1
/* This file is part of the Scotch distribution. It does
2
** not have the stardard Scotch header with the INRIA
3
** copyright notice because it is a very slight adaptation
4
** of the qsort routine of glibc 2.4, taylored to match
5
** Scotch needs. As Scotch is distributed according to the
6
** CeCILL-C license, which is LGPL-compatible, no further
7
** notices are required.
10
/* Copyright (C) 1991,1992,1996,1997,1999,2004 Free Software Foundation, Inc.
11
This file is part of the GNU C Library.
12
Written by Douglas C. Schmidt (schmidt@ics.uci.edu).
14
The GNU C Library is free software; you can redistribute it and/or
15
modify it under the terms of the GNU Lesser General Public
16
License as published by the Free Software Foundation; either
17
version 2.1 of the License, or (at your option) any later version.
19
The GNU C Library is distributed in the hope that it will be useful,
20
but WITHOUT ANY WARRANTY; without even the implied warranty of
21
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
22
Lesser General Public License for more details.
24
You should have received a copy of the GNU Lesser General Public
25
License along with the GNU C Library; if not, write to the Free
26
Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
29
/* If you consider tuning this algorithm, you should consult first:
30
Engineering a sort function; Jon Bentley and M. Douglas McIlroy;
31
Software - Practice and Experience; Vol. 23 (11), 1249-1265, 1993. */
37
#define max_thresh (MAX_THRESH * INTSORTSIZE) /* Variable turned into constant */
39
/* Stack node declarations used to store unfulfilled partition obligations. */
46
/* The next 4 #defines implement a very fast in-line stack abstraction. */
47
/* The stack needs log (total_elements) entries (we could even subtract
48
log(MAX_THRESH)). Since total_elements has type size_t, we get as
49
upper bound for log (total_elements):
50
bits per byte (CHAR_BIT) * sizeof(size_t). */
51
#define STACK_SIZE (CHAR_BIT * sizeof (INT))
52
#define PUSH(low, high) ((void) ((top->lo = (low)), (top->hi = (high)), ++top))
53
#define POP(low, high) ((void) (--top, (low = top->lo), (high = top->hi)))
54
#define STACK_NOT_EMPTY (stack < top)
56
#endif /* MAX_THRESH */
58
/* Order size using quicksort. This implementation incorporates
59
four optimizations discussed in Sedgewick:
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 SIZE_MAX is allocated on the
64
stack. Assuming a 32-bit (64 bit) integer for size_t, this needs
65
only 32 * sizeof(stack_node) == 256 bytes (for 64 bit: 1024 bytes).
66
Pretty cheap, actually.
68
2. Chose the pivot element using a median-of-three decision tree.
69
This reduces the probability of selecting a bad pivot value and
70
eliminates certain extraneous comparisons.
72
3. Only quicksorts TOTAL_ELEMS / MAX_THRESH partitions, leaving
73
insertion sort to order the MAX_THRESH items within each partition.
74
This is a big win, since insertion sort is faster for small, mostly
75
sorted array segments.
77
4. The larger of the two sub-partitions is always pushed onto the
78
stack first, with the algorithm then concentrating on the
79
smaller partition. This *guarantees* no more than log (total_elems)
80
stack size is needed (actually O(1) in this case)! */
83
** INTSORTNAME : Name of function
84
** INTSORTSIZE : Size of elements to sort
85
** INTSORTSWAP : Swapping macro
86
** INTSORTCMP : Comparison function
91
void * const pbase, /*+ Array to sort +*/
92
const INT total_elems) /*+ Number of entries to sort +*/
94
register char *base_ptr = (char *) pbase;
97
/* Avoid lossage with unsigned arithmetic below. */
100
if (total_elems > MAX_THRESH)
103
char *hi = &lo[INTSORTSIZE * (total_elems - 1)];
104
stack_node stack[STACK_SIZE];
105
stack_node *top = stack;
109
while (STACK_NOT_EMPTY)
114
/* Select median value from among LO, MID, and HI. Rearrange
115
LO and HI so the three values are sorted. This lowers the
116
probability of picking a pathological pivot value and
117
skips a comparison for both the LEFT_PTR and RIGHT_PTR in
120
char *mid = lo + INTSORTSIZE * ((hi - lo) / INTSORTSIZE >> 1);
122
if (INTSORTCMP ((void *) mid, (void *) lo))
123
INTSORTSWAP (mid, lo);
124
if (INTSORTCMP ((void *) hi, (void *) mid))
125
INTSORTSWAP (mid, hi);
128
if (INTSORTCMP ((void *) mid, (void *) lo))
129
INTSORTSWAP (mid, lo);
132
left_ptr = lo + INTSORTSIZE;
133
right_ptr = hi - INTSORTSIZE;
135
/* Here's the famous ``collapse the walls'' section of quicksort.
136
Gotta like those tight inner loops! They are the main reason
137
that this algorithm runs much faster than others. */
140
while (INTSORTCMP ((void *) left_ptr, (void *) mid))
141
left_ptr += INTSORTSIZE;
143
while (INTSORTCMP ((void *) mid, (void *) right_ptr))
144
right_ptr -= INTSORTSIZE;
146
if (left_ptr < right_ptr)
148
INTSORTSWAP (left_ptr, right_ptr);
151
else if (mid == right_ptr)
153
left_ptr += INTSORTSIZE;
154
right_ptr -= INTSORTSIZE;
156
else if (left_ptr == right_ptr)
158
left_ptr += INTSORTSIZE;
159
right_ptr -= INTSORTSIZE;
163
while (left_ptr <= right_ptr);
165
/* Set up pointers for next iteration. First determine whether
166
left and right partitions are below the threshold size. If so,
167
ignore one or both. Otherwise, push the larger partition's
168
bounds on the stack and continue sorting the smaller one. */
170
if ((size_t) (right_ptr - lo) <= max_thresh)
172
if ((size_t) (hi - left_ptr) <= max_thresh)
173
/* Ignore both small partitions. */
176
/* Ignore small left partition. */
179
else if ((size_t) (hi - left_ptr) <= max_thresh)
180
/* Ignore small right partition. */
182
else if ((right_ptr - lo) > (hi - left_ptr))
184
/* Push larger left partition indices. */
185
PUSH (lo, right_ptr);
190
/* Push larger right partition indices. */
197
/* Once the BASE_PTR array is partially sorted by quicksort the rest
198
is completely sorted using insertion sort, since this is efficient
199
for partitions below MAX_THRESH size. BASE_PTR points to the beginning
200
of the array to sort, and END_PTR points at the very last element in
201
the array (*not* one beyond it!). */
203
#define min(x, y) ((x) < (y) ? (x) : (y))
206
char *const end_ptr = &base_ptr[INTSORTSIZE * (total_elems - 1)];
207
char *tmp_ptr = base_ptr;
208
char *thresh = min(end_ptr, base_ptr + max_thresh);
209
register char *run_ptr;
211
/* Find smallest element in first threshold and place it at the
212
array's beginning. This is the smallest array element,
213
and the operation speeds up insertion sort's inner loop. */
215
for (run_ptr = tmp_ptr + INTSORTSIZE; run_ptr <= thresh; run_ptr += INTSORTSIZE)
216
if (INTSORTCMP ((void *) run_ptr, (void *) tmp_ptr))
219
if (tmp_ptr != base_ptr)
220
INTSORTSWAP (tmp_ptr, base_ptr);
222
/* Insertion sort, running from left-hand-side up to right-hand-side. */
224
run_ptr = base_ptr + INTSORTSIZE;
225
while ((run_ptr += INTSORTSIZE) <= end_ptr)
227
tmp_ptr = run_ptr - INTSORTSIZE;
228
while (INTSORTCMP ((void *) run_ptr, (void *) tmp_ptr))
229
tmp_ptr -= INTSORTSIZE;
231
tmp_ptr += INTSORTSIZE;
232
if (tmp_ptr != run_ptr)
236
trav = run_ptr + INTSORTSIZE;
237
while (--trav >= run_ptr)
242
for (hi = lo = trav; (lo -= INTSORTSIZE) >= tmp_ptr; hi = lo)
added
removed
src/libscotch/common_sort.c
