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 1.15 by root, Fri Nov 2 12:39:20 2007 UTC vs.
Revision 1.45 by root, Sun Mar 27 23:53:23 2011 UTC

1/* 1/*
2 * Copyright (c) 2000-2007 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,
… …
21 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- 21 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
22 * 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
23 * OF THE POSSIBILITY OF SUCH DAMAGE. 23 * OF THE POSSIBILITY OF SUCH DAMAGE.
24 * 24 *
25 * 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
26 * 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,
27 * 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
28 * 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
29 * 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
30 * BSD license, indicate your decision by deleting the provisions above and 30 * version of this file under the BSD license, indicate your decision
31 * 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
32 * 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
33 * of this file under either the BSD or the GPL. 34 * either the BSD or the GPL.
34 */ 35 */
35 36
36#include "lzfP.h" 37#include "lzfP.h"
37 38
38#define HSIZE (1 << (HLOG)) 39#define HSIZE (1 << (HLOG))
39 40
40/* 41/*
41 * don't play with this unless you benchmark! 42 * don't play with this unless you benchmark!
42 * decompression is not dependent on the hash function 43 * the data format is not dependent on the hash function.
43 * the hashing function might seem strange, just believe me 44 * the hash function might seem strange, just believe me,
44 * it works ;) 45 * it works ;)
45 */ 46 */
46#ifndef FRST 47#ifndef FRST
47# define FRST(p) (((p[0]) << 8) | p[1]) 48# define FRST(p) (((p[0]) << 8) | p[1])
48# 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
49# 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))
50/*# 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))
51#endif 64# endif
52/* 65#endif
53 * IDX works because it is very similar to a multiplicative hash, e.g.
54 * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1))
55 * the latter is also quite fast on newer CPUs, and compresses similarly.
56 *
57 * the next one is also quite good, albeit slow ;)
58 * (int)(cos(h & 0xffffff) * 1e6)
59 */
60 66
61#if 0 67#if 0
62/* original lzv-like hash function, much worse and thus slower */ 68/* original lzv-like hash function, much worse and thus slower */
63# define FRST(p) (p[0] << 5) ^ p[1] 69# define FRST(p) (p[0] << 5) ^ p[1]
64# define NEXT(v,p) ((v) << 5) ^ p[2] 70# define NEXT(v,p) ((v) << 5) ^ p[2]
… …
67 73
68#define MAX_LIT (1 << 5) 74#define MAX_LIT (1 << 5)
69#define MAX_OFF (1 << 13) 75#define MAX_OFF (1 << 13)
70#define MAX_REF ((1 << 8) + (1 << 3)) 76#define MAX_REF ((1 << 8) + (1 << 3))
71 77
72#if (__i386 || __amd64) && __GNUC__ >= 3 78#if __GNUC__ >= 3
73# define lzf_movsb(dst, src, len) \ 79# define expect(expr,value) __builtin_expect ((expr),(value))
74 asm ("rep movsb" \ 80# define inline inline
75 : "=D" (dst), "=S" (src), "=c" (len) \ 81#else
76 : "0" (dst), "1" (src), "2" (len)); 82# define expect(expr,value) (expr)
83# define inline static
77#endif 84#endif
85
86#define expect_false(expr) expect ((expr) != 0, 0)
87#define expect_true(expr) expect ((expr) != 0, 1)
78 88
79/* 89/*
80 * compressed format 90 * compressed format
81 * 91 *
82 * 000LLLLL <L+1> ; literal 92 * 000LLLLL <L+1> ; literal, L+1=1..33 octets
83 * LLLooooo oooooooo ; backref L 93 * LLLooooo oooooooo ; backref L+1=1..7 octets, o+1=1..4096 offset
84 * 111ooooo LLLLLLLL oooooooo ; backref L+7 94 * 111ooooo LLLLLLLL oooooooo ; backref L+8 octets, o+1=1..4096 offset
85 * 95 *
86 */ 96 */
87 97
88unsigned int 98unsigned int
89lzf_compress (const void *const in_data, unsigned int in_len, 99lzf_compress (const void *const in_data, unsigned int in_len,
… …
94 ) 104 )
95{ 105{
96#if !LZF_STATE_ARG 106#if !LZF_STATE_ARG
97 LZF_STATE htab; 107 LZF_STATE htab;
98#endif 108#endif
99 const u8 **hslot;
100 const u8 *ip = (const u8 *)in_data; 109 const u8 *ip = (const u8 *)in_data;
101 u8 *op = (u8 *)out_data; 110 u8 *op = (u8 *)out_data;
102 const u8 *in_end = ip + in_len; 111 const u8 *in_end = ip + in_len;
103 u8 *out_end = op + out_len; 112 u8 *out_end = op + out_len;
104 const u8 *ref; 113 const u8 *ref;
105 114
106 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
107 unsigned long off; 125 unsigned long off;
108 int lit = 0; 126#endif
127 unsigned int hval;
128 int lit;
129
130 if (!in_len || !out_len)
131 return 0;
109 132
110#if INIT_HTAB 133#if INIT_HTAB
111 memset (htab, 0, sizeof (htab)); 134 memset (htab, 0, sizeof (htab));
112# if 0
113 for (hslot = htab; hslot < htab + HSIZE; hslot++)
114 *hslot++ = ip;
115# endif 135#endif
116#endif
117 136
118 for (;;) 137 lit = 0; op++; /* start run */
138
139 hval = FRST (ip);
140 while (ip < in_end - 2)
119 { 141 {
120 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 )
121 { 161 {
122 hval = NEXT (hval, ip);
123 hslot = htab + IDX (hval);
124 ref = *hslot; *hslot = ip;
125
126 if (1
127#if INIT_HTAB && !USE_MEMCPY
128 && ref < ip /* the next test will actually take care of this, but this is faster */
129#endif
130 && (off = ip - ref - 1) < MAX_OFF
131 && ip + 4 < in_end
132 && ref > (u8 *)in_data
133#if STRICT_ALIGN
134 && ref[0] == ip[0]
135 && ref[1] == ip[1]
136 && ref[2] == ip[2]
137#else
138 && *(u16 *)ref == *(u16 *)ip
139 && ref[2] == ip[2]
140#endif
141 )
142 {
143 /* match found at *ref++ */ 162 /* match found at *ref++ */
144 unsigned int len = 2; 163 unsigned int len = 2;
145 unsigned int maxlen = in_end - ip - len; 164 unsigned int maxlen = in_end - ip - len;
146 maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; 165 maxlen = maxlen > MAX_REF ? MAX_REF : maxlen;
147 166
148 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 */
149 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 }
150 198
151 do 199 do
152 len++; 200 len++;
153 while (len < maxlen && ref[len] == ip[len]); 201 while (len < maxlen && ref[len] == ip[len]);
154 202
155 if (lit)
156 { 203 break;
157 *op++ = lit - 1;
158 lit = -lit;
159 do
160 *op++ = ip[lit];
161 while (++lit);
162 } 204 }
163 205
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);
164 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;
165 ip++; 251 ip++;
166
167 if (len < 7)
168 {
169 *op++ = (off >> 8) + (len << 5);
170 } 252 }
171 else
172 {
173 *op++ = (off >> 8) + ( 7 << 5);
174 *op++ = len - 7;
175 }
176
177 *op++ = off;
178
179#if ULTRA_FAST || VERY_FAST
180 ip += len;
181#if VERY_FAST && !ULTRA_FAST
182 --ip;
183#endif
184 hval = FRST (ip);
185
186 hval = NEXT (hval, ip);
187 htab[IDX (hval)] = ip;
188 ip++;
189
190#if VERY_FAST && !ULTRA_FAST
191 hval = NEXT (hval, ip);
192 htab[IDX (hval)] = ip;
193 ip++;
194#endif
195#else
196 do
197 {
198 hval = NEXT (hval, ip);
199 htab[IDX (hval)] = ip;
200 ip++;
201 }
202 while (len--); 253 while (len--);
203#endif 254#endif
204 continue;
205 }
206 } 255 }
207 else if (ip == in_end) 256 else
208 break;
209
210 /* one more literal byte we must copy */
211 lit++;
212 ip++;
213
214 if (lit == MAX_LIT)
215 { 257 {
216 if (op + 1 + MAX_LIT >= out_end) 258 /* one more literal byte we must copy */
259 if (expect_false (op >= out_end))
217 return 0; 260 return 0;
218 261
219 *op++ = MAX_LIT - 1; 262 lit++; *op++ = *ip++;
220 263
221#ifdef lzf_movsb 264 if (expect_false (lit == MAX_LIT))
222 ip -= lit;
223 lzf_movsb (op, ip, lit);
224#else
225 lit = -lit;
226 do 265 {
227 *op++ = ip[lit]; 266 op [- lit - 1] = lit - 1; /* stop run */
228 while (++lit); 267 lit = 0; op++; /* start run */
229#endif 268 }
230 } 269 }
231 } 270 }
232 271
233 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)
234 { 276 {
235 if (op + lit + 1 >= out_end) 277 lit++; *op++ = *ip++;
236 return 0;
237 278
238 *op++ = lit - 1; 279 if (expect_false (lit == MAX_LIT))
239#ifdef lzf_movsb 280 {
240 ip -= lit; 281 op [- lit - 1] = lit - 1; /* stop run */
241 lzf_movsb (op, ip, lit); 282 lit = 0; op++; /* start run */
242#else 283 }
243 lit = -lit;
244 do
245 *op++ = ip[lit];
246 while (++lit);
247#endif
248 } 284 }
249 285
286 op [- lit - 1] = lit - 1; /* end run */
287 op -= !lit; /* undo run if length is zero */
288
250 return op - (u8 *) out_data; 289 return op - (u8 *)out_data;
251} 290}
252 291

Diff Legend

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