1 | /* |
1 | /* |
2 | * Copyright (c) 2000-2007 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 | * |
13 | * |
14 | * 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 |
15 | * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- |
15 | * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- |
16 | * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO |
16 | * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO |
17 | * 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- |
18 | * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
18 | * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
… | |
… | |
38 | |
38 | |
39 | #define HSIZE (1 << (HLOG)) |
39 | #define HSIZE (1 << (HLOG)) |
40 | |
40 | |
41 | /* |
41 | /* |
42 | * don't play with this unless you benchmark! |
42 | * don't play with this unless you benchmark! |
43 | * decompression is not dependent on the hash function |
43 | * the data format is not dependent on the hash function. |
44 | * the hashing function might seem strange, just believe me |
44 | * the hash function might seem strange, just believe me, |
45 | * it works ;) |
45 | * it works ;) |
46 | */ |
46 | */ |
47 | #ifndef FRST |
47 | #ifndef FRST |
48 | # define FRST(p) (((p[0]) << 8) | p[1]) |
48 | # define FRST(p) (((p[0]) << 8) | p[1]) |
49 | # define NEXT(v,p) (((v) << 8) | p[2]) |
49 | # define NEXT(v,p) (((v) << 8) | p[2]) |
|
|
50 | # if MULTIPLICATION_IS_SLOW |
50 | # if ULTRA_FAST |
51 | # if ULTRA_FAST |
51 | # define IDX(h) ((( h >> (3*8 - HLOG)) - h ) & (HSIZE - 1)) |
52 | # define IDX(h) ((( h >> (3*8 - HLOG)) - h ) & (HSIZE - 1)) |
52 | # elif VERY_FAST |
53 | # elif VERY_FAST |
53 | # define IDX(h) ((( h >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) |
54 | # define IDX(h) ((( h >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) |
|
|
55 | # else |
|
|
56 | # define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) |
|
|
57 | # endif |
54 | # else |
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 | */ |
55 | # define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (HSIZE - 1)) |
63 | # define IDX(h) (((h * 0x1e35a7bdU) >> (32 - HLOG - 8)) & (HSIZE - 1)) |
56 | # endif |
64 | # endif |
57 | #endif |
65 | #endif |
58 | /* |
|
|
59 | * IDX works because it is very similar to a multiplicative hash, e.g. |
|
|
60 | * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1)) |
|
|
61 | * the latter is also quite fast on newer CPUs, and compresses similarly. |
|
|
62 | * |
|
|
63 | * the next one is also quite good, albeit slow ;) |
|
|
64 | * (int)(cos(h & 0xffffff) * 1e6) |
|
|
65 | */ |
|
|
66 | |
66 | |
67 | #if 0 |
67 | #if 0 |
68 | /* original lzv-like hash function, much worse and thus slower */ |
68 | /* original lzv-like hash function, much worse and thus slower */ |
69 | # define FRST(p) (p[0] << 5) ^ p[1] |
69 | # define FRST(p) (p[0] << 5) ^ p[1] |
70 | # define NEXT(v,p) ((v) << 5) ^ p[2] |
70 | # define NEXT(v,p) ((v) << 5) ^ p[2] |
71 | # define IDX(h) ((h) & (HSIZE - 1)) |
71 | # define IDX(h) ((h) & (HSIZE - 1)) |
72 | #endif |
72 | #endif |
73 | |
73 | |
74 | #define MAX_LIT (1 << 5) |
|
|
75 | #define MAX_OFF (1 << 13) |
|
|
76 | #define MAX_REF ((1 << 8) + (1 << 3)) |
|
|
77 | |
|
|
78 | #if __GNUC__ >= 3 |
74 | #if __GNUC__ >= 3 |
79 | # define expect(expr,value) __builtin_expect ((expr),(value)) |
75 | # define expect(expr,value) __builtin_expect ((expr),(value)) |
80 | # define inline inline |
76 | # define inline inline |
81 | #else |
77 | #else |
82 | # define expect(expr,value) (expr) |
78 | # define expect(expr,value) (expr) |
… | |
… | |
87 | #define expect_true(expr) expect ((expr) != 0, 1) |
83 | #define expect_true(expr) expect ((expr) != 0, 1) |
88 | |
84 | |
89 | /* |
85 | /* |
90 | * compressed format |
86 | * compressed format |
91 | * |
87 | * |
92 | * 000LLLLL <L+1> ; literal |
88 | * 000LLLLL <L+1> ; literal, L+1=1..33 octets |
93 | * LLLooooo oooooooo ; backref L |
89 | * LLLooooo oooooooo ; backref L+1=1..7 octets, o+1=1..4096 offset |
94 | * 111ooooo LLLLLLLL oooooooo ; backref L+7 |
90 | * 111ooooo LLLLLLLL oooooooo ; backref L+8 octets, o+1=1..4096 offset |
95 | * |
91 | * |
96 | */ |
92 | */ |
97 | |
93 | |
98 | unsigned int |
94 | unsigned int |
99 | lzf_compress (const void *const in_data, unsigned int in_len, |
95 | lzf_compress (const void *const in_data, unsigned int in_len, |
… | |
… | |
104 | ) |
100 | ) |
105 | { |
101 | { |
106 | #if !LZF_STATE_ARG |
102 | #if !LZF_STATE_ARG |
107 | LZF_STATE htab; |
103 | LZF_STATE htab; |
108 | #endif |
104 | #endif |
109 | const u8 **hslot; |
|
|
110 | const u8 *ip = (const u8 *)in_data; |
105 | const u8 *ip = (const u8 *)in_data; |
111 | u8 *op = (u8 *)out_data; |
106 | u8 *op = (u8 *)out_data; |
112 | const u8 *in_end = ip + in_len; |
107 | const u8 *in_end = ip + in_len; |
113 | u8 *out_end = op + out_len; |
108 | u8 *out_end = op + out_len; |
114 | const u8 *ref; |
109 | const u8 *ref; |
115 | |
110 | |
116 | /* off requires a type wide enough to hold a general pointer difference. |
111 | /* 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 |
112 | * 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 |
113 | * works for differences within a single object). We also assume that |
119 | * no bit pattern traps. Since the only platform that is both non-POSIX |
114 | * 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 |
115 | * and fails to support both assumptions is windows 64 bit, we make a |
121 | * special workaround for it. |
116 | * special workaround for it. |
122 | */ |
117 | */ |
123 | #if defined (WIN32) && defined (_M_X64) |
118 | #if defined (_WIN32) && defined (_M_X64) |
124 | unsigned _int64 off; /* workaround for missing POSIX compliance */ |
119 | /* workaround for missing POSIX compliance */ |
|
|
120 | #if __GNUC__ |
|
|
121 | unsigned long long off; |
|
|
122 | #else |
|
|
123 | unsigned __int64 off; |
|
|
124 | #endif |
125 | #else |
125 | #else |
126 | unsigned long off; |
126 | unsigned long off; |
127 | #endif |
127 | #endif |
128 | unsigned int hval; |
128 | unsigned int hval; |
129 | int lit; |
129 | int lit; |
130 | |
130 | |
131 | if (!in_len || !out_len) |
131 | if (!in_len || !out_len) |
132 | return 0; |
132 | return 0; |
133 | |
133 | |
134 | #if INIT_HTAB |
134 | #if INIT_HTAB |
135 | memset (htab, 0, sizeof (htab)); |
135 | memset (htab, 0, sizeof (LZF_STATE)); |
136 | # if 0 |
|
|
137 | for (hslot = htab; hslot < htab + HSIZE; hslot++) |
|
|
138 | *hslot++ = ip; |
|
|
139 | # endif |
|
|
140 | #endif |
136 | #endif |
141 | |
137 | |
142 | lit = 0; op++; /* start run */ |
138 | lit = 0; op++; /* start run */ |
143 | |
139 | |
144 | hval = FRST (ip); |
140 | hval = FRST (ip); |
145 | while (ip < in_end - 2) |
141 | while (ip < in_end - 2) |
146 | { |
142 | { |
|
|
143 | LZF_HSLOT *hslot; |
|
|
144 | |
147 | hval = NEXT (hval, ip); |
145 | hval = NEXT (hval, ip); |
148 | hslot = htab + IDX (hval); |
146 | hslot = htab + IDX (hval); |
149 | ref = *hslot; *hslot = ip; |
147 | ref = *hslot + LZF_HSLOT_BIAS; *hslot = ip - LZF_HSLOT_BIAS; |
150 | |
148 | |
151 | if (1 |
149 | if (1 |
152 | #if INIT_HTAB |
150 | #if INIT_HTAB |
153 | && ref < ip /* the next test will actually take care of this, but this is faster */ |
151 | && ref < ip /* the next test will actually take care of this, but this is faster */ |
154 | #endif |
152 | #endif |
155 | && (off = ip - ref - 1) < MAX_OFF |
153 | && (off = ip - ref - 1) < LZF_MAX_OFF |
156 | && ip + 4 < in_end |
|
|
157 | && ref > (u8 *)in_data |
154 | && ref > (u8 *)in_data |
|
|
155 | && ref[2] == ip[2] |
158 | #if STRICT_ALIGN |
156 | #if STRICT_ALIGN |
159 | && ref[0] == ip[0] |
157 | && ((ref[1] << 8) | ref[0]) == ((ip[1] << 8) | ip[0]) |
160 | && ref[1] == ip[1] |
|
|
161 | && ref[2] == ip[2] |
|
|
162 | #else |
158 | #else |
163 | && *(u16 *)ref == *(u16 *)ip |
159 | && *(u16 *)ref == *(u16 *)ip |
164 | && ref[2] == ip[2] |
|
|
165 | #endif |
160 | #endif |
166 | ) |
161 | ) |
167 | { |
162 | { |
168 | /* match found at *ref++ */ |
163 | /* match found at *ref++ */ |
169 | unsigned int len = 2; |
164 | unsigned int len = 2; |
170 | unsigned int maxlen = in_end - ip - len; |
165 | unsigned int maxlen = in_end - ip - len; |
171 | maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; |
166 | maxlen = maxlen > LZF_MAX_REF ? LZF_MAX_REF : maxlen; |
|
|
167 | |
|
|
168 | if (expect_false (op + 3 + 1 >= out_end)) /* first a faster conservative test */ |
|
|
169 | if (op - !lit + 3 + 1 >= out_end) /* second the exact but rare test */ |
|
|
170 | return 0; |
172 | |
171 | |
173 | op [- lit - 1] = lit - 1; /* stop run */ |
172 | op [- lit - 1] = lit - 1; /* stop run */ |
174 | op -= !lit; /* undo run if length is zero */ |
173 | op -= !lit; /* undo run if length is zero */ |
175 | |
|
|
176 | if (expect_false (op + 3 + 1 >= out_end)) |
|
|
177 | return 0; |
|
|
178 | |
174 | |
179 | for (;;) |
175 | for (;;) |
180 | { |
176 | { |
181 | if (expect_true (maxlen > 16)) |
177 | if (expect_true (maxlen > 16)) |
182 | { |
178 | { |
… | |
… | |
206 | while (len < maxlen && ref[len] == ip[len]); |
202 | while (len < maxlen && ref[len] == ip[len]); |
207 | |
203 | |
208 | break; |
204 | break; |
209 | } |
205 | } |
210 | |
206 | |
211 | len -= 2; |
207 | len -= 2; /* len is now #octets - 1 */ |
212 | ip++; |
208 | ip++; |
213 | |
209 | |
214 | if (len < 7) |
210 | if (len < 7) |
215 | { |
211 | { |
216 | *op++ = (off >> 8) + (len << 5); |
212 | *op++ = (off >> 8) + (len << 5); |
… | |
… | |
221 | *op++ = len - 7; |
217 | *op++ = len - 7; |
222 | } |
218 | } |
223 | |
219 | |
224 | *op++ = off; |
220 | *op++ = off; |
225 | |
221 | |
|
|
222 | lit = 0; op++; /* start run */ |
|
|
223 | |
|
|
224 | ip += len + 1; |
|
|
225 | |
|
|
226 | if (expect_false (ip >= in_end - 2)) |
|
|
227 | break; |
|
|
228 | |
226 | #if ULTRA_FAST || VERY_FAST |
229 | #if ULTRA_FAST || VERY_FAST |
227 | ip += len; |
|
|
228 | #if VERY_FAST && !ULTRA_FAST |
|
|
229 | --ip; |
230 | --ip; |
|
|
231 | # if VERY_FAST && !ULTRA_FAST |
|
|
232 | --ip; |
230 | #endif |
233 | # endif |
231 | hval = FRST (ip); |
234 | hval = FRST (ip); |
232 | |
235 | |
233 | hval = NEXT (hval, ip); |
236 | hval = NEXT (hval, ip); |
234 | htab[IDX (hval)] = ip; |
237 | htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
235 | ip++; |
238 | ip++; |
236 | |
239 | |
237 | #if VERY_FAST && !ULTRA_FAST |
240 | # if VERY_FAST && !ULTRA_FAST |
238 | hval = NEXT (hval, ip); |
241 | hval = NEXT (hval, ip); |
239 | htab[IDX (hval)] = ip; |
242 | htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
240 | ip++; |
243 | ip++; |
241 | #endif |
244 | # endif |
242 | #else |
245 | #else |
|
|
246 | ip -= len + 1; |
|
|
247 | |
243 | do |
248 | do |
244 | { |
249 | { |
245 | hval = NEXT (hval, ip); |
250 | hval = NEXT (hval, ip); |
246 | htab[IDX (hval)] = ip; |
251 | htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
247 | ip++; |
252 | ip++; |
248 | } |
253 | } |
249 | while (len--); |
254 | while (len--); |
250 | #endif |
255 | #endif |
251 | |
|
|
252 | lit = 0; op++; /* start run */ |
|
|
253 | } |
256 | } |
254 | else |
257 | else |
255 | { |
258 | { |
256 | /* one more literal byte we must copy */ |
259 | /* one more literal byte we must copy */ |
257 | if (expect_false (op >= out_end)) |
260 | if (expect_false (op >= out_end)) |
258 | return 0; |
261 | return 0; |
259 | |
262 | |
260 | lit++; *op++ = *ip++; |
263 | lit++; *op++ = *ip++; |
261 | |
264 | |
262 | if (expect_false (lit == MAX_LIT)) |
265 | if (expect_false (lit == LZF_MAX_LIT)) |
263 | { |
266 | { |
264 | op [- lit - 1] = lit - 1; /* stop run */ |
267 | op [- lit - 1] = lit - 1; /* stop run */ |
265 | lit = 0; op++; /* start run */ |
268 | lit = 0; op++; /* start run */ |
266 | } |
269 | } |
267 | } |
270 | } |
… | |
… | |
272 | |
275 | |
273 | while (ip < in_end) |
276 | while (ip < in_end) |
274 | { |
277 | { |
275 | lit++; *op++ = *ip++; |
278 | lit++; *op++ = *ip++; |
276 | |
279 | |
277 | if (expect_false (lit == MAX_LIT)) |
280 | if (expect_false (lit == LZF_MAX_LIT)) |
278 | { |
281 | { |
279 | op [- lit - 1] = lit - 1; /* stop run */ |
282 | op [- lit - 1] = lit - 1; /* stop run */ |
280 | lit = 0; op++; /* start run */ |
283 | lit = 0; op++; /* start run */ |
281 | } |
284 | } |
282 | } |
285 | } |