Rev 9013 | Details | Compare with Previous | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line |
|---|---|---|---|
| 9013 | turbocat | 1 | /* Copyright (C) 1994 DJ Delorie, see COPYING.DJ for details */ |
| 2 | #include |
||
| 3 | |||
| 4 | /*- |
||
| 5 | * Copyright (c) 1980, 1983 The Regents of the University of California. |
||
| 6 | * All rights reserved. |
||
| 7 | * |
||
| 8 | * Redistribution and use in source and binary forms are permitted |
||
| 9 | * provided that: (1) source distributions retain this entire copyright |
||
| 10 | * notice and comment, and (2) distributions including binaries display |
||
| 11 | * the following acknowledgement: ``This product includes software |
||
| 12 | * developed by the University of California, Berkeley and its contributors'' |
||
| 13 | * in the documentation or other materials provided with the distribution |
||
| 14 | * and in all advertising materials mentioning features or use of this |
||
| 15 | * software. Neither the name of the University nor the names of its |
||
| 16 | * contributors may be used to endorse or promote products derived |
||
| 17 | * from this software without specific prior written permission. |
||
| 18 | * THIS SOFTWARE IS PROVIDED ``AS IS'' AND WITHOUT ANY EXPRESS OR |
||
| 19 | * IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED |
||
| 20 | * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. |
||
| 21 | */ |
||
| 22 | |||
| 23 | /* |
||
| 24 | * qsort.c: |
||
| 25 | * Our own version of the system qsort routine which is faster by an average |
||
| 26 | * of 25%, with lows and highs of 10% and 50%. |
||
| 27 | * The THRESHold below is the insertion sort threshold, and has been adjusted |
||
| 28 | * for records of size 48 bytes. |
||
| 29 | * The MTHREShold is where we stop finding a better median. |
||
| 30 | */ |
||
| 31 | |||
| 9765 | turbocat | 32 | #define THRESH 4 /* threshold for insertion */ |
| 33 | #define MTHRESH 6 /* threshold for median */ |
||
| 9013 | turbocat | 34 | |
| 9765 | turbocat | 35 | static int (*qcmp)(const void*, const void*); /* the comparison routine */ |
| 36 | static int qsz; /* size of each record */ |
||
| 37 | static int thresh; /* THRESHold in chars */ |
||
| 38 | static int mthresh; /* MTHRESHold in chars */ |
||
| 9013 | turbocat | 39 | |
| 40 | /* |
||
| 41 | * qst: |
||
| 42 | * Do a quicksort |
||
| 43 | * First, find the median element, and put that one in the first place as the |
||
| 44 | * discriminator. (This "median" is just the median of the first, last and |
||
| 45 | * middle elements). (Using this median instead of the first element is a big |
||
| 46 | * win). Then, the usual partitioning/swapping, followed by moving the |
||
| 47 | * discriminator into the right place. Then, figure out the sizes of the two |
||
| 48 | * partions, do the smaller one recursively and the larger one via a repeat of |
||
| 49 | * this code. Stopping when there are less than THRESH elements in a partition |
||
| 50 | * and cleaning up with an insertion sort (in our caller) is a huge win. |
||
| 51 | * All data swaps are done in-line, which is space-losing but time-saving. |
||
| 52 | * (And there are only three places where this is done). |
||
| 53 | */ |
||
| 54 | |||
| 9765 | turbocat | 55 | static void qst(char* base, char* max) |
| 9013 | turbocat | 56 | { |
| 9765 | turbocat | 57 | char c, *i, *j, *jj; |
| 58 | int ii; |
||
| 59 | char *mid, *tmp; |
||
| 60 | int lo, hi; |
||
| 9013 | turbocat | 61 | |
| 62 | /* |
||
| 9765 | turbocat | 63 | * At the top here, lo is the number of characters of elements in the |
| 64 | * current partition. (Which should be max - base). |
||
| 65 | * Find the median of the first, last, and middle element and make |
||
| 66 | * that the middle element. Set j to largest of first and middle. |
||
| 67 | * If max is larger than that guy, then it's that guy, else compare |
||
| 68 | * max with loser of first and take larger. Things are set up to |
||
| 69 | * prefer the middle, then the first in case of ties. |
||
| 9013 | turbocat | 70 | */ |
| 9765 | turbocat | 71 | lo = max - base; /* number of elements as chars */ |
| 72 | do { |
||
| 73 | mid = i = base + qsz * ((lo / qsz) >> 1); |
||
| 74 | if (lo >= mthresh) { |
||
| 75 | j = (qcmp((jj = base), i) > 0 ? jj : i); |
||
| 76 | if (qcmp(j, (tmp = max - qsz)) > 0) { |
||
| 77 | /* switch to first loser */ |
||
| 78 | j = (j == jj ? i : jj); |
||
| 79 | if (qcmp(j, tmp) < 0) |
||
| 80 | j = tmp; |
||
| 81 | } |
||
| 82 | if (j != i) { |
||
| 83 | ii = qsz; |
||
| 84 | do { |
||
| 85 | c = *i; |
||
| 86 | *i++ = *j; |
||
| 87 | *j++ = c; |
||
| 88 | } while (--ii); |
||
| 89 | } |
||
| 90 | } |
||
| 91 | /* |
||
| 92 | * Semi-standard quicksort partitioning/swapping |
||
| 93 | */ |
||
| 94 | for (i = base, j = max - qsz;;) { |
||
| 95 | while (i < mid && qcmp(i, mid) <= 0) |
||
| 96 | i += qsz; |
||
| 97 | while (j > mid) { |
||
| 98 | if (qcmp(mid, j) <= 0) { |
||
| 99 | j -= qsz; |
||
| 100 | continue; |
||
| 101 | } |
||
| 102 | tmp = i + qsz; /* value of i after swap */ |
||
| 103 | if (i == mid) { |
||
| 104 | /* j <-> mid, new mid is j */ |
||
| 105 | mid = jj = j; |
||
| 106 | } else { |
||
| 107 | /* i <-> j */ |
||
| 108 | jj = j; |
||
| 109 | j -= qsz; |
||
| 110 | } |
||
| 111 | goto swap; |
||
| 112 | } |
||
| 113 | if (i == mid) { |
||
| 114 | break; |
||
| 115 | } else { |
||
| 116 | /* i <-> mid, new mid is i */ |
||
| 117 | jj = mid; |
||
| 118 | tmp = mid = i; /* value of i after swap */ |
||
| 119 | j -= qsz; |
||
| 120 | } |
||
| 121 | swap: |
||
| 122 | ii = qsz; |
||
| 123 | do { |
||
| 124 | c = *i; |
||
| 125 | *i++ = *jj; |
||
| 126 | *jj++ = c; |
||
| 127 | } while (--ii); |
||
| 128 | i = tmp; |
||
| 129 | } |
||
| 130 | /* |
||
| 131 | * Look at sizes of the two partitions, do the smaller |
||
| 132 | * one first by recursion, then do the larger one by |
||
| 133 | * making sure lo is its size, base and max are update |
||
| 134 | * correctly, and branching back. But only repeat |
||
| 135 | * (recursively or by branching) if the partition is |
||
| 136 | * of at least size THRESH. |
||
| 137 | */ |
||
| 138 | i = (j = mid) + qsz; |
||
| 139 | if ((lo = j - base) <= (hi = max - i)) { |
||
| 140 | if (lo >= thresh) |
||
| 141 | qst(base, j); |
||
| 142 | base = i; |
||
| 143 | lo = hi; |
||
| 144 | } else { |
||
| 145 | if (hi >= thresh) |
||
| 146 | qst(i, max); |
||
| 147 | max = j; |
||
| 148 | } |
||
| 149 | } while (lo >= thresh); |
||
| 9013 | turbocat | 150 | } |
| 151 | |||
| 152 | /* |
||
| 153 | * qsort: |
||
| 154 | * First, set up some global parameters for qst to share. Then, quicksort |
||
| 155 | * with qst(), and then a cleanup insertion sort ourselves. Sound simple? |
||
| 156 | * It's not... |
||
| 157 | */ |
||
| 158 | |||
| 9765 | turbocat | 159 | void qsort(void* base0, size_t n, size_t size, int (*compar)(const void*, const void*)) |
| 9013 | turbocat | 160 | { |
| 9765 | turbocat | 161 | char* base = (char*)base0; |
| 162 | char c, *i, *j, *lo, *hi; |
||
| 163 | char *min, *max; |
||
| 9013 | turbocat | 164 | |
| 9765 | turbocat | 165 | if (n <= 1) |
| 166 | return; |
||
| 167 | qsz = size; |
||
| 168 | qcmp = compar; |
||
| 169 | thresh = qsz * THRESH; |
||
| 170 | mthresh = qsz * MTHRESH; |
||
| 171 | max = base + n * qsz; |
||
| 172 | if (n >= THRESH) { |
||
| 173 | qst(base, max); |
||
| 174 | hi = base + thresh; |
||
| 175 | } else { |
||
| 176 | hi = max; |
||
| 9013 | turbocat | 177 | } |
| 9765 | turbocat | 178 | /* |
| 179 | * First put smallest element, which must be in the first THRESH, in |
||
| 180 | * the first position as a sentinel. This is done just by searching |
||
| 181 | * the first THRESH elements (or the first n if n < THRESH), finding |
||
| 182 | * the min, and swapping it into the first position. |
||
| 183 | */ |
||
| 184 | for (j = lo = base; (lo += qsz) < hi;) |
||
| 185 | if (qcmp(j, lo) > 0) |
||
| 186 | j = lo; |
||
| 187 | if (j != base) { |
||
| 188 | /* swap j into place */ |
||
| 189 | for (i = base, hi = base + qsz; i < hi;) { |
||
| 190 | c = *j; |
||
| 191 | *j++ = *i; |
||
| 192 | *i++ = c; |
||
| 193 | } |
||
| 9013 | turbocat | 194 | } |
| 9765 | turbocat | 195 | /* |
| 196 | * With our sentinel in place, we now run the following hyper-fast |
||
| 197 | * insertion sort. For each remaining element, min, from [1] to [n-1], |
||
| 198 | * set hi to the index of the element AFTER which this one goes. |
||
| 199 | * Then, do the standard insertion sort shift on a character at a time |
||
| 200 | * basis for each element in the frob. |
||
| 201 | */ |
||
| 202 | for (min = base; (hi = min += qsz) < max;) { |
||
| 203 | while (qcmp(hi -= qsz, min) > 0) |
||
| 204 | /* void */; |
||
| 205 | if ((hi += qsz) != min) { |
||
| 206 | for (lo = min + qsz; --lo >= min;) { |
||
| 207 | c = *lo; |
||
| 208 | for (i = j = lo; (j -= qsz) >= hi; i = j) |
||
| 209 | *i = *j; |
||
| 210 | *i = c; |
||
| 211 | } |
||
| 212 | } |
||
| 213 | } |
||
| 9013 | turbocat | 214 | } |