0
|
1 /* Plug-compatible replacement for UNIX qsort.
|
|
2 Copyright (C) 1989 Free Software Foundation, Inc.
|
|
3 Written by Douglas C. Schmidt (schmidt@ics.uci.edu)
|
|
4
|
|
5 This file is part of GNU CC.
|
|
6
|
|
7 GNU QSORT is free software; you can redistribute it and/or modify
|
|
8 it under the terms of the GNU General Public License as published by
|
|
9 the Free Software Foundation; either version 2, or (at your option)
|
|
10 any later version.
|
|
11
|
|
12 GNU QSORT is distributed in the hope that it will be useful,
|
|
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
15 GNU General Public License for more details.
|
|
16
|
|
17 You should have received a copy of the GNU General Public License
|
|
18 along with GNU QSORT; see the file COPYING. If not, write to
|
|
19 the Free the Free Software Foundation, Inc., 59 Temple Place - Suite 330,
|
|
20 Boston, MA 02111-1307, USA. */
|
|
21
|
|
22 /* Synched up with: FSF 19.28. */
|
|
23
|
|
24 #ifdef sparc
|
|
25 #include <alloca.h>
|
|
26 #endif
|
|
27
|
|
28 /* Invoke the comparison function, returns either 0, < 0, or > 0. */
|
|
29 #define CMP(A,B) ((*cmp)((A),(B)))
|
|
30
|
|
31 /* Byte-wise swap two items of size SIZE. */
|
|
32 #define SWAP(A,B,SIZE) do {int sz = (SIZE); char *a = (A); char *b = (B); \
|
|
33 do { char _temp = *a;*a++ = *b;*b++ = _temp;} while (--sz);} while (0)
|
|
34
|
|
35 /* Copy SIZE bytes from item B to item A. */
|
|
36 #define COPY(A,B,SIZE) {int sz = (SIZE); do { *(A)++ = *(B)++; } while (--sz); }
|
|
37
|
|
38 /* This should be replaced by a standard ANSI macro. */
|
|
39 #define BYTES_PER_WORD 8
|
|
40
|
|
41 /* The next 4 #defines implement a very fast in-line stack abstraction. */
|
|
42 #define STACK_SIZE (BYTES_PER_WORD * sizeof (long))
|
|
43 #define PUSH(LOW,HIGH) do {top->lo = LOW;top++->hi = HIGH;} while (0)
|
|
44 #define POP(LOW,HIGH) do {LOW = (--top)->lo;HIGH = top->hi;} while (0)
|
|
45 #define STACK_NOT_EMPTY (stack < top)
|
|
46
|
|
47 /* Discontinue quicksort algorithm when partition gets below this size.
|
|
48 This particular magic number was chosen to work best on a Sun 4/260. */
|
|
49 #define MAX_THRESH 4
|
|
50
|
|
51 /* Stack node declarations used to store unfulfilled partition obligations. */
|
|
52 typedef struct
|
|
53 {
|
|
54 char *lo;
|
|
55 char *hi;
|
|
56 } stack_node;
|
|
57
|
|
58 /* Order size using quicksort. This implementation incorporates
|
|
59 four optimizations discussed in Sedgewick:
|
|
60
|
|
61 1. Non-recursive, using an explicit stack of pointer that store the
|
|
62 next array partition to sort. To save time, this maximum amount
|
|
63 of space required to store an array of MAX_INT is allocated on the
|
|
64 stack. Assuming a 32-bit integer, this needs only 32 *
|
|
65 sizeof (stack_node) == 136 bits. Pretty cheap, actually.
|
|
66
|
|
67 2. Chose the pivot element using a median-of-three decision tree.
|
|
68 This reduces the probability of selecting a bad pivot value and
|
|
69 eliminates certain extraneous comparisons.
|
|
70
|
|
71 3. Only quicksorts TOTAL_ELEMS / MAX_THRESH partitions, leaving
|
|
72 insertion sort to order the MAX_THRESH items within each partition.
|
|
73 This is a big win, since insertion sort is faster for small, mostly
|
|
74 sorted array segments.
|
|
75
|
|
76 4. The larger of the two sub-partitions is always pushed onto the
|
|
77 stack first, with the algorithm then concentrating on the
|
|
78 smaller partition. This *guarantees* no more than log (n)
|
|
79 stack size is needed (actually O(1) in this case)! */
|
|
80
|
|
81 int
|
|
82 qsort (base_ptr, total_elems, size, cmp)
|
|
83 char *base_ptr;
|
|
84 int total_elems;
|
|
85 int size;
|
|
86 int (*cmp)();
|
|
87 {
|
|
88 /* Allocating SIZE bytes for a pivot buffer facilitates a better
|
|
89 algorithm below since we can do comparisons directly on the pivot. */
|
|
90 char *pivot_buffer = (char *) alloca (size);
|
|
91 int max_thresh = MAX_THRESH * size;
|
|
92
|
|
93 if (total_elems > MAX_THRESH)
|
|
94 {
|
|
95 char *lo = base_ptr;
|
|
96 char *hi = lo + size * (total_elems - 1);
|
|
97 stack_node stack[STACK_SIZE]; /* Largest size needed for 32-bit int!!! */
|
|
98 stack_node *top = stack + 1;
|
|
99
|
|
100 while (STACK_NOT_EMPTY)
|
|
101 {
|
|
102 char *left_ptr;
|
|
103 char *right_ptr;
|
|
104 {
|
|
105 char *pivot = pivot_buffer;
|
|
106 {
|
|
107 /* Select median value from among LO, MID, and HI. Rearrange
|
|
108 LO and HI so the three values are sorted. This lowers the
|
|
109 probability of picking a pathological pivot value and
|
|
110 skips a comparison for both the LEFT_PTR and RIGHT_PTR. */
|
|
111
|
|
112 char *mid = lo + size * ((hi - lo) / size >> 1);
|
|
113
|
|
114 if (CMP (mid, lo) < 0)
|
|
115 SWAP (mid, lo, size);
|
|
116 if (CMP (hi, mid) < 0)
|
|
117 SWAP (mid, hi, size);
|
|
118 else
|
|
119 goto jump_over;
|
|
120 if (CMP (mid, lo) < 0)
|
|
121 SWAP (mid, lo, size);
|
|
122 jump_over:
|
|
123 COPY (pivot, mid, size);
|
|
124 pivot = pivot_buffer;
|
|
125 }
|
|
126 left_ptr = lo + size;
|
|
127 right_ptr = hi - size;
|
|
128
|
|
129 /* Here's the famous ``collapse the walls'' section of quicksort.
|
|
130 Gotta like those tight inner loops! They are the main reason
|
|
131 that this algorithm runs much faster than others. */
|
|
132 do
|
|
133 {
|
|
134 while (CMP (left_ptr, pivot) < 0)
|
|
135 left_ptr += size;
|
|
136
|
|
137 while (CMP (pivot, right_ptr) < 0)
|
|
138 right_ptr -= size;
|
|
139
|
|
140 if (left_ptr < right_ptr)
|
|
141 {
|
|
142 SWAP (left_ptr, right_ptr, size);
|
|
143 left_ptr += size;
|
|
144 right_ptr -= size;
|
|
145 }
|
|
146 else if (left_ptr == right_ptr)
|
|
147 {
|
|
148 left_ptr += size;
|
|
149 right_ptr -= size;
|
|
150 break;
|
|
151 }
|
|
152 }
|
|
153 while (left_ptr <= right_ptr);
|
|
154
|
|
155 }
|
|
156
|
|
157 /* Set up pointers for next iteration. First determine whether
|
|
158 left and right partitions are below the threshold size. If so,
|
|
159 ignore one or both. Otherwise, push the larger partition's
|
|
160 bounds on the stack and continue sorting the smaller one. */
|
|
161
|
|
162 if ((right_ptr - lo) <= max_thresh)
|
|
163 {
|
|
164 if ((hi - left_ptr) <= max_thresh) /* Ignore both small partitions. */
|
|
165 POP (lo, hi);
|
|
166 else /* Ignore small left partition. */
|
|
167 lo = left_ptr;
|
|
168 }
|
|
169 else if ((hi - left_ptr) <= max_thresh) /* Ignore small right partition. */
|
|
170 hi = right_ptr;
|
|
171 else if ((right_ptr - lo) > (hi - left_ptr)) /* Push larger left partition indices. */
|
|
172 {
|
|
173 PUSH (lo, right_ptr);
|
|
174 lo = left_ptr;
|
|
175 }
|
|
176 else /* Push larger right partition indices. */
|
|
177 {
|
|
178 PUSH (left_ptr, hi);
|
|
179 hi = right_ptr;
|
|
180 }
|
|
181 }
|
|
182 }
|
|
183
|
|
184 /* Once the BASE_PTR array is partially sorted by quicksort the rest
|
|
185 is completely sorted using insertion sort, since this is efficient
|
|
186 for partitions below MAX_THRESH size. BASE_PTR points to the beginning
|
|
187 of the array to sort, and END_PTR points at the very last element in
|
|
188 the array (*not* one beyond it!). */
|
|
189
|
|
190 #define MIN(X,Y) ((X) < (Y) ? (X) : (Y))
|
|
191
|
|
192 {
|
|
193 char *end_ptr = base_ptr + size * (total_elems - 1);
|
|
194 char *run_ptr;
|
|
195 char *tmp_ptr = base_ptr;
|
|
196 char *thresh = MIN (end_ptr, base_ptr + max_thresh);
|
|
197
|
|
198 /* Find smallest element in first threshold and place it at the
|
|
199 array's beginning. This is the smallest array element,
|
|
200 and the operation speeds up insertion sort's inner loop. */
|
|
201
|
|
202 for (run_ptr = tmp_ptr + size; run_ptr <= thresh; run_ptr += size)
|
|
203 if (CMP (run_ptr, tmp_ptr) < 0)
|
|
204 tmp_ptr = run_ptr;
|
|
205
|
|
206 if (tmp_ptr != base_ptr)
|
|
207 SWAP (tmp_ptr, base_ptr, size);
|
|
208
|
|
209 /* Insertion sort, running from left-hand-side up to `right-hand-side.'
|
|
210 Pretty much straight out of the original GNU qsort routine. */
|
|
211
|
|
212 for (run_ptr = base_ptr + size; (tmp_ptr = run_ptr += size) <= end_ptr; )
|
|
213 {
|
|
214
|
|
215 while (CMP (run_ptr, tmp_ptr -= size) < 0)
|
|
216 ;
|
|
217
|
|
218 if ((tmp_ptr += size) != run_ptr)
|
|
219 {
|
|
220 char *trav;
|
|
221
|
|
222 for (trav = run_ptr + size; --trav >= run_ptr;)
|
|
223 {
|
|
224 char c = *trav;
|
|
225 char *hi, *lo;
|
|
226
|
|
227 for (hi = lo = trav; (lo -= size) >= tmp_ptr; hi = lo)
|
|
228 *hi = *lo;
|
|
229 *hi = c;
|
|
230 }
|
|
231 }
|
|
232
|
|
233 }
|
|
234 }
|
|
235 return 1;
|
|
236 }
|
|
237
|