Speed up toku_hash_rehash_everything by remembering the hash function

git-svn-id: file:///svn/tokudb@120 c7de825b-a66e-492c-adef-691d508d4ae1

newbrt/Makefile

@@ -38,13 +38,13 @@ pma-test: pma.o memory.o key.o ybt.o
 pma.o: pma.h yerror.h pma-internal.h memory.h key.h ybt.h brttypes.h  ../include/ydb-constants.h
 ybt.o: ybt.h brttypes.h
 ybt-test: ybt-test.o ybt.o memory.o
-cachetable.o: cachetable.h
+cachetable.o: cachetable.h hashfun.h
 brt-test: ybt.o brt.o hashtable.o pma.o memory.o brt-serialize.o cachetable.o header-io.o ybt.o key.o
 brt-test.o brt.o: brt.h hashtable.h pma.h brttypes.h cachetable.h
 brt-serialize-test.o: pma.h yerror.h brt.h memory.h hashtable.h brttypes.h brt-internal.h
 brt.o: brt.h mdict.h pma.h brttypes.h memory.h brt-internal.h cachetable.h hashtable.h
 mdict.o: pma.h
-hashtable.o: hashtable.h brttypes.h memory.h key.h yerror.h ../include/ydb-constants.h
+hashtable.o: hashtable.h brttypes.h memory.h key.h yerror.h ../include/ydb-constants.h hashfun.h
 memory.o: memory.h
 hashtest: hashtable.o memory.o
 brt-serialize.o: brt.h cachetable.h memory.h mdict.h pma.h brttypes.h brt-internal.h hashtable.h

newbrt/hashfun.h

@@ -5,6 +5,9 @@ static inline unsigned int hash_key (const unsigned char *key, unsigned long key
     unsigned int hash=0;
     for (i=0; i<keylen; i++, key++) {
 	hash *= 16777619;
+	// GCC 4.1.2 -O2 and -O3 translates the following shifts back into the multiply shown on the line above here.
+	// So much for optimizing this multiplication...
+	//hash += (hash<<1) + (hash<<4) + (hash<<7) + (hash<<8) + (hash<<24);
 	hash ^= *key;
     }
     return hash;

newbrt/hashtable.c

@@ -24,8 +24,8 @@ int toku_hashtable_create (HASHTABLE *h) {
     return 0;
 }
 
-static void hash_find_internal (HASHTABLE tab, const unsigned char *key, ITEMLEN keylen, HASHELT *hashelt, HASHELT **prev_ptr) {
-    unsigned int h = hash_key (key, keylen) % tab->arraysize;
+static void hash_find_internal (HASHTABLE tab, unsigned int hash, const unsigned char *key, ITEMLEN keylen, HASHELT *hashelt, HASHELT **prev_ptr) {
+    unsigned int h = hash % tab->arraysize;
     HASHELT he;
     HASHELT *prev = &tab->array[h];
     for (he=*prev; he; prev=&he->next, he=*prev) {
@@ -41,7 +41,7 @@ static void hash_find_internal (HASHTABLE tab, const unsigned char *key, ITEMLEN
     
 int toku_hash_find (HASHTABLE tab, bytevec key, ITEMLEN keylen, bytevec *data, ITEMLEN *datalen) {
     HASHELT he, *prev_ptr;
-    hash_find_internal(tab, key, keylen, &he, &prev_ptr);
+    hash_find_internal(tab, hash_key (key, keylen), key, keylen, &he, &prev_ptr);
     if (he==0) {
 	return -1;
     } else {
@@ -59,7 +59,9 @@ int toku_hash_rehash_everything (HASHTABLE tab, int newarraysize) {
     for (i=0; i<tab->arraysize; i++) {
 	HASHELT he;
 	while ((he=tab->array[i])!=0) {
-	    unsigned int h = hash_key((unsigned char *)he->key, he->keylen)%newarraysize;
+	    //unsigned int hk = hash_key((unsigned char *)he->key, he->keylen);
+	    unsigned int h = he->hash%newarraysize;
+	    //assert(he->hash==hk);
 	    tab->array[i] = he->next;
 	    he->next = newarray[h];
 	    newarray[h] = he;
@@ -75,10 +77,11 @@ int toku_hash_rehash_everything (HASHTABLE tab, int newarraysize) {
 
 int toku_hash_insert (HASHTABLE tab, const void *key, ITEMLEN keylen, const void *val, ITEMLEN vallen)
 {
-    unsigned int h = hash_key (key,keylen)%tab->arraysize;
+    unsigned int hk = hash_key (key,keylen);
+    unsigned int h = hk%tab->arraysize;
     {
 	HASHELT he,*prev_ptr;
-	hash_find_internal(tab, key, keylen, &he, &prev_ptr);
+	hash_find_internal(tab, hk, key, keylen, &he, &prev_ptr);
 	if (he!=0) {
 	    return BRT_ALREADY_THERE;
 	}
@@ -90,6 +93,7 @@ int toku_hash_insert (HASHTABLE tab, const void *key, ITEMLEN keylen, const void
 	he->keylen = keylen;
 	he->val = memdup(val, vallen);
 	he->vallen = vallen;
+	he->hash = hk;
 	he->next = tab->array[h];
 	tab->array[h]=he;
 	tab->n_keys++;
@@ -103,7 +107,7 @@ int toku_hash_insert (HASHTABLE tab, const void *key, ITEMLEN keylen, const void
 int toku_hash_delete (HASHTABLE tab, const void *key, ITEMLEN keylen) {
     HASHELT he, *prev_ptr;
     //printf("%s:%d deleting %s (bucket %d)\n", __FILE__, __LINE__, key, hash_key(key,keylen)%tab->arraysize);
-    hash_find_internal(tab, key, keylen, &he, &prev_ptr);
+    hash_find_internal(tab, hash_key (key, keylen), key, keylen, &he, &prev_ptr);
     if (he==0) return DB_NOTFOUND;
     else {
 	//printf("%s:%d deleting %s %s\n", __FILE__, __LINE__, he->key, he->val);

newbrt/hashtable.h

@@ -31,6 +31,7 @@ typedef struct hashelt *HASHELT;
 struct hashelt {
     char *key; ITEMLEN keylen;  /* key is NULL for empty elements */
     char *val; ITEMLEN vallen;
+    unsigned int hash;
     HASHELT next;
 };