| 1 |
/* |
| 2 |
* bs.cpp - BZZ-coder from DjVuLibre, a general purpose compressor |
| 3 |
* based on the Burrows-Wheeler (or "block sorting") transform. |
| 4 |
*/ |
| 5 |
|
| 6 |
#include "../base/mdjvucfg.h" |
| 7 |
#include <minidjvu/minidjvu.h> |
| 8 |
|
| 9 |
#include <stdlib.h> |
| 10 |
#include <stdio.h> |
| 11 |
#include <string.h> |
| 12 |
|
| 13 |
#include "bs.h" |
| 14 |
|
| 15 |
// ======================================== |
| 16 |
// --- Global Definitions |
| 17 |
|
| 18 |
|
| 19 |
// Sorting tresholds |
| 20 |
enum { FREQMAX=4, CTXIDS=3 }; |
| 21 |
// Limits on block sizes |
| 22 |
enum { MINBLOCK=10, MAXBLOCK=4096 }; |
| 23 |
|
| 24 |
|
| 25 |
|
| 26 |
#ifdef OVERFLOW |
| 27 |
#undef OVERFLOW |
| 28 |
#endif |
| 29 |
// Overflow required when encoding |
| 30 |
static const int OVERFLOW=32; |
| 31 |
|
| 32 |
// Sorting tresholds |
| 33 |
static const int RANKSORT_THRESH=10; |
| 34 |
static const int QUICKSORT_STACK=512; |
| 35 |
static const int PRESORT_THRESH=10; |
| 36 |
static const int PRESORT_DEPTH=8; |
| 37 |
static const int RADIX_THRESH=32768; |
| 38 |
|
| 39 |
static const int FREQS0=100000; |
| 40 |
static const int FREQS1=1000000; |
| 41 |
|
| 42 |
// ======================================== |
| 43 |
// -- Sorting Routines |
| 44 |
|
| 45 |
|
| 46 |
class _BSort // DJVU_CLASS |
| 47 |
{ |
| 48 |
public: |
| 49 |
~_BSort(); |
| 50 |
_BSort(unsigned char *data, int size); |
| 51 |
void run(int &markerpos); |
| 52 |
private: |
| 53 |
// Members |
| 54 |
int size; |
| 55 |
unsigned char *data; |
| 56 |
unsigned int *posn; |
| 57 |
int *rank; |
| 58 |
// Helpers |
| 59 |
inline int GT(int p1, int p2, int depth); |
| 60 |
inline int GTD(int p1, int p2, int depth); |
| 61 |
// -- final in-depth sort |
| 62 |
void ranksort(int lo, int hi, int d); |
| 63 |
// -- doubling sort |
| 64 |
int pivot3r(int *rr, int lo, int hi); |
| 65 |
void quicksort3r(int lo, int hi, int d); |
| 66 |
// -- presort to depth PRESORT_DEPTH |
| 67 |
unsigned char pivot3d(unsigned char *dd, int lo, int hi); |
| 68 |
void quicksort3d(int lo, int hi, int d); |
| 69 |
// -- radixsort |
| 70 |
void radixsort16(void); |
| 71 |
void radixsort8(void); |
| 72 |
}; |
| 73 |
|
| 74 |
|
| 75 |
// blocksort -- the main entry point |
| 76 |
|
| 77 |
static void blocksort(unsigned char *data, int size, int &markerpos) |
| 78 |
{ |
| 79 |
_BSort bsort(data, size); |
| 80 |
bsort.run(markerpos); |
| 81 |
} |
| 82 |
|
| 83 |
|
| 84 |
// _BSort construction |
| 85 |
|
| 86 |
_BSort::_BSort(unsigned char *xdata, int xsize) |
| 87 |
: size(xsize), data(xdata) |
| 88 |
{ |
| 89 |
posn = (unsigned int *) calloc (size, sizeof(unsigned int)); |
| 90 |
rank = (int *) calloc (size+1, sizeof(int)); |
| 91 |
|
| 92 |
assert(size>0 && size<0x1000000); |
| 93 |
rank[size] = -1; |
| 94 |
} |
| 95 |
|
| 96 |
_BSort::~_BSort() |
| 97 |
{ |
| 98 |
free(posn); |
| 99 |
free(rank); |
| 100 |
} |
| 101 |
|
| 102 |
|
| 103 |
|
| 104 |
// GT -- compare suffixes using rank information |
| 105 |
|
| 106 |
inline int _BSort::GT(int p1, int p2, int depth) |
| 107 |
{ |
| 108 |
int r1, r2; |
| 109 |
int twod = depth + depth; |
| 110 |
while (1) |
| 111 |
{ |
| 112 |
r1=rank[p1+depth]; r2=rank[p2+depth]; |
| 113 |
p1+=twod; p2+=twod; |
| 114 |
if (r1!=r2) |
| 115 |
return (r1>r2); |
| 116 |
r1=rank[p1]; r2=rank[p2]; |
| 117 |
if (r1!=r2) |
| 118 |
return (r1>r2); |
| 119 |
r1=rank[p1+depth]; r2=rank[p2+depth]; |
| 120 |
p1+=twod; p2+=twod; |
| 121 |
if (r1!=r2) |
| 122 |
return (r1>r2); |
| 123 |
r1=rank[p1]; r2=rank[p2]; |
| 124 |
if (r1!=r2) |
| 125 |
return (r1>r2); |
| 126 |
r1=rank[p1+depth]; r2=rank[p2+depth]; |
| 127 |
p1+=twod; p2+=twod; |
| 128 |
if (r1!=r2) |
| 129 |
return (r1>r2); |
| 130 |
r1=rank[p1]; r2=rank[p2]; |
| 131 |
if (r1!=r2) |
| 132 |
return (r1>r2); |
| 133 |
r1=rank[p1+depth]; r2=rank[p2+depth]; |
| 134 |
p1+=twod; p2+=twod; |
| 135 |
if (r1!=r2) |
| 136 |
return (r1>r2); |
| 137 |
r1=rank[p1]; r2=rank[p2]; |
| 138 |
if (r1!=r2) |
| 139 |
return (r1>r2); |
| 140 |
}; |
| 141 |
} |
| 142 |
|
| 143 |
|
| 144 |
// _BSort::ranksort -- |
| 145 |
// -- a simple insertion sort based on GT |
| 146 |
|
| 147 |
void _BSort::ranksort(int lo, int hi, int depth) |
| 148 |
{ |
| 149 |
int i,j; |
| 150 |
for (i=lo+1; i<=hi; i++) |
| 151 |
{ |
| 152 |
int tmp = posn[i]; |
| 153 |
for(j=i-1; j>=lo && GT(posn[j], tmp, depth); j--) |
| 154 |
posn[j+1] = posn[j]; |
| 155 |
posn[j+1] = tmp; |
| 156 |
} |
| 157 |
for(i=lo;i<=hi;i++) |
| 158 |
rank[posn[i]]=i; |
| 159 |
} |
| 160 |
|
| 161 |
// pivot -- return suitable pivot |
| 162 |
|
| 163 |
int _BSort::pivot3r(int *rr, int lo, int hi) |
| 164 |
{ |
| 165 |
int c1, c2, c3; |
| 166 |
if (hi-lo > 256) |
| 167 |
{ |
| 168 |
c1 = pivot3r(rr, lo, (6*lo+2*hi)/8); |
| 169 |
c2 = pivot3r(rr, (5*lo+3*hi)/8, (3*lo+5*hi)/8); |
| 170 |
c3 = pivot3r(rr, (2*lo+6*hi)/8, hi); |
| 171 |
} |
| 172 |
else |
| 173 |
{ |
| 174 |
c1 = rr[posn[lo]]; |
| 175 |
c2 = rr[posn[(lo+hi)/2]]; |
| 176 |
c3 = rr[posn[hi]]; |
| 177 |
} |
| 178 |
// Extract median |
| 179 |
if (c1>c3) |
| 180 |
{ int tmp=c1; c1=c3; c3=tmp; } |
| 181 |
if (c2<=c1) |
| 182 |
return c1; |
| 183 |
else if (c2>=c3) |
| 184 |
return c3; |
| 185 |
else |
| 186 |
return c2; |
| 187 |
} |
| 188 |
|
| 189 |
|
| 190 |
// _BSort::quicksort3r -- Three way quicksort algorithm |
| 191 |
// Sort suffixes based on rank at pos+depth |
| 192 |
// The algorithm breaks into ranksort when size is |
| 193 |
// smaller than RANKSORT_THRESH |
| 194 |
|
| 195 |
static inline int mini(int a, int b) |
| 196 |
{ |
| 197 |
return (a<=b) ? a : b; |
| 198 |
} |
| 199 |
|
| 200 |
static inline void vswap(int i, int j, int n, unsigned int *x) |
| 201 |
{ |
| 202 |
while (n-- > 0) |
| 203 |
{ |
| 204 |
int tmp = x[i]; x[i++]=x[j]; x[j++]=tmp; |
| 205 |
} |
| 206 |
} |
| 207 |
|
| 208 |
void _BSort::quicksort3r(int lo, int hi, int depth) |
| 209 |
{ |
| 210 |
/* Initialize stack */ |
| 211 |
int slo[QUICKSORT_STACK]; |
| 212 |
int shi[QUICKSORT_STACK]; |
| 213 |
int sp = 1; |
| 214 |
slo[0] = lo; |
| 215 |
shi[0] = hi; |
| 216 |
// Recursion elimination loop |
| 217 |
while (--sp>=0) |
| 218 |
{ |
| 219 |
lo = slo[sp]; |
| 220 |
hi = shi[sp]; |
| 221 |
// Test for insertion sort |
| 222 |
if (hi-lo<RANKSORT_THRESH) |
| 223 |
{ |
| 224 |
ranksort(lo, hi, depth); |
| 225 |
} |
| 226 |
else |
| 227 |
{ |
| 228 |
int tmp; |
| 229 |
int *rr=rank+depth; |
| 230 |
int med = pivot3r(rr,lo,hi); |
| 231 |
// -- positions are organized as follows: |
| 232 |
// [lo..l1[ [l1..l[ ]h..h1] ]h1..hi] |
| 233 |
// = < > = |
| 234 |
int l1 = lo; |
| 235 |
int h1 = hi; |
| 236 |
while (rr[posn[l1]]==med && l1<h1) { l1++; } |
| 237 |
while (rr[posn[h1]]==med && l1<h1) { h1--; } |
| 238 |
int l = l1; |
| 239 |
int h = h1; |
| 240 |
// -- partition set |
| 241 |
for (;;) |
| 242 |
{ |
| 243 |
while (l<=h) |
| 244 |
{ |
| 245 |
int c = rr[posn[l]] - med; |
| 246 |
if (c > 0) break; |
| 247 |
if (c == 0) { tmp=posn[l]; posn[l]=posn[l1]; posn[l1++]=tmp; } |
| 248 |
l++; |
| 249 |
} |
| 250 |
while (l<=h) |
| 251 |
{ |
| 252 |
int c = rr[posn[h]] - med; |
| 253 |
if (c < 0) break; |
| 254 |
if (c == 0) { tmp=posn[h]; posn[h]=posn[h1]; posn[h1--]=tmp; } |
| 255 |
h--; |
| 256 |
} |
| 257 |
if (l>h) break; |
| 258 |
tmp=posn[l]; posn[l]=posn[h]; posn[h]=tmp; |
| 259 |
} |
| 260 |
// -- reorganize as follows |
| 261 |
// [lo..l1[ [l1..h1] ]h1..hi] |
| 262 |
// < = > |
| 263 |
tmp = mini(l1-lo, l-l1); |
| 264 |
vswap(lo, l-tmp, tmp, posn); |
| 265 |
l1 = lo + (l-l1); |
| 266 |
tmp = mini(hi-h1, h1-h); |
| 267 |
vswap(hi-tmp+1, h+1, tmp, posn); |
| 268 |
h1 = hi - (h1-h); |
| 269 |
// -- process segments |
| 270 |
assert(sp+2<QUICKSORT_STACK); |
| 271 |
// ----- middle segment (=?) [l1, h1] |
| 272 |
for(int i=l1;i<=h1;i++) |
| 273 |
rank[posn[i]] = h1; |
| 274 |
// ----- lower segment (<) [lo, l1[ |
| 275 |
if (l1 > lo) |
| 276 |
{ |
| 277 |
for(int i=lo;i<l1;i++) |
| 278 |
rank[posn[i]]=l1-1; |
| 279 |
slo[sp]=lo; |
| 280 |
shi[sp]=l1-1; |
| 281 |
if (slo[sp] < shi[sp]) |
| 282 |
sp++; |
| 283 |
} |
| 284 |
// ----- upper segment (>) ]h1, hi] |
| 285 |
if (h1 < hi) |
| 286 |
{ |
| 287 |
slo[sp]=h1+1; |
| 288 |
shi[sp]=hi; |
| 289 |
if (slo[sp] < shi[sp]) |
| 290 |
sp++; |
| 291 |
} |
| 292 |
} |
| 293 |
} |
| 294 |
} |
| 295 |
|
| 296 |
|
| 297 |
|
| 298 |
|
| 299 |
|
| 300 |
|
| 301 |
// GTD -- compare suffixes using data information |
| 302 |
// (up to depth PRESORT_DEPTH) |
| 303 |
|
| 304 |
inline int _BSort::GTD(int p1, int p2, int depth) |
| 305 |
{ |
| 306 |
unsigned char c1, c2; |
| 307 |
p1+=depth; p2+=depth; |
| 308 |
while (depth < PRESORT_DEPTH) |
| 309 |
{ |
| 310 |
// Perform two |
| 311 |
c1=data[p1]; c2=data[p2]; |
| 312 |
if (c1!=c2) |
| 313 |
return (c1>c2); |
| 314 |
c1=data[p1+1]; c2=data[p2+1]; |
| 315 |
p1+=2; p2+=2; depth+=2; |
| 316 |
if (c1!=c2) |
| 317 |
return (c1>c2); |
| 318 |
} |
| 319 |
if (p1<size && p2<size) |
| 320 |
return 0; |
| 321 |
return (p1<p2); |
| 322 |
} |
| 323 |
|
| 324 |
// pivot3d -- return suitable pivot |
| 325 |
|
| 326 |
unsigned char _BSort::pivot3d(unsigned char *rr, int lo, int hi) |
| 327 |
{ |
| 328 |
unsigned char c1, c2, c3; |
| 329 |
if (hi-lo > 256) |
| 330 |
{ |
| 331 |
c1 = pivot3d(rr, lo, (6*lo+2*hi)/8); |
| 332 |
c2 = pivot3d(rr, (5*lo+3*hi)/8, (3*lo+5*hi)/8); |
| 333 |
c3 = pivot3d(rr, (2*lo+6*hi)/8, hi); |
| 334 |
} |
| 335 |
else |
| 336 |
{ |
| 337 |
c1 = rr[posn[lo]]; |
| 338 |
c2 = rr[posn[(lo+hi)/2]]; |
| 339 |
c3 = rr[posn[hi]]; |
| 340 |
} |
| 341 |
// Extract median |
| 342 |
if (c1>c3) |
| 343 |
{ int tmp=c1; c1=c3; c3=tmp; } |
| 344 |
if (c2<=c1) |
| 345 |
return c1; |
| 346 |
else if (c2>=c3) |
| 347 |
return c3; |
| 348 |
else |
| 349 |
return c2; |
| 350 |
} |
| 351 |
|
| 352 |
|
| 353 |
// _BSort::quicksort3d -- Three way quicksort algorithm |
| 354 |
// Sort suffixes based on strings until reaching |
| 355 |
// depth rank at pos+depth |
| 356 |
// The algorithm breaks into ranksort when size is |
| 357 |
// smaller than PRESORT_THRESH |
| 358 |
|
| 359 |
void _BSort::quicksort3d(int lo, int hi, int depth) |
| 360 |
{ |
| 361 |
/* Initialize stack */ |
| 362 |
int slo[QUICKSORT_STACK]; |
| 363 |
int shi[QUICKSORT_STACK]; |
| 364 |
int sd[QUICKSORT_STACK]; |
| 365 |
int sp = 1; |
| 366 |
slo[0] = lo; |
| 367 |
shi[0] = hi; |
| 368 |
sd[0] = depth; |
| 369 |
// Recursion elimination loop |
| 370 |
while (--sp>=0) |
| 371 |
{ |
| 372 |
lo = slo[sp]; |
| 373 |
hi = shi[sp]; |
| 374 |
depth = sd[sp]; |
| 375 |
// Test for insertion sort |
| 376 |
if (depth >= PRESORT_DEPTH) |
| 377 |
{ |
| 378 |
for (int i=lo; i<=hi; i++) |
| 379 |
rank[posn[i]] = hi; |
| 380 |
} |
| 381 |
else if (hi-lo<PRESORT_THRESH) |
| 382 |
{ |
| 383 |
int i,j; |
| 384 |
for (i=lo+1; i<=hi; i++) |
| 385 |
{ |
| 386 |
int tmp = posn[i]; |
| 387 |
for(j=i-1; j>=lo && GTD(posn[j], tmp, depth); j--) |
| 388 |
posn[j+1] = posn[j]; |
| 389 |
posn[j+1] = tmp; |
| 390 |
} |
| 391 |
for(i=hi;i>=lo;i=j) |
| 392 |
{ |
| 393 |
int tmp = posn[i]; |
| 394 |
rank[tmp] = i; |
| 395 |
for (j=i-1; j>=lo && !GTD(tmp,posn[j],depth); j--) |
| 396 |
rank[posn[j]] = i; |
| 397 |
} |
| 398 |
} |
| 399 |
else |
| 400 |
{ |
| 401 |
int tmp; |
| 402 |
unsigned char *dd=data+depth; |
| 403 |
unsigned char med = pivot3d(dd,lo,hi); |
| 404 |
// -- positions are organized as follows: |
| 405 |
// [lo..l1[ [l1..l[ ]h..h1] ]h1..hi] |
| 406 |
// = < > = |
| 407 |
int l1 = lo; |
| 408 |
int h1 = hi; |
| 409 |
while (dd[posn[l1]]==med && l1<h1) { l1++; } |
| 410 |
while (dd[posn[h1]]==med && l1<h1) { h1--; } |
| 411 |
int l = l1; |
| 412 |
int h = h1; |
| 413 |
// -- partition set |
| 414 |
for (;;) |
| 415 |
{ |
| 416 |
while (l<=h) |
| 417 |
{ |
| 418 |
int c = (int)dd[posn[l]] - (int)med; |
| 419 |
if (c > 0) break; |
| 420 |
if (c == 0) { tmp=posn[l]; posn[l]=posn[l1]; posn[l1++]=tmp; } |
| 421 |
l++; |
| 422 |
} |
| 423 |
while (l<=h) |
| 424 |
{ |
| 425 |
int c = (int)dd[posn[h]] - (int)med; |
| 426 |
if (c < 0) break; |
| 427 |
if (c == 0) { tmp=posn[h]; posn[h]=posn[h1]; posn[h1--]=tmp; } |
| 428 |
h--; |
| 429 |
} |
| 430 |
if (l>h) break; |
| 431 |
tmp=posn[l]; posn[l]=posn[h]; posn[h]=tmp; |
| 432 |
} |
| 433 |
// -- reorganize as follows |
| 434 |
// [lo..l1[ [l1..h1] ]h1..hi] |
| 435 |
// < = > |
| 436 |
tmp = mini(l1-lo, l-l1); |
| 437 |
vswap(lo, l-tmp, tmp, posn); |
| 438 |
l1 = lo + (l-l1); |
| 439 |
tmp = mini(hi-h1, h1-h); |
| 440 |
vswap(hi-tmp+1, h+1, tmp, posn); |
| 441 |
h1 = hi - (h1-h); |
| 442 |
// -- process segments |
| 443 |
assert(sp+3<QUICKSORT_STACK); |
| 444 |
// ----- middle segment (=?) [l1, h1] |
| 445 |
l = l1; h = h1; |
| 446 |
if (med==0) // special case for marker [slow] |
| 447 |
for (int i=l; i<=h; i++) |
| 448 |
if ((int)posn[i]+depth == size-1) |
| 449 |
{ |
| 450 |
tmp=posn[i]; posn[i]=posn[l]; posn[l]=tmp; |
| 451 |
rank[tmp]=l++; break; |
| 452 |
} |
| 453 |
if (l<h) |
| 454 |
{ |
| 455 |
slo[sp] = l; shi[sp] = h; sd[sp++] = depth+1; |
| 456 |
} |
| 457 |
else if (l==h) |
| 458 |
{ |
| 459 |
rank[posn[h]] = h; |
| 460 |
} |
| 461 |
// ----- lower segment (<) [lo, l1[ |
| 462 |
l = lo; |
| 463 |
h = l1-1; |
| 464 |
if (l<h) |
| 465 |
{ |
| 466 |
slo[sp] = l; shi[sp] = h; sd[sp++] = depth; |
| 467 |
} |
| 468 |
else if (l==h) |
| 469 |
{ |
| 470 |
rank[posn[h]] = h; |
| 471 |
} |
| 472 |
// ----- upper segment (>) ]h1, hi] |
| 473 |
l = h1+1; |
| 474 |
h = hi; |
| 475 |
if (l<h) |
| 476 |
{ |
| 477 |
slo[sp] = l; shi[sp] = h; sd[sp++] = depth; |
| 478 |
} |
| 479 |
else if (l==h) |
| 480 |
{ |
| 481 |
rank[posn[h]] = h; |
| 482 |
} |
| 483 |
} |
| 484 |
} |
| 485 |
} |
| 486 |
|
| 487 |
|
| 488 |
|
| 489 |
|
| 490 |
// _BSort::radixsort8 -- 8 bit radix sort |
| 491 |
|
| 492 |
void _BSort::radixsort8(void) |
| 493 |
{ |
| 494 |
int i; |
| 495 |
// Initialize frequency array |
| 496 |
int lo[256], hi[256]; |
| 497 |
for (i=0; i<256; i++) |
| 498 |
hi[i] = lo[i] = 0; |
| 499 |
// Count occurences |
| 500 |
for (i=0; i<size-1; i++) |
| 501 |
hi[data[i]] ++; |
| 502 |
// Compute positions (lo) |
| 503 |
int last = 1; |
| 504 |
for (i=0; i<256; i++) |
| 505 |
{ |
| 506 |
lo[i] = last; |
| 507 |
hi[i] = last + hi[i] - 1; |
| 508 |
last = hi[i] + 1; |
| 509 |
} |
| 510 |
for (i=0; i<size-1; i++) |
| 511 |
{ |
| 512 |
posn[ lo[data[i]]++ ] = i; |
| 513 |
rank[ i ] = hi[data[i]]; |
| 514 |
} |
| 515 |
// Process marker "$" |
| 516 |
posn[0] = size-1; |
| 517 |
rank[size-1] = 0; |
| 518 |
// Extra element |
| 519 |
rank[size] = -1; |
| 520 |
} |
| 521 |
|
| 522 |
|
| 523 |
// _BSort::radixsort16 -- 16 bit radix sort |
| 524 |
|
| 525 |
void _BSort::radixsort16(void) |
| 526 |
{ |
| 527 |
int i; |
| 528 |
// Initialize frequency array |
| 529 |
int *ftab = (int *) calloc(65536, sizeof(int)); |
| 530 |
for (i=0; i<65536; i++) |
| 531 |
ftab[i] = 0; |
| 532 |
// Count occurences |
| 533 |
unsigned char c1 = data[0]; |
| 534 |
for (i=0; i<size-1; i++) |
| 535 |
{ |
| 536 |
unsigned char c2 = data[i+1]; |
| 537 |
ftab[(c1<<8)|c2] ++; |
| 538 |
c1 = c2; |
| 539 |
} |
| 540 |
// Generate upper position |
| 541 |
for (i=1;i<65536;i++) |
| 542 |
ftab[i] += ftab[i-1]; |
| 543 |
// Fill rank array with upper bound |
| 544 |
c1 = data[0]; |
| 545 |
for (i=0; i<size-2; i++) |
| 546 |
{ |
| 547 |
unsigned char c2 = data[i+1]; |
| 548 |
rank[i] = ftab[(c1<<8)|c2]; |
| 549 |
c1 = c2; |
| 550 |
} |
| 551 |
// Fill posn array (backwards) |
| 552 |
c1 = data[size-2]; |
| 553 |
for (i=size-3; i>=0; i--) |
| 554 |
{ |
| 555 |
unsigned char c2 = data[i]; |
| 556 |
posn[ ftab[(c2<<8)|c1]-- ] = i; |
| 557 |
c1 = c2; |
| 558 |
} |
| 559 |
// Fixup marker stuff |
| 560 |
assert(data[size-1]==0); |
| 561 |
c1 = data[size-2]; |
| 562 |
posn[0] = size-1; |
| 563 |
posn[ ftab[(c1<<8)] ] = size-2; |
| 564 |
rank[size-1] = 0; |
| 565 |
rank[size-2] = ftab[(c1<<8)]; |
| 566 |
// Extra element |
| 567 |
rank[size] = -1; |
| 568 |
} |
| 569 |
|
| 570 |
|
| 571 |
|
| 572 |
// _BSort::run -- main sort loop |
| 573 |
|
| 574 |
void _BSort::run(int &markerpos) |
| 575 |
{ |
| 576 |
int lo, hi; |
| 577 |
assert(size>0); |
| 578 |
assert(data[size-1]==0); |
| 579 |
// Step 1: Radix sort |
| 580 |
int depth = 0; |
| 581 |
if (size > RADIX_THRESH) |
| 582 |
{ |
| 583 |
radixsort16(); |
| 584 |
depth=2; |
| 585 |
} |
| 586 |
else |
| 587 |
{ |
| 588 |
radixsort8(); |
| 589 |
depth=1; |
| 590 |
} |
| 591 |
// Step 2: Perform presort to depth PRESORT_DEPTH |
| 592 |
for (lo=0; lo<size; lo++) |
| 593 |
{ |
| 594 |
hi = rank[posn[lo]]; |
| 595 |
if (lo < hi) |
| 596 |
quicksort3d(lo, hi, depth); |
| 597 |
lo = hi; |
| 598 |
} |
| 599 |
depth = PRESORT_DEPTH; |
| 600 |
// Step 3: Perform rank doubling |
| 601 |
int again = 1; |
| 602 |
while (again) |
| 603 |
{ |
| 604 |
again = 0; |
| 605 |
int sorted_lo = 0; |
| 606 |
for (lo=0; lo<size; lo++) |
| 607 |
{ |
| 608 |
hi = rank[posn[lo]&0xffffff]; |
| 609 |
if (lo == hi) |
| 610 |
{ |
| 611 |
lo += (posn[lo]>>24) & 0xff; |
| 612 |
} |
| 613 |
else |
| 614 |
{ |
| 615 |
if (hi-lo < RANKSORT_THRESH) |
| 616 |
{ |
| 617 |
ranksort(lo, hi, depth); |
| 618 |
} |
| 619 |
else |
| 620 |
{ |
| 621 |
again += 1; |
| 622 |
while (sorted_lo < lo-1) |
| 623 |
{ |
| 624 |
int step = mini(255, lo-1-sorted_lo); |
| 625 |
posn[sorted_lo] = (posn[sorted_lo]&0xffffff) | (step<<24); |
| 626 |
sorted_lo += step+1; |
| 627 |
} |
| 628 |
quicksort3r(lo, hi, depth); |
| 629 |
sorted_lo = hi + 1; |
| 630 |
} |
| 631 |
lo = hi; |
| 632 |
} |
| 633 |
} |
| 634 |
// Finish threading |
| 635 |
while (sorted_lo < lo-1) |
| 636 |
{ |
| 637 |
int step = mini(255, lo-1-sorted_lo); |
| 638 |
posn[sorted_lo] = (posn[sorted_lo]&0xffffff) | (step<<24); |
| 639 |
sorted_lo += step+1; |
| 640 |
} |
| 641 |
// Double depth |
| 642 |
depth += depth; |
| 643 |
} |
| 644 |
// Step 4: Permute data |
| 645 |
int i; |
| 646 |
markerpos = -1; |
| 647 |
for (i=0; i<size; i++) |
| 648 |
rank[i] = data[i]; |
| 649 |
for (i=0; i<size; i++) |
| 650 |
{ |
| 651 |
int j = posn[i] & 0xffffff; |
| 652 |
if (j>0) |
| 653 |
{ |
| 654 |
data[i] = rank[j-1]; |
| 655 |
} |
| 656 |
else |
| 657 |
{ |
| 658 |
data[i] = 0; |
| 659 |
markerpos = i; |
| 660 |
} |
| 661 |
} |
| 662 |
assert(markerpos>=0 && markerpos<size); |
| 663 |
} |
| 664 |
|
| 665 |
|
| 666 |
// ======================================== |
| 667 |
// -- Encoding |
| 668 |
|
| 669 |
static void encode_raw(ZPEncoder &zp, int bits, int x) |
| 670 |
{ |
| 671 |
int n = 1; |
| 672 |
int m = (1<<bits); |
| 673 |
while (n < m) |
| 674 |
{ |
| 675 |
x = (x & (m-1)) << 1; |
| 676 |
int b = (x >> bits); |
| 677 |
zp.encode_without_context(b); |
| 678 |
n = (n<<1) | b; |
| 679 |
} |
| 680 |
} |
| 681 |
|
| 682 |
static void encode_binary(ZPEncoder &zp, ZPBitContext *ctx, int bits, int x) |
| 683 |
{ |
| 684 |
// Require 2^bits-1 contexts |
| 685 |
int n = 1; |
| 686 |
int m = (1<<bits); |
| 687 |
ctx = ctx - 1; |
| 688 |
while (n < m) |
| 689 |
{ |
| 690 |
x = (x & (m-1)) << 1; |
| 691 |
int b = (x >> bits); |
| 692 |
zp.encode(b, ctx[n]); |
| 693 |
n = (n<<1) | b; |
| 694 |
} |
| 695 |
} |
| 696 |
|
| 697 |
unsigned int BSEncoder::encode() |
| 698 |
{ |
| 699 |
///////////////////////////////// |
| 700 |
//////////// Block Sort Tranform |
| 701 |
|
| 702 |
int markerpos = size-1; |
| 703 |
blocksort(data,size,markerpos); |
| 704 |
|
| 705 |
///////////////////////////////// |
| 706 |
//////////// Encode Output Stream |
| 707 |
|
| 708 |
// Header |
| 709 |
ZPEncoder &zp = gzp; |
| 710 |
encode_raw(gzp, 24, size); |
| 711 |
// Determine and Encode Estimation Speed |
| 712 |
int fshift = 0; |
| 713 |
if (size < FREQS0) |
| 714 |
{ |
| 715 |
fshift=0; zp.encode_without_context(0); |
| 716 |
} |
| 717 |
else if (size < FREQS1) |
| 718 |
{ |
| 719 |
fshift = 1; |
| 720 |
zp.encode_without_context(1); |
| 721 |
zp.encode_without_context(0); |
| 722 |
} |
| 723 |
else |
| 724 |
{ |
| 725 |
fshift = 2; |
| 726 |
zp.encode_without_context(1); |
| 727 |
zp.encode_without_context(1); |
| 728 |
} |
| 729 |
// MTF |
| 730 |
unsigned char mtf[256]; |
| 731 |
unsigned char rmtf[256]; |
| 732 |
unsigned int freq[FREQMAX]; |
| 733 |
int m = 0; |
| 734 |
for (m=0; m<256; m++) |
| 735 |
mtf[m] = m; |
| 736 |
for (m=0; m<256; m++) |
| 737 |
rmtf[mtf[m]] = m; |
| 738 |
int fadd = 4; |
| 739 |
for (m=0; m<FREQMAX; m++) |
| 740 |
freq[m] = 0; |
| 741 |
// Encode |
| 742 |
int i; |
| 743 |
int mtfno = 3; |
| 744 |
for (i=0; i<size; i++) |
| 745 |
{ |
| 746 |
// Get MTF data |
| 747 |
int c = data[i]; |
| 748 |
int ctxid = CTXIDS-1; |
| 749 |
if (ctxid>mtfno) ctxid=mtfno; |
| 750 |
mtfno = rmtf[c]; |
| 751 |
if (i==markerpos) |
| 752 |
mtfno = 256; |
| 753 |
// Encode using ZPCodec |
| 754 |
int b; |
| 755 |
ZPBitContext *cx = ctx; |
| 756 |
b = (mtfno==0); |
| 757 |
zp.encode(b, cx[ctxid]); |
| 758 |
if (b) goto rotate; cx+=CTXIDS; |
| 759 |
b = (mtfno==1); |
| 760 |
zp.encode(b, cx[ctxid]); |
| 761 |
if (b) goto rotate; cx+=CTXIDS; |
| 762 |
b = (mtfno<4); |
| 763 |
zp.encode(b, cx[0]); |
| 764 |
if (b) { encode_binary(zp,cx+1,1,mtfno-2); goto rotate; } |
| 765 |
cx+=1+1; |
| 766 |
b = (mtfno<8); |
| 767 |
zp.encode(b, cx[0]); |
| 768 |
if (b) { encode_binary(zp,cx+1,2,mtfno-4); goto rotate; } |
| 769 |
cx+=1+3; |
| 770 |
b = (mtfno<16); |
| 771 |
zp.encode(b, cx[0]); |
| 772 |
if (b) { encode_binary(zp,cx+1,3,mtfno-8); goto rotate; } |
| 773 |
cx+=1+7; |
| 774 |
b = (mtfno<32); |
| 775 |
zp.encode(b, cx[0]); |
| 776 |
if (b) { encode_binary(zp,cx+1,4,mtfno-16); goto rotate; } |
| 777 |
cx+=1+15; |
| 778 |
b = (mtfno<64); |
| 779 |
zp.encode(b, cx[0]); |
| 780 |
if (b) { encode_binary(zp,cx+1,5,mtfno-32); goto rotate; } |
| 781 |
cx+=1+31; |
| 782 |
b = (mtfno<128); |
| 783 |
zp.encode(b, cx[0]); |
| 784 |
if (b) { encode_binary(zp,cx+1,6,mtfno-64); goto rotate; } |
| 785 |
cx+=1+63; |
| 786 |
b = (mtfno<256); |
| 787 |
zp.encode(b, cx[0]); |
| 788 |
if (b) { encode_binary(zp,cx+1,7,mtfno-128); goto rotate; } |
| 789 |
continue; |
| 790 |
// Rotate MTF according to empirical frequencies (new!) |
| 791 |
rotate: |
| 792 |
// Adjust frequencies for overflow |
| 793 |
fadd = fadd + (fadd>>fshift); |
| 794 |
if (fadd > 0x10000000) |
| 795 |
{ |
| 796 |
fadd = fadd>>24; |
| 797 |
freq[0] >>= 24; |
| 798 |
freq[1] >>= 24; |
| 799 |
freq[2] >>= 24; |
| 800 |
freq[3] >>= 24; |
| 801 |
for (int k=4; k<FREQMAX; k++) |
| 802 |
freq[k] = freq[k]>>24; |
| 803 |
} |
| 804 |
// Relocate new char according to new freq |
| 805 |
unsigned int fc = fadd; |
| 806 |
if (mtfno < FREQMAX) |
| 807 |
fc += freq[mtfno]; |
| 808 |
int k; |
| 809 |
for (k=mtfno; k>=FREQMAX; k--) |
| 810 |
{ |
| 811 |
mtf[k] = mtf[k-1]; |
| 812 |
rmtf[mtf[k]] = k; |
| 813 |
} |
| 814 |
for (; k>0 && fc>=freq[k-1]; k--) |
| 815 |
{ |
| 816 |
mtf[k] = mtf[k-1]; |
| 817 |
freq[k] = freq[k-1]; |
| 818 |
rmtf[mtf[k]] = k; |
| 819 |
} |
| 820 |
mtf[k] = c; |
| 821 |
freq[k] = fc; |
| 822 |
rmtf[mtf[k]] = k; |
| 823 |
} |
| 824 |
// Terminate |
| 825 |
return 0; |
| 826 |
} |
| 827 |
|
| 828 |
// ======================================== |
| 829 |
// --- Construction |
| 830 |
|
| 831 |
BSEncoder::BSEncoder(FILE *f,const int xencoding) |
| 832 |
: data(NULL), gzp(f) |
| 833 |
{ |
| 834 |
const int encoding=(xencoding<MINBLOCK)?MINBLOCK:xencoding; |
| 835 |
// Record block size |
| 836 |
blocksize = encoding * 1024; |
| 837 |
// Initialize context array |
| 838 |
memset(ctx, 0, sizeof(ctx)); |
| 839 |
} |
| 840 |
|
| 841 |
void BSEncoder::close() |
| 842 |
{ |
| 843 |
// Flush |
| 844 |
flush(); |
| 845 |
// Encode EOF marker |
| 846 |
encode_raw(gzp, 24, 0); |
| 847 |
// Free allocated memory |
| 848 |
free(data); |
| 849 |
data = NULL; |
| 850 |
gzp.close(); |
| 851 |
} |
| 852 |
|
| 853 |
BSEncoder::~BSEncoder() |
| 854 |
{ |
| 855 |
if (data) close(); |
| 856 |
} |
| 857 |
|
| 858 |
// ======================================== |
| 859 |
// -- ByteStream interface |
| 860 |
|
| 861 |
void BSEncoder::flush() |
| 862 |
{ |
| 863 |
if (bptr>0) |
| 864 |
{ |
| 865 |
assert(bptr<(int)blocksize); |
| 866 |
memset(data+bptr, 0, OVERFLOW); |
| 867 |
size = bptr+1; |
| 868 |
encode(); |
| 869 |
} |
| 870 |
size = bptr = 0; |
| 871 |
} |
| 872 |
|
| 873 |
size_t BSEncoder::write(void *buffer, size_t sz) |
| 874 |
{ |
| 875 |
// Trivial checks |
| 876 |
if (sz == 0) |
| 877 |
return 0; |
| 878 |
// Loop |
| 879 |
int copied = 0; |
| 880 |
while (sz > 0) |
| 881 |
{ |
| 882 |
// Initialize |
| 883 |
if (!data) |
| 884 |
{ |
| 885 |
bptr = 0; |
| 886 |
data = (unsigned char *) calloc(blocksize+OVERFLOW, sizeof(unsigned char)); |
| 887 |
} |
| 888 |
// Compute remaining |
| 889 |
int bytes = blocksize - 1 - bptr; |
| 890 |
if (bytes > (int)sz) |
| 891 |
bytes = sz; |
| 892 |
// Store date (todo: rle) |
| 893 |
memcpy(data+bptr, buffer, bytes); |
| 894 |
buffer = (void*)((char*)buffer + bytes); |
| 895 |
bptr += bytes; |
| 896 |
sz -= bytes; |
| 897 |
copied += bytes; |
| 898 |
offset += bytes; |
| 899 |
// Flush when needed |
| 900 |
if (bptr + 1 >= (int)blocksize) |
| 901 |
flush(); |
| 902 |
} |
| 903 |
// return |
| 904 |
return copied; |
| 905 |
} |
| 906 |
|
| 907 |
size_t BSEncoder::writall(void *buffer, size_t wsize) |
| 908 |
{ |
| 909 |
size_t total = 0; |
| 910 |
while (wsize > 0) |
| 911 |
{ |
| 912 |
size_t nitems = write(buffer, wsize); |
| 913 |
total += nitems; |
| 914 |
wsize -= nitems; |
| 915 |
buffer = (void*)((char*)buffer + nitems); |
| 916 |
} |
| 917 |
return total; |
| 918 |
} |
| 919 |
|
| 920 |
void BSEncoder::write8 (unsigned int card) |
| 921 |
{ |
| 922 |
unsigned char c[1]; |
| 923 |
c[0] = (card) & 0xff; |
| 924 |
write((void*)c, sizeof(c)); |
| 925 |
} |
| 926 |
|
| 927 |
void BSEncoder::write16(unsigned int card) |
| 928 |
{ |
| 929 |
unsigned char c[2]; |
| 930 |
c[0] = (card>>8) & 0xff; |
| 931 |
c[1] = (card) & 0xff; |
| 932 |
writall((void*)c, sizeof(c)); |
| 933 |
} |
| 934 |
|
| 935 |
void BSEncoder::write24(unsigned int card) |
| 936 |
{ |
| 937 |
unsigned char c[3]; |
| 938 |
c[0] = (card>>16) & 0xff; |
| 939 |
c[1] = (card>>8) & 0xff; |
| 940 |
c[2] = (card) & 0xff; |
| 941 |
writall((void*)c, sizeof(c)); |
| 942 |
} |
| 943 |
|
| 944 |
void BSEncoder::write32(unsigned int card) |
| 945 |
{ |
| 946 |
unsigned char c[4]; |
| 947 |
c[0] = (card>>24) & 0xff; |
| 948 |
c[1] = (card>>16) & 0xff; |
| 949 |
c[2] = (card>>8) & 0xff; |
| 950 |
c[3] = (card) & 0xff; |
| 951 |
writall((void*)c, sizeof(c)); |
| 952 |
} |
| 953 |
|
| 954 |
|
| 955 |
|