Skip to content

M
MariaDB
Commit 81dc029a authored by unknown's avatar unknown
Browse files
Options

mf_qsort.c: 

  qsort implementation backported from 4.0 tree, the old one was buggy


mysys/mf_qsort.c:
  qsort implementation backported from 4.0 tree, the old one was buggy
parent 4c3ed5e9
Show whitespace changes
Inline Side-by-side
Showing with 164 additions and 196 deletions
mysys/mf_qsort.c
View file @ 81dc029a
/* Copyright (C) 1991, 1992, 1996, 1997 Free Software Foundation, Inc. /* Copyright (C) 2000 MySQL AB
This file is part of the GNU C Library.
Written by Douglas C. Schmidt (schmidt@ics.uci.edu).
The GNU C Library is free software; you can redistribute it and/or This program is free software; you can redistribute it and/or modify
modify it under the terms of the GNU Library General Public License as it under the terms of the GNU General Public License as published by
published by the Free Software Foundation; either version 2 of the the Free Software Foundation; either version 2 of the License, or
License, or (at your option) any later version. (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful, This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
Library General Public License for more details. GNU General Public License for more details.
You should have received a copy of the GNU Library General Public You should have received a copy of the GNU General Public License
License along with the GNU C Library; see the file COPYING.LIB. If not, along with this program; if not, write to the Free Software
write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */
Boston, MA 02111-1307, USA. */
/* /*
Modifications by monty: qsort implementation optimized for comparison of pointers
- Uses mysys include files Inspired by the qsort implementations by Douglas C. Schmidt,
- Small fixes to make the it a bit faster and Bentley & McIlroy's "Engineering a Sort Function".
- Can be compiled with a cmp function that takes one extra argument.
*/ */
#include "mysys_priv.h" #include "mysys_priv.h"
#ifndef SCO
#include <m_string.h> #include <m_string.h>
#endif
/* Envoke the comparison function, returns either 0, < 0, or > 0. */ /* We need to use qsort with 2 different compare functions */
#ifdef QSORT_EXTRA_CMP_ARGUMENT #ifdef QSORT_EXTRA_CMP_ARGUMENT
#define CMP(A,B) ((*cmp)(cmp_argument,(A),(B))) #define CMP(A,B) ((*cmp)(cmp_argument,(A),(B)))
#else #else
#define CMP(A,B) ((*cmp)((A),(B))) #define CMP(A,B) ((*cmp)((A),(B)))
#endif #endif
/* Byte-wise swap two items of size SIZE. */ #define SWAP(A, B, size,swap_ptrs) \
#define SWAP(a, b, size) \ do { \
do \ if (swap_ptrs) \
{ \
reg1 char **a = (char**) (A), **b = (char**) (B); \
char *tmp = *a; *a++ = *b; *b++ = tmp; \
} \
else \
{ \ { \
register size_t __size = (size); \ reg1 char *a = (A), *b = (B); \
register char *__a = (a), *__b = (b); \ reg3 char *end= a+size; \
do \ do \
{ \ { \
char __tmp = *__a; \ char tmp = *a; *a++ = *b; *b++ = tmp; \
*__a++ = *__b; \ } while (a < end); \
*__b++ = __tmp; \ } \
} while (--__size > 0); \ } while (0)
} while (0)
/* Put the median in the middle argument */
/* Discontinue quicksort algorithm when partition gets below this size. #define MEDIAN(low, mid, high) \
This particular magic number was chosen to work best on a Sun 4/260. */ { \
#define MAX_THRESH 8 if (CMP(high,low) < 0) \
SWAP(high, low, size, ptr_cmp); \
/* Stack node declarations used to store unfulfilled partition obligations. */ if (CMP(mid, low) < 0) \
typedef struct _qsort_stack_node SWAP(mid, low, size, ptr_cmp); \
{ else if (CMP(high, mid) < 0) \
char *lo; SWAP(mid, high, size, ptr_cmp); \
char *hi; }
} stack_node;
/* The next 4 #defines implement a very fast in-line stack abstraction. */
#define STACK_SIZE (8 * sizeof(unsigned long int))
#define PUSH(LOW,HIGH) do {top->lo = LOW;top++->hi = HIGH;} while (0)
#define POP(LOW,HIGH) do {LOW = (--top)->lo;HIGH = top->hi;} while (0)
#define STACK_NOT_EMPTY (stack < top)
/* Order size using quicksort. This implementation incorporates
four optimizations discussed in Sedgewick:
1. Non-recursive, using an explicit stack of pointer that store the
next array partition to sort. To save time, this maximum amount
of space required to store an array of MAX_INT is allocated on the
stack. Assuming a 32-bit integer, this needs only 32 *
sizeof (stack_node) == 136 bits. Pretty cheap, actually.
2. Chose the pivot element using a median-of-three decision tree. /* The following node is used to store ranges to avoid recursive calls */
This reduces the probability of selecting a bad pivot value and
eliminates certain extraneous comparisons.
3. Only quicksorts TOTAL_ELEMS / MAX_THRESH partitions, leaving typedef struct st_stack
insertion sort to order the MAX_THRESH items within each partition. {
This is a big win, since insertion sort is faster for small, mostly char *low,*high;
sorted array segments. } stack_node;
4. The larger of the two sub-partitions is always pushed onto the #define PUSH(LOW,HIGH) {stack_ptr->low = LOW; stack_ptr++->high = HIGH;}
stack first, with the algorithm then concentrating on the #define POP(LOW,HIGH) {LOW = (--stack_ptr)->low; HIGH = stack_ptr->high;}
smaller partition. This *guarantees* no more than log (n)
stack size is needed (actually O(1) in this case)! */
/* The following stack size is enough for ulong ~0 elements */
#define STACK_SIZE (8 * sizeof(unsigned long int))
#define THRESHOLD_FOR_INSERT_SORT 10
#if defined(QSORT_TYPE_IS_VOID) #if defined(QSORT_TYPE_IS_VOID)
#define SORT_RETURN return #define SORT_RETURN return
#else #else
#define SORT_RETURN return 0 #define SORT_RETURN return 0
#endif #endif
/****************************************************************************
** 'standard' quicksort with the following extensions:
**
** Can be compiled with the qsort2_cmp compare function
** Store ranges on stack to avoid recursion
** Use insert sort on small ranges
** Optimize for sorting of pointers (used often by MySQL)
** Use median comparison to find partition element
*****************************************************************************/
#ifdef QSORT_EXTRA_CMP_ARGUMENT #ifdef QSORT_EXTRA_CMP_ARGUMENT
qsort_t qsort2(void *base_ptr, size_t total_elems, size_t size, qsort2_cmp cmp, qsort_t qsort2(void *base_ptr, size_t count, size_t size, qsort2_cmp cmp,
void *cmp_argument) void *cmp_argument)
#else #else
qsort_t qsort(void *base_ptr, size_t total_elems, size_t size, qsort_cmp cmp) qsort_t qsort(void *base_ptr, size_t count, size_t size, qsort_cmp cmp)
#endif #endif
{ {
/* Allocating SIZE bytes for a pivot buffer facilitates a better char *low, *high, *pivot;
algorithm below since we can do comparisons directly on the pivot. stack_node stack[STACK_SIZE], *stack_ptr;
*/ my_bool ptr_cmp;
size_t max_thresh = (size_t) (MAX_THRESH * size); /* Handle the simple case first */
if (total_elems <= 1) /* This will also make the rest of the code simpler */
SORT_RETURN; /* Crashes on MSDOS if continues */ if (count <= 1)
SORT_RETURN;
if (total_elems > MAX_THRESH) low = (char*) base_ptr;
{ high = low+ size * (count - 1);
char *lo = (char*) base_ptr; stack_ptr = stack + 1;
char *hi = &lo[size * (total_elems - 1)];
stack_node stack[STACK_SIZE]; /* Largest size needed for 32-bit int!!! */
stack_node *top = stack + 1;
char *pivot = (char *) my_alloca ((int) size);
#ifdef HAVE_purify #ifdef HAVE_purify
stack[0].lo=stack[0].hi=0; /* The first element in the stack will be accessed for the last POP */
stack[0].low=stack[0].high=0;
#endif #endif
pivot = (char *) my_alloca((int) size);
ptr_cmp= size == sizeof(char*) && !((low - (char*) 0)& (sizeof(char*)-1));
/* The following loop sorts elements between high and low */
do do
{ {
char *left_ptr,*right_ptr; char *low_ptr, *high_ptr, *mid;
/* Select median value from among LO, MID, and HI. Rearrange
LO and HI so the three values are sorted. This lowers the
probability of picking a pathological pivot value and
skips a comparison for both the LEFT_PTR and RIGHT_PTR. */
char *mid = lo + size * (((ulong) (hi - lo) / (ulong) size) >> 1);
if (CMP(hi,lo) < 0)
SWAP (hi, lo, size);
if (CMP (mid, lo) < 0)
SWAP (mid, lo, size);
else if (CMP (hi, mid) < 0)
SWAP (mid, hi, size);
memcpy (pivot, mid, size);
left_ptr = lo + size;
right_ptr = hi - size;
/* Here's the famous ``collapse the walls'' section of quicksort. count=((size_t) (high - low) / size)+1;
Gotta like those tight inner loops! They are the main reason /* If count is small, then an insert sort is faster than qsort */
that this algorithm runs much faster than others. */ if (count < THRESHOLD_FOR_INSERT_SORT)
do
{ {
while (CMP (left_ptr, pivot) < 0) for (low_ptr = low + size; low_ptr <= high; low_ptr += size)
left_ptr += size;
while (CMP (pivot, right_ptr) < 0)
right_ptr -= size;
if (left_ptr < right_ptr)
{ {
SWAP (left_ptr, right_ptr, size); char *ptr;
left_ptr += size; for (ptr = low_ptr; ptr > low && CMP(ptr - size, ptr) > 0;
right_ptr -= size; ptr -= size)
SWAP(ptr, ptr - size, size, ptr_cmp);
} }
else if (left_ptr == right_ptr) POP(low, high);
continue;
}
/* Try to find a good middle element */
mid= low + size * (count >> 1);
if (count > 40) /* Must be bigger than 24 */
{ {
left_ptr += size; size_t step = size* (count / 8);
right_ptr -= size; MEDIAN(low, low + step, low+step*2);
break; MEDIAN(mid - step, mid, mid+step);
MEDIAN(high - 2 * step, high-step, high);
/* Put best median in 'mid' */
MEDIAN(low+step, mid, high-step);
low_ptr = low;
high_ptr = high;
} }
else else
break; /* left_ptr > right_ptr */ {
MEDIAN(low, mid, high);
/* The low and high argument are already in sorted against 'pivot' */
low_ptr = low + size;
high_ptr = high - size;
} }
while (left_ptr <= right_ptr); memcpy(pivot, mid, size);
/* Set up pointers for next iteration. First determine whether do
left and right partitions are below the threshold size. If so,
ignore one or both. Otherwise, push the larger partition's
bounds on the stack and continue sorting the smaller one. */
if ((size_t) (right_ptr - lo) <= max_thresh)
{ {
if ((size_t) (hi - left_ptr) <= max_thresh) while (CMP(low_ptr, pivot) < 0)
POP (lo, hi); /* Ignore both small partitions. */ low_ptr += size;
else while (CMP(pivot, high_ptr) < 0)
lo = left_ptr; /* Ignore small left part. */ high_ptr -= size;
}
else if ((size_t) (hi - left_ptr) <= max_thresh) if (low_ptr < high_ptr)
hi = right_ptr; /* Ignore small right partition. */
else if ((right_ptr - lo) > (hi - left_ptr))
{ {
PUSH (lo, right_ptr); /* Push larger left part */ SWAP(low_ptr, high_ptr, size, ptr_cmp);
lo = left_ptr; low_ptr += size;
high_ptr -= size;
} }
else else
{ {
PUSH (left_ptr, hi); /* Push larger right part */ if (low_ptr == high_ptr)
hi = right_ptr; {
low_ptr += size;
high_ptr -= size;
} }
} while (STACK_NOT_EMPTY); break;
my_afree(pivot);
} }
}
while (low_ptr <= high_ptr);
/* Once the BASE_PTR array is partially sorted by quicksort the rest /*
is completely sorted using insertion sort, since this is efficient Prepare for next iteration.
for partitions below MAX_THRESH size. BASE_PTR points to the beginning Skip partitions of size 1 as these doesn't have to be sorted
of the array to sort, and END_PTR points at the very last element in Push the larger partition and sort the smaller one first.
the array (*not* one beyond it!). */ This ensures that the stack is keept small.
*/
{
char *end_ptr = (char*) base_ptr + size * (total_elems - 1);
char *tmp_ptr = (char*) base_ptr;
char *thresh = min (end_ptr, (char*) base_ptr + max_thresh);
register char *run_ptr;
/* Find smallest element in first threshold and place it at the
array's beginning. This is the smallest array element,
and the operation speeds up insertion sort's inner loop. */
for (run_ptr = tmp_ptr + size; run_ptr <= thresh; run_ptr += size)
if (CMP (run_ptr, tmp_ptr) < 0)
tmp_ptr = run_ptr;
if (tmp_ptr != (char*) base_ptr)
SWAP (tmp_ptr, (char*) base_ptr, size);
/* Insertion sort, running from left-hand-side up to right-hand-side. */
for (run_ptr = (char*) base_ptr + size; if ((int) (high_ptr - low) <= 0)
(run_ptr += size) <= end_ptr; )
{
if (CMP (run_ptr, (tmp_ptr = run_ptr-size)) < 0)
{ {
char *trav; if ((int) (high - low_ptr) <= 0)
while (CMP (run_ptr, tmp_ptr -= size) < 0) ;
tmp_ptr += size;
/* Shift down all smaller elements, put found element in 'run_ptr' */
for (trav = run_ptr + size; --trav >= run_ptr;)
{ {
char c = *trav; POP(low, high); /* Nothing more to sort */
char *hi, *lo;
for (hi = lo = trav; (lo -= size) >= tmp_ptr; hi = lo)
*hi = *lo;
*hi = c;
} }
else
low = low_ptr; /* Ignore small left part. */
} }
else if ((int) (high - low_ptr) <= 0)
high = high_ptr; /* Ignore small right part. */
else if ((high_ptr - low) > (high - low_ptr))
{
PUSH(low, high_ptr); /* Push larger left part */
low = low_ptr;
} }
else
{
PUSH(low_ptr, high); /* Push larger right part */
high = high_ptr;
} }
} while (stack_ptr > stack);
my_afree(pivot);
SORT_RETURN; SORT_RETURN;
} }
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7