1 | /* |
1 | /* |
2 | * Copyright (c) 2000-2009 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 |
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] |
… | |
… | |
87 | #define expect_true(expr) expect ((expr) != 0, 1) |
87 | #define expect_true(expr) expect ((expr) != 0, 1) |
88 | |
88 | |
89 | /* |
89 | /* |
90 | * compressed format |
90 | * compressed format |
91 | * |
91 | * |
92 | * 000LLLLL <L+1> ; literal |
92 | * 000LLLLL <L+1> ; literal, L+1=1..33 octets |
93 | * LLLooooo oooooooo ; backref L |
93 | * LLLooooo oooooooo ; backref L+1=1..7 octets, o+1=1..4096 offset |
94 | * 111ooooo LLLLLLLL oooooooo ; backref L+7 |
94 | * 111ooooo LLLLLLLL oooooooo ; backref L+8 octets, o+1=1..4096 offset |
95 | * |
95 | * |
96 | */ |
96 | */ |
97 | |
97 | |
98 | unsigned int |
98 | unsigned int |
99 | lzf_compress (const void *const in_data, unsigned int in_len, |
99 | lzf_compress (const void *const in_data, unsigned int in_len, |
… | |
… | |
104 | ) |
104 | ) |
105 | { |
105 | { |
106 | #if !LZF_STATE_ARG |
106 | #if !LZF_STATE_ARG |
107 | LZF_STATE htab; |
107 | LZF_STATE htab; |
108 | #endif |
108 | #endif |
109 | const u8 **hslot; |
|
|
110 | const u8 *ip = (const u8 *)in_data; |
109 | const u8 *ip = (const u8 *)in_data; |
111 | u8 *op = (u8 *)out_data; |
110 | u8 *op = (u8 *)out_data; |
112 | const u8 *in_end = ip + in_len; |
111 | const u8 *in_end = ip + in_len; |
113 | u8 *out_end = op + out_len; |
112 | u8 *out_end = op + out_len; |
114 | const u8 *ref; |
113 | const u8 *ref; |
115 | |
114 | |
116 | /* off requires a type wide enough to hold a general pointer difference. |
115 | /* 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 |
116 | * 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 |
117 | * 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 |
118 | * 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 |
119 | * and fails to support both assumptions is windows 64 bit, we make a |
121 | * special workaround for it. |
120 | * special workaround for it. |
122 | */ |
121 | */ |
123 | #if defined (WIN32) && defined (_M_X64) |
122 | #if defined (_WIN32) && defined (_M_X64) |
124 | unsigned _int64 off; /* workaround for missing POSIX compliance */ |
123 | /* workaround for missing POSIX compliance */ |
|
|
124 | #if #if __GNUC__ |
|
|
125 | unsigned long long off; |
|
|
126 | #else |
|
|
127 | unsigned __int64 off; |
|
|
128 | #endif |
125 | #else |
129 | #else |
126 | unsigned long off; |
130 | unsigned long off; |
127 | #endif |
131 | #endif |
128 | unsigned int hval; |
132 | unsigned int hval; |
129 | int lit; |
133 | int lit; |
… | |
… | |
131 | if (!in_len || !out_len) |
135 | if (!in_len || !out_len) |
132 | return 0; |
136 | return 0; |
133 | |
137 | |
134 | #if INIT_HTAB |
138 | #if INIT_HTAB |
135 | memset (htab, 0, sizeof (htab)); |
139 | memset (htab, 0, sizeof (htab)); |
136 | # if 0 |
|
|
137 | for (hslot = htab; hslot < htab + HSIZE; hslot++) |
|
|
138 | *hslot++ = ip; |
|
|
139 | # endif |
|
|
140 | #endif |
140 | #endif |
141 | |
141 | |
142 | lit = 0; op++; /* start run */ |
142 | lit = 0; op++; /* start run */ |
143 | |
143 | |
144 | hval = FRST (ip); |
144 | hval = FRST (ip); |
145 | while (ip < in_end - 2) |
145 | while (ip < in_end - 2) |
146 | { |
146 | { |
|
|
147 | LZF_HSLOT *hslot; |
|
|
148 | |
147 | hval = NEXT (hval, ip); |
149 | hval = NEXT (hval, ip); |
148 | hslot = htab + IDX (hval); |
150 | hslot = htab + IDX (hval); |
149 | ref = *hslot; *hslot = ip; |
151 | ref = *hslot + LZF_HSLOT_BIAS; *hslot = ip - LZF_HSLOT_BIAS; |
150 | |
152 | |
151 | if (1 |
153 | if (1 |
152 | #if INIT_HTAB |
154 | #if INIT_HTAB |
153 | && ref < ip /* the next test will actually take care of this, but this is faster */ |
155 | && ref < ip /* the next test will actually take care of this, but this is faster */ |
154 | #endif |
156 | #endif |
155 | && (off = ip - ref - 1) < MAX_OFF |
157 | && (off = ip - ref - 1) < MAX_OFF |
156 | && ip + 4 < in_end |
|
|
157 | && ref > (u8 *)in_data |
158 | && ref > (u8 *)in_data |
|
|
159 | && ref[2] == ip[2] |
158 | #if STRICT_ALIGN |
160 | #if STRICT_ALIGN |
159 | && ref[0] == ip[0] |
161 | && ((ref[1] << 8) | ref[0]) == ((ip[1] << 8) | ip[0]) |
160 | && ref[1] == ip[1] |
|
|
161 | && ref[2] == ip[2] |
|
|
162 | #else |
162 | #else |
163 | && *(u16 *)ref == *(u16 *)ip |
163 | && *(u16 *)ref == *(u16 *)ip |
164 | && ref[2] == ip[2] |
|
|
165 | #endif |
164 | #endif |
166 | ) |
165 | ) |
167 | { |
166 | { |
168 | /* match found at *ref++ */ |
167 | /* match found at *ref++ */ |
169 | unsigned int len = 2; |
168 | unsigned int len = 2; |
… | |
… | |
221 | *op++ = (off >> 8) + ( 7 << 5); |
220 | *op++ = (off >> 8) + ( 7 << 5); |
222 | *op++ = len - 7; |
221 | *op++ = len - 7; |
223 | } |
222 | } |
224 | |
223 | |
225 | *op++ = off; |
224 | *op++ = off; |
|
|
225 | |
226 | lit = 0; op++; /* start run */ |
226 | lit = 0; op++; /* start run */ |
227 | |
227 | |
228 | ip += len + 1; |
228 | ip += len + 1; |
229 | |
229 | |
230 | if (expect_false (ip >= in_end - 2)) |
230 | if (expect_false (ip >= in_end - 2)) |
… | |
… | |
236 | --ip; |
236 | --ip; |
237 | # endif |
237 | # endif |
238 | hval = FRST (ip); |
238 | hval = FRST (ip); |
239 | |
239 | |
240 | hval = NEXT (hval, ip); |
240 | hval = NEXT (hval, ip); |
241 | htab[IDX (hval)] = ip; |
241 | htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
242 | ip++; |
242 | ip++; |
243 | |
243 | |
244 | # if VERY_FAST && !ULTRA_FAST |
244 | # if VERY_FAST && !ULTRA_FAST |
245 | hval = NEXT (hval, ip); |
245 | hval = NEXT (hval, ip); |
246 | htab[IDX (hval)] = ip; |
246 | htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
247 | ip++; |
247 | ip++; |
248 | # endif |
248 | # endif |
249 | #else |
249 | #else |
250 | ip -= len + 1; |
250 | ip -= len + 1; |
251 | |
251 | |
252 | do |
252 | do |
253 | { |
253 | { |
254 | hval = NEXT (hval, ip); |
254 | hval = NEXT (hval, ip); |
255 | htab[IDX (hval)] = ip; |
255 | htab[IDX (hval)] = ip - LZF_HSLOT_BIAS; |
256 | ip++; |
256 | ip++; |
257 | } |
257 | } |
258 | while (len--); |
258 | while (len--); |
259 | #endif |
259 | #endif |
260 | } |
260 | } |