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 1.4 by pcg, Tue Dec 23 04:52:00 2003 UTC vs.
Revision 1.41 by root, Tue Jun 1 02:30:06 2010 UTC

1/* 1/*
2 * Copyright (c) 2000-2003 Marc Alexander Lehmann <pcg@goof.com> 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-
… …
22 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 19 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
23 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 20 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
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.
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.
27 */ 35 */
28 36
29#include "lzfP.h" 37#include "lzfP.h"
30 38
31#define HSIZE (1 << (HLOG)) 39#define HSIZE (1 << (HLOG))
32 40
33/* 41/*
34 * don't play with this unless you benchmark! 42 * don't play with this unless you benchmark!
35 * decompression is not dependent on the hash function 43 * the data format is not dependent on the hash function.
36 * the hashing function might seem strange, just believe me 44 * the hash function might seem strange, just believe me,
37 * it works ;) 45 * it works ;)
38 */ 46 */
47#ifndef FRST
39#define FRST(p) (((p[0]) << 8) + p[1]) 48# define FRST(p) (((p[0]) << 8) | p[1])
40#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
41#define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) + h*3) & (HSIZE - 1)) 55# define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (HSIZE - 1))
56# endif
57#endif
42/* 58/*
43 * 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.
44 * (h * 57321 >> (3*8 - HLOG)) 60 * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1))
61 * the latter is also quite fast on newer CPUs, and compresses similarly.
62 *
45 * the next one is also quite good, albeit slow ;) 63 * the next one is also quite good, albeit slow ;)
46 * (int)(cos(h & 0xffffff) * 1e6) 64 * (int)(cos(h & 0xffffff) * 1e6)
47 */ 65 */
48 66
49#if 0 67#if 0
50/* original lzv-like hash function */ 68/* original lzv-like hash function, much worse and thus slower */
51# define FRST(p) (p[0] << 5) ^ p[1] 69# define FRST(p) (p[0] << 5) ^ p[1]
52# define NEXT(v,p) ((v) << 5) ^ p[2] 70# define NEXT(v,p) ((v) << 5) ^ p[2]
53# define IDX(h) ((h) & (HSIZE - 1)) 71# define IDX(h) ((h) & (HSIZE - 1))
54#endif 72#endif
55 73
56#define MAX_LIT (1 << 5) 74#define MAX_LIT (1 << 5)
57#define MAX_OFF (1 << 13) 75#define MAX_OFF (1 << 13)
58#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)
59 88
60/* 89/*
61 * compressed format 90 * compressed format
62 * 91 *
63 * 000LLLLL <L+1> ; literal 92 * 000LLLLL <L+1> ; literal, L+1=1..33 octets
64 * LLLOOOOO oooooooo ; backref L 93 * LLLooooo oooooooo ; backref L+1=1..7 octets, o+1=1..4096 offset
65 * 111OOOOO LLLLLLLL oooooooo ; backref L+7 94 * 111ooooo LLLLLLLL oooooooo ; backref L+8 octets, o+1=1..4096 offset
66 * 95 *
67 */ 96 */
68 97
69unsigned int 98unsigned int
70lzf_compress (const void *const in_data, unsigned int in_len, 99lzf_compress (const void *const in_data, unsigned int in_len,
71 void *out_data, unsigned int out_len 100 void *out_data, unsigned int out_len
72#if !LZF_STATE_ARG 101#if LZF_STATE_ARG
73 , LZF_STATE *htab 102 , LZF_STATE htab
74#endif 103#endif
75 ) 104 )
76{ 105{
77#if LZF_STATE_ARG 106#if !LZF_STATE_ARG
78 LZF_STATE htab; 107 LZF_STATE htab;
79#endif 108#endif
80 const u8 **hslot; 109 const u8 **hslot;
81 const u8 *ip = (const u8 *)in_data; 110 const u8 *ip = (const u8 *)in_data;
82 u8 *op = (u8 *)out_data; 111 u8 *op = (u8 *)out_data;
83 const u8 *in_end = ip + in_len; 112 const u8 *in_end = ip + in_len;
84 u8 *out_end = op + out_len; 113 u8 *out_end = op + out_len;
85 const u8 *ref; 114 const u8 *ref;
86 115
87 unsigned int hval = FRST (ip); 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
88 unsigned long off; 126 unsigned long off;
89 int lit = 0; 127#endif
128 unsigned int hval;
129 int lit;
130
131 if (!in_len || !out_len)
132 return 0;
90 133
91#if INIT_HTAB 134#if INIT_HTAB
92# if USE_MEMCPY
93 memset (htab, 0, sizeof (htab)); 135 memset (htab, 0, sizeof (htab));
94# else 136# if 0
95 for (hslot = htab; hslot < htab + HSIZE; hslot++) 137 for (hslot = htab; hslot < htab + HSIZE; hslot++)
96 *hslot++ = ip; 138 *hslot++ = ip;
97# endif
98#endif 139# endif
140#endif
99 141
100 do 142 lit = 0; op++; /* start run */
143
144 hval = FRST (ip);
145 while (ip < in_end - 2)
101 { 146 {
102 hval = NEXT (hval, ip); 147 hval = NEXT (hval, ip);
103 hslot = htab + IDX (hval); 148 hslot = htab + IDX (hval);
104 ref = *hslot; *hslot = ip; 149 ref = *hslot; *hslot = ip;
105 150
106 if (1 151 if (1
107#if INIT_HTAB && !USE_MEMCPY 152#if INIT_HTAB
108 && ref < ip /* the next test will actually take care of this, but it is faster */ 153 && ref < ip /* the next test will actually take care of this, but this is faster */
109#endif 154#endif
110 && (off = ip - ref - 1) < MAX_OFF 155 && (off = ip - ref - 1) < MAX_OFF
111 && ip + 4 < in_end 156 && ip + 4 < in_end
112 && ref > (u8 *)in_data 157 && ref > (u8 *)in_data
158 && ref[2] == ip[2]
113#if STRICT_ALIGN 159#if STRICT_ALIGN
114 && ref[0] == ip[0] 160 && (ref[1] << 8) | ref[0] == (ip[1] << 8) | ip[0]
115 && ref[1] == ip[1]
116 && ref[2] == ip[2]
117#else 161#else
118 && *(u16 *)ref == *(u16 *)ip 162 && *(u16 *)ref == *(u16 *)ip
119 && ref[2] == ip[2]
120#endif 163#endif
121 ) 164 )
122 { 165 {
123 /* match found at *ref++ */ 166 /* match found at *ref++ */
124 unsigned int len = 2; 167 unsigned int len = 2;
125 unsigned int maxlen = in_end - ip - len; 168 unsigned int maxlen = in_end - ip - len;
126 maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; 169 maxlen = maxlen > MAX_REF ? MAX_REF : maxlen;
127 170
128 do 171 if (expect_false (op + 3 + 1 >= out_end)) /* first a faster conservative test */
129 len++; 172 if (op - !lit + 3 + 1 >= out_end) /* second the exact but rare test */
173 return 0;
174
175 op [- lit - 1] = lit - 1; /* stop run */
176 op -= !lit; /* undo run if length is zero */
177
178 for (;;)
179 {
180 if (expect_true (maxlen > 16))
181 {
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 len++; if (ref [len] != ip [len]) break;
186
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 len++; if (ref [len] != ip [len]) break;
191
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 len++; if (ref [len] != ip [len]) break;
196
197 len++; if (ref [len] != ip [len]) break;
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 }
202
203 do
204 len++;
130 while (len < maxlen && ref[len] == ip[len]); 205 while (len < maxlen && ref[len] == ip[len]);
131 206
132 if (op + lit + 1 + 3 >= out_end) 207 break;
133 return 0; 208 }
134 209
135 if (lit) 210 len -= 2; /* len is now #octets - 1 */
136 { 211 ip++;
137 *op++ = lit - 1;
138 lit = -lit;
139 do
140 *op++ = ip[lit];
141 while (++lit);
142 }
143 212
144 len -= 2;
145 ip++;
146
147 if (len < 7) 213 if (len < 7)
148 { 214 {
149 *op++ = (off >> 8) + (len << 5); 215 *op++ = (off >> 8) + (len << 5);
150 } 216 }
151 else 217 else
152 { 218 {
153 *op++ = (off >> 8) + ( 7 << 5); 219 *op++ = (off >> 8) + ( 7 << 5);
154 *op++ = len - 7; 220 *op++ = len - 7;
155 } 221 }
156 222
157 *op++ = off; 223 *op++ = off;
224 lit = 0; op++; /* start run */
158 225
159#if ULTRA_FAST
160 ip += len; 226 ip += len + 1;
227
228 if (expect_false (ip >= in_end - 2))
229 break;
230
231#if ULTRA_FAST || VERY_FAST
232 --ip;
233# if VERY_FAST && !ULTRA_FAST
234 --ip;
235# endif
161 hval = FRST (ip); 236 hval = FRST (ip);
237
162 hval = NEXT (hval, ip); 238 hval = NEXT (hval, ip);
163 htab[IDX (hval)] = ip; 239 htab[IDX (hval)] = ip;
164 ip++; 240 ip++;
165#else 241
166 do 242# if VERY_FAST && !ULTRA_FAST
167 {
168 hval = NEXT (hval, ip); 243 hval = NEXT (hval, ip);
169 htab[IDX (hval)] = ip; 244 htab[IDX (hval)] = ip;
170 ip++; 245 ip++;
171 }
172 while (len--);
173#endif 246# endif
174 } 247#else
248 ip -= len + 1;
249
250 do
251 {
252 hval = NEXT (hval, ip);
253 htab[IDX (hval)] = ip;
254 ip++;
255 }
256 while (len--);
257#endif
258 }
175 else 259 else
176 { 260 {
177 /* one more literal byte we must copy */ 261 /* one more literal byte we must copy */
178 lit++; 262 if (expect_false (op >= out_end))
179 ip++; 263 return 0;
180 264
181 if (lit == MAX_LIT) 265 lit++; *op++ = *ip++;
182 {
183 if (op + 1 + MAX_LIT >= out_end)
184 return 0;
185 266
186 *op++ = MAX_LIT - 1; 267 if (expect_false (lit == MAX_LIT))
187#if USE_MEMCPY 268 {
188 memcpy (op, ip - MAX_LIT, MAX_LIT); 269 op [- lit - 1] = lit - 1; /* stop run */
189 op += MAX_LIT; 270 lit = 0; op++; /* start run */
190 lit = 0; 271 }
191#else 272 }
192 lit = -lit;
193 do
194 *op++ = ip[lit];
195 while (++lit);
196#endif
197 }
198 }
199 } 273 }
274
275 if (op + 3 > out_end) /* at most 3 bytes can be missing here */
276 return 0;
277
200 while (ip < in_end); 278 while (ip < in_end)
201
202 if (lit)
203 { 279 {
204 if (op + lit + 1 >= out_end) 280 lit++; *op++ = *ip++;
205 return 0;
206 281
207 *op++ = lit - 1; 282 if (expect_false (lit == MAX_LIT))
208 lit = -lit; 283 {
209 do 284 op [- lit - 1] = lit - 1; /* stop run */
210 *op++ = ip[lit]; 285 lit = 0; op++; /* start run */
211 while (++lit); 286 }
212 } 287 }
213 288
289 op [- lit - 1] = lit - 1; /* end run */
290 op -= !lit; /* undo run if length is zero */
291
214 return op - (u8 *) out_data; 292 return op - (u8 *)out_data;
215} 293}
294

Diff Legend

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