| 1 | /* |
1 | /* |
| 2 | * Copyright (c) 2000-2007 Marc Alexander Lehmann <schmorp@schmorp.de> |
2 | * Copyright (c) 2000-2010 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, |
| … | |
… | |
| 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 ULTRA_FAST |
50 | # if ULTRA_FAST |
| 51 | # define IDX(h) (((h >> (3*8 - HLOG)) - h ) & (HSIZE - 1)) |
51 | # define IDX(h) ((( h >> (3*8 - HLOG)) - h ) & (HSIZE - 1)) |
| 52 | # elif VERY_FAST |
52 | # elif VERY_FAST |
| 53 | # define IDX(h) (((h >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) |
53 | # define IDX(h) ((( h >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) |
| 54 | # else |
54 | # else |
| 55 | # define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) |
55 | # define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) |
| 56 | # endif |
56 | # endif |
| 57 | /*# define IDX(h) ((ip[0] * 121 ^ ip[1] * 33 ^ ip[2] * 1) & (HSIZE-1))*/ |
|
|
| 58 | #endif |
57 | #endif |
| 59 | /* |
58 | /* |
| 60 | * IDX works because it is very similar to a multiplicative hash, e.g. |
59 | * IDX works because it is very similar to a multiplicative hash, e.g. |
| 61 | * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1)) |
60 | * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1)) |
| 62 | * the latter is also quite fast on newer CPUs, and compresses similarly. |
61 | * the latter is also quite fast on newer CPUs, and compresses similarly. |
| … | |
… | |
| 88 | #define expect_true(expr) expect ((expr) != 0, 1) |
87 | #define expect_true(expr) expect ((expr) != 0, 1) |
| 89 | |
88 | |
| 90 | /* |
89 | /* |
| 91 | * compressed format |
90 | * compressed format |
| 92 | * |
91 | * |
| 93 | * 000LLLLL <L+1> ; literal |
92 | * 000LLLLL <L+1> ; literal, L+1=1..33 octets |
| 94 | * LLLooooo oooooooo ; backref L |
93 | * LLLooooo oooooooo ; backref L+1=1..7 octets, o+1=1..4096 offset |
| 95 | * 111ooooo LLLLLLLL oooooooo ; backref L+7 |
94 | * 111ooooo LLLLLLLL oooooooo ; backref L+8 octets, o+1=1..4096 offset |
| 96 | * |
95 | * |
| 97 | */ |
96 | */ |
| 98 | |
97 | |
| 99 | unsigned int |
98 | unsigned int |
| 100 | lzf_compress (const void *const in_data, unsigned int in_len, |
99 | lzf_compress (const void *const in_data, unsigned int in_len, |
| … | |
… | |
| 105 | ) |
104 | ) |
| 106 | { |
105 | { |
| 107 | #if !LZF_STATE_ARG |
106 | #if !LZF_STATE_ARG |
| 108 | LZF_STATE htab; |
107 | LZF_STATE htab; |
| 109 | #endif |
108 | #endif |
| 110 | const u8 **hslot; |
|
|
| 111 | const u8 *ip = (const u8 *)in_data; |
109 | const u8 *ip = (const u8 *)in_data; |
| 112 | u8 *op = (u8 *)out_data; |
110 | u8 *op = (u8 *)out_data; |
| 113 | const u8 *in_end = ip + in_len; |
111 | const u8 *in_end = ip + in_len; |
| 114 | u8 *out_end = op + out_len; |
112 | u8 *out_end = op + out_len; |
| 115 | const u8 *ref; |
113 | const u8 *ref; |
| 116 | |
114 | |
|
|
115 | /* off requires a type wide enough to hold a general pointer difference. |
|
|
116 | * ISO C doesn't have that (size_t might not be enough and ptrdiff_t only |
|
|
117 | * works for differences within a single object). We also assume that no |
|
|
118 | * no bit pattern traps. Since the only platform that is both non-POSIX |
|
|
119 | * and fails to support both assumptions is windows 64 bit, we make a |
|
|
120 | * special workaround for it. |
|
|
121 | */ |
|
|
122 | #if defined (WIN32) && defined (_M_X64) |
|
|
123 | unsigned _int64 off; /* workaround for missing POSIX compliance */ |
|
|
124 | #else |
|
|
125 | unsigned long off; |
|
|
126 | #endif |
| 117 | unsigned int hval; |
127 | unsigned int hval; |
| 118 | unsigned long off; |
|
|
| 119 | int lit; |
128 | int lit; |
| 120 | |
129 | |
| 121 | if (!in_len || !out_len) |
130 | if (!in_len || !out_len) |
| 122 | return 0; |
131 | return 0; |
| 123 | |
132 | |
| 124 | #if INIT_HTAB |
133 | #if INIT_HTAB |
| 125 | memset (htab, 0, sizeof (htab)); |
134 | memset (htab, 0, sizeof (htab)); |
| 126 | # if 0 |
|
|
| 127 | for (hslot = htab; hslot < htab + HSIZE; hslot++) |
|
|
| 128 | *hslot++ = ip; |
|
|
| 129 | # endif |
|
|
| 130 | #endif |
135 | #endif |
| 131 | |
136 | |
| 132 | lit = 0; op++; /* start run */ |
137 | lit = 0; op++; /* start run */ |
| 133 | |
138 | |
| 134 | hval = FRST (ip); |
139 | hval = FRST (ip); |
| 135 | while (ip < in_end - 2) |
140 | while (ip < in_end - 2) |
| 136 | { |
141 | { |
|
|
142 | LZF_HSLOT *hslot; |
|
|
143 | |
| 137 | hval = NEXT (hval, ip); |
144 | hval = NEXT (hval, ip); |
| 138 | hslot = htab + IDX (hval); |
145 | hslot = htab + IDX (hval); |
| 139 | ref = *hslot; *hslot = ip; |
146 | ref = *hslot + LZF_HSLOT_BIAS; *hslot = ip - LZF_HSLOT_BIAS; |
| 140 | |
147 | |
| 141 | if (1 |
148 | if (1 |
| 142 | #if INIT_HTAB && !USE_MEMCPY |
149 | #if INIT_HTAB |
| 143 | && ref < ip /* the next test will actually take care of this, but this is faster */ |
150 | && ref < ip /* the next test will actually take care of this, but this is faster */ |
| 144 | #endif |
151 | #endif |
| 145 | && (off = ip - ref - 1) < MAX_OFF |
152 | && (off = ip - ref - 1) < MAX_OFF |
| 146 | && ip + 4 < in_end |
|
|
| 147 | && ref > (u8 *)in_data |
153 | && ref > (u8 *)in_data |
|
|
154 | && ref[2] == ip[2] |
| 148 | #if STRICT_ALIGN |
155 | #if STRICT_ALIGN |
| 149 | && ref[0] == ip[0] |
156 | && ((ref[1] << 8) | ref[0]) == ((ip[1] << 8) | ip[0]) |
| 150 | && ref[1] == ip[1] |
|
|
| 151 | && ref[2] == ip[2] |
|
|
| 152 | #else |
157 | #else |
| 153 | && *(u16 *)ref == *(u16 *)ip |
158 | && *(u16 *)ref == *(u16 *)ip |
| 154 | && ref[2] == ip[2] |
|
|
| 155 | #endif |
159 | #endif |
| 156 | ) |
160 | ) |
| 157 | { |
161 | { |
| 158 | /* match found at *ref++ */ |
162 | /* match found at *ref++ */ |
| 159 | unsigned int len = 2; |
163 | unsigned int len = 2; |
| 160 | unsigned int maxlen = in_end - ip - len; |
164 | unsigned int maxlen = in_end - ip - len; |
| 161 | maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; |
165 | maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; |
| 162 | |
166 | |
| 163 | if (expect_false (op + 1 + 3 >= out_end)) |
167 | if (expect_false (op + 3 + 1 >= out_end)) /* first a faster conservative test */ |
|
|
168 | if (op - !lit + 3 + 1 >= out_end) /* second the exact but rare test */ |
| 164 | return 0; |
169 | return 0; |
| 165 | |
170 | |
| 166 | op [- lit - 1] = lit - 1; /* stop run */ |
171 | op [- lit - 1] = lit - 1; /* stop run */ |
| 167 | op -= !lit; /* undo run if length is zero */ |
172 | op -= !lit; /* undo run if length is zero */ |
| 168 | |
173 | |
| 169 | for (;;) |
174 | for (;;) |
| … | |
… | |
| 196 | while (len < maxlen && ref[len] == ip[len]); |
201 | while (len < maxlen && ref[len] == ip[len]); |
| 197 | |
202 | |
| 198 | break; |
203 | break; |
| 199 | } |
204 | } |
| 200 | |
205 | |
| 201 | len -= 2; |
206 | len -= 2; /* len is now #octets - 1 */ |
| 202 | ip++; |
207 | ip++; |
| 203 | |
208 | |
| 204 | if (len < 7) |
209 | if (len < 7) |
| 205 | { |
210 | { |
| 206 | *op++ = (off >> 8) + (len << 5); |
211 | *op++ = (off >> 8) + (len << 5); |
| … | |
… | |
| 211 | *op++ = len - 7; |
216 | *op++ = len - 7; |
| 212 | } |
217 | } |
| 213 | |
218 | |
| 214 | *op++ = off; |
219 | *op++ = off; |
| 215 | |
220 | |
|
|
221 | lit = 0; op++; /* start run */ |
|
|
222 | |
|
|
223 | ip += len + 1; |
|
|
224 | |
|
|
225 | if (expect_false (ip >= in_end - 2)) |
|
|
226 | break; |
|
|
227 | |
| 216 | #if ULTRA_FAST || VERY_FAST |
228 | #if ULTRA_FAST || VERY_FAST |
| 217 | ip += len; |
|
|
| 218 | #if VERY_FAST && !ULTRA_FAST |
|
|
| 219 | --ip; |
229 | --ip; |
|
|
230 | # if VERY_FAST && !ULTRA_FAST |
|
|
231 | --ip; |
| 220 | #endif |
232 | # endif |
| 221 | hval = FRST (ip); |
233 | hval = FRST (ip); |
| 222 | |
234 | |
| 223 | hval = NEXT (hval, ip); |
235 | hval = NEXT (hval, ip); |
| 224 | htab[IDX (hval)] = ip; |
236 | htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
| 225 | ip++; |
237 | ip++; |
| 226 | |
238 | |
| 227 | #if VERY_FAST && !ULTRA_FAST |
239 | # if VERY_FAST && !ULTRA_FAST |
| 228 | hval = NEXT (hval, ip); |
240 | hval = NEXT (hval, ip); |
| 229 | htab[IDX (hval)] = ip; |
241 | htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
| 230 | ip++; |
242 | ip++; |
| 231 | #endif |
243 | # endif |
| 232 | #else |
244 | #else |
|
|
245 | ip -= len + 1; |
|
|
246 | |
| 233 | do |
247 | do |
| 234 | { |
248 | { |
| 235 | hval = NEXT (hval, ip); |
249 | hval = NEXT (hval, ip); |
| 236 | htab[IDX (hval)] = ip; |
250 | htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
| 237 | ip++; |
251 | ip++; |
| 238 | } |
252 | } |
| 239 | while (len--); |
253 | while (len--); |
| 240 | #endif |
254 | #endif |
| 241 | lit = 0; op++; /* start run */ |
|
|
| 242 | continue; |
|
|
| 243 | } |
255 | } |
| 244 | |
256 | else |
|
|
257 | { |
| 245 | /* one more literal byte we must copy */ |
258 | /* one more literal byte we must copy */ |
| 246 | |
|
|
| 247 | if (expect_false (op >= out_end)) |
259 | if (expect_false (op >= out_end)) |
| 248 | return 0; |
260 | return 0; |
| 249 | |
261 | |
| 250 | lit++; |
262 | lit++; *op++ = *ip++; |
|
|
263 | |
|
|
264 | if (expect_false (lit == MAX_LIT)) |
|
|
265 | { |
|
|
266 | op [- lit - 1] = lit - 1; /* stop run */ |
|
|
267 | lit = 0; op++; /* start run */ |
|
|
268 | } |
|
|
269 | } |
|
|
270 | } |
|
|
271 | |
|
|
272 | if (op + 3 > out_end) /* at most 3 bytes can be missing here */ |
|
|
273 | return 0; |
|
|
274 | |
|
|
275 | while (ip < in_end) |
|
|
276 | { |
| 251 | *op++ = *ip++; |
277 | lit++; *op++ = *ip++; |
| 252 | |
278 | |
| 253 | if (expect_false (lit == MAX_LIT)) |
279 | if (expect_false (lit == MAX_LIT)) |
| 254 | { |
280 | { |
| 255 | op [- lit - 1] = lit - 1; /* stop run */ |
281 | op [- lit - 1] = lit - 1; /* stop run */ |
| 256 | lit = 0; op++; /* start run */ |
282 | lit = 0; op++; /* start run */ |
| 257 | } |
283 | } |
| 258 | } |
284 | } |
| 259 | |
285 | |
| 260 | if (op + lit + 2 >= out_end) |
|
|
| 261 | return 0; |
|
|
| 262 | |
|
|
| 263 | while (ip < in_end) |
|
|
| 264 | { |
|
|
| 265 | lit++; |
|
|
| 266 | *op++ = *ip++; |
|
|
| 267 | } |
|
|
| 268 | |
|
|
| 269 | op [- lit - 1] = lit - 1; |
286 | op [- lit - 1] = lit - 1; /* end run */ |
|
|
287 | op -= !lit; /* undo run if length is zero */ |
| 270 | |
288 | |
| 271 | return op - (u8 *)out_data; |
289 | return op - (u8 *)out_data; |
| 272 | } |
290 | } |
| 273 | |
291 | |
