1 |
/* |
2 |
* Copyright (c) 2000-2010,2012 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 |
* 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 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 |
56 |
# define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (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)) |
64 |
# endif |
65 |
#endif |
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, L+1=1..33 octets |
93 |
* LLLooooo oooooooo ; backref L+1=1..7 octets, o+1=1..4096 offset |
94 |
* 111ooooo LLLLLLLL oooooooo ; backref L+8 octets, o+1=1..4096 offset |
95 |
* |
96 |
*/ |
97 |
|
98 |
unsigned int |
99 |
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 *ip = (const u8 *)in_data; |
110 |
u8 *op = (u8 *)out_data; |
111 |
const u8 *in_end = ip + in_len; |
112 |
u8 *out_end = op + out_len; |
113 |
const u8 *ref; |
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 |
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 |
/* workaround for missing POSIX compliance */ |
124 |
#if #if __GNUC__ |
125 |
unsigned long long off; |
126 |
#else |
127 |
unsigned __int64 off; |
128 |
#endif |
129 |
#else |
130 |
unsigned long off; |
131 |
#endif |
132 |
unsigned int hval; |
133 |
int lit; |
134 |
|
135 |
if (!in_len || !out_len) |
136 |
return 0; |
137 |
|
138 |
#if INIT_HTAB |
139 |
memset (htab, 0, sizeof (htab)); |
140 |
#endif |
141 |
|
142 |
lit = 0; op++; /* start run */ |
143 |
|
144 |
hval = FRST (ip); |
145 |
while (ip < in_end - 2) |
146 |
{ |
147 |
LZF_HSLOT *hslot; |
148 |
|
149 |
hval = NEXT (hval, ip); |
150 |
hslot = htab + IDX (hval); |
151 |
ref = *hslot + LZF_HSLOT_BIAS; *hslot = ip - LZF_HSLOT_BIAS; |
152 |
|
153 |
if (1 |
154 |
#if INIT_HTAB |
155 |
&& ref < ip /* the next test will actually take care of this, but this is faster */ |
156 |
#endif |
157 |
&& (off = ip - ref - 1) < MAX_OFF |
158 |
&& ref > (u8 *)in_data |
159 |
&& ref[2] == ip[2] |
160 |
#if STRICT_ALIGN |
161 |
&& ((ref[1] << 8) | ref[0]) == ((ip[1] << 8) | ip[0]) |
162 |
#else |
163 |
&& *(u16 *)ref == *(u16 *)ip |
164 |
#endif |
165 |
) |
166 |
{ |
167 |
/* match found at *ref++ */ |
168 |
unsigned int len = 2; |
169 |
unsigned int maxlen = in_end - ip - len; |
170 |
maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; |
171 |
|
172 |
if (expect_false (op + 3 + 1 >= out_end)) /* first a faster conservative test */ |
173 |
if (op - !lit + 3 + 1 >= out_end) /* second the exact but rare test */ |
174 |
return 0; |
175 |
|
176 |
op [- lit - 1] = lit - 1; /* stop run */ |
177 |
op -= !lit; /* undo run if length is zero */ |
178 |
|
179 |
for (;;) |
180 |
{ |
181 |
if (expect_true (maxlen > 16)) |
182 |
{ |
183 |
len++; if (ref [len] != ip [len]) break; |
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 |
|
188 |
len++; if (ref [len] != ip [len]) break; |
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 |
|
193 |
len++; if (ref [len] != ip [len]) break; |
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 |
|
198 |
len++; if (ref [len] != ip [len]) break; |
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 |
} |
203 |
|
204 |
do |
205 |
len++; |
206 |
while (len < maxlen && ref[len] == ip[len]); |
207 |
|
208 |
break; |
209 |
} |
210 |
|
211 |
len -= 2; /* len is now #octets - 1 */ |
212 |
ip++; |
213 |
|
214 |
if (len < 7) |
215 |
{ |
216 |
*op++ = (off >> 8) + (len << 5); |
217 |
} |
218 |
else |
219 |
{ |
220 |
*op++ = (off >> 8) + ( 7 << 5); |
221 |
*op++ = len - 7; |
222 |
} |
223 |
|
224 |
*op++ = off; |
225 |
|
226 |
lit = 0; op++; /* start run */ |
227 |
|
228 |
ip += len + 1; |
229 |
|
230 |
if (expect_false (ip >= in_end - 2)) |
231 |
break; |
232 |
|
233 |
#if ULTRA_FAST || VERY_FAST |
234 |
--ip; |
235 |
# if VERY_FAST && !ULTRA_FAST |
236 |
--ip; |
237 |
# endif |
238 |
hval = FRST (ip); |
239 |
|
240 |
hval = NEXT (hval, ip); |
241 |
htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
242 |
ip++; |
243 |
|
244 |
# if VERY_FAST && !ULTRA_FAST |
245 |
hval = NEXT (hval, ip); |
246 |
htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
247 |
ip++; |
248 |
# endif |
249 |
#else |
250 |
ip -= len + 1; |
251 |
|
252 |
do |
253 |
{ |
254 |
hval = NEXT (hval, ip); |
255 |
htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
256 |
ip++; |
257 |
} |
258 |
while (len--); |
259 |
#endif |
260 |
} |
261 |
else |
262 |
{ |
263 |
/* one more literal byte we must copy */ |
264 |
if (expect_false (op >= out_end)) |
265 |
return 0; |
266 |
|
267 |
lit++; *op++ = *ip++; |
268 |
|
269 |
if (expect_false (lit == MAX_LIT)) |
270 |
{ |
271 |
op [- lit - 1] = lit - 1; /* stop run */ |
272 |
lit = 0; op++; /* start run */ |
273 |
} |
274 |
} |
275 |
} |
276 |
|
277 |
if (op + 3 > out_end) /* at most 3 bytes can be missing here */ |
278 |
return 0; |
279 |
|
280 |
while (ip < in_end) |
281 |
{ |
282 |
lit++; *op++ = *ip++; |
283 |
|
284 |
if (expect_false (lit == MAX_LIT)) |
285 |
{ |
286 |
op [- lit - 1] = lit - 1; /* stop run */ |
287 |
lit = 0; op++; /* start run */ |
288 |
} |
289 |
} |
290 |
|
291 |
op [- lit - 1] = lit - 1; /* end run */ |
292 |
op -= !lit; /* undo run if length is zero */ |
293 |
|
294 |
return op - (u8 *)out_data; |
295 |
} |
296 |
|