blob 7338a40c5964ae6ddfb855465249fc1a2fa5a2a3
1 #include "cache.h"
2 #include "diff.h"
3 #include "diffcore.h"
5 /*
6 * Idea here is very simple.
8 * We have total of (sz-N+1) N-byte overlapping sequences in buf whose
9 * size is sz. If the same N-byte sequence appears in both source and
10 * destination, we say the byte that starts that sequence is shared
11 * between them (i.e. copied from source to destination).
13 * For each possible N-byte sequence, if the source buffer has more
14 * instances of it than the destination buffer, that means the
15 * difference are the number of bytes not copied from source to
16 * destination. If the counts are the same, everything was copied
17 * from source to destination. If the destination has more,
18 * everything was copied, and destination added more.
20 * We are doing an approximation so we do not really have to waste
21 * memory by actually storing the sequence. We just hash them into
22 * somewhere around 2^16 hashbuckets and count the occurrences.
24 * The length of the sequence is arbitrarily set to 8 for now.
27 /* Wild guess at the initial hash size */
28 #define INITIAL_HASH_SIZE 9
30 /* We leave more room in smaller hash but do not let it
31 * grow to have unused hole too much.
33 #define INITIAL_FREE(sz_log2) ((1<<(sz_log2))*(sz_log2-3)/(sz_log2))
35 /* A prime rather carefully chosen between 2^16..2^17, so that
36 * HASHBASE < INITIAL_FREE(17). We want to keep the maximum hashtable
37 * size under the current 2<<17 maximum, which can hold this many
38 * different values before overflowing to hashtable of size 2<<18.
40 #define HASHBASE 107927
42 struct spanhash {
43 unsigned int hashval;
44 unsigned int cnt;
46 struct spanhash_top {
47 int alloc_log2;
48 int free;
49 struct spanhash data[FLEX_ARRAY];
52 static struct spanhash *spanhash_find(struct spanhash_top *top,
53 unsigned int hashval)
55 int sz = 1 << top->alloc_log2;
56 int bucket = hashval & (sz - 1);
57 while (1) {
58 struct spanhash *h = &(top->data[bucket++]);
59 if (!h->cnt)
60 return NULL;
61 if (h->hashval == hashval)
62 return h;
63 if (sz <= bucket)
64 bucket = 0;
68 static struct spanhash_top *spanhash_rehash(struct spanhash_top *orig)
70 struct spanhash_top *new;
71 int i;
72 int osz = 1 << orig->alloc_log2;
73 int sz = osz << 1;
75 new = xmalloc(sizeof(*orig) + sizeof(struct spanhash) * sz);
76 new->alloc_log2 = orig->alloc_log2 + 1;
77 new->free = INITIAL_FREE(new->alloc_log2);
78 memset(new->data, 0, sizeof(struct spanhash) * sz);
79 for (i = 0; i < osz; i++) {
80 struct spanhash *o = &(orig->data[i]);
81 int bucket;
82 if (!o->cnt)
83 continue;
84 bucket = o->hashval & (sz - 1);
85 while (1) {
86 struct spanhash *h = &(new->data[bucket++]);
87 if (!h->cnt) {
88 h->hashval = o->hashval;
89 h->cnt = o->cnt;
90 new->free--;
91 break;
93 if (sz <= bucket)
94 bucket = 0;
97 free(orig);
98 return new;
101 static struct spanhash_top *add_spanhash(struct spanhash_top *top,
102 unsigned int hashval, int cnt)
104 int bucket, lim;
105 struct spanhash *h;
107 lim = (1 << top->alloc_log2);
108 bucket = hashval & (lim - 1);
109 while (1) {
110 h = &(top->data[bucket++]);
111 if (!h->cnt) {
112 h->hashval = hashval;
113 h->cnt = cnt;
114 top->free--;
115 if (top->free < 0)
116 return spanhash_rehash(top);
119 if (h->hashval == hashval) {
120 h->cnt += cnt;
123 if (lim <= bucket)
124 bucket = 0;
128 static struct spanhash_top *hash_chars(unsigned char *buf, unsigned int sz)
130 int i, n;
131 unsigned int accum1, accum2, hashval;
132 struct spanhash_top *hash;
134 i = INITIAL_HASH_SIZE;
135 hash = xmalloc(sizeof(*hash) + sizeof(struct spanhash) * (1<<i));
136 hash->alloc_log2 = i;
137 hash->free = INITIAL_FREE(i);
138 memset(hash->data, 0, sizeof(struct spanhash) * (1<<i));
140 n = 0;
141 accum1 = accum2 = 0;
142 while (sz) {
143 unsigned int c = *buf++;
144 unsigned int old_1 = accum1;
145 sz--;
146 accum1 = (accum1 << 7) ^ (accum2 >> 25);
147 accum2 = (accum2 << 7) ^ (old_1 >> 25);
148 accum1 += c;
149 if (++n < 64 && c != '\n')
150 continue;
151 hashval = (accum1 + accum2 * 0x61) % HASHBASE;
152 hash = add_spanhash(hash, hashval, n);
153 n = 0;
154 accum1 = accum2 = 0;
156 return hash;
159 int diffcore_count_changes(void *src, unsigned long src_size,
160 void *dst, unsigned long dst_size,
161 void **src_count_p,
162 void **dst_count_p,
163 unsigned long delta_limit,
164 unsigned long *src_copied,
167 int i, ssz;
168 struct spanhash_top *src_count, *dst_count;
169 unsigned long sc, la;
171 src_count = dst_count = NULL;
172 if (src_count_p)
173 src_count = *src_count_p;
174 if (!src_count) {
175 src_count = hash_chars(src, src_size);
176 if (src_count_p)
177 *src_count_p = src_count;
179 if (dst_count_p)
180 dst_count = *dst_count_p;
181 if (!dst_count) {
182 dst_count = hash_chars(dst, dst_size);
183 if (dst_count_p)
184 *dst_count_p = dst_count;
186 sc = la = 0;
188 ssz = 1 << src_count->alloc_log2;
189 for (i = 0; i < ssz; i++) {
190 struct spanhash *s = &(src_count->data[i]);
191 struct spanhash *d;
192 unsigned dst_cnt, src_cnt;
193 if (!s->cnt)
194 continue;
195 src_cnt = s->cnt;
196 d = spanhash_find(dst_count, s->hashval);
197 dst_cnt = d ? d->cnt : 0;
198 if (src_cnt < dst_cnt) {
199 la += dst_cnt - src_cnt;
200 sc += src_cnt;
202 else
203 sc += dst_cnt;
206 if (!src_count_p)
207 free(src_count);
208 if (!dst_count_p)
209 free(dst_count);
210 *src_copied = sc;