spiceparser/sort.c

   1 /********************
   2     This file is part of the software library CADLIB written by Conrad Ziesler
   3     Copyright 2003, Conrad Ziesler, all rights reserved.
   4
   5 *************************
   6     This program is free software; you can redistribute it and/or modify
   7     it under the terms of the GNU General Public License as published by
   8     the Free Software Foundation; either version 2 of the License, or
   9     (at your option) any later version.
  10
  11     This program is distributed in the hope that it will be useful,
  12     but WITHOUT ANY WARRANTY; without even the implied warranty of
  13     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14     GNU General Public License for more details.
  15
  16     You should have received a copy of the GNU General Public License
  17     along with this program; if not, write to the Free Software
  18     Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  19
  20 ******************/
  21 /* sort.c, sorting routines
  22    Conrad Ziesler
  23
  24    this code is adapted from gnu glibc qsort.c, and so their header is as follows:
  25    reason: stdlib sort interface does not allow for indirection in cmp without using hackish global
  26    variables. also, we rely heavily on list already, so this makes the interface cleaner
  27 */
  28
  29
  30
  31 /* Copyright (C) 1991, 1992, 1996, 1997 Free Software Foundation, Inc.
  32    This file is part of the GNU C Library.
  33    Written by Douglas C. Schmidt (schmidt@ics.uci.edu).
  34
  35    The GNU C Library is free software; you can redistribute it and/or
  36    modify it under the terms of the GNU Library General Public License as
  37    published by the Free Software Foundation; either version 2 of the
  38    License, or (at your option) any later version.
  39
  40    The GNU C Library is distributed in the hope that it will be useful,
  41    but WITHOUT ANY WARRANTY; without even the implied warranty of
  42    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  43    Library General Public License for more details.
  44
  45    You should have received a copy of the GNU Library General Public
  46    License along with the GNU C Library; see the file COPYING.LIB.  If not,
  47    write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  48    Boston, MA 02111-1307, USA.  */
  49
  50
  51 #include "debug.h"
  52
  53 #include "list.h"
  54 #include "list_search.h"
  55
  56
  57 /* Byte-wise swap two items of size SIZE. */
  58 #define SWAP(a, b, size)                                                      \
  59   do                                                                          \
  60     {                                                                         \
  61       register size_t __size = (size);                                        \
  62       register char *__a = (a), *__b = (b);                                   \
  63       do                                                                      \
  64         {                                                                     \
  65           char __tmp = *__a;                                                  \
  66           *__a++ = *__b;                                                      \
  67           *__b++ = __tmp;                                                     \
  68         } while (--__size > 0);                                               \
  69     } while (0)
  70
  71 /* Discontinue quicksort algorithm when partition gets below this size.
  72    This particular magic number was chosen to work best on a Sun 4/260. */
  73 #define MAX_THRESH 4
  74
  75 /* Stack node declarations used to store unfulfilled partition obligations. */
  76 typedef struct
  77   {
  78     char *lo;
  79     char *hi;
  80   } stack_node;
  81
  82 /* The next 4 #defines implement a very fast in-line stack abstraction. */
  83 #define STACK_SIZE      (8 * sizeof(unsigned long int))
  84 #define PUSH(low, high) ((void) ((top->lo = (low)), (top->hi = (high)), ++top))
  85 #define POP(low, high)  ((void) (--top, (low = top->lo), (high = top->hi)))
  86 #define STACK_NOT_EMPTY (stack < top)
  87
  88
  89 /* Order size using quicksort.  This implementation incorporates
  90    four optimizations discussed in Sedgewick:
  91
  92    1. Non-recursive, using an explicit stack of pointer that store the
  93       next array partition to sort.  To save time, this maximum amount
  94       of space required to store an array of MAX_INT is allocated on the
  95       stack.  Assuming a 32-bit integer, this needs only 32 *
  96       sizeof(stack_node) == 136 bits.  Pretty cheap, actually.
  97
  98    2. Chose the pivot element using a median-of-three decision tree.
  99       This reduces the probability of selecting a bad pivot value and
 100       eliminates certain extraneous comparisons.
 101
 102    3. Only quicksorts TOTAL_ELEMS / MAX_THRESH partitions, leaving
 103       insertion sort to order the MAX_THRESH items within each partition.
 104       This is a big win, since insertion sort is faster for small, mostly
 105       sorted array segments.
 106
 107    4. The larger of the two sub-partitions is always pushed onto the
 108       stack first, with the algorithm then concentrating on the
 109       smaller partition.  This *guarantees* no more than log (n)
 110       stack size is needed (actually O(1) in this case)!  */
 111
 112
 113 #define CMP(a,b) (sf->cmpf(sf,a,b))
 114
 115
 116 void list_qsort (sort_func_t *sf)
 117 {
 118   size_t total_elems=sf->tosort->q;
 119   size_t size=sf->tosort->s;
 120   void *const pbase=sf->tosort->d;
 121   register char *base_ptr = (char *) pbase;
 122
 123   /* Allocating SIZE bytes for a pivot buffer facilitates a better
 124      algorithm below since we can do comparisons directly on the pivot. */
 125   char *pivot_buffer = (char *) malloc (size);
 126   const size_t max_thresh = MAX_THRESH * size;
 127   sf->aiskey=0;
 128   assert(pivot_buffer!=NULL);
 129   memset(pivot_buffer,0,size);
 130
 131   if (total_elems == 0)
 132     /* Avoid lossage with unsigned arithmetic below.  */
 133     { free(pivot_buffer); return; }
 134
 135   if (total_elems > MAX_THRESH)
 136     {
 137       char *lo = base_ptr;
 138       char *hi = &lo[size * (total_elems - 1)];
 139       /* Largest size needed for 32-bit int!!! */
 140       stack_node stack[STACK_SIZE];
 141       stack_node *top = stack + 1;
 142
 143       while (STACK_NOT_EMPTY)
 144         {
 145           char *left_ptr;
 146           char *right_ptr;
 147
 148           char *pivot = pivot_buffer;
 149
 150           /* Select median value from among LO, MID, and HI. Rearrange
 151              LO and HI so the three values are sorted. This lowers the
 152              probability of picking a pathological pivot value and
 153              skips a comparison for both the LEFT_PTR and RIGHT_PTR. */
 154
 155           char *mid = lo + size * ((hi - lo) / size >> 1);
 156
 157           if (CMP ((void *) mid, (void *) lo) < 0)
 158             SWAP (mid, lo, size);
 159           if (CMP ((void *) hi, (void *) mid) < 0)
 160             SWAP (mid, hi, size);
 161           else
 162             goto jump_over;
 163           if (CMP ((void *) mid, (void *) lo) < 0)
 164             SWAP (mid, lo, size);
 165         jump_over:;
 166           memcpy (pivot, mid, size);
 167           pivot = pivot_buffer;
 168
 169           left_ptr  = lo + size;
 170           right_ptr = hi - size;
 171
 172           /* Here's the famous ``collapse the walls'' section of quicksort.
 173              Gotta like those tight inner loops!  They are the main reason
 174              that this algorithm runs much faster than others. */
 175           do
 176             {
 177               while (CMP ((void *) left_ptr, (void *) pivot) < 0)
 178                 left_ptr += size;
 179
 180               while (CMP ((void *) pivot, (void *) right_ptr) < 0)
 181                 right_ptr -= size;
 182
 183               if (left_ptr < right_ptr)
 184                 {
 185                   SWAP (left_ptr, right_ptr, size);
 186                   left_ptr += size;
 187                   right_ptr -= size;
 188                 }
 189               else if (left_ptr == right_ptr)
 190                 {
 191                   left_ptr += size;
 192                   right_ptr -= size;
 193                   break;
 194                 }
 195             }
 196           while (left_ptr <= right_ptr);
 197
 198           /* Set up pointers for next iteration.  First determine whether
 199              left and right partitions are below the threshold size.  If so,
 200              ignore one or both.  Otherwise, push the larger partition's
 201              bounds on the stack and continue sorting the smaller one. */
 202
 203           if ((size_t) (right_ptr - lo) <= max_thresh)
 204             {
 205               if ((size_t) (hi - left_ptr) <= max_thresh)
 206                 /* Ignore both small partitions. */
 207                 POP (lo, hi);
 208               else
 209                 /* Ignore small left partition. */
 210                 lo = left_ptr;
 211             }
 212           else if ((size_t) (hi - left_ptr) <= max_thresh)
 213             /* Ignore small right partition. */
 214             hi = right_ptr;
 215           else if ((right_ptr - lo) > (hi - left_ptr))
 216             {
 217               /* Push larger left partition indices. */
 218               PUSH (lo, right_ptr);
 219               lo = left_ptr;
 220             }
 221           else
 222             {
 223               /* Push larger right partition indices. */
 224               PUSH (left_ptr, hi);
 225               hi = right_ptr;
 226             }
 227         }
 228     }
 229
 230   /* Once the BASE_PTR array is partially sorted by quicksort the rest
 231      is completely sorted using insertion sort, since this is efficient
 232      for partitions below MAX_THRESH size. BASE_PTR points to the beginning
 233      of the array to sort, and END_PTR points at the very last element in
 234      the array (*not* one beyond it!). */
 235
 236 #define min(x, y) ((x) < (y) ? (x) : (y))
 237
 238   {
 239     char *const end_ptr = &base_ptr[size * (total_elems - 1)];
 240     char *tmp_ptr = base_ptr;
 241     char *thresh = min(end_ptr, base_ptr + max_thresh);
 242     register char *run_ptr;
 243
 244     /* Find smallest element in first threshold and place it at the
 245        array's beginning.  This is the smallest array element,
 246        and the operation speeds up insertion sort's inner loop. */
 247
 248     for (run_ptr = tmp_ptr + size; run_ptr <= thresh; run_ptr += size)
 249       if (CMP ((void *) run_ptr, (void *) tmp_ptr) < 0)
 250         tmp_ptr = run_ptr;
 251
 252     if (tmp_ptr != base_ptr)
 253       SWAP (tmp_ptr, base_ptr, size);
 254
 255     /* Insertion sort, running from left-hand-side up to right-hand-side.  */
 256
 257     run_ptr = base_ptr + size;
 258     while ((run_ptr += size) <= end_ptr)
 259       {
 260         tmp_ptr = run_ptr - size;
 261         while (CMP ((void *) run_ptr, (void *) tmp_ptr) < 0)
 262           tmp_ptr -= size;
 263
 264         tmp_ptr += size;
 265         if (tmp_ptr != run_ptr)
 266           {
 267             char *trav;
 268
 269             trav = run_ptr + size;
 270             while (--trav >= run_ptr)
 271               {
 272                 char c = *trav;
 273                 char *hi, *lo;
 274
 275                 for (hi = lo = trav; (lo -= size) >= tmp_ptr; hi = lo)
 276                   *hi = *lo;
 277                 *hi = c;
 278               }
 279           }
 280       }
 281   }
 282     { free(pivot_buffer); return; }
 283 }
 284
 285
 286
 287
 288
 289
 290
 291
 292
 293 /**** support for search list bindings.  we copy and modify the qsort
 294       to reduce overhead  of extra bindings  -- conrad
 295 ***/
 296
 297
 298
 299 /* swap optimized for integers */
 300 #define SWAPINT(a, b)                                                         \
 301   do                                                                          \
 302     {                                                                         \
 303       register int *__a = (void *)(a), *__b = (void *)(b), __tmp;             \
 304       __tmp=*__a; *__a=*__b; *__a=__tmp;                                      \
 305     } while (0)
 306
 307
 308 #define CMPIND(a,b) (psp_cmpf( psp->user, source_d+(source_s*(*((int*)a))), source_d+(source_s*(*((int*)b)))))
 309 #define size (sizeof(int))
 310
 311 void list_search_qsort (list_search_t *lsp, list_psearch_t *psp)
 312 {
 313   char *source_d;
 314   int source_s;
 315   int (*psp_cmpf)(const void *user ,const void *a, const void *b);
 316   size_t total_elems=lsp->list.q;
 317
 318   void *const pbase=psp->isearchdata;
 319   register char *base_ptr;
 320   int pivot_buffer_el=0;
 321   char *pivot_buffer = (char *) &pivot_buffer_el;
 322   const size_t max_thresh = MAX_THRESH * size;
 323
 324   base_ptr= (char *) pbase;
 325   /* copy out of pointers to stack variables so compile can do memory scheduling */
 326   source_d=lsp->list.d;
 327   source_s=lsp->list.s;
 328   psp_cmpf=psp->cmpf;
 329
 330   if (total_elems == 0) return;
 331
 332   if (total_elems > MAX_THRESH)
 333     {
 334       char *lo = base_ptr;
 335       char *hi = &lo[size * (total_elems - 1)];
 336       /* Largest size needed for 32-bit int!!! */
 337       stack_node stack[STACK_SIZE];
 338       stack_node *top = stack + 1;
 339
 340       while (STACK_NOT_EMPTY)
 341         {
 342           char *left_ptr;
 343           char *right_ptr;
 344
 345           char *pivot = pivot_buffer;
 346
 347           /* Select median value from among LO, MID, and HI. Rearrange
 348              LO and HI so the three values are sorted. This lowers the
 349              probability of picking a pathological pivot value and
 350              skips a comparison for both the LEFT_PTR and RIGHT_PTR. */
 351
 352           char *mid = lo + size * ((hi - lo) / size >> 1);
 353
 354           assert(hi>=lo);
 355
 356           if (CMPIND ((void *) mid, (void *) lo) < 0)
 357             SWAPINT (mid, lo);
 358           if (CMPIND ((void *) hi, (void *) mid) < 0)
 359             SWAPINT (mid, hi);
 360           else
 361             goto jump_over;
 362           if (CMPIND ((void *) mid, (void *) lo) < 0)
 363             SWAPINT (mid, lo);
 364         jump_over:;
 365           (*((int*)pivot)) = (*((int*)mid));
 366           pivot = pivot_buffer;
 367
 368           left_ptr  = lo + size;
 369           right_ptr = hi - size;
 370
 371           /* Here's the famous ``collapse the walls'' section of quicksort.
 372              Gotta like those tight inner loops!  They are the main reason
 373              that this algorithm runs much faster than others. */
 374           do
 375             {
 376               while (CMPIND ((void *) left_ptr, (void *) pivot) < 0)
 377                 left_ptr += size;
 378
 379               while (CMPIND ((void *) pivot, (void *) right_ptr) < 0)
 380                 right_ptr -= size;
 381
 382               if (left_ptr < right_ptr)
 383                 {
 384                   SWAPINT (left_ptr, right_ptr);
 385                   left_ptr += size;
 386                   right_ptr -= size;
 387                 }
 388               else if (left_ptr == right_ptr)
 389                 {
 390                   left_ptr += size;
 391                   right_ptr -= size;
 392                   break;
 393                 }
 394             }
 395           while (left_ptr <= right_ptr);
 396
 397           /* Set up pointers for next iteration.  First determine whether
 398              left and right partitions are below the threshold size.  If so,
 399              ignore one or both.  Otherwise, push the larger partition's
 400              bounds on the stack and continue sorting the smaller one. */
 401
 402           if ((size_t) (right_ptr - lo) <= max_thresh)
 403             {
 404               if ((size_t) (hi - left_ptr) <= max_thresh)
 405                 /* Ignore both small partitions. */
 406                 POP (lo, hi);
 407               else
 408                 /* Ignore small left partition. */
 409                 lo = left_ptr;
 410             }
 411           else if ((size_t) (hi - left_ptr) <= max_thresh)
 412             /* Ignore small right partition. */
 413             hi = right_ptr;
 414           else if ((right_ptr - lo) > (hi - left_ptr))
 415             {
 416               /* Push larger left partition indices. */
 417               PUSH (lo, right_ptr);
 418               lo = left_ptr;
 419             }
 420           else
 421             {
 422               /* Push larger right partition indices. */
 423               PUSH (left_ptr, hi);
 424               hi = right_ptr;
 425             }
 426         }
 427     }
 428
 429   /* Once the BASE_PTR array is partially sorted by quicksort the rest
 430      is completely sorted using insertion sort, since this is efficient
 431      for partitions below MAX_THRESH size. BASE_PTR points to the beginning
 432      of the array to sort, and END_PTR points at the very last element in
 433      the array (*not* one beyond it!). */
 434
 435 #define min(x, y) ((x) < (y) ? (x) : (y))
 436
 437   {
 438     char *const end_ptr = &base_ptr[size * (total_elems - 1)];
 439     char *tmp_ptr = base_ptr;
 440     char *thresh = min(end_ptr, base_ptr + max_thresh);
 441     register char *run_ptr;
 442
 443     /* Find smallest element in first threshold and place it at the
 444        array's beginning.  This is the smallest array element,
 445        and the operation speeds up insertion sort's inner loop. */
 446
 447     for (run_ptr = tmp_ptr + size; run_ptr <= thresh; run_ptr += size)
 448       if (CMPIND ((void *) run_ptr, (void *) tmp_ptr) < 0)
 449         tmp_ptr = run_ptr;
 450
 451     if (tmp_ptr != base_ptr)
 452       SWAPINT (tmp_ptr, base_ptr);
 453
 454     /* Insertion sort, running from left-hand-side up to right-hand-side.  */
 455
 456     run_ptr = base_ptr + size;
 457     while ((run_ptr += size) <= end_ptr)
 458       {
 459         tmp_ptr = run_ptr - size;
 460         while (CMPIND ((void *) run_ptr, (void *) tmp_ptr) < 0)
 461           tmp_ptr -= size;
 462
 463         tmp_ptr += size;
 464         if (tmp_ptr != run_ptr)
 465           {
 466             char *trav;
 467
 468             trav = run_ptr + size;
 469             while (--trav >= run_ptr)
 470               {
 471                 char c = *trav;
 472                 char *hi, *lo;
 473
 474                 for (hi = lo = trav; (lo -= size) >= tmp_ptr; hi = lo)
 475                   *hi = *lo;
 476                 *hi = c;
 477               }
 478           }
 479       }
 480   }
 481     {  return; }
 482 }