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 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, 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 | |
88 | unsigned int |
98 | unsigned int |
89 | lzf_compress (const void *const in_data, unsigned int in_len, |
99 | lzf_compress (const void *const in_data, unsigned int in_len, |
… | |
… | |
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 | /* 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 |
107 | unsigned long off; |
126 | unsigned long off; |
108 | int lit = 0; |
127 | #endif |
|
|
128 | unsigned int hval; |
|
|
129 | int lit; |
|
|
130 | |
|
|
131 | if (!in_len || !out_len) |
|
|
132 | return 0; |
109 | |
133 | |
110 | #if INIT_HTAB |
134 | #if INIT_HTAB |
111 | memset (htab, 0, sizeof (htab)); |
135 | memset (htab, 0, sizeof (htab)); |
112 | # if 0 |
136 | # if 0 |
113 | for (hslot = htab; hslot < htab + HSIZE; hslot++) |
137 | for (hslot = htab; hslot < htab + HSIZE; hslot++) |
114 | *hslot++ = ip; |
138 | *hslot++ = ip; |
115 | # endif |
139 | # endif |
116 | #endif |
140 | #endif |
117 | |
141 | |
118 | for (;;) |
142 | lit = 0; op++; /* start run */ |
|
|
143 | |
|
|
144 | hval = FRST (ip); |
|
|
145 | while (ip < in_end - 2) |
119 | { |
146 | { |
120 | if (ip < in_end - 2) |
147 | hval = NEXT (hval, ip); |
|
|
148 | hslot = htab + IDX (hval); |
|
|
149 | ref = *hslot; *hslot = ip; |
|
|
150 | |
|
|
151 | if (1 |
|
|
152 | #if INIT_HTAB |
|
|
153 | && ref < ip /* the next test will actually take care of this, but this is faster */ |
|
|
154 | #endif |
|
|
155 | && (off = ip - ref - 1) < MAX_OFF |
|
|
156 | && ip + 4 < in_end |
|
|
157 | && ref > (u8 *)in_data |
|
|
158 | && ref[2] == ip[2] |
|
|
159 | #if STRICT_ALIGN |
|
|
160 | && (ref[1] << 8) | ref[0] == (ip[1] << 8) | ip[0] |
|
|
161 | #else |
|
|
162 | && *(u16 *)ref == *(u16 *)ip |
|
|
163 | #endif |
|
|
164 | ) |
121 | { |
165 | { |
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++ */ |
166 | /* match found at *ref++ */ |
144 | unsigned int len = 2; |
167 | unsigned int len = 2; |
145 | unsigned int maxlen = in_end - ip - len; |
168 | unsigned int maxlen = in_end - ip - len; |
146 | maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; |
169 | maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; |
147 | |
170 | |
148 | if (op + lit + 1 + 3 >= out_end) |
171 | if (expect_false (op + 3 + 1 >= out_end)) /* first a faster conservative test */ |
|
|
172 | if (op - !lit + 3 + 1 >= out_end) /* second the exact but rare test */ |
149 | return 0; |
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 | } |
150 | |
202 | |
151 | do |
203 | do |
152 | len++; |
204 | len++; |
153 | while (len < maxlen && ref[len] == ip[len]); |
205 | while (len < maxlen && ref[len] == ip[len]); |
154 | |
206 | |
155 | if (lit) |
|
|
156 | { |
207 | break; |
157 | *op++ = lit - 1; |
|
|
158 | lit = -lit; |
|
|
159 | do |
|
|
160 | *op++ = ip[lit]; |
|
|
161 | while (++lit); |
|
|
162 | } |
208 | } |
163 | |
209 | |
164 | len -= 2; |
210 | len -= 2; /* len is now #octets - 1 */ |
165 | ip++; |
211 | ip++; |
166 | |
212 | |
167 | if (len < 7) |
213 | if (len < 7) |
168 | { |
214 | { |
169 | *op++ = (off >> 8) + (len << 5); |
215 | *op++ = (off >> 8) + (len << 5); |
170 | } |
216 | } |
171 | else |
217 | else |
172 | { |
218 | { |
173 | *op++ = (off >> 8) + ( 7 << 5); |
219 | *op++ = (off >> 8) + ( 7 << 5); |
174 | *op++ = len - 7; |
220 | *op++ = len - 7; |
175 | } |
221 | } |
176 | |
222 | |
177 | *op++ = off; |
223 | *op++ = off; |
|
|
224 | lit = 0; op++; /* start run */ |
|
|
225 | |
|
|
226 | ip += len + 1; |
|
|
227 | |
|
|
228 | if (expect_false (ip >= in_end - 2)) |
|
|
229 | break; |
178 | |
230 | |
179 | #if ULTRA_FAST || VERY_FAST |
231 | #if ULTRA_FAST || VERY_FAST |
180 | ip += len; |
232 | --ip; |
181 | #if VERY_FAST && !ULTRA_FAST |
233 | # if VERY_FAST && !ULTRA_FAST |
182 | --ip; |
234 | --ip; |
183 | #endif |
235 | # endif |
184 | hval = FRST (ip); |
236 | hval = FRST (ip); |
185 | |
237 | |
|
|
238 | hval = NEXT (hval, ip); |
|
|
239 | htab[IDX (hval)] = ip; |
|
|
240 | ip++; |
|
|
241 | |
|
|
242 | # if VERY_FAST && !ULTRA_FAST |
|
|
243 | hval = NEXT (hval, ip); |
|
|
244 | htab[IDX (hval)] = ip; |
|
|
245 | ip++; |
|
|
246 | # endif |
|
|
247 | #else |
|
|
248 | ip -= len + 1; |
|
|
249 | |
|
|
250 | do |
|
|
251 | { |
186 | hval = NEXT (hval, ip); |
252 | hval = NEXT (hval, ip); |
187 | htab[IDX (hval)] = ip; |
253 | htab[IDX (hval)] = ip; |
188 | ip++; |
254 | 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 | } |
255 | } |
202 | while (len--); |
256 | while (len--); |
203 | #endif |
257 | #endif |
204 | continue; |
|
|
205 | } |
|
|
206 | } |
258 | } |
207 | else if (ip == in_end) |
259 | else |
208 | break; |
|
|
209 | |
|
|
210 | /* one more literal byte we must copy */ |
|
|
211 | lit++; |
|
|
212 | ip++; |
|
|
213 | |
|
|
214 | if (lit == MAX_LIT) |
|
|
215 | { |
260 | { |
216 | if (op + 1 + MAX_LIT >= out_end) |
261 | /* one more literal byte we must copy */ |
|
|
262 | if (expect_false (op >= out_end)) |
217 | return 0; |
263 | return 0; |
218 | |
264 | |
219 | *op++ = MAX_LIT - 1; |
265 | lit++; *op++ = *ip++; |
220 | |
266 | |
221 | #ifdef lzf_movsb |
267 | if (expect_false (lit == MAX_LIT)) |
222 | ip -= lit; |
|
|
223 | lzf_movsb (op, ip, lit); |
|
|
224 | #else |
|
|
225 | lit = -lit; |
|
|
226 | do |
268 | { |
227 | *op++ = ip[lit]; |
269 | op [- lit - 1] = lit - 1; /* stop run */ |
228 | while (++lit); |
270 | lit = 0; op++; /* start run */ |
229 | #endif |
271 | } |
230 | } |
272 | } |
231 | } |
273 | } |
232 | |
274 | |
233 | if (lit) |
275 | if (op + 3 > out_end) /* at most 3 bytes can be missing here */ |
|
|
276 | return 0; |
|
|
277 | |
|
|
278 | while (ip < in_end) |
234 | { |
279 | { |
235 | if (op + lit + 1 >= out_end) |
280 | lit++; *op++ = *ip++; |
236 | return 0; |
|
|
237 | |
281 | |
238 | *op++ = lit - 1; |
282 | if (expect_false (lit == MAX_LIT)) |
239 | #ifdef lzf_movsb |
283 | { |
240 | ip -= lit; |
284 | op [- lit - 1] = lit - 1; /* stop run */ |
241 | lzf_movsb (op, ip, lit); |
285 | lit = 0; op++; /* start run */ |
242 | #else |
286 | } |
243 | lit = -lit; |
|
|
244 | do |
|
|
245 | *op++ = ip[lit]; |
|
|
246 | while (++lit); |
|
|
247 | #endif |
|
|
248 | } |
287 | } |
249 | |
288 | |
|
|
289 | op [- lit - 1] = lit - 1; /* end run */ |
|
|
290 | op -= !lit; /* undo run if length is zero */ |
|
|
291 | |
250 | return op - (u8 *) out_data; |
292 | return op - (u8 *)out_data; |
251 | } |
293 | } |
252 | |
294 | |