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 1.11 by root, Fri Feb 16 22:13:43 2007 UTC vs.
Revision 1.31 by root, Tue May 6 22:49:18 2008 UTC

1/* 1/*
2 * Copyright (c) 2000-2005 Marc Alexander Lehmann <schmorp@schmorp.de> 2 * Copyright (c) 2000-2007 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))
… …
47 * it works ;) 45 * it works ;)
48 */ 46 */
49#ifndef FRST 47#ifndef FRST
50# define FRST(p) (((p[0]) << 8) | p[1]) 48# define FRST(p) (((p[0]) << 8) | p[1])
51# define NEXT(v,p) (((v) << 8) | p[2]) 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
52# 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
53#endif 57#endif
54/* 58/*
55 * 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.
56 * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1)) 60 * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1))
57 * the latter is also quite fast on newer CPUs, and sligthly better 61 * the latter is also quite fast on newer CPUs, and compresses similarly.
58 * 62 *
59 * the next one is also quite good, albeit slow ;) 63 * the next one is also quite good, albeit slow ;)
60 * (int)(cos(h & 0xffffff) * 1e6) 64 * (int)(cos(h & 0xffffff) * 1e6)
61 */ 65 */
62 66
… …
67# define IDX(h) ((h) & (HSIZE - 1)) 71# define IDX(h) ((h) & (HSIZE - 1))
68#endif 72#endif
69 73
70#define MAX_LIT (1 << 5) 74#define MAX_LIT (1 << 5)
71#define MAX_OFF (1 << 13) 75#define MAX_OFF (1 << 13)
72#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)
73 88
74/* 89/*
75 * compressed format 90 * compressed format
76 * 91 *
77 * 000LLLLL <L+1> ; literal 92 * 000LLLLL <L+1> ; literal
… …
96 u8 *op = (u8 *)out_data; 111 u8 *op = (u8 *)out_data;
97 const u8 *in_end = ip + in_len; 112 const u8 *in_end = ip + in_len;
98 u8 *out_end = op + out_len; 113 u8 *out_end = op + out_len;
99 const u8 *ref; 114 const u8 *ref;
100 115
101 unsigned int hval = FRST (ip); 116 unsigned int hval;
117#if WIN32
118 unsigned _int64 off; /* workaround for microsoft bug (they claim to support POSIX) */
119#else
102 unsigned long off; 120 unsigned long off;
103 int lit = 0; 121#endif
122 int lit;
123
124 if (!in_len || !out_len)
125 return 0;
104 126
105#if INIT_HTAB 127#if INIT_HTAB
106# if USE_MEMCPY
107 memset (htab, 0, sizeof (htab)); 128 memset (htab, 0, sizeof (htab));
108# else 129# if 0
109 for (hslot = htab; hslot < htab + HSIZE; hslot++) 130 for (hslot = htab; hslot < htab + HSIZE; hslot++)
110 *hslot++ = ip; 131 *hslot++ = ip;
111# endif 132# endif
112#endif 133#endif
113 134
114 for (;;) 135 lit = 0; op++; /* start run */
136
137 hval = FRST (ip);
138 while (ip < in_end - 2)
115 { 139 {
116 if (ip < in_end - 2) 140 hval = NEXT (hval, ip);
141 hslot = htab + IDX (hval);
142 ref = *hslot; *hslot = ip;
143
144 if (1
145#if INIT_HTAB
146 && ref < ip /* the next test will actually take care of this, but this is faster */
147#endif
148 && (off = ip - ref - 1) < MAX_OFF
149 && ip + 4 < in_end
150 && ref > (u8 *)in_data
151#if STRICT_ALIGN
152 && ref[0] == ip[0]
153 && ref[1] == ip[1]
154 && ref[2] == ip[2]
155#else
156 && *(u16 *)ref == *(u16 *)ip
157 && ref[2] == ip[2]
158#endif
159 )
117 { 160 {
118 hval = NEXT (hval, ip);
119 hslot = htab + IDX (hval);
120 ref = *hslot; *hslot = ip;
121
122 if (1
123#if INIT_HTAB && !USE_MEMCPY
124 && ref < ip /* the next test will actually take care of this, but this is faster */
125#endif
126 && (off = ip - ref - 1) < MAX_OFF
127 && ip + 4 < in_end
128 && ref > (u8 *)in_data
129#if STRICT_ALIGN
130 && ref[0] == ip[0]
131 && ref[1] == ip[1]
132 && ref[2] == ip[2]
133#else
134 && *(u16 *)ref == *(u16 *)ip
135 && ref[2] == ip[2]
136#endif
137 )
138 {
139 /* match found at *ref++ */ 161 /* match found at *ref++ */
140 unsigned int len = 2; 162 unsigned int len = 2;
141 unsigned int maxlen = in_end - ip - len; 163 unsigned int maxlen = in_end - ip - len;
142 maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; 164 maxlen = maxlen > MAX_REF ? MAX_REF : maxlen;
143 165
144 if (op + lit + 1 + 3 >= out_end) 166 op [- lit - 1] = lit - 1; /* stop run */
167 op -= !lit; /* undo run if length is zero */
168
169 if (expect_false (op + 3 + 1 >= out_end))
145 return 0; 170 return 0;
171
172 for (;;)
173 {
174 if (expect_true (maxlen > 16))
175 {
176 len++; if (ref [len] != ip [len]) break;
177 len++; if (ref [len] != ip [len]) break;
178 len++; if (ref [len] != ip [len]) break;
179 len++; if (ref [len] != ip [len]) break;
180
181 len++; if (ref [len] != ip [len]) break;
182 len++; if (ref [len] != ip [len]) break;
183 len++; if (ref [len] != ip [len]) break;
184 len++; if (ref [len] != ip [len]) break;
185
186 len++; if (ref [len] != ip [len]) break;
187 len++; if (ref [len] != ip [len]) break;
188 len++; if (ref [len] != ip [len]) break;
189 len++; if (ref [len] != ip [len]) break;
190
191 len++; if (ref [len] != ip [len]) break;
192 len++; if (ref [len] != ip [len]) break;
193 len++; if (ref [len] != ip [len]) break;
194 len++; if (ref [len] != ip [len]) break;
195 }
146 196
147 do 197 do
148 len++; 198 len++;
149 while (len < maxlen && ref[len] == ip[len]); 199 while (len < maxlen && ref[len] == ip[len]);
150 200
151 if (lit)
152 { 201 break;
153 *op++ = lit - 1;
154 lit = -lit;
155 do
156 *op++ = ip[lit];
157 while (++lit);
158 } 202 }
159 203
160 len -= 2; 204 len -= 2;
161 ip++; 205 ip++;
162 206
163 if (len < 7) 207 if (len < 7)
164 { 208 {
165 *op++ = (off >> 8) + (len << 5); 209 *op++ = (off >> 8) + (len << 5);
166 } 210 }
167 else 211 else
168 { 212 {
169 *op++ = (off >> 8) + ( 7 << 5); 213 *op++ = (off >> 8) + ( 7 << 5);
170 *op++ = len - 7; 214 *op++ = len - 7;
171 } 215 }
172 216
173 *op++ = off; 217 *op++ = off;
174 218
175#if ULTRA_FAST || VERY_FAST 219#if ULTRA_FAST || VERY_FAST
176 ip += len; 220 ip += len;
177#if VERY_FAST && !ULTRA_FAST 221#if VERY_FAST && !ULTRA_FAST
178 --ip; 222 --ip;
179#endif 223#endif
180 hval = FRST (ip); 224 hval = FRST (ip);
181 225
226 hval = NEXT (hval, ip);
227 htab[IDX (hval)] = ip;
228 ip++;
229
230#if VERY_FAST && !ULTRA_FAST
231 hval = NEXT (hval, ip);
232 htab[IDX (hval)] = ip;
233 ip++;
234#endif
235#else
236 do
237 {
182 hval = NEXT (hval, ip); 238 hval = NEXT (hval, ip);
183 htab[IDX (hval)] = ip; 239 htab[IDX (hval)] = ip;
184 ip++; 240 ip++;
185
186#if VERY_FAST && !ULTRA_FAST
187 hval = NEXT (hval, ip);
188 htab[IDX (hval)] = ip;
189 ip++;
190#endif
191#else
192 do
193 {
194 hval = NEXT (hval, ip);
195 htab[IDX (hval)] = ip;
196 ip++;
197 } 241 }
198 while (len--); 242 while (len--);
199#endif 243#endif
200 continue; 244
201 } 245 lit = 0; op++; /* start run */
202 } 246 }
203 else if (ip == in_end) 247 else
204 break;
205
206 /* one more literal byte we must copy */
207 lit++;
208 ip++;
209
210 if (lit == MAX_LIT)
211 { 248 {
212 if (op + 1 + MAX_LIT >= out_end) 249 /* one more literal byte we must copy */
250 if (expect_false (op >= out_end))
213 return 0; 251 return 0;
214 252
215 *op++ = MAX_LIT - 1; 253 lit++; *op++ = *ip++;
216#if USE_MEMCPY 254
217 memcpy (op, ip - MAX_LIT, MAX_LIT); 255 if (expect_false (lit == MAX_LIT))
218 op += MAX_LIT;
219 lit = 0;
220#else
221 lit = -lit;
222 do 256 {
223 *op++ = ip[lit]; 257 op [- lit - 1] = lit - 1; /* stop run */
224 while (++lit); 258 lit = 0; op++; /* start run */
225#endif 259 }
226 } 260 }
227 } 261 }
228 262
229 if (lit) 263 if (op + 3 > out_end) /* at most 3 bytes can be missing here */
264 return 0;
265
266 while (ip < in_end)
230 { 267 {
231 if (op + lit + 1 >= out_end) 268 lit++; *op++ = *ip++;
232 return 0;
233 269
234 *op++ = lit - 1; 270 if (expect_false (lit == MAX_LIT))
235 lit = -lit; 271 {
236 do 272 op [- lit - 1] = lit - 1; /* stop run */
237 *op++ = ip[lit]; 273 lit = 0; op++; /* start run */
238 while (++lit); 274 }
239 } 275 }
240 276
277 op [- lit - 1] = lit - 1; /* end run */
278 op -= !lit; /* undo run if length is zero */
279
241 return op - (u8 *) out_data; 280 return op - (u8 *)out_data;
242} 281}
282

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines