| 1 | /* |
| 2 | Simple DirectMedia Layer |
| 3 | Copyright (C) 1997-2025 Sam Lantinga <slouken@libsdl.org> |
| 4 | |
| 5 | This software is provided 'as-is', without any express or implied |
| 6 | warranty. In no event will the authors be held liable for any damages |
| 7 | arising from the use of this software. |
| 8 | |
| 9 | Permission is granted to anyone to use this software for any purpose, |
| 10 | including commercial applications, and to alter it and redistribute it |
| 11 | freely, subject to the following restrictions: |
| 12 | |
| 13 | 1. The origin of this software must not be misrepresented; you must not |
| 14 | claim that you wrote the original software. If you use this software |
| 15 | in a product, an acknowledgment in the product documentation would be |
| 16 | appreciated but is not required. |
| 17 | 2. Altered source versions must be plainly marked as such, and must not be |
| 18 | misrepresented as being the original software. |
| 19 | 3. This notice may not be removed or altered from any source distribution. |
| 20 | */ |
| 21 | #include "SDL_internal.h" |
| 22 | |
| 23 | // SDL3 always uses its own internal qsort implementation, below, so |
| 24 | // it can guarantee stable sorts across platforms and not have to |
| 25 | // tapdance to support the various qsort_r interfaces, or bridge from |
| 26 | // the C runtime's non-SDLCALL compare functions. |
| 27 | |
| 28 | #ifdef assert |
| 29 | #undef assert |
| 30 | #endif |
| 31 | #define assert SDL_assert |
| 32 | #ifdef malloc |
| 33 | #undef malloc |
| 34 | #endif |
| 35 | #define malloc SDL_malloc |
| 36 | #ifdef free |
| 37 | #undef free |
| 38 | #endif |
| 39 | #define free SDL_free |
| 40 | #ifdef memcpy |
| 41 | #undef memcpy |
| 42 | #endif |
| 43 | #define memcpy SDL_memcpy |
| 44 | #ifdef memmove |
| 45 | #undef memmove |
| 46 | #endif |
| 47 | #define memmove SDL_memmove |
| 48 | |
| 49 | /* |
| 50 | This code came from Gareth McCaughan, under the zlib license. |
| 51 | Specifically this: https://www.mccaughan.org.uk/software/qsort.c-1.16 |
| 52 | |
| 53 | Everything below this comment until the HAVE_QSORT #endif was from Gareth |
| 54 | (any minor changes will be noted inline). |
| 55 | |
| 56 | Thank you to Gareth for relicensing this code under the zlib license for our |
| 57 | benefit! |
| 58 | |
| 59 | Update for SDL3: we have modified this from a qsort function to qsort_r. |
| 60 | |
| 61 | --ryan. |
| 62 | */ |
| 63 | |
| 64 | /* This is a drop-in replacement for the C library's |qsort()| routine. |
| 65 | * |
| 66 | * It is intended for use where you know or suspect that your |
| 67 | * platform's qsort is bad. If that isn't the case, then you |
| 68 | * should probably use the qsort your system gives you in preference |
| 69 | * to mine -- it will likely have been tested and tuned better. |
| 70 | * |
| 71 | * Features: |
| 72 | * - Median-of-three pivoting (and more) |
| 73 | * - Truncation and final polishing by a single insertion sort |
| 74 | * - Early truncation when no swaps needed in pivoting step |
| 75 | * - Explicit recursion, guaranteed not to overflow |
| 76 | * - A few little wrinkles stolen from the GNU |qsort()|. |
| 77 | * (For the avoidance of doubt, no code was stolen, only |
| 78 | * broad ideas.) |
| 79 | * - separate code for non-aligned / aligned / word-size objects |
| 80 | * |
| 81 | * Earlier releases of this code used an idiosyncratic licence |
| 82 | * I wrote myself, because I'm an idiot. The code is now released |
| 83 | * under the "zlib/libpng licence"; you will find the actual |
| 84 | * terms in the next comment. I request (but do not require) |
| 85 | * that if you make any changes beyond the name of the exported |
| 86 | * routine and reasonable tweaks to the TRUNC_* and |
| 87 | * PIVOT_THRESHOLD values, you modify the _ID string so as |
| 88 | * to make it clear that you have changed the code. |
| 89 | * |
| 90 | * If you find problems with this code, or find ways of |
| 91 | * making it significantly faster, please let me know! |
| 92 | * My e-mail address, valid as of early 2016 and for the |
| 93 | * foreseeable future, is |
| 94 | * gareth.mccaughan@pobox.com |
| 95 | * Thanks! |
| 96 | * |
| 97 | * Gareth McCaughan |
| 98 | */ |
| 99 | |
| 100 | /* Copyright (c) 1998-2021 Gareth McCaughan |
| 101 | * |
| 102 | * This software is provided 'as-is', without any express or implied |
| 103 | * warranty. In no event will the authors be held liable for any |
| 104 | * damages arising from the use of this software. |
| 105 | * |
| 106 | * Permission is granted to anyone to use this software for any purpose, |
| 107 | * including commercial applications, and to alter it and redistribute it |
| 108 | * freely, subject to the following restrictions: |
| 109 | * |
| 110 | * 1. The origin of this software must not be misrepresented; |
| 111 | * you must not claim that you wrote the original software. |
| 112 | * If you use this software in a product, an acknowledgment |
| 113 | * in the product documentation would be appreciated but |
| 114 | * is not required. |
| 115 | * |
| 116 | * 2. Altered source versions must be plainly marked as such, |
| 117 | * and must not be misrepresented as being the original software. |
| 118 | * |
| 119 | * 3. This notice may not be removed or altered from any source |
| 120 | * distribution. |
| 121 | */ |
| 122 | |
| 123 | /* Revision history since release: |
| 124 | * 1998-03-19 v1.12 First release I have any records of. |
| 125 | * 2007-09-02 v1.13 Fix bug kindly reported by Dan Bodoh |
| 126 | * (premature termination of recursion). |
| 127 | * Add a few clarifying comments. |
| 128 | * Minor improvements to debug output. |
| 129 | * 2016-02-21 v1.14 Replace licence with 2-clause BSD, |
| 130 | * and clarify a couple of things in |
| 131 | * comments. No code changes. |
| 132 | * 2016-03-10 v1.15 Fix bug kindly reported by Ryan Gordon |
| 133 | * (pre-insertion-sort messed up). |
| 134 | * Disable DEBUG_QSORT by default. |
| 135 | * Tweak comments very slightly. |
| 136 | * 2021-02-20 v1.16 Fix bug kindly reported by Ray Gardner |
| 137 | * (error in recursion leading to possible |
| 138 | * stack overflow). |
| 139 | * When checking alignment, avoid casting |
| 140 | * pointer to possibly-smaller integer. |
| 141 | */ |
| 142 | |
| 143 | /* BEGIN SDL CHANGE ... commented this out with an #if 0 block. --ryan. */ |
| 144 | #if 0 |
| 145 | #include <assert.h> |
| 146 | #include <stdint.h> |
| 147 | #include <stdlib.h> |
| 148 | #include <string.h> |
| 149 | |
| 150 | #undef DEBUG_QSORT |
| 151 | |
| 152 | static char _ID[]="<qsort.c gjm WITH CHANGES FOR SDL3 1.16 2021-02-20>" ; |
| 153 | #endif |
| 154 | /* END SDL CHANGE ... commented this out with an #if 0 block. --ryan. */ |
| 155 | |
| 156 | /* How many bytes are there per word? (Must be a power of 2, |
| 157 | * and must in fact equal sizeof(int).) |
| 158 | */ |
| 159 | #define WORD_BYTES sizeof(int) |
| 160 | |
| 161 | /* How big does our stack need to be? Answer: one entry per |
| 162 | * bit in a |size_t|. (Actually, a bit less because we don't |
| 163 | * recurse all the way down to size-1 subarrays.) |
| 164 | */ |
| 165 | #define STACK_SIZE (8*sizeof(size_t)) |
| 166 | |
| 167 | /* Different situations have slightly different requirements, |
| 168 | * and we make life epsilon easier by using different truncation |
| 169 | * points for the three different cases. |
| 170 | * So far, I have tuned TRUNC_words and guessed that the same |
| 171 | * value might work well for the other two cases. Of course |
| 172 | * what works well on my machine might work badly on yours. |
| 173 | */ |
| 174 | #define TRUNC_nonaligned 12 |
| 175 | #define TRUNC_aligned 12 |
| 176 | #define TRUNC_words 12*WORD_BYTES /* nb different meaning */ |
| 177 | |
| 178 | /* We use a simple pivoting algorithm for shortish sub-arrays |
| 179 | * and a more complicated one for larger ones. The threshold |
| 180 | * is PIVOT_THRESHOLD. |
| 181 | */ |
| 182 | #define PIVOT_THRESHOLD 40 |
| 183 | |
| 184 | typedef struct { char * first; char * last; } stack_entry; |
| 185 | #define pushLeft {stack[stacktop].first=ffirst;stack[stacktop++].last=last;} |
| 186 | #define pushRight {stack[stacktop].first=first;stack[stacktop++].last=llast;} |
| 187 | #define doLeft {first=ffirst;llast=last;continue;} |
| 188 | #define doRight {ffirst=first;last=llast;continue;} |
| 189 | #define pop {if (--stacktop<0) break;\ |
| 190 | first=ffirst=stack[stacktop].first;\ |
| 191 | last=llast=stack[stacktop].last;\ |
| 192 | continue;} |
| 193 | |
| 194 | /* Some comments on the implementation. |
| 195 | * 1. When we finish partitioning the array into "low" |
| 196 | * and "high", we forget entirely about short subarrays, |
| 197 | * because they'll be done later by insertion sort. |
| 198 | * Doing lots of little insertion sorts might be a win |
| 199 | * on large datasets for locality-of-reference reasons, |
| 200 | * but it makes the code much nastier and increases |
| 201 | * bookkeeping overhead. |
| 202 | * 2. We always save the longer and get to work on the |
| 203 | * shorter. This guarantees that whenever we push |
| 204 | * a k'th entry onto the stack we are about to get |
| 205 | * working on something of size <= N/2^k where N is |
| 206 | * the original array size; so the stack can't need |
| 207 | * more than log_2(max-array-size) entries. |
| 208 | * 3. We choose a pivot by looking at the first, last |
| 209 | * and middle elements. We arrange them into order |
| 210 | * because it's easy to do that in conjunction with |
| 211 | * choosing the pivot, and it makes things a little |
| 212 | * easier in the partitioning step. Anyway, the pivot |
| 213 | * is the middle of these three. It's still possible |
| 214 | * to construct datasets where the algorithm takes |
| 215 | * time of order n^2, but it simply never happens in |
| 216 | * practice. |
| 217 | * 3' Newsflash: On further investigation I find that |
| 218 | * it's easy to construct datasets where median-of-3 |
| 219 | * simply isn't good enough. So on large-ish subarrays |
| 220 | * we do a more sophisticated pivoting: we take three |
| 221 | * sets of 3 elements, find their medians, and then |
| 222 | * take the median of those. |
| 223 | * 4. We copy the pivot element to a separate place |
| 224 | * because that way we can always do our comparisons |
| 225 | * directly against a pointer to that separate place, |
| 226 | * and don't have to wonder "did we move the pivot |
| 227 | * element?". This makes the inner loop better. |
| 228 | * 5. It's possible to make the pivoting even more |
| 229 | * reliable by looking at more candidates when n |
| 230 | * is larger. (Taking this to its logical conclusion |
| 231 | * results in a variant of quicksort that doesn't |
| 232 | * have that n^2 worst case.) However, the overhead |
| 233 | * from the extra bookkeeping means that it's just |
| 234 | * not worth while. |
| 235 | * 6. This is pretty clean and portable code. Here are |
| 236 | * all the potential portability pitfalls and problems |
| 237 | * I know of: |
| 238 | * - In one place (the insertion sort) I construct |
| 239 | * a pointer that points just past the end of the |
| 240 | * supplied array, and assume that (a) it won't |
| 241 | * compare equal to any pointer within the array, |
| 242 | * and (b) it will compare equal to a pointer |
| 243 | * obtained by stepping off the end of the array. |
| 244 | * These might fail on some segmented architectures. |
| 245 | * - I assume that there are 8 bits in a |char| when |
| 246 | * computing the size of stack needed. This would |
| 247 | * fail on machines with 9-bit or 16-bit bytes. |
| 248 | * - I assume that if |((int)base&(sizeof(int)-1))==0| |
| 249 | * and |(size&(sizeof(int)-1))==0| then it's safe to |
| 250 | * get at array elements via |int*|s, and that if |
| 251 | * actually |size==sizeof(int)| as well then it's |
| 252 | * safe to treat the elements as |int|s. This might |
| 253 | * fail on systems that convert pointers to integers |
| 254 | * in non-standard ways. |
| 255 | * - I assume that |8*sizeof(size_t)<=INT_MAX|. This |
| 256 | * would be false on a machine with 8-bit |char|s, |
| 257 | * 16-bit |int|s and 4096-bit |size_t|s. :-) |
| 258 | */ |
| 259 | |
| 260 | /* The recursion logic is the same in each case. |
| 261 | * We keep chopping up until we reach subarrays of size |
| 262 | * strictly less than Trunc; we leave these unsorted. */ |
| 263 | #define Recurse(Trunc) \ |
| 264 | { size_t l=last-ffirst,r=llast-first; \ |
| 265 | if (l<Trunc) { \ |
| 266 | if (r>=Trunc) doRight \ |
| 267 | else pop \ |
| 268 | } \ |
| 269 | else if (l<=r) { pushRight; doLeft } \ |
| 270 | else if (r>=Trunc) { pushLeft; doRight }\ |
| 271 | else doLeft \ |
| 272 | } |
| 273 | |
| 274 | /* and so is the pivoting logic (note: last is inclusive): */ |
| 275 | #define Pivot(swapper,sz) \ |
| 276 | if ((size_t)(last-first)>PIVOT_THRESHOLD*sz) mid=pivot_big(first,mid,last,sz,compare,userdata);\ |
| 277 | else { \ |
| 278 | if (compare(userdata,first,mid)<0) { \ |
| 279 | if (compare(userdata,mid,last)>0) { \ |
| 280 | swapper(mid,last); \ |
| 281 | if (compare(userdata,first,mid)>0) swapper(first,mid);\ |
| 282 | } \ |
| 283 | } \ |
| 284 | else { \ |
| 285 | if (compare(userdata,mid,last)>0) swapper(first,last)\ |
| 286 | else { \ |
| 287 | swapper(first,mid); \ |
| 288 | if (compare(userdata,mid,last)>0) swapper(mid,last);\ |
| 289 | } \ |
| 290 | } \ |
| 291 | first+=sz; last-=sz; \ |
| 292 | } |
| 293 | |
| 294 | #ifdef DEBUG_QSORT |
| 295 | #include <stdio.h> |
| 296 | #endif |
| 297 | |
| 298 | /* and so is the partitioning logic: */ |
| 299 | #define Partition(swapper,sz) { \ |
| 300 | do { \ |
| 301 | while (compare(userdata,first,pivot)<0) first+=sz; \ |
| 302 | while (compare(userdata,pivot,last)<0) last-=sz; \ |
| 303 | if (first<last) { \ |
| 304 | swapper(first,last); \ |
| 305 | first+=sz; last-=sz; } \ |
| 306 | else if (first==last) { first+=sz; last-=sz; break; }\ |
| 307 | } while (first<=last); \ |
| 308 | } |
| 309 | |
| 310 | /* and so is the pre-insertion-sort operation of putting |
| 311 | * the smallest element into place as a sentinel. |
| 312 | * Doing this makes the inner loop nicer. I got this |
| 313 | * idea from the GNU implementation of qsort(). |
| 314 | * We find the smallest element from the first |nmemb|, |
| 315 | * or the first |limit|, whichever is smaller; |
| 316 | * therefore we must have ensured that the globally smallest |
| 317 | * element is in the first |limit| (because our |
| 318 | * quicksort recursion bottoms out only once we |
| 319 | * reach subarrays smaller than |limit|). |
| 320 | */ |
| 321 | #define PreInsertion(swapper,limit,sz) \ |
| 322 | first=base; \ |
| 323 | last=first + ((nmemb>limit ? limit : nmemb)-1)*sz;\ |
| 324 | while (last!=base) { \ |
| 325 | if (compare(userdata,first,last)>0) first=last; \ |
| 326 | last-=sz; } \ |
| 327 | if (first!=base) swapper(first,(char*)base); |
| 328 | |
| 329 | /* and so is the insertion sort, in the first two cases: */ |
| 330 | #define Insertion(swapper) \ |
| 331 | last=((char*)base)+nmemb*size; \ |
| 332 | for (first=((char*)base)+size;first!=last;first+=size) { \ |
| 333 | char *test; \ |
| 334 | /* Find the right place for |first|. \ |
| 335 | * My apologies for var reuse. */ \ |
| 336 | for (test=first-size;compare(userdata,test,first)>0;test-=size) ; \ |
| 337 | test+=size; \ |
| 338 | if (test!=first) { \ |
| 339 | /* Shift everything in [test,first) \ |
| 340 | * up by one, and place |first| \ |
| 341 | * where |test| is. */ \ |
| 342 | memcpy(pivot,first,size); \ |
| 343 | memmove(test+size,test,first-test); \ |
| 344 | memcpy(test,pivot,size); \ |
| 345 | } \ |
| 346 | } |
| 347 | |
| 348 | #define SWAP_nonaligned(a,b) { \ |
| 349 | register char *aa=(a),*bb=(b); \ |
| 350 | register size_t sz=size; \ |
| 351 | do { register char t=*aa; *aa++=*bb; *bb++=t; } while (--sz); } |
| 352 | |
| 353 | #define SWAP_aligned(a,b) { \ |
| 354 | register int *aa=(int*)(a),*bb=(int*)(b); \ |
| 355 | register size_t sz=size; \ |
| 356 | do { register int t=*aa;*aa++=*bb; *bb++=t; } while (sz-=WORD_BYTES); } |
| 357 | |
| 358 | #define SWAP_words(a,b) { \ |
| 359 | register int t=*((int*)a); *((int*)a)=*((int*)b); *((int*)b)=t; } |
| 360 | |
| 361 | /* ---------------------------------------------------------------------- */ |
| 362 | |
| 363 | static char * pivot_big(char *first, char *mid, char *last, size_t size, |
| 364 | int (SDLCALL *compare)(void *, const void *, const void *), void *userdata) { |
| 365 | size_t d=(((last-first)/size)>>3)*size; |
| 366 | #ifdef DEBUG_QSORT |
| 367 | fprintf(stderr, "pivot_big: first=%p last=%p size=%lu n=%lu\n" , first, (unsigned long)last, size, (unsigned long)((last-first+1)/size)); |
| 368 | #endif |
| 369 | char *m1,*m2,*m3; |
| 370 | { char *a=first, *b=first+d, *c=first+2*d; |
| 371 | #ifdef DEBUG_QSORT |
| 372 | fprintf(stderr,"< %d %d %d @ %p %p %p\n" ,*(int*)a,*(int*)b,*(int*)c, a,b,c); |
| 373 | #endif |
| 374 | m1 = compare(userdata,a,b)<0 ? |
| 375 | (compare(userdata,b,c)<0 ? b : (compare(userdata,a,c)<0 ? c : a)) |
| 376 | : (compare(userdata,a,c)<0 ? a : (compare(userdata,b,c)<0 ? c : b)); |
| 377 | } |
| 378 | { char *a=mid-d, *b=mid, *c=mid+d; |
| 379 | #ifdef DEBUG_QSORT |
| 380 | fprintf(stderr,". %d %d %d @ %p %p %p\n" ,*(int*)a,*(int*)b,*(int*)c, a,b,c); |
| 381 | #endif |
| 382 | m2 = compare(userdata,a,b)<0 ? |
| 383 | (compare(userdata,b,c)<0 ? b : (compare(userdata,a,c)<0 ? c : a)) |
| 384 | : (compare(userdata,a,c)<0 ? a : (compare(userdata,b,c)<0 ? c : b)); |
| 385 | } |
| 386 | { char *a=last-2*d, *b=last-d, *c=last; |
| 387 | #ifdef DEBUG_QSORT |
| 388 | fprintf(stderr,"> %d %d %d @ %p %p %p\n" ,*(int*)a,*(int*)b,*(int*)c, a,b,c); |
| 389 | #endif |
| 390 | m3 = compare(userdata,a,b)<0 ? |
| 391 | (compare(userdata,b,c)<0 ? b : (compare(userdata,a,c)<0 ? c : a)) |
| 392 | : (compare(userdata,a,c)<0 ? a : (compare(userdata,b,c)<0 ? c : b)); |
| 393 | } |
| 394 | #ifdef DEBUG_QSORT |
| 395 | fprintf(stderr,"-> %d %d %d @ %p %p %p\n" ,*(int*)m1,*(int*)m2,*(int*)m3, m1,m2,m3); |
| 396 | #endif |
| 397 | return compare(userdata,m1,m2)<0 ? |
| 398 | (compare(userdata,m2,m3)<0 ? m2 : (compare(userdata,m1,m3)<0 ? m3 : m1)) |
| 399 | : (compare(userdata,m1,m3)<0 ? m1 : (compare(userdata,m2,m3)<0 ? m3 : m2)); |
| 400 | } |
| 401 | |
| 402 | /* ---------------------------------------------------------------------- */ |
| 403 | |
| 404 | static void qsort_r_nonaligned(void *base, size_t nmemb, size_t size, |
| 405 | int (SDLCALL *compare)(void *, const void *, const void *), void *userdata) { |
| 406 | |
| 407 | stack_entry stack[STACK_SIZE]; |
| 408 | int stacktop=0; |
| 409 | char *first,*last; |
| 410 | char *pivot=malloc(size); |
| 411 | size_t trunc=TRUNC_nonaligned*size; |
| 412 | assert(pivot != NULL); |
| 413 | |
| 414 | first=(char*)base; last=first+(nmemb-1)*size; |
| 415 | |
| 416 | if ((size_t)(last-first)>=trunc) { |
| 417 | char *ffirst=first, *llast=last; |
| 418 | while (1) { |
| 419 | /* Select pivot */ |
| 420 | { char * mid=first+size*((last-first)/size >> 1); |
| 421 | Pivot(SWAP_nonaligned,size); |
| 422 | memcpy(pivot,mid,size); |
| 423 | } |
| 424 | /* Partition. */ |
| 425 | Partition(SWAP_nonaligned,size); |
| 426 | /* Prepare to recurse/iterate. */ |
| 427 | Recurse(trunc) |
| 428 | } |
| 429 | } |
| 430 | PreInsertion(SWAP_nonaligned,TRUNC_nonaligned,size); |
| 431 | Insertion(SWAP_nonaligned); |
| 432 | free(pivot); |
| 433 | } |
| 434 | |
| 435 | static void qsort_r_aligned(void *base, size_t nmemb, size_t size, |
| 436 | int (SDLCALL *compare)(void *,const void *, const void *), void *userdata) { |
| 437 | |
| 438 | stack_entry stack[STACK_SIZE]; |
| 439 | int stacktop=0; |
| 440 | char *first,*last; |
| 441 | char *pivot=malloc(size); |
| 442 | size_t trunc=TRUNC_aligned*size; |
| 443 | assert(pivot != NULL); |
| 444 | |
| 445 | first=(char*)base; last=first+(nmemb-1)*size; |
| 446 | |
| 447 | if ((size_t)(last-first)>=trunc) { |
| 448 | char *ffirst=first,*llast=last; |
| 449 | while (1) { |
| 450 | /* Select pivot */ |
| 451 | { char * mid=first+size*((last-first)/size >> 1); |
| 452 | Pivot(SWAP_aligned,size); |
| 453 | memcpy(pivot,mid,size); |
| 454 | } |
| 455 | /* Partition. */ |
| 456 | Partition(SWAP_aligned,size); |
| 457 | /* Prepare to recurse/iterate. */ |
| 458 | Recurse(trunc) |
| 459 | } |
| 460 | } |
| 461 | PreInsertion(SWAP_aligned,TRUNC_aligned,size); |
| 462 | Insertion(SWAP_aligned); |
| 463 | free(pivot); |
| 464 | } |
| 465 | |
| 466 | static void qsort_r_words(void *base, size_t nmemb, |
| 467 | int (SDLCALL *compare)(void *,const void *, const void *), void *userdata) { |
| 468 | |
| 469 | stack_entry stack[STACK_SIZE]; |
| 470 | int stacktop=0; |
| 471 | char *first,*last; |
| 472 | char *pivot=malloc(WORD_BYTES); |
| 473 | assert(pivot != NULL); |
| 474 | |
| 475 | first=(char*)base; last=first+(nmemb-1)*WORD_BYTES; |
| 476 | |
| 477 | if (last-first>=TRUNC_words) { |
| 478 | char *ffirst=first, *llast=last; |
| 479 | while (1) { |
| 480 | #ifdef DEBUG_QSORT |
| 481 | fprintf(stderr,"Doing %d:%d: " , |
| 482 | (first-(char*)base)/WORD_BYTES, |
| 483 | (last-(char*)base)/WORD_BYTES); |
| 484 | #endif |
| 485 | /* Select pivot */ |
| 486 | { char * mid=first+WORD_BYTES*((last-first) / (2*WORD_BYTES)); |
| 487 | Pivot(SWAP_words,WORD_BYTES); |
| 488 | *(int*)pivot=*(int*)mid; |
| 489 | #ifdef DEBUG_QSORT |
| 490 | fprintf(stderr,"pivot = %p = #%lu = %d\n" , mid, (unsigned long)(((int*)mid)-((int*)base)), *(int*)mid); |
| 491 | #endif |
| 492 | } |
| 493 | /* Partition. */ |
| 494 | Partition(SWAP_words,WORD_BYTES); |
| 495 | #ifdef DEBUG_QSORT |
| 496 | fprintf(stderr, "after partitioning first=#%lu last=#%lu\n" , (first-(char*)base)/4lu, (last-(char*)base)/4lu); |
| 497 | #endif |
| 498 | /* Prepare to recurse/iterate. */ |
| 499 | Recurse(TRUNC_words) |
| 500 | } |
| 501 | } |
| 502 | PreInsertion(SWAP_words,TRUNC_words/WORD_BYTES,WORD_BYTES); |
| 503 | /* Now do insertion sort. */ |
| 504 | last=((char*)base)+nmemb*WORD_BYTES; |
| 505 | for (first=((char*)base)+WORD_BYTES;first!=last;first+=WORD_BYTES) { |
| 506 | /* Find the right place for |first|. My apologies for var reuse */ |
| 507 | int *pl=(int*)(first-WORD_BYTES),*pr=(int*)first; |
| 508 | *(int*)pivot=*(int*)first; |
| 509 | for (;compare(userdata,pl,pivot)>0;pr=pl,--pl) { |
| 510 | *pr=*pl; } |
| 511 | if (pr!=(int*)first) *pr=*(int*)pivot; |
| 512 | } |
| 513 | free(pivot); |
| 514 | } |
| 515 | |
| 516 | /* ---------------------------------------------------------------------- */ |
| 517 | |
| 518 | void SDL_qsort_r(void *base, size_t nmemb, size_t size, |
| 519 | SDL_CompareCallback_r compare, void *userdata) { |
| 520 | |
| 521 | if (nmemb<=1) return; |
| 522 | if (((uintptr_t)base|size)&(WORD_BYTES-1)) |
| 523 | qsort_r_nonaligned(base,nmemb,size,compare,userdata); |
| 524 | else if (size!=WORD_BYTES) |
| 525 | qsort_r_aligned(base,nmemb,size,compare,userdata); |
| 526 | else |
| 527 | qsort_r_words(base,nmemb,compare,userdata); |
| 528 | } |
| 529 | |
| 530 | static int SDLCALL qsort_non_r_bridge(void *userdata, const void *a, const void *b) |
| 531 | { |
| 532 | int (SDLCALL *compare)(const void *, const void *) = (int (SDLCALL *)(const void *, const void *)) userdata; |
| 533 | return compare(a, b); |
| 534 | } |
| 535 | |
| 536 | void SDL_qsort(void *base, size_t nmemb, size_t size, SDL_CompareCallback compare) |
| 537 | { |
| 538 | SDL_qsort_r(base, nmemb, size, qsort_non_r_bridge, compare); |
| 539 | } |
| 540 | |
| 541 | // Don't use the C runtime for such a simple function, since we want to allow SDLCALL callbacks and userdata. |
| 542 | // SDL's replacement: Taken from the Public Domain C Library (PDCLib): |
| 543 | // Permission is granted to use, modify, and / or redistribute at will. |
| 544 | void *SDL_bsearch_r(const void *key, const void *base, size_t nmemb, size_t size, SDL_CompareCallback_r compare, void *userdata) |
| 545 | { |
| 546 | const void *pivot; |
| 547 | size_t corr; |
| 548 | int rc; |
| 549 | |
| 550 | while (nmemb) { |
| 551 | /* algorithm needs -1 correction if remaining elements are an even number. */ |
| 552 | corr = nmemb % 2; |
| 553 | nmemb /= 2; |
| 554 | pivot = (const char *)base + (nmemb * size); |
| 555 | rc = compare(userdata, key, pivot); |
| 556 | |
| 557 | if (rc > 0) { |
| 558 | base = (const char *)pivot + size; |
| 559 | /* applying correction */ |
| 560 | nmemb -= (1 - corr); |
| 561 | } else if (rc == 0) { |
| 562 | return (void *)pivot; |
| 563 | } |
| 564 | } |
| 565 | |
| 566 | return NULL; |
| 567 | } |
| 568 | |
| 569 | void *SDL_bsearch(const void *key, const void *base, size_t nmemb, size_t size, SDL_CompareCallback compare) |
| 570 | { |
| 571 | // qsort_non_r_bridge just happens to match calling conventions, so reuse it. |
| 572 | return SDL_bsearch_r(key, base, nmemb, size, qsort_non_r_bridge, compare); |
| 573 | } |
| 574 | |
| 575 | |