1 | /* |
2 | * Copyright (c) 2000-2010 Marc Alexander Lehmann <[email protected]> |
3 | * |
4 | * Redistribution and use in source and binary forms, with or without modifica- |
5 | * tion, are permitted provided that the following conditions are met: |
6 | * |
7 | * 1. Redistributions of source code must retain the above copyright notice, |
8 | * this list of conditions and the following disclaimer. |
9 | * |
10 | * 2. Redistributions in binary form must reproduce the above copyright |
11 | * notice, this list of conditions and the following disclaimer in the |
12 | * documentation and/or other materials provided with the distribution. |
13 | * |
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- |
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- |
18 | * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
19 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; |
20 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, |
21 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- |
22 | * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED |
23 | * OF THE POSSIBILITY OF SUCH DAMAGE. |
24 | * |
25 | * Alternatively, the contents of this file may be used under the terms of |
26 | * the GNU General Public License ("GPL") version 2 or any later version, |
27 | * in which case the provisions of the GPL are applicable instead of |
28 | * the above. If you wish to allow the use of your version of this file |
29 | * only under the terms of the GPL and not to allow others to use your |
30 | * version of this file under the BSD license, indicate your decision |
31 | * by deleting the provisions above and replace them with the notice |
32 | * and other provisions required by the GPL. If you do not delete the |
33 | * provisions above, a recipient may use your version of this file under |
34 | * either the BSD or the GPL. |
35 | */ |
36 | |
37 | #include "lzfP.h" |
38 | |
39 | #define HSIZE (1 << (HLOG)) |
40 | |
41 | /* |
42 | * don't play with this unless you benchmark! |
43 | * the data format is not dependent on the hash function. |
44 | * the hash function might seem strange, just believe me, |
45 | * it works ;) |
46 | */ |
47 | #ifndef FRST |
48 | # define FRST(p) (((p[0]) << 8) | p[1]) |
49 | # define NEXT(v,p) (((v) << 8) | p[2]) |
50 | # if ULTRA_FAST |
51 | # define IDX(h) ((( h >> (3*8 - HLOG)) - h ) & (HSIZE - 1)) |
52 | # elif VERY_FAST |
53 | # define IDX(h) ((( h >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) |
54 | # else |
55 | # define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) |
56 | # endif |
57 | #endif |
58 | /* |
59 | * IDX works because it is very similar to a multiplicative hash, e.g. |
60 | * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1)) |
61 | * the latter is also quite fast on newer CPUs, and compresses similarly. |
62 | * |
63 | * the next one is also quite good, albeit slow ;) |
64 | * (int)(cos(h & 0xffffff) * 1e6) |
65 | */ |
66 | |
67 | #if 0 |
68 | /* original lzv-like hash function, much worse and thus slower */ |
69 | # define FRST(p) (p[0] << 5) ^ p[1] |
70 | # define NEXT(v,p) ((v) << 5) ^ p[2] |
71 | # define IDX(h) ((h) & (HSIZE - 1)) |
72 | #endif |
73 | |
74 | #define MAX_LIT (1 << 5) |
75 | #define MAX_OFF (1 << 13) |
76 | #define MAX_REF ((1 << 8) + (1 << 3)) |
77 | |
78 | #if __GNUC__ >= 3 |
79 | # define expect(expr,value) __builtin_expect ((expr),(value)) |
80 | # define inline inline |
81 | #else |
82 | # define expect(expr,value) (expr) |
83 | # define inline static |
84 | #endif |
85 | |
86 | #define expect_false(expr) expect ((expr) != 0, 0) |
87 | #define expect_true(expr) expect ((expr) != 0, 1) |
88 | |
89 | #if defined(__has_attribute) |
90 | # if __has_attribute(no_sanitize) |
91 | # define NO_SANITIZE(sanitizer) __attribute__((no_sanitize(sanitizer))) |
92 | # endif |
93 | #endif |
94 | |
95 | #if !defined(NO_SANITIZE) |
96 | # define NO_SANITIZE(sanitizer) |
97 | #endif |
98 | |
99 | /* |
100 | * compressed format |
101 | * |
102 | * 000LLLLL <L+1> ; literal, L+1=1..33 octets |
103 | * LLLooooo oooooooo ; backref L+1=1..7 octets, o+1=1..4096 offset |
104 | * 111ooooo LLLLLLLL oooooooo ; backref L+8 octets, o+1=1..4096 offset |
105 | * |
106 | */ |
107 | NO_SANITIZE("alignment" ) |
108 | size_t |
109 | lzf_compress (const void *const in_data, size_t in_len, |
110 | void *out_data, size_t out_len |
111 | #if LZF_STATE_ARG |
112 | , LZF_STATE htab |
113 | #endif |
114 | ) |
115 | { |
116 | #if !LZF_STATE_ARG |
117 | LZF_STATE htab; |
118 | #endif |
119 | const u8 *ip = (const u8 *)in_data; |
120 | u8 *op = (u8 *)out_data; |
121 | const u8 *in_end = ip + in_len; |
122 | u8 *out_end = op + out_len; |
123 | const u8 *ref; |
124 | |
125 | /* off requires a type wide enough to hold a general pointer difference. |
126 | * ISO C doesn't have that (size_t might not be enough and ptrdiff_t only |
127 | * works for differences within a single object). We also assume that no |
128 | * no bit pattern traps. Since the only platform that is both non-POSIX |
129 | * and fails to support both assumptions is windows 64 bit, we make a |
130 | * special workaround for it. |
131 | */ |
132 | #if defined (WIN32) && defined (_M_X64) |
133 | unsigned _int64 off; /* workaround for missing POSIX compliance */ |
134 | #else |
135 | size_t off; |
136 | #endif |
137 | unsigned int hval; |
138 | int lit; |
139 | |
140 | if (!in_len || !out_len) |
141 | return 0; |
142 | |
143 | #if INIT_HTAB |
144 | memset (htab, 0, sizeof (htab)); |
145 | #endif |
146 | |
147 | lit = 0; op++; /* start run */ |
148 | |
149 | hval = FRST (ip); |
150 | while (ip < in_end - 2) |
151 | { |
152 | LZF_HSLOT *hslot; |
153 | |
154 | hval = NEXT (hval, ip); |
155 | hslot = htab + IDX (hval); |
156 | ref = *hslot ? (*hslot + LZF_HSLOT_BIAS) : NULL; /* avoid applying zero offset to null pointer */ |
157 | *hslot = ip - LZF_HSLOT_BIAS; |
158 | |
159 | if (1 |
160 | #if INIT_HTAB |
161 | && ref < ip /* the next test will actually take care of this, but this is faster */ |
162 | #endif |
163 | && (off = ip - ref - 1) < MAX_OFF |
164 | && ref > (u8 *)in_data |
165 | && ref[2] == ip[2] |
166 | #if STRICT_ALIGN |
167 | && ((ref[1] << 8) | ref[0]) == ((ip[1] << 8) | ip[0]) |
168 | #else |
169 | && *(u16 *)ref == *(u16 *)ip |
170 | #endif |
171 | ) |
172 | { |
173 | /* match found at *ref++ */ |
174 | unsigned int len = 2; |
175 | size_t maxlen = in_end - ip - len; |
176 | maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; |
177 | |
178 | if (expect_false (op + 3 + 1 >= out_end)) /* first a faster conservative test */ |
179 | if (op - !lit + 3 + 1 >= out_end) /* second the exact but rare test */ |
180 | return 0; |
181 | |
182 | op [- lit - 1] = lit - 1; /* stop run */ |
183 | op -= !lit; /* undo run if length is zero */ |
184 | |
185 | for (;;) |
186 | { |
187 | if (expect_true (maxlen > 16)) |
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 | |
199 | len++; if (ref [len] != ip [len]) break; |
200 | len++; if (ref [len] != ip [len]) break; |
201 | len++; if (ref [len] != ip [len]) break; |
202 | len++; if (ref [len] != ip [len]) break; |
203 | |
204 | len++; if (ref [len] != ip [len]) break; |
205 | len++; if (ref [len] != ip [len]) break; |
206 | len++; if (ref [len] != ip [len]) break; |
207 | len++; if (ref [len] != ip [len]) break; |
208 | } |
209 | |
210 | do |
211 | len++; |
212 | while (len < maxlen && ref[len] == ip[len]); |
213 | |
214 | break; |
215 | } |
216 | |
217 | len -= 2; /* len is now #octets - 1 */ |
218 | ip++; |
219 | |
220 | if (len < 7) |
221 | { |
222 | *op++ = (off >> 8) + (len << 5); |
223 | } |
224 | else |
225 | { |
226 | *op++ = (off >> 8) + ( 7 << 5); |
227 | *op++ = len - 7; |
228 | } |
229 | |
230 | *op++ = off; |
231 | |
232 | lit = 0; op++; /* start run */ |
233 | |
234 | ip += len + 1; |
235 | |
236 | if (expect_false (ip >= in_end - 2)) |
237 | break; |
238 | |
239 | #if ULTRA_FAST || VERY_FAST |
240 | --ip; |
241 | # if VERY_FAST && !ULTRA_FAST |
242 | --ip; |
243 | # endif |
244 | hval = FRST (ip); |
245 | |
246 | hval = NEXT (hval, ip); |
247 | htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
248 | ip++; |
249 | |
250 | # if VERY_FAST && !ULTRA_FAST |
251 | hval = NEXT (hval, ip); |
252 | htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
253 | ip++; |
254 | # endif |
255 | #else |
256 | ip -= len + 1; |
257 | |
258 | do |
259 | { |
260 | hval = NEXT (hval, ip); |
261 | htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
262 | ip++; |
263 | } |
264 | while (len--); |
265 | #endif |
266 | } |
267 | else |
268 | { |
269 | /* one more literal byte we must copy */ |
270 | if (expect_false (op >= out_end)) |
271 | return 0; |
272 | |
273 | lit++; *op++ = *ip++; |
274 | |
275 | if (expect_false (lit == MAX_LIT)) |
276 | { |
277 | op [- lit - 1] = lit - 1; /* stop run */ |
278 | lit = 0; op++; /* start run */ |
279 | } |
280 | } |
281 | } |
282 | |
283 | if (op + 3 > out_end) /* at most 3 bytes can be missing here */ |
284 | return 0; |
285 | |
286 | while (ip < in_end) |
287 | { |
288 | lit++; *op++ = *ip++; |
289 | |
290 | if (expect_false (lit == MAX_LIT)) |
291 | { |
292 | op [- lit - 1] = lit - 1; /* stop run */ |
293 | lit = 0; op++; /* start run */ |
294 | } |
295 | } |
296 | |
297 | op [- lit - 1] = lit - 1; /* end run */ |
298 | op -= !lit; /* undo run if length is zero */ |
299 | |
300 | return op - (u8 *)out_data; |
301 | } |
302 | |
303 | |