The universal hash table generator now uses prime table sizes instead of

powers of two. This slows down all operations a little as we now need
to perform division instead of just AND-ing with a mask, but it allows
us to use the new hash functions in hashfunc.h which are significantly
faster than the original ones (at the expense of having bad distribution
modulo non-primes).

Also changed the limit logic to avoid rehashing when the table is already
too small or too large.

diff --git a/lib/hash-test.c b/lib/hash-test.c
index c03d9fa40736c5dd4eb78c96ed1987232b6feed6..2f83561352de834bb4fcbcf49c6e390b3f28830d 100644 (file)
--- a/lib/hash-test.c
+++ b/lib/hash-test.c
@@ -5,7 +5,7 @@
 #include <stdio.h>
 #include <string.h>
 
-#if 1
+#if 0
 
 /* TEST 1: integers */
 
@@ -111,7 +111,7 @@ static void test(void)
   log(L_INFO, "OK");
 }
 
-#elif 0
+#elif 1
 
 /* TEST 3: internal strings + pools */
 
@@ -141,13 +141,13 @@ static void test(void)
 
   pool = mp_new(16384);
   test_init();
-  for (i=0; i<1024; i+=2)
+  for (i=0; i<1048576; i+=2)
     {
       char x[32];
       sprintf(x, "abc%d", i);
       test_new(x);
     }
-  for (i=0; i<1024; i++)
+  for (i=0; i<1048576; i++)
     {
       char x[32];
       struct node *n;

diff --git a/lib/hashtable.h b/lib/hashtable.h
index f96a07fd1f9aa61f685ed260619e36a1044e63ed..17b6d08e1799d43a7ced814da9653e9b37215e2d 100644 (file)
--- a/lib/hashtable.h
+++ b/lib/hashtable.h
@@ -76,8 +76,7 @@
  *  ... and a couple of extra parameters:
  *
  *  HASH_NOCASE                string comparisons should be case-insensitive.
- *  HASH_DEFAULT_SIZE=n        initially, use hash table of `n' entries.
- *                     The `n' has to be a power of two.
+ *  HASH_DEFAULT_SIZE=n        initially, use hash table of approx. `n' entries.
  *  HASH_CONSERVE_SPACE        use as little space as possible.
  *  HASH_FN_BITS=n     The hash function gives only `n' significant bits.
  *  HASH_ATOMIC_TYPE=t Atomic values are of type `t' instead of int.
@@ -128,7 +127,7 @@ typedef struct P(bucket) {
 
 struct P(table) {
   uns hash_size;
-  uns hash_count, hash_max, hash_min, hash_hard_max, hash_mask;
+  uns hash_count, hash_max, hash_min, hash_hard_max;
   P(bucket) **ht;
 } P(table);
 
@@ -284,13 +283,17 @@ static inline void P(free) (void *x)
 
 static void P(alloc_table) (void)
 {
+  T.hash_size = nextprime(T.hash_size);
   T.ht = xmalloc(sizeof(void *) * T.hash_size);
   bzero(T.ht, sizeof(void *) * T.hash_size);
-  T.hash_max = T.hash_size * 2;
-  if (T.hash_max > T.hash_hard_max)
-    T.hash_max = T.hash_hard_max;
-  T.hash_min = T.hash_size / 4;
-  T.hash_mask = T.hash_size - 1;
+  if (2*T.hash_size < T.hash_hard_max)
+    T.hash_max = 2*T.hash_size;
+  else
+    T.hash_max = ~0U;
+  if (T.hash_size/2 > HASH_DEFAULT_SIZE)
+    T.hash_min = T.hash_size/4;
+  else
+    T.hash_min = 0;
 }
 
 static void P(init) (void)
@@ -349,7 +352,7 @@ static void P(rehash) (uns size)
       while (b)
        {
          nb = b->next;
-         h = P(bucket_hash)(b) & T.hash_mask;
+         h = P(bucket_hash)(b) % T.hash_size;
          b->next = newt[h];
          newt[h] = b;
          b = nb;
@@ -362,7 +365,7 @@ static void P(rehash) (uns size)
 static P(node) * P(find) (HASH_KEY_DECL)
 {
   uns h0 = P(hash) (HASH_KEY( ));
-  uns h = h0 & T.hash_mask;
+  uns h = h0 % T.hash_size;
   P(bucket) *b;
 
   for (b=T.ht[h]; b; b=b->next)
@@ -385,7 +388,7 @@ static P(node) * P(new) (HASH_KEY_DECL)
   P(bucket) *b;
 
   h0 = P(hash) (HASH_KEY( ));
-  h = h0 & T.hash_mask;
+  h = h0 % T.hash_size;
   b = P(alloc) (sizeof(struct P(bucket)) + HASH_EXTRA_SIZE(HASH_KEY( )));
   b->next = T.ht[h];
   T.ht[h] = b;
@@ -404,7 +407,7 @@ static P(node) * P(new) (HASH_KEY_DECL)
 static P(node) * P(lookup) (HASH_KEY_DECL)
 {
   uns h0 = P(hash) (HASH_KEY( ));
-  uns h = h0 & T.hash_mask;
+  uns h = h0 % T.hash_size;
   P(bucket) *b;
 
   for (b=T.ht[h]; b; b=b->next)
@@ -435,7 +438,7 @@ static P(node) * P(lookup) (HASH_KEY_DECL)
 static int P(delete) (HASH_KEY_DECL)
 {
   uns h0 = P(hash) (HASH_KEY( ));
-  uns h = h0 & T.hash_mask;
+  uns h = h0 % T.hash_size;
   P(bucket) *b, **bb;
 
   for (bb=&T.ht[h]; b=*bb; bb=&b->next)
@@ -462,7 +465,7 @@ static void P(remove) (P(node) *n)
 {
   P(bucket) *x = SKIP_BACK(struct P(bucket), n, n);
   uns h0 = P(bucket_hash)(x);
-  uns h = h0 & T.hash_mask;
+  uns h = h0 % T.hash_size;
   P(bucket) *b, **bb;
 
   for (bb=&T.ht[h]; (b=*bb) && b != x; bb=&b->next)