| 1 | /*
|
|---|
| 2 | * Copyright (c) 2000-2008 Marc Alexander Lehmann <[email protected]>
|
|---|
| 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 "lzfP.h"
|
|---|
| 38 | #ifdef VBOX
|
|---|
| 39 | # include "lzf.h" /* need the prototype */
|
|---|
| 40 | #endif
|
|---|
| 41 |
|
|---|
| 42 | #define HSIZE (1 << (HLOG))
|
|---|
| 43 |
|
|---|
| 44 | /*
|
|---|
| 45 | * don't play with this unless you benchmark!
|
|---|
| 46 | * decompression is not dependent on the hash function
|
|---|
| 47 | * the hashing function might seem strange, just believe me
|
|---|
| 48 | * it works ;)
|
|---|
| 49 | */
|
|---|
| 50 | #ifndef FRST
|
|---|
| 51 | # define FRST(p) (((p[0]) << 8) | p[1])
|
|---|
| 52 | # define NEXT(v,p) (((v) << 8) | p[2])
|
|---|
| 53 | # if ULTRA_FAST
|
|---|
| 54 | # define IDX(h) ((( h >> (3*8 - HLOG)) - h ) & (HSIZE - 1))
|
|---|
| 55 | # elif VERY_FAST
|
|---|
| 56 | # define IDX(h) ((( h >> (3*8 - HLOG)) - h*5) & (HSIZE - 1))
|
|---|
| 57 | # else
|
|---|
| 58 | # define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (HSIZE - 1))
|
|---|
| 59 | # endif
|
|---|
| 60 | #endif
|
|---|
| 61 | /*
|
|---|
| 62 | * IDX works because it is very similar to a multiplicative hash, e.g.
|
|---|
| 63 | * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1))
|
|---|
| 64 | * the latter is also quite fast on newer CPUs, and compresses similarly.
|
|---|
| 65 | *
|
|---|
| 66 | * the next one is also quite good, albeit slow ;)
|
|---|
| 67 | * (int)(cos(h & 0xffffff) * 1e6)
|
|---|
| 68 | */
|
|---|
| 69 |
|
|---|
| 70 | #if 0
|
|---|
| 71 | /* original lzv-like hash function, much worse and thus slower */
|
|---|
| 72 | # define FRST(p) (p[0] << 5) ^ p[1]
|
|---|
| 73 | # define NEXT(v,p) ((v) << 5) ^ p[2]
|
|---|
| 74 | # define IDX(h) ((h) & (HSIZE - 1))
|
|---|
| 75 | #endif
|
|---|
| 76 |
|
|---|
| 77 | #define MAX_LIT (1 << 5)
|
|---|
| 78 | #define MAX_OFF (1 << 13)
|
|---|
| 79 | #define MAX_REF ((1 << 8) + (1 << 3))
|
|---|
| 80 |
|
|---|
| 81 | #if __GNUC__ >= 3
|
|---|
| 82 | # define expect(expr,value) __builtin_expect ((expr),(value))
|
|---|
| 83 | # define inline inline
|
|---|
| 84 | #else
|
|---|
| 85 | # define expect(expr,value) (expr)
|
|---|
| 86 | # define inline static
|
|---|
| 87 | #endif
|
|---|
| 88 |
|
|---|
| 89 | #define expect_false(expr) expect ((expr) != 0, 0)
|
|---|
| 90 | #define expect_true(expr) expect ((expr) != 0, 1)
|
|---|
| 91 |
|
|---|
| 92 | /*
|
|---|
| 93 | * compressed format
|
|---|
| 94 | *
|
|---|
| 95 | * 000LLLLL <L+1> ; literal
|
|---|
| 96 | * LLLooooo oooooooo ; backref L
|
|---|
| 97 | * 111ooooo LLLLLLLL oooooooo ; backref L+7
|
|---|
| 98 | *
|
|---|
| 99 | */
|
|---|
| 100 |
|
|---|
| 101 | unsigned int
|
|---|
| 102 | lzf_compress (const void *const in_data, unsigned int in_len,
|
|---|
| 103 | void *out_data, unsigned int out_len
|
|---|
| 104 | #if LZF_STATE_ARG
|
|---|
| 105 | , LZF_STATE htab
|
|---|
| 106 | #endif
|
|---|
| 107 | )
|
|---|
| 108 | {
|
|---|
| 109 | #if !LZF_STATE_ARG
|
|---|
| 110 | LZF_STATE htab;
|
|---|
| 111 | #endif
|
|---|
| 112 | const u8 **hslot;
|
|---|
| 113 | const u8 *ip = (const u8 *)in_data;
|
|---|
| 114 | u8 *op = (u8 *)out_data;
|
|---|
| 115 | const u8 *in_end = ip + in_len;
|
|---|
| 116 | u8 *out_end = op + out_len;
|
|---|
| 117 | const u8 *ref;
|
|---|
| 118 |
|
|---|
| 119 | /* off requires a type wide enough to hold a general pointer difference.
|
|---|
| 120 | * ISO C doesn't have that (size_t might not be enough and ptrdiff_t only
|
|---|
| 121 | * works for differences within a single object). We also assume that no
|
|---|
| 122 | * no bit pattern traps. Since the only platform that is both non-POSIX
|
|---|
| 123 | * and fails to support both assumptions is windows 64 bit, we make a
|
|---|
| 124 | * special workaround for it.
|
|---|
| 125 | */
|
|---|
| 126 | #if defined (WIN32) && defined (_M_X64)
|
|---|
| 127 | unsigned _int64 off; /* workaround for missing POSIX compliance */
|
|---|
| 128 | #else
|
|---|
| 129 | unsigned long off;
|
|---|
| 130 | #endif
|
|---|
| 131 | unsigned int hval;
|
|---|
| 132 | int lit;
|
|---|
| 133 |
|
|---|
| 134 | if (!in_len || !out_len)
|
|---|
| 135 | return 0;
|
|---|
| 136 |
|
|---|
| 137 | #if INIT_HTAB
|
|---|
| 138 | memset (htab, 0, sizeof (htab));
|
|---|
| 139 | # if 0
|
|---|
| 140 | for (hslot = htab; hslot < htab + HSIZE; hslot++)
|
|---|
| 141 | *hslot++ = ip;
|
|---|
| 142 | # endif
|
|---|
| 143 | #endif
|
|---|
| 144 |
|
|---|
| 145 | lit = 0; op++; /* start run */
|
|---|
| 146 |
|
|---|
| 147 | hval = FRST (ip);
|
|---|
| 148 | while (ip < in_end - 2)
|
|---|
| 149 | {
|
|---|
| 150 | hval = NEXT (hval, ip);
|
|---|
| 151 | hslot = htab + IDX (hval);
|
|---|
| 152 | ref = *hslot; *hslot = ip;
|
|---|
| 153 |
|
|---|
| 154 | if (1
|
|---|
| 155 | #if INIT_HTAB
|
|---|
| 156 | && ref < ip /* the next test will actually take care of this, but this is faster */
|
|---|
| 157 | #endif
|
|---|
| 158 | && (off = ip - ref - 1) < MAX_OFF
|
|---|
| 159 | && ip + 4 < in_end
|
|---|
| 160 | && ref > (u8 *)in_data
|
|---|
| 161 | #if STRICT_ALIGN
|
|---|
| 162 | && ref[0] == ip[0]
|
|---|
| 163 | && ref[1] == ip[1]
|
|---|
| 164 | && ref[2] == ip[2]
|
|---|
| 165 | #else
|
|---|
| 166 | && *(u16 *)ref == *(u16 *)ip
|
|---|
| 167 | && ref[2] == ip[2]
|
|---|
| 168 | #endif
|
|---|
| 169 | )
|
|---|
| 170 | {
|
|---|
| 171 | /* match found at *ref++ */
|
|---|
| 172 | unsigned int len = 2;
|
|---|
| 173 | unsigned int maxlen = in_end - ip - len;
|
|---|
| 174 | maxlen = maxlen > MAX_REF ? MAX_REF : maxlen;
|
|---|
| 175 |
|
|---|
| 176 | op [- lit - 1] = lit - 1; /* stop run */
|
|---|
| 177 | op -= !lit; /* undo run if length is zero */
|
|---|
| 178 |
|
|---|
| 179 | if (expect_false (op + 3 + 1 >= out_end))
|
|---|
| 180 | return 0;
|
|---|
| 181 |
|
|---|
| 182 | for (;;)
|
|---|
| 183 | {
|
|---|
| 184 | if (expect_true (maxlen > 16))
|
|---|
| 185 | {
|
|---|
| 186 | len++; if (ref [len] != ip [len]) break;
|
|---|
| 187 | len++; if (ref [len] != ip [len]) break;
|
|---|
| 188 | len++; if (ref [len] != ip [len]) break;
|
|---|
| 189 | len++; if (ref [len] != ip [len]) break;
|
|---|
| 190 |
|
|---|
| 191 | len++; if (ref [len] != ip [len]) break;
|
|---|
| 192 | len++; if (ref [len] != ip [len]) break;
|
|---|
| 193 | len++; if (ref [len] != ip [len]) break;
|
|---|
| 194 | len++; if (ref [len] != ip [len]) break;
|
|---|
| 195 |
|
|---|
| 196 | len++; if (ref [len] != ip [len]) break;
|
|---|
| 197 | len++; if (ref [len] != ip [len]) break;
|
|---|
| 198 | len++; if (ref [len] != ip [len]) break;
|
|---|
| 199 | len++; if (ref [len] != ip [len]) break;
|
|---|
| 200 |
|
|---|
| 201 | len++; if (ref [len] != ip [len]) break;
|
|---|
| 202 | len++; if (ref [len] != ip [len]) break;
|
|---|
| 203 | len++; if (ref [len] != ip [len]) break;
|
|---|
| 204 | len++; if (ref [len] != ip [len]) break;
|
|---|
| 205 | }
|
|---|
| 206 |
|
|---|
| 207 | do
|
|---|
| 208 | len++;
|
|---|
| 209 | while (len < maxlen && ref[len] == ip[len]);
|
|---|
| 210 |
|
|---|
| 211 | break;
|
|---|
| 212 | }
|
|---|
| 213 |
|
|---|
| 214 | len -= 2; /* len is now #octets - 1 */
|
|---|
| 215 | ip++;
|
|---|
| 216 |
|
|---|
| 217 | if (len < 7)
|
|---|
| 218 | {
|
|---|
| 219 | *op++ = (off >> 8) + (len << 5);
|
|---|
| 220 | }
|
|---|
| 221 | else
|
|---|
| 222 | {
|
|---|
| 223 | *op++ = (off >> 8) + ( 7 << 5);
|
|---|
| 224 | *op++ = len - 7;
|
|---|
| 225 | }
|
|---|
| 226 |
|
|---|
| 227 | *op++ = off;
|
|---|
| 228 | lit = 0; op++; /* start run */
|
|---|
| 229 |
|
|---|
| 230 | ip += len + 1;
|
|---|
| 231 |
|
|---|
| 232 | if (expect_false (ip >= in_end - 2))
|
|---|
| 233 | break;
|
|---|
| 234 |
|
|---|
| 235 | #if ULTRA_FAST || VERY_FAST
|
|---|
| 236 | --ip;
|
|---|
| 237 | # if VERY_FAST && !ULTRA_FAST
|
|---|
| 238 | --ip;
|
|---|
| 239 | # endif
|
|---|
| 240 | hval = FRST (ip);
|
|---|
| 241 |
|
|---|
| 242 | hval = NEXT (hval, ip);
|
|---|
| 243 | htab[IDX (hval)] = ip;
|
|---|
| 244 | ip++;
|
|---|
| 245 |
|
|---|
| 246 | # if VERY_FAST && !ULTRA_FAST
|
|---|
| 247 | hval = NEXT (hval, ip);
|
|---|
| 248 | htab[IDX (hval)] = ip;
|
|---|
| 249 | ip++;
|
|---|
| 250 | # endif
|
|---|
| 251 | #else
|
|---|
| 252 | ip -= len + 1;
|
|---|
| 253 |
|
|---|
| 254 | do
|
|---|
| 255 | {
|
|---|
| 256 | hval = NEXT (hval, ip);
|
|---|
| 257 | htab[IDX (hval)] = ip;
|
|---|
| 258 | ip++;
|
|---|
| 259 | }
|
|---|
| 260 | while (len--);
|
|---|
| 261 | #endif
|
|---|
| 262 | }
|
|---|
| 263 | else
|
|---|
| 264 | {
|
|---|
| 265 | /* one more literal byte we must copy */
|
|---|
| 266 | if (expect_false (op >= out_end))
|
|---|
| 267 | return 0;
|
|---|
| 268 |
|
|---|
| 269 | lit++; *op++ = *ip++;
|
|---|
| 270 |
|
|---|
| 271 | if (expect_false (lit == MAX_LIT))
|
|---|
| 272 | {
|
|---|
| 273 | op [- lit - 1] = lit - 1; /* stop run */
|
|---|
| 274 | lit = 0; op++; /* start run */
|
|---|
| 275 | }
|
|---|
| 276 | }
|
|---|
| 277 | }
|
|---|
| 278 |
|
|---|
| 279 | if (op + 3 > out_end) /* at most 3 bytes can be missing here */
|
|---|
| 280 | return 0;
|
|---|
| 281 |
|
|---|
| 282 | while (ip < in_end)
|
|---|
| 283 | {
|
|---|
| 284 | lit++; *op++ = *ip++;
|
|---|
| 285 |
|
|---|
| 286 | if (expect_false (lit == MAX_LIT))
|
|---|
| 287 | {
|
|---|
| 288 | op [- lit - 1] = lit - 1; /* stop run */
|
|---|
| 289 | lit = 0; op++; /* start run */
|
|---|
| 290 | }
|
|---|
| 291 | }
|
|---|
| 292 |
|
|---|
| 293 | op [- lit - 1] = lit - 1; /* end run */
|
|---|
| 294 | op -= !lit; /* undo run if length is zero */
|
|---|
| 295 |
|
|---|
| 296 | return op - (u8 *)out_data;
|
|---|
| 297 | }
|
|---|
| 298 |
|
|---|
