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, |
… | |
… | |
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 |
|
|
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 |
50 | # 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)) |
51 | /*# 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)) |
52 | #endif |
64 | # endif |
53 | /* |
65 | #endif |
54 | * IDX works because it is very similar to a multiplicative hash, e.g. |
|
|
55 | * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1)) |
|
|
56 | * the latter is also quite fast on newer CPUs, and compresses similarly. |
|
|
57 | * |
|
|
58 | * the next one is also quite good, albeit slow ;) |
|
|
59 | * (int)(cos(h & 0xffffff) * 1e6) |
|
|
60 | */ |
|
|
61 | |
66 | |
62 | #if 0 |
67 | #if 0 |
63 | /* original lzv-like hash function, much worse and thus slower */ |
68 | /* original lzv-like hash function, much worse and thus slower */ |
64 | # define FRST(p) (p[0] << 5) ^ p[1] |
69 | # define FRST(p) (p[0] << 5) ^ p[1] |
65 | # define NEXT(v,p) ((v) << 5) ^ p[2] |
70 | # define NEXT(v,p) ((v) << 5) ^ p[2] |
… | |
… | |
68 | |
73 | |
69 | #define MAX_LIT (1 << 5) |
74 | #define MAX_LIT (1 << 5) |
70 | #define MAX_OFF (1 << 13) |
75 | #define MAX_OFF (1 << 13) |
71 | #define MAX_REF ((1 << 8) + (1 << 3)) |
76 | #define MAX_REF ((1 << 8) + (1 << 3)) |
72 | |
77 | |
73 | #if (__i386 || __amd64) && __GNUC__ >= 3 |
|
|
74 | # define lzf_movsb(dst, src, len) \ |
|
|
75 | asm ("rep movsb" \ |
|
|
76 | : "=D" (dst), "=S" (src), "=c" (len) \ |
|
|
77 | : "0" (dst), "1" (src), "2" (len)); |
|
|
78 | #endif |
|
|
79 | |
|
|
80 | #if __GNUC__ >= 3 |
78 | #if __GNUC__ >= 3 |
81 | # define expect(expr,value) __builtin_expect ((expr),(value)) |
79 | # define expect(expr,value) __builtin_expect ((expr),(value)) |
82 | # define inline inline |
80 | # define inline inline |
83 | #else |
81 | #else |
84 | # define expect(expr,value) (expr) |
82 | # define expect(expr,value) (expr) |
… | |
… | |
89 | #define expect_true(expr) expect ((expr) != 0, 1) |
87 | #define expect_true(expr) expect ((expr) != 0, 1) |
90 | |
88 | |
91 | /* |
89 | /* |
92 | * compressed format |
90 | * compressed format |
93 | * |
91 | * |
94 | * 000LLLLL <L+1> ; literal |
92 | * 000LLLLL <L+1> ; literal, L+1=1..33 octets |
95 | * LLLooooo oooooooo ; backref L |
93 | * LLLooooo oooooooo ; backref L+1=1..7 octets, o+1=1..4096 offset |
96 | * 111ooooo LLLLLLLL oooooooo ; backref L+7 |
94 | * 111ooooo LLLLLLLL oooooooo ; backref L+8 octets, o+1=1..4096 offset |
97 | * |
95 | * |
98 | */ |
96 | */ |
99 | |
97 | |
100 | unsigned int |
98 | unsigned int |
101 | lzf_compress (const void *const in_data, unsigned int in_len, |
99 | lzf_compress (const void *const in_data, unsigned int in_len, |
… | |
… | |
106 | ) |
104 | ) |
107 | { |
105 | { |
108 | #if !LZF_STATE_ARG |
106 | #if !LZF_STATE_ARG |
109 | LZF_STATE htab; |
107 | LZF_STATE htab; |
110 | #endif |
108 | #endif |
111 | const u8 **hslot; |
|
|
112 | const u8 *ip = (const u8 *)in_data; |
109 | const u8 *ip = (const u8 *)in_data; |
113 | u8 *op = (u8 *)out_data; |
110 | u8 *op = (u8 *)out_data; |
114 | const u8 *in_end = ip + in_len; |
111 | const u8 *in_end = ip + in_len; |
115 | u8 *out_end = op + out_len; |
112 | u8 *out_end = op + out_len; |
116 | const u8 *ref; |
113 | const u8 *ref; |
117 | |
114 | |
118 | 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 |
119 | unsigned long off; |
130 | unsigned long off; |
120 | int lit = 0; |
131 | #endif |
|
|
132 | unsigned int hval; |
|
|
133 | int lit; |
|
|
134 | |
|
|
135 | if (!in_len || !out_len) |
|
|
136 | return 0; |
121 | |
137 | |
122 | #if INIT_HTAB |
138 | #if INIT_HTAB |
123 | memset (htab, 0, sizeof (htab)); |
139 | memset (htab, 0, sizeof (htab)); |
124 | # if 0 |
140 | #endif |
125 | for (hslot = htab; hslot < htab + HSIZE; hslot++) |
141 | |
126 | *hslot++ = ip; |
142 | lit = 0; op++; /* start run */ |
|
|
143 | |
|
|
144 | hval = FRST (ip); |
|
|
145 | while (ip < in_end - 2) |
|
|
146 | { |
|
|
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 | ) |
|
|
166 | { |
|
|
167 | /* match found at *ref++ */ |
|
|
168 | unsigned int len = 2; |
|
|
169 | unsigned int maxlen = in_end - ip - len; |
|
|
170 | maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; |
|
|
171 | |
|
|
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 */ |
|
|
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 | } |
|
|
203 | |
|
|
204 | do |
|
|
205 | len++; |
|
|
206 | while (len < maxlen && ref[len] == ip[len]); |
|
|
207 | |
|
|
208 | break; |
|
|
209 | } |
|
|
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); |
|
|
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; |
127 | # endif |
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++; |
128 | #endif |
248 | # endif |
|
|
249 | #else |
|
|
250 | ip -= len + 1; |
129 | |
251 | |
130 | for (;;) |
252 | do |
|
|
253 | { |
|
|
254 | hval = NEXT (hval, ip); |
|
|
255 | htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
|
|
256 | ip++; |
|
|
257 | } |
|
|
258 | while (len--); |
|
|
259 | #endif |
|
|
260 | } |
|
|
261 | else |
|
|
262 | { |
|
|
263 | /* one more literal byte we must copy */ |
|
|
264 | if (expect_false (op >= out_end)) |
|
|
265 | return 0; |
|
|
266 | |
|
|
267 | lit++; *op++ = *ip++; |
|
|
268 | |
|
|
269 | if (expect_false (lit == MAX_LIT)) |
|
|
270 | { |
|
|
271 | op [- lit - 1] = lit - 1; /* stop run */ |
|
|
272 | lit = 0; op++; /* start run */ |
|
|
273 | } |
|
|
274 | } |
|
|
275 | } |
|
|
276 | |
|
|
277 | if (op + 3 > out_end) /* at most 3 bytes can be missing here */ |
|
|
278 | return 0; |
|
|
279 | |
|
|
280 | while (ip < in_end) |
131 | { |
281 | { |
132 | if (expect_true (ip < in_end - 2)) |
282 | lit++; *op++ = *ip++; |
133 | { |
|
|
134 | hval = NEXT (hval, ip); |
|
|
135 | hslot = htab + IDX (hval); |
|
|
136 | ref = *hslot; *hslot = ip; |
|
|
137 | |
|
|
138 | if (1 |
|
|
139 | #if INIT_HTAB && !USE_MEMCPY |
|
|
140 | && ref < ip /* the next test will actually take care of this, but this is faster */ |
|
|
141 | #endif |
|
|
142 | && (off = ip - ref - 1) < MAX_OFF |
|
|
143 | && ip + 4 < in_end |
|
|
144 | && ref > (u8 *)in_data |
|
|
145 | #if STRICT_ALIGN |
|
|
146 | && ref[0] == ip[0] |
|
|
147 | && ref[1] == ip[1] |
|
|
148 | && ref[2] == ip[2] |
|
|
149 | #else |
|
|
150 | && *(u16 *)ref == *(u16 *)ip |
|
|
151 | && ref[2] == ip[2] |
|
|
152 | #endif |
|
|
153 | ) |
|
|
154 | { |
|
|
155 | /* match found at *ref++ */ |
|
|
156 | unsigned int len = 2; |
|
|
157 | unsigned int maxlen = in_end - ip - len; |
|
|
158 | maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; |
|
|
159 | |
|
|
160 | if (expect_false (op + lit + 1 + 3 >= out_end)) |
|
|
161 | return 0; |
|
|
162 | |
|
|
163 | if (lit) |
|
|
164 | { |
|
|
165 | *op++ = lit - 1; |
|
|
166 | lit = -lit; |
|
|
167 | do |
|
|
168 | *op++ = ip[lit]; |
|
|
169 | while (++lit); |
|
|
170 | } |
|
|
171 | |
|
|
172 | for (;;) |
|
|
173 | { |
|
|
174 | if (expect_true (ip < in_end - 2 - 8 && maxlen > 8)) |
|
|
175 | { |
|
|
176 | len++; if (ref [len] != ip [len]) break; |
|
|
177 | len++; if (ref [len] != ip [len]) break; |
|
|
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 | len++; if (ref [len] != ip [len]) break; |
|
|
183 | len++; if (ref [len] != ip [len]) break; |
|
|
184 | } |
|
|
185 | |
|
|
186 | do |
|
|
187 | len++; |
|
|
188 | while (len < maxlen && ref[len] == ip[len]); |
|
|
189 | |
|
|
190 | break; |
|
|
191 | } |
|
|
192 | |
|
|
193 | len -= 2; |
|
|
194 | ip++; |
|
|
195 | |
|
|
196 | if (len < 7) |
|
|
197 | { |
|
|
198 | *op++ = (off >> 8) + (len << 5); |
|
|
199 | } |
|
|
200 | else |
|
|
201 | { |
|
|
202 | *op++ = (off >> 8) + ( 7 << 5); |
|
|
203 | *op++ = len - 7; |
|
|
204 | } |
|
|
205 | |
|
|
206 | *op++ = off; |
|
|
207 | |
|
|
208 | #if ULTRA_FAST || VERY_FAST |
|
|
209 | ip += len; |
|
|
210 | #if VERY_FAST && !ULTRA_FAST |
|
|
211 | --ip; |
|
|
212 | #endif |
|
|
213 | hval = FRST (ip); |
|
|
214 | |
|
|
215 | hval = NEXT (hval, ip); |
|
|
216 | htab[IDX (hval)] = ip; |
|
|
217 | ip++; |
|
|
218 | |
|
|
219 | #if VERY_FAST && !ULTRA_FAST |
|
|
220 | hval = NEXT (hval, ip); |
|
|
221 | htab[IDX (hval)] = ip; |
|
|
222 | ip++; |
|
|
223 | #endif |
|
|
224 | #else |
|
|
225 | do |
|
|
226 | { |
|
|
227 | hval = NEXT (hval, ip); |
|
|
228 | htab[IDX (hval)] = ip; |
|
|
229 | ip++; |
|
|
230 | } |
|
|
231 | while (len--); |
|
|
232 | #endif |
|
|
233 | continue; |
|
|
234 | } |
|
|
235 | } |
|
|
236 | else if (expect_false (ip == in_end)) |
|
|
237 | break; |
|
|
238 | |
|
|
239 | /* one more literal byte we must copy */ |
|
|
240 | lit++; |
|
|
241 | ip++; |
|
|
242 | |
283 | |
243 | if (expect_false (lit == MAX_LIT)) |
284 | if (expect_false (lit == MAX_LIT)) |
244 | { |
285 | { |
245 | if (op + 1 + MAX_LIT >= out_end) |
286 | op [- lit - 1] = lit - 1; /* stop run */ |
246 | return 0; |
287 | lit = 0; op++; /* start run */ |
247 | |
|
|
248 | *op++ = MAX_LIT - 1; |
|
|
249 | |
|
|
250 | #ifdef lzf_movsb |
|
|
251 | ip -= lit; |
|
|
252 | lzf_movsb (op, ip, lit); |
|
|
253 | #else |
|
|
254 | lit = -lit; |
|
|
255 | do |
|
|
256 | *op++ = ip[lit]; |
|
|
257 | while (++lit); |
|
|
258 | #endif |
|
|
259 | } |
288 | } |
260 | } |
289 | } |
261 | |
290 | |
262 | if (lit) |
291 | op [- lit - 1] = lit - 1; /* end run */ |
263 | { |
292 | op -= !lit; /* undo run if length is zero */ |
264 | if (op + lit + 1 >= out_end) |
|
|
265 | return 0; |
|
|
266 | |
293 | |
267 | *op++ = lit - 1; |
|
|
268 | #ifdef lzf_movsb |
|
|
269 | ip -= lit; |
|
|
270 | lzf_movsb (op, ip, lit); |
|
|
271 | #else |
|
|
272 | lit = -lit; |
|
|
273 | do |
|
|
274 | *op++ = ip[lit]; |
|
|
275 | while (++lit); |
|
|
276 | #endif |
|
|
277 | } |
|
|
278 | |
|
|
279 | return op - (u8 *) out_data; |
294 | return op - (u8 *)out_data; |
280 | } |
295 | } |
281 | |
296 | |