lzf.c revision 225736 
1/*
2 * Copyright (c) 2000-2008 Marc Alexander Lehmann <schmorp@schmorp.de>
3 *
4 * Redistribution and use in source and binary forms, with or without modifica-
5 * tion, are permitted provided that the following conditions are met:
6 *
7 *   1.  Redistributions of source code must retain the above copyright notice,
8 *       this list of conditions and the following disclaimer.
9 *
10 *   2.  Redistributions in binary form must reproduce the above copyright
11 *       notice, this list of conditions and the following disclaimer in the
12 *       documentation and/or other materials provided with the distribution.
13 *
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-
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-
18 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
19 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
20 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
21 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
22 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
23 * OF THE POSSIBILITY OF SUCH DAMAGE.
24 *
25 * Alternatively, the contents of this file may be used under the terms of
26 * the GNU General Public License ("GPL") version 2 or any later version,
27 * in which case the provisions of the GPL are applicable instead of
28 * the above. If you wish to allow the use of your version of this file
29 * only under the terms of the GPL and not to allow others to use your
30 * version of this file under the BSD license, indicate your decision
31 * by deleting the provisions above and replace them with the notice
32 * and other provisions required by the GPL. If you do not delete the
33 * provisions above, a recipient may use your version of this file under
34 * either the BSD or the GPL.
35 */
36
37#include "lzf.h"
38
39#define HSIZE (1 << (HLOG))
40
41/*
42 * don't play with this unless you benchmark!
43 * decompression is not dependent on the hash function
44 * the hashing function might seem strange, just believe me
45 * it works ;)
46 */
47#ifndef FRST
48# define FRST(p) (((p[0]) << 8) | p[1])
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
55#  define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (HSIZE - 1))
56# endif
57#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
67#if 0
68/* original lzv-like hash function, much worse and thus slower */
69# define FRST(p) (p[0] << 5) ^ p[1]
70# define NEXT(v,p) ((v) << 5) ^ p[2]
71# define IDX(h) ((h) & (HSIZE - 1))
72#endif
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
79# define expect(expr,value)         __builtin_expect ((expr),(value))
80# define inline                     inline
81#else
82# define expect(expr,value)         (expr)
83# define inline                     static
84#endif
85
86#define expect_false(expr) expect ((expr) != 0, 0)
87#define expect_true(expr)  expect ((expr) != 0, 1)
88
89/*
90 * compressed format
91 *
92 * 000LLLLL <L+1>    ; literal
93 * LLLooooo oooooooo ; backref L
94 * 111ooooo LLLLLLLL oooooooo ; backref L+7
95 *
96 */
97
98unsigned int
99lzf_compress (const void *const in_data, unsigned int in_len,
100	      void *out_data, unsigned int out_len
101#if LZF_STATE_ARG
102              , LZF_STATE htab
103#endif
104              )
105{
106#if !LZF_STATE_ARG
107  LZF_STATE htab;
108#endif
109  const u8 **hslot;
110  const u8 *ip = (const u8 *)in_data;
111        u8 *op = (u8 *)out_data;
112  const u8 *in_end  = ip + in_len;
113        u8 *out_end = op + out_len;
114  const u8 *ref;
115
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
126  unsigned long off;
127#endif
128  unsigned int hval;
129  int lit;
130
131  if (!in_len || !out_len)
132    return 0;
133
134#if INIT_HTAB
135  memset (htab, 0, sizeof (htab));
136# if 0
137  for (hslot = htab; hslot < htab + HSIZE; hslot++)
138    *hslot++ = ip;
139# endif
140#endif
141
142  lit = 0; op++; /* start run */
143
144  hval = FRST (ip);
145  while (ip < in_end - 2)
146    {
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 > (const 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          && *(const u16 *)ref == *(const u16 *)ip
164          && ref[2] == ip[2]
165#endif
166        )
167        {
168          /* match found at *ref++ */
169          unsigned int len = 2;
170          unsigned int maxlen = in_end - ip - len;
171          maxlen = maxlen > MAX_REF ? MAX_REF : maxlen;
172
173          if (expect_false (op + 3 + 1 >= out_end)) /* first a faster conservative test */
174            if (op - !lit + 3 + 1 >= out_end) /* second the exact but rare test */
175              return 0;
176
177          op [- lit - 1] = lit - 1; /* stop run */
178          op -= !lit; /* undo run if length is zero */
179
180          for (;;)
181            {
182              if (expect_true (maxlen > 16))
183                {
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                  len++; if (ref [len] != ip [len]) break;
188
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                  len++; if (ref [len] != ip [len]) break;
193
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                  len++; if (ref [len] != ip [len]) break;
198
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                  len++; if (ref [len] != ip [len]) break;
203                }
204
205              do
206                len++;
207              while (len < maxlen && ref[len] == ip[len]);
208
209              break;
210            }
211
212          len -= 2; /* len is now #octets - 1 */
213          ip++;
214
215          if (len < 7)
216            {
217              *op++ = (off >> 8) + (len << 5);
218            }
219          else
220            {
221              *op++ = (off >> 8) + (  7 << 5);
222              *op++ = len - 7;
223            }
224
225          *op++ = off;
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;
237# endif
238          hval = FRST (ip);
239
240          hval = NEXT (hval, ip);
241          htab[IDX (hval)] = ip;
242          ip++;
243
244# if VERY_FAST && !ULTRA_FAST
245          hval = NEXT (hval, ip);
246          htab[IDX (hval)] = ip;
247          ip++;
248# endif
249#else
250          ip -= len + 1;
251
252          do
253            {
254              hval = NEXT (hval, ip);
255              htab[IDX (hval)] = ip;
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)
281    {
282      lit++; *op++ = *ip++;
283
284      if (expect_false (lit == MAX_LIT))
285        {
286          op [- lit - 1] = lit - 1; /* stop run */
287          lit = 0; op++; /* start run */
288        }
289    }
290
291  op [- lit - 1] = lit - 1; /* end run */
292  op -= !lit; /* undo run if length is zero */
293
294  return op - (u8 *)out_data;
295}
296
297#if AVOID_ERRNO
298# define SET_ERRNO(n)
299#else
300# include <errno.h>
301# define SET_ERRNO(n) errno = (n)
302#endif
303
304#if (__i386 || __amd64) && __GNUC__ >= 3
305# define lzf_movsb(dst, src, len)                \
306   asm ("rep movsb"                              \
307        : "=D" (dst), "=S" (src), "=c" (len)     \
308        :  "0" (dst),  "1" (src),  "2" (len));
309#endif
310
311unsigned int
312lzf_decompress (const void *const in_data,  unsigned int in_len,
313                void             *out_data, unsigned int out_len)
314{
315  u8 const *ip = (const u8 *)in_data;
316  u8       *op = (u8 *)out_data;
317  u8 const *const in_end  = ip + in_len;
318  u8       *const out_end = op + out_len;
319
320  do
321    {
322      unsigned int ctrl = *ip++;
323
324      if (ctrl < (1 << 5)) /* literal run */
325        {
326          ctrl++;
327
328          if (op + ctrl > out_end)
329            {
330              SET_ERRNO (E2BIG);
331              return 0;
332            }
333
334#if CHECK_INPUT
335          if (ip + ctrl > in_end)
336            {
337              SET_ERRNO (EINVAL);
338              return 0;
339            }
340#endif
341
342#ifdef lzf_movsb
343          lzf_movsb (op, ip, ctrl);
344#else
345          do
346            *op++ = *ip++;
347          while (--ctrl);
348#endif
349        }
350      else /* back reference */
351        {
352          unsigned int len = ctrl >> 5;
353
354          u8 *ref = op - ((ctrl & 0x1f) << 8) - 1;
355
356#if CHECK_INPUT
357          if (ip >= in_end)
358            {
359              SET_ERRNO (EINVAL);
360              return 0;
361            }
362#endif
363          if (len == 7)
364            {
365              len += *ip++;
366#if CHECK_INPUT
367              if (ip >= in_end)
368                {
369                  SET_ERRNO (EINVAL);
370                  return 0;
371                }
372#endif
373            }
374
375          ref -= *ip++;
376
377          if (op + len + 2 > out_end)
378            {
379              SET_ERRNO (E2BIG);
380              return 0;
381            }
382
383          if (ref < (u8 *)out_data)
384            {
385              SET_ERRNO (EINVAL);
386              return 0;
387            }
388
389#ifdef lzf_movsb
390          len += 2;
391          lzf_movsb (op, ref, len);
392#else
393          *op++ = *ref++;
394          *op++ = *ref++;
395
396          do
397            *op++ = *ref++;
398          while (--len);
399#endif
400        }
401    }
402  while (ip < in_end);
403
404  return op - (u8 *)out_data;
405}
406
407