Skip to content

M
mariadb
Commit c32ee4ee authored by monty@donna.mysql.fi's avatar monty@donna.mysql.fi
Browse files
Options

New qsort implementation

parent ab9c0df7
Hide whitespace changes
Inline Side-by-side
Showing with 152 additions and 196 deletions
mysys/mf_qsort.c
View file @ c32ee4ee
......@@ -15,245 +15,201 @@
Software Foundation, Inc., 59 Temple Place - Suite 330, Boston,
MA 02111-1307, USA */
/* Plug-compatible replacement for UNIX qsort.
Copyright (C) 1989 Free Software Foundation, Inc.
Written by Douglas C. Schmidt (schmidt@ics.uci.edu)
Optimized and modyfied for mysys by monty.
/*
qsort implementation optimized for comparison of pointers
Inspired by the qsort implementations by Douglas C. Schmidt,
and Bentley & McIlroy's "Engineering a Sort Function".
*/
This file is part of GNU CC.
GNU QSORT is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 1, or (at your option)
any later version.
GNU QSORT is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with GNU QSORT; see the file COPYING. If not, write to
the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. */
#include "mysys_priv.h"
/* 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
#define CMP(A,B) ((*cmp)(cmp_argument,(A),(B)))
#else
#define CMP(A,B) ((*cmp)((A),(B)))
#endif
/* Byte-wise swap two items of size SIZE. */
#define SWAP(A,B,SIZE) do {int sz=(int)(SIZE); char *a = (A); char *b = (B); \
do { char _temp = *a;*a++ = *b;*b++ = _temp;} while (--sz);} while (0)
/* Copy SIZE bytes from item B to item A. */
#define COPY(A,B,SIZE) {int sz = (int) (SIZE); do { *(A)++ = *(B)++; } while (--sz); }
/* This should be replaced by a standard ANSI macro. */
#define BYTES_PER_WORD 8
/* The next 4 #defines implement a very fast in-line stack abstraction. */
#define STACK_SIZE (BYTES_PER_WORD * sizeof (long))
#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)
#define SWAP(A, B, size,swap_ptrs) \
do { \
if (swap_ptrs) \
{ \
reg1 char **a = (char**) (A), **b = (char**) (B); \
char *tmp = *a; *a++ = *b; *b++ = tmp; \
} \
else \
{ \
reg1 char *a = (A), *b = (B); \
reg3 char *end= a+size; \
do \
{ \
char tmp = *a; *a++ = *b; *b++ = tmp; \
} while (a < end); \
} \
} while (0)
/* Put the median in the middle argument */
#define MEDIAN(low, mid, high) \
{ \
if (CMP(high,low) < 0) \
SWAP(high, low, size, ptr_cmp); \
if (CMP(mid, low) < 0) \
SWAP(mid, low, size, ptr_cmp); \
else if (CMP(high, mid) < 0) \
SWAP(mid, high, size, ptr_cmp); \
}
/* Discontinue quicksort algorithm when partition gets below this size.
This particular magic number was chosen to work best on a Sparc SLC. */
#define MAX_THRESH 12
/* The following node is used to store ranges to avoid recursive calls */
/* Stack node declarations used to store unfulfilled partition obligations. */
typedef struct
typedef struct st_stack
{
char *lo;
char *hi;
} stack_node;
char *low,*high;
} STACK;
/* 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.
This reduces the probability of selecting a bad pivot value and
eliminates certain extraneous comparisons.
3. Only quicksorts TOTAL_ELEMS / MAX_THRESH partitions, leaving
insertion sort to order the MAX_THRESH items within each partition.
This is a big win, since insertion sort is faster for small, mostly
sorted array segements.
4. The larger of the two sub-partitions is always pushed onto the
stack first, with the algorithm then concentrating on the
smaller partition. This *guarantees* no more than log (n)
stack size is needed (actually O(1) in this case)! */
#define PUSH(LOW,HIGH) {stack_ptr->low = LOW; stack_ptr++->high = HIGH;}
#define POP(LOW,HIGH) {LOW = (--stack_ptr)->low; HIGH = stack_ptr->high;}
/* 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)
#define SORT_RETURN return
#else
#define SORT_RETURN return 0
#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
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)
#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
{
/* Allocating SIZE bytes for a pivot buffer facilitates a better
algorithm below since we can do comparisons directly on the pivot.
*/
int max_thresh = (int) (MAX_THRESH * size);
if (total_elems <= 1)
SORT_RETURN; /* Crashes on MSDOS if continues */
if (total_elems > MAX_THRESH)
{
char *lo = base_ptr;
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_buffer = (char *) my_alloca ((int) size);
char *low, *high, *pivot;
STACK stack[STACK_SIZE], *stack_ptr;
my_bool ptr_cmp;
/* Handle the simple case first */
/* This will also make the rest of the code simpler */
if (count <= 1)
SORT_RETURN;
low = (char*) base_ptr;
high = low+ size * (count - 1);
stack_ptr = stack + 1;
#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].lo=stack[0].hi=0;
#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
{
char *low_ptr, *high_ptr, *mid;
while (STACK_NOT_EMPTY)
count=((size_t) (high - low) / size)+1;
/* If count is small, then an insert sort is faster than qsort */
if (count < THRESHOLD_FOR_INSERT_SORT)
{
char *left_ptr;
char *right_ptr;
for (low_ptr = low + size; low_ptr <= high; low_ptr += size)
{
char *pivot = pivot_buffer;
{
/* 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 * (((uint) (hi - lo) / (uint) 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);
COPY (pivot, mid, size);
pivot = pivot_buffer;
}
left_ptr = lo + size;
right_ptr = hi - size;
/* Here's the famous ``collapse the walls'' section of quicksort.
Gotta like those tight inner loops! They are the main reason
that this algorithm runs much faster than others. */
do
{
while (CMP (left_ptr, pivot) < 0)
left_ptr += size;
while (CMP (pivot, right_ptr) < 0)
right_ptr -= size;
if (left_ptr < right_ptr)
{
SWAP (left_ptr, right_ptr, size);
left_ptr += size;
right_ptr -= size;
}
else if (left_ptr == right_ptr)
{
left_ptr += size;
right_ptr -= size;
break;
}
}
while (left_ptr <= right_ptr);
char *ptr;
for (ptr = low_ptr; ptr > low && CMP(ptr - size, ptr) > 0;
ptr -= size)
SWAP(ptr, ptr - size, size, ptr_cmp);
}
POP(low, high);
continue;
}
/* Set up pointers for next iteration. First determine whether
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. */
/* Try to find a good middle element */
mid= low + size * (count >> 1);
if (count > 40) /* Must be bigger than 24 */
{
size_t step = size* (count / 8);
MEDIAN(low, low + step, low+step*2);
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
{
MEDIAN(low, mid, high);
/* The low and high argument are already in sorted against 'pivot' */
low_ptr = low + size;
high_ptr = high - size;
}
memcpy(pivot, mid, size);
if ((right_ptr - lo) <= max_thresh)
{
if ((hi - left_ptr) <= max_thresh) /* Ignore both small parts. */
POP (lo, hi);
else /* Ignore small left part. */
lo = left_ptr;
}
else if ((hi - left_ptr) <= max_thresh) /* Ignore small right part. */
hi = right_ptr;
else if ((right_ptr - lo) > (hi - left_ptr)) /* Push larger left part */
do
{
while (CMP(low_ptr, pivot) < 0)
low_ptr += size;
while (CMP(pivot, high_ptr) < 0)
high_ptr -= size;
if (low_ptr < high_ptr)
{
PUSH (lo, right_ptr);
lo = left_ptr;
SWAP(low_ptr, high_ptr, size, ptr_cmp);
low_ptr += size;
high_ptr -= size;
}
else /* Push larger right part */
else
{
PUSH (left_ptr, hi);
hi = right_ptr;
if (low_ptr == high_ptr)
{
low_ptr += size;
high_ptr -= size;
}
break;
}
}
my_afree(pivot_buffer);
}
/* Once the BASE_PTR array is partially sorted by quicksort the rest
is completely sorted using insertion sort, since this is efficient
for partitions below MAX_THRESH size. BASE_PTR points to the beginning
of the array to sort, and END_PTR points at the very last element in
the array (*not* one beyond it!). */
{
char *end_ptr = (char*) base_ptr + size * (total_elems - 1);
char *run_ptr;
char *tmp_ptr = (char*) base_ptr;
char *thresh = min (end_ptr, (char*) base_ptr + max_thresh);
while (low_ptr <= high_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. */
/*
Prepare for next iteration.
Skip partitions of size 1 as these doesn't have to be sorted
Push the larger partition and sort the smaller one first.
This ensures that the stack is keept small.
*/
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.'
Pretty much straight out of the original GNU qsort routine. */
for (run_ptr = (char*) base_ptr + size;
(tmp_ptr = run_ptr += size) <= end_ptr; )
if ((int) (high_ptr - low) <= 0)
{
while (CMP (run_ptr, tmp_ptr -= size) < 0) ;
if ((tmp_ptr += size) != run_ptr)
if ((int) (high - low_ptr) <= 0)
{
char *trav;
for (trav = run_ptr + size; --trav >= run_ptr;)
{
char c = *trav;
char *hi, *lo;
for (hi = lo = trav; (lo -= size) >= tmp_ptr; hi = lo)
*hi = *lo;
*hi = c;
}
POP(low, high); /* Nothing more to sort */
}
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;
}
sql/sql_handler.cc
View file @ c32ee4ee
......@@ -104,7 +104,7 @@ int mysql_ha_read(THD *thd, TABLE_LIST *tables,
List_iterator<Item> it(list);
it++;
insert_fields(thd,tables,tables->name,&it);
insert_fields(thd,tables,tables->db,tables->name,&it);
table->file->index_init(keyno);
......
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