summaryrefslogtreecommitdiffstats
path: root/src
diff options
context:
space:
mode:
authorBenoît Ganne <bganne@cisco.com>2020-01-07 14:53:09 +0100
committerAndrew Yourtchenko <ayourtch@gmail.com>2020-01-08 09:43:25 +0000
commit5dfc03f110fd2ea742495c2ca755dfc511a024c9 (patch)
treed06ffb40407ddf58380af4b826fb7383b3af50cd /src
parent5c944eef7012e7c5b363399ed92966fb659196b7 (diff)
rdma: add FEATURE.yaml
Type: docs Change-Id: I91d701814e4bd9953616d2b7c76ae2ea7c07074b Signed-off-by: Benoît Ganne <bganne@cisco.com>
Diffstat (limited to 'src')
-rw-r--r--src/plugins/rdma/FEATURE.yaml8
1 files changed, 8 insertions, 0 deletions
diff --git a/src/plugins/rdma/FEATURE.yaml b/src/plugins/rdma/FEATURE.yaml
new file mode 100644
index 00000000000..66271089b2e
--- /dev/null
+++ b/src/plugins/rdma/FEATURE.yaml
@@ -0,0 +1,8 @@
+---
+name: rdma device driver
+maintainer: Benoît Ganne <bganne@cisco.com>
+features:
+ - rdma driver for Mellanox ConnectX-4/5 adapters
+description: "rdma device driver support"
+state: production
+properties: [API, CLI, STATS, MULTITHREAD]
#n159'>159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 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 childb 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))
    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:
 */