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 1.10 by root, Thu Apr 14 20:38:50 2005 UTC vs.
Revision 1.45 by root, Sun Mar 27 23:53:23 2011 UTC

1/* 1/*
2 * Copyright (c) 2000-2005 Marc Alexander Lehmann <schmorp@schmorp.de> 2 * Copyright (c) 2000-2010 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 */
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 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
52# define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) 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))
53#endif 64# endif
54/* 65#endif
55 * IDX works because it is very similar to a multiplicative hash, e.g.
56 * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1))
57 * the latter is also quite fast on newer CPUs, and sligthly better
58 *
59 * the next one is also quite good, albeit slow ;)
60 * (int)(cos(h & 0xffffff) * 1e6)
61 */
62 66
63#if 0 67#if 0
64/* original lzv-like hash function, much worse and thus slower */ 68/* original lzv-like hash function, much worse and thus slower */
65# define FRST(p) (p[0] << 5) ^ p[1] 69# define FRST(p) (p[0] << 5) ^ p[1]
66# define NEXT(v,p) ((v) << 5) ^ p[2] 70# define NEXT(v,p) ((v) << 5) ^ p[2]
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, L+1=1..33 octets
78 * LLLooooo oooooooo ; backref L 93 * LLLooooo oooooooo ; backref L+1=1..7 octets, o+1=1..4096 offset
79 * 111ooooo LLLLLLLL oooooooo ; backref L+7 94 * 111ooooo LLLLLLLL oooooooo ; backref L+8 octets, o+1=1..4096 offset
80 * 95 *
81 */ 96 */
82 97
83unsigned int 98unsigned int
84lzf_compress (const void *const in_data, unsigned int in_len, 99lzf_compress (const void *const in_data, unsigned int in_len,
85 void *out_data, unsigned int out_len 100 void *out_data, unsigned int out_len
86#if LZF_STATE_ARG 101#if LZF_STATE_ARG
87 , LZF_STATE *htab 102 , LZF_STATE htab
88#endif 103#endif
89 ) 104 )
90{ 105{
91#if !LZF_STATE_ARG 106#if !LZF_STATE_ARG
92 LZF_STATE htab; 107 LZF_STATE htab;
93#endif 108#endif
94 const u8 **hslot;
95 const u8 *ip = (const u8 *)in_data; 109 const u8 *ip = (const u8 *)in_data;
96 u8 *op = (u8 *)out_data; 110 u8 *op = (u8 *)out_data;
97 const u8 *in_end = ip + in_len; 111 const u8 *in_end = ip + in_len;
98 u8 *out_end = op + out_len; 112 u8 *out_end = op + out_len;
99 const u8 *ref; 113 const u8 *ref;
100 114
101 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 no
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 unsigned _int64 off; /* workaround for missing POSIX compliance */
124#else
102 unsigned long off; 125 unsigned long off;
103 int lit = 0; 126#endif
127 unsigned int hval;
128 int lit;
129
130 if (!in_len || !out_len)
131 return 0;
104 132
105#if INIT_HTAB 133#if INIT_HTAB
106# if USE_MEMCPY
107 memset (htab, 0, sizeof (htab)); 134 memset (htab, 0, sizeof (htab));
108# else
109 for (hslot = htab; hslot < htab + HSIZE; hslot++)
110 *hslot++ = ip;
111# endif 135#endif
112#endif
113 136
114 for (;;) 137 lit = 0; op++; /* start run */
138
139 hval = FRST (ip);
140 while (ip < in_end - 2)
115 { 141 {
116 if (ip < in_end - 2) 142 LZF_HSLOT *hslot;
143
144 hval = NEXT (hval, ip);
145 hslot = htab + IDX (hval);
146 ref = *hslot + LZF_HSLOT_BIAS; *hslot = ip - LZF_HSLOT_BIAS;
147
148 if (1
149#if INIT_HTAB
150 && ref < ip /* the next test will actually take care of this, but this is faster */
151#endif
152 && (off = ip - ref - 1) < MAX_OFF
153 && ref > (u8 *)in_data
154 && ref[2] == ip[2]
155#if STRICT_ALIGN
156 && ((ref[1] << 8) | ref[0]) == ((ip[1] << 8) | ip[0])
157#else
158 && *(u16 *)ref == *(u16 *)ip
159#endif
160 )
117 { 161 {
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++ */ 162 /* match found at *ref++ */
140 unsigned int len = 2; 163 unsigned int len = 2;
141 unsigned int maxlen = in_end - ip - len; 164 unsigned int maxlen = in_end - ip - len;
142 maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; 165 maxlen = maxlen > MAX_REF ? MAX_REF : maxlen;
143 166
144 if (op + lit + 1 + 3 >= out_end) 167 if (expect_false (op + 3 + 1 >= out_end)) /* first a faster conservative test */
168 if (op - !lit + 3 + 1 >= out_end) /* second the exact but rare test */
145 return 0; 169 return 0;
170
171 op [- lit - 1] = lit - 1; /* stop run */
172 op -= !lit; /* undo run if length is zero */
173
174 for (;;)
175 {
176 if (expect_true (maxlen > 16))
177 {
178 len++; if (ref [len] != ip [len]) break;
179 len++; if (ref [len] != ip [len]) break;
180 len++; if (ref [len] != ip [len]) break;
181 len++; if (ref [len] != ip [len]) break;
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 }
146 198
147 do 199 do
148 len++; 200 len++;
149 while (len < maxlen && ref[len] == ip[len]); 201 while (len < maxlen && ref[len] == ip[len]);
150 202
151 if (lit)
152 { 203 break;
153 *op++ = lit - 1;
154 lit = -lit;
155 do
156 *op++ = ip[lit];
157 while (++lit);
158 } 204 }
159 205
206 len -= 2; /* len is now #octets - 1 */
207 ip++;
208
209 if (len < 7)
210 {
211 *op++ = (off >> 8) + (len << 5);
212 }
213 else
214 {
215 *op++ = (off >> 8) + ( 7 << 5);
160 len -= 2; 216 *op++ = len - 7;
217 }
218
219 *op++ = off;
220
221 lit = 0; op++; /* start run */
222
223 ip += len + 1;
224
225 if (expect_false (ip >= in_end - 2))
226 break;
227
228#if ULTRA_FAST || VERY_FAST
229 --ip;
230# if VERY_FAST && !ULTRA_FAST
231 --ip;
232# endif
233 hval = FRST (ip);
234
235 hval = NEXT (hval, ip);
236 htab[IDX (hval)] = ip - LZF_HSLOT_BIAS;
237 ip++;
238
239# if VERY_FAST && !ULTRA_FAST
240 hval = NEXT (hval, ip);
241 htab[IDX (hval)] = ip - LZF_HSLOT_BIAS;
242 ip++;
243# endif
244#else
245 ip -= len + 1;
246
247 do
248 {
249 hval = NEXT (hval, ip);
250 htab[IDX (hval)] = ip - LZF_HSLOT_BIAS;
161 ip++; 251 ip++;
162
163 if (len < 7)
164 {
165 *op++ = (off >> 8) + (len << 5);
166 } 252 }
167 else
168 {
169 *op++ = (off >> 8) + ( 7 << 5);
170 *op++ = len - 7;
171 }
172
173 *op++ = off;
174
175#if ULTRA_FAST || VERY_FAST
176 ip += len;
177#if VERY_FAST && !ULTRA_FAST
178 --ip;
179#endif
180 hval = FRST (ip);
181
182 hval = NEXT (hval, ip);
183 htab[IDX (hval)] = ip;
184 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 }
198 while (len--); 253 while (len--);
199#endif 254#endif
200 continue;
201 }
202 } 255 }
203 else if (ip == in_end) 256 else
204 break;
205
206 /* one more literal byte we must copy */
207 lit++;
208 ip++;
209
210 if (lit == MAX_LIT)
211 { 257 {
212 if (op + 1 + MAX_LIT >= out_end) 258 /* one more literal byte we must copy */
259 if (expect_false (op >= out_end))
213 return 0; 260 return 0;
214 261
215 *op++ = MAX_LIT - 1; 262 lit++; *op++ = *ip++;
216#if USE_MEMCPY 263
217 memcpy (op, ip - MAX_LIT, MAX_LIT); 264 if (expect_false (lit == MAX_LIT))
218 op += MAX_LIT;
219 lit = 0;
220#else
221 lit = -lit;
222 do 265 {
223 *op++ = ip[lit]; 266 op [- lit - 1] = lit - 1; /* stop run */
224 while (++lit); 267 lit = 0; op++; /* start run */
225#endif 268 }
226 } 269 }
227 } 270 }
228 271
229 if (lit) 272 if (op + 3 > out_end) /* at most 3 bytes can be missing here */
273 return 0;
274
275 while (ip < in_end)
230 { 276 {
231 if (op + lit + 1 >= out_end) 277 lit++; *op++ = *ip++;
232 return 0;
233 278
234 *op++ = lit - 1; 279 if (expect_false (lit == MAX_LIT))
235 lit = -lit; 280 {
236 do 281 op [- lit - 1] = lit - 1; /* stop run */
237 *op++ = ip[lit]; 282 lit = 0; op++; /* start run */
238 while (++lit); 283 }
239 } 284 }
240 285
286 op [- lit - 1] = lit - 1; /* end run */
287 op -= !lit; /* undo run if length is zero */
288
241 return op - (u8 *) out_data; 289 return op - (u8 *)out_data;
242} 290}
291

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines