1 /* Copyright (C) 1991-2020 Free Software Foundation, Inc. 2 This file is part of libctf (imported from Gnulib). 3 Written by Douglas C. Schmidt (schmidt@ics.uci.edu). 4 5 The GNU C Library is free software; you can redistribute it and/or 6 modify it under the terms of the GNU Lesser General Public 7 License as published by the Free Software Foundation; either 8 version 2.1 of the License, or (at your option) any later version. 9 10 The GNU C Library is distributed in the hope that it will be useful, 11 but WITHOUT ANY WARRANTY; without even the implied warranty of 12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 13 Lesser General Public License for more details. 14 15 You should have received a copy of the GNU Lesser General Public 16 License along with the GNU C Library; if not, see 17 <https://www.gnu.org/licenses/>. */ 18 19 /* If you consider tuning this algorithm, you should consult first: 20 Engineering a sort function; Jon Bentley and M. Douglas McIlroy; 21 Software - Practice and Experience; Vol. 23 (11), 1249-1265, 1993. */ 22 23 #ifndef _LIBC 24 # include <config.h> 25 #endif 26 27 #include <limits.h> 28 #include <stdlib.h> 29 #include <string.h> 30 #include "ctf-decls.h" 31 32 #ifndef _LIBC 33 # define _quicksort ctf_qsort_r 34 # define __compar_d_fn_t compar_d_fn_t 35 typedef int (*compar_d_fn_t) (const void *, const void *, void *); 36 #endif 37 38 /* Byte-wise swap two items of size SIZE. */ 39 #define SWAP(a, b, size) \ 40 do \ 41 { \ 42 size_t __size = (size); \ 43 char *__a = (a), *__b = (b); \ 44 do \ 45 { \ 46 char __tmp = *__a; \ 47 *__a++ = *__b; \ 48 *__b++ = __tmp; \ 49 } while (--__size > 0); \ 50 } while (0) 51 52 /* Discontinue quicksort algorithm when partition gets below this size. 53 This particular magic number was chosen to work best on a Sun 4/260. */ 54 #define MAX_THRESH 4 55 56 /* Stack node declarations used to store unfulfilled partition obligations. */ 57 typedef struct 58 { 59 char *lo; 60 char *hi; 61 } stack_node; 62 63 /* The next 4 #defines implement a very fast in-line stack abstraction. */ 64 /* The stack needs log (total_elements) entries (we could even subtract 65 log(MAX_THRESH)). Since total_elements has type size_t, we get as 66 upper bound for log (total_elements): 67 bits per byte (CHAR_BIT) * sizeof(size_t). */ 68 #define STACK_SIZE (CHAR_BIT * sizeof(size_t)) 69 #define PUSH(low, high) ((void) ((top->lo = (low)), (top->hi = (high)), ++top)) 70 #define POP(low, high) ((void) (--top, (low = top->lo), (high = top->hi))) 71 #define STACK_NOT_EMPTY (stack < top) 72 73 74 /* Order size using quicksort. This implementation incorporates 75 four optimizations discussed in Sedgewick: 76 77 1. Non-recursive, using an explicit stack of pointer that store the 78 next array partition to sort. To save time, this maximum amount 79 of space required to store an array of SIZE_MAX is allocated on the 80 stack. Assuming a 32-bit (64 bit) integer for size_t, this needs 81 only 32 * sizeof(stack_node) == 256 bytes (for 64 bit: 1024 bytes). 82 Pretty cheap, actually. 83 84 2. Chose the pivot element using a median-of-three decision tree. 85 This reduces the probability of selecting a bad pivot value and 86 eliminates certain extraneous comparisons. 87 88 3. Only quicksorts TOTAL_ELEMS / MAX_THRESH partitions, leaving 89 insertion sort to order the MAX_THRESH items within each partition. 90 This is a big win, since insertion sort is faster for small, mostly 91 sorted array segments. 92 93 4. The larger of the two sub-partitions is always pushed onto the 94 stack first, with the algorithm then concentrating on the 95 smaller partition. This *guarantees* no more than log (total_elems) 96 stack size is needed (actually O(1) in this case)! */ 97 98 void 99 _quicksort (void *const pbase, size_t total_elems, size_t size, 100 __compar_d_fn_t cmp, void *arg) 101 { 102 char *base_ptr = (char *) pbase; 103 104 const size_t max_thresh = MAX_THRESH * size; 105 106 if (total_elems == 0) 107 /* Avoid lossage with unsigned arithmetic below. */ 108 return; 109 110 if (total_elems > MAX_THRESH) 111 { 112 char *lo = base_ptr; 113 char *hi = &lo[size * (total_elems - 1)]; 114 stack_node stack[STACK_SIZE]; 115 stack_node *top = stack; 116 117 PUSH (NULL, NULL); 118 119 while (STACK_NOT_EMPTY) 120 { 121 char *left_ptr; 122 char *right_ptr; 123 124 /* Select median value from among LO, MID, and HI. Rearrange 125 LO and HI so the three values are sorted. This lowers the 126 probability of picking a pathological pivot value and 127 skips a comparison for both the LEFT_PTR and RIGHT_PTR in 128 the while loops. */ 129 130 char *mid = lo + size * ((hi - lo) / size >> 1); 131 132 if ((*cmp) ((void *) mid, (void *) lo, arg) < 0) 133 SWAP (mid, lo, size); 134 if ((*cmp) ((void *) hi, (void *) mid, arg) < 0) 135 SWAP (mid, hi, size); 136 else 137 goto jump_over; 138 if ((*cmp) ((void *) mid, (void *) lo, arg) < 0) 139 SWAP (mid, lo, size); 140 jump_over:; 141 142 left_ptr = lo + size; 143 right_ptr = hi - size; 144 145 /* Here's the famous ``collapse the walls'' section of quicksort. 146 Gotta like those tight inner loops! They are the main reason 147 that this algorithm runs much faster than others. */ 148 do 149 { 150 while ((*cmp) ((void *) left_ptr, (void *) mid, arg) < 0) 151 left_ptr += size; 152 153 while ((*cmp) ((void *) mid, (void *) right_ptr, arg) < 0) 154 right_ptr -= size; 155 156 if (left_ptr < right_ptr) 157 { 158 SWAP (left_ptr, right_ptr, size); 159 if (mid == left_ptr) 160 mid = right_ptr; 161 else if (mid == right_ptr) 162 mid = left_ptr; 163 left_ptr += size; 164 right_ptr -= size; 165 } 166 else if (left_ptr == right_ptr) 167 { 168 left_ptr += size; 169 right_ptr -= size; 170 break; 171 } 172 } 173 while (left_ptr <= right_ptr); 174 175 /* Set up pointers for next iteration. First determine whether 176 left and right partitions are below the threshold size. If so, 177 ignore one or both. Otherwise, push the larger partition's 178 bounds on the stack and continue sorting the smaller one. */ 179 180 if ((size_t) (right_ptr - lo) <= max_thresh) 181 { 182 if ((size_t) (hi - left_ptr) <= max_thresh) 183 /* Ignore both small partitions. */ 184 POP (lo, hi); 185 else 186 /* Ignore small left partition. */ 187 lo = left_ptr; 188 } 189 else if ((size_t) (hi - left_ptr) <= max_thresh) 190 /* Ignore small right partition. */ 191 hi = right_ptr; 192 else if ((right_ptr - lo) > (hi - left_ptr)) 193 { 194 /* Push larger left partition indices. */ 195 PUSH (lo, right_ptr); 196 lo = left_ptr; 197 } 198 else 199 { 200 /* Push larger right partition indices. */ 201 PUSH (left_ptr, hi); 202 hi = right_ptr; 203 } 204 } 205 } 206 207 /* Once the BASE_PTR array is partially sorted by quicksort the rest 208 is completely sorted using insertion sort, since this is efficient 209 for partitions below MAX_THRESH size. BASE_PTR points to the beginning 210 of the array to sort, and END_PTR points at the very last element in 211 the array (*not* one beyond it!). */ 212 213 #define min(x, y) ((x) < (y) ? (x) : (y)) 214 215 { 216 char *const end_ptr = &base_ptr[size * (total_elems - 1)]; 217 char *tmp_ptr = base_ptr; 218 char *thresh = min(end_ptr, base_ptr + max_thresh); 219 char *run_ptr; 220 221 /* Find smallest element in first threshold and place it at the 222 array's beginning. This is the smallest array element, 223 and the operation speeds up insertion sort's inner loop. */ 224 225 for (run_ptr = tmp_ptr + size; run_ptr <= thresh; run_ptr += size) 226 if ((*cmp) ((void *) run_ptr, (void *) tmp_ptr, arg) < 0) 227 tmp_ptr = run_ptr; 228 229 if (tmp_ptr != base_ptr) 230 SWAP (tmp_ptr, base_ptr, size); 231 232 /* Insertion sort, running from left-hand-side up to right-hand-side. */ 233 234 run_ptr = base_ptr + size; 235 while ((run_ptr += size) <= end_ptr) 236 { 237 tmp_ptr = run_ptr - size; 238 while ((*cmp) ((void *) run_ptr, (void *) tmp_ptr, arg) < 0) 239 tmp_ptr -= size; 240 241 tmp_ptr += size; 242 if (tmp_ptr != run_ptr) 243 { 244 char *trav; 245 246 trav = run_ptr + size; 247 while (--trav >= run_ptr) 248 { 249 char c = *trav; 250 char *hi, *lo; 251 252 for (hi = lo = trav; (lo -= size) >= tmp_ptr; hi = lo) 253 *hi = *lo; 254 *hi = c; 255 } 256 } 257 } 258 } 259 } 260