Diff for /embedaddon/rsync/hashtable.c between versions 1.1.1.3 and 1.1.1.4

version 1.1.1.3, 2016/11/01 09:54:32 version 1.1.1.4, 2021/03/17 00:32:36
Line 1 Line 1
 /*  /*
  * Routines to provide a memory-efficient hashtable.   * Routines to provide a memory-efficient hashtable.
  *   *
 * Copyright (C) 2007-2015 Wayne Davison * Copyright (C) 2007-2020 Wayne Davison
  *   *
  * This program is free software; you can redistribute it and/or modify   * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by   * it under the terms of the GNU General Public License as published by
Line 35  struct hashtable *hashtable_create(int size, int key64 Line 35  struct hashtable *hashtable_create(int size, int key64
                         size *= 2;                          size *= 2;
         }          }
   
        if (!(tbl = new(struct hashtable))        tbl = new(struct hashtable);
         || !(tbl->nodes = new_array0(char, size * node_size)))        tbl->nodes = new_array0(char, size * node_size);
                out_of_memory("hashtable_create"); 
         tbl->size = size;          tbl->size = size;
         tbl->entries = 0;          tbl->entries = 0;
         tbl->node_size = node_size;          tbl->node_size = node_size;
Line 66  void hashtable_destroy(struct hashtable *tbl) Line 65  void hashtable_destroy(struct hashtable *tbl)
         free(tbl);          free(tbl);
 }  }
   
/* This returns the node for the indicated key, either newly created or/* Returns the node that holds the indicated key if it exists. When it does not
 * already existing.  Returns NULL if not allocating and not found. */ * exist, it returns either NULL (when data_when_new is NULL), or it returns a
void *hashtable_find(struct hashtable *tbl, int64 key, int allocate_if_missing) * new node with its node->data set to the indicated value.
  *
  * If your code doesn't know the data value for a new node in advance (usually
  * because it doesn't know if a node is new or not) you should pass in a unique
  * (non-0) value that you can use to check if the returned node is new. You can
  * then overwrite the data with any value you want (even 0) since it only needs
  * to be different than whatever data_when_new value you use later on.
  *
  * This return is a void* just because it might be pointing at a ht_int32_node
  * or a ht_int64_node, and that makes the caller's assignment a little easier. */
 void *hashtable_find(struct hashtable *tbl, int64 key, void *data_when_new)
 {  {
         int key64 = tbl->key64;          int key64 = tbl->key64;
         struct ht_int32_node *node;          struct ht_int32_node *node;
Line 79  void *hashtable_find(struct hashtable *tbl, int64 key, Line 88  void *hashtable_find(struct hashtable *tbl, int64 key,
                 exit_cleanup(RERR_MESSAGEIO);                  exit_cleanup(RERR_MESSAGEIO);
         }          }
   
        if (allocate_if_missing && tbl->entries > HASH_LOAD_LIMIT(tbl->size)) {        if (data_when_new && tbl->entries > HASH_LOAD_LIMIT(tbl->size)) {
                 void *old_nodes = tbl->nodes;                  void *old_nodes = tbl->nodes;
                 int size = tbl->size * 2;                  int size = tbl->size * 2;
                 int i;                  int i;
   
                if (!(tbl->nodes = new_array0(char, size * tbl->node_size)))                tbl->nodes = new_array0(char, size * tbl->node_size);
                        out_of_memory("hashtable_node"); 
                 tbl->size = size;                  tbl->size = size;
                 tbl->entries = 0;                  tbl->entries = 0;
   
Line 99  void *hashtable_find(struct hashtable *tbl, int64 key, Line 107  void *hashtable_find(struct hashtable *tbl, int64 key,
                         int64 move_key = HT_KEY(move_node, key64);                          int64 move_key = HT_KEY(move_node, key64);
                         if (move_key == 0)                          if (move_key == 0)
                                 continue;                                  continue;
                        node = hashtable_find(tbl, move_key, 1);                        if (move_node->data)
                        node->data = move_node->data;                                hashtable_find(tbl, move_key, move_node->data);
                         else {
                                 node = hashtable_find(tbl, move_key, "");
                                 node->data = 0;
                         }
                 }                  }
   
                 free(old_nodes);                  free(old_nodes);
Line 155  void *hashtable_find(struct hashtable *tbl, int64 key, Line 167  void *hashtable_find(struct hashtable *tbl, int64 key,
                 if (nkey == key)                  if (nkey == key)
                         return node;                          return node;
                 if (nkey == 0) {                  if (nkey == 0) {
                        if (!allocate_if_missing)                        if (!data_when_new)
                                 return NULL;                                  return NULL;
                         break;                          break;
                 }                  }
Line 167  void *hashtable_find(struct hashtable *tbl, int64 key, Line 179  void *hashtable_find(struct hashtable *tbl, int64 key,
                 ((struct ht_int64_node*)node)->key = key;                  ((struct ht_int64_node*)node)->key = key;
         else          else
                 node->key = (int32)key;                  node->key = (int32)key;
           node->data = data_when_new;
         tbl->entries++;          tbl->entries++;
         return node;          return node;
   }
   
   #ifndef WORDS_BIGENDIAN
   # define HASH_LITTLE_ENDIAN 1
   # define HASH_BIG_ENDIAN 0
   #else
   # define HASH_LITTLE_ENDIAN 0
   # define HASH_BIG_ENDIAN 1
   #endif
   
   /*
    -------------------------------------------------------------------------------
    lookup3.c, by Bob Jenkins, May 2006, Public Domain.
   
    These are functions for producing 32-bit hashes for hash table lookup.
    hash_word(), hashlittle(), hashlittle2(), hashbig(), mix(), and final()
    are externally useful functions.  Routines to test the hash are included
    if SELF_TEST is defined.  You can use this free for any purpose.  It's in
    the public domain.  It has no warranty.
   
    You probably want to use hashlittle().  hashlittle() and hashbig()
    hash byte arrays.  hashlittle() is is faster than hashbig() on
    little-endian machines.  Intel and AMD are little-endian machines.
    On second thought, you probably want hashlittle2(), which is identical to
    hashlittle() except it returns two 32-bit hashes for the price of one.
    You could implement hashbig2() if you wanted but I haven't bothered here.
   
    If you want to find a hash of, say, exactly 7 integers, do
      a = i1;  b = i2;  c = i3;
      mix(a,b,c);
      a += i4; b += i5; c += i6;
      mix(a,b,c);
      a += i7;
      final(a,b,c);
    then use c as the hash value.  If you have a variable length array of
    4-byte integers to hash, use hash_word().  If you have a byte array (like
    a character string), use hashlittle().  If you have several byte arrays, or
    a mix of things, see the comments above hashlittle().
   
    Why is this so big?  I read 12 bytes at a time into 3 4-byte integers,
    then mix those integers.  This is fast (you can do a lot more thorough
    mixing with 12*3 instructions on 3 integers than you can with 3 instructions
    on 1 byte), but shoehorning those bytes into integers efficiently is messy.
   */
   
   #define hashsize(n) ((uint32_t)1<<(n))
   #define hashmask(n) (hashsize(n)-1)
   #define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
   
   /*
    -------------------------------------------------------------------------------
    mix -- mix 3 32-bit values reversibly.
   
    This is reversible, so any information in (a,b,c) before mix() is
    still in (a,b,c) after mix().
   
    If four pairs of (a,b,c) inputs are run through mix(), or through
    mix() in reverse, there are at least 32 bits of the output that
    are sometimes the same for one pair and different for another pair.
    This was tested for:
    * pairs that differed by one bit, by two bits, in any combination
      of top bits of (a,b,c), or in any combination of bottom bits of
      (a,b,c).
    * "differ" is defined as +, -, ^, or ~^.  For + and -, I transformed
      the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
      is commonly produced by subtraction) look like a single 1-bit
      difference.
    * the base values were pseudorandom, all zero but one bit set, or
      all zero plus a counter that starts at zero.
   
    Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
    satisfy this are
        4  6  8 16 19  4
        9 15  3 18 27 15
       14  9  3  7 17  3
    Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
    for "differ" defined as + with a one-bit base and a two-bit delta.  I
    used http://burtleburtle.net/bob/hash/avalanche.html to choose
    the operations, constants, and arrangements of the variables.
   
    This does not achieve avalanche.  There are input bits of (a,b,c)
    that fail to affect some output bits of (a,b,c), especially of a.  The
    most thoroughly mixed value is c, but it doesn't really even achieve
    avalanche in c.
   
    This allows some parallelism.  Read-after-writes are good at doubling
    the number of bits affected, so the goal of mixing pulls in the opposite
    direction as the goal of parallelism.  I did what I could.  Rotates
    seem to cost as much as shifts on every machine I could lay my hands
    on, and rotates are much kinder to the top and bottom bits, so I used
    rotates.
    -------------------------------------------------------------------------------
   */
   #define mix(a,b,c) \
   { \
     a -= c;  a ^= rot(c, 4);  c += b; \
     b -= a;  b ^= rot(a, 6);  a += c; \
     c -= b;  c ^= rot(b, 8);  b += a; \
     a -= c;  a ^= rot(c,16);  c += b; \
     b -= a;  b ^= rot(a,19);  a += c; \
     c -= b;  c ^= rot(b, 4);  b += a; \
   }
   
   /*
    -------------------------------------------------------------------------------
    final -- final mixing of 3 32-bit values (a,b,c) into c
   
    Pairs of (a,b,c) values differing in only a few bits will usually
    produce values of c that look totally different.  This was tested for
    * pairs that differed by one bit, by two bits, in any combination
      of top bits of (a,b,c), or in any combination of bottom bits of
      (a,b,c).
    * "differ" is defined as +, -, ^, or ~^.  For + and -, I transformed
      the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
      is commonly produced by subtraction) look like a single 1-bit
      difference.
    * the base values were pseudorandom, all zero but one bit set, or
      all zero plus a counter that starts at zero.
   
    These constants passed:
     14 11 25 16 4 14 24
     12 14 25 16 4 14 24
    and these came close:
      4  8 15 26 3 22 24
     10  8 15 26 3 22 24
     11  8 15 26 3 22 24
    -------------------------------------------------------------------------------
   */
   #define final(a,b,c) \
   { \
     c ^= b; c -= rot(b,14); \
     a ^= c; a -= rot(c,11); \
     b ^= a; b -= rot(a,25); \
     c ^= b; c -= rot(b,16); \
     a ^= c; a -= rot(c,4);  \
     b ^= a; b -= rot(a,14); \
     c ^= b; c -= rot(b,24); \
   }
   
   
   /*
    -------------------------------------------------------------------------------
    hashlittle() -- hash a variable-length key into a 32-bit value
      k       : the key (the unaligned variable-length array of bytes)
      length  : the length of the key, counting by bytes
      val2    : IN: can be any 4-byte value OUT: second 32 bit hash.
    Returns a 32-bit value.  Every bit of the key affects every bit of
    the return value.  Two keys differing by one or two bits will have
    totally different hash values.  Note that the return value is better
    mixed than val2, so use that first.
   
    The best hash table sizes are powers of 2.  There is no need to do
    mod a prime (mod is sooo slow!).  If you need less than 32 bits,
    use a bitmask.  For example, if you need only 10 bits, do
      h = (h & hashmask(10));
    In which case, the hash table should have hashsize(10) elements.
   
    If you are hashing n strings (uint8_t **)k, do it like this:
      for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h);
   
    By Bob Jenkins, 2006.  bob_jenkins@burtleburtle.net.  You may use this
    code any way you wish, private, educational, or commercial.  It's free.
   
    Use for hash table lookup, or anything where one collision in 2^^32 is
    acceptable.  Do NOT use for cryptographic purposes.
    -------------------------------------------------------------------------------
   */
   
   uint32_t hashlittle(const void *key, size_t length)
   {
     uint32_t a,b,c;                                          /* internal state */
     union { const void *ptr; size_t i; } u;     /* needed for Mac Powerbook G4 */
   
     /* Set up the internal state */
     a = b = c = 0xdeadbeef + ((uint32_t)length);
   
     u.ptr = key;
     if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
       const uint32_t *k = (const uint32_t *)key;         /* read 32-bit chunks */
       const uint8_t  *k8;
   
       /*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
       while (length > 12)
       {
         a += k[0];
         b += k[1];
         c += k[2];
         mix(a,b,c);
         length -= 12;
         k += 3;
       }
   
       /*----------------------------- handle the last (probably partial) block */
       k8 = (const uint8_t *)k;
       switch(length)
       {
       case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
       case 11: c+=((uint32_t)k8[10])<<16;  /* fall through */
       case 10: c+=((uint32_t)k8[9])<<8;    /* fall through */
       case 9 : c+=k8[8];                   /* fall through */
       case 8 : b+=k[1]; a+=k[0]; break;
       case 7 : b+=((uint32_t)k8[6])<<16;   /* fall through */
       case 6 : b+=((uint32_t)k8[5])<<8;    /* fall through */
       case 5 : b+=k8[4];                   /* fall through */
       case 4 : a+=k[0]; break;
       case 3 : a+=((uint32_t)k8[2])<<16;   /* fall through */
       case 2 : a+=((uint32_t)k8[1])<<8;    /* fall through */
       case 1 : a+=k8[0]; break;
       case 0 : return c;
       }
     } else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
       const uint16_t *k = (const uint16_t *)key;         /* read 16-bit chunks */
       const uint8_t  *k8;
   
       /*--------------- all but last block: aligned reads and different mixing */
       while (length > 12)
       {
         a += k[0] + (((uint32_t)k[1])<<16);
         b += k[2] + (((uint32_t)k[3])<<16);
         c += k[4] + (((uint32_t)k[5])<<16);
         mix(a,b,c);
         length -= 12;
         k += 6;
       }
   
       /*----------------------------- handle the last (probably partial) block */
       k8 = (const uint8_t *)k;
       switch(length)
       {
       case 12: c+=k[4]+(((uint32_t)k[5])<<16);
                b+=k[2]+(((uint32_t)k[3])<<16);
                a+=k[0]+(((uint32_t)k[1])<<16);
                break;
       case 11: c+=((uint32_t)k8[10])<<16;     /* fall through */
       case 10: c+=k[4];
                b+=k[2]+(((uint32_t)k[3])<<16);
                a+=k[0]+(((uint32_t)k[1])<<16);
                break;
       case 9 : c+=k8[8];                      /* fall through */
       case 8 : b+=k[2]+(((uint32_t)k[3])<<16);
                a+=k[0]+(((uint32_t)k[1])<<16);
                break;
       case 7 : b+=((uint32_t)k8[6])<<16;      /* fall through */
       case 6 : b+=k[2];
                a+=k[0]+(((uint32_t)k[1])<<16);
                break;
       case 5 : b+=k8[4];                      /* fall through */
       case 4 : a+=k[0]+(((uint32_t)k[1])<<16);
                break;
       case 3 : a+=((uint32_t)k8[2])<<16;      /* fall through */
       case 2 : a+=k[0];
                break;
       case 1 : a+=k8[0];
                break;
       case 0 : return c;                     /* zero length requires no mixing */
       }
   
     } else {                        /* need to read the key one byte at a time */
       const uint8_t *k = (const uint8_t *)key;
   
       /*--------------- all but the last block: affect some 32 bits of (a,b,c) */
       while (length > 12)
       {
         a += k[0];
         a += ((uint32_t)k[1])<<8;
         a += ((uint32_t)k[2])<<16;
         a += ((uint32_t)k[3])<<24;
         b += k[4];
         b += ((uint32_t)k[5])<<8;
         b += ((uint32_t)k[6])<<16;
         b += ((uint32_t)k[7])<<24;
         c += k[8];
         c += ((uint32_t)k[9])<<8;
         c += ((uint32_t)k[10])<<16;
         c += ((uint32_t)k[11])<<24;
         mix(a,b,c);
         length -= 12;
         k += 12;
       }
   
       /*-------------------------------- last block: affect all 32 bits of (c) */
       switch(length)                   /* all the case statements fall through */
       {
       case 12: c+=((uint32_t)k[11])<<24;
                /* FALLTHROUGH */
       case 11: c+=((uint32_t)k[10])<<16;
                /* FALLTHROUGH */
       case 10: c+=((uint32_t)k[9])<<8;
                /* FALLTHROUGH */
       case 9 : c+=k[8];
                /* FALLTHROUGH */
       case 8 : b+=((uint32_t)k[7])<<24;
                /* FALLTHROUGH */
       case 7 : b+=((uint32_t)k[6])<<16;
                /* FALLTHROUGH */
       case 6 : b+=((uint32_t)k[5])<<8;
                /* FALLTHROUGH */
       case 5 : b+=k[4];
                /* FALLTHROUGH */
       case 4 : a+=((uint32_t)k[3])<<24;
                /* FALLTHROUGH */
       case 3 : a+=((uint32_t)k[2])<<16;
                /* FALLTHROUGH */
       case 2 : a+=((uint32_t)k[1])<<8;
                /* FALLTHROUGH */
       case 1 : a+=k[0];
                break;
       case 0 : return c;
       }
     }
   
     final(a,b,c);
     return c;
 }  }

Removed from v.1.1.1.3  
changed lines
  Added in v.1.1.1.4


FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>