1 : /*
2 : * Copyright (c) 2000-2008 Marc Alexander Lehmann <schmorp@schmorp.de>
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 : * decompression is not dependent on the hash function
44 : * the hashing 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 : /*
90 : * compressed format
91 : *
92 : * 000LLLLL <L+1> ; literal
93 : * LLLooooo oooooooo ; backref L
94 : * 111ooooo LLLLLLLL oooooooo ; backref L+7
95 : *
96 : */
97 :
98 : unsigned int
99 61673 : lzf_compress (const void *const in_data, unsigned int in_len,
100 : void *out_data, unsigned int out_len
101 : #if LZF_STATE_ARG
102 : , LZF_STATE htab
103 : #endif
104 : )
105 : {
106 : #if !LZF_STATE_ARG
107 : LZF_STATE htab;
108 : #endif
109 : const u8 **hslot;
110 61673 : const u8 *ip = (const u8 *)in_data;
111 61673 : u8 *op = (u8 *)out_data;
112 61673 : const u8 *in_end = ip + in_len;
113 61673 : u8 *out_end = op + out_len;
114 : const u8 *ref;
115 :
116 : /* off requires a type wide enough to hold a general pointer difference.
117 : * ISO C doesn't have that (size_t might not be enough and ptrdiff_t only
118 : * works for differences within a single object). We also assume that no
119 : * no bit pattern traps. Since the only platform that is both non-POSIX
120 : * and fails to support both assumptions is windows 64 bit, we make a
121 : * special workaround for it.
122 : */
123 : #if defined (WIN32) && defined (_M_X64)
124 : unsigned _int64 off; /* workaround for missing POSIX compliance */
125 : #else
126 : unsigned long off;
127 : #endif
128 : unsigned int hval;
129 : int lit;
130 :
131 61673 : if (!in_len || !out_len)
132 0 : return 0;
133 :
134 : #if INIT_HTAB
135 : memset (htab, 0, sizeof (htab));
136 : # if 0
137 : for (hslot = htab; hslot < htab + HSIZE; hslot++)
138 : *hslot++ = ip;
139 : # endif
140 : #endif
141 :
142 61673 : lit = 0; op++; /* start run */
143 :
144 61673 : hval = FRST (ip);
145 6347731 : while (ip < in_end - 2)
146 : {
147 6277158 : hval = NEXT (hval, ip);
148 6277158 : hslot = htab + IDX (hval);
149 6277158 : ref = *hslot; *hslot = ip;
150 :
151 11268889 : if (1
152 : #if INIT_HTAB
153 : && ref < ip /* the next test will actually take care of this, but this is faster */
154 : #endif
155 6277158 : && (off = ip - ref - 1) < MAX_OFF
156 1404008 : && ip + 4 < in_end
157 1404008 : && ref > (u8 *)in_data
158 : #if STRICT_ALIGN
159 : && ref[0] == ip[0]
160 : && ref[1] == ip[1]
161 : && ref[2] == ip[2]
162 : #else
163 4001962 : && *(u16 *)ref == *(u16 *)ip
164 3751240 : && ref[2] == ip[2]
165 : #endif
166 : )
167 : {
168 : /* match found at *ref++ */
169 1146260 : unsigned int len = 2;
170 1146260 : unsigned int maxlen = in_end - ip - len;
171 1146260 : maxlen = maxlen > MAX_REF ? MAX_REF : maxlen;
172 :
173 1146260 : op [- lit - 1] = lit - 1; /* stop run */
174 1146260 : op -= !lit; /* undo run if length is zero */
175 :
176 1146260 : if (expect_false (op + 3 + 1 >= out_end))
177 4821 : return 0;
178 :
179 : for (;;)
180 : {
181 1141439 : if (expect_true (maxlen > 16))
182 : {
183 1084384 : len++; if (ref [len] != ip [len]) break;
184 279843 : len++; if (ref [len] != ip [len]) break;
185 170212 : len++; if (ref [len] != ip [len]) break;
186 134927 : len++; if (ref [len] != ip [len]) break;
187 :
188 51436 : len++; if (ref [len] != ip [len]) break;
189 6907 : len++; if (ref [len] != ip [len]) break;
190 1003 : len++; if (ref [len] != ip [len]) break;
191 423 : len++; if (ref [len] != ip [len]) break;
192 :
193 364 : len++; if (ref [len] != ip [len]) break;
194 323 : len++; if (ref [len] != ip [len]) break;
195 312 : len++; if (ref [len] != ip [len]) break;
196 308 : len++; if (ref [len] != ip [len]) break;
197 :
198 308 : len++; if (ref [len] != ip [len]) break;
199 308 : len++; if (ref [len] != ip [len]) break;
200 308 : len++; if (ref [len] != ip [len]) break;
201 308 : len++; if (ref [len] != ip [len]) break;
202 : }
203 :
204 : do
205 92972 : len++;
206 92972 : while (len < maxlen && ref[len] == ip[len]);
207 :
208 : break;
209 : }
210 :
211 1141439 : len -= 2; /* len is now #octets - 1 */
212 1141439 : ip++;
213 :
214 1141439 : if (len < 7)
215 : {
216 1140386 : *op++ = (off >> 8) + (len << 5);
217 : }
218 : else
219 : {
220 1053 : *op++ = (off >> 8) + ( 7 << 5);
221 1053 : *op++ = len - 7;
222 : }
223 :
224 1141439 : *op++ = off;
225 1141439 : lit = 0; op++; /* start run */
226 :
227 1141439 : ip += len + 1;
228 :
229 1141439 : if (expect_false (ip >= in_end - 2))
230 : break;
231 :
232 : #if ULTRA_FAST || VERY_FAST
233 1135460 : --ip;
234 : # if VERY_FAST && !ULTRA_FAST
235 1135460 : --ip;
236 : # endif
237 1135460 : hval = FRST (ip);
238 :
239 1135460 : hval = NEXT (hval, ip);
240 1135460 : htab[IDX (hval)] = ip;
241 1135460 : ip++;
242 :
243 : # if VERY_FAST && !ULTRA_FAST
244 1135460 : hval = NEXT (hval, ip);
245 1135460 : htab[IDX (hval)] = ip;
246 1135460 : ip++;
247 : # endif
248 : #else
249 : ip -= len + 1;
250 :
251 : do
252 : {
253 : hval = NEXT (hval, ip);
254 : htab[IDX (hval)] = ip;
255 : ip++;
256 : }
257 : while (len--);
258 : #endif
259 : }
260 : else
261 : {
262 : /* one more literal byte we must copy */
263 5130898 : if (expect_false (op >= out_end))
264 41973 : return 0;
265 :
266 5088925 : lit++; *op++ = *ip++;
267 :
268 5088925 : if (expect_false (lit == MAX_LIT))
269 : {
270 74226 : op [- lit - 1] = lit - 1; /* stop run */
271 74226 : lit = 0; op++; /* start run */
272 : }
273 : }
274 : }
275 :
276 14879 : if (op + 3 > out_end) /* at most 3 bytes can be missing here */
277 2938 : return 0;
278 :
279 35813 : while (ip < in_end)
280 : {
281 23872 : lit++; *op++ = *ip++;
282 :
283 23872 : if (expect_false (lit == MAX_LIT))
284 : {
285 0 : op [- lit - 1] = lit - 1; /* stop run */
286 0 : lit = 0; op++; /* start run */
287 : }
288 : }
289 :
290 11941 : op [- lit - 1] = lit - 1; /* end run */
291 11941 : op -= !lit; /* undo run if length is zero */
292 :
293 11941 : return op - (u8 *)out_data;
294 : }
295 :
|
