv;
+ rv = (kv->kb.k_u64[4] == key[0]) ? kv->values[4] : rv;
+
+ return rv;
+}
+
+static inline u64
+pfhash_search_kv_8v8 (pfhash_t * p, u32 bucket_contents, u64 * key)
+{
+ pfhash_kv_8v8_t *kv;
+ u64 rv = (u64) ~ 0;
+
+ if (PREDICT_FALSE (bucket_contents == PFHASH_BUCKET_OVERFLOW))
+ {
+ uword *hp;
+ hp = hash_get_mem (p->overflow_hash, key);
+ if (hp)
+ return hp[0];
+ return (u64) ~ 0;
+ }
+
+ kv = &p->kvp[bucket_contents].kv8v8;
+
+ rv = (kv->kb.k_u64[0] == key[0]) ? kv->values[0] : rv;
+ rv = (kv->kb.k_u64[1] == key[0]) ? kv->values[1] : rv;
+ rv = (kv->kb.k_u64[2] == key[0]) ? kv->values[2] : rv;
+ rv = (kv->kb.k_u64[3] == key[0]) ? kv->values[3] : rv;
+
+ return rv;
+}
+
+static inline u32
+pfhash_search_kv_4 (pfhash_t * p, u32 bucket_contents, u32 * key)
+{
+ u32x4 vector_key;
+ u32x4 is_equal[2];
+ u32 zbm[2], winner_index;
+ pfhash_kv_4_t *kv;
+
+ if (PREDICT_FALSE (bucket_contents == PFHASH_BUCKET_OVERFLOW))
+ {
+ uword *hp;
+ hp = hash_get_mem (p->overflow_hash, key);
+ if (hp)
+ return hp[0];
+ return (u32) ~ 0;
+ }
+
+ kv = &p->kvp[bucket_contents].kv4;
+
+ vector_key = u32x4_splat (key[0]);
+
+ is_equal[0] = (kv->kb.k_u32x4[0] == vector_key);
+ is_equal[1] = (kv->kb.k_u32x4[1] == vector_key);
+ zbm[0] = ~u32x4_zero_byte_mask (is_equal[0]) & 0xFFFF;
+ zbm[1] = ~u32x4_zero_byte_mask (is_equal[1]) & 0xFFFF;
+
+ if (PREDICT_FALSE ((zbm[0] == 0) && (zbm[1] == 0)))
+ return (u32) ~ 0;
+
+ winner_index = min_log2 (zbm[0]) >> 2;
+ winner_index = zbm[1] ? (4 + (min_log2 (zbm[1]) >> 2)) : winner_index;
+
+ return kv->values[winner_index];
+}
+
+#endif /* CLIB_HAVE_VEC128 */
+
+#endif /* included_clib_pfhash_h */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/phash.c b/extras/deprecated/vppinfra/phash.c
new file mode 100644
index 00000000000..52c29b33f78
--- /dev/null
+++ b/extras/deprecated/vppinfra/phash.c
@@ -0,0 +1,1017 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ Copyright (c) 2005 Eliot Dresselhaus
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice shall be
+ included in all copies or substantial portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+*/
+
+/* This is all stolen from Bob Jenkins and reworked for clib. Thanks
+ once again Bob for the great work. */
+
+/*
+------------------------------------------------------------------------------
+perfect.c: code to generate code for a hash for perfect hashing.
+(c) Bob Jenkins, September 1996, December 1999
+You may use this code in any way you wish, and it is free. No warranty.
+I hereby place this in the public domain.
+Source is http://burtleburtle.net/bob/c/perfect.c
+
+This generates a minimal perfect hash function. That means, given a
+set of n keys, this determines a hash function that maps each of
+those keys into a value in 0..n-1 with no collisions.
+
+The perfect hash function first uses a normal hash function on the key
+to determine (a,b) such that the pair (a,b) is distinct for all
+keys, then it computes a^scramble[tab[b]] to get the final perfect hash.
+tab[] is an array of 1-byte values and scramble[] is a 256-term array of
+2-byte or 4-byte values. If there are n keys, the length of tab[] is a
+power of two between n/3 and n.
+
+I found the idea of computing distinct (a,b) values in "Practical minimal
+perfect hash functions for large databases", Fox, Heath, Chen, and Daoud,
+Communications of the ACM, January 1992. They found the idea in Chichelli
+(CACM Jan 1980). Beyond that, our methods differ.
+
+The key is hashed to a pair (a,b) where a in 0..*alen*-1 and b in
+0..*blen*-1. A fast hash function determines both a and b
+simultaneously. Any decent hash function is likely to produce
+hashes so that (a,b) is distinct for all pairs. I try the hash
+using different values of *salt* until all pairs are distinct.
+
+The final hash is (a XOR scramble[tab[b]]). *scramble* is a
+predetermined mapping of 0..255 into 0..smax-1. *tab* is an
+array that we fill in in such a way as to make the hash perfect.
+
+First we fill in all values of *tab* that are used by more than one
+key. We try all possible values for each position until one works.
+
+This leaves m unmapped keys and m values that something could hash to.
+If you treat unmapped keys as lefthand nodes and unused hash values
+as righthand nodes, and draw a line connecting each key to each hash
+value it could map to, you get a bipartite graph. We attempt to
+find a perfect matching in this graph. If we succeed, we have
+determined a perfect hash for the whole set of keys.
+
+*scramble* is used because (a^tab[i]) clusters keys around *a*.
+------------------------------------------------------------------------------
+*/
+
+#include <vppinfra/bitmap.h>
+#include <vppinfra/format.h>
+#include <vppinfra/phash.h>
+#include <vppinfra/random.h>
+
+static void
+init_keys_direct_u32 (phash_main_t * pm)
+{
+ int n_keys_left, b_mask, a_shift;
+ u32 seed;
+ phash_key_t *k;
+
+ seed = pm->hash_seed;
+ b_mask = (1 << pm->b_bits) - 1;
+ a_shift = BITS (seed) - pm->a_bits;
+
+ k = pm->keys;
+ n_keys_left = vec_len (pm->keys);
+
+ while (n_keys_left >= 2)
+ {
+ u32 x0, y0, z0;
+ u32 x1, y1, z1;
+
+ x0 = y0 = z0 = seed;
+ x1 = y1 = z1 = seed;
+ x0 += (u32) k[0].key;
+ x1 += (u32) k[1].key;
+
+ hash_mix32 (x0, y0, z0);
+ hash_mix32 (x1, y1, z1);
+
+ k[0].b = z0 & b_mask;
+ k[1].b = z1 & b_mask;
+ k[0].a = z0 >> a_shift;
+ k[1].a = z1 >> a_shift;
+ if (PREDICT_FALSE (a_shift >= BITS (z0)))
+ k[0].a = k[1].a = 0;
+
+ k += 2;
+ n_keys_left -= 2;
+ }
+
+ if (n_keys_left >= 1)
+ {
+ u32 x0, y0, z0;
+
+ x0 = y0 = z0 = seed;
+ x0 += k[0].key;
+
+ hash_mix32 (x0, y0, z0);
+
+ k[0].b = z0 & b_mask;
+ k[0].a = z0 >> a_shift;
+ if (PREDICT_FALSE (a_shift >= BITS (z0)))
+ k[0].a = 0;
+
+ k += 1;
+ n_keys_left -= 1;
+ }
+}
+
+static void
+init_keys_direct_u64 (phash_main_t * pm)
+{
+ int n_keys_left, b_mask, a_shift;
+ u64 seed;
+ phash_key_t *k;
+
+ seed = pm->hash_seed;
+ b_mask = (1 << pm->b_bits) - 1;
+ a_shift = BITS (seed) - pm->a_bits;
+
+ k = pm->keys;
+ n_keys_left = vec_len (pm->keys);
+
+ while (n_keys_left >= 2)
+ {
+ u64 x0, y0, z0;
+ u64 x1, y1, z1;
+
+ x0 = y0 = z0 = seed;
+ x1 = y1 = z1 = seed;
+ x0 += (u64) k[0].key;
+ x1 += (u64) k[1].key;
+
+ hash_mix64 (x0, y0, z0);
+ hash_mix64 (x1, y1, z1);
+
+ k[0].b = z0 & b_mask;
+ k[1].b = z1 & b_mask;
+ k[0].a = z0 >> a_shift;
+ k[1].a = z1 >> a_shift;
+ if (PREDICT_FALSE (a_shift >= BITS (z0)))
+ k[0].a = k[1].a = 0;
+
+ k += 2;
+ n_keys_left -= 2;
+ }
+
+ if (n_keys_left >= 1)
+ {
+ u64 x0, y0, z0;
+
+ x0 = y0 = z0 = seed;
+ x0 += k[0].key;
+
+ hash_mix64 (x0, y0, z0);
+
+ k[0].b = z0 & b_mask;
+ k[0].a = z0 >> a_shift;
+ if (PREDICT_FALSE (a_shift >= BITS (z0)))
+ k[0].a = 0;
+
+ k += 1;
+ n_keys_left -= 1;
+ }
+}
+
+static void
+init_keys_indirect_u32 (phash_main_t * pm)
+{
+ int n_keys_left, b_mask, a_shift;
+ u32 seed;
+ phash_key_t *k;
+
+ seed = pm->hash_seed;
+ b_mask = (1 << pm->b_bits) - 1;
+ a_shift = BITS (seed) - pm->a_bits;
+
+ k = pm->keys;
+ n_keys_left = vec_len (pm->keys);
+
+ while (n_keys_left >= 2)
+ {
+ u32 xyz[6];
+ u32 x0, y0, z0;
+ u32 x1, y1, z1;
+
+ pm->key_seed2 (pm->private, k[0].key, k[1].key, &xyz);
+
+ x0 = y0 = z0 = seed;
+ x1 = y1 = z1 = seed;
+ x0 += xyz[0];
+ y0 += xyz[1];
+ z0 += xyz[2];
+ x1 += xyz[3];
+ y1 += xyz[4];
+ z1 += xyz[5];
+
+ hash_mix32 (x0, y0, z0);
+ hash_mix32 (x1, y1, z1);
+
+ k[0].b = z0 & b_mask;
+ k[1].b = z1 & b_mask;
+ k[0].a = z0 >> a_shift;
+ k[1].a = z1 >> a_shift;
+ if (PREDICT_FALSE (a_shift >= BITS (z0)))
+ k[0].a = k[1].a = 0;
+
+ k += 2;
+ n_keys_left -= 2;
+ }
+
+ if (n_keys_left >= 1)
+ {
+ u32 xyz[3];
+ u32 x0, y0, z0;
+
+ pm->key_seed1 (pm->private, k[0].key, &xyz);
+
+ x0 = y0 = z0 = seed;
+ x0 += xyz[0];
+ y0 += xyz[1];
+ z0 += xyz[2];
+
+ hash_mix32 (x0, y0, z0);
+
+ k[0].b = z0 & b_mask;
+ k[0].a = z0 >> a_shift;
+ if (PREDICT_FALSE (a_shift >= BITS (z0)))
+ k[0].a = 0;
+
+ k += 1;
+ n_keys_left -= 1;
+ }
+}
+
+static void
+init_keys_indirect_u64 (phash_main_t * pm)
+{
+ int n_keys_left, b_mask, a_shift;
+ u64 seed;
+ phash_key_t *k;
+
+ seed = pm->hash_seed;
+ b_mask = (1 << pm->b_bits) - 1;
+ a_shift = BITS (seed) - pm->a_bits;
+
+ k = pm->keys;
+ n_keys_left = vec_len (pm->keys);
+
+ while (n_keys_left >= 2)
+ {
+ u64 xyz[6];
+ u64 x0, y0, z0;
+ u64 x1, y1, z1;
+
+ pm->key_seed2 (pm->private, k[0].key, k[1].key, &xyz);
+
+ x0 = y0 = z0 = seed;
+ x1 = y1 = z1 = seed;
+ x0 += xyz[0];
+ y0 += xyz[1];
+ z0 += xyz[2];
+ x1 += xyz[3];
+ y1 += xyz[4];
+ z1 += xyz[5];
+
+ hash_mix64 (x0, y0, z0);
+ hash_mix64 (x1, y1, z1);
+
+ k[0].b = z0 & b_mask;
+ k[1].b = z1 & b_mask;
+ k[0].a = z0 >> a_shift;
+ k[1].a = z1 >> a_shift;
+ if (PREDICT_FALSE (a_shift >= BITS (z0)))
+ k[0].a = k[1].a = 0;
+
+ k += 2;
+ n_keys_left -= 2;
+ }
+
+ if (n_keys_left >= 1)
+ {
+ u64 xyz[3];
+ u64 x0, y0, z0;
+
+ pm->key_seed1 (pm->private, k[0].key, &xyz);
+
+ x0 = y0 = z0 = seed;
+ x0 += xyz[0];
+ y0 += xyz[1];
+ z0 += xyz[2];
+
+ hash_mix64 (x0, y0, z0);
+
+ k[0].b = z0 & b_mask;
+ k[0].a = z0 >> a_shift;
+ if (PREDICT_FALSE (a_shift >= BITS (z0)))
+ k[0].a = 0;
+
+ k += 1;
+ n_keys_left -= 1;
+ }
+}
+
+/*
+ * insert keys into table according to key->b
+ * check if the initial hash might work
+ */
+static int
+init_tabb (phash_main_t * pm)
+{
+ int no_collisions;
+ phash_tabb_t *tb;
+ phash_key_t *k, *l;
+
+ if (pm->key_seed1)
+ {
+ if (pm->flags & PHASH_FLAG_MIX64)
+ init_keys_indirect_u64 (pm);
+ else
+ init_keys_indirect_u32 (pm);
+ }
+ else
+ {
+ if (pm->flags & PHASH_FLAG_MIX64)
+ init_keys_direct_u64 (pm);
+ else
+ init_keys_direct_u32 (pm);
+ }
+
+ if (!pm->tabb)
+ vec_resize (pm->tabb, 1 << pm->b_bits);
+ else
+ vec_foreach (tb, pm->tabb) phash_tabb_free (tb);
+
+ /* Two keys with the same (a,b) guarantees a collision */
+ no_collisions = 1;
+ vec_foreach (k, pm->keys)
+ {
+ u32 i, *ki;
+
+ tb = pm->tabb + k->b;
+ ki = tb->keys;
+ for (i = 0; i < vec_len (ki); i++)
+ {
+ l = pm->keys + ki[i];
+ if (k->a == l->a)
+ {
+ /* Given keys are supposed to be unique. */
+ if (pm->key_is_equal
+ && pm->key_is_equal (pm->private, l->key, k->key))
+ clib_error ("duplicate keys");
+ no_collisions = 0;
+ goto done;
+ }
+ }
+
+ vec_add1 (tb->keys, k - pm->keys);
+ }
+
+done:
+ return no_collisions;
+}
+
+/* Try to apply an augmenting list */
+static int
+apply (phash_main_t * pm, u32 tail, u32 rollback)
+{
+ phash_key_t *k;
+ phash_tabb_t *pb;
+ phash_tabq_t *q_child, *q_parent;
+ u32 ki, i, hash, child, parent;
+ u32 stabb; /* scramble[tab[b]] */
+ int no_collision;
+
+ no_collision = 1;
+
+ /* Walk from child to parent until root is reached. */
+ for (child = tail - 1; child; child = parent)
+ {
+ q_child = &pm->tabq[child];
+ parent = q_child->parent_q;
+ q_parent = &pm->tabq[parent];
+
+ /* find parent's list of siblings */
+ ASSERT (q_parent->b_q < vec_len (pm->tabb));
+ pb = pm->tabb + q_parent->b_q;
+
+ /* erase old hash values */
+ stabb = pm->scramble[pb->val_b];
+ for (i = 0; i < vec_len (pb->keys); i++)
+ {
+ ki = pb->keys[i];
+ k = pm->keys + ki;
+ hash = k->a ^ stabb;
+
+ /* Erase hash for all of child's siblings. */
+ if (ki == pm->tabh[hash])
+ pm->tabh[hash] = ~0;
+ }
+
+ /* change pb->val_b, which will change the hashes of all parent siblings */
+ pb->val_b = rollback ? q_child->oldval_q : q_child->newval_q;
+
+ /* set new hash values */
+ stabb = pm->scramble[pb->val_b];
+ for (i = 0; i < vec_len (pb->keys); i++)
+ {
+ ki = pb->keys[i];
+ k = pm->keys + ki;
+
+ hash = k->a ^ stabb;
+ if (rollback)
+ {
+ if (parent == 0)
+ continue; /* root never had a hash */
+ }
+ else if (pm->tabh[hash] != ~0)
+ {
+ /* Very rare case: roll back any changes. */
+ apply (pm, tail, /* rollback changes */ 1);
+ no_collision = 0;
+ goto done;
+ }
+ pm->tabh[hash] = ki;
+ }
+ }
+
+done:
+ return no_collision;
+}
+
+
+/*
+-------------------------------------------------------------------------------
+augment(): Add item to the mapping.
+
+Construct a spanning tree of *b*s with *item* as root, where each
+parent can have all its hashes changed (by some new val_b) with
+at most one collision, and each child is the b of that collision.
+
+I got this from Tarjan's "Data Structures and Network Algorithms". The
+path from *item* to a *b* that can be remapped with no collision is
+an "augmenting path". Change values of tab[b] along the path so that
+the unmapped key gets mapped and the unused hash value gets used.
+
+Assuming 1 key per b, if m out of n hash values are still unused,
+you should expect the transitive closure to cover n/m nodes before
+an unused node is found. Sum(i=1..n)(n/i) is about nlogn, so expect
+this approach to take about nlogn time to map all single-key b's.
+-------------------------------------------------------------------------------
+
+high_water: a value higher than any now in tabb[].water_b.
+*/
+static int
+augment (phash_main_t * pm, u32 b_root, u32 high_water)
+{
+ u32 q; /* current position walking through the queue */
+ u32 tail; /* tail of the queue. 0 is the head of the queue. */
+ phash_tabb_t *tb_parent, *tb_child, *tb_hit;
+ phash_key_t *k_parent, *k_child;
+ u32 v, v_limit; /* possible value for myb->val_b */
+ u32 i, ki, hash;
+
+ v_limit =
+ 1 << ((pm->flags & PHASH_FLAG_USE_SCRAMBLE) ? pm->s_bits : BITS (u8));
+
+ /* Initialize the root of the spanning tree. */
+ pm->tabq[0].b_q = b_root;
+ tail = 1;
+
+ /* construct the spanning tree by walking the queue, add children to tail */
+ for (q = 0; q < tail; q++)
+ {
+ if ((pm->flags & PHASH_FLAG_FAST_MODE)
+ && !(pm->flags & PHASH_FLAG_MINIMAL) && q == 1)
+ break; /* don't do transitive closure */
+
+ tb_parent = pm->tabb + pm->tabq[q].b_q; /* the b for this node */
+
+ for (v = 0; v < v_limit; v++)
+ {
+ tb_child = 0;
+
+ for (i = 0; i < vec_len (tb_parent->keys); i++)
+ {
+ ki = tb_parent->keys[i];
+ k_parent = pm->keys + ki;
+
+ hash = k_parent->a ^ pm->scramble[v];
+ if (hash >= pm->hash_max)
+ goto try_next_v; /* hash code out of bounds => we can't use this v */
+
+ ki = pm->tabh[hash];
+ if (ki == ~0)
+ continue;
+
+ k_child = pm->keys + ki;
+ tb_hit = pm->tabb + k_child->b;
+
+ if (tb_child)
+ {
+ /* Hit at most one child b. */
+ if (tb_child == tb_hit)
+ goto try_next_v;
+ }
+ else
+ {
+ /* Remember this as child b. */
+ tb_child = tb_hit;
+ if (tb_hit->water_b == high_water)
+ goto try_next_v; /* already explored */
+ }
+ }
+
+ /* tb_parent with v has either one or zero collisions. */
+
+ /* add child b to the queue of reachable things */
+ if (tb_child)
+ tb_child->water_b = high_water;
+ pm->tabq[tail].b_q = tb_child ? tb_child - pm->tabb : ~0;
+ pm->tabq[tail].newval_q = v; /* how to make parent (myb) use this hash */
+ pm->tabq[tail].oldval_q = tb_parent->val_b; /* need this for rollback */
+ pm->tabq[tail].parent_q = q;
+ ++tail;
+
+ /* Found a v with no collisions? */
+ if (!tb_child)
+ {
+ /* Try to apply the augmenting path. */
+ if (apply (pm, tail, /* rollback */ 0))
+ return 1; /* success, item was added to the perfect hash */
+ --tail; /* don't know how to handle such a child! */
+ }
+
+ try_next_v:
+ ;
+ }
+ }
+ return 0;
+}
+
+
+static phash_tabb_t *sort_tabb;
+
+static int
+phash_tabb_compare (void *a1, void *a2)
+{
+ u32 *b1 = a1;
+ u32 *b2 = a2;
+ phash_tabb_t *tb1, *tb2;
+
+ tb1 = sort_tabb + b1[0];
+ tb2 = sort_tabb + b2[0];
+
+ return ((int) vec_len (tb2->keys) - (int) vec_len (tb1->keys));
+}
+
+/* find a mapping that makes this a perfect hash */
+static int
+perfect (phash_main_t * pm)
+{
+ u32 i;
+
+ /* clear any state from previous attempts */
+ if (vec_bytes (pm->tabh))
+ clib_memset (pm->tabh, ~0, vec_bytes (pm->tabh));
+
+ vec_validate (pm->tabb_sort, vec_len (pm->tabb) - 1);
+ for (i = 0; i < vec_len (pm->tabb_sort); i++)
+ pm->tabb_sort[i] = i;
+
+ sort_tabb = pm->tabb;
+
+ vec_sort_with_function (pm->tabb_sort, phash_tabb_compare);
+
+ /* In descending order by number of keys, map all *b*s */
+ for (i = 0; i < vec_len (pm->tabb_sort); i++)
+ {
+ if (!augment (pm, pm->tabb_sort[i], i + 1))
+ return 0;
+ }
+
+ /* Success! We found a perfect hash of all keys into 0..nkeys-1. */
+ return 1;
+}
+
+
+/*
+ * Find initial a_bits = log2 (a_max), b_bits = log2 (b_max).
+ * Initial a_max and b_max values were found empirically. Some factors:
+ *
+ * If s_max<256 there is no scramble, so tab[b] needs to cover 0..s_max-1.
+ *
+ * a_max and b_max must be powers of 2 because the values in 0..a_max-1 and
+ * 0..b_max-1 are produced by applying a bitmask to the initial hash function.
+ *
+ * a_max must be less than s_max, in fact less than n_keys, because otherwise
+ * there would often be no i such that a^scramble[i] is in 0..n_keys-1 for
+ * all the *a*s associated with a given *b*, so there would be no legal
+ * value to assign to tab[b]. This only matters when we're doing a minimal
+ * perfect hash.
+ *
+ * It takes around 800 trials to find distinct (a,b) with nkey=s_max*(5/8)
+ * and a_max*b_max = s_max*s_max/32.
+ *
+ * Values of b_max less than s_max/4 never work, and s_max/2 always works.
+ *
+ * We want b_max as small as possible because it is the number of bytes in
+ * the huge array we must create for the perfect hash.
+ *
+ * When nkey <= s_max*(5/8), b_max=s_max/4 works much more often with
+ * a_max=s_max/8 than with a_max=s_max/4. Above s_max*(5/8), b_max=s_max/4
+ * doesn't seem to care whether a_max=s_max/8 or a_max=s_max/4. I think it
+ * has something to do with 5/8 = 1/8 * 5. For example examine 80000,
+ * 85000, and 90000 keys with different values of a_max. This only matters
+ * if we're doing a minimal perfect hash.
+ *
+ * When a_max*b_max <= 1<<U32BITS, the initial hash must produce one integer.
+ * Bigger than that it must produce two integers, which increases the
+ * cost of the hash per character hashed.
+ */
+static void
+guess_initial_parameters (phash_main_t * pm)
+{
+ u32 s_bits, s_max, a_max, b_max, n_keys;
+ int is_minimal, is_fast_mode;
+ const u32 b_max_use_scramble_threshold = 4096;
+
+ is_minimal = (pm->flags & PHASH_FLAG_MINIMAL) != 0;
+ is_fast_mode = (pm->flags & PHASH_FLAG_FAST_MODE) != 0;
+
+ n_keys = vec_len (pm->keys);
+ s_bits = max_log2 (n_keys);
+ s_max = 1 << s_bits;
+ a_max = 0;
+
+ if (is_minimal)
+ {
+ switch (s_bits)
+ {
+ case 0:
+ a_max = 1;
+ b_max = 1;
+ case 1:
+ case 2:
+ case 3:
+ case 4:
+ case 5:
+ case 6:
+ case 7:
+ case 8:
+ /*
+ * Was: a_max = is_minimal ? s_max / 2 : s_max;
+ * However, we know that is_minimal must be true, so the
+ * if-arm of the ternary expression is always executed.
+ */
+ a_max = s_max / 2;
+ b_max = s_max / 2;
+ break;
+ case 9:
+ case 10:
+ case 11:
+ case 12:
+ case 13:
+ case 14:
+ case 15:
+ case 16:
+ case 17:
+ if (is_fast_mode)
+ {
+ a_max = s_max / 2;
+ b_max = s_max / 4;
+ }
+ else if (s_max / 4 < b_max_use_scramble_threshold)
+ {
+ if (n_keys <= s_max * 0.52)
+ a_max = b_max = s_max / 8;
+ else
+ a_max = b_max = s_max / 4;
+ }
+ else
+ {
+ a_max = ((n_keys <= s_max * (5.0 / 8.0)) ? s_max / 8 :
+ (n_keys <=
+ s_max * (3.0 / 4.0)) ? s_max / 4 : s_max / 2);
+ b_max = s_max / 4; /* always give the small size a shot */
+ }
+ break;
+ case 18:
+ if (is_fast_mode)
+ a_max = b_max = s_max / 2;
+ else
+ {
+ a_max = s_max / 8; /* never require the multiword hash */
+ b_max = (n_keys <= s_max * (5.0 / 8.0)) ? s_max / 4 : s_max / 2;
+ }
+ break;
+ case 19:
+ case 20:
+ a_max = (n_keys <= s_max * (5.0 / 8.0)) ? s_max / 8 : s_max / 2;
+ b_max = (n_keys <= s_max * (5.0 / 8.0)) ? s_max / 4 : s_max / 2;
+ break;
+ default:
+ /* Just find a hash as quick as possible.
+ We'll be thrashing virtual memory at this size. */
+ a_max = b_max = s_max / 2;
+ break;
+ }
+ }
+ else
+ {
+ /* Non-minimal perfect hash. */
+ if (is_fast_mode && n_keys > s_max * 0.8)
+ {
+ s_max *= 2;
+ s_bits += 1;
+ }
+
+ if (s_max / 4 <= (1 << 14))
+ b_max = ((n_keys <= s_max * 0.56) ? s_max / 32 :
+ (n_keys <= s_max * 0.74) ? s_max / 16 : s_max / 8);
+ else
+ b_max = ((n_keys <= s_max * 0.6) ? s_max / 16 :
+ (n_keys <= s_max * 0.8) ? s_max / 8 : s_max / 4);
+
+ if (is_fast_mode && b_max < s_max / 8)
+ b_max = s_max / 8;
+
+ if (a_max < 1)
+ a_max = 1;
+ if (b_max < 1)
+ b_max = 1;
+ }
+
+ ASSERT (s_max == (1 << s_bits));
+ ASSERT (is_pow2 (a_max));
+ ASSERT (is_pow2 (b_max));
+ pm->s_bits = s_bits;
+ pm->a_bits = min_log2 (a_max);
+ pm->b_bits = min_log2 (b_max);
+ if (b_max >= b_max_use_scramble_threshold)
+ pm->flags |= PHASH_FLAG_USE_SCRAMBLE;
+}
+
+/* compute p(x), where p is a permutation of 0..(1<<nbits)-1 */
+/* permute(0)=0. This is intended and useful. */
+always_inline u32
+scramble_permute (u32 x, u32 nbits)
+{
+ int i;
+ int mask = (1 << nbits) - 1;
+ int const2 = 1 + nbits / 2;
+ int const3 = 1 + nbits / 3;
+ int const4 = 1 + nbits / 4;
+ int const5 = 1 + nbits / 5;
+ for (i = 0; i < 20; i++)
+ {
+ x = (x + (x << const2)) & mask;
+ x = (x ^ (x >> const3));
+ x = (x + (x << const4)) & mask;
+ x = (x ^ (x >> const5));
+ }
+ return x;
+}
+
+/* initialize scramble[] with distinct random values in 0..smax-1 */
+static void
+scramble_init (phash_main_t * pm)
+{
+ u32 i;
+
+ /* fill scramble[] with distinct random integers in 0..smax-1 */
+ vec_validate (pm->scramble, (1 << (pm->s_bits < 8 ? 8 : pm->s_bits)) - 1);
+ for (i = 0; i < vec_len (pm->scramble); i++)
+ pm->scramble[i] = scramble_permute (i, pm->s_bits);
+}
+
+/* Try to find a perfect hash function. */
+clib_error_t *
+phash_find_perfect_hash (phash_main_t * pm)
+{
+ clib_error_t *error = 0;
+ u32 max_a_bits, n_tries_this_a_b, want_minimal;
+
+ /* guess initial values for s_max, a_max and b_max */
+ guess_initial_parameters (pm);
+
+ want_minimal = pm->flags & PHASH_FLAG_MINIMAL;
+
+new_s:
+ if (pm->b_bits == 0)
+ pm->a_bits = pm->s_bits;
+
+ max_a_bits = pm->s_bits - want_minimal;
+ if (max_a_bits < 1)
+ max_a_bits = 1;
+
+ pm->hash_max = want_minimal ? vec_len (pm->keys) : (1 << pm->s_bits);
+
+ scramble_init (pm);
+
+ /* Allocate working memory. */
+ vec_free (pm->tabh);
+ vec_validate_init_empty (pm->tabh, pm->hash_max - 1, ~0);
+ vec_free (pm->tabq);
+ vec_validate (pm->tabq, 1 << pm->b_bits);
+
+ /* Actually find the perfect hash */
+ n_tries_this_a_b = 0;
+ while (1)
+ {
+ /* Choose random hash seeds until keys become unique. */
+ pm->hash_seed = random_u64 (&pm->random_seed);
+ pm->n_seed_trials++;
+ if (init_tabb (pm))
+ {
+ /* Found unique (A, B). */
+
+ /* Hash may already be perfect. */
+ if (pm->b_bits == 0)
+ goto done;
+
+ pm->n_perfect_calls++;
+ if (perfect (pm))
+ goto done;
+
+ goto increase_b;
+ }
+
+ /* Keep trying with different seed value. */
+ n_tries_this_a_b++;
+ if (n_tries_this_a_b < 2048)
+ continue;
+
+ /* Try to put more bits in (A,B) to make distinct (A,B) more likely */
+ if (pm->a_bits < max_a_bits)
+ pm->a_bits++;
+ else if (pm->b_bits < pm->s_bits)
+ {
+ increase_b:
+ vec_resize (pm->tabb, vec_len (pm->tabb));
+ vec_resize (pm->tabq, vec_len (pm->tabq));
+ pm->b_bits++;
+ }
+ else
+ {
+ /* Can't increase (A, B) any more, so try increasing S. */
+ goto new_s;
+ }
+ }
+
+done:
+ /* Construct mapping table for hash lookups. */
+ if (!error)
+ {
+ u32 b, v;
+
+ pm->a_shift = ((pm->flags & PHASH_FLAG_MIX64) ? 64 : 32) - pm->a_bits;
+ pm->b_mask = (1 << pm->b_bits) - 1;
+
+ vec_resize (pm->tab, vec_len (pm->tabb));
+ for (b = 0; b < vec_len (pm->tabb); b++)
+ {
+ v = pm->tabb[b].val_b;
+
+ /* Apply scramble now for small enough value of b_bits. */
+ if (!(pm->flags & PHASH_FLAG_USE_SCRAMBLE))
+ v = pm->scramble[v];
+
+ pm->tab[b] = v;
+ }
+ }
+
+ /* Free working memory. */
+ phash_main_free_working_memory (pm);
+
+ return error;
+}
+
+/* Slow hash computation for general keys. */
+uword
+phash_hash_slow (phash_main_t * pm, uword key)
+{
+ u32 a, b, v;
+
+ if (pm->flags & PHASH_FLAG_MIX64)
+ {
+ u64 x0, y0, z0;
+
+ x0 = y0 = z0 = pm->hash_seed;
+
+ if (pm->key_seed1)
+ {
+ u64 xyz[3];
+ pm->key_seed1 (pm->private, key, &xyz);
+ x0 += xyz[0];
+ y0 += xyz[1];
+ z0 += xyz[2];
+ }
+ else
+ x0 += key;
+
+ hash_mix64 (x0, y0, z0);
+
+ a = z0 >> pm->a_shift;
+ b = z0 & pm->b_mask;
+ }
+ else
+ {
+ u32 x0, y0, z0;
+
+ x0 = y0 = z0 = pm->hash_seed;
+
+ if (pm->key_seed1)
+ {
+ u32 xyz[3];
+ pm->key_seed1 (pm->private, key, &xyz);
+ x0 += xyz[0];
+ y0 += xyz[1];
+ z0 += xyz[2];
+ }
+ else
+ x0 += key;
+
+ hash_mix32 (x0, y0, z0);
+
+ a = z0 >> pm->a_shift;
+ b = z0 & pm->b_mask;
+ }
+
+ v = pm->tab[b];
+ if (pm->flags & PHASH_FLAG_USE_SCRAMBLE)
+ v = pm->scramble[v];
+ return a ^ v;
+}
+
+/* Verify that perfect hash is perfect. */
+clib_error_t *
+phash_validate (phash_main_t * pm)
+{
+ phash_key_t *k;
+ uword *unique_bitmap = 0;
+ clib_error_t *error = 0;
+
+ vec_foreach (k, pm->keys)
+ {
+ uword h = phash_hash_slow (pm, k->key);
+
+ if (h >= pm->hash_max)
+ {
+ error = clib_error_return (0, "hash out of range %wd", h);
+ goto done;
+ }
+
+ if (clib_bitmap_get (unique_bitmap, h))
+ {
+ error = clib_error_return (0, "hash non-unique");
+ goto done;
+ }
+
+ unique_bitmap = clib_bitmap_ori (unique_bitmap, h);
+ }
+
+done:
+ clib_bitmap_free (unique_bitmap);
+ return error;
+}
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/phash.h b/extras/deprecated/vppinfra/phash.h
new file mode 100644
index 00000000000..3dc59c724f7
--- /dev/null
+++ b/extras/deprecated/vppinfra/phash.h
@@ -0,0 +1,194 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ Copyright (c) 2005 Eliot Dresselhaus
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice shall be
+ included in all copies or substantial portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+*/
+
+#ifndef included_phash_h
+#define included_phash_h
+
+#include <vppinfra/hash.h> /* for Bob's mixing functions */
+
+typedef struct
+{
+ /* Maybe either pointer to vector or inline word. */
+ uword key;
+
+ /* Hash code (A, B). */
+ u32 a, b;
+} phash_key_t;
+
+/* Table indexed by B. */
+typedef struct
+{
+ /* Vector of key indices with this same value of B. */
+ u32 *keys;
+
+ /* hash=a^tabb[b].val_b */
+ u32 val_b;
+
+ /* High watermark of who has visited this map node. */
+ u32 water_b;
+} phash_tabb_t;
+
+always_inline void
+phash_tabb_free (phash_tabb_t * b)
+{
+ vec_free (b->keys);
+ b->val_b = b->water_b = 0;
+}
+
+typedef struct
+{
+ /* b that currently occupies this hash */
+ u32 b_q;
+
+ /* Queue position of parent that could use this hash. */
+ u32 parent_q;
+
+ /* What to change parent tab[b] to use this hash. */
+ u32 newval_q;
+
+ /* Original value of tab[b]. */
+ u32 oldval_q;
+} phash_tabq_t;
+
+typedef struct
+{
+ u8 a_bits, b_bits, s_bits, a_shift;
+ u32 b_mask;
+ u32 *tab;
+ u32 *scramble;
+
+ /* Seed value for hash mixer. */
+ u64 hash_seed;
+
+ u32 flags;
+
+ /* Key functions want 64 bit keys.
+ Use hash_mix64 rather than hash_mix32. */
+#define PHASH_FLAG_MIX64 (1 << 0)
+#define PHASH_FLAG_MIX32 (0 << 0)
+
+ /* When b_bits is large enough (>= 12) we scramble. */
+#define PHASH_FLAG_USE_SCRAMBLE (1 << 1)
+
+ /* Slow mode gives smaller tables but at the expense of more run time. */
+#define PHASH_FLAG_SLOW_MODE (0 << 2)
+#define PHASH_FLAG_FAST_MODE (1 << 2)
+
+ /* Generate minimal perfect hash instead of perfect hash. */
+#define PHASH_FLAG_NON_MINIMAL (0 << 3)
+#define PHASH_FLAG_MINIMAL (1 << 3)
+
+ /* vec_len (keys) for minimal hash;
+ 1 << s_bits for non-minimal hash. */
+ u32 hash_max;
+
+ /* Vector of keys. */
+ phash_key_t *keys;
+
+ /* Used by callbacks to identify keys. */
+ void *private;
+
+ /* Key comparison callback. */
+ int (*key_is_equal) (void *private, uword key1, uword key2);
+
+ /* Callback to reduce single key -> hash seeds. */
+ void (*key_seed1) (void *private, uword key, void *seed);
+
+ /* Callback to reduce two key2 -> hash seeds. */
+ void (*key_seed2) (void *private, uword key1, uword key2, void *seed);
+
+ /* Stuff used to compute perfect hash. */
+ u32 random_seed;
+
+ /* Stuff indexed by B. */
+ phash_tabb_t *tabb;
+
+ /* Table of B ordered by number of keys in tabb[b]. */
+ u32 *tabb_sort;
+
+ /* Unique key (or ~0 if none) for a given hash
+ H = A ^ scramble[tab[B].val_b]. */
+ u32 *tabh;
+
+ /* Stuff indexed by q. */
+ phash_tabq_t *tabq;
+
+ /* Stats. */
+ u32 n_seed_trials, n_perfect_calls;
+} phash_main_t;
+
+always_inline void
+phash_main_free_working_memory (phash_main_t * pm)
+{
+ vec_free (pm->tabb);
+ vec_free (pm->tabq);
+ vec_free (pm->tabh);
+ vec_free (pm->tabb_sort);
+ if (!(pm->flags & PHASH_FLAG_USE_SCRAMBLE))
+ vec_free (pm->scramble);
+}
+
+always_inline void
+phash_main_free (phash_main_t * pm)
+{
+ phash_main_free_working_memory (pm);
+ vec_free (pm->tab);
+ vec_free (pm->keys);
+ clib_memset (pm, 0, sizeof (pm[0]));
+}
+
+/* Slow hash computation for general keys. */
+uword phash_hash_slow (phash_main_t * pm, uword key);
+
+/* Main routine to compute perfect hash. */
+clib_error_t *phash_find_perfect_hash (phash_main_t * pm);
+
+/* Validates that hash is indeed perfect. */
+clib_error_t *phash_validate (phash_main_t * pm);
+
+/* Unit test. */
+int phash_test_main (unformat_input_t * input);
+
+#endif /* included_phash_h */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/pipeline.h b/extras/deprecated/vppinfra/pipeline.h
new file mode 100644
index 00000000000..5a9799b455e
--- /dev/null
+++ b/extras/deprecated/vppinfra/pipeline.h
@@ -0,0 +1,176 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ * pipeline.h: software pipeline infrastructure
+ *
+ * Copyright (c) 2010 Eliot Dresselhaus
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be
+ * included in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#ifndef included_clib_pipeline_h
+#define included_clib_pipeline_h
+
+#define clib_pipeline_stage(F,TYPE,ARG,I,BODY) \
+ always_inline void F##_inline (void * _, u32 I) \
+ { TYPE ARG = _; { BODY; } } \
+ never_inline void F##_no_inline (TYPE ARG, u32 I) \
+ { F##_inline (ARG, I); }
+
+#define clib_pipeline_stage_static(F,TYPE,ARG,I,BODY) \
+ static_always_inline void F##_inline (void * _, u32 I) \
+ { TYPE ARG = _; { BODY; } } \
+ never_inline void F##_no_inline (TYPE ARG, u32 I) \
+ { F##_inline (ARG, I); }
+
+#define clib_pipeline_stage_no_inline(F,TYPE,ARG,I,BODY) \
+ never_inline void F##_no_inline (void * _, u32 I) \
+ { TYPE ARG = _; { BODY; } } \
+ never_inline void F##_inline (TYPE ARG, u32 I) \
+ { F##_no_inline (ARG, I); }
+
+#define _clib_pipeline_var(v) _clib_pipeline_##v
+
+#define clib_pipeline_stage_execute(F,A,I,S) \
+ F##_##S (A, _clib_pipeline_var(i) - (I))
+
+#define clib_pipeline_main_stage(F,A,I) \
+ clib_pipeline_stage_execute (F, A, I, inline)
+#define clib_pipeline_init_stage(F,A,I) \
+ if (_clib_pipeline_var(i) >= (I)) clib_pipeline_stage_execute (F, A, I, no_inline)
+#define clib_pipeline_exit_stage(F,A,I) \
+ if (_clib_pipeline_var(i) >= (I) && _clib_pipeline_var(i) - (I) < _clib_pipeline_var(n_vectors)) \
+ clib_pipeline_stage_execute (F, A, I, no_inline)
+
+#define clib_pipeline_init_loop \
+ for (_clib_pipeline_var(i) = 0; \
+ _clib_pipeline_var(i) < \
+ clib_min (_clib_pipeline_var(n_stages) - 1, \
+ _clib_pipeline_var(n_vectors)); \
+ _clib_pipeline_var(i)++)
+
+#define clib_pipeline_main_loop \
+ for (; _clib_pipeline_var(i) < _clib_pipeline_var(n_vectors); \
+ _clib_pipeline_var(i)++)
+
+#define clib_pipeline_exit_loop \
+ for (; _clib_pipeline_var(i) < (_clib_pipeline_var(n_vectors) \
+ + _clib_pipeline_var(n_stages) - 1); \
+ _clib_pipeline_var(i)++)
+
+#define clib_pipeline_run_2_stage(N,ARG,STAGE0,STAGE1) \
+do { \
+ uword _clib_pipeline_var(n_vectors) = (N); \
+ uword _clib_pipeline_var(n_stages) = 2; \
+ uword _clib_pipeline_var(i); \
+ \
+ clib_pipeline_init_loop \
+ { \
+ clib_pipeline_init_stage (STAGE0, ARG, 0); \
+ } \
+ \
+ clib_pipeline_main_loop \
+ { \
+ clib_pipeline_main_stage (STAGE0, ARG, 0); \
+ clib_pipeline_main_stage (STAGE1, ARG, 1); \
+ } \
+ \
+ clib_pipeline_exit_loop \
+ { \
+ clib_pipeline_exit_stage (STAGE1, ARG, 1); \
+ } \
+} while (0)
+
+#define clib_pipeline_run_3_stage(N,ARG,STAGE0,STAGE1,STAGE2) \
+do { \
+ uword _clib_pipeline_var(n_vectors) = (N); \
+ uword _clib_pipeline_var(n_stages) = 3; \
+ uword _clib_pipeline_var(i); \
+ \
+ clib_pipeline_init_loop \
+ { \
+ clib_pipeline_init_stage (STAGE0, ARG, 0); \
+ clib_pipeline_init_stage (STAGE1, ARG, 1); \
+ } \
+ \
+ clib_pipeline_main_loop \
+ { \
+ clib_pipeline_main_stage (STAGE0, ARG, 0); \
+ clib_pipeline_main_stage (STAGE1, ARG, 1); \
+ clib_pipeline_main_stage (STAGE2, ARG, 2); \
+ } \
+ \
+ clib_pipeline_exit_loop \
+ { \
+ clib_pipeline_exit_stage (STAGE1, ARG, 1); \
+ clib_pipeline_exit_stage (STAGE2, ARG, 2); \
+ } \
+} while (0)
+
+#define clib_pipeline_run_4_stage(N,ARG,STAGE0,STAGE1,STAGE2,STAGE3) \
+do { \
+ uword _clib_pipeline_var(n_vectors) = (N); \
+ uword _clib_pipeline_var(n_stages) = 4; \
+ uword _clib_pipeline_var(i); \
+ \
+ clib_pipeline_init_loop \
+ { \
+ clib_pipeline_init_stage (STAGE0, ARG, 0); \
+ clib_pipeline_init_stage (STAGE1, ARG, 1); \
+ clib_pipeline_init_stage (STAGE2, ARG, 2); \
+ } \
+ \
+ clib_pipeline_main_loop \
+ { \
+ clib_pipeline_main_stage (STAGE0, ARG, 0); \
+ clib_pipeline_main_stage (STAGE1, ARG, 1); \
+ clib_pipeline_main_stage (STAGE2, ARG, 2); \
+ clib_pipeline_main_stage (STAGE3, ARG, 3); \
+ } \
+ \
+ clib_pipeline_exit_loop \
+ { \
+ clib_pipeline_exit_stage (STAGE1, ARG, 1); \
+ clib_pipeline_exit_stage (STAGE2, ARG, 2); \
+ clib_pipeline_exit_stage (STAGE3, ARG, 3); \
+ } \
+} while (0)
+
+#endif /* included_clib_pipeline_h */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/qhash.c b/extras/deprecated/vppinfra/qhash.c
new file mode 100644
index 00000000000..3b5a175065d
--- /dev/null
+++ b/extras/deprecated/vppinfra/qhash.c
@@ -0,0 +1,859 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ Copyright (c) 2006 Eliot Dresselhaus
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice shall be
+ included in all copies or substantial portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+*/
+
+#include <vppinfra/qhash.h>
+
+#define QHASH_ALL_VALID ((1 << QHASH_KEYS_PER_BUCKET) - 1)
+
+void *
+_qhash_resize (void *v, uword length, uword elt_bytes)
+{
+ qhash_t *h;
+ uword l;
+
+ l = clib_max (max_log2 (length), 2 + QHASH_LOG2_KEYS_PER_BUCKET);
+
+ /* Round up if less than 1/2 full. */
+ l += ((f64) length / (f64) (1 << l)) < .5;
+
+ v = _vec_resize (0, 1 << l, elt_bytes << l, sizeof (h[0]),
+ /* align */ sizeof (uword));
+
+ h = qhash_header (v);
+ h->n_elts = 0;
+ h->log2_hash_size = l;
+ h->hash_keys =
+ clib_mem_alloc_aligned_no_fail (sizeof (h->hash_keys[0]) << l,
+ CLIB_CACHE_LINE_BYTES);
+ vec_resize (h->hash_key_valid_bitmap,
+ 1 << (l - QHASH_LOG2_KEYS_PER_BUCKET));
+ clib_memset (v, ~0, elt_bytes << l);
+
+ return v;
+}
+
+static u8 min_log2_table[256];
+
+static inline uword
+qhash_min_log2 (uword x)
+{
+ ASSERT (is_pow2 (x));
+ ASSERT (x < 256);
+ return min_log2_table[x];
+}
+
+static void
+qhash_min_log2_init ()
+{
+ int i;
+ for (i = 0; i < 256; i++)
+ min_log2_table[i] = min_log2 (i);
+}
+
+always_inline uword
+qhash_get_valid_elt_mask (qhash_t * h, uword i)
+{
+ return h->hash_key_valid_bitmap[i / QHASH_KEYS_PER_BUCKET];
+}
+
+always_inline void
+qhash_set_valid_elt_mask (qhash_t * h, uword i, uword mask)
+{
+ h->hash_key_valid_bitmap[i / QHASH_KEYS_PER_BUCKET] = mask;
+}
+
+always_inline uword
+qhash_search_bucket (uword * hash_keys, uword search_key, uword m)
+{
+ uword t;
+#define _(i) ((hash_keys[i] == search_key) << i)
+ t = (_(0) | _(1) | _(2) | _(3));
+ if (QHASH_KEYS_PER_BUCKET > 4)
+ t |= (_(4) | _(5) | _(6) | _(7));
+ if (QHASH_KEYS_PER_BUCKET > 8)
+ t |= (_(8) | _(9) | _(10) | _(11) | _(12) | _(13) | _(14) | _(15));
+#undef _
+ return m & t;
+}
+
+/* Lookup multiple keys in the same hash table. */
+void
+qhash_get_multiple (void *v,
+ uword * search_keys,
+ uword n_search_keys, u32 * result_indices)
+{
+ qhash_t *h = qhash_header (v);
+ uword *k, *hash_keys;
+ uword n_left, bucket_mask;
+ u32 *r;
+
+ if (!v)
+ {
+ clib_memset (result_indices, ~0,
+ sizeof (result_indices[0]) * n_search_keys);
+ return;
+ }
+
+ bucket_mask = pow2_mask (h->log2_hash_size) & ~(QHASH_KEYS_PER_BUCKET - 1);
+
+ k = search_keys;
+ n_left = n_search_keys;
+ hash_keys = h->hash_keys;
+ r = result_indices;
+
+ while (n_left >= 2)
+ {
+ u32 a0, b0, c0, bi0, valid0, match0;
+ u32 a1, b1, c1, bi1, valid1, match1;
+ uword k0, k1, *h0, *h1;
+
+ k0 = k[0];
+ k1 = k[1];
+ n_left -= 2;
+ k += 2;
+
+ a0 = a1 = h->hash_seeds[0];
+ b0 = b1 = h->hash_seeds[1];
+ c0 = c1 = h->hash_seeds[2];
+ a0 ^= k0;
+ a1 ^= k1;
+#if uword_bits == 64
+ b0 ^= k0 >> 32;
+ b1 ^= k1 >> 32;
+#endif
+
+ hash_mix32_step_1 (a0, b0, c0);
+ hash_mix32_step_1 (a1, b1, c1);
+ hash_mix32_step_2 (a0, b0, c0);
+ hash_mix32_step_2 (a1, b1, c1);
+ hash_mix32_step_3 (a0, b0, c0);
+ hash_mix32_step_3 (a1, b1, c1);
+
+ bi0 = c0 & bucket_mask;
+ bi1 = c1 & bucket_mask;
+
+ h0 = hash_keys + bi0;
+ h1 = hash_keys + bi1;
+
+ /* Search two buckets. */
+ valid0 = qhash_get_valid_elt_mask (h, bi0);
+ valid1 = qhash_get_valid_elt_mask (h, bi1);
+
+ match0 = qhash_search_bucket (h0, k0, valid0);
+ match1 = qhash_search_bucket (h1, k1, valid1);
+
+ bi0 += qhash_min_log2 (match0);
+ bi1 += qhash_min_log2 (match1);
+
+ r[0] = match0 ? bi0 : ~0;
+ r[1] = match1 ? bi1 : ~0;
+ r += 2;
+
+ /* Full buckets trigger search of overflow hash. */
+ if (PREDICT_FALSE (!match0 && valid0 == QHASH_ALL_VALID))
+ {
+ uword *p = hash_get (h->overflow_hash, k0);
+ r[-2] = p ? p[0] : ~0;
+ }
+
+ /* Full buckets trigger search of overflow hash. */
+ if (PREDICT_FALSE (!match1 && valid1 == QHASH_ALL_VALID))
+ {
+ uword *p = hash_get (h->overflow_hash, k1);
+ r[-1] = p ? p[0] : ~0;
+ }
+ }
+
+ while (n_left >= 1)
+ {
+ u32 a0, b0, c0, bi0, valid0, match0;
+ uword k0, *h0;
+
+ k0 = k[0];
+ n_left -= 1;
+ k += 1;
+
+ a0 = h->hash_seeds[0];
+ b0 = h->hash_seeds[1];
+ c0 = h->hash_seeds[2];
+ a0 ^= k0;
+#if uword_bits == 64
+ b0 ^= k0 >> 32;
+#endif
+
+ hash_mix32 (a0, b0, c0);
+
+ bi0 = c0 & bucket_mask;
+
+ h0 = hash_keys + bi0;
+
+ /* Search one bucket. */
+ valid0 = qhash_get_valid_elt_mask (h, bi0);
+ match0 = qhash_search_bucket (h0, k0, valid0);
+
+ bi0 += qhash_min_log2 (match0);
+
+ r[0] = match0 ? bi0 : ~0;
+ r += 1;
+
+ /* Full buckets trigger search of overflow hash. */
+ if (PREDICT_FALSE (!match0 && valid0 == QHASH_ALL_VALID))
+ {
+ uword *p = hash_get (h->overflow_hash, k0);
+ r[-1] = p ? p[0] : ~0;
+ }
+ }
+}
+
+/* Lookup multiple keys in the same hash table.
+ Returns index of first matching key. */
+u32
+qhash_get_first_match (void *v,
+ uword * search_keys,
+ uword n_search_keys, uword * matching_key)
+{
+ qhash_t *h = qhash_header (v);
+ uword *k, *hash_keys;
+ uword n_left, match_mask, bucket_mask;
+
+ if (!v)
+ return ~0;
+
+ match_mask = 0;
+ bucket_mask = pow2_mask (h->log2_hash_size) & ~(QHASH_KEYS_PER_BUCKET - 1);
+
+ k = search_keys;
+ n_left = n_search_keys;
+ hash_keys = h->hash_keys;
+ while (n_left >= 2)
+ {
+ u32 a0, b0, c0, bi0, valid0;
+ u32 a1, b1, c1, bi1, valid1;
+ uword k0, k1, *h0, *h1;
+
+ k0 = k[0];
+ k1 = k[1];
+ n_left -= 2;
+ k += 2;
+
+ a0 = a1 = h->hash_seeds[0];
+ b0 = b1 = h->hash_seeds[1];
+ c0 = c1 = h->hash_seeds[2];
+ a0 ^= k0;
+ a1 ^= k1;
+#if uword_bits == 64
+ b0 ^= k0 >> 32;
+ b1 ^= k1 >> 32;
+#endif
+
+ hash_mix32_step_1 (a0, b0, c0);
+ hash_mix32_step_1 (a1, b1, c1);
+ hash_mix32_step_2 (a0, b0, c0);
+ hash_mix32_step_2 (a1, b1, c1);
+ hash_mix32_step_3 (a0, b0, c0);
+ hash_mix32_step_3 (a1, b1, c1);
+
+ bi0 = c0 & bucket_mask;
+ bi1 = c1 & bucket_mask;
+
+ h0 = hash_keys + bi0;
+ h1 = hash_keys + bi1;
+
+ /* Search two buckets. */
+ valid0 = qhash_get_valid_elt_mask (h, bi0);
+ valid1 = qhash_get_valid_elt_mask (h, bi1);
+ match_mask = qhash_search_bucket (h0, k0, valid0);
+ match_mask |= (qhash_search_bucket (h1, k1, valid1)
+ << QHASH_KEYS_PER_BUCKET);
+ if (match_mask)
+ {
+ uword bi, is_match1;
+
+ bi = qhash_min_log2 (match_mask);
+ is_match1 = bi >= QHASH_KEYS_PER_BUCKET;
+
+ bi += ((is_match1 ? bi1 : bi0)
+ - (is_match1 << QHASH_LOG2_KEYS_PER_BUCKET));
+ *matching_key = (k - 2 - search_keys) + is_match1;
+ return bi;
+ }
+
+ /* Full buckets trigger search of overflow hash. */
+ if (PREDICT_FALSE (valid0 == QHASH_ALL_VALID
+ || valid1 == QHASH_ALL_VALID))
+ {
+ uword *p = 0;
+ uword ki = k - 2 - search_keys;
+
+ if (valid0 == QHASH_ALL_VALID)
+ p = hash_get (h->overflow_hash, k0);
+
+ if (!p && valid1 == QHASH_ALL_VALID)
+ {
+ p = hash_get (h->overflow_hash, k1);
+ ki++;
+ }
+
+ if (p)
+ {
+ *matching_key = ki;
+ return p[0];
+ }
+ }
+ }
+
+ while (n_left >= 1)
+ {
+ u32 a0, b0, c0, bi0, valid0;
+ uword k0, *h0;
+
+ k0 = k[0];
+ n_left -= 1;
+ k += 1;
+
+ a0 = h->hash_seeds[0];
+ b0 = h->hash_seeds[1];
+ c0 = h->hash_seeds[2];
+ a0 ^= k0;
+#if uword_bits == 64
+ b0 ^= k0 >> 32;
+#endif
+
+ hash_mix32 (a0, b0, c0);
+
+ bi0 = c0 & bucket_mask;
+
+ h0 = hash_keys + bi0;
+
+ /* Search one bucket. */
+ valid0 = qhash_get_valid_elt_mask (h, bi0);
+ match_mask = qhash_search_bucket (h0, k0, valid0);
+ if (match_mask)
+ {
+ uword bi;
+ bi = bi0 + qhash_min_log2 (match_mask);
+ *matching_key = (k - 1 - search_keys);
+ return bi;
+ }
+
+ /* Full buckets trigger search of overflow hash. */
+ if (PREDICT_FALSE (valid0 == QHASH_ALL_VALID))
+ {
+ uword *p = hash_get (h->overflow_hash, k0);
+ if (p)
+ {
+ *matching_key = (k - 1 - search_keys);
+ return p[0];
+ }
+ }
+ }
+
+ return ~0;
+}
+
+static void *
+qhash_set_overflow (void *v, uword elt_bytes,
+ uword key, uword bi, uword * n_elts, u32 * result)
+{
+ qhash_t *h = qhash_header (v);
+ uword *p = hash_get (h->overflow_hash, key);
+ uword i;
+
+ bi /= QHASH_KEYS_PER_BUCKET;
+
+ if (p)
+ i = p[0];
+ else
+ {
+ uword l = vec_len (h->overflow_free_indices);
+ if (l > 0)
+ {
+ i = h->overflow_free_indices[l - 1];
+ _vec_len (h->overflow_free_indices) = l - 1;
+ }
+ else
+ i = (1 << h->log2_hash_size) + hash_elts (h->overflow_hash);
+ hash_set (h->overflow_hash, key, i);
+ vec_validate (h->overflow_counts, bi);
+ h->overflow_counts[bi] += 1;
+ *n_elts += 1;
+
+ l = vec_len (v);
+ if (i >= l)
+ {
+ uword dl = round_pow2 (1 + i - l, 8);
+ v = _vec_resize (v, dl, (l + dl) * elt_bytes, sizeof (h[0]),
+ /* align */ sizeof (uword));
+ clib_memset (v + l * elt_bytes, ~0, dl * elt_bytes);
+ }
+ }
+
+ *result = i;
+
+ return v;
+}
+
+static uword
+qhash_unset_overflow (void *v, uword key, uword bi, uword * n_elts)
+{
+ qhash_t *h = qhash_header (v);
+ uword *p = hash_get (h->overflow_hash, key);
+ uword result;
+
+ bi /= QHASH_KEYS_PER_BUCKET;
+
+ if (p)
+ {
+ result = p[0];
+ hash_unset (h->overflow_hash, key);
+ ASSERT (bi < vec_len (h->overflow_counts));
+ ASSERT (h->overflow_counts[bi] > 0);
+ ASSERT (*n_elts > 0);
+ vec_add1 (h->overflow_free_indices, result);
+ h->overflow_counts[bi] -= 1;
+ *n_elts -= 1;
+ }
+ else
+ result = ~0;
+
+ return result;
+}
+
+always_inline uword
+qhash_find_free (uword i, uword valid_mask)
+{
+ return first_set (~valid_mask & pow2_mask (QHASH_KEYS_PER_BUCKET));
+}
+
+void *
+_qhash_set_multiple (void *v,
+ uword elt_bytes,
+ uword * search_keys,
+ uword n_search_keys, u32 * result_indices)
+{
+ qhash_t *h = qhash_header (v);
+ uword *k, *hash_keys;
+ uword n_left, n_elts, bucket_mask;
+ u32 *r;
+
+ if (vec_len (v) < n_search_keys)
+ v = _qhash_resize (v, n_search_keys, elt_bytes);
+
+ if (qhash_min_log2 (2) != 1)
+ {
+ qhash_min_log2_init ();
+ ASSERT (qhash_min_log2 (2) == 1);
+ }
+
+ ASSERT (v != 0);
+
+ bucket_mask = pow2_mask (h->log2_hash_size) & ~(QHASH_KEYS_PER_BUCKET - 1);
+
+ hash_keys = h->hash_keys;
+ k = search_keys;
+ r = result_indices;
+ n_left = n_search_keys;
+ n_elts = h->n_elts;
+
+ while (n_left >= 2)
+ {
+ u32 a0, b0, c0, bi0, match0, valid0, free0;
+ u32 a1, b1, c1, bi1, match1, valid1, free1;
+ uword k0, *h0;
+ uword k1, *h1;
+
+ k0 = k[0];
+ k1 = k[1];
+
+ /* Keys must be unique. */
+ ASSERT (k0 != k1);
+
+ n_left -= 2;
+ k += 2;
+
+ a0 = a1 = h->hash_seeds[0];
+ b0 = b1 = h->hash_seeds[1];
+ c0 = c1 = h->hash_seeds[2];
+ a0 ^= k0;
+ a1 ^= k1;
+#if uword_bits == 64
+ b0 ^= k0 >> 32;
+ b1 ^= k1 >> 32;
+#endif
+
+ hash_mix32_step_1 (a0, b0, c0);
+ hash_mix32_step_1 (a1, b1, c1);
+ hash_mix32_step_2 (a0, b0, c0);
+ hash_mix32_step_2 (a1, b1, c1);
+ hash_mix32_step_3 (a0, b0, c0);
+ hash_mix32_step_3 (a1, b1, c1);
+
+ bi0 = c0 & bucket_mask;
+ bi1 = c1 & bucket_mask;
+
+ h0 = hash_keys + bi0;
+ h1 = hash_keys + bi1;
+
+ /* Search two buckets. */
+ valid0 = qhash_get_valid_elt_mask (h, bi0);
+ valid1 = qhash_get_valid_elt_mask (h, bi1);
+
+ match0 = qhash_search_bucket (h0, k0, valid0);
+ match1 = qhash_search_bucket (h1, k1, valid1);
+
+ /* Find first free element starting at hash offset into bucket. */
+ free0 = qhash_find_free (c0 & (QHASH_KEYS_PER_BUCKET - 1), valid0);
+
+ valid1 = valid1 | (bi0 == bi1 ? free0 : 0);
+ free1 = qhash_find_free (c1 & (QHASH_KEYS_PER_BUCKET - 1), valid1);
+
+ n_elts += (match0 == 0) + (match1 == 0);
+
+ match0 = match0 ? match0 : free0;
+ match1 = match1 ? match1 : free1;
+
+ valid0 |= match0;
+ valid1 |= match1;
+
+ h0 += qhash_min_log2 (match0);
+ h1 += qhash_min_log2 (match1);
+
+ if (PREDICT_FALSE (!match0 || !match1))
+ goto slow_path2;
+
+ h0[0] = k0;
+ h1[0] = k1;
+ r[0] = h0 - hash_keys;
+ r[1] = h1 - hash_keys;
+ r += 2;
+ qhash_set_valid_elt_mask (h, bi0, valid0);
+ qhash_set_valid_elt_mask (h, bi1, valid1);
+ continue;
+
+ slow_path2:
+ if (!match0)
+ {
+ n_elts -= 1;
+ v = qhash_set_overflow (v, elt_bytes, k0, bi0, &n_elts, &r[0]);
+ }
+ else
+ {
+ h0[0] = k0;
+ r[0] = h0 - hash_keys;
+ qhash_set_valid_elt_mask (h, bi0, valid0);
+ }
+ if (!match1)
+ {
+ n_elts -= 1;
+ v = qhash_set_overflow (v, elt_bytes, k1, bi1, &n_elts, &r[1]);
+ }
+ else
+ {
+ h1[0] = k1;
+ r[1] = h1 - hash_keys;
+ qhash_set_valid_elt_mask (h, bi1, valid1);
+ }
+ r += 2;
+ }
+
+ while (n_left >= 1)
+ {
+ u32 a0, b0, c0, bi0, match0, valid0, free0;
+ uword k0, *h0;
+
+ k0 = k[0];
+ n_left -= 1;
+ k += 1;
+
+ a0 = h->hash_seeds[0];
+ b0 = h->hash_seeds[1];
+ c0 = h->hash_seeds[2];
+ a0 ^= k0;
+#if uword_bits == 64
+ b0 ^= k0 >> 32;
+#endif
+
+ hash_mix32 (a0, b0, c0);
+
+ bi0 = c0 & bucket_mask;
+
+ h0 = hash_keys + bi0;
+
+ valid0 = qhash_get_valid_elt_mask (h, bi0);
+
+ /* Find first free element starting at hash offset into bucket. */
+ free0 = qhash_find_free (c0 & (QHASH_KEYS_PER_BUCKET - 1), valid0);
+
+ match0 = qhash_search_bucket (h0, k0, valid0);
+
+ n_elts += (match0 == 0);
+
+ match0 = match0 ? match0 : free0;
+
+ valid0 |= match0;
+
+ h0 += qhash_min_log2 (match0);
+
+ if (PREDICT_FALSE (!match0))
+ goto slow_path1;
+
+ h0[0] = k0;
+ r[0] = h0 - hash_keys;
+ r += 1;
+ qhash_set_valid_elt_mask (h, bi0, valid0);
+ continue;
+
+ slow_path1:
+ n_elts -= 1;
+ v = qhash_set_overflow (v, elt_bytes, k0, bi0, &n_elts, &r[0]);
+ r += 1;
+ }
+
+ h = qhash_header (v);
+ h->n_elts = n_elts;
+
+ return v;
+}
+
+static uword
+unset_slow_path (void *v, uword elt_bytes,
+ uword k0, uword bi0, uword valid0, uword match0,
+ uword * n_elts)
+{
+ qhash_t *h = qhash_header (v);
+ uword i, j = 0, k, l, t = ~0;
+ hash_pair_t *p, *found;
+
+ if (!match0)
+ {
+ if (valid0 == QHASH_ALL_VALID)
+ t = qhash_unset_overflow (v, k0, bi0, n_elts);
+ return t;
+ }
+
+ i = bi0 / QHASH_KEYS_PER_BUCKET;
+ t = bi0 + qhash_min_log2 (match0);
+
+ if (valid0 == QHASH_ALL_VALID
+ && i < vec_len (h->overflow_counts) && h->overflow_counts[i] > 0)
+ {
+ found = 0;
+ /* *INDENT-OFF* */
+ hash_foreach_pair (p, h->overflow_hash, ({
+ j = qhash_hash_mix (h, p->key) / QHASH_KEYS_PER_BUCKET;
+ if (j == i)
+ {
+ found = p;
+ break;
+ }
+ }));
+ /* *INDENT-ON* */
+ ASSERT (found != 0);
+ ASSERT (j == i);
+
+ l = found->value[0];
+ k = found->key;
+ hash_unset3 (h->overflow_hash, k, &j);
+ vec_add1 (h->overflow_free_indices, j);
+ h->overflow_counts[i] -= 1;
+
+ qhash_set_valid_elt_mask (h, bi0, valid0);
+
+ h->hash_keys[t] = k;
+ clib_memswap (v + t * elt_bytes, v + l * elt_bytes, elt_bytes);
+ t = l;
+ }
+ else
+ qhash_set_valid_elt_mask (h, bi0, valid0 ^ match0);
+
+ return t;
+}
+
+void
+_qhash_unset_multiple (void *v,
+ uword elt_bytes,
+ uword * search_keys,
+ uword n_search_keys, u32 * result_indices)
+{
+ qhash_t *h = qhash_header (v);
+ uword *k, *hash_keys;
+ uword n_left, n_elts, bucket_mask;
+ u32 *r;
+
+ if (!v)
+ {
+ uword i;
+ for (i = 0; i < n_search_keys; i++)
+ result_indices[i] = ~0;
+ }
+
+ bucket_mask = pow2_mask (h->log2_hash_size) & ~(QHASH_KEYS_PER_BUCKET - 1);
+
+ hash_keys = h->hash_keys;
+ k = search_keys;
+ r = result_indices;
+ n_left = n_search_keys;
+ n_elts = h->n_elts;
+
+ while (n_left >= 2)
+ {
+ u32 a0, b0, c0, bi0, match0, valid0;
+ u32 a1, b1, c1, bi1, match1, valid1;
+ uword k0, *h0;
+ uword k1, *h1;
+
+ k0 = k[0];
+ k1 = k[1];
+
+ /* Keys must be unique. */
+ ASSERT (k0 != k1);
+
+ n_left -= 2;
+ k += 2;
+
+ a0 = a1 = h->hash_seeds[0];
+ b0 = b1 = h->hash_seeds[1];
+ c0 = c1 = h->hash_seeds[2];
+ a0 ^= k0;
+ a1 ^= k1;
+#if uword_bits == 64
+ b0 ^= k0 >> 32;
+ b1 ^= k1 >> 32;
+#endif
+
+ hash_mix32_step_1 (a0, b0, c0);
+ hash_mix32_step_1 (a1, b1, c1);
+ hash_mix32_step_2 (a0, b0, c0);
+ hash_mix32_step_2 (a1, b1, c1);
+ hash_mix32_step_3 (a0, b0, c0);
+ hash_mix32_step_3 (a1, b1, c1);
+
+ bi0 = c0 & bucket_mask;
+ bi1 = c1 & bucket_mask;
+
+ h0 = hash_keys + bi0;
+ h1 = hash_keys + bi1;
+
+ /* Search two buckets. */
+ valid0 = qhash_get_valid_elt_mask (h, bi0);
+ valid1 = qhash_get_valid_elt_mask (h, bi1);
+
+ match0 = qhash_search_bucket (h0, k0, valid0);
+ match1 = qhash_search_bucket (h1, k1, valid1);
+
+ n_elts -= (match0 != 0) + (match1 != 0);
+
+ if (PREDICT_FALSE (valid0 == QHASH_ALL_VALID
+ || valid1 == QHASH_ALL_VALID))
+ goto slow_path2;
+
+ valid0 ^= match0;
+ qhash_set_valid_elt_mask (h, bi0, valid0);
+
+ valid1 = bi0 == bi1 ? valid0 : valid1;
+ valid1 ^= match1;
+
+ qhash_set_valid_elt_mask (h, bi1, valid1);
+
+ r[0] = match0 ? bi0 + qhash_min_log2 (match0) : ~0;
+ r[1] = match1 ? bi1 + qhash_min_log2 (match1) : ~0;
+ r += 2;
+ continue;
+
+ slow_path2:
+ r[0] = unset_slow_path (v, elt_bytes, k0, bi0, valid0, match0, &n_elts);
+ if (bi0 == bi1)
+ {
+ /* Search again in same bucket to test new overflow element. */
+ valid1 = qhash_get_valid_elt_mask (h, bi0);
+ if (!match1)
+ {
+ match1 = qhash_search_bucket (h1, k1, valid1);
+ n_elts -= (match1 != 0);
+ }
+ }
+ r[1] = unset_slow_path (v, elt_bytes, k1, bi1, valid1, match1, &n_elts);
+ r += 2;
+ }
+
+ while (n_left >= 1)
+ {
+ u32 a0, b0, c0, bi0, match0, valid0;
+ uword k0, *h0;
+
+ k0 = k[0];
+ n_left -= 1;
+ k += 1;
+
+ a0 = h->hash_seeds[0];
+ b0 = h->hash_seeds[1];
+ c0 = h->hash_seeds[2];
+ a0 ^= k0;
+#if uword_bits == 64
+ b0 ^= k0 >> 32;
+#endif
+
+ hash_mix32 (a0, b0, c0);
+
+ bi0 = c0 & bucket_mask;
+
+ h0 = hash_keys + bi0;
+
+ valid0 = qhash_get_valid_elt_mask (h, bi0);
+
+ match0 = qhash_search_bucket (h0, k0, valid0);
+ n_elts -= (match0 != 0);
+ qhash_set_valid_elt_mask (h, bi0, valid0 ^ match0);
+
+ r[0] = match0 ? bi0 + qhash_min_log2 (match0) : ~0;
+ r += 1;
+
+ if (PREDICT_FALSE (valid0 == QHASH_ALL_VALID))
+ r[-1] = unset_slow_path (v, elt_bytes, k0, bi0, valid0, match0,
+ &n_elts);
+ }
+
+ h->n_elts = n_elts;
+}
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/qhash.h b/extras/deprecated/vppinfra/qhash.h
new file mode 100644
index 00000000000..9dbbd971ade
--- /dev/null
+++ b/extras/deprecated/vppinfra/qhash.h
@@ -0,0 +1,169 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ Copyright (c) 2006 Eliot Dresselhaus
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice shall be
+ included in all copies or substantial portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+*/
+
+#ifndef included_qhash_h
+#define included_qhash_h
+
+#include <vppinfra/cache.h>
+#include <vppinfra/hash.h>
+
+/* Word hash tables. */
+typedef struct
+{
+ /* Number of elements in hash. */
+ u32 n_elts;
+
+ u32 log2_hash_size;
+
+ /* Jenkins hash seeds. */
+ u32 hash_seeds[3];
+
+ /* Fall back CLIB hash for overflow in fixed sized buckets. */
+ uword *overflow_hash;
+
+ u32 *overflow_counts, *overflow_free_indices;
+
+ u8 *hash_key_valid_bitmap;
+
+ uword *hash_keys;
+} qhash_t;
+
+always_inline qhash_t *
+qhash_header (void *v)
+{
+ return vec_header (v, sizeof (qhash_t));
+}
+
+always_inline uword
+qhash_elts (void *v)
+{
+ return v ? qhash_header (v)->n_elts : 0;
+}
+
+always_inline uword
+qhash_n_overflow (void *v)
+{
+ return v ? hash_elts (qhash_header (v)->overflow_hash) : 0;
+}
+
+#define QHASH_LOG2_KEYS_PER_BUCKET 2
+#define QHASH_KEYS_PER_BUCKET (1 << QHASH_LOG2_KEYS_PER_BUCKET)
+
+always_inline uword
+qhash_hash_mix (qhash_t * h, uword key)
+{
+ u32 a, b, c;
+
+ a = h->hash_seeds[0];
+ b = h->hash_seeds[1];
+ c = h->hash_seeds[2];
+
+ a ^= key;
+#if uword_bits == 64
+ b ^= key >> 32;
+#endif
+
+ hash_mix32 (a, b, c);
+
+ return c & pow2_mask (h->log2_hash_size);
+}
+
+#define qhash_resize(v,n) (v) = _qhash_resize ((v), (n), sizeof ((v)[0]))
+
+#define qhash_foreach(var,v,body)
+
+#define qhash_set_multiple(v,keys,n,results) \
+ (v) = _qhash_set_multiple ((v), sizeof ((v)[0]), (keys), (n), (results))
+
+#define qhash_unset_multiple(v,keys,n,results) \
+ _qhash_unset_multiple ((v), sizeof ((v)[0]), (keys), (n), (results))
+
+#define qhash_get(v,key) \
+({ \
+ uword _qhash_get_k = (key); \
+ qhash_get_first_match ((v), &_qhash_get_k, 1, &_qhash_get_k); \
+})
+
+#define qhash_set(v,k) \
+({ \
+ uword _qhash_set_k = (k); \
+ qhash_set_multiple ((v), &_qhash_set_k, 1, &_qhash_set_k); \
+ _qhash_set_k; \
+})
+
+#define qhash_unset(v,k) \
+({ \
+ uword _qhash_unset_k = (k); \
+ qhash_unset_multiple ((v), &_qhash_unset_k, 1, &_qhash_unset_k); \
+ _qhash_unset_k; \
+})
+
+void *_qhash_resize (void *v, uword length, uword elt_bytes);
+
+/* Lookup multiple keys in the same hash table. */
+void
+qhash_get_multiple (void *v,
+ uword * search_keys,
+ uword n_search_keys, u32 * result_indices);
+
+/* Lookup multiple keys in the same hash table.
+ Returns index of first matching key. */
+u32
+qhash_get_first_match (void *v,
+ uword * search_keys,
+ uword n_search_keys, uword * matching_key);
+
+/* Set/unset helper functions. */
+void *_qhash_set_multiple (void *v,
+ uword elt_bytes,
+ uword * search_keys,
+ uword n_search_keys, u32 * result_indices);
+void
+_qhash_unset_multiple (void *v,
+ uword elt_bytes,
+ uword * search_keys,
+ uword n_search_keys, u32 * result_indices);
+
+#endif /* included_qhash_h */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/slist.c b/extras/deprecated/vppinfra/slist.c
new file mode 100644
index 00000000000..5598871c884
--- /dev/null
+++ b/extras/deprecated/vppinfra/slist.c
@@ -0,0 +1,336 @@
+/*
+ Copyright (c) 2012 Cisco and/or its affiliates.
+
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+*/
+
+#include <vppinfra/slist.h>
+
+/*
+ * skip-list implementation
+ *
+ * Good news / bad news. As balanced binary tree schemes go,
+ * this one seems pretty fast and is reasonably simple. There's a very
+ * limited amount that can be done to mitigate sdram read latency.
+ *
+ * Each active clib_slist_elt_t is on from 1 to N lists. Each active element
+ * is always on the "level-0" list. Since most elements are *only* on
+ * level 0, we keep the level 0 (and level 1) in the element. For those
+ * elements on more than two lists, we switch to a vector. Hence, the
+ * "n" union in slib_slist_elt_t.
+ *
+ * The low-order bit of elt->n.next0[0] is 1 for inlined next indices,
+ * 0 for vector indices (since the allocator always aligns to at least
+ * a 4-byte boundary). We can only represent 2e9 items, but since the
+ * practical performance limit is O(1e7), it doesn't matter.
+ *
+ * We create a "head" element which (by construction) is always
+ * lexically lighter than any other element. This makes a large number
+ * of irritating special cases go away.
+ *
+ * User code is in charge of comparing a supplied key with
+ * the key component of a user pool element. The user tells this code
+ * to add or delete (opaque key, 32-bit integer) pairs to the skip-list.
+ *
+ * The algorithm adds new elements to one or more lists.
+ * For levels greater than zero, the probability of a new element landing on
+ * a list is branching_factor**N. Branching_factor = 0.2 seems to work
+ * OK, yielding about 50 compares per search at O(1e7) items.
+ */
+
+clib_error_t *
+clib_slist_init (clib_slist_t * sp, f64 branching_factor,
+ clib_slist_key_compare_function_t compare,
+ format_function_t format_user_element)
+{
+ clib_slist_elt_t *head;
+ clib_memset (sp, 0, sizeof (sp[0]));
+ sp->branching_factor = branching_factor;
+ sp->format_user_element = format_user_element;
+ sp->compare = compare;
+ sp->seed = 0xdeaddabe;
+ pool_get (sp->elts, head);
+ vec_add1 (head->n.nexts, (u32) ~ 0);
+ head->user_pool_index = (u32) ~ 0;
+ vec_validate (sp->path, 1);
+ vec_validate (sp->occupancy, 0);
+
+ return 0;
+}
+
+/*
+ * slist_search_internal
+ */
+static inline clib_slist_search_result_t
+slist_search_internal (clib_slist_t * sp, void *key, int need_full_path)
+{
+ int level, comp_result;
+ clib_slist_elt_t *search_elt, *head_elt;
+
+ sp->ncompares = 0;
+ /*
+ * index 0 is the magic listhead element which is
+ * lexically lighter than / to the left of every element
+ */
+ search_elt = head_elt = pool_elt_at_index (sp->elts, 0);
+
+ /*
+ * Initial negotiating position, only the head_elt is
+ * lighter than the supplied key
+ */
+ clib_memset (sp->path, 0, vec_len (head_elt->n.nexts) * sizeof (u32));
+
+ /* Walk the fastest lane first */
+ level = vec_len (head_elt->n.nexts) - 1;
+ _vec_len (sp->path) = level + 1;
+
+ while (1)
+ {
+ u32 next_index_this_level;
+ clib_slist_elt_t *prefetch_elt;
+
+ /*
+ * Prefetching the next element at this level makes a measurable
+ * difference, but doesn't fix the dependent read stall problem
+ */
+ prefetch_elt = sp->elts +
+ clib_slist_get_next_at_level (search_elt, level);
+
+ CLIB_PREFETCH (prefetch_elt, CLIB_CACHE_LINE_BYTES, READ);
+
+ /* Compare the key with the current element */
+ comp_result = (search_elt == head_elt) ? 1 :
+ sp->compare (key, search_elt->user_pool_index);
+
+ sp->ncompares++;
+ /* key "lighter" than this element */
+ if (comp_result < 0)
+ {
+ /*
+ * Back up to previous item on this list
+ * and search the next finer-grained list
+ * starting there.
+ */
+ search_elt = pool_elt_at_index (sp->elts, sp->path[level]);
+ next_list:
+ if (level > 0)
+ {
+ level--;
+ continue;
+ }
+ else
+ {
+ return CLIB_SLIST_NO_MATCH;
+ }
+ }
+ /* Match */
+ if (comp_result == 0)
+ {
+ /*
+ * If we're trying to delete an element, we need to
+ * track down all of the elements which point at it.
+ * Otherwise, don't bother with it
+ */
+ if (need_full_path && level > 0)
+ {
+ search_elt = pool_elt_at_index (sp->elts, sp->path[level]);
+ level--;
+ continue;
+ }
+ level = vec_len (head_elt->n.nexts);
+ sp->path[level] = search_elt - sp->elts;
+ _vec_len (sp->path) = level + 1;
+ return CLIB_SLIST_MATCH;
+ }
+ /*
+ * comp_result positive, key is to the right of
+ * this element
+ */
+ sp->path[level] = search_elt - sp->elts;
+
+ /* Out of list at this level? */
+ next_index_this_level =
+ clib_slist_get_next_at_level (search_elt, level);
+ if (next_index_this_level == (u32) ~ 0)
+ goto next_list;
+
+ /* No, try the next element */
+ search_elt = pool_elt_at_index (sp->elts, next_index_this_level);
+ }
+ return 0; /* notreached */
+}
+
+u32
+clib_slist_search (clib_slist_t * sp, void *key, u32 * ncompares)
+{
+ clib_slist_search_result_t rv;
+
+ rv = slist_search_internal (sp, key, 0 /* dont need full path */ );
+ if (rv == CLIB_SLIST_MATCH)
+ {
+ clib_slist_elt_t *elt;
+ elt = pool_elt_at_index (sp->elts, sp->path[vec_len (sp->path) - 1]);
+ if (ncompares)
+ *ncompares = sp->ncompares;
+ return elt->user_pool_index;
+ }
+ return (u32) ~ 0;
+}
+
+void
+clib_slist_add (clib_slist_t * sp, void *key, u32 user_pool_index)
+{
+ clib_slist_elt_t *new_elt;
+ clib_slist_search_result_t search_result;
+ int level;
+
+ search_result = slist_search_internal (sp, key,
+ 0 /* don't need full path */ );
+
+ /* Special case: key exists, just replace user_pool_index */
+ if (PREDICT_FALSE (search_result == CLIB_SLIST_MATCH))
+ {
+ clib_slist_elt_t *elt;
+ elt = pool_elt_at_index (sp->elts, sp->path[0]);
+ elt->user_pool_index = user_pool_index;
+ return;
+ }
+
+ pool_get (sp->elts, new_elt);
+ new_elt->n.nexts = 0;
+ new_elt->user_pool_index = user_pool_index;
+
+ /* sp->path lists elements to the left of key, by level */
+ for (level = 0; level < vec_len (sp->path); level++)
+ {
+ clib_slist_elt_t *prev_elt_this_level;
+ u32 prev_elt_next_index_this_level;
+
+ /* Add to list at the current level */
+ prev_elt_this_level = pool_elt_at_index (sp->elts, sp->path[level]);
+ prev_elt_next_index_this_level = clib_slist_get_next_at_level
+ (prev_elt_this_level, level);
+
+ clib_slist_set_next_at_level (new_elt, prev_elt_next_index_this_level,
+ level);
+
+ clib_slist_set_next_at_level (prev_elt_this_level, new_elt - sp->elts,
+ level);
+ sp->occupancy[level]++;
+
+ /* Randomly add to the next-higher level */
+ if (random_f64 (&sp->seed) > sp->branching_factor)
+ break;
+ }
+ {
+ /* Time to add a new ply? */
+ clib_slist_elt_t *head_elt = pool_elt_at_index (sp->elts, 0);
+ int top_level = vec_len (head_elt->n.nexts) - 1;
+ if (((f64) sp->occupancy[top_level]) * sp->branching_factor > 1.0)
+ {
+ vec_add1 (sp->occupancy, 0);
+ vec_add1 (head_elt->n.nexts, (u32) ~ 0);
+ /* full match case returns n+1 items */
+ vec_validate (sp->path, vec_len (head_elt->n.nexts));
+ }
+ }
+}
+
+clib_slist_search_result_t
+clib_slist_del (clib_slist_t * sp, void *key)
+{
+ clib_slist_search_result_t search_result;
+ clib_slist_elt_t *del_elt;
+ int level;
+
+ search_result = slist_search_internal (sp, key, 1 /* need full path */ );
+
+ if (PREDICT_FALSE (search_result == CLIB_SLIST_NO_MATCH))
+ return search_result;
+
+ del_elt = pool_elt_at_index (sp->elts, sp->path[vec_len (sp->path) - 1]);
+ ASSERT (vec_len (sp->path) > 1);
+
+ for (level = 0; level < vec_len (sp->path) - 1; level++)
+ {
+ clib_slist_elt_t *path_elt;
+ u32 path_elt_next_index;
+
+ path_elt = pool_elt_at_index (sp->elts, sp->path[level]);
+ path_elt_next_index = clib_slist_get_next_at_level (path_elt, level);
+
+ /* Splice the item out of the list if it's adjacent to the victim */
+ if (path_elt_next_index == del_elt - sp->elts)
+ {
+ sp->occupancy[level]--;
+ path_elt_next_index = clib_slist_get_next_at_level (del_elt, level);
+ clib_slist_set_next_at_level (path_elt, path_elt_next_index, level);
+ }
+ }
+
+ /* If this element is on more than two lists it has a vector of nexts */
+ if (!(del_elt->n.next0[0] & 1))
+ vec_free (del_elt->n.nexts);
+ pool_put (sp->elts, del_elt);
+ return CLIB_SLIST_MATCH;
+}
+
+u8 *
+format_slist (u8 * s, va_list * args)
+{
+ clib_slist_t *sl = va_arg (*args, clib_slist_t *);
+ int verbose = va_arg (*args, int);
+ int i;
+ clib_slist_elt_t *head_elt, *elt;
+
+ s = format (s, "slist 0x%x, %u items, branching_factor %.2f\n", sl,
+ sl->occupancy ? sl->occupancy[0] : 0, sl->branching_factor);
+
+ if (pool_elts (sl->elts) == 0)
+ return s;
+
+ head_elt = pool_elt_at_index (sl->elts, 0);
+
+ for (i = 0; i < vec_len (head_elt->n.nexts); i++)
+ {
+ s = format (s, "level %d: %d elts\n", i,
+ sl->occupancy ? sl->occupancy[i] : 0);
+
+ if (verbose && head_elt->n.nexts[i] != (u32) ~ 0)
+ {
+ elt = pool_elt_at_index (sl->elts, head_elt->n.nexts[i]);
+ while (elt)
+ {
+ u32 next_index;
+ s = format (s, "%U(%d) ", sl->format_user_element,
+ elt->user_pool_index, elt - sl->elts);
+ next_index = clib_slist_get_next_at_level (elt, i);
+ ASSERT (next_index != 0x7fffffff);
+ if (next_index == (u32) ~ 0)
+ break;
+ else
+ elt = pool_elt_at_index (sl->elts, next_index);
+ }
+ }
+ s = format (s, "\n");
+ }
+ return s;
+}
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/slist.h b/extras/deprecated/vppinfra/slist.h
new file mode 100644
index 00000000000..a7c77e27c96
--- /dev/null
+++ b/extras/deprecated/vppinfra/slist.h
@@ -0,0 +1,145 @@
+/*
+ Copyright (c) 2012 Cisco and/or its affiliates.
+
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+*/
+
+#ifndef included_slist_h
+#define included_slist_h
+
+#include <stdarg.h>
+#include <vppinfra/clib.h>
+#include <vppinfra/vec.h>
+#include <vppinfra/pool.h>
+#include <vppinfra/error.h>
+#include <vppinfra/format.h>
+#include <vppinfra/cache.h>
+
+typedef word (clib_slist_key_compare_function_t)
+ (void *key, u32 elt_pool_index);
+
+typedef enum
+{
+ CLIB_SLIST_MATCH = 0,
+ CLIB_SLIST_NO_MATCH
+} clib_slist_search_result_t;
+
+typedef struct
+{
+ /* Vector of next elements. Every valid instance has at least one */
+ union
+ {
+ u32 next0[2];
+ u32 *nexts;
+ } n;
+
+ /* Index of item in user's pool */
+ u32 user_pool_index;
+ /* $$$ pad to even divisor of cache line */
+} clib_slist_elt_t;
+
+static inline u32
+clib_slist_get_next_at_level (clib_slist_elt_t * elt, int level)
+{
+ if (elt->n.next0[0] & 1)
+ {
+ ASSERT (level < 2);
+ if (level == 1)
+ return elt->n.next0[1];
+ /* preserve ~0 (end of list) */
+ return (elt->n.next0[0] == (u32) ~ 0) ? elt->n.next0[0] :
+ (elt->n.next0[0] >> 1);
+ }
+ else
+ {
+ ASSERT (level < vec_len (elt->n.nexts));
+ return elt->n.nexts[level];
+ }
+}
+
+static inline void
+clib_slist_set_next_at_level (clib_slist_elt_t * elt, u32 index, int level)
+{
+ u32 old_level0_value[2];
+ /* level0 and not a vector */
+ if (level < 2 && (elt->n.next0[0] == 0 || elt->n.next0[0] & 1))
+ {
+ if (level == 0)
+ {
+ elt->n.next0[0] = (index << 1) | 1;
+ return;
+ }
+ elt->n.next0[1] = index;
+ return;
+ }
+ /* have to save old level0 values? */
+ if (elt->n.next0[0] & 1)
+ {
+ old_level0_value[0] = (elt->n.next0[0] == (u32) ~ 0) ?
+ elt->n.next0[0] : elt->n.next0[0] >> 1;
+ old_level0_value[1] = elt->n.next0[1];
+ elt->n.nexts = 0;
+ vec_add1 (elt->n.nexts, old_level0_value[0]);
+ vec_add1 (elt->n.nexts, old_level0_value[1]);
+ }
+ vec_validate (elt->n.nexts, level);
+ elt->n.nexts[level] = index;
+}
+
+
+typedef struct
+{
+ /* pool of skip-list elements */
+ clib_slist_elt_t *elts;
+
+ /* last search path */
+ u32 *path;
+
+ /* last search number of compares */
+ u32 ncompares;
+
+ /* occupancy stats */
+ u32 *occupancy;
+
+ /* Comparison function */
+ clib_slist_key_compare_function_t *compare;
+
+ /* Format function */
+ format_function_t *format_user_element;
+
+ /* items appear in successive plies with Pr (1 / branching_factor) */
+ f64 branching_factor;
+
+ /* random seed */
+ u32 seed;
+} clib_slist_t;
+
+clib_error_t *clib_slist_init (clib_slist_t * sp, f64 branching_factor,
+ clib_slist_key_compare_function_t compare,
+ format_function_t format_user_element);
+
+format_function_t format_slist;
+
+void clib_slist_add (clib_slist_t * sp, void *key, u32 user_pool_index);
+clib_slist_search_result_t clib_slist_del (clib_slist_t * sp, void *key);
+u32 clib_slist_search (clib_slist_t * sp, void *key, u32 * ncompares);
+
+#endif /* included_slist_h */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/test_cuckoo_bihash.c b/extras/deprecated/vppinfra/test_cuckoo_bihash.c
new file mode 100644
index 00000000000..e992e26fa7d
--- /dev/null
+++ b/extras/deprecated/vppinfra/test_cuckoo_bihash.c
@@ -0,0 +1,451 @@
+/*
+ * Copyright (c) 2017 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+__thread int thread_id = 0;
+
+#include <vppinfra/time.h>
+#include <vppinfra/cache.h>
+#include <vppinfra/error.h>
+#include <vppinfra/heap.h>
+#include <vppinfra/format.h>
+#include <vppinfra/pool.h>
+#include <vppinfra/error.h>
+#include <vppinfra/hash.h>
+#include <vppinfra/cache.h>
+
+#define os_get_cpu_number() (thread_id)
+
+#include <vppinfra/cuckoo_8_8.h>
+#include <vppinfra/cuckoo_template.h>
+#include <vppinfra/cuckoo_template.c>
+
+#include <vppinfra/bihash_8_8.h>
+#include <vppinfra/bihash_template.h>
+#include <vppinfra/bihash_template.c>
+
+#include <pthread.h>
+
+#define MAX_THREADS 255
+
+typedef struct
+{
+ void *tm;
+ int thread_idx;
+ u64 nlookups;
+} thread_data_t;
+
+typedef struct
+{
+ u64 deadline;
+ u64 seed;
+ u32 nbuckets;
+ u32 nitems;
+ u32 runtime;
+ int verbose;
+ int non_random_keys;
+ int nthreads;
+ int search_iter;
+ uword *key_hash;
+ u64 *keys;
+ CVT (clib_cuckoo) ch;
+ BVT (clib_bihash) bh;
+ clib_time_t clib_time;
+ u64 *key_add_del_sequence;
+ u8 *key_op_sequence;
+ u64 *key_search_sequence[MAX_THREADS];
+ unformat_input_t *input;
+ u64 nadds;
+ u64 ndels;
+ pthread_t bwriter_thread;
+ pthread_t breader_threads[MAX_THREADS];
+ pthread_t cwriter_thread;
+ pthread_t creader_threads[MAX_THREADS];
+ thread_data_t wthread_data;
+ thread_data_t rthread_data[MAX_THREADS];
+} test_main_t;
+
+test_main_t test_main;
+
+uword
+vl (void *v)
+{
+ return vec_len (v);
+}
+
+#define w_thread(x, guts) \
+ void *x##writer_thread (void *v) \
+ { \
+ test_main_t *tm = v; \
+ uword counter = 0; \
+ u64 nadds = 0; \
+ u64 ndels = 0; \
+ u64 deadline = tm->deadline; \
+ do \
+ { \
+ for (counter = 0; counter < vec_len (tm->key_add_del_sequence); \
+ ++counter) \
+ { \
+ u64 idx = tm->key_add_del_sequence[counter]; \
+ u8 op = tm->key_op_sequence[counter]; \
+ if (op) \
+ { \
+ ++nadds; \
+ } \
+ else \
+ { \
+ ++ndels; \
+ } \
+ guts; \
+ if (clib_cpu_time_now () > deadline) \
+ { \
+ break; \
+ } \
+ } \
+ } \
+ while (clib_cpu_time_now () < deadline); \
+ tm->nadds = nadds; \
+ tm->ndels = ndels; \
+ return NULL; \
+ }
+
+/* *INDENT-OFF* */
+w_thread (b, {
+ BVT (clib_bihash_kv) kv;
+ kv.key = tm->keys[idx];
+ kv.value = *hash_get (tm->key_hash, kv.key);
+ BV (clib_bihash_add_del) (&tm->bh, &kv, op);
+});
+/* *INDENT-ON* */
+
+/* *INDENT-OFF* */
+w_thread (c, {
+ CVT (clib_cuckoo_kv) kv;
+ kv.key = tm->keys[idx];
+ kv.value = *hash_get (tm->key_hash, kv.key);
+ CV (clib_cuckoo_add_del) (&tm->ch, &kv, op, 0);
+});
+/* *INDENT-ON* */
+
+#define r_thread(x, guts) \
+ void *x##reader_thread (void *v) \
+ { \
+ thread_data_t *data = v; \
+ thread_id = data->thread_idx; \
+ test_main_t *tm = data->tm; \
+ uword thread_idx = data->thread_idx; \
+ u64 *idx; \
+ uword nlookups = 0; \
+ u64 deadline = tm->deadline; \
+ do \
+ { \
+ vec_foreach (idx, tm->key_search_sequence[thread_idx]) \
+ { \
+ guts; \
+ ++nlookups; \
+ if (clib_cpu_time_now () > deadline) \
+ { \
+ break; \
+ } \
+ } \
+ } \
+ while (clib_cpu_time_now () < deadline); \
+ data->nlookups = nlookups; \
+ return NULL; \
+ }
+
+/* *INDENT-OFF* */
+r_thread (c, {
+ CVT (clib_cuckoo_kv) kv;
+ kv.key = tm->keys[*idx];
+ kv.value = *hash_get (tm->key_hash, kv.key);
+ CV (clib_cuckoo_search) (&tm->ch, &kv, &kv);
+});
+/* *INDENT-ON* */
+
+/* *INDENT-OFF* */
+r_thread (b, {
+ BVT (clib_bihash_kv) kv;
+ kv.key = tm->keys[*idx];
+ kv.value = *hash_get (tm->key_hash, kv.key);
+ BV (clib_bihash_search) (&tm->bh, &kv, &kv);
+});
+/* *INDENT-ON* */
+
+#define run_threads(x) \
+ do \
+ { \
+ \
+ before = clib_time_now (&tm->clib_time); \
+ tm->deadline = clib_cpu_time_now () + \
+ tm->runtime * tm->clib_time.clocks_per_second; \
+ fformat (stdout, #x "-> Start threads..., runtime is %llu second(s)\n", \
+ (long long unsigned)tm->runtime); \
+ \
+ /* \
+ fformat (stdout, #x "-> Writer thread only...\n"); \
+ if (0 != \
+ pthread_create (&tm->x##writer_thread, NULL, x##writer_thread, tm)) \
+ { \
+ perror ("pthread_create()"); \
+ abort (); \
+ } \
+ \
+ if (0 != pthread_join (tm->x##writer_thread, NULL)) \
+ { \
+ perror ("pthread_join()"); \
+ abort (); \
+ } \
+ \
+ delta = clib_time_now (&tm->clib_time) - before; \
+ fformat (stdout, #x "-> %wu adds, %wu dels in %.6f seconds\n", \
+ tm->nadds, tm->ndels, delta); \
+ tm->nadds = 0; \
+ tm->ndels = 0; \
+ */ \
+ \
+ fformat (stdout, #x "-> Writer + %d readers\n", tm->nthreads); \
+ before = clib_time_now (&tm->clib_time); \
+ tm->deadline = clib_cpu_time_now () + \
+ tm->runtime * tm->clib_time.clocks_per_second; \
+ if (0 != \
+ pthread_create (&tm->x##writer_thread, NULL, x##writer_thread, tm)) \
+ { \
+ perror ("pthread_create()"); \
+ abort (); \
+ } \
+ \
+ for (i = 0; i < tm->nthreads; i++) \
+ { \
+ tm->rthread_data[i].nlookups = 0; \
+ if (0 != pthread_create (&tm->x##reader_threads[i], NULL, \
+ x##reader_thread, &tm->rthread_data[i])) \
+ { \
+ perror ("pthread_create()"); \
+ abort (); \
+ } \
+ } \
+ \
+ if (0 != pthread_join (tm->x##writer_thread, NULL)) \
+ { \
+ perror ("pthread_join()"); \
+ abort (); \
+ } \
+ \
+ for (i = 0; i < tm->nthreads; i++) \
+ { \
+ if (0 != pthread_join (tm->x##reader_threads[i], NULL)) \
+ { \
+ perror ("pthread_join()"); \
+ abort (); \
+ } \
+ } \
+ \
+ delta = clib_time_now (&tm->clib_time) - before; \
+ \
+ total_searches = 0; \
+ for (i = 0; i < tm->nthreads; ++i) \
+ { \
+ u64 nlookups = tm->rthread_data[i].nlookups; \
+ fformat (stdout, #x "-> Thread #%d: %u searches\n", i, nlookups); \
+ total_searches += nlookups; \
+ } \
+ \
+ if (delta > 0) \
+ { \
+ ops = (tm->nadds + tm->ndels) / (f64)delta; \
+ fformat (stdout, #x "-> %.f add/dels per second\n", ops); \
+ sps = ((f64)total_searches) / delta; \
+ fformat (stdout, #x "-> %.f searches per second\n", sps); \
+ } \
+ \
+ fformat (stdout, \
+ #x "-> %wu adds, %wu dels, %lld searches in %.6f seconds\n", \
+ tm->nadds, tm->ndels, total_searches, delta); \
+ } \
+ while (0);
+
+static void
+cb (CVT (clib_cuckoo) * h, void *ctx)
+{
+ fformat (stdout, "Garbage callback called...\n");
+}
+
+static clib_error_t *
+test_cuckoo_bihash (test_main_t * tm)
+{
+ int i;
+ uword *p;
+ uword total_searches;
+ f64 before, delta;
+ f64 ops = 0, sps = 0;
+ f64 bops = 0, bsps = 0;
+ f64 cops = 0, csps = 0;
+ CVT (clib_cuckoo) * ch;
+ BVT (clib_bihash) * bh;
+
+ ch = &tm->ch;
+ bh = &tm->bh;
+
+ CV (clib_cuckoo_init) (ch, "test", 1, cb, NULL);
+ BV (clib_bihash_init) (bh, (char *) "test", tm->nbuckets, 256 << 20);
+
+ fformat (stdout, "Pick %lld unique %s keys...\n", tm->nitems,
+ tm->non_random_keys ? "non-random" : "random");
+
+ for (i = 0; i < tm->nitems; i++)
+ {
+ u64 rndkey;
+
+ if (tm->non_random_keys == 0)
+ {
+
+ again:
+ rndkey = random_u64 (&tm->seed);
+
+ p = hash_get (tm->key_hash, rndkey);
+ if (p)
+ goto again;
+ }
+ else
+ rndkey = (u64) (i + 1) << 16;
+
+ hash_set (tm->key_hash, rndkey, i + 1);
+ vec_add1 (tm->keys, rndkey);
+
+ int j;
+ for (j = 0; j < tm->nthreads; ++j)
+ {
+ u64 *x = tm->key_search_sequence[j];
+ vec_add1 (x, random_u64 (&tm->seed) % tm->nitems);
+ tm->key_search_sequence[j] = x;
+ }
+ vec_add1 (tm->key_add_del_sequence,
+ random_u64 (&tm->seed) % tm->nitems);
+ vec_add1 (tm->key_op_sequence, (rndkey % 10 < 8) ? 1 : 0);
+ }
+
+ int thread_counter = 0;
+ tm->wthread_data.tm = tm;
+ tm->wthread_data.thread_idx = thread_counter;
+ for (i = 0; i < tm->nthreads; ++i)
+ {
+ tm->rthread_data[i].tm = tm;
+ tm->rthread_data[i].thread_idx = thread_counter;
+ tm->rthread_data[i].nlookups = 0;
+ ++thread_counter;
+ }
+
+ int iter;
+ for (iter = 0; iter < tm->search_iter; ++iter)
+ {
+ fformat (stdout, "Bihash test #%d\n", iter);
+ run_threads (b);
+ bops = ops;
+ bsps = sps;
+ fformat (stdout, "%U", BV (format_bihash), bh, 0);
+ fformat (stdout, "Cuckoo test #%d\n", iter);
+ run_threads (c);
+ cops = ops;
+ csps = sps;
+ fformat (stdout, "%U", CV (format_cuckoo), ch, 0);
+ fformat (stdout,
+ "Bihash add/del speed is %.2f%% of cuckoo add/del speed\n",
+ bops / cops * 100);
+ fformat (stdout,
+ "Bihash search speed is %.2f%% of cuckoo search speed\n",
+ bsps / csps * 100);
+ }
+ return 0;
+}
+
+clib_error_t *
+test_cuckoo_bihash_main (test_main_t * tm)
+{
+ unformat_input_t *i = tm->input;
+ clib_error_t *error;
+
+ while (unformat_check_input (i) != UNFORMAT_END_OF_INPUT)
+ {
+ if (unformat (i, "seed %u", &tm->seed))
+ ;
+ else if (unformat (i, "nbuckets %d", &tm->nbuckets))
+ ;
+ else if (unformat (i, "non-random-keys"))
+ tm->non_random_keys = 1;
+ else if (unformat (i, "nitems %d", &tm->nitems))
+ ;
+ else if (unformat (i, "search_iter %d", &tm->search_iter))
+ ;
+ else if (unformat (i, "verbose %d", &tm->verbose))
+ ;
+ else if (unformat (i, "runtime %d", &tm->runtime))
+ ;
+ else if (unformat (i, "nthreads %d", &tm->nthreads))
+ ;
+ else if (unformat (i, "verbose"))
+ tm->verbose = 1;
+ else
+ return clib_error_return (0, "unknown input '%U'",
+ format_unformat_error, i);
+ }
+
+ error = test_cuckoo_bihash (tm);
+
+ return error;
+}
+
+#ifdef CLIB_UNIX
+int
+main (int argc, char *argv[])
+{
+ unformat_input_t i;
+ clib_error_t *error;
+ test_main_t *tm = &test_main;
+ clib_memset (&test_main, 0, sizeof (test_main));
+
+ clib_mem_init (0, 3ULL << 30);
+
+ tm->input = &i;
+ tm->seed = 0xdeaddabe;
+
+ tm->nbuckets = 2;
+ tm->nitems = 5;
+ tm->verbose = 1;
+ tm->nthreads = 1;
+ clib_time_init (&tm->clib_time);
+ tm->runtime = 1;
+ tm->search_iter = 1;
+ tm->key_hash = hash_create (0, sizeof (uword));
+
+ unformat_init_command_line (&i, argv);
+ error = test_cuckoo_bihash_main (tm);
+ unformat_free (&i);
+
+ if (error)
+ {
+ clib_error_report (error);
+ return 1;
+ }
+ return 0;
+}
+#endif /* CLIB_UNIX */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/test_cuckoo_template.c b/extras/deprecated/vppinfra/test_cuckoo_template.c
new file mode 100644
index 00000000000..9619dc2e802
--- /dev/null
+++ b/extras/deprecated/vppinfra/test_cuckoo_template.c
@@ -0,0 +1,318 @@
+/*
+ * Copyright (c) 2017 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <vppinfra/time.h>
+#include <vppinfra/cache.h>
+#include <vppinfra/error.h>
+#include <vppinfra/heap.h>
+#include <vppinfra/format.h>
+#include <vppinfra/pool.h>
+#include <vppinfra/error.h>
+#include <vppinfra/hash.h>
+#include <vppinfra/cache.h>
+
+#include <vppinfra/time.h>
+#include <vppinfra/cache.h>
+#include <vppinfra/error.h>
+
+#include <vppinfra/cuckoo_8_8.h>
+#include <vppinfra/cuckoo_template.h>
+#include <vppinfra/cuckoo_template.c>
+
+typedef struct
+{
+ u64 seed;
+ u32 nbuckets;
+ u32 nitems;
+ u32 search_iter;
+ int careful_delete_tests;
+ int verbose;
+ int non_random_keys;
+ uword *key_hash;
+ u64 *keys;
+ CVT (clib_cuckoo) hash;
+ clib_time_t clib_time;
+
+ unformat_input_t *input;
+
+} test_main_t;
+
+test_main_t test_main;
+
+uword
+vl (void *v)
+{
+ return vec_len (v);
+}
+
+void
+do_search (test_main_t * tm, CVT (clib_cuckoo) * h)
+{
+ int i, j;
+ CVT (clib_cuckoo_kv) kv;
+ for (j = 0; j < tm->search_iter; j++)
+ {
+ for (i = 0; i < tm->nitems; i++)
+ {
+ kv.key = tm->keys[i];
+ if (CV (clib_cuckoo_search) (h, &kv, &kv) < 0)
+ if (CV (clib_cuckoo_search) (h, &kv, &kv) < 0)
+ clib_warning ("[%d] search for key %llu failed unexpectedly\n",
+ i, tm->keys[i]);
+ if (kv.value != (u64) (i + 1))
+ clib_warning
+ ("[%d] search for key %llu returned %llu, not %llu\n", i,
+ tm->keys[i], kv.value, (u64) (i + 1));
+ }
+ }
+}
+
+static void
+cb (CVT (clib_cuckoo) * h, void *ctx)
+{
+ fformat (stdout, "Garbage callback called...");
+ if (clib_cpu_time_now () % 3)
+ {
+ fformat (stdout, "collecting garbage...\n");
+ CV (clib_cuckoo_garbage_collect) (h);
+ }
+ else
+ {
+ fformat (stdout, "ignoring for now...\n");
+ }
+}
+
+static clib_error_t *
+test_cuckoo (test_main_t * tm)
+{
+ int i;
+ uword *p;
+ uword total_searches;
+ f64 before, delta;
+ CVT (clib_cuckoo) * h;
+ CVT (clib_cuckoo_kv) kv;
+
+ h = &tm->hash;
+
+ CV (clib_cuckoo_init) (h, "test", tm->nbuckets, cb, NULL);
+
+ fformat (stdout, "Pick %lld unique %s keys...\n", tm->nitems,
+ tm->non_random_keys ? "non-random" : "random");
+
+ for (i = 0; i < tm->nitems; i++)
+ {
+ u64 rndkey;
+
+ if (tm->non_random_keys == 0)
+ {
+
+ again:
+ rndkey = random_u64 (&tm->seed);
+
+ p = hash_get (tm->key_hash, rndkey);
+ if (p)
+ goto again;
+ }
+ else
+ rndkey = (u64) (i + 1) << 16;
+
+ hash_set (tm->key_hash, rndkey, i + 1);
+ vec_add1 (tm->keys, rndkey);
+ }
+
+ fformat (stdout, "Add items...\n");
+ for (i = 0; i < tm->nitems; i++)
+ {
+ kv.key = tm->keys[i];
+ kv.value = i + 1;
+
+ CV (clib_cuckoo_add_del) (h, &kv, 1 /* is_add */ ,
+ 0 /* overwrite */ );
+
+ if (tm->verbose > 1)
+ {
+ fformat (stdout, "--------------------\n");
+ fformat (stdout, "After adding key %llu value %lld...\n",
+ tm->keys[i], (u64) (i + 1));
+ fformat (stdout, "%U", CV (format_cuckoo), h,
+ 2 /* very verbose */ );
+ }
+
+ CVT (clib_cuckoo_kv) kv2;
+ int rv = CV (clib_cuckoo_search) (h, &kv, &kv2);
+ ASSERT (CLIB_CUCKOO_ERROR_SUCCESS == rv);
+ }
+
+ fformat (stdout, "%U", CV (format_cuckoo), h, 0 /* very verbose */ );
+
+ fformat (stdout, "Search for items %d times...\n", tm->search_iter);
+
+ before = clib_time_now (&tm->clib_time);
+
+ do_search (tm, h);
+
+ delta = clib_time_now (&tm->clib_time) - before;
+ total_searches = (uword) tm->search_iter * (uword) tm->nitems;
+
+ if (delta > 0)
+ fformat (stdout, "%.f searches per second\n",
+ ((f64) total_searches) / delta);
+
+ fformat (stdout, "%lld searches in %.6f seconds\n", total_searches, delta);
+
+#if 0
+ int j;
+ fformat (stdout, "Standard E-hash search for items %d times...\n",
+ tm->search_iter);
+
+ before = clib_time_now (&tm->clib_time);
+
+ for (j = 0; j < tm->search_iter; j++)
+ {
+ for (i = 0; i < tm->nitems; i++)
+ {
+ p = hash_get (tm->key_hash, tm->keys[i]);
+ if (p == 0 || p[0] != (uword) (i + 1))
+ clib_warning ("ugh, couldn't find %lld\n", tm->keys[i]);
+ }
+ }
+
+ delta = clib_time_now (&tm->clib_time) - before;
+ total_searches = (uword) tm->search_iter * (uword) tm->nitems;
+
+ fformat (stdout, "%lld searches in %.6f seconds\n", total_searches, delta);
+
+ if (delta > 0)
+ fformat (stdout, "%.f searches per second\n",
+ ((f64) total_searches) / delta);
+
+#endif
+ fformat (stdout, "Delete items...\n");
+
+ for (i = 0; i < tm->nitems; i++)
+ {
+ int j;
+ int rv;
+
+ kv.key = tm->keys[i];
+ kv.value = (u64) (i + 1);
+ rv = CV (clib_cuckoo_add_del) (h, &kv, 0 /* is_add */ ,
+ 0 /* dont_overwrite */ );
+
+ if (rv < 0)
+ clib_warning ("delete key %lld not ok but should be", tm->keys[i]);
+
+ if (tm->careful_delete_tests)
+ {
+ for (j = 0; j < tm->nitems; j++)
+ {
+ kv.key = tm->keys[j];
+ rv = CV (clib_cuckoo_search) (h, &kv, &kv);
+ if (j <= i && rv >= 0)
+ {
+ clib_warning
+ ("i %d j %d search ok but should not be, value %lld", i,
+ j, kv.value);
+ }
+ if (j > i && rv < 0)
+ {
+ clib_warning ("i %d j %d search not ok but should be", i,
+ j);
+ }
+ }
+ }
+ }
+
+ fformat (stdout, "After deletions, should be empty...\n");
+
+ fformat (stdout, "%U", CV (format_cuckoo), h, 0 /* very verbose */ );
+ return 0;
+}
+
+clib_error_t *
+test_cuckoo_main (test_main_t * tm)
+{
+ unformat_input_t *i = tm->input;
+ clib_error_t *error;
+
+ while (unformat_check_input (i) != UNFORMAT_END_OF_INPUT)
+ {
+ if (unformat (i, "seed %u", &tm->seed))
+ ;
+ else if (unformat (i, "nbuckets %d", &tm->nbuckets))
+ ;
+ else if (unformat (i, "non-random-keys"))
+ tm->non_random_keys = 1;
+ else if (unformat (i, "nitems %d", &tm->nitems))
+ ;
+ else if (unformat (i, "careful %d", &tm->careful_delete_tests))
+ ;
+ else if (unformat (i, "verbose %d", &tm->verbose))
+ ;
+ else if (unformat (i, "search %d", &tm->search_iter))
+ ;
+ else if (unformat (i, "verbose"))
+ tm->verbose = 1;
+ else
+ return clib_error_return (0, "unknown input '%U'",
+ format_unformat_error, i);
+ }
+
+ error = test_cuckoo (tm);
+
+ return error;
+}
+
+#ifdef CLIB_UNIX
+int
+main (int argc, char *argv[])
+{
+ unformat_input_t i;
+ clib_error_t *error;
+ test_main_t *tm = &test_main;
+
+ clib_mem_init (0, 3ULL << 30);
+
+ tm->input = &i;
+ tm->seed = 0xdeaddabe;
+
+ tm->nbuckets = 2;
+ tm->nitems = 100000;
+ tm->verbose = 1;
+ tm->search_iter = 10000;
+ tm->careful_delete_tests = 0;
+ tm->key_hash = hash_create (0, sizeof (uword));
+ clib_time_init (&tm->clib_time);
+
+ unformat_init_command_line (&i, argv);
+ error = test_cuckoo_main (tm);
+ unformat_free (&i);
+
+ if (error)
+ {
+ clib_error_report (error);
+ return 1;
+ }
+ return 0;
+}
+#endif /* CLIB_UNIX */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/test_flowhash_template.c b/extras/deprecated/vppinfra/test_flowhash_template.c
new file mode 100644
index 00000000000..19ac4edf2e2
--- /dev/null
+++ b/extras/deprecated/vppinfra/test_flowhash_template.c
@@ -0,0 +1,257 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <vppinfra/time.h>
+#include <vppinfra/cache.h>
+#include <vppinfra/error.h>
+
+#include <vppinfra/heap.h>
+#include <vppinfra/format.h>
+#include <vppinfra/random.h>
+#include <vppinfra/hash.h>
+
+#include <vppinfra/flowhash_8_8.h>
+
+/* Not actually tested here. But included for compilation purposes. */
+#include <vppinfra/flowhash_24_16.h>
+
+typedef struct
+{
+ u64 seed;
+ u32 fixed_entries;
+ u32 collision_buckets;
+ u32 nitems;
+ u32 iterations;
+ u32 prefetch;
+ int non_random_keys;
+ uword *key_hash;
+ flowhash_lkey_8_8_t *keys;
+ flowhash_8_8_t *hash;
+ clib_time_t clib_time;
+ unformat_input_t *input;
+} test_main_t;
+
+test_main_t test_main;
+
+static clib_error_t *
+test_flowhash (test_main_t * tm)
+{
+ f64 before, delta;
+ u64 total;
+ u32 overflow;
+ int i, j;
+ uword *p;
+ tm->hash = flowhash_alloc_8_8 (tm->fixed_entries, tm->collision_buckets);
+ if (tm->hash == NULL)
+ return clib_error_return (0, "Could not alloc hash");
+
+ fformat (stdout, "Allocated hash memory size: %llu\n",
+ flowhash_memory_size (tm->hash));
+
+ fformat (stdout, "Pick %lld unique %s keys...\n",
+ tm->nitems, tm->non_random_keys ? "non-random" : "random");
+
+ for (i = 0; i < tm->nitems; i++)
+ {
+ flowhash_lkey_8_8_t rndkey;
+ if (tm->non_random_keys == 0)
+ {
+ again:
+ rndkey.as_u64[0] = random_u64 (&tm->seed);
+ if ((p = hash_get (tm->key_hash, rndkey.as_u64[0])))
+ goto again;
+ }
+ else
+ rndkey.as_u64[0] = (u64) (i + 1) << 16;
+
+ hash_set (tm->key_hash, rndkey.as_u64[0], i + 1);
+ vec_add1 (tm->keys, rndkey);
+ }
+
+ hash_free (tm->key_hash);
+
+ /* Additions */
+ overflow = 0;
+ before = clib_time_now (&tm->clib_time);
+ fformat (stdout, "Adding %u items...\n", tm->nitems);
+ for (i = 0; i < tm->nitems; i++)
+ {
+ u32 hash = flowhash_hash_8_8 (&tm->keys[i]);
+ u32 ei;
+ flowhash_get_8_8 (tm->hash, &tm->keys[i], hash, 1, &ei);
+ if (flowhash_is_overflow (ei))
+ overflow++;
+
+ /* Set value (No matter if success) */
+ flowhash_value (tm->hash, ei)->as_u64[0] = i + 1;
+
+ /* Save value until time > 1 */
+ flowhash_timeout (tm->hash, ei) = 1;
+ }
+
+ delta = clib_time_now (&tm->clib_time) - before;
+ total = tm->nitems;
+ fformat (stdout, "%lld additions in %.6f seconds\n", total, delta);
+ if (delta > 0)
+ fformat (stdout, "%.f additions per second\n", ((f64) total) / delta);
+
+ fformat (stdout, "%u elements in table\n", flowhash_elts_8_8 (tm->hash, 1));
+ fformat (stdout, "Flowhash counters:\n");
+ fformat (stdout, " collision-lookup: %lu\n",
+ tm->hash->collision_lookup_counter);
+ fformat (stdout, " not-enough-buckets: %lu\n",
+ tm->hash->not_enough_buckets_counter);
+ fformat (stdout, " overflows: %lu\n", overflow);
+
+ /* Lookups (very similar to additions) */
+ overflow = 0;
+ before = clib_time_now (&tm->clib_time);
+ fformat (stdout, "Looking up %u items %u times...\n", tm->nitems,
+ tm->iterations);
+
+ for (j = 0; j < tm->iterations; j++)
+ {
+ i = 0;
+ if (tm->prefetch)
+ for (; i < tm->nitems - tm->prefetch; i++)
+ {
+ u32 ei;
+ u32 hash = flowhash_hash_8_8 (&tm->keys[i + tm->prefetch]);
+ flowhash_prefetch (tm->hash, hash);
+ hash = flowhash_hash_8_8 (&tm->keys[i]);
+ flowhash_get_8_8 (tm->hash, &tm->keys[i], hash, 1, &ei);
+ if (flowhash_is_overflow (ei))
+ overflow++;
+ else if (flowhash_timeout (tm->hash, ei) != 1)
+ clib_warning ("Key not found: %lld\n", tm->keys[i].as_u64[0]);
+ else if (flowhash_value (tm->hash, ei)->as_u64[0] != i + 1)
+ clib_warning ("Value mismatch for key %lld\n",
+ tm->keys[i].as_u64[0]);
+ }
+
+ for (; i < tm->nitems; i++)
+ {
+ u32 ei;
+ u32 hash = flowhash_hash_8_8 (&tm->keys[i]);
+ flowhash_get_8_8 (tm->hash, &tm->keys[i], hash, 1, &ei);
+ if (flowhash_is_overflow (ei))
+ overflow++;
+ else if (flowhash_timeout (tm->hash, ei) != 1)
+ clib_warning ("Key not found: %lld\n", tm->keys[i].as_u64[0]);
+ else if (flowhash_value (tm->hash, ei)->as_u64[0] != i + 1)
+ clib_warning ("Value mismatch for key %lld\n",
+ tm->keys[i].as_u64[0]);
+ }
+ }
+
+ delta = clib_time_now (&tm->clib_time) - before;
+ total = tm->nitems * tm->iterations;
+ fformat (stdout, "%lld lookups in %.6f seconds\n", total, delta);
+ if (delta > 0)
+ fformat (stdout, "%.f lookups per second\n", ((f64) total) / delta);
+
+ /* Delete */
+ for (i = 0; i < tm->nitems; i++)
+ {
+ u32 hash = flowhash_hash_8_8 (&tm->keys[i]);
+ u32 ei;
+ flowhash_get_8_8 (tm->hash, &tm->keys[i], hash, 1, &ei);
+ flowhash_timeout (tm->hash, ei) = 0;
+ }
+
+ fformat (stdout, "%u elements in table\n", flowhash_elts_8_8 (tm->hash, 1));
+
+ vec_free (tm->keys);
+ flowhash_free_8_8 (tm->hash);
+
+ return NULL;
+}
+
+clib_error_t *
+test_flowhash_main (test_main_t * tm)
+{
+ unformat_input_t *i = tm->input;
+ clib_error_t *error;
+
+ while (unformat_check_input (i) != UNFORMAT_END_OF_INPUT)
+ {
+ if (unformat (i, "seed %u", &tm->seed))
+ ;
+ else if (unformat (i, "fixed-entries %d", &tm->fixed_entries))
+ ;
+ else if (unformat (i, "collision-buckets %d", &tm->collision_buckets))
+ ;
+ else if (unformat (i, "non-random-keys"))
+ tm->non_random_keys = 1;
+ else if (unformat (i, "nitems %d", &tm->nitems))
+ ;
+ else if (unformat (i, "prefetch %d", &tm->prefetch))
+ ;
+ else if (unformat (i, "iterations %d", &tm->iterations))
+ ;
+ else
+ return clib_error_return (0, "unknown input '%U'",
+ format_unformat_error, i);
+ }
+
+ error = test_flowhash (tm);
+ return error;
+}
+
+#ifdef CLIB_UNIX
+int
+main (int argc, char *argv[])
+{
+
+ unformat_input_t i;
+ clib_error_t *error;
+ test_main_t *tm = &test_main;
+
+ clib_mem_init (0, 3ULL << 30);
+
+ tm->fixed_entries = 8 << 20;
+ tm->collision_buckets = 1 << 20;
+ tm->seed = 0xdeadf00l;
+ tm->iterations = 1;
+ tm->input = &i;
+ tm->nitems = 1000;
+ tm->non_random_keys = 0;
+ tm->key_hash = hash_create (0, sizeof (uword));
+ tm->prefetch = 0;
+ clib_time_init (&tm->clib_time);
+
+ unformat_init_command_line (&i, argv);
+ error = test_flowhash_main (tm);
+ unformat_free (&i);
+
+ if (error)
+ {
+ clib_error_report (error);
+ return 1;
+ }
+ return 0;
+
+ return 0;
+}
+#endif /* CLIB_UNIX */
+
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/test_phash.c b/extras/deprecated/vppinfra/test_phash.c
new file mode 100644
index 00000000000..47711c28dbc
--- /dev/null
+++ b/extras/deprecated/vppinfra/test_phash.c
@@ -0,0 +1,151 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ Copyright (c) 2005 Eliot Dresselhaus
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice shall be
+ included in all copies or substantial portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+*/
+
+#include <vppinfra/phash.h>
+#include <vppinfra/format.h>
+#include <vppinfra/random.h>
+
+static int verbose;
+#define if_verbose(format,args...) \
+ if (verbose) { clib_warning(format, ## args); }
+
+int
+test_phash_main (unformat_input_t * input)
+{
+ phash_main_t _pm = { 0 }, *pm = &_pm;
+ int n_keys, random_keys;
+ u32 seed;
+ clib_error_t *error;
+
+ random_keys = 1;
+ n_keys = 1000;
+ while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
+ {
+ if (0 == unformat (input, "keys %d", &n_keys)
+ && 0 == unformat (input, "verbose %=", &verbose, 1)
+ && 0 == unformat (input, "random-keys %=", &random_keys, 1)
+ && 0 == unformat (input, "sequential-keys %=", &random_keys, 0)
+ && 0 == unformat (input, "seed %d", &pm->random_seed)
+ && 0 == unformat (input, "64-bit %|", &pm->flags, PHASH_FLAG_MIX64)
+ && 0 == unformat (input, "32-bit %|", &pm->flags, PHASH_FLAG_MIX32)
+ && 0 == unformat (input, "fast %|", &pm->flags,
+ PHASH_FLAG_FAST_MODE)
+ && 0 == unformat (input, "slow %|", &pm->flags,
+ PHASH_FLAG_SLOW_MODE)
+ && 0 == unformat (input, "minimal %|", &pm->flags,
+ PHASH_FLAG_MINIMAL)
+ && 0 == unformat (input, "non-minimal %|", &pm->flags,
+ PHASH_FLAG_NON_MINIMAL))
+ clib_error ("unknown input `%U'", format_unformat_error, input);
+ }
+
+ if (!pm->random_seed)
+ pm->random_seed = random_default_seed ();
+
+ if_verbose
+ ("%d %d-bit keys, random seed %d, %s mode, looking for %sminimal hash",
+ n_keys, (pm->flags & PHASH_FLAG_MIX64) ? 64 : 32, pm->random_seed,
+ (pm->flags & PHASH_FLAG_FAST_MODE) ? "fast" : "slow",
+ (pm->flags & PHASH_FLAG_MINIMAL) ? "" : "non-");
+
+ seed = pm->random_seed;
+
+ /* Initialize random keys. */
+ {
+ phash_key_t *k;
+
+ vec_resize (pm->keys, n_keys);
+ vec_foreach (k, pm->keys)
+ {
+ k->key = k - pm->keys;
+ if (random_keys)
+ {
+ if (pm->flags & PHASH_FLAG_MIX64)
+ k->key = random_u64 (&seed);
+ else
+ k->key = random_u32 (&seed);
+ }
+ }
+ }
+
+ error = phash_find_perfect_hash (pm);
+ if (error)
+ {
+ clib_error_report (error);
+ return 1;
+ }
+ else
+ {
+ if_verbose ("(%d,%d) (a,b) bits, %d seeds tried, %d tree walks",
+ pm->a_bits, pm->b_bits,
+ pm->n_seed_trials, pm->n_perfect_calls);
+
+ error = phash_validate (pm);
+ if (error)
+ {
+ clib_error_report (error);
+ return 1;
+ }
+ }
+
+ return 0;
+}
+
+#ifdef CLIB_UNIX
+int
+main (int argc, char *argv[])
+{
+ unformat_input_t i;
+ int res;
+
+ clib_mem_init (0, 64ULL << 20);
+
+ verbose = (argc > 1);
+ unformat_init_command_line (&i, argv);
+ res = test_phash_main (&i);
+ unformat_free (&i);
+ return res;
+}
+#endif
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/test_qhash.c b/extras/deprecated/vppinfra/test_qhash.c
new file mode 100644
index 00000000000..a520fa4bd77
--- /dev/null
+++ b/extras/deprecated/vppinfra/test_qhash.c
@@ -0,0 +1,335 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include <vppinfra/bitmap.h>
+#include <vppinfra/os.h>
+#include <vppinfra/qhash.h>
+#include <vppinfra/random.h>
+#include <vppinfra/time.h>
+
+typedef struct
+{
+ u32 n_iter, seed, n_keys, n_hash_keys, verbose;
+
+ u32 max_vector;
+
+ uword *hash;
+
+ uword *keys_in_hash_bitmap;
+
+ u32 *qhash;
+
+ uword *keys;
+
+ uword *lookup_keys;
+ uword *lookup_key_indices;
+ u32 *lookup_results;
+
+ u32 *get_multiple_results;
+
+ clib_time_t time;
+
+ f64 overflow_fraction, ave_elts;
+ f64 get_time, hash_get_time;
+ f64 set_time, set_count;
+ f64 unset_time, unset_count;
+ f64 hash_set_time, hash_unset_time;
+} test_qhash_main_t;
+
+clib_error_t *
+test_qhash_main (unformat_input_t * input)
+{
+ clib_error_t *error = 0;
+ test_qhash_main_t _tm, *tm = &_tm;
+ uword i, iter;
+
+ clib_memset (tm, 0, sizeof (tm[0]));
+ tm->n_iter = 10;
+ tm->seed = 1;
+ tm->n_keys = 10;
+ tm->max_vector = 1;
+
+ while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
+ {
+ if (unformat (input, "iter %d", &tm->n_iter))
+ ;
+ else if (unformat (input, "seed %d", &tm->seed))
+ ;
+ else if (unformat (input, "keys %d", &tm->n_keys))
+ ;
+ else if (unformat (input, "size %d", &tm->n_hash_keys))
+ ;
+ else if (unformat (input, "vector %d", &tm->max_vector))
+ ;
+ else if (unformat (input, "verbose"))
+ tm->verbose = 1;
+ else
+ {
+ error = clib_error_create ("unknown input `%U'\n",
+ format_unformat_error, input);
+ goto done;
+ }
+ }
+
+ if (!tm->seed)
+ tm->seed = random_default_seed ();
+
+ clib_time_init (&tm->time);
+
+ clib_warning ("iter %d, seed %u, keys %d, max vector %d, ",
+ tm->n_iter, tm->seed, tm->n_keys, tm->max_vector);
+
+ vec_resize (tm->keys, tm->n_keys);
+ vec_resize (tm->get_multiple_results, tm->n_keys);
+ for (i = 0; i < vec_len (tm->keys); i++)
+ tm->keys[i] = random_uword (&tm->seed);
+
+ if (!tm->n_hash_keys)
+ tm->n_hash_keys = 2 * max_pow2 (tm->n_keys);
+ tm->n_hash_keys = clib_max (tm->n_keys, tm->n_hash_keys);
+ qhash_resize (tm->qhash, tm->n_hash_keys);
+
+ {
+ qhash_t *h = qhash_header (tm->qhash);
+ int i;
+ for (i = 0; i < ARRAY_LEN (h->hash_seeds); i++)
+ h->hash_seeds[i] = random_uword (&tm->seed);
+ }
+
+ vec_resize (tm->lookup_keys, tm->max_vector);
+ vec_resize (tm->lookup_key_indices, tm->max_vector);
+ vec_resize (tm->lookup_results, tm->max_vector);
+
+ for (iter = 0; iter < tm->n_iter; iter++)
+ {
+ uword *p, j, n, is_set;
+
+ n = tm->max_vector;
+
+ is_set = random_u32 (&tm->seed) & 1;
+ is_set |= hash_elts (tm->hash) < (tm->n_keys / 4);
+ if (hash_elts (tm->hash) > (3 * tm->n_keys) / 4)
+ is_set = 0;
+
+ _vec_len (tm->lookup_keys) = n;
+ _vec_len (tm->lookup_key_indices) = n;
+ j = 0;
+ while (j < n)
+ {
+ i = random_u32 (&tm->seed) % vec_len (tm->keys);
+ if (clib_bitmap_get (tm->keys_in_hash_bitmap, i) != is_set)
+ {
+ f64 t[2];
+ tm->lookup_key_indices[j] = i;
+ tm->lookup_keys[j] = tm->keys[i];
+ t[0] = clib_time_now (&tm->time);
+ if (is_set)
+ hash_set (tm->hash, tm->keys[i], i);
+ else
+ hash_unset (tm->hash, tm->keys[i]);
+ t[1] = clib_time_now (&tm->time);
+ if (is_set)
+ tm->hash_set_time += t[1] - t[0];
+ else
+ tm->hash_unset_time += t[1] - t[0];
+ tm->keys_in_hash_bitmap
+ = clib_bitmap_set (tm->keys_in_hash_bitmap, i, is_set);
+ j++;
+ }
+ }
+
+ {
+ f64 t[2];
+
+ if (is_set)
+ {
+ t[0] = clib_time_now (&tm->time);
+ qhash_set_multiple (tm->qhash,
+ tm->lookup_keys,
+ vec_len (tm->lookup_keys),
+ tm->lookup_results);
+ t[1] = clib_time_now (&tm->time);
+ tm->set_time += t[1] - t[0];
+ tm->set_count += vec_len (tm->lookup_keys);
+ for (i = 0; i < vec_len (tm->lookup_keys); i++)
+ {
+ uword r = tm->lookup_results[i];
+ *vec_elt_at_index (tm->qhash, r) = tm->lookup_key_indices[i];
+ }
+ }
+ else
+ {
+ t[0] = clib_time_now (&tm->time);
+ qhash_unset_multiple (tm->qhash,
+ tm->lookup_keys,
+ vec_len (tm->lookup_keys),
+ tm->lookup_results);
+ t[1] = clib_time_now (&tm->time);
+ tm->unset_time += t[1] - t[0];
+ tm->unset_count += vec_len (tm->lookup_keys);
+
+ for (i = 0; i < vec_len (tm->lookup_keys); i++)
+ {
+ uword r = tm->lookup_results[i];
+ *vec_elt_at_index (tm->qhash, r) = ~0;
+ }
+ }
+ }
+
+ if (qhash_elts (tm->qhash) != hash_elts (tm->hash))
+ os_panic ();
+
+ {
+ qhash_t *h;
+ uword i, k, l, count;
+
+ h = qhash_header (tm->qhash);
+
+ for (i = k = 0; k < vec_len (h->hash_key_valid_bitmap); k++)
+ i += count_set_bits (h->hash_key_valid_bitmap[k]);
+ k = hash_elts (h->overflow_hash);
+ l = qhash_elts (tm->qhash);
+ if (i + k != l)
+ os_panic ();
+
+ count = hash_elts (h->overflow_hash);
+ for (i = 0; i < (1 << h->log2_hash_size); i++)
+ count += tm->qhash[i] != ~0;
+ if (count != qhash_elts (tm->qhash))
+ os_panic ();
+
+ {
+ u32 *tmp = 0;
+
+ /* *INDENT-OFF* */
+ hash_foreach (k, l, h->overflow_hash, ({
+ j = qhash_hash_mix (h, k) / QHASH_KEYS_PER_BUCKET;
+ vec_validate (tmp, j);
+ tmp[j] += 1;
+ }));
+ /* *INDENT-ON* */
+
+ for (k = 0; k < vec_len (tmp); k++)
+ {
+ if (k >= vec_len (h->overflow_counts))
+ os_panic ();
+ if (h->overflow_counts[k] != tmp[k])
+ os_panic ();
+ }
+ for (; k < vec_len (h->overflow_counts); k++)
+ if (h->overflow_counts[k] != 0)
+ os_panic ();
+
+ vec_free (tmp);
+ }
+ }
+
+ {
+ f64 t[2];
+
+ t[0] = clib_time_now (&tm->time);
+ qhash_get_multiple (tm->qhash, tm->keys, vec_len (tm->keys),
+ tm->get_multiple_results);
+ t[1] = clib_time_now (&tm->time);
+ tm->get_time += t[1] - t[0];
+
+ for (i = 0; i < vec_len (tm->keys); i++)
+ {
+ u32 r;
+
+ t[0] = clib_time_now (&tm->time);
+ p = hash_get (tm->hash, tm->keys[i]);
+ t[1] = clib_time_now (&tm->time);
+ tm->hash_get_time += t[1] - t[0];
+
+ r = qhash_get (tm->qhash, tm->keys[i]);
+ if (p)
+ {
+ if (p[0] != i)
+ os_panic ();
+ if (*vec_elt_at_index (tm->qhash, r) != i)
+ os_panic ();
+ }
+ else
+ {
+ if (r != ~0)
+ os_panic ();
+ }
+ if (r != tm->get_multiple_results[i])
+ os_panic ();
+ }
+ }
+
+ tm->overflow_fraction +=
+ ((f64) qhash_n_overflow (tm->qhash) / qhash_elts (tm->qhash));
+ tm->ave_elts += qhash_elts (tm->qhash);
+ }
+
+ fformat (stderr, "%d iter %.6e overflow, %.4f ave. elts\n",
+ tm->n_iter,
+ tm->overflow_fraction / tm->n_iter, tm->ave_elts / tm->n_iter);
+
+ tm->get_time /= tm->n_iter * vec_len (tm->keys);
+ tm->hash_get_time /= tm->n_iter * vec_len (tm->keys);
+
+ tm->set_time /= tm->set_count;
+ tm->unset_time /= tm->unset_count;
+ tm->hash_set_time /= tm->set_count;
+ tm->hash_unset_time /= tm->unset_count;
+
+ fformat (stderr,
+ "get/set/unset clocks %.2e %.2e %.2e clib %.2e %.2e %.2e ratio %.2f %.2f %.2f\n",
+ tm->get_time * tm->time.clocks_per_second,
+ tm->set_time * tm->time.clocks_per_second,
+ tm->unset_time * tm->time.clocks_per_second,
+ tm->hash_get_time * tm->time.clocks_per_second,
+ tm->hash_set_time * tm->time.clocks_per_second,
+ tm->hash_unset_time * tm->time.clocks_per_second,
+ tm->hash_get_time / tm->get_time, tm->hash_set_time / tm->set_time,
+ tm->hash_unset_time / tm->unset_time);
+
+
+done:
+ return error;
+}
+
+#ifdef CLIB_UNIX
+int
+main (int argc, char *argv[])
+{
+ unformat_input_t i;
+ clib_error_t *error;
+
+ clib_mem_init (0, 64ULL << 20);
+
+ unformat_init_command_line (&i, argv);
+ error = test_qhash_main (&i);
+ unformat_free (&i);
+ if (error)
+ {
+ clib_error_report (error);
+ return 1;
+ }
+ else
+ return 0;
+}
+#endif /* CLIB_UNIX */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/test_slist.c b/extras/deprecated/vppinfra/test_slist.c
new file mode 100644
index 00000000000..3c3cbf73ca9
--- /dev/null
+++ b/extras/deprecated/vppinfra/test_slist.c
@@ -0,0 +1,228 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifdef CLIB_UNIX
+#include <unistd.h>
+#include <stdlib.h>
+#include <stdio.h>
+#endif
+
+#include <vppinfra/slist.h>
+
+typedef struct
+{
+ u32 *random_pool;
+ u32 seed;
+ u32 iter;
+ u32 verbose;
+ f64 branching_factor;
+ clib_slist_t slist;
+} test_main_t;
+
+test_main_t test_main;
+
+#define foreach_simple_test \
+_(2) \
+_(4) \
+_(3) \
+_(1)
+
+
+void
+run_test (test_main_t * tm)
+{
+ int i;
+ u32 *tv;
+ u32 ncompares;
+ u64 total_compares = 0;
+
+ if (1)
+ {
+ /*
+ * Add a bunch of random numbers to the skip-list,
+ * sorting them.
+ */
+ for (i = 0; i < tm->iter; i++)
+ {
+ pool_get (tm->random_pool, tv);
+ *tv = random_u32 (&tm->seed);
+ clib_slist_add (&tm->slist, tv, tv - tm->random_pool);
+ }
+ /* make sure we can find each one */
+ for (i = 0; i < tm->iter; i++)
+ {
+ u32 search_result;
+ tv = pool_elt_at_index (tm->random_pool, i);
+
+ search_result = clib_slist_search (&tm->slist, tv, &ncompares);
+ ASSERT (search_result == i);
+
+ total_compares += ncompares;
+ }
+
+ fformat (stdout, "%.2f avg compares/search\n",
+ (f64) total_compares / (f64) i);
+
+ fformat (stdout, "%U\n", format_slist, &tm->slist,
+ tm->iter < 1000 /* verbose */ );
+
+ /* delete half of them */
+ for (i = tm->iter / 2; i < tm->iter; i++)
+ {
+ tv = pool_elt_at_index (tm->random_pool, i);
+ (void) clib_slist_del (&tm->slist, tv);
+ }
+
+ /* make sure we can find the set we should find, and no others */
+ for (i = 0; i < tm->iter; i++)
+ {
+ u32 search_result;
+ tv = pool_elt_at_index (tm->random_pool, i);
+
+ search_result = clib_slist_search (&tm->slist, tv, &ncompares);
+ if (i >= tm->iter / 2)
+ ASSERT (search_result == (u32) ~ 0);
+ else
+ ASSERT (search_result == i);
+
+ }
+
+ fformat (stdout, "%U\n", format_slist, &tm->slist,
+ tm->iter < 1000 /* verbose */ );
+
+ /* delete the rest */
+ for (i = 0; i < tm->iter; i++)
+ {
+ tv = pool_elt_at_index (tm->random_pool, i);
+
+ (void) clib_slist_del (&tm->slist, tv);
+ }
+
+ fformat (stdout, "%U\n", format_slist, &tm->slist,
+ tm->iter < 1000 /* verbose */ );
+ }
+ else
+ {
+
+#define _(n) \
+ do { \
+ pool_get (tm->random_pool, tv); \
+ *tv = n; \
+ clib_slist_add (&tm->slist, tv, tv - tm->random_pool); \
+ fformat(stdout, "%U\n", format_slist, &tm->slist, 1 /* verbose */); \
+ } while (0);
+ foreach_simple_test;
+#undef _
+ }
+
+ return;
+}
+
+word
+test_compare (void *key, u32 elt_index)
+{
+ u32 *k = (u32 *) key;
+ u32 elt = test_main.random_pool[elt_index];
+
+ if (*k < elt)
+ return -1;
+ if (*k > elt)
+ return 1;
+ return 0;
+}
+
+u8 *
+test_format (u8 * s, va_list * args)
+{
+ u32 elt_index = va_arg (*args, u32);
+ u32 elt = test_main.random_pool[elt_index];
+
+ return format (s, "%u", elt);
+}
+
+void
+initialize_slist (test_main_t * tm)
+{
+ clib_slist_init (&tm->slist, tm->branching_factor,
+ test_compare, test_format);
+}
+
+int
+test_slist_main (unformat_input_t * input)
+{
+ test_main_t *tm = &test_main;
+ u32 tmp;
+
+ tm->seed = 0xbabeb00b;
+ tm->iter = 100000;
+ tm->verbose = 1;
+ tm->branching_factor = 1.0 / 5.0;
+
+ while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
+ {
+ if (unformat (input, "seed %d", &tm->seed))
+ continue;
+ else if (unformat (input, "iter %d", &tm->iter))
+ continue;
+ else if (unformat (input, "verbose"))
+ tm->verbose = 1;
+ else if (unformat (input, "branch %d", &tmp))
+ {
+ if (tmp > 0)
+ tm->branching_factor = 1.0 / (f64) tmp;
+ else
+ fformat (stderr, "warning: branch = 0, ignored\n");
+ }
+ else
+ {
+ clib_error ("unknown input `%U'", format_unformat_error, input);
+ goto usage;
+ }
+ }
+ initialize_slist (tm);
+ run_test (tm);
+
+ return 0;
+
+usage:
+ fformat (stderr, "usage: test_slist seed <seed> iter <iter> [verbose]\n");
+ return 1;
+
+}
+
+#ifdef CLIB_UNIX
+int
+main (int argc, char *argv[])
+{
+ unformat_input_t i;
+ int ret;
+
+ clib_mem_init (0, (u64) 4 << 30);
+
+ unformat_init_command_line (&i, argv);
+ ret = test_slist_main (&i);
+ unformat_free (&i);
+
+ return ret;
+}
+#endif /* CLIB_UNIX */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/test_timing_wheel.c b/extras/deprecated/vppinfra/test_timing_wheel.c
new file mode 100644
index 00000000000..48020d520a0
--- /dev/null
+++ b/extras/deprecated/vppinfra/test_timing_wheel.c
@@ -0,0 +1,391 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+#include <vppinfra/bitmap.h>
+#include <vppinfra/error.h>
+#include <vppinfra/format.h>
+#include <vppinfra/pool.h>
+#include <vppinfra/random.h>
+#include <vppinfra/time.h>
+#include <vppinfra/timing_wheel.h>
+#include <vppinfra/zvec.h>
+
+#include <vppinfra/math.h>
+
+#if __GNUC__ < 4
+#define SQRT(a) a
+#else
+#define SQRT(a) sqrt(a)
+#endif
+
+typedef struct
+{
+ uword n_iter;
+
+ u32 n_events;
+ u32 seed;
+ u32 verbose;
+
+ /* Time is "synthetic" e.g. not taken from CPU timer. */
+ u32 synthetic_time;
+
+ clib_time_t time;
+ timing_wheel_t timing_wheel;
+
+ u64 *events;
+
+ f64 max_time;
+ f64 wait_time;
+
+ f64 total_iterate_time;
+ f64 time_iterate_start;
+
+ f64 time_per_status_update;
+ f64 time_next_status_update;
+} test_timing_wheel_main_t;
+
+typedef struct
+{
+ f64 dt;
+ f64 fraction;
+ u64 count;
+} test_timing_wheel_tmp_t;
+
+static void
+set_event (test_timing_wheel_main_t * tm, uword i)
+{
+ timing_wheel_t *w = &tm->timing_wheel;
+ u64 cpu_time;
+
+ cpu_time = w->current_time_index << w->log2_clocks_per_bin;
+ if (tm->synthetic_time)
+ cpu_time += random_u32 (&tm->seed) % tm->n_iter;
+ else
+ cpu_time +=
+ random_f64 (&tm->seed) * tm->max_time * tm->time.clocks_per_second;
+
+ timing_wheel_insert (w, cpu_time, i);
+ timing_wheel_validate (w);
+ tm->events[i] = cpu_time;
+}
+
+static int
+test_timing_wheel_tmp_cmp (void *a1, void *a2)
+{
+ test_timing_wheel_tmp_t *f1 = a1;
+ test_timing_wheel_tmp_t *f2 = a2;
+
+ return f1->dt < f2->dt ? -1 : (f1->dt > f2->dt ? +1 : 0);
+}
+
+clib_error_t *
+test_timing_wheel_main (unformat_input_t * input)
+{
+ clib_error_t *error = 0;
+ test_timing_wheel_main_t _tm, *tm = &_tm;
+ timing_wheel_t *w = &tm->timing_wheel;
+ uword iter, i;
+
+ clib_memset (tm, 0, sizeof (tm[0]));
+ tm->n_iter = 10;
+ tm->time_per_status_update = 0;
+ tm->n_events = 100;
+ tm->seed = 1;
+ tm->synthetic_time = 1;
+ tm->max_time = 1;
+ tm->wait_time = 1e-3;
+
+ w->validate = 0;
+ w->n_wheel_elt_time_bits = 32;
+
+ while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
+ {
+ if (unformat (input, "iter %wd", &tm->n_iter))
+ ;
+ else if (unformat (input, "events %d", &tm->n_events))
+ ;
+ else
+ if (unformat (input, "elt-time-bits %d", &w->n_wheel_elt_time_bits))
+ ;
+ else if (unformat (input, "seed %d", &tm->seed))
+ ;
+ else if (unformat (input, "verbose"))
+ tm->verbose = 1;
+ else if (unformat (input, "validate"))
+ w->validate = 1;
+
+ else if (unformat (input, "real-time"))
+ tm->synthetic_time = 0;
+ else if (unformat (input, "synthetic-time"))
+ tm->synthetic_time = 1;
+ else if (unformat (input, "max-time %f", &tm->max_time))
+ ;
+ else if (unformat (input, "wait-time %f", &tm->wait_time))
+ ;
+ else if (unformat (input, "iter-time %f", &tm->total_iterate_time))
+ ;
+ else if (unformat (input, "print %f", &tm->time_per_status_update))
+ ;
+
+ else
+ {
+ error = clib_error_create ("unknown input `%U'\n",
+ format_unformat_error, input);
+ goto done;
+ }
+ }
+
+ if (!tm->seed)
+ tm->seed = random_default_seed ();
+
+ clib_time_init (&tm->time);
+
+ if (tm->synthetic_time)
+ {
+ w->min_sched_time = tm->time.seconds_per_clock;
+ w->max_sched_time = w->min_sched_time * 256;
+ timing_wheel_init (w, 0, tm->time.clocks_per_second);
+ }
+ else
+ {
+ timing_wheel_init (w, clib_cpu_time_now (), tm->time.clocks_per_second);
+ }
+
+ clib_warning ("iter %wd, events %d, seed %u, %U",
+ tm->n_iter, tm->n_events, tm->seed,
+ format_timing_wheel, &tm->timing_wheel, /* verbose */ 0);
+
+ /* Make some events. */
+ vec_resize (tm->events, tm->n_events);
+ for (i = 0; i < vec_len (tm->events); i++)
+ set_event (tm, i);
+
+ {
+ u32 *expired = 0;
+ f64 ave_error = 0;
+ f64 rms_error = 0;
+ f64 max_error = 0, min_error = 1e30;
+ u32 *error_hist = 0;
+ uword n_expired = 0;
+ uword *expired_bitmap[2] = { 0 };
+ uword n_events_in_wheel = vec_len (tm->events);
+
+ vec_resize (expired, 32);
+ vec_resize (error_hist, 1024);
+
+ tm->time_iterate_start = clib_time_now (&tm->time);
+ tm->time_next_status_update =
+ tm->time_iterate_start + tm->time_per_status_update;
+
+ if (tm->total_iterate_time != 0)
+ tm->n_iter = ~0;
+
+ for (iter = 0; iter < tm->n_iter || n_events_in_wheel > 0; iter++)
+ {
+ u64 cpu_time, min_next_time[2];
+
+ if (tm->synthetic_time)
+ cpu_time = iter << w->log2_clocks_per_bin;
+ else
+ cpu_time = clib_cpu_time_now ();
+
+ _vec_len (expired) = 0;
+ expired =
+ timing_wheel_advance (w, cpu_time, expired, &min_next_time[0]);
+ timing_wheel_validate (w);
+
+ /* Update bitmap of expired events. */
+ if (w->validate)
+ {
+ for (i = 0; i < vec_len (tm->events); i++)
+ {
+ uword is_expired;
+
+ is_expired =
+ (cpu_time >> w->log2_clocks_per_bin) >=
+ (tm->events[i] >> w->log2_clocks_per_bin);
+ expired_bitmap[0] =
+ clib_bitmap_set (expired_bitmap[0], i, is_expired);
+
+ /* Validate min next time. */
+ if (is_expired)
+ ASSERT (min_next_time[0] > tm->events[i]);
+ else
+ ASSERT (min_next_time[0] <= tm->events[i]);
+ }
+ }
+
+ n_expired += vec_len (expired);
+ for (i = 0; i < vec_len (expired); i++)
+ {
+ word j, idt;
+ i64 dt_cpu;
+ f64 fdt_cpu;
+
+ j = expired[i];
+ expired_bitmap[1] = clib_bitmap_ori (expired_bitmap[1], j);
+
+ dt_cpu = cpu_time - tm->events[j];
+
+ /* Event must be scheduled in correct bin. */
+ if (tm->synthetic_time)
+ ASSERT (dt_cpu >= 0 && dt_cpu <= (1 << w->log2_clocks_per_bin));
+
+ fdt_cpu = dt_cpu * tm->time.seconds_per_clock;
+
+ ave_error += fdt_cpu;
+ rms_error += fdt_cpu * fdt_cpu;
+
+ if (fdt_cpu > max_error)
+ max_error = fdt_cpu;
+ if (fdt_cpu < min_error)
+ min_error = fdt_cpu;
+
+ idt =
+ (cpu_time >> w->log2_clocks_per_bin) -
+ (tm->events[j] >> w->log2_clocks_per_bin);
+ idt = zvec_signed_to_unsigned (idt);
+ vec_validate (error_hist, idt);
+ error_hist[idt] += 1;
+ }
+
+ if (w->validate)
+ for (i = 0; i < vec_len (tm->events); i++)
+ {
+ int is_expired = clib_bitmap_get (expired_bitmap[0], i);
+ int is_expired_w = clib_bitmap_get (expired_bitmap[1], i);
+ ASSERT (is_expired == is_expired_w);
+ }
+
+ min_next_time[1] = ~0;
+ for (i = 0; i < vec_len (tm->events); i++)
+ {
+ if (!clib_bitmap_get (expired_bitmap[1], i))
+ min_next_time[1] = clib_min (min_next_time[1], tm->events[i]);
+ }
+ if (min_next_time[0] != min_next_time[1])
+ clib_error ("min next time wrong 0x%Lx != 0x%Lx", min_next_time[0],
+ min_next_time[1]);
+
+ if (tm->time_per_status_update != 0
+ && clib_time_now (&tm->time) >= tm->time_next_status_update)
+ {
+ f64 ave = 0, rms = 0;
+
+ tm->time_next_status_update += tm->time_per_status_update;
+ if (n_expired > 0)
+ {
+ ave = ave_error / n_expired;
+ rms = SQRT (rms_error / n_expired - ave * ave);
+ }
+
+ clib_warning
+ ("%12wd iter done %10wd expired; ave. error %.4e +- %.4e, range %.4e %.4e",
+ iter, n_expired, ave, rms, min_error, max_error);
+ }
+
+ if (tm->total_iterate_time != 0
+ && (clib_time_now (&tm->time) - tm->time_iterate_start
+ >= tm->total_iterate_time))
+ tm->n_iter = iter;
+
+ /* Add new events to wheel to replace expired ones. */
+ n_events_in_wheel -= vec_len (expired);
+ if (iter < tm->n_iter)
+ {
+ for (i = 0; i < vec_len (expired); i++)
+ {
+ uword j = expired[i];
+ set_event (tm, j);
+ expired_bitmap[1] =
+ clib_bitmap_andnoti (expired_bitmap[1], j);
+ }
+ n_events_in_wheel += vec_len (expired);
+ }
+ }
+
+ ave_error /= n_expired;
+ rms_error = SQRT (rms_error / n_expired - ave_error * ave_error);
+
+ clib_warning
+ ("%wd iter done %wd expired; ave. error %.4e +- %.4e, range %.4e %.4e",
+ 1 + iter, n_expired, ave_error, rms_error, min_error, max_error);
+
+ {
+ test_timing_wheel_tmp_t *fs, *f;
+ f64 total_fraction;
+
+ fs = 0;
+ for (i = 0; i < vec_len (error_hist); i++)
+ {
+ if (error_hist[i] == 0)
+ continue;
+ vec_add2 (fs, f, 1);
+ f->dt =
+ (((i64) zvec_unsigned_to_signed (i) << w->log2_clocks_per_bin) *
+ tm->time.seconds_per_clock);
+ f->fraction = (f64) error_hist[i] / (f64) n_expired;
+ f->count = error_hist[i];
+ }
+
+ vec_sort_with_function (fs, test_timing_wheel_tmp_cmp);
+
+ total_fraction = 0;
+ vec_foreach (f, fs)
+ {
+ total_fraction += f->fraction;
+ if (f == fs)
+ fformat (stdout, "%=12s %=16s %=16s %s\n", "Error max", "Fraction",
+ "Total", "Count");
+ fformat (stdout, "%12.4e %16.4f%% %16.4f%% %Ld\n", f->dt,
+ f->fraction * 100, total_fraction * 100, f->count);
+ }
+ }
+
+ clib_warning ("%U", format_timing_wheel, w, /* verbose */ 1);
+ }
+
+done:
+ return error;
+}
+
+#ifdef CLIB_UNIX
+int
+main (int argc, char *argv[])
+{
+ unformat_input_t i;
+ clib_error_t *error;
+
+ clib_mem_init (0, 64ULL << 20);
+
+ unformat_init_command_line (&i, argv);
+ error = test_timing_wheel_main (&i);
+ unformat_free (&i);
+ if (error)
+ {
+ clib_error_report (error);
+ return 1;
+ }
+ else
+ return 0;
+}
+#endif /* CLIB_UNIX */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/test_vhash.c b/extras/deprecated/vppinfra/test_vhash.c
new file mode 100644
index 00000000000..594d46c38df
--- /dev/null
+++ b/extras/deprecated/vppinfra/test_vhash.c
@@ -0,0 +1,759 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ Copyright (c) 2010 Eliot Dresselhaus
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice shall be
+ included in all copies or substantial portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+*/
+
+#if 0
+#ifdef __OPTIMIZE__
+#undef CLIB_DEBUG
+#endif
+#endif
+
+#include <vppinfra/bitmap.h>
+#include <vppinfra/error.h>
+#include <vppinfra/os.h>
+#include <vppinfra/random.h>
+#include <vppinfra/time.h>
+#include <vppinfra/vhash.h>
+
+#ifdef CLIB_HAVE_VEC128
+
+typedef struct
+{
+ u32 n_iter;
+ u32 seed;
+ u32 verbose;
+ u32 n_keys;
+ u32 log2_size;
+ u32 n_key_u32;
+
+ u32 n_vectors_div_4;
+ u32 n_vectors_mod_4;
+
+ u32 *keys;
+ u32 *results;
+
+ u32 *vhash_get_key_indices;
+ u32 *vhash_get_results;
+
+ u32 *vhash_key_indices;
+ u32 *vhash_results;
+
+ vhash_t vhash;
+
+ uword **key_hash;
+
+ struct
+ {
+ u64 n_clocks;
+ u64 n_vectors;
+ u64 n_calls;
+ } get_stats, set_stats, unset_stats;
+} test_vhash_main_t;
+
+always_inline u32
+test_vhash_key_gather (void *_tm, u32 vi, u32 wi, u32 n_key_u32s)
+{
+ test_vhash_main_t *tm = _tm;
+ ASSERT (n_key_u32s == tm->n_key_u32);
+ ASSERT (wi < n_key_u32s);
+ vi = vec_elt (tm->vhash_key_indices, vi);
+ return vec_elt (tm->keys, vi * n_key_u32s + wi);
+}
+
+always_inline u32x4
+test_vhash_4key_gather (void *_tm, u32 vi, u32 wi, u32 n_key_u32s)
+{
+ test_vhash_main_t *tm = _tm;
+ u32 *p;
+ u32x4_union_t x;
+
+ ASSERT (n_key_u32s == tm->n_key_u32);
+ ASSERT (wi < n_key_u32s);
+
+ p = vec_elt_at_index (tm->vhash_key_indices, vi + 0);
+ x.as_u32[0] = tm->keys[p[0] * n_key_u32s + wi];
+ x.as_u32[1] = tm->keys[p[1] * n_key_u32s + wi];
+ x.as_u32[2] = tm->keys[p[2] * n_key_u32s + wi];
+ x.as_u32[3] = tm->keys[p[3] * n_key_u32s + wi];
+ return x.as_u32x4;
+}
+
+always_inline u32
+test_vhash_get_result (void *_tm,
+ u32 vector_index, u32 result_index, u32 n_key_u32s)
+{
+ test_vhash_main_t *tm = _tm;
+ u32 *p = vec_elt_at_index (tm->vhash_results, vector_index);
+ p[0] = result_index;
+ return result_index;
+}
+
+always_inline u32x4
+test_vhash_get_4result (void *_tm,
+ u32 vector_index, u32x4 results, u32 n_key_u32s)
+{
+ test_vhash_main_t *tm = _tm;
+ u32 *p = vec_elt_at_index (tm->vhash_results, vector_index);
+ *(u32x4 *) p = results;
+ return results;
+}
+
+always_inline u32
+test_vhash_set_result (void *_tm,
+ u32 vector_index, u32 old_result, u32 n_key_u32s)
+{
+ test_vhash_main_t *tm = _tm;
+ u32 *p = vec_elt_at_index (tm->vhash_results, vector_index);
+ u32 new_result = p[0];
+ p[0] = old_result;
+ return new_result;
+}
+
+always_inline u32
+test_vhash_unset_result (void *_tm, u32 i, u32 old_result, u32 n_key_u32s)
+{
+ test_vhash_main_t *tm = _tm;
+ u32 *p = vec_elt_at_index (tm->vhash_results, i);
+ p[0] = old_result;
+ return 0;
+}
+
+#define _(N_KEY_U32) \
+ always_inline u32 \
+ test_vhash_key_gather_##N_KEY_U32 (void * _tm, u32 vi, u32 i) \
+ { return test_vhash_key_gather (_tm, vi, i, N_KEY_U32); } \
+ \
+ always_inline u32x4 \
+ test_vhash_key_gather_4_##N_KEY_U32 (void * _tm, u32 vi, u32 i) \
+ { return test_vhash_4key_gather (_tm, vi, i, N_KEY_U32); } \
+ \
+ clib_pipeline_stage \
+ (test_vhash_gather_keys_stage_##N_KEY_U32, \
+ test_vhash_main_t *, tm, i, \
+ { \
+ vhash_gather_4key_stage \
+ (&tm->vhash, \
+ /* vector_index */ i, \
+ test_vhash_key_gather_4_##N_KEY_U32, \
+ tm, \
+ N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage_no_inline \
+ (test_vhash_gather_keys_mod_stage_##N_KEY_U32, \
+ test_vhash_main_t *, tm, i, \
+ { \
+ vhash_gather_key_stage \
+ (&tm->vhash, \
+ /* vector_index */ tm->n_vectors_div_4, \
+ /* n_vectors */ tm->n_vectors_mod_4, \
+ test_vhash_key_gather_##N_KEY_U32, \
+ tm, \
+ N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage \
+ (test_vhash_hash_finalize_stage_##N_KEY_U32, \
+ test_vhash_main_t *, tm, i, \
+ { \
+ vhash_finalize_stage (&tm->vhash, i, N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage_no_inline \
+ (test_vhash_hash_finalize_mod_stage_##N_KEY_U32, \
+ test_vhash_main_t *, tm, i, \
+ { \
+ vhash_finalize_stage (&tm->vhash, tm->n_vectors_div_4, N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage \
+ (test_vhash_get_stage_##N_KEY_U32, \
+ test_vhash_main_t *, tm, i, \
+ { \
+ vhash_get_4_stage (&tm->vhash, \
+ /* vector_index */ i, \
+ test_vhash_get_4result, \
+ tm, N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage_no_inline \
+ (test_vhash_get_mod_stage_##N_KEY_U32, \
+ test_vhash_main_t *, tm, i, \
+ { \
+ vhash_get_stage (&tm->vhash, \
+ /* vector_index */ tm->n_vectors_div_4, \
+ /* n_vectors */ tm->n_vectors_mod_4, \
+ test_vhash_get_result, \
+ tm, N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage \
+ (test_vhash_set_stage_##N_KEY_U32, \
+ test_vhash_main_t *, tm, i, \
+ { \
+ vhash_set_stage (&tm->vhash, \
+ /* vector_index */ i, \
+ /* n_vectors */ VECTOR_WORD_TYPE_LEN (u32), \
+ test_vhash_set_result, \
+ tm, N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage_no_inline \
+ (test_vhash_set_mod_stage_##N_KEY_U32, \
+ test_vhash_main_t *, tm, i, \
+ { \
+ vhash_set_stage (&tm->vhash, \
+ /* vector_index */ tm->n_vectors_div_4, \
+ /* n_vectors */ tm->n_vectors_mod_4, \
+ test_vhash_set_result, \
+ tm, N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage \
+ (test_vhash_unset_stage_##N_KEY_U32, \
+ test_vhash_main_t *, tm, i, \
+ { \
+ vhash_unset_stage (&tm->vhash, \
+ /* vector_index */ i, \
+ /* n_vectors */ VECTOR_WORD_TYPE_LEN (u32), \
+ test_vhash_unset_result, \
+ tm, N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage_no_inline \
+ (test_vhash_unset_mod_stage_##N_KEY_U32, \
+ test_vhash_main_t *, tm, i, \
+ { \
+ vhash_unset_stage (&tm->vhash, \
+ /* vector_index */ tm->n_vectors_div_4, \
+ /* n_vectors */ tm->n_vectors_mod_4, \
+ test_vhash_unset_result, \
+ tm, N_KEY_U32); \
+ })
+
+_(1);
+_(2);
+_(3);
+_(4);
+_(5);
+_(6);
+
+#undef _
+
+#define _(N_KEY_U32) \
+ clib_pipeline_stage \
+ (test_vhash_hash_mix_stage_##N_KEY_U32, \
+ test_vhash_main_t *, tm, i, \
+ { \
+ vhash_mix_stage (&tm->vhash, i, N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage_no_inline \
+ (test_vhash_hash_mix_mod_stage_##N_KEY_U32, \
+ test_vhash_main_t *, tm, i, \
+ { \
+ vhash_mix_stage (&tm->vhash, tm->n_vectors_div_4, N_KEY_U32); \
+ })
+
+_(4);
+_(5);
+_(6);
+
+#undef _
+
+typedef enum
+{
+ GET, SET, UNSET,
+} test_vhash_op_t;
+
+static void
+test_vhash_op (test_vhash_main_t * tm,
+ u32 * key_indices,
+ u32 * results, uword n_keys, test_vhash_op_t op)
+{
+ vhash_validate_sizes (&tm->vhash, tm->n_key_u32, n_keys);
+
+ tm->vhash_results = results;
+ tm->vhash_key_indices = key_indices;
+ tm->n_vectors_div_4 = n_keys / 4;
+ tm->n_vectors_mod_4 = n_keys % 4;
+
+ if (tm->n_vectors_div_4 > 0)
+ {
+ switch (tm->n_key_u32)
+ {
+ default:
+ ASSERT (0);
+ break;
+
+#define _(N_KEY_U32) \
+ case N_KEY_U32: \
+ if (op == GET) \
+ clib_pipeline_run_3_stage \
+ (tm->n_vectors_div_4, \
+ tm, \
+ test_vhash_gather_keys_stage_##N_KEY_U32, \
+ test_vhash_hash_finalize_stage_##N_KEY_U32, \
+ test_vhash_get_stage_##N_KEY_U32); \
+ else if (op == SET) \
+ clib_pipeline_run_3_stage \
+ (tm->n_vectors_div_4, \
+ tm, \
+ test_vhash_gather_keys_stage_##N_KEY_U32, \
+ test_vhash_hash_finalize_stage_##N_KEY_U32, \
+ test_vhash_set_stage_##N_KEY_U32); \
+ else \
+ clib_pipeline_run_3_stage \
+ (tm->n_vectors_div_4, \
+ tm, \
+ test_vhash_gather_keys_stage_##N_KEY_U32, \
+ test_vhash_hash_finalize_stage_##N_KEY_U32, \
+ test_vhash_unset_stage_##N_KEY_U32); \
+ break;
+
+ _(1);
+ _(2);
+ _(3);
+
+#undef _
+
+#define _(N_KEY_U32) \
+ case N_KEY_U32: \
+ if (op == GET) \
+ clib_pipeline_run_4_stage \
+ (tm->n_vectors_div_4, \
+ tm, \
+ test_vhash_gather_keys_stage_##N_KEY_U32, \
+ test_vhash_hash_mix_stage_##N_KEY_U32, \
+ test_vhash_hash_finalize_stage_##N_KEY_U32, \
+ test_vhash_get_stage_##N_KEY_U32); \
+ else if (op == SET) \
+ clib_pipeline_run_4_stage \
+ (tm->n_vectors_div_4, \
+ tm, \
+ test_vhash_gather_keys_stage_##N_KEY_U32, \
+ test_vhash_hash_mix_stage_##N_KEY_U32, \
+ test_vhash_hash_finalize_stage_##N_KEY_U32, \
+ test_vhash_set_stage_##N_KEY_U32); \
+ else \
+ clib_pipeline_run_4_stage \
+ (tm->n_vectors_div_4, \
+ tm, \
+ test_vhash_gather_keys_stage_##N_KEY_U32, \
+ test_vhash_hash_mix_stage_##N_KEY_U32, \
+ test_vhash_hash_finalize_stage_##N_KEY_U32, \
+ test_vhash_unset_stage_##N_KEY_U32); \
+ break;
+
+ _(4);
+ _(5);
+ _(6);
+
+#undef _
+ }
+ }
+
+
+ if (tm->n_vectors_mod_4 > 0)
+ {
+ switch (tm->n_key_u32)
+ {
+ default:
+ ASSERT (0);
+ break;
+
+#define _(N_KEY_U32) \
+ case N_KEY_U32: \
+ if (op == GET) \
+ clib_pipeline_run_3_stage \
+ (1, \
+ tm, \
+ test_vhash_gather_keys_mod_stage_##N_KEY_U32, \
+ test_vhash_hash_finalize_mod_stage_##N_KEY_U32, \
+ test_vhash_get_mod_stage_##N_KEY_U32); \
+ else if (op == SET) \
+ clib_pipeline_run_3_stage \
+ (1, \
+ tm, \
+ test_vhash_gather_keys_mod_stage_##N_KEY_U32, \
+ test_vhash_hash_finalize_mod_stage_##N_KEY_U32, \
+ test_vhash_set_mod_stage_##N_KEY_U32); \
+ else \
+ clib_pipeline_run_3_stage \
+ (1, \
+ tm, \
+ test_vhash_gather_keys_mod_stage_##N_KEY_U32, \
+ test_vhash_hash_finalize_mod_stage_##N_KEY_U32, \
+ test_vhash_unset_mod_stage_##N_KEY_U32); \
+ break;
+
+ _(1);
+ _(2);
+ _(3);
+
+#undef _
+
+#define _(N_KEY_U32) \
+ case N_KEY_U32: \
+ if (op == GET) \
+ clib_pipeline_run_4_stage \
+ (1, \
+ tm, \
+ test_vhash_gather_keys_mod_stage_##N_KEY_U32, \
+ test_vhash_hash_mix_mod_stage_##N_KEY_U32, \
+ test_vhash_hash_finalize_mod_stage_##N_KEY_U32, \
+ test_vhash_get_mod_stage_##N_KEY_U32); \
+ else if (op == SET) \
+ clib_pipeline_run_4_stage \
+ (1, \
+ tm, \
+ test_vhash_gather_keys_mod_stage_##N_KEY_U32, \
+ test_vhash_hash_mix_mod_stage_##N_KEY_U32, \
+ test_vhash_hash_finalize_mod_stage_##N_KEY_U32, \
+ test_vhash_set_mod_stage_##N_KEY_U32); \
+ else \
+ clib_pipeline_run_4_stage \
+ (1, \
+ tm, \
+ test_vhash_gather_keys_mod_stage_##N_KEY_U32, \
+ test_vhash_hash_mix_mod_stage_##N_KEY_U32, \
+ test_vhash_hash_finalize_mod_stage_##N_KEY_U32, \
+ test_vhash_unset_mod_stage_##N_KEY_U32); \
+ break;
+
+ _(4);
+ _(5);
+ _(6);
+
+#undef _
+ }
+ }
+}
+
+int
+test_vhash_main (unformat_input_t * input)
+{
+ clib_error_t *error = 0;
+ test_vhash_main_t _tm, *tm = &_tm;
+ vhash_t *vh = &tm->vhash;
+ uword i, j;
+
+ clib_memset (tm, 0, sizeof (tm[0]));
+ tm->n_iter = 100;
+ tm->seed = 1;
+ tm->n_keys = 1;
+ tm->n_key_u32 = 1;
+ tm->log2_size = 8;
+ tm->verbose = 0;
+
+ while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
+ {
+ if (unformat (input, "iter %d", &tm->n_iter))
+ ;
+ else if (unformat (input, "seed %d", &tm->seed))
+ ;
+ else if (unformat (input, "n-keys %d", &tm->n_keys))
+ ;
+ else if (unformat (input, "log2-size %d", &tm->log2_size))
+ ;
+ else if (unformat (input, "key-words %d", &tm->n_key_u32))
+ ;
+ else if (unformat (input, "verbose %=", &tm->verbose, 1))
+ ;
+ else
+ {
+ error = clib_error_create ("unknown input `%U'\n",
+ format_unformat_error, input);
+ goto done;
+ }
+ }
+
+ if (tm->seed == 0)
+ tm->seed = random_default_seed ();
+
+ clib_warning ("iter %d seed %d n-keys %d log2-size %d key-words %d",
+ tm->n_iter, tm->seed, tm->n_keys, tm->log2_size,
+ tm->n_key_u32);
+
+ {
+ u32 seeds[3];
+ seeds[0] = seeds[1] = seeds[2] = 0xdeadbeef;
+ vhash_init (vh, tm->log2_size, tm->n_key_u32, seeds);
+ }
+
+ /* Choose unique keys. */
+ vec_resize (tm->keys, tm->n_keys * tm->n_key_u32);
+ vec_resize (tm->key_hash, tm->n_key_u32);
+ for (i = j = 0; i < vec_len (tm->keys); i++, j++)
+ {
+ j = j == tm->n_key_u32 ? 0 : j;
+ do
+ {
+ tm->keys[i] = random_u32 (&tm->seed);
+ }
+ while (hash_get (tm->key_hash[j], tm->keys[i]));
+ hash_set (tm->key_hash[j], tm->keys[i], 0);
+ }
+
+ vec_resize (tm->results, tm->n_keys);
+ for (i = 0; i < vec_len (tm->results); i++)
+ {
+ do
+ {
+ tm->results[i] = random_u32 (&tm->seed);
+ }
+ while (tm->results[i] == ~0);
+ }
+
+ vec_resize_aligned (tm->vhash_get_results, tm->n_keys,
+ CLIB_CACHE_LINE_BYTES);
+ vec_clone (tm->vhash_get_key_indices, tm->results);
+ for (i = 0; i < vec_len (tm->vhash_get_key_indices); i++)
+ tm->vhash_get_key_indices[i] = i;
+
+ {
+ uword *is_set_bitmap = 0;
+ uword *to_set_bitmap = 0;
+ uword *to_unset_bitmap = 0;
+ u32 *to_set = 0, *to_unset = 0;
+ u32 *to_set_results = 0, *to_unset_results = 0;
+ u64 t[2];
+
+ for (i = 0; i < tm->n_iter; i++)
+ {
+ vec_reset_length (to_set);
+ vec_reset_length (to_unset);
+ vec_reset_length (to_set_results);
+ vec_reset_length (to_unset_results);
+
+ do
+ {
+ to_set_bitmap = clib_bitmap_random (to_set_bitmap,
+ tm->n_keys, &tm->seed);
+ }
+ while (clib_bitmap_is_zero (to_set_bitmap));
+ to_unset_bitmap = clib_bitmap_dup_and (to_set_bitmap, is_set_bitmap);
+ to_set_bitmap = clib_bitmap_andnot (to_set_bitmap, to_unset_bitmap);
+
+ /* *INDENT-OFF* */
+ clib_bitmap_foreach (j, to_set_bitmap, ({
+ vec_add1 (to_set, j);
+ vec_add1 (to_set_results, tm->results[j]);
+ }));
+ /* *INDENT-ON* */
+ /* *INDENT-OFF* */
+ clib_bitmap_foreach (j, to_unset_bitmap, ({
+ vec_add1 (to_unset, j);
+ vec_add1 (to_unset_results, 0xdeadbeef);
+ }));
+ /* *INDENT-ON* */
+
+ if (vec_len (to_set) > 0)
+ {
+ t[0] = clib_cpu_time_now ();
+ test_vhash_op (tm, to_set, to_set_results, vec_len (to_set), SET);
+ t[1] = clib_cpu_time_now ();
+ tm->set_stats.n_clocks += t[1] - t[0];
+ tm->set_stats.n_vectors += vec_len (to_set);
+ tm->set_stats.n_calls += 1;
+ is_set_bitmap = clib_bitmap_or (is_set_bitmap, to_set_bitmap);
+ }
+
+ t[0] = clib_cpu_time_now ();
+ test_vhash_op (tm, tm->vhash_get_key_indices,
+ tm->vhash_get_results,
+ vec_len (tm->vhash_get_key_indices), GET);
+ t[1] = clib_cpu_time_now ();
+ tm->get_stats.n_clocks += t[1] - t[0];
+ tm->get_stats.n_vectors += vec_len (tm->vhash_get_key_indices);
+ tm->get_stats.n_calls += 1;
+
+ for (j = 0; j < vec_len (tm->vhash_get_results); j++)
+ {
+ u32 r0 = tm->vhash_get_results[j];
+ u32 r1 = tm->results[j];
+ if (clib_bitmap_get (is_set_bitmap, j))
+ {
+ if (r0 != r1)
+ os_panic ();
+ }
+ else
+ {
+ if (r0 != ~0)
+ os_panic ();
+ }
+ }
+
+ if (vh->n_elts != clib_bitmap_count_set_bits (is_set_bitmap))
+ os_panic ();
+
+ if (vec_len (to_unset) > 0)
+ {
+ t[0] = clib_cpu_time_now ();
+ test_vhash_op (tm, to_unset, to_unset_results,
+ vec_len (to_unset), UNSET);
+ t[1] = clib_cpu_time_now ();
+ tm->unset_stats.n_clocks += t[1] - t[0];
+ tm->unset_stats.n_vectors += vec_len (to_unset);
+ tm->unset_stats.n_calls += 1;
+ is_set_bitmap =
+ clib_bitmap_andnot (is_set_bitmap, to_unset_bitmap);
+ }
+
+ t[0] = clib_cpu_time_now ();
+ test_vhash_op (tm, tm->vhash_get_key_indices,
+ tm->vhash_get_results,
+ vec_len (tm->vhash_get_key_indices), GET);
+ t[1] = clib_cpu_time_now ();
+ tm->get_stats.n_clocks += t[1] - t[0];
+ tm->get_stats.n_vectors += vec_len (tm->vhash_get_key_indices);
+ tm->get_stats.n_calls += 1;
+
+ for (j = 0; j < vec_len (tm->vhash_get_results); j++)
+ {
+ u32 r0 = tm->vhash_get_results[j];
+ u32 r1 = tm->results[j];
+ if (clib_bitmap_get (is_set_bitmap, j))
+ {
+ if (r0 != r1)
+ os_panic ();
+ }
+ else
+ {
+ if (r0 != ~0)
+ os_panic ();
+ }
+ }
+
+ if (vh->n_elts != clib_bitmap_count_set_bits (is_set_bitmap))
+ os_panic ();
+ }
+
+ vhash_resize (vh, tm->log2_size + 1);
+
+ test_vhash_op (tm, tm->vhash_get_key_indices,
+ tm->vhash_get_results,
+ vec_len (tm->vhash_get_key_indices), GET);
+
+ for (j = 0; j < vec_len (tm->vhash_get_results); j++)
+ {
+ u32 r0 = tm->vhash_get_results[j];
+ u32 r1 = tm->results[j];
+ if (clib_bitmap_get (is_set_bitmap, j))
+ {
+ if (r0 != r1)
+ os_panic ();
+ }
+ else
+ {
+ if (r0 != ~0)
+ os_panic ();
+ }
+ }
+
+ if (vh->n_elts != clib_bitmap_count_set_bits (is_set_bitmap))
+ os_panic ();
+ }
+
+ {
+ clib_time_t ct;
+
+ clib_time_init (&ct);
+
+ clib_warning ("%.4e clocks/get %.4e gets/call %.4e gets/sec",
+ (f64) tm->get_stats.n_clocks /
+ (f64) tm->get_stats.n_vectors,
+ (f64) tm->get_stats.n_vectors / (f64) tm->get_stats.n_calls,
+ (f64) tm->get_stats.n_vectors /
+ (f64) (tm->get_stats.n_clocks * ct.seconds_per_clock));
+ if (tm->set_stats.n_calls > 0)
+ clib_warning ("%.4e clocks/set %.4e sets/call %.4e sets/sec",
+ (f64) tm->set_stats.n_clocks /
+ (f64) tm->set_stats.n_vectors,
+ (f64) tm->set_stats.n_vectors /
+ (f64) tm->set_stats.n_calls,
+ (f64) tm->set_stats.n_vectors /
+ (f64) (tm->set_stats.n_clocks * ct.seconds_per_clock));
+ if (tm->unset_stats.n_calls > 0)
+ clib_warning ("%.4e clocks/unset %.4e unsets/call %.4e unsets/sec",
+ (f64) tm->unset_stats.n_clocks /
+ (f64) tm->unset_stats.n_vectors,
+ (f64) tm->unset_stats.n_vectors /
+ (f64) tm->unset_stats.n_calls,
+ (f64) tm->unset_stats.n_vectors /
+ (f64) (tm->unset_stats.n_clocks * ct.seconds_per_clock));
+ }
+
+done:
+ if (error)
+ clib_error_report (error);
+ return 0;
+}
+
+#endif /* CLIB_HAVE_VEC128 */
+
+#ifndef CLIB_HAVE_VEC128
+int
+test_vhash_main (unformat_input_t * input)
+{
+ clib_error ("compiled without vector support");
+ return 0;
+}
+#endif
+
+#ifdef CLIB_UNIX
+int
+main (int argc, char *argv[])
+{
+ unformat_input_t i;
+ int r;
+
+ clib_mem_init (0, 64ULL << 20);
+
+ unformat_init_command_line (&i, argv);
+ r = test_vhash_main (&i);
+ unformat_free (&i);
+ return r;
+}
+#endif
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/test_zvec.c b/extras/deprecated/vppinfra/test_zvec.c
new file mode 100644
index 00000000000..7d777fabf83
--- /dev/null
+++ b/extras/deprecated/vppinfra/test_zvec.c
@@ -0,0 +1,119 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ Copyright (c) 2005 Eliot Dresselhaus
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice shall be
+ included in all copies or substantial portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+*/
+
+#include <vppinfra/zvec.h>
+#include <vppinfra/format.h>
+#include <vppinfra/random.h>
+
+static int verbose;
+#define if_verbose(format,args...) \
+ if (verbose) { clib_warning(format, ## args); }
+
+int
+test_zvec_main (unformat_input_t * input)
+{
+ uword n_iterations;
+ uword i;
+ u32 seed;
+
+ n_iterations = 1024;
+ seed = 0;
+
+ while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT)
+ {
+ if (0 == unformat (input, "iter %d", &n_iterations)
+ && 0 == unformat (input, "seed %d", &seed))
+ clib_error ("unknown input `%U'", format_unformat_error, input);
+ }
+
+ if_verbose ("%d iterations, seed %d\n", n_iterations, seed);
+
+ for (i = 0; i < n_iterations; i++)
+ {
+ uword coding, data, d[2], limit, n_zdata_bits[2];
+
+ if (seed)
+ coding = random_u32 (&seed);
+ else
+ coding = i;
+
+ limit = coding - 1;
+ if (limit > (1 << 16))
+ limit = 1 << 16;
+ for (data = 0; data <= limit; data++)
+ {
+ d[0] = zvec_encode (coding, data, &n_zdata_bits[0]);
+
+ if (coding != 0)
+ ASSERT ((d[0] >> n_zdata_bits[0]) == 0);
+
+ d[1] = zvec_decode (coding, d[0], &n_zdata_bits[1]);
+ ASSERT (data == d[1]);
+
+ ASSERT (n_zdata_bits[0] == n_zdata_bits[1]);
+ }
+ }
+
+ return 0;
+}
+
+#ifdef CLIB_UNIX
+int
+main (int argc, char *argv[])
+{
+ unformat_input_t i;
+ int ret;
+
+ clib_mem_init (0, 64ULL << 20);
+
+ verbose = (argc > 1);
+ unformat_init_command_line (&i, argv);
+ ret = test_zvec_main (&i);
+ unformat_free (&i);
+
+ return ret;
+}
+#endif /* CLIB_UNIX */
+
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/unix-kelog.c b/extras/deprecated/vppinfra/unix-kelog.c
new file mode 100644
index 00000000000..88428ee8f2e
--- /dev/null
+++ b/extras/deprecated/vppinfra/unix-kelog.c
@@ -0,0 +1,415 @@
+/*
+ Copyright (c) 2010 Cisco and/or its affiliates.
+
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+*/
+
+#include <vppinfra/error.h>
+#include <vppinfra/unix.h>
+#include <vppinfra/elog.h>
+#include <vppinfra/format.h>
+#include <vppinfra/os.h>
+
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
+#include <time.h>
+
+typedef enum
+{
+ RUNNING = 0,
+ WAKEUP,
+} sched_event_type_t;
+
+typedef struct
+{
+ u32 cpu;
+ u8 *task;
+ u32 pid;
+ f64 timestamp;
+ sched_event_type_t type;
+} sched_event_t;
+
+void
+kelog_init (elog_main_t * em, char *kernel_tracer, u32 n_events)
+{
+ int enable_fd, current_tracer_fd, data_fd;
+ int len;
+ struct timespec ts, ts2;
+ char *trace_enable = "/debug/tracing/tracing_enabled";
+ char *current_tracer = "/debug/tracing/current_tracer";
+ char *trace_data = "/debug/tracing/trace";
+ f64 realtime, monotonic;
+ f64 freq, secs_per_clock;
+
+ ASSERT (kernel_tracer);
+
+ /*$$$$ fixme */
+ n_events = 1 << 18;
+
+ /* init first so we won't hurt ourselves if we bail */
+ elog_init (em, n_events);
+
+ enable_fd = open (trace_enable, O_RDWR);
+ if (enable_fd < 0)
+ {
+ clib_warning ("Couldn't open %s", trace_enable);
+ return;
+ }
+ /* disable kernel tracing */
+ if (write (enable_fd, "0\n", 2) != 2)
+ {
+ clib_unix_warning ("disable tracing");
+ close (enable_fd);
+ return;
+ }
+
+ /*
+ * open + clear the data buffer.
+ * see .../linux/kernel/trace/trace.c:tracing_open()
+ */
+ data_fd = open (trace_data, O_RDWR | O_TRUNC);
+ if (data_fd < 0)
+ {
+ clib_warning ("Couldn't open+clear %s", trace_data);
+ return;
+ }
+ close (data_fd);
+
+ /* configure tracing */
+ current_tracer_fd = open (current_tracer, O_RDWR);
+
+ if (current_tracer_fd < 0)
+ {
+ clib_warning ("Couldn't open %s", current_tracer);
+ close (enable_fd);
+ return;
+ }
+
+ len = strlen (kernel_tracer);
+
+ if (write (current_tracer_fd, kernel_tracer, len) != len)
+ {
+ clib_unix_warning ("configure trace");
+ close (current_tracer_fd);
+ close (enable_fd);
+ return;
+ }
+
+ close (current_tracer_fd);
+
+ /*
+ * The kernel event log uses CLOCK_MONOTONIC timestamps,
+ * not CLOCK_REALTIME timestamps. These differ by a constant
+ * but the constant is not available in user mode.
+ * This estimate will be off by one syscall round-trip.
+ */
+ clib_time_init (&em->cpu_timer);
+ em->init_time.cpu = em->cpu_timer.init_cpu_time;
+ syscall (SYS_clock_gettime, CLOCK_MONOTONIC, &ts);
+
+ /* enable kernel tracing */
+ if (write (enable_fd, "1\n", 2) != 2)
+ {
+ clib_unix_warning ("enable tracing");
+ close (enable_fd);
+ return;
+ }
+
+ close (enable_fd);
+}
+
+
+u8 *
+format_sched_event (u8 * s, va_list * va)
+{
+ sched_event_t *e = va_arg (*va, sched_event_t *);
+
+ s = format (s, "cpu %d task %10s type %s timestamp %12.6f\n",
+ e->cpu, e->task, e->type ? "WAKEUP " : "RUNNING", e->timestamp);
+
+ return s;
+}
+
+sched_event_t *
+parse_sched_switch_trace (u8 * tdata, u32 * index)
+{
+ u8 *cp = tdata + *index;
+ u8 *limit = tdata + vec_len (tdata);
+ int colons;
+ static sched_event_t event;
+ sched_event_t *e = &event;
+ static u8 *task_name;
+ u32 secs, usecs;
+ int i;
+
+again:
+ /* eat leading w/s */
+ while (cp < limit && (*cp == ' ' && *cp == '\t'))
+ cp++;
+ if (cp == limit)
+ return 0;
+
+ /* header line */
+ if (*cp == '#')
+ {
+ while (cp < limit && (*cp != '\n'))
+ cp++;
+ if (*cp == '\n')
+ {
+ cp++;
+ goto again;
+ }
+ clib_warning ("bugger 0");
+ return 0;
+ }
+
+ while (cp < limit && *cp != ']')
+ cp++;
+
+ if (*cp == 0)
+ return 0;
+
+ if (*cp != ']')
+ {
+ clib_warning ("bugger 0.1");
+ return 0;
+ }
+
+ cp++;
+ while (cp < limit && (*cp == ' ' && *cp == '\t'))
+ cp++;
+ if (cp == limit)
+ {
+ clib_warning ("bugger 0.2");
+ return 0;
+ }
+
+ secs = atoi (cp);
+
+ while (cp < limit && (*cp != '.'))
+ cp++;
+
+ if (cp == limit)
+ {
+ clib_warning ("bugger 0.3");
+ return 0;
+ }
+
+ cp++;
+
+ usecs = atoi (cp);
+
+ e->timestamp = ((f64) secs) + ((f64) usecs) * 1e-6;
+
+ /* eat up to third colon */
+ for (i = 0; i < 3; i++)
+ {
+ while (cp < limit && *cp != ':')
+ cp++;
+ cp++;
+ }
+ --cp;
+ if (*cp != ':')
+ {
+ clib_warning ("bugger 1");
+ return 0;
+ }
+ /* aim at '>' (switch-to) / '+' (wakeup) */
+ cp += 5;
+ if (cp >= limit)
+ {
+ clib_warning ("bugger 2");
+ return 0;
+ }
+ if (*cp == '>')
+ e->type = RUNNING;
+ else if (*cp == '+')
+ e->type = WAKEUP;
+ else
+ {
+ clib_warning ("bugger 3");
+ return 0;
+ }
+
+ cp += 3;
+ if (cp >= limit)
+ {
+ clib_warning ("bugger 4");
+ return 0;
+ }
+
+ e->cpu = atoi (cp);
+ cp += 4;
+
+ if (cp >= limit)
+ {
+ clib_warning ("bugger 4");
+ return 0;
+ }
+ while (cp < limit && (*cp == ' ' || *cp == '\t'))
+ cp++;
+
+ e->pid = atoi (cp);
+
+ for (i = 0; i < 2; i++)
+ {
+ while (cp < limit && *cp != ':')
+ cp++;
+ cp++;
+ }
+ --cp;
+ if (*cp != ':')
+ {
+ clib_warning ("bugger 5");
+ return 0;
+ }
+
+ cp += 3;
+ if (cp >= limit)
+ {
+ clib_warning ("bugger 6");
+ return 0;
+ }
+ while (cp < limit && (*cp != ' ' && *cp != '\n'))
+ {
+ vec_add1 (task_name, *cp);
+ cp++;
+ }
+ vec_add1 (task_name, 0);
+ /* _vec_len() = 0 in caller */
+ e->task = task_name;
+
+ if (cp < limit)
+ cp++;
+
+ *index = cp - tdata;
+ return e;
+}
+
+static u32
+elog_id_for_pid (elog_main_t * em, u8 * name, u32 pid)
+{
+ uword *p, r;
+ mhash_t *h = &em->string_table_hash;
+
+ if (!em->string_table_hash.hash)
+ mhash_init (h, sizeof (uword), sizeof (pid));
+
+ p = mhash_get (h, &pid);
+ if (p)
+ return p[0];
+ r = elog_string (em, "%s(%d)", name, pid);
+ mhash_set (h, &pid, r, /* old_value */ 0);
+ return r;
+}
+
+void
+kelog_collect_sched_switch_trace (elog_main_t * em)
+{
+ int enable_fd, data_fd;
+ char *trace_enable = "/debug/tracing/tracing_enabled";
+ char *trace_data = "/debug/tracing/trace";
+ u8 *data = 0;
+ u8 *dp;
+ int bytes, total_bytes;
+ u32 pos;
+ sched_event_t *evt;
+ u64 nsec_to_add;
+ u32 index;
+ f64 clocks_per_sec;
+
+ enable_fd = open (trace_enable, O_RDWR);
+ if (enable_fd < 0)
+ {
+ clib_warning ("Couldn't open %s", trace_enable);
+ return;
+ }
+ /* disable kernel tracing */
+ if (write (enable_fd, "0\n", 2) != 2)
+ {
+ clib_unix_warning ("disable tracing");
+ close (enable_fd);
+ return;
+ }
+ close (enable_fd);
+
+ /* Read the trace data */
+ data_fd = open (trace_data, O_RDWR);
+ if (data_fd < 0)
+ {
+ clib_warning ("Couldn't open %s", trace_data);
+ return;
+ }
+
+ /*
+ * Extract trace into a vector. Note that seq_printf() [kernel]
+ * is not guaranteed to produce 4096 bytes at a time.
+ */
+ vec_validate (data, 4095);
+ total_bytes = 0;
+ pos = 0;
+ while (1)
+ {
+ bytes = read (data_fd, data + pos, 4096);
+ if (bytes <= 0)
+ break;
+
+ total_bytes += bytes;
+ _vec_len (data) = total_bytes;
+
+ pos = vec_len (data);
+ vec_validate (data, vec_len (data) + 4095);
+ }
+ vec_add1 (data, 0);
+
+ /* Synthesize events */
+ em->is_enabled = 1;
+
+ index = 0;
+ while ((evt = parse_sched_switch_trace (data, &index)))
+ {
+ u64 fake_cpu_clock;
+
+ fake_cpu_clock = evt->timestamp * em->cpu_timer.clocks_per_second;
+ {
+ ELOG_TYPE_DECLARE (e) =
+ {
+ .format = "%d: %s %s",.format_args = "i4T4t4",.n_enum_strings =
+ 2,.enum_strings =
+ {
+ "running", "wakeup",}
+ ,};
+ struct
+ {
+ u32 cpu, string_table_offset, which;
+ } *ed;
+
+ ed = elog_event_data_not_inline (em, &__ELOG_TYPE_VAR (e),
+ &em->default_track, fake_cpu_clock);
+ ed->cpu = evt->cpu;
+ ed->string_table_offset = elog_id_for_pid (em, evt->task, evt->pid);
+ ed->which = evt->type;
+ }
+ _vec_len (evt->task) = 0;
+ }
+ em->is_enabled = 0;
+}
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/vhash.c b/extras/deprecated/vppinfra/vhash.c
new file mode 100644
index 00000000000..9120f502c91
--- /dev/null
+++ b/extras/deprecated/vppinfra/vhash.c
@@ -0,0 +1,772 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ Copyright (c) 2010 Eliot Dresselhaus
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice shall be
+ included in all copies or substantial portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+*/
+
+#include <vppinfra/vhash.h>
+
+#ifdef CLIB_HAVE_VEC128
+
+/* Overflow search buckets have an extra u32x4 for saving key_hash data.
+ This makes it easier to refill main search bucket from overflow vector. */
+typedef struct
+{
+ /* 4 results for this bucket. */
+ u32x4_union_t result;
+
+ /* 4 hash codes for this bucket. These are used to refill main
+ search buckets from overflow buckets when space becomes available. */
+ u32x4_union_t key_hash;
+
+ /* n_key_u32s u32x4s of key data follow. */
+ u32x4_union_t key[0];
+} vhash_overflow_search_bucket_t;
+
+always_inline void
+set_overflow_result (vhash_overflow_search_bucket_t * b,
+ u32 i, u32 result, u32 key_hash)
+{
+ b->result.as_u32[i] = result;
+ b->key_hash.as_u32[i] = key_hash;
+}
+
+always_inline void
+free_overflow_bucket (vhash_overflow_buckets_t * ob,
+ vhash_overflow_search_bucket_t * b, u32 i)
+{
+ u32 o = (u32x4_union_t *) b - ob->search_buckets;
+ ASSERT (o < vec_len (ob->search_buckets));
+ vec_add1 (ob->free_indices, 4 * o + i);
+}
+
+always_inline vhash_overflow_search_bucket_t *
+get_overflow_search_bucket (vhash_overflow_buckets_t * obs, u32 i,
+ u32 n_key_u32s)
+{
+ return ((vhash_overflow_search_bucket_t *)
+ vec_elt_at_index (obs->search_buckets, i));
+}
+
+always_inline vhash_overflow_search_bucket_t *
+next_overflow_bucket (vhash_overflow_search_bucket_t * b, u32 n_key_u32s)
+{
+ return (vhash_overflow_search_bucket_t *) & b->key[n_key_u32s];
+}
+
+#define foreach_vhash_overflow_bucket(b,ob,n_key_u32s) \
+ for ((b) = (vhash_overflow_search_bucket_t *) ob->search_buckets; \
+ (u32x4_union_t *) (b) < vec_end (ob->search_buckets); \
+ b = next_overflow_bucket (b, n_key_u32s))
+
+u32
+vhash_get_overflow (vhash_t * h, u32 key_hash, u32 vi, u32 n_key_u32s)
+{
+ vhash_overflow_buckets_t *ob = vhash_get_overflow_buckets (h, key_hash);
+ vhash_overflow_search_bucket_t *b;
+ u32 i, result = 0;
+
+ foreach_vhash_overflow_bucket (b, ob, n_key_u32s)
+ {
+ u32x4 r = b->result.as_u32x4;
+
+ for (i = 0; i < n_key_u32s; i++)
+ r &= vhash_bucket_compare (h, &b->key[0], i, vi);
+
+ result = vhash_merge_results (r);
+ if (result)
+ break;
+ }
+
+ return result;
+}
+
+u32
+vhash_set_overflow (vhash_t * h,
+ u32 key_hash, u32 vi, u32 new_result, u32 n_key_u32s)
+{
+ vhash_overflow_buckets_t *ob = vhash_get_overflow_buckets (h, key_hash);
+ vhash_overflow_search_bucket_t *b;
+ u32 i_set, i, old_result;
+
+ foreach_vhash_overflow_bucket (b, ob, n_key_u32s)
+ {
+ u32x4 r;
+
+ r = b->result.as_u32x4;
+ for (i = 0; i < n_key_u32s; i++)
+ r &= vhash_bucket_compare (h, &b->key[0], i, vi);
+
+ old_result = vhash_merge_results (r);
+ if (old_result)
+ {
+ i_set = vhash_non_empty_result_index (r);
+ set_overflow_result (b, i_set, new_result, key_hash);
+ return old_result;
+ }
+ }
+
+ /* Check free list. */
+ if (vec_len (ob->free_indices) == 0)
+ {
+ /* Out of free overflow buckets. Resize. */
+ u32 j, *p;
+ i = vec_len (ob->search_buckets);
+ vec_resize_aligned (ob->search_buckets,
+ sizeof (b[0]) / sizeof (u32x4) + n_key_u32s,
+ CLIB_CACHE_LINE_BYTES);
+ vec_add2 (ob->free_indices, p, 4);
+ for (j = 0; j < 4; j++)
+ p[j] = 4 * i + j;
+ }
+
+ i = vec_pop (ob->free_indices);
+
+ i_set = i & 3;
+ b = ((vhash_overflow_search_bucket_t *)
+ vec_elt_at_index (ob->search_buckets, i / 4));
+
+ /* Insert result. */
+ set_overflow_result (b, i_set, new_result, key_hash);
+
+ /* Insert key. */
+ for (i = 0; i < n_key_u32s; i++)
+ b->key[i].as_u32[i_set] = vhash_get_key_word (h, i, vi);
+
+ ob->n_overflow++;
+ h->n_elts++;
+
+ return /* old result was invalid */ 0;
+}
+
+u32
+vhash_unset_overflow (vhash_t * h, u32 key_hash, u32 vi, u32 n_key_u32s)
+{
+ vhash_overflow_buckets_t *ob = vhash_get_overflow_buckets (h, key_hash);
+ vhash_overflow_search_bucket_t *b;
+ u32 i_set, i, old_result;
+
+ foreach_vhash_overflow_bucket (b, ob, n_key_u32s)
+ {
+ u32x4 r;
+
+ r = b->result.as_u32x4;
+ for (i = 0; i < n_key_u32s; i++)
+ r &= vhash_bucket_compare (h, &b->key[0], i, vi);
+
+ old_result = vhash_merge_results (r);
+ if (old_result)
+ {
+ i_set = vhash_non_empty_result_index (r);
+
+ /* Invalidate result and invert key hash so that this will
+ never match since all keys in this overflow bucket have
+ matching key hashs. */
+ set_overflow_result (b, i_set, 0, ~key_hash);
+
+ free_overflow_bucket (ob, b, i_set);
+
+ ASSERT (ob->n_overflow > 0);
+ ob->n_overflow--;
+ h->n_elts--;
+ return old_result;
+ }
+ }
+
+ /* Could not find key. */
+ return 0;
+}
+
+void
+vhash_unset_refill_from_overflow (vhash_t * h,
+ vhash_search_bucket_t * sb,
+ u32 key_hash, u32 n_key_u32s)
+{
+ vhash_overflow_buckets_t *obs = vhash_get_overflow_buckets (h, key_hash);
+ vhash_overflow_search_bucket_t *ob;
+ u32 i, j, i_refill, bucket_mask = h->bucket_mask.as_u32[0];
+
+ /* Find overflow element with matching key hash. */
+ foreach_vhash_overflow_bucket (ob, obs, n_key_u32s)
+ {
+ for (i = 0; i < 4; i++)
+ {
+ if (!ob->result.as_u32[i])
+ continue;
+ if ((ob->key_hash.as_u32[i] & bucket_mask)
+ != (key_hash & bucket_mask))
+ continue;
+
+ i_refill = vhash_empty_result_index (sb->result.as_u32x4);
+ sb->result.as_u32[i_refill] = ob->result.as_u32[i];
+ for (j = 0; j < n_key_u32s; j++)
+ sb->key[j].as_u32[i_refill] = ob->key[j].as_u32[i];
+ set_overflow_result (ob, i, 0, ~key_hash);
+ free_overflow_bucket (obs, ob, i);
+ return;
+ }
+ }
+}
+
+void
+vhash_init (vhash_t * h, u32 log2_n_keys, u32 n_key_u32, u32 * hash_seeds)
+{
+ uword i, j, m;
+ vhash_search_bucket_t *b;
+
+ clib_memset (h, 0, sizeof (h[0]));
+
+ /* Must have at least 4 keys (e.g. one search bucket). */
+ log2_n_keys = clib_max (log2_n_keys, 2);
+
+ h->log2_n_keys = log2_n_keys;
+ h->n_key_u32 = n_key_u32;
+ m = pow2_mask (h->log2_n_keys) & ~3;
+ for (i = 0; i < VECTOR_WORD_TYPE_LEN (u32); i++)
+ h->bucket_mask.as_u32[i] = m;
+
+ /* Allocate and zero search buckets. */
+ i = (sizeof (b[0]) / sizeof (u32x4) + n_key_u32) << (log2_n_keys - 2);
+ vec_validate_aligned (h->search_buckets, i - 1, CLIB_CACHE_LINE_BYTES);
+
+ for (i = 0; i < ARRAY_LEN (h->find_first_zero_table); i++)
+ h->find_first_zero_table[i] = min_log2 (first_set (~i));
+
+ for (i = 0; i < ARRAY_LEN (h->hash_seeds); i++)
+ for (j = 0; j < VECTOR_WORD_TYPE_LEN (u32); j++)
+ h->hash_seeds[i].as_u32[j] = hash_seeds[i];
+}
+
+static_always_inline u32
+vhash_main_key_gather (void *_vm, u32 vi, u32 wi, u32 n_key_u32)
+{
+ vhash_main_t *vm = _vm;
+ return vec_elt (vm->keys, vi * n_key_u32 + wi);
+}
+
+static_always_inline u32x4
+vhash_main_4key_gather (void *_vm, u32 vi, u32 wi, u32 n_key_u32s)
+{
+ vhash_main_t *vm = _vm;
+ u32x4_union_t x;
+
+ ASSERT (n_key_u32s == vm->n_key_u32);
+ ASSERT (wi < n_key_u32s);
+
+ x.as_u32[0] = vec_elt (vm->keys, (vi + 0) * n_key_u32s + wi);
+ x.as_u32[1] = vec_elt (vm->keys, (vi + 1) * n_key_u32s + wi);
+ x.as_u32[2] = vec_elt (vm->keys, (vi + 2) * n_key_u32s + wi);
+ x.as_u32[3] = vec_elt (vm->keys, (vi + 3) * n_key_u32s + wi);
+ return x.as_u32x4;
+}
+
+static_always_inline u32
+vhash_main_set_result (void *_vm, u32 vi, u32 old_result, u32 n_key_u32)
+{
+ vhash_main_t *vm = _vm;
+ u32 *p = vec_elt_at_index (vm->results, vi);
+ u32 new_result = p[0];
+ p[0] = old_result;
+ return new_result;
+}
+
+static_always_inline u32
+vhash_main_get_result (void *_vm, u32 vi, u32 old_result, u32 n_key_u32)
+{
+ vhash_main_t *vm = _vm;
+ vec_elt (vm->results, vi) = old_result;
+ return old_result;
+}
+
+static_always_inline u32x4
+vhash_main_get_4result (void *_vm, u32 vi, u32x4 old_result, u32 n_key_u32)
+{
+ vhash_main_t *vm = _vm;
+ u32x4 *p = (u32x4 *) vec_elt_at_index (vm->results, vi);
+ p[0] = old_result;
+ return old_result;
+}
+
+#define _(N_KEY_U32) \
+ static_always_inline u32 \
+ vhash_main_key_gather_##N_KEY_U32 (void * _vm, u32 vi, u32 i) \
+ { return vhash_main_key_gather (_vm, vi, i, N_KEY_U32); } \
+ \
+ static_always_inline u32x4 \
+ vhash_main_4key_gather_##N_KEY_U32 (void * _vm, u32 vi, u32 i) \
+ { return vhash_main_4key_gather (_vm, vi, i, N_KEY_U32); } \
+ \
+ clib_pipeline_stage_static \
+ (vhash_main_gather_keys_stage_##N_KEY_U32, \
+ vhash_main_t *, vm, i, \
+ { \
+ vhash_gather_4key_stage \
+ (vm->vhash, \
+ /* vector_index */ i, \
+ vhash_main_4key_gather_##N_KEY_U32, \
+ vm, \
+ N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage_no_inline \
+ (vhash_main_gather_keys_mod_stage_##N_KEY_U32, \
+ vhash_main_t *, vm, i, \
+ { \
+ vhash_gather_key_stage \
+ (vm->vhash, \
+ /* vector_index */ vm->n_vectors_div_4, \
+ /* n_vectors */ vm->n_vectors_mod_4, \
+ vhash_main_key_gather_##N_KEY_U32, \
+ vm, \
+ N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage \
+ (vhash_main_hash_finalize_stage_##N_KEY_U32, \
+ vhash_main_t *, vm, i, \
+ { \
+ vhash_finalize_stage (vm->vhash, i, N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage_no_inline \
+ (vhash_main_hash_finalize_mod_stage_##N_KEY_U32, \
+ vhash_main_t *, vm, i, \
+ { \
+ vhash_finalize_stage (vm->vhash, vm->n_vectors_div_4, N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage_static \
+ (vhash_main_get_stage_##N_KEY_U32, \
+ vhash_main_t *, vm, i, \
+ { \
+ vhash_get_4_stage (vm->vhash, \
+ /* vector_index */ i, \
+ vhash_main_get_4result, \
+ vm, N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage_no_inline \
+ (vhash_main_get_mod_stage_##N_KEY_U32, \
+ vhash_main_t *, vm, i, \
+ { \
+ vhash_get_stage (vm->vhash, \
+ /* vector_index */ vm->n_vectors_div_4, \
+ /* n_vectors */ vm->n_vectors_mod_4, \
+ vhash_main_get_result, \
+ vm, N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage_static \
+ (vhash_main_set_stage_##N_KEY_U32, \
+ vhash_main_t *, vm, i, \
+ { \
+ vhash_set_stage (vm->vhash, \
+ /* vector_index */ i, \
+ /* n_vectors */ VECTOR_WORD_TYPE_LEN (u32), \
+ vhash_main_set_result, \
+ vm, N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage_no_inline \
+ (vhash_main_set_mod_stage_##N_KEY_U32, \
+ vhash_main_t *, vm, i, \
+ { \
+ vhash_set_stage (vm->vhash, \
+ /* vector_index */ vm->n_vectors_div_4, \
+ /* n_vectors */ vm->n_vectors_mod_4, \
+ vhash_main_set_result, \
+ vm, N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage_static \
+ (vhash_main_unset_stage_##N_KEY_U32, \
+ vhash_main_t *, vm, i, \
+ { \
+ vhash_unset_stage (vm->vhash, \
+ /* vector_index */ i, \
+ /* n_vectors */ VECTOR_WORD_TYPE_LEN (u32), \
+ vhash_main_get_result, \
+ vm, N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage_no_inline \
+ (vhash_main_unset_mod_stage_##N_KEY_U32, \
+ vhash_main_t *, vm, i, \
+ { \
+ vhash_unset_stage (vm->vhash, \
+ /* vector_index */ vm->n_vectors_div_4, \
+ /* n_vectors */ vm->n_vectors_mod_4, \
+ vhash_main_get_result, \
+ vm, N_KEY_U32); \
+ })
+
+_(1);
+_(2);
+_(3);
+_(4);
+_(5);
+_(6);
+
+#undef _
+
+#define _(N_KEY_U32) \
+ clib_pipeline_stage \
+ (vhash_main_hash_mix_stage_##N_KEY_U32, \
+ vhash_main_t *, vm, i, \
+ { \
+ vhash_mix_stage (vm->vhash, i, N_KEY_U32); \
+ }) \
+ \
+ clib_pipeline_stage_no_inline \
+ (vhash_main_hash_mix_mod_stage_##N_KEY_U32, \
+ vhash_main_t *, vm, i, \
+ { \
+ vhash_mix_stage (vm->vhash, vm->n_vectors_div_4, N_KEY_U32); \
+ })
+
+_(4);
+_(5);
+_(6);
+
+#undef _
+
+typedef enum
+{
+ GET, SET, UNSET,
+} vhash_main_op_t;
+
+static void
+vhash_main_op (vhash_main_t * vm, vhash_main_op_t op)
+{
+ u32 n_keys = vec_len (vm->results);
+
+ vm->n_key_u32 = vm->vhash->n_key_u32;
+
+ vhash_validate_sizes (vm->vhash, vm->n_key_u32, n_keys);
+
+ vm->n_vectors_div_4 = n_keys / 4;
+ vm->n_vectors_mod_4 = n_keys % 4;
+
+ if (vm->n_vectors_div_4 > 0)
+ {
+ switch (vm->n_key_u32)
+ {
+ default:
+ ASSERT (0);
+ break;
+
+#define _(N_KEY_U32) \
+ case N_KEY_U32: \
+ if (op == GET) \
+ clib_pipeline_run_3_stage \
+ (vm->n_vectors_div_4, \
+ vm, \
+ vhash_main_gather_keys_stage_##N_KEY_U32, \
+ vhash_main_hash_finalize_stage_##N_KEY_U32, \
+ vhash_main_get_stage_##N_KEY_U32); \
+ else if (op == SET) \
+ clib_pipeline_run_3_stage \
+ (vm->n_vectors_div_4, \
+ vm, \
+ vhash_main_gather_keys_stage_##N_KEY_U32, \
+ vhash_main_hash_finalize_stage_##N_KEY_U32, \
+ vhash_main_set_stage_##N_KEY_U32); \
+ else \
+ clib_pipeline_run_3_stage \
+ (vm->n_vectors_div_4, \
+ vm, \
+ vhash_main_gather_keys_stage_##N_KEY_U32, \
+ vhash_main_hash_finalize_stage_##N_KEY_U32, \
+ vhash_main_unset_stage_##N_KEY_U32); \
+ break;
+
+ _(1);
+ _(2);
+ _(3);
+
+#undef _
+
+#define _(N_KEY_U32) \
+ case N_KEY_U32: \
+ if (op == GET) \
+ clib_pipeline_run_4_stage \
+ (vm->n_vectors_div_4, \
+ vm, \
+ vhash_main_gather_keys_stage_##N_KEY_U32, \
+ vhash_main_hash_mix_stage_##N_KEY_U32, \
+ vhash_main_hash_finalize_stage_##N_KEY_U32, \
+ vhash_main_get_stage_##N_KEY_U32); \
+ else if (op == SET) \
+ clib_pipeline_run_4_stage \
+ (vm->n_vectors_div_4, \
+ vm, \
+ vhash_main_gather_keys_stage_##N_KEY_U32, \
+ vhash_main_hash_mix_stage_##N_KEY_U32, \
+ vhash_main_hash_finalize_stage_##N_KEY_U32, \
+ vhash_main_set_stage_##N_KEY_U32); \
+ else \
+ clib_pipeline_run_4_stage \
+ (vm->n_vectors_div_4, \
+ vm, \
+ vhash_main_gather_keys_stage_##N_KEY_U32, \
+ vhash_main_hash_mix_stage_##N_KEY_U32, \
+ vhash_main_hash_finalize_stage_##N_KEY_U32, \
+ vhash_main_unset_stage_##N_KEY_U32); \
+ break;
+
+ _(4);
+ _(5);
+ _(6);
+
+#undef _
+ }
+ }
+
+
+ if (vm->n_vectors_mod_4 > 0)
+ {
+ switch (vm->n_key_u32)
+ {
+ default:
+ ASSERT (0);
+ break;
+
+#define _(N_KEY_U32) \
+ case N_KEY_U32: \
+ if (op == GET) \
+ clib_pipeline_run_3_stage \
+ (1, \
+ vm, \
+ vhash_main_gather_keys_mod_stage_##N_KEY_U32, \
+ vhash_main_hash_finalize_mod_stage_##N_KEY_U32, \
+ vhash_main_get_mod_stage_##N_KEY_U32); \
+ else if (op == SET) \
+ clib_pipeline_run_3_stage \
+ (1, \
+ vm, \
+ vhash_main_gather_keys_mod_stage_##N_KEY_U32, \
+ vhash_main_hash_finalize_mod_stage_##N_KEY_U32, \
+ vhash_main_set_mod_stage_##N_KEY_U32); \
+ else \
+ clib_pipeline_run_3_stage \
+ (1, \
+ vm, \
+ vhash_main_gather_keys_mod_stage_##N_KEY_U32, \
+ vhash_main_hash_finalize_mod_stage_##N_KEY_U32, \
+ vhash_main_unset_mod_stage_##N_KEY_U32); \
+ break;
+
+ _(1);
+ _(2);
+ _(3);
+
+#undef _
+
+#define _(N_KEY_U32) \
+ case N_KEY_U32: \
+ if (op == GET) \
+ clib_pipeline_run_4_stage \
+ (1, \
+ vm, \
+ vhash_main_gather_keys_mod_stage_##N_KEY_U32, \
+ vhash_main_hash_mix_mod_stage_##N_KEY_U32, \
+ vhash_main_hash_finalize_mod_stage_##N_KEY_U32, \
+ vhash_main_get_mod_stage_##N_KEY_U32); \
+ else if (op == SET) \
+ clib_pipeline_run_4_stage \
+ (1, \
+ vm, \
+ vhash_main_gather_keys_mod_stage_##N_KEY_U32, \
+ vhash_main_hash_mix_mod_stage_##N_KEY_U32, \
+ vhash_main_hash_finalize_mod_stage_##N_KEY_U32, \
+ vhash_main_set_mod_stage_##N_KEY_U32); \
+ else \
+ clib_pipeline_run_4_stage \
+ (1, \
+ vm, \
+ vhash_main_gather_keys_mod_stage_##N_KEY_U32, \
+ vhash_main_hash_mix_mod_stage_##N_KEY_U32, \
+ vhash_main_hash_finalize_mod_stage_##N_KEY_U32, \
+ vhash_main_unset_mod_stage_##N_KEY_U32); \
+ break;
+
+ _(4);
+ _(5);
+ _(6);
+
+#undef _
+ }
+ }
+}
+
+void
+vhash_main_get (vhash_main_t * vm)
+{
+ vhash_main_op (vm, GET);
+}
+
+void
+vhash_main_set (vhash_main_t * vm)
+{
+ vhash_main_op (vm, SET);
+}
+
+void
+vhash_main_unset (vhash_main_t * vm)
+{
+ vhash_main_op (vm, UNSET);
+}
+
+u32
+vhash_resize_incremental (vhash_resize_t * vr, u32 vector_index,
+ u32 n_keys_this_call)
+{
+ vhash_t *old = vr->old;
+ vhash_main_t *vm = &vr->new;
+ vhash_t *new = vm->vhash;
+ uword i, j, n_key_u32;
+
+ n_key_u32 = old->n_key_u32;
+
+ if (vector_index == 0)
+ {
+ u32 hash_seeds[3];
+ hash_seeds[0] = old->hash_seeds[0].as_u32[0];
+ hash_seeds[1] = old->hash_seeds[1].as_u32[0];
+ hash_seeds[2] = old->hash_seeds[2].as_u32[0];
+ vhash_init (new, old->log2_n_keys + 1, n_key_u32, hash_seeds);
+ }
+
+ vec_reset_length (vm->keys);
+ vec_reset_length (vm->results);
+
+ if (0 == (vector_index >> old->log2_n_keys))
+ {
+ for (i = vector_index; 0 == (i >> (old->log2_n_keys - 2)); i++)
+ {
+ vhash_search_bucket_t *b =
+ vhash_get_search_bucket_with_index (old, 4 * i, n_key_u32);
+ u32 r, *k;
+
+#define _(I) \
+ if ((r = b->result.as_u32[I]) != 0) \
+ { \
+ vec_add1 (vm->results, r - 1); \
+ vec_add2 (vm->keys, k, n_key_u32); \
+ for (j = 0; j < n_key_u32; j++) \
+ k[j] = b->key[j].as_u32[I]; \
+ }
+
+ _(0);
+ _(1);
+ _(2);
+ _(3);
+
+#undef _
+
+ if (vec_len (vm->results) >= n_keys_this_call)
+ {
+ vhash_main_op (vm, SET);
+ return i;
+ }
+ }
+ }
+
+ /* Add overflow buckets. */
+ {
+ vhash_overflow_buckets_t *ob;
+ vhash_overflow_search_bucket_t *b;
+
+ for (ob = old->overflow_buckets;
+ ob < old->overflow_buckets + ARRAY_LEN (old->overflow_buckets); ob++)
+ {
+ foreach_vhash_overflow_bucket (b, ob, old->n_key_u32)
+ {
+ u32 r, *k;
+
+#define _(I) \
+ if ((r = b->result.as_u32[I]) != 0) \
+ { \
+ vec_add1 (vm->results, r - 1); \
+ vec_add2 (vm->keys, k, n_key_u32); \
+ for (j = 0; j < n_key_u32; j++) \
+ k[j] = b->key[j].as_u32[I]; \
+ }
+
+ _(0);
+ _(1);
+ _(2);
+ _(3);
+
+#undef _
+ }
+ }
+ }
+
+ vhash_main_op (vm, SET);
+
+ /* Let caller know we are done. */
+ return ~0;
+}
+
+void
+vhash_resize (vhash_t * old, u32 log2_n_keys)
+{
+ static vhash_resize_t vr;
+ vhash_t new;
+ u32 i = 0;
+
+ vr.old = old;
+ vr.new.vhash = &new;
+
+ while (1)
+ {
+ i = vhash_resize_incremental (&vr, i, 1024);
+ if (i == ~0)
+ break;
+ }
+
+ vhash_free (old);
+ *old = new;
+}
+
+#endif /* CLIB_HAVE_VEC128 */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/vhash.h b/extras/deprecated/vppinfra/vhash.h
new file mode 100644
index 00000000000..85dfb788308
--- /dev/null
+++ b/extras/deprecated/vppinfra/vhash.h
@@ -0,0 +1,850 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ Copyright (c) 2010 Eliot Dresselhaus
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice shall be
+ included in all copies or substantial portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+*/
+
+#ifndef included_clib_vhash_h
+#define included_clib_vhash_h
+
+#include <vppinfra/vector.h>
+
+#ifdef CLIB_HAVE_VEC128
+
+#include <vppinfra/cache.h>
+#include <vppinfra/hash.h>
+#include <vppinfra/pipeline.h>
+
+/* Gathers 32 bits worth of key with given index. */
+typedef u32 (vhash_key_function_t) (void *state, u32 vector_index,
+ u32 key_word_index);
+typedef u32x4 (vhash_4key_function_t) (void *state, u32 vector_index,
+ u32 key_word_index);
+/* Sets/gets result of hash lookup. */
+typedef u32 (vhash_result_function_t) (void *state, u32 vector_index,
+ u32 result, u32 n_key_u32);
+typedef u32x4 (vhash_4result_function_t) (void *state, u32 vector_index,
+ u32x4 results, u32 n_key_u32);
+
+typedef struct
+{
+ u32x4_union_t hashed_key[3];
+} vhash_hashed_key_t;
+
+/* Search buckets are really this structure. */
+typedef struct
+{
+ /* 4 results for this bucket.
+ Zero is used to mark empty results. This means user can't use the result ~0
+ since user results differ from internal results stored in buckets by 1.
+ e.g. internal result = user result + 1. */
+ u32x4_union_t result;
+
+ /* n_key_u32s u32x4s of key data follow. */
+ u32x4_union_t key[0];
+} vhash_search_bucket_t;
+
+typedef struct
+{
+ u32x4_union_t *search_buckets;
+
+ /* Vector of bucket free indices. */
+ u32 *free_indices;
+
+ /* Number of entries in this overflow bucket. */
+ u32 n_overflow;
+} vhash_overflow_buckets_t;
+
+typedef struct
+{
+ /* 2^log2_n_keys keys grouped in groups of 4.
+ Each bucket contains 4 results plus 4 keys for a
+ total of (1 + n_key_u32) u32x4s. */
+ u32x4_union_t *search_buckets;
+
+ /* When a bucket of 4 results/keys are full we search
+ the overflow. hash_key is used to select which overflow
+ bucket. */
+ vhash_overflow_buckets_t overflow_buckets[16];
+
+ /* Total count of occupied elements in hash table. */
+ u32 n_elts;
+
+ u32 log2_n_keys;
+
+ /* Number of 32 bit words in a hash key. */
+ u32 n_key_u32;
+
+ u32x4_union_t bucket_mask;
+
+ /* table[i] = min_log2 (first_set (~i)). */
+ u8 find_first_zero_table[16];
+
+ /* Hash seeds for Jenkins hash. */
+ u32x4_union_t hash_seeds[3];
+
+ /* Key work space is a vector of length
+ n_key_u32s << log2_n_key_word_len_u32x. */
+ u32 log2_n_key_word_len_u32x;
+
+ /* Work space to store keys between pipeline stages. */
+ u32x4_union_t *key_work_space;
+
+ /* Hash work space to store Jenkins hash values between
+ pipeline stages. */
+ vhash_hashed_key_t *hash_work_space;
+} vhash_t;
+
+always_inline vhash_overflow_buckets_t *
+vhash_get_overflow_buckets (vhash_t * h, u32 key)
+{
+ u32 i = (((key & h->bucket_mask.as_u32[0]) >> 2) & 0xf);
+ ASSERT (i < ARRAY_LEN (h->overflow_buckets));
+ return h->overflow_buckets + i;
+}
+
+always_inline uword
+vhash_is_non_empty_overflow_bucket (vhash_t * h, u32 key)
+{
+ u32 i = (((key & h->bucket_mask.as_u32[0]) >> 2) & 0xf);
+ ASSERT (i < ARRAY_LEN (h->overflow_buckets));
+ return h->overflow_buckets[i].n_overflow > 0;
+}
+
+always_inline void
+vhash_free_overflow_buckets (vhash_overflow_buckets_t * obs)
+{
+ vec_free (obs->search_buckets);
+ vec_free (obs->free_indices);
+}
+
+always_inline void
+vhash_free (vhash_t * h)
+{
+ uword i;
+ for (i = 0; i < ARRAY_LEN (h->overflow_buckets); i++)
+ vhash_free_overflow_buckets (&h->overflow_buckets[i]);
+ vec_free (h->search_buckets);
+ vec_free (h->key_work_space);
+ vec_free (h->hash_work_space);
+}
+
+always_inline void
+vhash_set_key_word (vhash_t * h, u32 wi, u32 vi, u32 value)
+{
+ u32 i0 = (wi << h->log2_n_key_word_len_u32x) + (vi / 4);
+ u32 i1 = vi % 4;
+ vec_elt (h->key_work_space, i0).as_u32[i1] = value;
+}
+
+always_inline void
+vhash_set_key_word_u32x (vhash_t * h, u32 wi, u32 vi, u32x value)
+{
+ u32 i0 = (wi << h->log2_n_key_word_len_u32x) + (vi / 4);
+ vec_elt (h->key_work_space, i0).as_u32x4 = value;
+}
+
+always_inline u32
+vhash_get_key_word (vhash_t * h, u32 wi, u32 vi)
+{
+ u32 i0 = (wi << h->log2_n_key_word_len_u32x) + (vi / 4);
+ u32 i1 = vi % 4;
+ return vec_elt (h->key_work_space, i0).as_u32[i1];
+}
+
+always_inline u32x
+vhash_get_key_word_u32x (vhash_t * h, u32 wi, u32 vi)
+{
+ u32 i0 = (wi << h->log2_n_key_word_len_u32x) + vi;
+ return vec_elt (h->key_work_space, i0).as_u32x4;
+}
+
+always_inline void
+vhash_validate_sizes (vhash_t * h, u32 n_key_u32, u32 n_vectors)
+{
+ u32 n, l;
+
+ n = max_pow2 (n_vectors) / 4;
+ n = clib_max (n, 8);
+
+ h->log2_n_key_word_len_u32x = l = min_log2 (n);
+ vec_validate_aligned (h->key_work_space, (n_key_u32 << l) - 1,
+ CLIB_CACHE_LINE_BYTES);
+ vec_validate_aligned (h->hash_work_space, n - 1, CLIB_CACHE_LINE_BYTES);
+}
+
+always_inline void
+vhash_gather_key_stage (vhash_t * h,
+ u32 vector_index,
+ u32 n_vectors,
+ vhash_key_function_t key_function,
+ void *state, u32 n_key_u32s)
+{
+ u32 i, j, vi;
+
+ /* Gather keys for 4 packets (for 128 bit vector length e.g. u32x4). */
+ for (i = 0; i < n_vectors; i++)
+ {
+ vi = vector_index * 4 + i;
+ for (j = 0; j < n_key_u32s; j++)
+ vhash_set_key_word (h, j, vi, key_function (state, vi, j));
+ }
+}
+
+always_inline void
+vhash_gather_4key_stage (vhash_t * h,
+ u32 vector_index,
+ vhash_4key_function_t key_function,
+ void *state, u32 n_key_u32s)
+{
+ u32 j, vi;
+ vi = vector_index * 4;
+ for (j = 0; j < n_key_u32s; j++)
+ vhash_set_key_word_u32x (h, j, vi, key_function (state, vi, j));
+}
+
+always_inline void
+vhash_mix_stage (vhash_t * h, u32 vector_index, u32 n_key_u32s)
+{
+ i32 i, n_left;
+ u32x a, b, c;
+
+ /* Only need to do this for keys longer than 12 bytes. */
+ ASSERT (n_key_u32s > 3);
+
+ a = h->hash_seeds[0].as_u32x4;
+ b = h->hash_seeds[1].as_u32x4;
+ c = h->hash_seeds[2].as_u32x4;
+ for (i = 0, n_left = n_key_u32s - 3; n_left > 0; n_left -= 3, i += 3)
+ {
+ a +=
+ vhash_get_key_word_u32x (h, n_key_u32s - 1 - (i + 0), vector_index);
+ if (n_left > 1)
+ b +=
+ vhash_get_key_word_u32x (h, n_key_u32s - 1 - (i + 1), vector_index);
+ if (n_left > 2)
+ c +=
+ vhash_get_key_word_u32x (h, n_key_u32s - 1 - (i + 2), vector_index);
+
+ hash_v3_mix_u32x (a, b, c);
+ }
+
+ /* Save away a, b, c for later finalize. */
+ {
+ vhash_hashed_key_t *hk =
+ vec_elt_at_index (h->hash_work_space, vector_index);
+ hk->hashed_key[0].as_u32x4 = a;
+ hk->hashed_key[1].as_u32x4 = b;
+ hk->hashed_key[2].as_u32x4 = c;
+ }
+}
+
+always_inline vhash_search_bucket_t *
+vhash_get_search_bucket_with_index (vhash_t * h, u32 i, u32 n_key_u32s)
+{
+ return ((vhash_search_bucket_t *)
+ vec_elt_at_index (h->search_buckets,
+ (i / 4) *
+ ((sizeof (vhash_search_bucket_t) /
+ sizeof (u32x4)) + n_key_u32s)));
+}
+
+always_inline vhash_search_bucket_t *
+vhash_get_search_bucket (vhash_t * h, u32 key_hash, u32 n_key_u32s)
+{
+ u32 i = key_hash & h->bucket_mask.as_u32[0];
+ return vhash_get_search_bucket_with_index (h, i, n_key_u32s);
+}
+
+always_inline u32x4
+vhash_get_4_search_bucket_byte_offsets (vhash_t * h, u32x4 key_hash,
+ u32 n_key_u32s)
+{
+ vhash_search_bucket_t *b;
+ u32 n_bytes_per_bucket = sizeof (b[0]) + n_key_u32s * sizeof (b->key[0]);
+ u32x4 r = key_hash & h->bucket_mask.as_u32x4;
+
+ /* Multiply with shifts and adds to get bucket byte offset. */
+#define _(x) u32x4_ishift_left (r, (x) - 2)
+ if (n_bytes_per_bucket == (1 << 5))
+ r = _(5);
+ else if (n_bytes_per_bucket == ((1 << 5) + (1 << 4)))
+ r = _(5) + _(4);
+ else if (n_bytes_per_bucket == (1 << 6))
+ r = _(6);
+ else if (n_bytes_per_bucket == ((1 << 6) + (1 << 4)))
+ r = _(6) + _(4);
+ else if (n_bytes_per_bucket == ((1 << 6) + (1 << 5)))
+ r = _(6) + _(5);
+ else if (n_bytes_per_bucket == ((1 << 6) + (1 << 5) + (1 << 4)))
+ r = _(6) + _(5) + _(4);
+ else
+ ASSERT (0);
+#undef _
+ return r;
+}
+
+always_inline void
+vhash_finalize_stage (vhash_t * h, u32 vector_index, u32 n_key_u32s)
+{
+ i32 n_left;
+ u32x a, b, c;
+ vhash_hashed_key_t *hk =
+ vec_elt_at_index (h->hash_work_space, vector_index);
+
+ if (n_key_u32s <= 3)
+ {
+ a = h->hash_seeds[0].as_u32x4;
+ b = h->hash_seeds[1].as_u32x4;
+ c = h->hash_seeds[2].as_u32x4;
+ n_left = n_key_u32s;
+ }
+ else
+ {
+ a = hk->hashed_key[0].as_u32x4;
+ b = hk->hashed_key[1].as_u32x4;
+ c = hk->hashed_key[2].as_u32x4;
+ n_left = 3;
+ }
+
+ if (n_left > 0)
+ a += vhash_get_key_word_u32x (h, 0, vector_index);
+ if (n_left > 1)
+ b += vhash_get_key_word_u32x (h, 1, vector_index);
+ if (n_left > 2)
+ c += vhash_get_key_word_u32x (h, 2, vector_index);
+
+ hash_v3_finalize_u32x (a, b, c);
+
+ /* Only save away last 32 bits of hash code. */
+ hk->hashed_key[2].as_u32x4 = c;
+
+ /* Prefetch buckets. This costs a bit for small tables but saves
+ big for large ones. */
+ {
+ vhash_search_bucket_t *b0, *b1, *b2, *b3;
+ u32x4_union_t kh;
+
+ kh.as_u32x4 = vhash_get_4_search_bucket_byte_offsets (h, c, n_key_u32s);
+ hk->hashed_key[1].as_u32x4 = kh.as_u32x4;
+
+ b0 = (void *) h->search_buckets + kh.as_u32[0];
+ b1 = (void *) h->search_buckets + kh.as_u32[1];
+ b2 = (void *) h->search_buckets + kh.as_u32[2];
+ b3 = (void *) h->search_buckets + kh.as_u32[3];
+
+ CLIB_PREFETCH (b0, sizeof (b0[0]) + n_key_u32s * sizeof (b0->key[0]),
+ READ);
+ CLIB_PREFETCH (b1, sizeof (b1[0]) + n_key_u32s * sizeof (b1->key[0]),
+ READ);
+ CLIB_PREFETCH (b2, sizeof (b2[0]) + n_key_u32s * sizeof (b2->key[0]),
+ READ);
+ CLIB_PREFETCH (b3, sizeof (b3[0]) + n_key_u32s * sizeof (b3->key[0]),
+ READ);
+ }
+}
+
+always_inline u32
+vhash_merge_results (u32x4 r)
+{
+ r = r | u32x4_word_shift_right (r, 2);
+ r = r | u32x4_word_shift_right (r, 1);
+ return u32x4_get0 (r);
+}
+
+/* Bucket is full if none of its 4 results are 0. */
+always_inline u32
+vhash_search_bucket_is_full (u32x4 r)
+{
+ return u32x4_zero_byte_mask (r) == 0;
+}
+
+always_inline u32
+vhash_non_empty_result_index (u32x4 x)
+{
+ u32 empty_mask = u32x4_zero_byte_mask (x);
+ ASSERT (empty_mask != 0xffff);
+ return min_log2 (0xffff & ~empty_mask) / 4;
+}
+
+always_inline u32
+vhash_empty_result_index (u32x4 x)
+{
+ u32 empty_mask = u32x4_zero_byte_mask (x);
+ ASSERT (empty_mask != 0);
+ return min_log2 (0xffff & empty_mask) / 4;
+}
+
+always_inline u32x4
+vhash_bucket_compare (vhash_t * h,
+ u32x4_union_t * bucket, u32 key_word_index, u32 vi)
+{
+ u32 k = vhash_get_key_word (h, key_word_index, vi);
+ u32x4 x = { k, k, k, k };
+ return (bucket[key_word_index].as_u32x4 == x);
+}
+
+#define vhash_bucket_compare_4(h,wi,vi,b0,b1,b2,b3,cmp0,cmp1,cmp2,cmp3) \
+do { \
+ u32x4 _k4 = vhash_get_key_word_u32x ((h), (wi), (vi)); \
+ u32x4 _k0 = u32x4_splat_word (_k4, 0); \
+ u32x4 _k1 = u32x4_splat_word (_k4, 1); \
+ u32x4 _k2 = u32x4_splat_word (_k4, 2); \
+ u32x4 _k3 = u32x4_splat_word (_k4, 3); \
+ \
+ cmp0 = (b0->key[wi].as_u32x4 == _k0); \
+ cmp1 = (b1->key[wi].as_u32x4 == _k1); \
+ cmp2 = (b2->key[wi].as_u32x4 == _k2); \
+ cmp3 = (b3->key[wi].as_u32x4 == _k3); \
+} while (0)
+
+u32 vhash_get_overflow (vhash_t * h, u32 key_hash, u32 vi, u32 n_key_u32s);
+
+always_inline void
+vhash_get_stage (vhash_t * h,
+ u32 vector_index,
+ u32 n_vectors,
+ vhash_result_function_t result_function,
+ void *state, u32 n_key_u32s)
+{
+ u32 i, j;
+ vhash_hashed_key_t *hk =
+ vec_elt_at_index (h->hash_work_space, vector_index);
+ vhash_search_bucket_t *b;
+
+ for (i = 0; i < n_vectors; i++)
+ {
+ u32 vi = vector_index * 4 + i;
+ u32 key_hash = hk->hashed_key[2].as_u32[i];
+ u32 result;
+ u32x4 r, r0;
+
+ b = vhash_get_search_bucket (h, key_hash, n_key_u32s);
+
+ r = r0 = b->result.as_u32x4;
+ for (j = 0; j < n_key_u32s; j++)
+ r &= vhash_bucket_compare (h, &b->key[0], j, vi);
+
+ /* At this point only one of 4 results should be non-zero.
+ So we can or all 4 together and get the valid result (if there is one). */
+ result = vhash_merge_results (r);
+
+ if (!result && vhash_search_bucket_is_full (r0))
+ result = vhash_get_overflow (h, key_hash, vi, n_key_u32s);
+
+ result_function (state, vi, result - 1, n_key_u32s);
+ }
+}
+
+always_inline void
+vhash_get_4_stage (vhash_t * h,
+ u32 vector_index,
+ vhash_4result_function_t result_function,
+ void *state, u32 n_key_u32s)
+{
+ u32 i, vi;
+ vhash_hashed_key_t *hk =
+ vec_elt_at_index (h->hash_work_space, vector_index);
+ vhash_search_bucket_t *b0, *b1, *b2, *b3;
+ u32x4 r0, r1, r2, r3, r0_before, r1_before, r2_before, r3_before;
+ u32x4_union_t kh;
+
+ kh.as_u32x4 = hk->hashed_key[1].as_u32x4;
+
+ b0 = (void *) h->search_buckets + kh.as_u32[0];
+ b1 = (void *) h->search_buckets + kh.as_u32[1];
+ b2 = (void *) h->search_buckets + kh.as_u32[2];
+ b3 = (void *) h->search_buckets + kh.as_u32[3];
+
+ r0 = r0_before = b0->result.as_u32x4;
+ r1 = r1_before = b1->result.as_u32x4;
+ r2 = r2_before = b2->result.as_u32x4;
+ r3 = r3_before = b3->result.as_u32x4;
+
+ vi = vector_index * 4;
+
+ for (i = 0; i < n_key_u32s; i++)
+ {
+ u32x4 c0, c1, c2, c3;
+ vhash_bucket_compare_4 (h, i, vector_index,
+ b0, b1, b2, b3, c0, c1, c2, c3);
+ r0 &= c0;
+ r1 &= c1;
+ r2 &= c2;
+ r3 &= c3;
+ }
+
+ u32x4_transpose (r0, r1, r2, r3);
+
+ /* Gather together 4 results. */
+ {
+ u32x4_union_t r;
+ u32x4 ones = { 1, 1, 1, 1 };
+ u32 not_found_mask;
+
+ r.as_u32x4 = r0 | r1 | r2 | r3;
+ not_found_mask = u32x4_zero_byte_mask (r.as_u32x4);
+ not_found_mask &= ((vhash_search_bucket_is_full (r0_before) << (4 * 0))
+ | (vhash_search_bucket_is_full (r1_before) << (4 * 1))
+ | (vhash_search_bucket_is_full (r2_before) << (4 * 2))
+ | (vhash_search_bucket_is_full (r3_before) <<
+ (4 * 3)));
+ if (not_found_mask)
+ {
+ u32x4_union_t key_hash;
+
+ key_hash.as_u32x4 =
+ hk->hashed_key[2].as_u32x4 & h->bucket_mask.as_u32x4;
+
+ /* Slow path: one of the buckets may have been full and we need to search overflow. */
+ if (not_found_mask & (1 << (4 * 0)))
+ r.as_u32[0] = vhash_get_overflow (h, key_hash.as_u32[0],
+ vi + 0, n_key_u32s);
+ if (not_found_mask & (1 << (4 * 1)))
+ r.as_u32[1] = vhash_get_overflow (h, key_hash.as_u32[1],
+ vi + 1, n_key_u32s);
+ if (not_found_mask & (1 << (4 * 2)))
+ r.as_u32[2] = vhash_get_overflow (h, key_hash.as_u32[2],
+ vi + 2, n_key_u32s);
+ if (not_found_mask & (1 << (4 * 3)))
+ r.as_u32[3] = vhash_get_overflow (h, key_hash.as_u32[3],
+ vi + 3, n_key_u32s);
+ }
+
+ result_function (state, vi, r.as_u32x4 - ones, n_key_u32s);
+ }
+}
+
+u32
+vhash_set_overflow (vhash_t * h,
+ u32 key_hash, u32 vi, u32 new_result, u32 n_key_u32s);
+
+always_inline void
+vhash_set_stage (vhash_t * h,
+ u32 vector_index,
+ u32 n_vectors,
+ vhash_result_function_t result_function,
+ void *state, u32 n_key_u32s)
+{
+ u32 i, j, n_new_elts = 0;
+ vhash_hashed_key_t *hk =
+ vec_elt_at_index (h->hash_work_space, vector_index);
+ vhash_search_bucket_t *b;
+
+ for (i = 0; i < n_vectors; i++)
+ {
+ u32 vi = vector_index * 4 + i;
+ u32 key_hash = hk->hashed_key[2].as_u32[i];
+ u32 old_result, new_result;
+ u32 i_set;
+ u32x4 r, r0, cmp;
+
+ b = vhash_get_search_bucket (h, key_hash, n_key_u32s);
+
+ cmp = vhash_bucket_compare (h, &b->key[0], 0, vi);
+ for (j = 1; j < n_key_u32s; j++)
+ cmp &= vhash_bucket_compare (h, &b->key[0], j, vi);
+
+ r0 = b->result.as_u32x4;
+ r = r0 & cmp;
+
+ /* At this point only one of 4 results should be non-zero.
+ So we can or all 4 together and get the valid result (if there is one). */
+ old_result = vhash_merge_results (r);
+
+ if (!old_result && vhash_search_bucket_is_full (r0))
+ old_result = vhash_get_overflow (h, key_hash, vi, n_key_u32s);
+
+ /* Get new result; possibly do something with old result. */
+ new_result = result_function (state, vi, old_result - 1, n_key_u32s);
+
+ /* User cannot use ~0 as a hash result since a result of 0 is
+ used to mark unused bucket entries. */
+ ASSERT (new_result + 1 != 0);
+ new_result += 1;
+
+ /* Set over-writes existing result. */
+ if (old_result)
+ {
+ i_set = vhash_non_empty_result_index (r);
+ b->result.as_u32[i_set] = new_result;
+ }
+ else
+ {
+ /* Set allocates new result. */
+ u32 valid_mask;
+
+ valid_mask = (((b->result.as_u32[0] != 0) << 0)
+ | ((b->result.as_u32[1] != 0) << 1)
+ | ((b->result.as_u32[2] != 0) << 2)
+ | ((b->result.as_u32[3] != 0) << 3));
+
+ /* Rotate 4 bit valid mask so that key_hash corresponds to bit 0. */
+ i_set = key_hash & 3;
+ valid_mask =
+ ((valid_mask >> i_set) | (valid_mask << (4 - i_set))) & 0xf;
+
+ /* Insert into first empty position in bucket after key_hash. */
+ i_set = (i_set + h->find_first_zero_table[valid_mask]) & 3;
+
+ if (valid_mask != 0xf)
+ {
+ n_new_elts += 1;
+
+ b->result.as_u32[i_set] = new_result;
+
+ /* Insert new key into search bucket. */
+ for (j = 0; j < n_key_u32s; j++)
+ b->key[j].as_u32[i_set] = vhash_get_key_word (h, j, vi);
+ }
+ else
+ vhash_set_overflow (h, key_hash, vi, new_result, n_key_u32s);
+ }
+ }
+
+ h->n_elts += n_new_elts;
+}
+
+u32 vhash_unset_overflow (vhash_t * h, u32 key_hash, u32 vi, u32 n_key_u32s);
+
+void
+vhash_unset_refill_from_overflow (vhash_t * h,
+ vhash_search_bucket_t * b,
+ u32 key_hash, u32 n_key_u32s);
+
+/* Note: Eliot tried doing 4 unsets at once and could not get a speed up
+ and abandoned vhash_unset_4_stage. */
+always_inline void
+vhash_unset_stage (vhash_t * h,
+ u32 vector_index,
+ u32 n_vectors,
+ vhash_result_function_t result_function,
+ void *state, u32 n_key_u32s)
+{
+ u32 i, j, n_elts_unset = 0;
+ vhash_hashed_key_t *hk =
+ vec_elt_at_index (h->hash_work_space, vector_index);
+ vhash_search_bucket_t *b;
+
+ for (i = 0; i < n_vectors; i++)
+ {
+ u32 vi = vector_index * 4 + i;
+ u32 key_hash = hk->hashed_key[2].as_u32[i];
+ u32 old_result;
+ u32x4 cmp, r0;
+
+ b = vhash_get_search_bucket (h, key_hash, n_key_u32s);
+
+ cmp = vhash_bucket_compare (h, &b->key[0], 0, vi);
+ for (j = 1; j < n_key_u32s; j++)
+ cmp &= vhash_bucket_compare (h, &b->key[0], j, vi);
+
+ r0 = b->result.as_u32x4;
+
+ /* At this point cmp is all ones where key matches and zero otherwise.
+ So, this will invalidate results for matching key and do nothing otherwise. */
+ b->result.as_u32x4 = r0 & ~cmp;
+
+ old_result = vhash_merge_results (r0 & cmp);
+
+ n_elts_unset += old_result != 0;
+
+ if (vhash_search_bucket_is_full (r0))
+ {
+ if (old_result)
+ vhash_unset_refill_from_overflow (h, b, key_hash, n_key_u32s);
+ else
+ old_result = vhash_unset_overflow (h, key_hash, vi, n_key_u32s);
+ }
+
+ result_function (state, vi, old_result - 1, n_key_u32s);
+ }
+ ASSERT (h->n_elts >= n_elts_unset);
+ h->n_elts -= n_elts_unset;
+}
+
+void vhash_init (vhash_t * h, u32 log2_n_keys, u32 n_key_u32,
+ u32 * hash_seeds);
+
+void vhash_resize (vhash_t * old, u32 log2_n_keys);
+
+typedef struct
+{
+ vhash_t *vhash;
+
+ union
+ {
+ struct
+ {
+ u32 *keys;
+ u32 *results;
+ };
+
+ /* Vector layout for get keys. */
+ struct
+ {
+ u32x4_union_t *get_keys;
+ u32x4_union_t *get_results;
+ };
+ };
+
+ u32 n_vectors_div_4;
+ u32 n_vectors_mod_4;
+
+ u32 n_key_u32;
+
+ u32 n_keys;
+} vhash_main_t;
+
+always_inline u32
+vhash_get_alloc_keys (vhash_main_t * vm, u32 n_keys, u32 n_key_u32)
+{
+ u32 i, n;
+
+ i = vm->n_keys;
+ vm->n_keys = i + n_keys;
+
+ n = (round_pow2 (vm->n_keys, 4) / 4) * n_key_u32;
+
+ vec_validate_aligned (vm->get_keys, n - 1, sizeof (vm->get_keys[0]));
+ vec_validate_aligned (vm->get_results, n - 1, sizeof (vm->get_results[0]));
+
+ return i;
+}
+
+always_inline void
+vhash_get_set_key_word (vhash_main_t * vm, u32 vi, u32 wi, u32 n_key_u32,
+ u32 value)
+{
+ u32x4_union_t *k = vec_elt_at_index (vm->get_keys, (vi / 4) * n_key_u32);
+ ASSERT (wi < n_key_u32);
+ k[wi].as_u32[vi % 4] = value;
+}
+
+always_inline u32
+vhash_get_fetch_result (vhash_main_t * vm, u32 vi)
+{
+ u32x4_union_t *r = vec_elt_at_index (vm->get_results, vi / 4);
+ return r->as_u32[vi % 4];
+}
+
+void vhash_main_get (vhash_main_t * vm);
+
+always_inline u32
+vhash_set_alloc_keys (vhash_main_t * vm, u32 n_keys, u32 n_key_u32)
+{
+ u32 i;
+
+ i = vm->n_keys;
+ vm->n_keys = i + n_keys;
+
+ vec_resize (vm->keys, n_keys * n_key_u32);
+ vec_resize (vm->results, n_keys);
+
+ return i;
+}
+
+always_inline void
+vhash_set_set_key_word (vhash_main_t * vm, u32 vi, u32 wi, u32 n_key_u32,
+ u32 value)
+{
+ u32 *k = vec_elt_at_index (vm->keys, vi * n_key_u32);
+ ASSERT (wi < n_key_u32);
+ k[wi] = value;
+}
+
+always_inline void
+vhash_set_set_result (vhash_main_t * vm, u32 vi, u32 result)
+{
+ u32 *r = vec_elt_at_index (vm->results, vi);
+ r[0] = result;
+}
+
+always_inline u32
+vhash_set_fetch_old_result (vhash_main_t * vm, u32 vi)
+{
+ u32 *r = vec_elt_at_index (vm->results, vi);
+ return r[0];
+}
+
+void vhash_main_set (vhash_main_t * vm);
+
+always_inline u32
+vhash_unset_alloc_keys (vhash_main_t * vm, u32 n_keys, u32 n_key_u32)
+{
+ return vhash_set_alloc_keys (vm, n_keys, n_key_u32);
+}
+
+always_inline void
+vhash_unset_set_key_word (vhash_main_t * vm, u32 vi, u32 wi, u32 n_key_u32,
+ u32 value)
+{
+ vhash_set_set_key_word (vm, vi, wi, n_key_u32, value);
+}
+
+always_inline void
+vhash_unset_set_result (vhash_main_t * vm, u32 vi, u32 result)
+{
+ vhash_set_set_result (vm, vi, result);
+}
+
+always_inline u32
+vhash_unset_fetch_old_result (vhash_main_t * vm, u32 vi)
+{
+ return vhash_set_fetch_old_result (vm, vi);
+}
+
+void vhash_main_unset (vhash_main_t * vm);
+
+typedef struct
+{
+ vhash_main_t new;
+
+ vhash_t *old;
+} vhash_resize_t;
+
+u32 vhash_resize_incremental (vhash_resize_t * vr, u32 vector_index,
+ u32 n_vectors);
+
+#endif /* CLIB_HAVE_VEC128 */
+
+#endif /* included_clib_vhash_h */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/xy.h b/extras/deprecated/vppinfra/xy.h
new file mode 100644
index 00000000000..fb562161a62
--- /dev/null
+++ b/extras/deprecated/vppinfra/xy.h
@@ -0,0 +1,56 @@
+/* (X,Y) coordinates. */
+
+/*
+ Copyright (c) 2008 Eliot Dresselhaus
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice shall be
+ included in all copies or substantial portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+*/
+
+#ifndef included_clib_xy_h
+#define included_clib_xy_h
+
+#include <vppinfra/types.h>
+
+/* Basic definitions: coordinates and points. */
+typedef double xy_float_t;
+typedef __complex__ double xy_t;
+typedef __complex__ int ixy_t;
+
+typedef __complex__ char i8xy_t;
+typedef __complex__ short i16xy_t;
+typedef __complex__ int i32xy_t;
+
+/* X/Y components of a point: can be used as either rvalue/lvalue. */
+#define xy_x(x) __real__ (x)
+#define xy_y(x) __imag__ (x)
+
+/* Unit vectors in x/y directions. */
+#define xy_x_unit_vector (1)
+#define xy_y_unit_vector (1I)
+
+#endif /* included_clib_xy_h */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/zvec.c b/extras/deprecated/vppinfra/zvec.c
new file mode 100644
index 00000000000..d062e5f7db1
--- /dev/null
+++ b/extras/deprecated/vppinfra/zvec.c
@@ -0,0 +1,442 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ Copyright (c) 2001, 2002, 2003, 2005 Eliot Dresselhaus
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice shall be
+ included in all copies or substantial portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+*/
+
+#include <vppinfra/bitmap.h>
+#include <vppinfra/bitops.h> /* for next_with_same_number_of_set_bits */
+#include <vppinfra/error.h> /* for ASSERT */
+#include <vppinfra/mem.h>
+#include <vppinfra/os.h> /* for os_panic */
+#include <vppinfra/vec.h>
+#include <vppinfra/zvec.h>
+
+/* Consider coding as bitmap, coding = 2^c_0 + 2^c_1 + ... + 2^c_n
+ With c_0 < c_1 < ... < c_n. coding == 0 represents c_n = BITS (uword).
+
+ Unsigned integers i = 0 ... are represented as follows:
+
+ 0 <= i < 2^c_0 (i << 1) | (1 << 0) binary: i 1
+ 2^c_0 <= i < 2^c_0 + 2^c_1 (i << 2) | (1 << 1) binary: i 1 0
+ ... binary: i 0 ... 0
+
+ Smaller numbers use less bits. Coding is chosen so that encoding
+ of given histogram of typical values gives smallest number of bits.
+ The number and position of coding bits c_i are used to best fit the
+ histogram of typical values.
+*/
+
+/* Decode given compressed data. Return number of compressed data
+ bits used. */
+uword
+zvec_decode (uword coding, uword zdata, uword * n_zdata_bits)
+{
+ uword c, d, result, n_bits;
+ uword explicit_end, implicit_end;
+
+ result = 0;
+ n_bits = 0;
+ while (1)
+ {
+ c = first_set (coding);
+ implicit_end = c == coding;
+ explicit_end = (zdata & 1) & ~implicit_end;
+ d = (zdata >> explicit_end) & (c - 1);
+ if (explicit_end | implicit_end)
+ {
+ result += d;
+ n_bits += min_log2 (c) + explicit_end;
+ break;
+ }
+ n_bits += 1;
+ result += c;
+ coding ^= c;
+ zdata >>= 1;
+ }
+
+ if (coding == 0)
+ n_bits = BITS (uword);
+
+ *n_zdata_bits = n_bits;
+ return result;
+}
+
+uword
+zvec_encode (uword coding, uword data, uword * n_result_bits)
+{
+ uword c, shift, result;
+ uword explicit_end, implicit_end;
+
+ /* Data must be in range. Note special coding == 0
+ would break for data - 1 <= coding. */
+ ASSERT (data <= coding - 1);
+
+ shift = 0;
+ while (1)
+ {
+ c = first_set (coding);
+ implicit_end = c == coding;
+ explicit_end = ((data & (c - 1)) == data);
+ if (explicit_end | implicit_end)
+ {
+ uword t = explicit_end & ~implicit_end;
+ result = ((data << t) | t) << shift;
+ *n_result_bits =
+ /* data bits */ (c == 0 ? BITS (uword) : min_log2 (c))
+ /* shift bits */ + shift + t;
+ return result;
+ }
+ data -= c;
+ coding ^= c;
+ shift++;
+ }
+
+ /* Never reached. */
+ ASSERT (0);
+ return ~0;
+}
+
+always_inline uword
+get_data (void *data, uword data_bytes, uword is_signed)
+{
+ if (data_bytes == 1)
+ return is_signed ? zvec_signed_to_unsigned (*(i8 *) data) : *(u8 *) data;
+ else if (data_bytes == 2)
+ return is_signed ? zvec_signed_to_unsigned (*(i16 *) data) : *(u16 *)
+ data;
+ else if (data_bytes == 4)
+ return is_signed ? zvec_signed_to_unsigned (*(i32 *) data) : *(u32 *)
+ data;
+ else if (data_bytes == 8)
+ return is_signed ? zvec_signed_to_unsigned (*(i64 *) data) : *(u64 *)
+ data;
+ else
+ {
+ os_panic ();
+ return ~0;
+ }
+}
+
+always_inline void
+put_data (void *data, uword data_bytes, uword is_signed, uword x)
+{
+ if (data_bytes == 1)
+ {
+ if (is_signed)
+ *(i8 *) data = zvec_unsigned_to_signed (x);
+ else
+ *(u8 *) data = x;
+ }
+ else if (data_bytes == 2)
+ {
+ if (is_signed)
+ *(i16 *) data = zvec_unsigned_to_signed (x);
+ else
+ *(u16 *) data = x;
+ }
+ else if (data_bytes == 4)
+ {
+ if (is_signed)
+ *(i32 *) data = zvec_unsigned_to_signed (x);
+ else
+ *(u32 *) data = x;
+ }
+ else if (data_bytes == 8)
+ {
+ if (is_signed)
+ *(i64 *) data = zvec_unsigned_to_signed (x);
+ else
+ *(u64 *) data = x;
+ }
+ else
+ {
+ os_panic ();
+ }
+}
+
+always_inline uword *
+zvec_encode_inline (uword * zvec,
+ uword * zvec_n_bits,
+ uword coding,
+ void *data,
+ uword data_stride,
+ uword n_data, uword data_bytes, uword is_signed)
+{
+ uword i;
+
+ i = *zvec_n_bits;
+ while (n_data >= 1)
+ {
+ uword d0, z0, l0;
+
+ d0 = get_data (data + 0 * data_stride, data_bytes, is_signed);
+ data += 1 * data_stride;
+ n_data -= 1;
+
+ z0 = zvec_encode (coding, d0, &l0);
+ zvec = clib_bitmap_set_multiple (zvec, i, z0, l0);
+ i += l0;
+ }
+
+ *zvec_n_bits = i;
+ return zvec;
+}
+
+#define _(TYPE,IS_SIGNED) \
+ uword * zvec_encode_##TYPE (uword * zvec, \
+ uword * zvec_n_bits, \
+ uword coding, \
+ void * data, \
+ uword data_stride, \
+ uword n_data) \
+ { \
+ return zvec_encode_inline (zvec, zvec_n_bits, \
+ coding, \
+ data, data_stride, n_data, \
+ /* data_bytes */ sizeof (TYPE), \
+ /* is_signed */ IS_SIGNED); \
+ }
+
+_(u8, /* is_signed */ 0);
+_(u16, /* is_signed */ 0);
+_(u32, /* is_signed */ 0);
+_(u64, /* is_signed */ 0);
+_(i8, /* is_signed */ 1);
+_(i16, /* is_signed */ 1);
+_(i32, /* is_signed */ 1);
+_(i64, /* is_signed */ 1);
+
+#undef _
+
+always_inline uword
+coding_max_n_bits (uword coding)
+{
+ uword n_bits;
+ (void) zvec_decode (coding, 0, &n_bits);
+ return n_bits;
+}
+
+always_inline void
+zvec_decode_inline (uword * zvec,
+ uword * zvec_n_bits,
+ uword coding,
+ void *data,
+ uword data_stride,
+ uword n_data, uword data_bytes, uword is_signed)
+{
+ uword i, n_max;
+
+ i = *zvec_n_bits;
+ n_max = coding_max_n_bits (coding);
+ while (n_data >= 1)
+ {
+ uword d0, z0, l0;
+
+ z0 = clib_bitmap_get_multiple (zvec, i, n_max);
+ d0 = zvec_decode (coding, z0, &l0);
+ i += l0;
+ put_data (data + 0 * data_stride, data_bytes, is_signed, d0);
+ data += 1 * data_stride;
+ n_data -= 1;
+ }
+ *zvec_n_bits = i;
+}
+
+#define _(TYPE,IS_SIGNED) \
+ void zvec_decode_##TYPE (uword * zvec, \
+ uword * zvec_n_bits, \
+ uword coding, \
+ void * data, \
+ uword data_stride, \
+ uword n_data) \
+ { \
+ return zvec_decode_inline (zvec, zvec_n_bits, \
+ coding, \
+ data, data_stride, n_data, \
+ /* data_bytes */ sizeof (TYPE), \
+ /* is_signed */ IS_SIGNED); \
+ }
+
+_(u8, /* is_signed */ 0);
+_(u16, /* is_signed */ 0);
+_(u32, /* is_signed */ 0);
+_(u64, /* is_signed */ 0);
+_(i8, /* is_signed */ 1);
+_(i16, /* is_signed */ 1);
+_(i32, /* is_signed */ 1);
+_(i64, /* is_signed */ 1);
+
+#undef _
+
+/* Compute number of bits needed to encode given histogram. */
+static uword
+zvec_coding_bits (uword coding, uword * histogram_counts, uword min_bits)
+{
+ uword n_type_bits, n_bits;
+ uword this_count, last_count, max_count_index;
+ uword i, b, l;
+
+ n_bits = 0;
+ n_type_bits = 1;
+ last_count = 0;
+ max_count_index = vec_len (histogram_counts) - 1;
+
+ /* Coding is not large enough to encode given data. */
+ if (coding <= max_count_index)
+ return ~0;
+
+ i = 0;
+ while (coding != 0)
+ {
+ b = first_set (coding);
+ l = min_log2 (b);
+ i += b;
+
+ this_count =
+ histogram_counts[i > max_count_index ? max_count_index : i - 1];
+
+ /* No more data to encode? */
+ if (this_count == last_count)
+ break;
+
+ /* Last coding is i 0 ... 0 so we don't need an extra type bit. */
+ if (coding == b)
+ n_type_bits--;
+
+ n_bits += (this_count - last_count) * (n_type_bits + l);
+
+ /* This coding cannot be minimal: so return. */
+ if (n_bits >= min_bits)
+ return ~0;
+
+ last_count = this_count;
+ coding ^= b;
+ n_type_bits++;
+ }
+
+ return n_bits;
+}
+
+uword
+_zvec_coding_from_histogram (void *histogram,
+ uword histogram_len,
+ uword histogram_elt_count_offset,
+ uword histogram_elt_bytes,
+ uword max_value_to_encode,
+ zvec_coding_info_t * coding_return)
+{
+ uword coding, min_coding;
+ uword min_coding_bits, coding_bits;
+ uword i, n_bits_set, total_count;
+ uword *counts;
+ zvec_histogram_count_t *h_count = histogram + histogram_elt_count_offset;
+
+ if (histogram_len < 1)
+ {
+ coding_return->coding = 0;
+ coding_return->min_coding_bits = 0;
+ coding_return->n_data = 0;
+ coding_return->n_codes = 0;
+ coding_return->ave_coding_bits = 0;
+ return 0;
+ }
+
+ total_count = 0;
+ counts = vec_new (uword, histogram_len);
+ for (i = 0; i < histogram_len; i++)
+ {
+ zvec_histogram_count_t this_count = h_count[0];
+ total_count += this_count;
+ counts[i] = total_count;
+ h_count =
+ (zvec_histogram_count_t *) ((void *) h_count + histogram_elt_bytes);
+ }
+
+ min_coding = 0;
+ min_coding_bits = ~0;
+
+ {
+ uword base_coding =
+ max_value_to_encode !=
+ ~0 ? (1 + max_value_to_encode) : vec_len (counts);
+ uword max_coding = max_pow2 (2 * base_coding);
+
+ for (n_bits_set = 1; n_bits_set <= 8; n_bits_set++)
+ {
+ for (coding = pow2_mask (n_bits_set);
+ coding < max_coding;
+ coding = next_with_same_number_of_set_bits (coding))
+ {
+ coding_bits = zvec_coding_bits (coding, counts, min_coding_bits);
+ if (coding_bits >= min_coding_bits)
+ continue;
+ min_coding_bits = coding_bits;
+ min_coding = coding;
+ }
+ }
+ }
+
+ if (coding_return)
+ {
+ coding_return->coding = min_coding;
+ coding_return->min_coding_bits = min_coding_bits;
+ coding_return->n_data = total_count;
+ coding_return->n_codes = vec_len (counts);
+ coding_return->ave_coding_bits =
+ (f64) min_coding_bits / (f64) total_count;
+ }
+
+ vec_free (counts);
+
+ return min_coding;
+}
+
+u8 *
+format_zvec_coding (u8 * s, va_list * args)
+{
+ zvec_coding_info_t *c = va_arg (*args, zvec_coding_info_t *);
+ return format (s,
+ "zvec coding 0x%x, %d elts, %d codes, %d bits total, %.4f ave bits/code",
+ c->coding, c->n_data, c->n_codes, c->min_coding_bits,
+ c->ave_coding_bits);
+}
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/extras/deprecated/vppinfra/zvec.h b/extras/deprecated/vppinfra/zvec.h
new file mode 100644
index 00000000000..7d35a3fe41f
--- /dev/null
+++ b/extras/deprecated/vppinfra/zvec.h
@@ -0,0 +1,166 @@
+/*
+ * Copyright (c) 2015 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*
+ Copyright (c) 2001, 2002, 2003 Eliot Dresselhaus
+
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+
+ The above copyright notice and this permission notice shall be
+ included in all copies or substantial portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+*/
+
+#ifndef included_zvec_h
+#define included_zvec_h
+
+#include <vppinfra/clib.h>
+#include <vppinfra/error.h> /* for ASSERT */
+#include <vppinfra/format.h>
+
+/* zvec: compressed vectors.
+
+ Data is entropy coded with 32 bit "codings".
+
+ Consider coding as bitmap, coding = 2^c_0 + 2^c_1 + ... + 2^c_n
+ With c_0 < c_1 < ... < c_n. coding == 0 represents c_n = BITS (uword).
+
+ Unsigned integers i = 0 ... are represented as follows:
+
+ 0 <= i < 2^c_0 (i << 1) | (1 << 0) binary: i 1
+ 2^c_0 <= i < 2^c_0 + 2^c_1 (i << 2) | (1 << 1) binary: i 1 0
+ ... binary: i 0 ... 0
+
+ Smaller numbers use less bits. Coding is chosen so that encoding
+ of given histogram of typical values gives smallest number of bits.
+ The number and position of coding bits c_i are used to best fit the
+ histogram of typical values.
+*/
+
+typedef struct
+{
+ /* Smallest coding for given histogram of typical data. */
+ u32 coding;
+
+ /* Number of data in histogram. */
+ u32 n_data;
+
+ /* Number of codes (unique values) in histogram. */
+ u32 n_codes;
+
+ /* Number of bits in smallest coding of data. */
+ u32 min_coding_bits;
+
+ /* Average number of bits per code. */
+ f64 ave_coding_bits;
+} zvec_coding_info_t;
+
+/* Encode/decode data. */
+uword zvec_encode (uword coding, uword data, uword * n_result_bits);
+uword zvec_decode (uword coding, uword zdata, uword * n_zdata_bits);
+
+format_function_t format_zvec_coding;
+
+typedef u32 zvec_histogram_count_t;
+
+#define zvec_coding_from_histogram(h,count_field,len,max_value_to_encode,zc) \
+ _zvec_coding_from_histogram ((h), (len), \
+ STRUCT_OFFSET_OF_VAR (h, count_field), \
+ sizeof (h[0]), \
+ max_value_to_encode, \
+ (zc))
+
+uword
+_zvec_coding_from_histogram (void *_histogram,
+ uword histogram_len,
+ uword histogram_elt_count_offset,
+ uword histogram_elt_bytes,
+ uword max_value_to_encode,
+ zvec_coding_info_t * coding_info_return);
+
+#define _(TYPE,IS_SIGNED) \
+ uword * zvec_encode_##TYPE (uword * zvec, uword * zvec_n_bits, uword coding, \
+ void * data, uword data_stride, uword n_data);
+
+_(u8, /* is_signed */ 0);
+_(u16, /* is_signed */ 0);
+_(u32, /* is_signed */ 0);
+_(u64, /* is_signed */ 0);
+_(i8, /* is_signed */ 1);
+_(i16, /* is_signed */ 1);
+_(i32, /* is_signed */ 1);
+_(i64, /* is_signed */ 1);
+
+#undef _
+
+#define _(TYPE,IS_SIGNED) \
+ void zvec_decode_##TYPE (uword * zvec, \
+ uword * zvec_n_bits, \
+ uword coding, \
+ void * data, \
+ uword data_stride, \
+ uword n_data)
+
+_(u8, /* is_signed */ 0);
+_(u16, /* is_signed */ 0);
+_(u32, /* is_signed */ 0);
+_(u64, /* is_signed */ 0);
+_(i8, /* is_signed */ 1);
+_(i16, /* is_signed */ 1);
+_(i32, /* is_signed */ 1);
+_(i64, /* is_signed */ 1);
+
+#undef _
+
+/* Signed <=> unsigned conversion.
+ -1, -2, -3, ... => 1, 3, 5, ... odds
+ 0, +1, +2, +3, ... => 0, 2, 4, 6, ... evens */
+always_inline uword
+zvec_signed_to_unsigned (word s)
+{
+ uword a = s < 0;
+ s = 2 * s + a;
+ return a ? -s : s;
+}
+
+always_inline word
+zvec_unsigned_to_signed (uword u)
+{
+ uword a = u & 1;
+ u >>= 1;
+ return a ? -u : u;
+}
+
+#endif /* included_zvec_h */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */