#include "rsync.h"
extern int verbose;
-extern int am_server;
extern int do_progress;
extern int checksum_seed;
extern int append_mode;
int updating_basis_file;
-typedef unsigned short tag;
-
-#define TABLESIZE (1<<16)
-#define NULL_TAG (-1)
-
static int false_alarms;
-static int tag_hits;
+static int hash_hits;
static int matches;
static int64 data_transfer;
static int total_false_alarms;
-static int total_tag_hits;
+static int total_hash_hits;
static int total_matches;
extern struct stats stats;
-struct target {
- tag t;
- int32 i;
-};
-
-static struct target *targets;
-
-static int32 *tag_table;
-
-#define gettag2(s1,s2) (((s1) + (s2)) & 0xFFFF)
-#define gettag(sum) gettag2((sum)&0xFFFF,(sum)>>16)
-
-static int compare_targets(struct target *t1,struct target *t2)
-{
- return (int)t1->t - (int)t2->t;
-}
+static uint32 tablesize;
+static int32 *hash_table;
+#define SUM2HASH(sum) ((sum)%tablesize)
static void build_hash_table(struct sum_struct *s)
{
int32 i;
+ uint32 prior_size = tablesize;
+
+ /* Dynamically calculate the hash table size so that the hash load
+ * for big files is about 80%. This number must be odd or s2 will
+ * not be able to span the entire set. */
+ tablesize = (uint32)(s->count/8) * 10 + 11;
+ if (tablesize < 65537)
+ tablesize = 65537; /* a prime number */
+ if (tablesize != prior_size) {
+ free(hash_table);
+ hash_table = new_array(int32, tablesize);
+ if (!hash_table)
+ out_of_memory("build_hash_table");
+ }
- if (!tag_table)
- tag_table = new_array(int32, TABLESIZE);
-
- targets = new_array(struct target, s->count);
- if (!tag_table || !targets)
- out_of_memory("build_hash_table");
+ memset(hash_table, 0xFF, tablesize * sizeof hash_table[0]);
for (i = 0; i < s->count; i++) {
- targets[i].i = i;
- targets[i].t = gettag(s->sums[i].sum1);
+ uint32 t = SUM2HASH(s->sums[i].sum1);
+ s->sums[i].chain = hash_table[t];
+ hash_table[t] = i;
}
-
- qsort(targets,s->count,sizeof(targets[0]),(int (*)())compare_targets);
-
- for (i = 0; i < TABLESIZE; i++)
- tag_table[i] = NULL_TAG;
-
- for (i = s->count; i-- > 0; )
- tag_table[targets[i].t] = i;
}
sum_update(map_ptr(buf, last_match + j, n1), n1);
}
-
if (i >= 0)
last_match = offset + s->sums[i].len;
else
}
do {
- tag t = gettag2(s1,s2);
int done_csum2 = 0;
- int32 j = tag_table[t];
+ int32 i;
if (verbose > 4)
rprintf(FINFO,"offset=%.0f sum=%08x\n",(double)offset,sum);
- if (j == NULL_TAG)
- goto null_tag;
+ i = hash_table[SUM2HASH(sum)];
+ if (i < 0)
+ goto null_hash;
- sum = (s1 & 0xffff) | (s2 << 16);
- tag_hits++;
+ hash_hits++;
do {
- int32 l, i = targets[j].i;
+ int32 l;
if (sum != s->sums[i].sum1)
continue;
&& !(s->sums[i].flags & SUMFLG_SAME_OFFSET))
continue;
- if (verbose > 3)
- rprintf(FINFO,"potential match at %.0f target=%.0f %.0f sum=%08x\n",
- (double)offset,(double)j,(double)i,sum);
+ if (verbose > 3) {
+ rprintf(FINFO,
+ "potential match at %.0f i=%ld sum=%08x\n",
+ (double)offset, (long)i, sum);
+ }
if (!done_csum2) {
map = (schar *)map_ptr(buf,offset,l);
* one with an identical offset, so we prefer that over
* the following want_i optimization. */
if (updating_basis_file) {
- do {
- int32 i2 = targets[j].i;
+ int32 i2;
+ for (i2 = i; i2 >= 0; i2 = s->sums[i2].chain) {
if (s->sums[i2].offset != offset)
continue;
if (i2 != i) {
* both the sender and the receiver. */
s->sums[i].flags |= SUMFLG_SAME_OFFSET;
goto set_want_i;
- } while (++j < s->count && targets[j].t == t);
+ }
}
/* we've found a match, but now check to see
s2 = sum >> 16;
matches++;
break;
- } while (++j < s->count && targets[j].t == t);
+ } while ((i = s->sums[i].chain) >= 0);
- null_tag:
+ null_hash:
backup = offset - last_match;
/* We sometimes read 1 byte prior to last_match... */
if (backup < 0)
s2 += s1;
} else
--k;
+ sum = (s1 & 0xffff) | (s2 << 16);
/* By matching early we avoid re-reading the
data 3 times in the case where a token
last_match = 0;
false_alarms = 0;
- tag_hits = 0;
+ hash_hits = 0;
matches = 0;
data_transfer = 0;
rprintf(FINFO,"sending file_sum\n");
write_buf(f,file_sum,MD4_SUM_LENGTH);
- if (targets) {
- free(targets);
- targets=NULL;
- }
-
if (verbose > 2)
- rprintf(FINFO, "false_alarms=%d tag_hits=%d matches=%d\n",
- false_alarms, tag_hits, matches);
+ rprintf(FINFO, "false_alarms=%d hash_hits=%d matches=%d\n",
+ false_alarms, hash_hits, matches);
- total_tag_hits += tag_hits;
+ total_hash_hits += hash_hits;
total_false_alarms += false_alarms;
total_matches += matches;
stats.literal_data += data_transfer;
return;
rprintf(FINFO,
- "total: matches=%d tag_hits=%d false_alarms=%d data=%.0f\n",
- total_matches,total_tag_hits,
- total_false_alarms,
+ "total: matches=%d hash_hits=%d false_alarms=%d data=%.0f\n",
+ total_matches, total_hash_hits, total_false_alarms,
(double)stats.literal_data);
}
Matt McCutchen's Web Site / rsync / rsync.git / blobdiff