1 | /* |
1 | /* |
2 | * Copyright (c) 2000-2007 Marc Alexander Lehmann <schmorp@schmorp.de> |
2 | * Copyright (c) 2000-2008 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, |
… | |
… | |
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 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 |
50 | # 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)) |
51 | /*# define IDX(h) ((ip[0] * 121 ^ ip[1] * 33 ^ ip[2] * 1) & (HSIZE-1))*/ |
56 | # endif |
52 | #endif |
57 | #endif |
53 | /* |
58 | /* |
54 | * 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. |
55 | * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1)) |
60 | * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1)) |
56 | * 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. |
… | |
… | |
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 |
78 | #if __GNUC__ >= 3 |
74 | # define lzf_movsb(dst, src, len) \ |
79 | # define expect(expr,value) __builtin_expect ((expr),(value)) |
75 | asm ("rep movsb" \ |
80 | # define inline inline |
76 | : "=D" (dst), "=S" (src), "=c" (len) \ |
81 | #else |
77 | : "0" (dst), "1" (src), "2" (len)); |
82 | # define expect(expr,value) (expr) |
|
|
83 | # define inline static |
78 | #endif |
84 | #endif |
|
|
85 | |
|
|
86 | #define expect_false(expr) expect ((expr) != 0, 0) |
|
|
87 | #define expect_true(expr) expect ((expr) != 0, 1) |
79 | |
88 | |
80 | /* |
89 | /* |
81 | * compressed format |
90 | * compressed format |
82 | * |
91 | * |
83 | * 000LLLLL <L+1> ; literal |
92 | * 000LLLLL <L+1> ; literal |
… | |
… | |
102 | u8 *op = (u8 *)out_data; |
111 | u8 *op = (u8 *)out_data; |
103 | const u8 *in_end = ip + in_len; |
112 | const u8 *in_end = ip + in_len; |
104 | u8 *out_end = op + out_len; |
113 | u8 *out_end = op + out_len; |
105 | const u8 *ref; |
114 | const u8 *ref; |
106 | |
115 | |
107 | 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 |
108 | unsigned long off; |
126 | unsigned long off; |
109 | int lit = 0; |
127 | #endif |
|
|
128 | unsigned int hval; |
|
|
129 | int lit; |
|
|
130 | |
|
|
131 | if (!in_len || !out_len) |
|
|
132 | return 0; |
110 | |
133 | |
111 | #if INIT_HTAB |
134 | #if INIT_HTAB |
112 | memset (htab, 0, sizeof (htab)); |
135 | memset (htab, 0, sizeof (htab)); |
113 | # if 0 |
136 | # if 0 |
114 | for (hslot = htab; hslot < htab + HSIZE; hslot++) |
137 | for (hslot = htab; hslot < htab + HSIZE; hslot++) |
115 | *hslot++ = ip; |
138 | *hslot++ = ip; |
116 | # endif |
139 | # endif |
117 | #endif |
140 | #endif |
118 | |
141 | |
119 | for (;;) |
142 | lit = 0; op++; /* start run */ |
|
|
143 | |
|
|
144 | hval = FRST (ip); |
|
|
145 | while (ip < in_end - 2) |
120 | { |
146 | { |
121 | 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 | #if STRICT_ALIGN |
|
|
159 | && ref[0] == ip[0] |
|
|
160 | && ref[1] == ip[1] |
|
|
161 | && ref[2] == ip[2] |
|
|
162 | #else |
|
|
163 | && *(u16 *)ref == *(u16 *)ip |
|
|
164 | && ref[2] == ip[2] |
|
|
165 | #endif |
|
|
166 | ) |
122 | { |
167 | { |
123 | hval = NEXT (hval, ip); |
|
|
124 | hslot = htab + IDX (hval); |
|
|
125 | ref = *hslot; *hslot = ip; |
|
|
126 | |
|
|
127 | if (1 |
|
|
128 | #if INIT_HTAB && !USE_MEMCPY |
|
|
129 | && ref < ip /* the next test will actually take care of this, but this is faster */ |
|
|
130 | #endif |
|
|
131 | && (off = ip - ref - 1) < MAX_OFF |
|
|
132 | && ip + 4 < in_end |
|
|
133 | && ref > (u8 *)in_data |
|
|
134 | #if STRICT_ALIGN |
|
|
135 | && ref[0] == ip[0] |
|
|
136 | && ref[1] == ip[1] |
|
|
137 | && ref[2] == ip[2] |
|
|
138 | #else |
|
|
139 | && *(u16 *)ref == *(u16 *)ip |
|
|
140 | && ref[2] == ip[2] |
|
|
141 | #endif |
|
|
142 | ) |
|
|
143 | { |
|
|
144 | /* match found at *ref++ */ |
168 | /* match found at *ref++ */ |
145 | unsigned int len = 2; |
169 | unsigned int len = 2; |
146 | unsigned int maxlen = in_end - ip - len; |
170 | unsigned int maxlen = in_end - ip - len; |
147 | maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; |
171 | maxlen = maxlen > MAX_REF ? MAX_REF : maxlen; |
148 | |
172 | |
149 | if (op + lit + 1 + 3 >= out_end) |
173 | op [- lit - 1] = lit - 1; /* stop run */ |
|
|
174 | op -= !lit; /* undo run if length is zero */ |
|
|
175 | |
|
|
176 | if (expect_false (op + 3 + 1 >= out_end)) |
150 | return 0; |
177 | return 0; |
|
|
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 | } |
151 | |
203 | |
152 | do |
204 | do |
153 | len++; |
205 | len++; |
154 | while (len < maxlen && ref[len] == ip[len]); |
206 | while (len < maxlen && ref[len] == ip[len]); |
155 | |
207 | |
156 | if (lit) |
|
|
157 | { |
208 | break; |
158 | *op++ = lit - 1; |
|
|
159 | lit = -lit; |
|
|
160 | do |
|
|
161 | *op++ = ip[lit]; |
|
|
162 | while (++lit); |
|
|
163 | } |
209 | } |
164 | |
210 | |
165 | len -= 2; |
211 | len -= 2; /* len is now #octets - 1 */ |
166 | ip++; |
212 | ip++; |
167 | |
213 | |
168 | if (len < 7) |
214 | if (len < 7) |
169 | { |
215 | { |
170 | *op++ = (off >> 8) + (len << 5); |
216 | *op++ = (off >> 8) + (len << 5); |
171 | } |
217 | } |
172 | else |
218 | else |
173 | { |
219 | { |
174 | *op++ = (off >> 8) + ( 7 << 5); |
220 | *op++ = (off >> 8) + ( 7 << 5); |
175 | *op++ = len - 7; |
221 | *op++ = len - 7; |
176 | } |
222 | } |
177 | |
223 | |
178 | *op++ = off; |
224 | *op++ = off; |
|
|
225 | lit = 0; op++; /* start run */ |
|
|
226 | |
|
|
227 | ip += len + 1; |
|
|
228 | |
|
|
229 | if (expect_false (ip > in_end - 2)) |
|
|
230 | break; |
179 | |
231 | |
180 | #if ULTRA_FAST || VERY_FAST |
232 | #if ULTRA_FAST || VERY_FAST |
181 | ip += len; |
233 | --ip; |
182 | #if VERY_FAST && !ULTRA_FAST |
234 | # if VERY_FAST && !ULTRA_FAST |
183 | --ip; |
235 | --ip; |
184 | #endif |
236 | # endif |
185 | hval = FRST (ip); |
237 | hval = FRST (ip); |
186 | |
238 | |
|
|
239 | hval = NEXT (hval, ip); |
|
|
240 | htab[IDX (hval)] = ip; |
|
|
241 | ip++; |
|
|
242 | |
|
|
243 | # if VERY_FAST && !ULTRA_FAST |
|
|
244 | hval = NEXT (hval, ip); |
|
|
245 | htab[IDX (hval)] = ip; |
|
|
246 | ip++; |
|
|
247 | # endif |
|
|
248 | #else |
|
|
249 | ip -= len + 1; |
|
|
250 | |
|
|
251 | do |
|
|
252 | { |
187 | hval = NEXT (hval, ip); |
253 | hval = NEXT (hval, ip); |
188 | htab[IDX (hval)] = ip; |
254 | htab[IDX (hval)] = ip; |
189 | ip++; |
255 | ip++; |
190 | |
|
|
191 | #if VERY_FAST && !ULTRA_FAST |
|
|
192 | hval = NEXT (hval, ip); |
|
|
193 | htab[IDX (hval)] = ip; |
|
|
194 | ip++; |
|
|
195 | #endif |
|
|
196 | #else |
|
|
197 | do |
|
|
198 | { |
|
|
199 | hval = NEXT (hval, ip); |
|
|
200 | htab[IDX (hval)] = ip; |
|
|
201 | ip++; |
|
|
202 | } |
256 | } |
203 | while (len--); |
257 | while (len--); |
204 | #endif |
258 | #endif |
205 | continue; |
|
|
206 | } |
|
|
207 | } |
259 | } |
208 | else if (ip == in_end) |
260 | else |
209 | break; |
|
|
210 | |
|
|
211 | /* one more literal byte we must copy */ |
|
|
212 | lit++; |
|
|
213 | ip++; |
|
|
214 | |
|
|
215 | if (lit == MAX_LIT) |
|
|
216 | { |
261 | { |
217 | if (op + 1 + MAX_LIT >= out_end) |
262 | /* one more literal byte we must copy */ |
|
|
263 | if (expect_false (op >= out_end)) |
218 | return 0; |
264 | return 0; |
219 | |
265 | |
220 | *op++ = MAX_LIT - 1; |
266 | lit++; *op++ = *ip++; |
221 | |
267 | |
222 | #ifdef lzf_movsb |
268 | if (expect_false (lit == MAX_LIT)) |
223 | ip -= lit; |
|
|
224 | lzf_movsb (op, ip, lit); |
|
|
225 | #else |
|
|
226 | lit = -lit; |
|
|
227 | do |
269 | { |
228 | *op++ = ip[lit]; |
270 | op [- lit - 1] = lit - 1; /* stop run */ |
229 | while (++lit); |
271 | lit = 0; op++; /* start run */ |
230 | #endif |
272 | } |
231 | } |
273 | } |
232 | } |
274 | } |
233 | |
275 | |
234 | if (lit) |
276 | if (op + 3 > out_end) /* at most 3 bytes can be missing here */ |
|
|
277 | return 0; |
|
|
278 | |
|
|
279 | while (ip < in_end) |
235 | { |
280 | { |
236 | if (op + lit + 1 >= out_end) |
281 | lit++; *op++ = *ip++; |
237 | return 0; |
|
|
238 | |
282 | |
239 | *op++ = lit - 1; |
283 | if (expect_false (lit == MAX_LIT)) |
240 | #ifdef lzf_movsb |
284 | { |
241 | ip -= lit; |
285 | op [- lit - 1] = lit - 1; /* stop run */ |
242 | lzf_movsb (op, ip, lit); |
286 | lit = 0; op++; /* start run */ |
243 | #else |
287 | } |
244 | lit = -lit; |
|
|
245 | do |
|
|
246 | *op++ = ip[lit]; |
|
|
247 | while (++lit); |
|
|
248 | #endif |
|
|
249 | } |
288 | } |
250 | |
289 | |
|
|
290 | op [- lit - 1] = lit - 1; /* end run */ |
|
|
291 | op -= !lit; /* undo run if length is zero */ |
|
|
292 | |
251 | return op - (u8 *) out_data; |
293 | return op - (u8 *)out_data; |
252 | } |
294 | } |
253 | |
295 | |