ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/liblzf/lzf_c.c
(Generate patch)

Comparing liblzf/lzf_c.c (file contents):
Revision 1.6 by pcg, Tue Apr 20 18:11:58 2004 UTC vs.
Revision 1.48 by root, Sat Jun 30 21:39:41 2012 UTC

1/* 1/*
2 * Copyright (c) 2000-2003 Marc Alexander Lehmann <pcg@goof.com> 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 *
14 * 3. The name of the author may not be used to endorse or promote products
15 * derived from this software without specific prior written permission.
16 * 13 *
17 * 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
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- 15 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 16 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * 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-
24 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- 21 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
25 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED 22 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
26 * OF THE POSSIBILITY OF SUCH DAMAGE. 23 * OF THE POSSIBILITY OF SUCH DAMAGE.
27 * 24 *
28 * Alternatively, the contents of this file may be used under the terms of 25 * Alternatively, the contents of this file may be used under the terms of
29 * the GNU General Public License version 2 (the "GPL"), in which case the 26 * the GNU General Public License ("GPL") version 2 or any later version,
30 * provisions of the GPL are applicable instead of the above. If you wish to 27 * in which case the provisions of the GPL are applicable instead of
31 * allow the use of your version of this file only under the terms of the 28 * the above. If you wish to allow the use of your version of this file
32 * GPL and not to allow others to use your version of this file under the 29 * only under the terms of the GPL and not to allow others to use your
33 * BSD license, indicate your decision by deleting the provisions above and 30 * version of this file under the BSD license, indicate your decision
34 * replace them with the notice and other provisions required by the GPL. If 31 * by deleting the provisions above and replace them with the notice
35 * you do not delete the provisions above, a recipient may use your version 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
36 * of this file under either the BSD or the GPL. 34 * either the BSD or the GPL.
37 */ 35 */
38 36
39#include "lzfP.h" 37#include "lzfP.h"
40 38
41#define HSIZE (1 << (HLOG)) 39#define HSIZE (1 << (HLOG))
42 40
43/* 41/*
44 * don't play with this unless you benchmark! 42 * don't play with this unless you benchmark!
45 * decompression is not dependent on the hash function 43 * the data format is not dependent on the hash function.
46 * the hashing function might seem strange, just believe me 44 * the hash function might seem strange, just believe me,
47 * it works ;) 45 * it works ;)
48 */ 46 */
47#ifndef FRST
49#define FRST(p) (((p[0]) << 8) + p[1]) 48# define FRST(p) (((p[0]) << 8) | p[1])
50#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
51#define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) + h*3) & (HSIZE - 1)) 56# define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (HSIZE - 1))
52/* 57# endif
53 * IDX works because it is very similar to a multiplicative hash, e.g. 58# else
54 * (h * 57321 >> (3*8 - HLOG)) 59/* this one was developed with sesse,
55 * the next one is also quite good, albeit slow ;) 60 * and is very similar to the one in snappy.
56 * (int)(cos(h & 0xffffff) * 1e6) 61 * it does need a modern enough cpu with a fast multiplication.
57 */ 62 */
63# define IDX(h) (((h * 0x1e35a7bdU) >> (32 - HLOG - 8)) & (HSIZE - 1))
64# endif
65#endif
58 66
59#if 0 67#if 0
60/* original lzv-like hash function */ 68/* original lzv-like hash function, much worse and thus slower */
61# define FRST(p) (p[0] << 5) ^ p[1] 69# define FRST(p) (p[0] << 5) ^ p[1]
62# define NEXT(v,p) ((v) << 5) ^ p[2] 70# define NEXT(v,p) ((v) << 5) ^ p[2]
63# define IDX(h) ((h) & (HSIZE - 1)) 71# define IDX(h) ((h) & (HSIZE - 1))
64#endif 72#endif
65 73
66#define MAX_LIT (1 << 5) 74#define MAX_LIT (1 << 5)
67#define MAX_OFF (1 << 13) 75#define MAX_OFF (1 << 13)
68#define MAX_REF ((1 << 8) + (1 << 3)) 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)
69 88
70/* 89/*
71 * compressed format 90 * compressed format
72 * 91 *
73 * 000LLLLL <L+1> ; literal 92 * 000LLLLL <L+1> ; literal, L+1=1..33 octets
74 * LLLOOOOO oooooooo ; backref L 93 * LLLooooo oooooooo ; backref L+1=1..7 octets, o+1=1..4096 offset
75 * 111OOOOO LLLLLLLL oooooooo ; backref L+7 94 * 111ooooo LLLLLLLL oooooooo ; backref L+8 octets, o+1=1..4096 offset
76 * 95 *
77 */ 96 */
78 97
79unsigned int 98unsigned int
80lzf_compress (const void *const in_data, unsigned int in_len, 99lzf_compress (const void *const in_data, unsigned int in_len,
81 void *out_data, unsigned int out_len 100 void *out_data, unsigned int out_len
82#if LZF_STATE_ARG 101#if LZF_STATE_ARG
83 , LZF_STATE *htab 102 , LZF_STATE htab
84#endif 103#endif
85 ) 104 )
86{ 105{
87#if !LZF_STATE_ARG 106#if !LZF_STATE_ARG
88 LZF_STATE htab; 107 LZF_STATE htab;
89#endif 108#endif
90 const u8 **hslot;
91 const u8 *ip = (const u8 *)in_data; 109 const u8 *ip = (const u8 *)in_data;
92 u8 *op = (u8 *)out_data; 110 u8 *op = (u8 *)out_data;
93 const u8 *in_end = ip + in_len; 111 const u8 *in_end = ip + in_len;
94 u8 *out_end = op + out_len; 112 u8 *out_end = op + out_len;
95 const u8 *ref; 113 const u8 *ref;
96 114
97 unsigned int hval = FRST (ip); 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 __GNUC__
125 unsigned long long off;
126 #else
127 unsigned __int64 off;
128 #endif
129#else
98 unsigned long off; 130 unsigned long off;
99 int lit = 0; 131#endif
132 unsigned int hval;
133 int lit;
134
135 if (!in_len || !out_len)
136 return 0;
100 137
101#if INIT_HTAB 138#if INIT_HTAB
102# if USE_MEMCPY
103 memset (htab, 0, sizeof (htab)); 139 memset (htab, 0, sizeof (htab));
104# else
105 for (hslot = htab; hslot < htab + HSIZE; hslot++)
106 *hslot++ = ip;
107# endif 140#endif
108#endif
109 141
110 for (;;) 142 lit = 0; op++; /* start run */
143
144 hval = FRST (ip);
145 while (ip < in_end - 2)
111 { 146 {
112 if (ip < in_end - 2) 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 )
113 { 166 {
114 hval = NEXT (hval, ip);
115 hslot = htab + IDX (hval);
116 ref = *hslot; *hslot = ip;
117
118 if (1
119#if INIT_HTAB && !USE_MEMCPY
120 && ref < ip /* the next test will actually take care of this, but this is faster */
121#endif
122 && (off = ip - ref - 1) < MAX_OFF
123 && ip + 4 < in_end
124 && ref > (u8 *)in_data
125#if STRICT_ALIGN
126 && ref[0] == ip[0]
127 && ref[1] == ip[1]
128 && ref[2] == ip[2]
129#else
130 && *(u16 *)ref == *(u16 *)ip
131 && ref[2] == ip[2]
132#endif
133 )
134 {
135 /* match found at *ref++ */ 167 /* match found at *ref++ */
136 unsigned int len = 2; 168 unsigned int len = 2;
137 unsigned int maxlen = in_end - ip - len; 169 unsigned int maxlen = in_end - ip - len;
138 maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; 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 }
139 203
140 do 204 do
141 len++; 205 len++;
142 while (len < maxlen && ref[len] == ip[len]); 206 while (len < maxlen && ref[len] == ip[len]);
143 207
144 if (op + lit + 1 + 3 >= out_end)
145 return 0;
146
147 if (lit)
148 { 208 break;
149 *op++ = lit - 1;
150 lit = -lit;
151 do
152 *op++ = ip[lit];
153 while (++lit);
154 } 209 }
155 210
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);
156 len -= 2; 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;
157 ip++; 256 ip++;
158
159 if (len < 7)
160 {
161 *op++ = (off >> 8) + (len << 5);
162 } 257 }
163 else
164 {
165 *op++ = (off >> 8) + ( 7 << 5);
166 *op++ = len - 7;
167 }
168
169 *op++ = off;
170
171#if ULTRA_FAST
172 ip += len;
173 hval = FRST (ip);
174 hval = NEXT (hval, ip);
175 htab[IDX (hval)] = ip;
176 ip++;
177#else
178 do
179 {
180 hval = NEXT (hval, ip);
181 htab[IDX (hval)] = ip;
182 ip++;
183 }
184 while (len--); 258 while (len--);
185#endif 259#endif
186 continue;
187 }
188 } 260 }
189 else if (ip == in_end) 261 else
190 break;
191
192 /* one more literal byte we must copy */
193 lit++;
194 ip++;
195
196 if (lit == MAX_LIT)
197 { 262 {
198 if (op + 1 + MAX_LIT >= out_end) 263 /* one more literal byte we must copy */
264 if (expect_false (op >= out_end))
199 return 0; 265 return 0;
200 266
201 *op++ = MAX_LIT - 1; 267 lit++; *op++ = *ip++;
202#if USE_MEMCPY 268
203 memcpy (op, ip - MAX_LIT, MAX_LIT); 269 if (expect_false (lit == MAX_LIT))
204 op += MAX_LIT;
205 lit = 0;
206#else
207 lit = -lit;
208 do 270 {
209 *op++ = ip[lit]; 271 op [- lit - 1] = lit - 1; /* stop run */
210 while (++lit); 272 lit = 0; op++; /* start run */
211#endif 273 }
212 } 274 }
213 } 275 }
214 276
215 if (lit) 277 if (op + 3 > out_end) /* at most 3 bytes can be missing here */
278 return 0;
279
280 while (ip < in_end)
216 { 281 {
217 if (op + lit + 1 >= out_end) 282 lit++; *op++ = *ip++;
218 return 0;
219 283
220 *op++ = lit - 1; 284 if (expect_false (lit == MAX_LIT))
221 lit = -lit; 285 {
222 do 286 op [- lit - 1] = lit - 1; /* stop run */
223 *op++ = ip[lit]; 287 lit = 0; op++; /* start run */
224 while (++lit); 288 }
225 } 289 }
226 290
291 op [- lit - 1] = lit - 1; /* end run */
292 op -= !lit; /* undo run if length is zero */
293
227 return op - (u8 *) out_data; 294 return op - (u8 *)out_data;
228} 295}
296

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines