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 1.28 by root, Fri Nov 30 10:34:26 2007 UTC

… …
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))
… …
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 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
49# 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))
50/*# define IDX(h) ((ip[0] * 121 ^ ip[1] * 33 ^ ip[2] * 1) & (HSIZE-1))*/ 56# endif
51#endif 57#endif
52/* 58/*
53 * 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.
54 * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1)) 60 * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1))
55 * the latter is also quite fast on newer CPUs, and compresses similarly. 61 * the latter is also quite fast on newer CPUs, and compresses similarly.
… …
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
… …
101 u8 *op = (u8 *)out_data; 111 u8 *op = (u8 *)out_data;
102 const u8 *in_end = ip + in_len; 112 const u8 *in_end = ip + in_len;
103 u8 *out_end = op + out_len; 113 u8 *out_end = op + out_len;
104 const u8 *ref; 114 const u8 *ref;
105 115
106 unsigned int hval = FRST (ip); 116 unsigned int hval;
107 unsigned long off; 117 unsigned long off;
108 int lit = 0; 118 int lit;
119
120 if (!in_len || !out_len)
121 return 0;
109 122
110#if INIT_HTAB 123#if INIT_HTAB
111 memset (htab, 0, sizeof (htab)); 124 memset (htab, 0, sizeof (htab));
112# if 0 125# if 0
113 for (hslot = htab; hslot < htab + HSIZE; hslot++) 126 for (hslot = htab; hslot < htab + HSIZE; hslot++)
114 *hslot++ = ip; 127 *hslot++ = ip;
115# endif 128# endif
116#endif 129#endif
117 130
118 for (;;) 131 lit = 0; op++; /* start run */
132
133 hval = FRST (ip);
134 while (ip < in_end - 2)
119 { 135 {
120 if (ip < in_end - 2) 136 hval = NEXT (hval, ip);
137 hslot = htab + IDX (hval);
138 ref = *hslot; *hslot = ip;
139
140 if (1
141#if INIT_HTAB
142 && ref < ip /* the next test will actually take care of this, but this is faster */
143#endif
144 && (off = ip - ref - 1) < MAX_OFF
145 && ip + 4 < in_end
146 && ref > (u8 *)in_data
147#if STRICT_ALIGN
148 && ref[0] == ip[0]
149 && ref[1] == ip[1]
150 && ref[2] == ip[2]
151#else
152 && *(u16 *)ref == *(u16 *)ip
153 && ref[2] == ip[2]
154#endif
155 )
121 { 156 {
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++ */ 157 /* match found at *ref++ */
144 unsigned int len = 2; 158 unsigned int len = 2;
145 unsigned int maxlen = in_end - ip - len; 159 unsigned int maxlen = in_end - ip - len;
146 maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; 160 maxlen = maxlen > MAX_REF ? MAX_REF : maxlen;
147 161
148 if (op + lit + 1 + 3 >= out_end) 162 if (maxlen > 8) maxlen = 8;//D
163
164 op [- lit - 1] = lit - 1; /* stop run */
165 op -= !lit; /* undo run if length is zero */
166
167 if (expect_false (op + 3 + 1 >= out_end))
149 return 0; 168 return 0;
169
170 for (;;)
171 {
172 if (expect_true (maxlen > 16))
173 {
174 len++; if (ref [len] != ip [len]) break;
175 len++; if (ref [len] != ip [len]) break;
176 len++; if (ref [len] != ip [len]) break;
177 len++; if (ref [len] != ip [len]) break;
178
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 len++; if (ref [len] != ip [len]) break;
183
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 len++; if (ref [len] != ip [len]) break;
188
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 len++; if (ref [len] != ip [len]) break;
193 }
150 194
151 do 195 do
152 len++; 196 len++;
153 while (len < maxlen && ref[len] == ip[len]); 197 while (len < maxlen && ref[len] == ip[len]);
154 198
155 if (lit)
156 { 199 break;
157 *op++ = lit - 1;
158 lit = -lit;
159 do
160 *op++ = ip[lit];
161 while (++lit);
162 } 200 }
163 201
164 len -= 2; 202 len -= 2;
165 ip++; 203 ip++;
166 204
167 if (len < 7) 205 if (len < 7)
168 { 206 {
169 *op++ = (off >> 8) + (len << 5); 207 *op++ = (off >> 8) + (len << 5);
170 } 208 }
171 else 209 else
172 { 210 {
173 *op++ = (off >> 8) + ( 7 << 5); 211 *op++ = (off >> 8) + ( 7 << 5);
174 *op++ = len - 7; 212 *op++ = len - 7;
175 } 213 }
176 214
177 *op++ = off; 215 *op++ = off;
178 216
179#if ULTRA_FAST || VERY_FAST 217#if ULTRA_FAST || VERY_FAST
180 ip += len; 218 ip += len;
181#if VERY_FAST && !ULTRA_FAST 219#if VERY_FAST && !ULTRA_FAST
182 --ip; 220 --ip;
183#endif 221#endif
184 hval = FRST (ip); 222 hval = FRST (ip);
185 223
224 hval = NEXT (hval, ip);
225 htab[IDX (hval)] = ip;
226 ip++;
227
228#if VERY_FAST && !ULTRA_FAST
229 hval = NEXT (hval, ip);
230 htab[IDX (hval)] = ip;
231 ip++;
232#endif
233#else
234 do
235 {
186 hval = NEXT (hval, ip); 236 hval = NEXT (hval, ip);
187 htab[IDX (hval)] = ip; 237 htab[IDX (hval)] = ip;
188 ip++; 238 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 } 239 }
202 while (len--); 240 while (len--);
203#endif 241#endif
204 continue; 242
205 } 243 lit = 0; op++; /* start run */
206 } 244 }
207 else if (ip == in_end) 245 else
208 break;
209
210 /* one more literal byte we must copy */
211 lit++;
212 ip++;
213
214 if (lit == MAX_LIT)
215 { 246 {
216 if (op + 1 + MAX_LIT >= out_end) 247 /* one more literal byte we must copy */
248 if (expect_false (op >= out_end))
217 return 0; 249 return 0;
218 250
219 *op++ = MAX_LIT - 1; 251 lit++; *op++ = *ip++;
220 252
221#ifdef lzf_movsb 253 if (expect_false (lit == MAX_LIT))
222 ip -= lit;
223 lzf_movsb (op, ip, lit);
224#else
225 lit = -lit;
226 do 254 {
227 *op++ = ip[lit]; 255 op [- lit - 1] = lit - 1; /* stop run */
228 while (++lit); 256 lit = 0; op++; /* start run */
229#endif 257 }
230 } 258 }
231 } 259 }
232 260
233 if (lit) 261 if (op + 2 > out_end) /* at most 2 bytes can be missing here */
262 return 0;
263
264 while (ip < in_end)
234 { 265 {
235 if (op + lit + 1 >= out_end) 266 lit++; *op++ = *ip++;
236 return 0;
237 267
238 *op++ = lit - 1; 268 if (expect_false (lit == MAX_LIT))
239#ifdef lzf_movsb 269 {
240 ip -= lit; 270 op [- lit - 1] = lit - 1; /* stop run */
241 lzf_movsb (op, ip, lit); 271 lit = 0; op++; /* start run */
242#else 272 }
243 lit = -lit;
244 do
245 *op++ = ip[lit];
246 while (++lit);
247#endif
248 } 273 }
249 274
275 op [- lit - 1] = lit - 1; /* end run */
276 op -= !lit; /* undo run if length is zero */
277
250 return op - (u8 *) out_data; 278 return op - (u8 *)out_data;
251} 279}
252 280

Diff Legend

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