*
* Copyright (C) 1996 Andrew Tridgell
* Copyright (C) 1996 Paul Mackerras
- * Copyright (C) 2003-2007 Wayne Davison
+ * Copyright (C) 2003-2009 Wayne Davison
*
* 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
extern struct stats stats;
-#define TABLESIZE (1<<16)
+#define TRADITIONAL_TABLESIZE (1<<16)
+static uint32 tablesize;
static int32 *hash_table;
#define SUM2HASH2(s1,s2) (((s1) + (s2)) & 0xFFFF)
#define SUM2HASH(sum) SUM2HASH2((sum)&0xFFFF,(sum)>>16)
-static int32 build_hash_table(struct sum_struct *s, int32 start)
-{
- int32 i, end = s->count;
+#define BIG_SUM2HASH(sum) ((sum)%tablesize)
- if (!hash_table) {
- hash_table = new_array(int32, TABLESIZE);
+static void build_hash_table(struct sum_struct *s)
+{
+ static uint32 alloc_size;
+ int32 i;
+
+ /* Dynamically calculate the hash table size so that the hash load
+ * for big files is about 80%. A number greater than the traditional
+ * size must be odd or s2 will not be able to span the entire set. */
+ tablesize = (uint32)(s->count/8) * 10 + 11;
+ if (tablesize < TRADITIONAL_TABLESIZE)
+ tablesize = TRADITIONAL_TABLESIZE;
+ if (tablesize > alloc_size || tablesize < alloc_size - 16*1024) {
+ if (hash_table)
+ free(hash_table);
+ hash_table = new_array(int32, tablesize);
if (!hash_table)
out_of_memory("build_hash_table");
+ alloc_size = tablesize;
}
- memset(hash_table, 0xFF, TABLESIZE * sizeof hash_table[0]);
-
- if (end - start > TABLESIZE*8/10)
- end = start + TABLESIZE*8/10;
-
- for (i = start; i < end; i++) {
- uint32 t = SUM2HASH(s->sums[i].sum1);
- s->sums[i].chain = hash_table[t];
- hash_table[t] = i;
- }
+ memset(hash_table, 0xFF, tablesize * sizeof hash_table[0]);
- if (verbose > 2) {
- rprintf(FINFO, "built hash table for entries %ld - %ld\n",
- (long)start, (long)end - 1);
+ if (tablesize == TRADITIONAL_TABLESIZE) {
+ for (i = 0; i < s->count; i++) {
+ uint32 t = SUM2HASH(s->sums[i].sum1);
+ s->sums[i].chain = hash_table[t];
+ hash_table[t] = i;
+ }
+ } else {
+ for (i = 0; i < s->count; i++) {
+ uint32 t = BIG_SUM2HASH(s->sums[i].sum1);
+ s->sums[i].chain = hash_table[t];
+ hash_table[t] = i;
+ }
}
-
- return end;
}
static OFF_T last_match;
-/**
- * Transmit a literal and/or match token.
+/* Transmit a literal and/or match token.
*
* This delightfully-named function is called either when we find a
* match and need to transmit all the unmatched data leading up to it,
* transmit it. As a result of this second case, it is called even if
* we have not matched at all!
*
- * @param i If >0, the number of a matched token. If 0, indicates we
- * have only literal data.
- **/
+ * If i >= 0, the number of a matched token. If < 0, indicates we have
+ * only literal data. A -1 will send a 0-token-int too, and a -2 sends
+ * only literal data, w/o any token-int. */
static void matched(int f, struct sum_struct *s, struct map_struct *buf,
OFF_T offset, int32 i)
{
static void hash_search(int f,struct sum_struct *s,
struct map_struct *buf, OFF_T len)
{
- OFF_T offset, end, reset = 0;
- int32 k, want_i, backup, sum_pos = 0;
+ OFF_T offset, aligned_offset, end;
+ int32 k, want_i, aligned_i, backup;
char sum2[SUM_LENGTH];
uint32 s1, s2, sum;
int more;
if (verbose > 3)
rprintf(FINFO, "sum=%.8x k=%ld\n", sum, (long)k);
- offset = 0;
+ offset = aligned_offset = aligned_i = 0;
end = len + 1 - s->sums[s->count-1].len;
int done_csum2 = 0;
int32 i;
- if (offset >= reset) {
- sum_pos = build_hash_table(s, sum_pos);
- reset = sum_pos * s->blength;
- }
-
if (verbose > 4) {
rprintf(FINFO, "offset=%.0f sum=%04x%04x\n",
(double)offset, s2 & 0xFFFF, s1 & 0xFFFF);
}
- i = hash_table[SUM2HASH2(s1,s2)];
- if (i < 0)
- goto null_hash;
+ if (tablesize == TRADITIONAL_TABLESIZE) {
+ if ((i = hash_table[SUM2HASH2(s1,s2)]) < 0)
+ goto null_hash;
+ sum = (s1 & 0xffff) | (s2 << 16);
+ } else {
+ sum = (s1 & 0xffff) | (s2 << 16);
+ if ((i = hash_table[BIG_SUM2HASH(sum)]) < 0)
+ goto null_hash;
+ }
- sum = (s1 & 0xffff) | (s2 << 16);
hash_hits++;
do {
int32 l;
/* When updating in-place, the best possible match is
* one with an identical offset, so we prefer that over
- * the following want_i optimization. */
+ * the adjacent want_i optimization. */
if (updating_basis_file) {
- int32 i2;
- for (i2 = i; i2 >= 0; i2 = s->sums[i2].chain) {
- if (s->sums[i2].offset != offset)
- continue;
- if (i2 != i) {
- if (sum != s->sums[i2].sum1)
- break;
- if (memcmp(sum2, s->sums[i2].sum2,
- s->s2length) != 0)
- break;
- i = i2;
+ /* All the generator's chunks start at blength boundaries. */
+ while (aligned_offset < offset) {
+ aligned_offset += s->blength;
+ aligned_i++;
+ }
+ if (offset == aligned_offset && aligned_i < s->count) {
+ if (i != aligned_i) {
+ if (sum != s->sums[aligned_i].sum1
+ || l != s->sums[aligned_i].len
+ || memcmp(sum2, s->sums[aligned_i].sum2, s->s2length) != 0)
+ goto check_want_i;
+ i = aligned_i;
}
- /* This chunk was at the same offset on
- * both the sender and the receiver. */
+ /* This identical chunk is in the same spot in the old and new file. */
s->sums[i].flags |= SUMFLG_SAME_OFFSET;
- goto set_want_i;
+ want_i = i;
}
}
+ check_want_i:
/* we've found a match, but now check to see
* if want_i can hint at a better match. */
if (i != want_i && want_i < s->count
* will be happy */
i = want_i;
}
- set_want_i:
want_i = i + 1;
matched(f,s,buf,offset,i);
}
if (len > 0 && s->count > 0) {
+ build_hash_table(s);
+
+ if (verbose > 2)
+ rprintf(FINFO,"built hash table\n");
+
hash_search(f, s, buf, len);
if (verbose > 2)