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 1.14 by root, Thu Jun 21 22:11:34 2007 UTC vs.
Revision 1.48 by root, Sat Jun 30 21:39:41 2012 UTC

1/* 1/*
2 * Copyright (c) 2000-2005 Marc Alexander Lehmann <schmorp@schmorp.de> 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 */
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))
53/*# define IDX(h) ((ip[0] * 121 ^ ip[1] * 33 ^ ip[2] * 1) & (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))
54#endif 64# endif
55/* 65#endif
56 * IDX works because it is very similar to a multiplicative hash, e.g.
57 * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1))
58 * the latter is also quite fast on newer CPUs, and compresses similarly.
59 *
60 * the next one is also quite good, albeit slow ;)
61 * (int)(cos(h & 0xffffff) * 1e6)
62 */
63 66
64#if 0 67#if 0
65/* original lzv-like hash function, much worse and thus slower */ 68/* original lzv-like hash function, much worse and thus slower */
66# define FRST(p) (p[0] << 5) ^ p[1] 69# define FRST(p) (p[0] << 5) ^ p[1]
67# define NEXT(v,p) ((v) << 5) ^ p[2] 70# define NEXT(v,p) ((v) << 5) ^ p[2]
70 73
71#define MAX_LIT (1 << 5) 74#define MAX_LIT (1 << 5)
72#define MAX_OFF (1 << 13) 75#define MAX_OFF (1 << 13)
73#define MAX_REF ((1 << 8) + (1 << 3)) 76#define MAX_REF ((1 << 8) + (1 << 3))
74 77
75#if (__i386 || __amd64) && __GNUC__ >= 3 78#if __GNUC__ >= 3
76# define lzf_movsb(dst, src, len) \ 79# define expect(expr,value) __builtin_expect ((expr),(value))
77 asm ("rep movsb" \ 80# define inline inline
78 : "=D" (dst), "=S" (src), "=c" (len) \ 81#else
79 : "0" (dst), "1" (src), "2" (len)); 82# define expect(expr,value) (expr)
83# define inline static
80#endif 84#endif
85
86#define expect_false(expr) expect ((expr) != 0, 0)
87#define expect_true(expr) expect ((expr) != 0, 1)
81 88
82/* 89/*
83 * compressed format 90 * compressed format
84 * 91 *
85 * 000LLLLL <L+1> ; literal 92 * 000LLLLL <L+1> ; literal, L+1=1..33 octets
86 * LLLooooo oooooooo ; backref L 93 * LLLooooo oooooooo ; backref L+1=1..7 octets, o+1=1..4096 offset
87 * 111ooooo LLLLLLLL oooooooo ; backref L+7 94 * 111ooooo LLLLLLLL oooooooo ; backref L+8 octets, o+1=1..4096 offset
88 * 95 *
89 */ 96 */
90 97
91unsigned int 98unsigned int
92lzf_compress (const void *const in_data, unsigned int in_len, 99lzf_compress (const void *const in_data, unsigned int in_len,
97 ) 104 )
98{ 105{
99#if !LZF_STATE_ARG 106#if !LZF_STATE_ARG
100 LZF_STATE htab; 107 LZF_STATE htab;
101#endif 108#endif
102 const u8 **hslot;
103 const u8 *ip = (const u8 *)in_data; 109 const u8 *ip = (const u8 *)in_data;
104 u8 *op = (u8 *)out_data; 110 u8 *op = (u8 *)out_data;
105 const u8 *in_end = ip + in_len; 111 const u8 *in_end = ip + in_len;
106 u8 *out_end = op + out_len; 112 u8 *out_end = op + out_len;
107 const u8 *ref; 113 const u8 *ref;
108 114
109 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
110 unsigned long off; 130 unsigned long off;
111 int lit = 0; 131#endif
132 unsigned int hval;
133 int lit;
134
135 if (!in_len || !out_len)
136 return 0;
112 137
113#if INIT_HTAB 138#if INIT_HTAB
114 memset (htab, 0, sizeof (htab)); 139 memset (htab, 0, sizeof (htab));
115# if 0
116 for (hslot = htab; hslot < htab + HSIZE; hslot++)
117 *hslot++ = ip;
118# endif 140#endif
119#endif
120 141
121 for (;;) 142 lit = 0; op++; /* start run */
143
144 hval = FRST (ip);
145 while (ip < in_end - 2)
122 { 146 {
123 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 )
124 { 166 {
125 hval = NEXT (hval, ip);
126 hslot = htab + IDX (hval);
127 ref = *hslot; *hslot = ip;
128
129 if (1
130#if INIT_HTAB && !USE_MEMCPY
131 && ref < ip /* the next test will actually take care of this, but this is faster */
132#endif
133 && (off = ip - ref - 1) < MAX_OFF
134 && ip + 4 < in_end
135 && ref > (u8 *)in_data
136#if STRICT_ALIGN
137 && ref[0] == ip[0]
138 && ref[1] == ip[1]
139 && ref[2] == ip[2]
140#else
141 && *(u16 *)ref == *(u16 *)ip
142 && ref[2] == ip[2]
143#endif
144 )
145 {
146 /* match found at *ref++ */ 167 /* match found at *ref++ */
147 unsigned int len = 2; 168 unsigned int len = 2;
148 unsigned int maxlen = in_end - ip - len; 169 unsigned int maxlen = in_end - ip - len;
149 maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; 170 maxlen = maxlen > MAX_REF ? MAX_REF : maxlen;
150 171
151 if (op + lit + 1 + 3 >= out_end) 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 */
152 return 0; 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 }
153 203
154 do 204 do
155 len++; 205 len++;
156 while (len < maxlen && ref[len] == ip[len]); 206 while (len < maxlen && ref[len] == ip[len]);
157 207
158 if (lit)
159 { 208 break;
160 *op++ = lit - 1;
161 lit = -lit;
162 do
163 *op++ = ip[lit];
164 while (++lit);
165 } 209 }
166 210
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);
167 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;
168 ip++; 256 ip++;
169
170 if (len < 7)
171 {
172 *op++ = (off >> 8) + (len << 5);
173 } 257 }
174 else
175 {
176 *op++ = (off >> 8) + ( 7 << 5);
177 *op++ = len - 7;
178 }
179
180 *op++ = off;
181
182#if ULTRA_FAST || VERY_FAST
183 ip += len;
184#if VERY_FAST && !ULTRA_FAST
185 --ip;
186#endif
187 hval = FRST (ip);
188
189 hval = NEXT (hval, ip);
190 htab[IDX (hval)] = ip;
191 ip++;
192
193#if VERY_FAST && !ULTRA_FAST
194 hval = NEXT (hval, ip);
195 htab[IDX (hval)] = ip;
196 ip++;
197#endif
198#else
199 do
200 {
201 hval = NEXT (hval, ip);
202 htab[IDX (hval)] = ip;
203 ip++;
204 }
205 while (len--); 258 while (len--);
206#endif 259#endif
207 continue;
208 }
209 } 260 }
210 else if (ip == in_end) 261 else
211 break;
212
213 /* one more literal byte we must copy */
214 lit++;
215 ip++;
216
217 if (lit == MAX_LIT)
218 { 262 {
219 if (op + 1 + MAX_LIT >= out_end) 263 /* one more literal byte we must copy */
264 if (expect_false (op >= out_end))
220 return 0; 265 return 0;
221 266
222 *op++ = MAX_LIT - 1; 267 lit++; *op++ = *ip++;
223 268
224#ifdef lzf_movsb 269 if (expect_false (lit == MAX_LIT))
225 ip -= lit;
226 lzf_movsb (op, ip, lit);
227#else
228 lit = -lit;
229 do 270 {
230 *op++ = ip[lit]; 271 op [- lit - 1] = lit - 1; /* stop run */
231 while (++lit); 272 lit = 0; op++; /* start run */
232#endif 273 }
233 } 274 }
234 } 275 }
235 276
236 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)
237 { 281 {
238 if (op + lit + 1 >= out_end) 282 lit++; *op++ = *ip++;
239 return 0;
240 283
241 *op++ = lit - 1; 284 if (expect_false (lit == MAX_LIT))
242#ifdef lzf_movsb 285 {
243 ip -= lit; 286 op [- lit - 1] = lit - 1; /* stop run */
244 lzf_movsb (op, ip, lit); 287 lit = 0; op++; /* start run */
245#else 288 }
246 lit = -lit;
247 do
248 *op++ = ip[lit];
249 while (++lit);
250#endif
251 } 289 }
252 290
291 op [- lit - 1] = lit - 1; /* end run */
292 op -= !lit; /* undo run if length is zero */
293
253 return op - (u8 *) out_data; 294 return op - (u8 *)out_data;
254} 295}
255 296

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines