| 1 | /* |
1 | /* |
| 2 | * Copyright (c) 2000-2007 Marc Alexander Lehmann <schmorp@schmorp.de> |
2 | * Copyright (c) 2000-2010,2012 Marc Alexander Lehmann <schmorp@schmorp.de> |
| 3 | * |
3 | * |
| 4 | * Redistribution and use in source and binary forms, with or without modifica- |
4 | * Redistribution and use in source and binary forms, with or without modifica- |
| 5 | * tion, are permitted provided that the following conditions are met: |
5 | * tion, are permitted provided that the following conditions are met: |
| 6 | * |
6 | * |
| 7 | * 1. Redistributions of source code must retain the above copyright notice, |
7 | * 1. Redistributions of source code must retain the above copyright notice, |
| 8 | * this list of conditions and the following disclaimer. |
8 | * this list of conditions and the following disclaimer. |
| 9 | * |
9 | * |
| 10 | * 2. Redistributions in binary form must reproduce the above copyright |
10 | * 2. Redistributions in binary form must reproduce the above copyright |
| 11 | * notice, this list of conditions and the following disclaimer in the |
11 | * notice, this list of conditions and the following disclaimer in the |
| 12 | * documentation and/or other materials provided with the distribution. |
12 | * documentation and/or other materials provided with the distribution. |
| 13 | * |
13 | * |
| 14 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED |
14 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED |
| 15 | * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- |
15 | * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- |
| 16 | * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO |
16 | * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO |
| 17 | * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- |
17 | * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- |
| 18 | * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
18 | * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
| … | |
… | |
| 38 | |
38 | |
| 39 | #define HSIZE (1 << (HLOG)) |
39 | #define HSIZE (1 << (HLOG)) |
| 40 | |
40 | |
| 41 | /* |
41 | /* |
| 42 | * don't play with this unless you benchmark! |
42 | * don't play with this unless you benchmark! |
| 43 | * decompression is not dependent on the hash function |
43 | * the data format is not dependent on the hash function. |
| 44 | * the hashing function might seem strange, just believe me |
44 | * the hash function might seem strange, just believe me, |
| 45 | * it works ;) |
45 | * it works ;) |
| 46 | */ |
46 | */ |
| 47 | #ifndef FRST |
47 | #ifndef FRST |
| 48 | # define FRST(p) (((p[0]) << 8) | p[1]) |
48 | # define FRST(p) (((p[0]) << 8) | p[1]) |
| 49 | # define NEXT(v,p) (((v) << 8) | p[2]) |
49 | # define NEXT(v,p) (((v) << 8) | p[2]) |
|
|
50 | # if MULTIPLICATION_IS_SLOW |
|
|
51 | # if ULTRA_FAST |
|
|
52 | # define IDX(h) ((( h >> (3*8 - HLOG)) - h ) & (HSIZE - 1)) |
|
|
53 | # elif VERY_FAST |
|
|
54 | # define IDX(h) ((( h >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) |
|
|
55 | # else |
| 50 | # define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) |
56 | # define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) |
| 51 | /*# define IDX(h) ((ip[0] * 121 ^ ip[1] * 33 ^ ip[2] * 1) & (HSIZE-1))*/ |
57 | # endif |
|
|
58 | # else |
|
|
59 | /* this one was developed with sesse, |
|
|
60 | * and is very similar to the one in snappy. |
|
|
61 | * it does need a modern enough cpu with a fast multiplication. |
|
|
62 | */ |
|
|
63 | # define IDX(h) (((h * 0x1e35a7bdU) >> (32 - HLOG - 8)) & (HSIZE - 1)) |
| 52 | #endif |
64 | # endif |
| 53 | /* |
65 | #endif |
| 54 | * IDX works because it is very similar to a multiplicative hash, e.g. |
|
|
| 55 | * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1)) |
|
|
| 56 | * the latter is also quite fast on newer CPUs, and compresses similarly. |
|
|
| 57 | * |
|
|
| 58 | * the next one is also quite good, albeit slow ;) |
|
|
| 59 | * (int)(cos(h & 0xffffff) * 1e6) |
|
|
| 60 | */ |
|
|
| 61 | |
66 | |
| 62 | #if 0 |
67 | #if 0 |
| 63 | /* original lzv-like hash function, much worse and thus slower */ |
68 | /* original lzv-like hash function, much worse and thus slower */ |
| 64 | # define FRST(p) (p[0] << 5) ^ p[1] |
69 | # define FRST(p) (p[0] << 5) ^ p[1] |
| 65 | # define NEXT(v,p) ((v) << 5) ^ p[2] |
70 | # define NEXT(v,p) ((v) << 5) ^ p[2] |
| 66 | # define IDX(h) ((h) & (HSIZE - 1)) |
71 | # define IDX(h) ((h) & (HSIZE - 1)) |
| 67 | #endif |
72 | #endif |
| 68 | |
73 | |
| 69 | #define MAX_LIT (1 << 5) |
74 | #if __GNUC__ >= 3 |
| 70 | #define MAX_OFF (1 << 13) |
75 | # define expect(expr,value) __builtin_expect ((expr),(value)) |
| 71 | #define MAX_REF ((1 << 8) + (1 << 3)) |
76 | # define inline inline |
| 72 | |
77 | #else |
| 73 | #if (__i386 || __amd64) && __GNUC__ >= 3 |
78 | # define expect(expr,value) (expr) |
| 74 | # define lzf_movsb(dst, src, len) \ |
79 | # define inline static |
| 75 | asm ("rep movsb" \ |
|
|
| 76 | : "=D" (dst), "=S" (src), "=c" (len) \ |
|
|
| 77 | : "0" (dst), "1" (src), "2" (len)); |
|
|
| 78 | #endif |
80 | #endif |
|
|
81 | |
|
|
82 | #define expect_false(expr) expect ((expr) != 0, 0) |
|
|
83 | #define expect_true(expr) expect ((expr) != 0, 1) |
| 79 | |
84 | |
| 80 | /* |
85 | /* |
| 81 | * compressed format |
86 | * compressed format |
| 82 | * |
87 | * |
| 83 | * 000LLLLL <L+1> ; literal |
88 | * 000LLLLL <L+1> ; literal, L+1=1..33 octets |
| 84 | * LLLooooo oooooooo ; backref L |
89 | * LLLooooo oooooooo ; backref L+1=1..7 octets, o+1=1..4096 offset |
| 85 | * 111ooooo LLLLLLLL oooooooo ; backref L+7 |
90 | * 111ooooo LLLLLLLL oooooooo ; backref L+8 octets, o+1=1..4096 offset |
| 86 | * |
91 | * |
| 87 | */ |
92 | */ |
| 88 | |
93 | |
| 89 | unsigned int |
94 | unsigned int |
| 90 | lzf_compress (const void *const in_data, unsigned int in_len, |
95 | lzf_compress (const void *const in_data, unsigned int in_len, |
| … | |
… | |
| 95 | ) |
100 | ) |
| 96 | { |
101 | { |
| 97 | #if !LZF_STATE_ARG |
102 | #if !LZF_STATE_ARG |
| 98 | LZF_STATE htab; |
103 | LZF_STATE htab; |
| 99 | #endif |
104 | #endif |
| 100 | const u8 **hslot; |
|
|
| 101 | const u8 *ip = (const u8 *)in_data; |
105 | const u8 *ip = (const u8 *)in_data; |
| 102 | u8 *op = (u8 *)out_data; |
106 | u8 *op = (u8 *)out_data; |
| 103 | const u8 *in_end = ip + in_len; |
107 | const u8 *in_end = ip + in_len; |
| 104 | u8 *out_end = op + out_len; |
108 | u8 *out_end = op + out_len; |
| 105 | const u8 *ref; |
109 | const u8 *ref; |
| 106 | |
110 | |
| 107 | unsigned int hval = FRST (ip); |
111 | /* off requires a type wide enough to hold a general pointer difference. |
|
|
112 | * ISO C doesn't have that (size_t might not be enough and ptrdiff_t only |
|
|
113 | * works for differences within a single object). We also assume that |
|
|
114 | * no bit pattern traps. Since the only platform that is both non-POSIX |
|
|
115 | * and fails to support both assumptions is windows 64 bit, we make a |
|
|
116 | * special workaround for it. |
|
|
117 | */ |
|
|
118 | #if defined (_WIN32) && defined (_M_X64) |
|
|
119 | /* workaround for missing POSIX compliance */ |
|
|
120 | #if __GNUC__ |
|
|
121 | unsigned long long off; |
|
|
122 | #else |
|
|
123 | unsigned __int64 off; |
|
|
124 | #endif |
|
|
125 | #else |
| 108 | unsigned long off; |
126 | unsigned long off; |
| 109 | int lit = 0; |
127 | #endif |
|
|
128 | unsigned int hval; |
|
|
129 | int lit; |
|
|
130 | |
|
|
131 | if (!in_len || !out_len) |
|
|
132 | return 0; |
| 110 | |
133 | |
| 111 | #if INIT_HTAB |
134 | #if INIT_HTAB |
| 112 | memset (htab, 0, sizeof (htab)); |
135 | memset (htab, 0, sizeof (LZF_STATE)); |
| 113 | # if 0 |
|
|
| 114 | for (hslot = htab; hslot < htab + HSIZE; hslot++) |
|
|
| 115 | *hslot++ = ip; |
|
|
| 116 | # endif |
136 | #endif |
| 117 | #endif |
|
|
| 118 | |
137 | |
| 119 | for (;;) |
138 | lit = 0; op++; /* start run */ |
|
|
139 | |
|
|
140 | hval = FRST (ip); |
|
|
141 | while (ip < in_end - 2) |
| 120 | { |
142 | { |
| 121 | if (ip < in_end - 2) |
143 | LZF_HSLOT *hslot; |
|
|
144 | |
|
|
145 | hval = NEXT (hval, ip); |
|
|
146 | hslot = htab + IDX (hval); |
|
|
147 | ref = *hslot + LZF_HSLOT_BIAS; *hslot = ip - LZF_HSLOT_BIAS; |
|
|
148 | |
|
|
149 | if (1 |
|
|
150 | #if INIT_HTAB |
|
|
151 | && ref < ip /* the next test will actually take care of this, but this is faster */ |
|
|
152 | #endif |
|
|
153 | && (off = ip - ref - 1) < LZF_MAX_OFF |
|
|
154 | && ref > (u8 *)in_data |
|
|
155 | && ref[2] == ip[2] |
|
|
156 | #if STRICT_ALIGN |
|
|
157 | && ((ref[1] << 8) | ref[0]) == ((ip[1] << 8) | ip[0]) |
|
|
158 | #else |
|
|
159 | && *(u16 *)ref == *(u16 *)ip |
|
|
160 | #endif |
|
|
161 | ) |
| 122 | { |
162 | { |
| 123 | hval = NEXT (hval, ip); |
|
|
| 124 | hslot = htab + IDX (hval); |
|
|
| 125 | ref = *hslot; *hslot = ip; |
|
|
| 126 | |
|
|
| 127 | if (1 |
|
|
| 128 | #if INIT_HTAB && !USE_MEMCPY |
|
|
| 129 | && ref < ip /* the next test will actually take care of this, but this is faster */ |
|
|
| 130 | #endif |
|
|
| 131 | && (off = ip - ref - 1) < MAX_OFF |
|
|
| 132 | && ip + 4 < in_end |
|
|
| 133 | && ref > (u8 *)in_data |
|
|
| 134 | #if STRICT_ALIGN |
|
|
| 135 | && ref[0] == ip[0] |
|
|
| 136 | && ref[1] == ip[1] |
|
|
| 137 | && ref[2] == ip[2] |
|
|
| 138 | #else |
|
|
| 139 | && *(u16 *)ref == *(u16 *)ip |
|
|
| 140 | && ref[2] == ip[2] |
|
|
| 141 | #endif |
|
|
| 142 | ) |
|
|
| 143 | { |
|
|
| 144 | /* match found at *ref++ */ |
163 | /* match found at *ref++ */ |
| 145 | unsigned int len = 2; |
164 | unsigned int len = 2; |
| 146 | unsigned int maxlen = in_end - ip - len; |
165 | unsigned int maxlen = in_end - ip - len; |
| 147 | maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; |
166 | maxlen = maxlen > LZF_MAX_REF ? LZF_MAX_REF : maxlen; |
| 148 | |
167 | |
| 149 | if (op + lit + 1 + 3 >= out_end) |
168 | if (expect_false (op + 3 + 1 >= out_end)) /* first a faster conservative test */ |
|
|
169 | if (op - !lit + 3 + 1 >= out_end) /* second the exact but rare test */ |
| 150 | return 0; |
170 | return 0; |
|
|
171 | |
|
|
172 | op [- lit - 1] = lit - 1; /* stop run */ |
|
|
173 | op -= !lit; /* undo run if length is zero */ |
|
|
174 | |
|
|
175 | for (;;) |
|
|
176 | { |
|
|
177 | if (expect_true (maxlen > 16)) |
|
|
178 | { |
|
|
179 | len++; if (ref [len] != ip [len]) break; |
|
|
180 | len++; if (ref [len] != ip [len]) break; |
|
|
181 | len++; if (ref [len] != ip [len]) break; |
|
|
182 | len++; if (ref [len] != ip [len]) break; |
|
|
183 | |
|
|
184 | len++; if (ref [len] != ip [len]) break; |
|
|
185 | len++; if (ref [len] != ip [len]) break; |
|
|
186 | len++; if (ref [len] != ip [len]) break; |
|
|
187 | len++; if (ref [len] != ip [len]) break; |
|
|
188 | |
|
|
189 | len++; if (ref [len] != ip [len]) break; |
|
|
190 | len++; if (ref [len] != ip [len]) break; |
|
|
191 | len++; if (ref [len] != ip [len]) break; |
|
|
192 | len++; if (ref [len] != ip [len]) break; |
|
|
193 | |
|
|
194 | len++; if (ref [len] != ip [len]) break; |
|
|
195 | len++; if (ref [len] != ip [len]) break; |
|
|
196 | len++; if (ref [len] != ip [len]) break; |
|
|
197 | len++; if (ref [len] != ip [len]) break; |
|
|
198 | } |
| 151 | |
199 | |
| 152 | do |
200 | do |
| 153 | len++; |
201 | len++; |
| 154 | while (len < maxlen && ref[len] == ip[len]); |
202 | while (len < maxlen && ref[len] == ip[len]); |
| 155 | |
203 | |
| 156 | if (lit) |
|
|
| 157 | { |
204 | break; |
| 158 | *op++ = lit - 1; |
|
|
| 159 | lit = -lit; |
|
|
| 160 | do |
|
|
| 161 | *op++ = ip[lit]; |
|
|
| 162 | while (++lit); |
|
|
| 163 | } |
205 | } |
| 164 | |
206 | |
|
|
207 | len -= 2; /* len is now #octets - 1 */ |
|
|
208 | ip++; |
|
|
209 | |
|
|
210 | if (len < 7) |
|
|
211 | { |
|
|
212 | *op++ = (off >> 8) + (len << 5); |
|
|
213 | } |
|
|
214 | else |
|
|
215 | { |
|
|
216 | *op++ = (off >> 8) + ( 7 << 5); |
| 165 | len -= 2; |
217 | *op++ = len - 7; |
|
|
218 | } |
|
|
219 | |
|
|
220 | *op++ = off; |
|
|
221 | |
|
|
222 | lit = 0; op++; /* start run */ |
|
|
223 | |
|
|
224 | ip += len + 1; |
|
|
225 | |
|
|
226 | if (expect_false (ip >= in_end - 2)) |
|
|
227 | break; |
|
|
228 | |
|
|
229 | #if ULTRA_FAST || VERY_FAST |
|
|
230 | --ip; |
|
|
231 | # if VERY_FAST && !ULTRA_FAST |
|
|
232 | --ip; |
|
|
233 | # endif |
|
|
234 | hval = FRST (ip); |
|
|
235 | |
|
|
236 | hval = NEXT (hval, ip); |
|
|
237 | htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
|
|
238 | ip++; |
|
|
239 | |
|
|
240 | # if VERY_FAST && !ULTRA_FAST |
|
|
241 | hval = NEXT (hval, ip); |
|
|
242 | htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
|
|
243 | ip++; |
|
|
244 | # endif |
|
|
245 | #else |
|
|
246 | ip -= len + 1; |
|
|
247 | |
|
|
248 | do |
|
|
249 | { |
|
|
250 | hval = NEXT (hval, ip); |
|
|
251 | htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
| 166 | ip++; |
252 | ip++; |
| 167 | |
|
|
| 168 | if (len < 7) |
|
|
| 169 | { |
|
|
| 170 | *op++ = (off >> 8) + (len << 5); |
|
|
| 171 | } |
253 | } |
| 172 | else |
|
|
| 173 | { |
|
|
| 174 | *op++ = (off >> 8) + ( 7 << 5); |
|
|
| 175 | *op++ = len - 7; |
|
|
| 176 | } |
|
|
| 177 | |
|
|
| 178 | *op++ = off; |
|
|
| 179 | |
|
|
| 180 | #if ULTRA_FAST || VERY_FAST |
|
|
| 181 | ip += len; |
|
|
| 182 | #if VERY_FAST && !ULTRA_FAST |
|
|
| 183 | --ip; |
|
|
| 184 | #endif |
|
|
| 185 | hval = FRST (ip); |
|
|
| 186 | |
|
|
| 187 | hval = NEXT (hval, ip); |
|
|
| 188 | htab[IDX (hval)] = ip; |
|
|
| 189 | ip++; |
|
|
| 190 | |
|
|
| 191 | #if VERY_FAST && !ULTRA_FAST |
|
|
| 192 | hval = NEXT (hval, ip); |
|
|
| 193 | htab[IDX (hval)] = ip; |
|
|
| 194 | ip++; |
|
|
| 195 | #endif |
|
|
| 196 | #else |
|
|
| 197 | do |
|
|
| 198 | { |
|
|
| 199 | hval = NEXT (hval, ip); |
|
|
| 200 | htab[IDX (hval)] = ip; |
|
|
| 201 | ip++; |
|
|
| 202 | } |
|
|
| 203 | while (len--); |
254 | while (len--); |
| 204 | #endif |
255 | #endif |
| 205 | continue; |
|
|
| 206 | } |
|
|
| 207 | } |
256 | } |
| 208 | else if (ip == in_end) |
257 | else |
| 209 | break; |
|
|
| 210 | |
|
|
| 211 | /* one more literal byte we must copy */ |
|
|
| 212 | lit++; |
|
|
| 213 | ip++; |
|
|
| 214 | |
|
|
| 215 | if (lit == MAX_LIT) |
|
|
| 216 | { |
258 | { |
| 217 | if (op + 1 + MAX_LIT >= out_end) |
259 | /* one more literal byte we must copy */ |
|
|
260 | if (expect_false (op >= out_end)) |
| 218 | return 0; |
261 | return 0; |
| 219 | |
262 | |
| 220 | *op++ = MAX_LIT - 1; |
263 | lit++; *op++ = *ip++; |
| 221 | |
264 | |
| 222 | #ifdef lzf_movsb |
265 | if (expect_false (lit == LZF_MAX_LIT)) |
| 223 | ip -= lit; |
|
|
| 224 | lzf_movsb (op, ip, lit); |
|
|
| 225 | #else |
|
|
| 226 | lit = -lit; |
|
|
| 227 | do |
266 | { |
| 228 | *op++ = ip[lit]; |
267 | op [- lit - 1] = lit - 1; /* stop run */ |
| 229 | while (++lit); |
268 | lit = 0; op++; /* start run */ |
| 230 | #endif |
269 | } |
| 231 | } |
270 | } |
| 232 | } |
271 | } |
| 233 | |
272 | |
| 234 | if (lit) |
273 | if (op + 3 > out_end) /* at most 3 bytes can be missing here */ |
|
|
274 | return 0; |
|
|
275 | |
|
|
276 | while (ip < in_end) |
| 235 | { |
277 | { |
| 236 | if (op + lit + 1 >= out_end) |
278 | lit++; *op++ = *ip++; |
| 237 | return 0; |
|
|
| 238 | |
279 | |
| 239 | *op++ = lit - 1; |
280 | if (expect_false (lit == LZF_MAX_LIT)) |
| 240 | #ifdef lzf_movsb |
281 | { |
| 241 | ip -= lit; |
282 | op [- lit - 1] = lit - 1; /* stop run */ |
| 242 | lzf_movsb (op, ip, lit); |
283 | lit = 0; op++; /* start run */ |
| 243 | #else |
284 | } |
| 244 | lit = -lit; |
|
|
| 245 | do |
|
|
| 246 | *op++ = ip[lit]; |
|
|
| 247 | while (++lit); |
|
|
| 248 | #endif |
|
|
| 249 | } |
285 | } |
| 250 | |
286 | |
|
|
287 | op [- lit - 1] = lit - 1; /* end run */ |
|
|
288 | op -= !lit; /* undo run if length is zero */ |
|
|
289 | |
| 251 | return op - (u8 *) out_data; |
290 | return op - (u8 *)out_data; |
| 252 | } |
291 | } |
| 253 | |
292 | |
