lzfse 0.0.1.pre.4
Sign up to get free protection for your applications and to get access to all the features.
- checksums.yaml +7 -0
- data/.yardopts +1 -0
- data/LICENSE +21 -0
- data/README.md +38 -0
- data/ext/lzfse/LICENSE +27 -0
- data/ext/lzfse/ext.c +194 -0
- data/ext/lzfse/ext.h +14 -0
- data/ext/lzfse/extconf.rb +10 -0
- data/ext/lzfse/lzfse.h +136 -0
- data/ext/lzfse/lzfse_decode.c +72 -0
- data/ext/lzfse/lzfse_decode_base.c +630 -0
- data/ext/lzfse/lzfse_encode.c +163 -0
- data/ext/lzfse/lzfse_encode_base.c +830 -0
- data/ext/lzfse/lzfse_encode_tables.h +218 -0
- data/ext/lzfse/lzfse_fse.c +216 -0
- data/ext/lzfse/lzfse_fse.h +631 -0
- data/ext/lzfse/lzfse_internal.h +616 -0
- data/ext/lzfse/lzfse_tunables.h +60 -0
- data/ext/lzfse/lzvn_decode_base.c +711 -0
- data/ext/lzfse/lzvn_decode_base.h +68 -0
- data/ext/lzfse/lzvn_encode_base.c +593 -0
- data/ext/lzfse/lzvn_encode_base.h +116 -0
- data/lib/lzfse.rb +25 -0
- data/lib/lzfse/exceptions.rb +12 -0
- data/lib/lzfse/version.rb +5 -0
- data/vendor/bundle/ruby/2.7.0/gems/rainbow-3.0.0/LICENSE +20 -0
- data/vendor/bundle/ruby/2.7.0/gems/regexp_parser-2.1.1/LICENSE +22 -0
- data/vendor/bundle/ruby/2.7.0/gems/yard-0.9.26/LICENSE +22 -0
- metadata +126 -0
@@ -0,0 +1,68 @@
|
|
1
|
+
/*
|
2
|
+
Copyright (c) 2015-2016, Apple Inc. All rights reserved.
|
3
|
+
|
4
|
+
Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
|
5
|
+
|
6
|
+
1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
|
7
|
+
|
8
|
+
2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer
|
9
|
+
in the documentation and/or other materials provided with the distribution.
|
10
|
+
|
11
|
+
3. Neither the name of the copyright holder(s) nor the names of any contributors may be used to endorse or promote products derived
|
12
|
+
from this software without specific prior written permission.
|
13
|
+
|
14
|
+
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
15
|
+
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
|
16
|
+
COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
|
17
|
+
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
18
|
+
HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
|
19
|
+
ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
20
|
+
*/
|
21
|
+
|
22
|
+
// LZVN low-level decoder (v2)
|
23
|
+
// Functions in the low-level API should switch to these at some point.
|
24
|
+
// Apr 2014
|
25
|
+
|
26
|
+
#ifndef LZVN_DECODE_BASE_H
|
27
|
+
#define LZVN_DECODE_BASE_H
|
28
|
+
|
29
|
+
#include "lzfse_internal.h"
|
30
|
+
|
31
|
+
/*! @abstract Base decoder state. */
|
32
|
+
typedef struct {
|
33
|
+
|
34
|
+
// Decoder I/O
|
35
|
+
|
36
|
+
// Next byte to read in source buffer
|
37
|
+
const unsigned char *src;
|
38
|
+
// Next byte after source buffer
|
39
|
+
const unsigned char *src_end;
|
40
|
+
|
41
|
+
// Next byte to write in destination buffer (by decoder)
|
42
|
+
unsigned char *dst;
|
43
|
+
// Valid range for destination buffer is [dst_begin, dst_end - 1]
|
44
|
+
unsigned char *dst_begin;
|
45
|
+
unsigned char *dst_end;
|
46
|
+
// Next byte to read in destination buffer (modified by caller)
|
47
|
+
unsigned char *dst_current;
|
48
|
+
|
49
|
+
// Decoder state
|
50
|
+
|
51
|
+
// Partially expanded match, or 0,0,0.
|
52
|
+
// In that case, src points to the next literal to copy, or the next op-code
|
53
|
+
// if L==0.
|
54
|
+
size_t L, M, D;
|
55
|
+
|
56
|
+
// Distance for last emitted match, or 0
|
57
|
+
lzvn_offset d_prev;
|
58
|
+
|
59
|
+
// Did we decode end-of-stream?
|
60
|
+
int end_of_stream;
|
61
|
+
|
62
|
+
} lzvn_decoder_state;
|
63
|
+
|
64
|
+
/*! @abstract Decode source to destination.
|
65
|
+
* Updates \p state (src,dst,d_prev). */
|
66
|
+
void lzvn_decode(lzvn_decoder_state *state);
|
67
|
+
|
68
|
+
#endif // LZVN_DECODE_BASE_H
|
@@ -0,0 +1,593 @@
|
|
1
|
+
/*
|
2
|
+
Copyright (c) 2015-2016, Apple Inc. All rights reserved.
|
3
|
+
|
4
|
+
Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
|
5
|
+
|
6
|
+
1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
|
7
|
+
|
8
|
+
2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer
|
9
|
+
in the documentation and/or other materials provided with the distribution.
|
10
|
+
|
11
|
+
3. Neither the name of the copyright holder(s) nor the names of any contributors may be used to endorse or promote products derived
|
12
|
+
from this software without specific prior written permission.
|
13
|
+
|
14
|
+
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
15
|
+
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
|
16
|
+
COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
|
17
|
+
(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
18
|
+
HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
|
19
|
+
ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
20
|
+
*/
|
21
|
+
|
22
|
+
// LZVN low-level encoder
|
23
|
+
|
24
|
+
#include "lzvn_encode_base.h"
|
25
|
+
|
26
|
+
#if defined(_MSC_VER) && !defined(__clang__)
|
27
|
+
# define restrict __restrict
|
28
|
+
#endif
|
29
|
+
|
30
|
+
// ===============================================================
|
31
|
+
// Coarse/fine copy, non overlapping buffers
|
32
|
+
|
33
|
+
/*! @abstract Copy at least \p nbytes bytes from \p src to \p dst, by blocks
|
34
|
+
* of 8 bytes (may go beyond range). No overlap.
|
35
|
+
* @return \p dst + \p nbytes. */
|
36
|
+
static inline unsigned char *lzvn_copy64(unsigned char *restrict dst,
|
37
|
+
const unsigned char *restrict src,
|
38
|
+
size_t nbytes) {
|
39
|
+
for (size_t i = 0; i < nbytes; i += 8)
|
40
|
+
store8(dst + i, load8(src + i));
|
41
|
+
return dst + nbytes;
|
42
|
+
}
|
43
|
+
|
44
|
+
/*! @abstract Copy exactly \p nbytes bytes from \p src to \p dst (respects range).
|
45
|
+
* No overlap.
|
46
|
+
* @return \p dst + \p nbytes. */
|
47
|
+
static inline unsigned char *lzvn_copy8(unsigned char *restrict dst,
|
48
|
+
const unsigned char *restrict src,
|
49
|
+
size_t nbytes) {
|
50
|
+
for (size_t i = 0; i < nbytes; i++)
|
51
|
+
dst[i] = src[i];
|
52
|
+
return dst + nbytes;
|
53
|
+
}
|
54
|
+
|
55
|
+
/*! @abstract Emit (L,0,0) instructions (final literal).
|
56
|
+
* We read at most \p L bytes from \p p.
|
57
|
+
* @param p input stream
|
58
|
+
* @param q1 the first byte after the output buffer.
|
59
|
+
* @return pointer to the next output, <= \p q1.
|
60
|
+
* @return \p q1 if output is full. In that case, output will be partially invalid.
|
61
|
+
*/
|
62
|
+
static inline unsigned char *emit_literal(const unsigned char *p,
|
63
|
+
unsigned char *q, unsigned char *q1,
|
64
|
+
size_t L) {
|
65
|
+
size_t x;
|
66
|
+
while (L > 15) {
|
67
|
+
x = L < 271 ? L : 271;
|
68
|
+
if (q + x + 10 >= q1)
|
69
|
+
goto OUT_FULL;
|
70
|
+
store2(q, 0xE0 + ((x - 16) << 8));
|
71
|
+
q += 2;
|
72
|
+
L -= x;
|
73
|
+
q = lzvn_copy8(q, p, x);
|
74
|
+
p += x;
|
75
|
+
}
|
76
|
+
if (L > 0) {
|
77
|
+
if (q + L + 10 >= q1)
|
78
|
+
goto OUT_FULL;
|
79
|
+
*q++ = 0xE0 + L; // 1110LLLL
|
80
|
+
q = lzvn_copy8(q, p, L);
|
81
|
+
}
|
82
|
+
return q;
|
83
|
+
|
84
|
+
OUT_FULL:
|
85
|
+
return q1;
|
86
|
+
}
|
87
|
+
|
88
|
+
/*! @abstract Emit (L,M,D) instructions. M>=3.
|
89
|
+
* @param p input stream pointing to the beginning of the literal. We read at
|
90
|
+
* most \p L+4 bytes from \p p.
|
91
|
+
* @param q1 the first byte after the output buffer.
|
92
|
+
* @return pointer to the next output, <= \p q1.
|
93
|
+
* @return \p q1 if output is full. In that case, output will be partially invalid.
|
94
|
+
*/
|
95
|
+
static inline unsigned char *emit(const unsigned char *p, unsigned char *q,
|
96
|
+
unsigned char *q1, size_t L, size_t M,
|
97
|
+
size_t D, size_t D_prev) {
|
98
|
+
size_t x;
|
99
|
+
while (L > 15) {
|
100
|
+
x = L < 271 ? L : 271;
|
101
|
+
if (q + x + 10 >= q1)
|
102
|
+
goto OUT_FULL;
|
103
|
+
store2(q, 0xE0 + ((x - 16) << 8));
|
104
|
+
q += 2;
|
105
|
+
L -= x;
|
106
|
+
q = lzvn_copy64(q, p, x);
|
107
|
+
p += x;
|
108
|
+
}
|
109
|
+
if (L > 3) {
|
110
|
+
if (q + L + 10 >= q1)
|
111
|
+
goto OUT_FULL;
|
112
|
+
*q++ = 0xE0 + L; // 1110LLLL
|
113
|
+
q = lzvn_copy64(q, p, L);
|
114
|
+
p += L;
|
115
|
+
L = 0;
|
116
|
+
}
|
117
|
+
x = M <= 10 - 2 * L ? M : 10 - 2 * L; // x = min(10-2*L,M)
|
118
|
+
M -= x;
|
119
|
+
x -= 3; // M = (x+3) + M' max value for x is 7-2*L
|
120
|
+
|
121
|
+
// Here L<4 literals remaining, we read them here
|
122
|
+
uint32_t literal = load4(p);
|
123
|
+
// P is not accessed after this point
|
124
|
+
|
125
|
+
// Relaxed capacity test covering all cases
|
126
|
+
if (q + 8 >= q1)
|
127
|
+
goto OUT_FULL;
|
128
|
+
|
129
|
+
if (D == D_prev) {
|
130
|
+
if (L == 0) {
|
131
|
+
*q++ = 0xF0 + (x + 3); // XM!
|
132
|
+
} else {
|
133
|
+
*q++ = (L << 6) + (x << 3) + 6; // LLxxx110
|
134
|
+
}
|
135
|
+
store4(q, literal);
|
136
|
+
q += L;
|
137
|
+
} else if (D < 2048 - 2 * 256) {
|
138
|
+
// Short dist D>>8 in 0..5
|
139
|
+
*q++ = (D >> 8) + (L << 6) + (x << 3); // LLxxxDDD
|
140
|
+
*q++ = D & 0xFF;
|
141
|
+
store4(q, literal);
|
142
|
+
q += L;
|
143
|
+
} else if (D >= (1 << 14) || M == 0 || (x + 3) + M > 34) {
|
144
|
+
// Long dist
|
145
|
+
*q++ = (L << 6) + (x << 3) + 7;
|
146
|
+
store2(q, D);
|
147
|
+
q += 2;
|
148
|
+
store4(q, literal);
|
149
|
+
q += L;
|
150
|
+
} else {
|
151
|
+
// Medium distance
|
152
|
+
x += M;
|
153
|
+
M = 0;
|
154
|
+
*q++ = 0xA0 + (x >> 2) + (L << 3);
|
155
|
+
store2(q, D << 2 | (x & 3));
|
156
|
+
q += 2;
|
157
|
+
store4(q, literal);
|
158
|
+
q += L;
|
159
|
+
}
|
160
|
+
|
161
|
+
// Issue remaining match
|
162
|
+
while (M > 15) {
|
163
|
+
if (q + 2 >= q1)
|
164
|
+
goto OUT_FULL;
|
165
|
+
x = M < 271 ? M : 271;
|
166
|
+
store2(q, 0xf0 + ((x - 16) << 8));
|
167
|
+
q += 2;
|
168
|
+
M -= x;
|
169
|
+
}
|
170
|
+
if (M > 0) {
|
171
|
+
if (q + 1 >= q1)
|
172
|
+
goto OUT_FULL;
|
173
|
+
*q++ = 0xF0 + M; // M = 0..15
|
174
|
+
}
|
175
|
+
|
176
|
+
return q;
|
177
|
+
|
178
|
+
OUT_FULL:
|
179
|
+
return q1;
|
180
|
+
}
|
181
|
+
|
182
|
+
// ===============================================================
|
183
|
+
// Conversions
|
184
|
+
|
185
|
+
/*! @abstract Return 32-bit value to store for offset x. */
|
186
|
+
static inline int32_t offset_to_s32(lzvn_offset x) { return (int32_t)x; }
|
187
|
+
|
188
|
+
/*! @abstract Get offset from 32-bit stored value x. */
|
189
|
+
static inline lzvn_offset offset_from_s32(int32_t x) { return (lzvn_offset)x; }
|
190
|
+
|
191
|
+
// ===============================================================
|
192
|
+
// Hash and Matching
|
193
|
+
|
194
|
+
/*! @abstract Get hash in range \c [0,LZVN_ENCODE_HASH_VALUES-1] from 3 bytes in i. */
|
195
|
+
static inline uint32_t hash3i(uint32_t i) {
|
196
|
+
i &= 0xffffff; // truncate to 24-bit input (slightly increases compression ratio)
|
197
|
+
uint32_t h = (i * (1 + (1 << 6) + (1 << 12))) >> 12;
|
198
|
+
return h & (LZVN_ENCODE_HASH_VALUES - 1);
|
199
|
+
}
|
200
|
+
|
201
|
+
/*! @abstract Return the number [0, 4] of zero bytes in \p x, starting from the
|
202
|
+
* least significant byte. */
|
203
|
+
static inline lzvn_offset trailing_zero_bytes(uint32_t x) {
|
204
|
+
return (x == 0) ? 4 : (__builtin_ctzl(x) >> 3);
|
205
|
+
}
|
206
|
+
|
207
|
+
/*! @abstract Return the number [0, 4] of matching chars between values at
|
208
|
+
* \p src+i and \p src+j, starting from the least significant byte.
|
209
|
+
* Assumes we can read 4 chars from each position. */
|
210
|
+
static inline lzvn_offset nmatch4(const unsigned char *src, lzvn_offset i,
|
211
|
+
lzvn_offset j) {
|
212
|
+
uint32_t vi = load4(src + i);
|
213
|
+
uint32_t vj = load4(src + j);
|
214
|
+
return trailing_zero_bytes(vi ^ vj);
|
215
|
+
}
|
216
|
+
|
217
|
+
/*! @abstract Check if l_begin, m_begin, m0_begin (m0_begin < m_begin) can be
|
218
|
+
* expanded to a match of length at least 3.
|
219
|
+
* @param m_begin new string to match.
|
220
|
+
* @param m0_begin candidate old string.
|
221
|
+
* @param src source buffer, with valid indices src_begin <= i < src_end.
|
222
|
+
* (src_begin may be <0)
|
223
|
+
* @return If a match can be found, return 1 and set all \p match fields,
|
224
|
+
* otherwise return 0.
|
225
|
+
* @note \p *match should be 0 before the call. */
|
226
|
+
static inline int lzvn_find_match(const unsigned char *src,
|
227
|
+
lzvn_offset src_begin,
|
228
|
+
lzvn_offset src_end, lzvn_offset l_begin,
|
229
|
+
lzvn_offset m0_begin, lzvn_offset m_begin,
|
230
|
+
lzvn_match_info *match) {
|
231
|
+
lzvn_offset n = nmatch4(src, m_begin, m0_begin);
|
232
|
+
if (n < 3)
|
233
|
+
return 0; // no match
|
234
|
+
|
235
|
+
lzvn_offset D = m_begin - m0_begin; // actual distance
|
236
|
+
if (D <= 0 || D > LZVN_ENCODE_MAX_DISTANCE)
|
237
|
+
return 0; // distance out of range
|
238
|
+
|
239
|
+
// Expand forward
|
240
|
+
lzvn_offset m_end = m_begin + n;
|
241
|
+
while (n == 4 && m_end + 4 < src_end) {
|
242
|
+
n = nmatch4(src, m_end, m_end - D);
|
243
|
+
m_end += n;
|
244
|
+
}
|
245
|
+
|
246
|
+
// Expand backwards over literal
|
247
|
+
while (m0_begin > src_begin && m_begin > l_begin &&
|
248
|
+
src[m_begin - 1] == src[m0_begin - 1]) {
|
249
|
+
m0_begin--;
|
250
|
+
m_begin--;
|
251
|
+
}
|
252
|
+
|
253
|
+
// OK, we keep it, update MATCH
|
254
|
+
lzvn_offset M = m_end - m_begin; // match length
|
255
|
+
match->m_begin = m_begin;
|
256
|
+
match->m_end = m_end;
|
257
|
+
match->K = M - ((D < 0x600) ? 2 : 3);
|
258
|
+
match->M = M;
|
259
|
+
match->D = D;
|
260
|
+
|
261
|
+
return 1; // OK
|
262
|
+
}
|
263
|
+
|
264
|
+
/*! @abstract Same as lzvn_find_match, but we already know that N bytes do
|
265
|
+
* match (N<=4). */
|
266
|
+
static inline int lzvn_find_matchN(const unsigned char *src,
|
267
|
+
lzvn_offset src_begin,
|
268
|
+
lzvn_offset src_end, lzvn_offset l_begin,
|
269
|
+
lzvn_offset m0_begin, lzvn_offset m_begin,
|
270
|
+
lzvn_offset n, lzvn_match_info *match) {
|
271
|
+
// We can skip the first comparison on 4 bytes
|
272
|
+
if (n < 3)
|
273
|
+
return 0; // no match
|
274
|
+
|
275
|
+
lzvn_offset D = m_begin - m0_begin; // actual distance
|
276
|
+
if (D <= 0 || D > LZVN_ENCODE_MAX_DISTANCE)
|
277
|
+
return 0; // distance out of range
|
278
|
+
|
279
|
+
// Expand forward
|
280
|
+
lzvn_offset m_end = m_begin + n;
|
281
|
+
while (n == 4 && m_end + 4 < src_end) {
|
282
|
+
n = nmatch4(src, m_end, m_end - D);
|
283
|
+
m_end += n;
|
284
|
+
}
|
285
|
+
|
286
|
+
// Expand backwards over literal
|
287
|
+
while (m0_begin > src_begin && m_begin > l_begin &&
|
288
|
+
src[m_begin - 1] == src[m0_begin - 1]) {
|
289
|
+
m0_begin--;
|
290
|
+
m_begin--;
|
291
|
+
}
|
292
|
+
|
293
|
+
// OK, we keep it, update MATCH
|
294
|
+
lzvn_offset M = m_end - m_begin; // match length
|
295
|
+
match->m_begin = m_begin;
|
296
|
+
match->m_end = m_end;
|
297
|
+
match->K = M - ((D < 0x600) ? 2 : 3);
|
298
|
+
match->M = M;
|
299
|
+
match->D = D;
|
300
|
+
|
301
|
+
return 1; // OK
|
302
|
+
}
|
303
|
+
|
304
|
+
// ===============================================================
|
305
|
+
// Encoder Backend
|
306
|
+
|
307
|
+
/*! @abstract Emit a match and update state.
|
308
|
+
* @return number of bytes written to \p dst. May be 0 if there is no more space
|
309
|
+
* in \p dst to emit the match. */
|
310
|
+
static inline lzvn_offset lzvn_emit_match(lzvn_encoder_state *state,
|
311
|
+
lzvn_match_info match) {
|
312
|
+
size_t L = (size_t)(match.m_begin - state->src_literal); // literal count
|
313
|
+
size_t M = (size_t)match.M; // match length
|
314
|
+
size_t D = (size_t)match.D; // match distance
|
315
|
+
size_t D_prev = (size_t)state->d_prev; // previously emitted match distance
|
316
|
+
unsigned char *dst = emit(state->src + state->src_literal, state->dst,
|
317
|
+
state->dst_end, L, M, D, D_prev);
|
318
|
+
// Check if DST is full
|
319
|
+
if (dst >= state->dst_end) {
|
320
|
+
return 0; // FULL
|
321
|
+
}
|
322
|
+
|
323
|
+
// Update state
|
324
|
+
lzvn_offset dst_used = dst - state->dst;
|
325
|
+
state->d_prev = match.D;
|
326
|
+
state->dst = dst;
|
327
|
+
state->src_literal = match.m_end;
|
328
|
+
return dst_used;
|
329
|
+
}
|
330
|
+
|
331
|
+
/*! @abstract Emit a n-bytes literal and update state.
|
332
|
+
* @return number of bytes written to \p dst. May be 0 if there is no more space
|
333
|
+
* in \p dst to emit the literal. */
|
334
|
+
static inline lzvn_offset lzvn_emit_literal(lzvn_encoder_state *state,
|
335
|
+
lzvn_offset n) {
|
336
|
+
size_t L = (size_t)n;
|
337
|
+
unsigned char *dst = emit_literal(state->src + state->src_literal, state->dst,
|
338
|
+
state->dst_end, L);
|
339
|
+
// Check if DST is full
|
340
|
+
if (dst >= state->dst_end)
|
341
|
+
return 0; // FULL
|
342
|
+
|
343
|
+
// Update state
|
344
|
+
lzvn_offset dst_used = dst - state->dst;
|
345
|
+
state->dst = dst;
|
346
|
+
state->src_literal += n;
|
347
|
+
return dst_used;
|
348
|
+
}
|
349
|
+
|
350
|
+
/*! @abstract Emit end-of-stream and update state.
|
351
|
+
* @return number of bytes written to \p dst. May be 0 if there is no more space
|
352
|
+
* in \p dst to emit the instruction. */
|
353
|
+
static inline lzvn_offset lzvn_emit_end_of_stream(lzvn_encoder_state *state) {
|
354
|
+
// Do we have 8 byte in dst?
|
355
|
+
if (state->dst_end < state->dst + 8)
|
356
|
+
return 0; // FULL
|
357
|
+
|
358
|
+
// Insert end marker and update state
|
359
|
+
store8(state->dst, 0x06); // end-of-stream command
|
360
|
+
state->dst += 8;
|
361
|
+
return 8; // dst_used
|
362
|
+
}
|
363
|
+
|
364
|
+
// ===============================================================
|
365
|
+
// Encoder Functions
|
366
|
+
|
367
|
+
/*! @abstract Initialize encoder table in \p state, uses current I/O parameters. */
|
368
|
+
static inline void lzvn_init_table(lzvn_encoder_state *state) {
|
369
|
+
lzvn_offset index = -LZVN_ENCODE_MAX_DISTANCE; // max match distance
|
370
|
+
if (index < state->src_begin)
|
371
|
+
index = state->src_begin;
|
372
|
+
uint32_t value = load4(state->src + index);
|
373
|
+
|
374
|
+
lzvn_encode_entry_type e;
|
375
|
+
for (int i = 0; i < 4; i++) {
|
376
|
+
e.indices[i] = offset_to_s32(index);
|
377
|
+
e.values[i] = value;
|
378
|
+
}
|
379
|
+
for (int u = 0; u < LZVN_ENCODE_HASH_VALUES; u++)
|
380
|
+
state->table[u] = e; // fill entire table
|
381
|
+
}
|
382
|
+
|
383
|
+
void lzvn_encode(lzvn_encoder_state *state) {
|
384
|
+
const lzvn_match_info NO_MATCH = {0};
|
385
|
+
|
386
|
+
for (; state->src_current < state->src_current_end; state->src_current++) {
|
387
|
+
// Get 4 bytes at src_current
|
388
|
+
uint32_t vi = load4(state->src + state->src_current);
|
389
|
+
|
390
|
+
// Compute new hash H at position I, and push value into position table
|
391
|
+
int h = hash3i(vi); // index of first entry
|
392
|
+
|
393
|
+
// Read table entries for H
|
394
|
+
lzvn_encode_entry_type e = state->table[h];
|
395
|
+
|
396
|
+
// Update entry with index=current and value=vi
|
397
|
+
lzvn_encode_entry_type updated_e; // rotate values, so we will replace the oldest
|
398
|
+
updated_e.indices[0] = offset_to_s32(state->src_current);
|
399
|
+
updated_e.indices[1] = e.indices[0];
|
400
|
+
updated_e.indices[2] = e.indices[1];
|
401
|
+
updated_e.indices[3] = e.indices[2];
|
402
|
+
updated_e.values[0] = vi;
|
403
|
+
updated_e.values[1] = e.values[0];
|
404
|
+
updated_e.values[2] = e.values[1];
|
405
|
+
updated_e.values[3] = e.values[2];
|
406
|
+
|
407
|
+
// Do not check matches if still in previously emitted match
|
408
|
+
if (state->src_current < state->src_literal)
|
409
|
+
goto after_emit;
|
410
|
+
|
411
|
+
// Update best with candidate if better
|
412
|
+
#define UPDATE(best, candidate) \
|
413
|
+
do { \
|
414
|
+
if (candidate.K > best.K || \
|
415
|
+
((candidate.K == best.K) && (candidate.m_end > best.m_end + 1))) { \
|
416
|
+
best = candidate; \
|
417
|
+
} \
|
418
|
+
} while (0)
|
419
|
+
// Check candidate. Keep if better.
|
420
|
+
#define CHECK_CANDIDATE(ik, nk) \
|
421
|
+
do { \
|
422
|
+
lzvn_match_info m1; \
|
423
|
+
if (lzvn_find_matchN(state->src, state->src_begin, state->src_end, \
|
424
|
+
state->src_literal, ik, state->src_current, nk, &m1)) { \
|
425
|
+
UPDATE(incoming, m1); \
|
426
|
+
} \
|
427
|
+
} while (0)
|
428
|
+
// Emit match M. Return if we don't have enough space in the destination buffer
|
429
|
+
#define EMIT_MATCH(m) \
|
430
|
+
do { \
|
431
|
+
if (lzvn_emit_match(state, m) == 0) \
|
432
|
+
return; \
|
433
|
+
} while (0)
|
434
|
+
// Emit literal of length L. Return if we don't have enough space in the
|
435
|
+
// destination buffer
|
436
|
+
#define EMIT_LITERAL(l) \
|
437
|
+
do { \
|
438
|
+
if (lzvn_emit_literal(state, l) == 0) \
|
439
|
+
return; \
|
440
|
+
} while (0)
|
441
|
+
|
442
|
+
lzvn_match_info incoming = NO_MATCH;
|
443
|
+
|
444
|
+
// Check candidates in order (closest first)
|
445
|
+
uint32_t diffs[4];
|
446
|
+
for (int k = 0; k < 4; k++)
|
447
|
+
diffs[k] = e.values[k] ^ vi; // XOR, 0 if equal
|
448
|
+
lzvn_offset ik; // index
|
449
|
+
lzvn_offset nk; // match byte count
|
450
|
+
|
451
|
+
// The values stored in e.xyzw are 32-bit signed indices, extended to signed
|
452
|
+
// type lzvn_offset
|
453
|
+
ik = offset_from_s32(e.indices[0]);
|
454
|
+
nk = trailing_zero_bytes(diffs[0]);
|
455
|
+
CHECK_CANDIDATE(ik, nk);
|
456
|
+
ik = offset_from_s32(e.indices[1]);
|
457
|
+
nk = trailing_zero_bytes(diffs[1]);
|
458
|
+
CHECK_CANDIDATE(ik, nk);
|
459
|
+
ik = offset_from_s32(e.indices[2]);
|
460
|
+
nk = trailing_zero_bytes(diffs[2]);
|
461
|
+
CHECK_CANDIDATE(ik, nk);
|
462
|
+
ik = offset_from_s32(e.indices[3]);
|
463
|
+
nk = trailing_zero_bytes(diffs[3]);
|
464
|
+
CHECK_CANDIDATE(ik, nk);
|
465
|
+
|
466
|
+
// Check candidate at previous distance
|
467
|
+
if (state->d_prev != 0) {
|
468
|
+
lzvn_match_info m1;
|
469
|
+
if (lzvn_find_match(state->src, state->src_begin, state->src_end,
|
470
|
+
state->src_literal, state->src_current - state->d_prev,
|
471
|
+
state->src_current, &m1)) {
|
472
|
+
m1.K = m1.M - 1; // fix K for D_prev
|
473
|
+
UPDATE(incoming, m1);
|
474
|
+
}
|
475
|
+
}
|
476
|
+
|
477
|
+
// Here we have the best candidate in incoming, may be NO_MATCH
|
478
|
+
|
479
|
+
// If no incoming match, and literal backlog becomes too high, emit pending
|
480
|
+
// match, or literals if there is no pending match
|
481
|
+
if (incoming.M == 0) {
|
482
|
+
if (state->src_current - state->src_literal >=
|
483
|
+
LZVN_ENCODE_MAX_LITERAL_BACKLOG) // at this point, we always have
|
484
|
+
// current >= literal
|
485
|
+
{
|
486
|
+
if (state->pending.M != 0) {
|
487
|
+
EMIT_MATCH(state->pending);
|
488
|
+
state->pending = NO_MATCH;
|
489
|
+
} else {
|
490
|
+
EMIT_LITERAL(271); // emit long literal (271 is the longest literal size we allow)
|
491
|
+
}
|
492
|
+
}
|
493
|
+
goto after_emit;
|
494
|
+
}
|
495
|
+
|
496
|
+
if (state->pending.M == 0) {
|
497
|
+
// NOTE. Here, we can also emit incoming right away. It will make the
|
498
|
+
// encoder 1.5x faster, at a cost of ~10% lower compression ratio:
|
499
|
+
// EMIT_MATCH(incoming);
|
500
|
+
// state->pending = NO_MATCH;
|
501
|
+
|
502
|
+
// No pending match, emit nothing, keep incoming
|
503
|
+
state->pending = incoming;
|
504
|
+
} else {
|
505
|
+
// Here we have both incoming and pending
|
506
|
+
if (state->pending.m_end <= incoming.m_begin) {
|
507
|
+
// No overlap: emit pending, keep incoming
|
508
|
+
EMIT_MATCH(state->pending);
|
509
|
+
state->pending = incoming;
|
510
|
+
} else {
|
511
|
+
// If pending is better, emit pending and discard incoming.
|
512
|
+
// Otherwise, emit incoming and discard pending.
|
513
|
+
if (incoming.K > state->pending.K)
|
514
|
+
state->pending = incoming;
|
515
|
+
EMIT_MATCH(state->pending);
|
516
|
+
state->pending = NO_MATCH;
|
517
|
+
}
|
518
|
+
}
|
519
|
+
|
520
|
+
after_emit:
|
521
|
+
|
522
|
+
// We commit state changes only after we tried to emit instructions, so we
|
523
|
+
// can restart in the same state in case dst was full and we quit the loop.
|
524
|
+
state->table[h] = updated_e;
|
525
|
+
|
526
|
+
} // i loop
|
527
|
+
|
528
|
+
// Do not emit pending match here. We do it only at the end of stream.
|
529
|
+
}
|
530
|
+
|
531
|
+
// ===============================================================
|
532
|
+
// API entry points
|
533
|
+
|
534
|
+
size_t lzvn_encode_scratch_size(void) { return LZVN_ENCODE_WORK_SIZE; }
|
535
|
+
|
536
|
+
static size_t lzvn_encode_partial(void *__restrict dst, size_t dst_size,
|
537
|
+
const void *__restrict src, size_t src_size,
|
538
|
+
size_t *src_used, void *__restrict work) {
|
539
|
+
// Min size checks to avoid accessing memory outside buffers.
|
540
|
+
if (dst_size < LZVN_ENCODE_MIN_DST_SIZE) {
|
541
|
+
*src_used = 0;
|
542
|
+
return 0;
|
543
|
+
}
|
544
|
+
// Max input size check (limit to offsets on uint32_t).
|
545
|
+
if (src_size > LZVN_ENCODE_MAX_SRC_SIZE) {
|
546
|
+
src_size = LZVN_ENCODE_MAX_SRC_SIZE;
|
547
|
+
}
|
548
|
+
|
549
|
+
// Setup encoder state
|
550
|
+
lzvn_encoder_state state;
|
551
|
+
memset(&state, 0, sizeof(state));
|
552
|
+
|
553
|
+
state.src = src;
|
554
|
+
state.src_begin = 0;
|
555
|
+
state.src_end = (lzvn_offset)src_size;
|
556
|
+
state.src_literal = 0;
|
557
|
+
state.src_current = 0;
|
558
|
+
state.dst = dst;
|
559
|
+
state.dst_begin = dst;
|
560
|
+
state.dst_end = (unsigned char *)dst + dst_size - 8; // reserve 8 bytes for end-of-stream
|
561
|
+
state.table = work;
|
562
|
+
|
563
|
+
// Do not encode if the input buffer is too small. We'll emit a literal instead.
|
564
|
+
if (src_size >= LZVN_ENCODE_MIN_SRC_SIZE) {
|
565
|
+
|
566
|
+
state.src_current_end = (lzvn_offset)src_size - LZVN_ENCODE_MIN_MARGIN;
|
567
|
+
lzvn_init_table(&state);
|
568
|
+
lzvn_encode(&state);
|
569
|
+
|
570
|
+
}
|
571
|
+
|
572
|
+
// No need to test the return value: src_literal will not be updated on failure,
|
573
|
+
// and we will fail later.
|
574
|
+
lzvn_emit_literal(&state, state.src_end - state.src_literal);
|
575
|
+
|
576
|
+
// Restore original size, so end-of-stream always succeeds, and emit it
|
577
|
+
state.dst_end = (unsigned char *)dst + dst_size;
|
578
|
+
lzvn_emit_end_of_stream(&state);
|
579
|
+
|
580
|
+
*src_used = state.src_literal;
|
581
|
+
return (size_t)(state.dst - state.dst_begin);
|
582
|
+
}
|
583
|
+
|
584
|
+
size_t lzvn_encode_buffer(void *__restrict dst, size_t dst_size,
|
585
|
+
const void *__restrict src, size_t src_size,
|
586
|
+
void *__restrict work) {
|
587
|
+
size_t src_used = 0;
|
588
|
+
size_t dst_used =
|
589
|
+
lzvn_encode_partial(dst, dst_size, src, src_size, &src_used, work);
|
590
|
+
if (src_used != src_size)
|
591
|
+
return 0; // could not encode entire input stream = fail
|
592
|
+
return dst_used; // return encoded size
|
593
|
+
}
|