为啥不考虑直接上C的代码呢？利用iso c binding 写几个interface即可
pasuka 发表于 2014-3-4 20:43
为啥不考虑直接上C的代码呢？利用iso c binding 写几个interface即可
做这点计算，Fortran还是足以应对的。
pasuka 发表于 2014-3-4 20:43
为啥不考虑直接上C的代码呢？利用iso c binding 写几个interface即可
欢迎大家把自己熟悉的代码拿过来做对比，我发这个贴只是抛砖引玉
算法一样时，计算机语言、编译器、系统等不同效率可能都有差别，所以可以让感兴趣的爱好者一起来参与
chuxf 发表于 2014-3-2 20:17
我这里找了三个快速排序算法。
楼主这个，我称为 QuickSortMod，Intel Fortran 扩展的 Qsort，以及 Juli R ...
简单测试了一下，分别调用QuickSortMod和Juli Rew 的qsort（修改为整数排序）以及gcc自带的qsort，随机整型数组的长度由10的1次方增加到10的7次方（受制于内存，未测试更大的动态数组）测试平台：
OS： Debian 7.2 32bit
CPU： ARMv6-compatible processor 700Mhz OC 1000Mhz
Memory: 437.7MB
CC： gcc-4.7
FC： gfortran-4.7
计算结果：
=========================================================================
Fortran version of Quick Sort I
Len of array: 10 Time =0.000 seconds.
Fortran version of Quick Sort II
Len of array: 10 Time =0.000 seconds.
Fortran call C version of qsort
Len of array: 10 Time =0.000 seconds.
Fortran version of Quick Sort I
Len of array: 100 Time =0.000 seconds.
Fortran version of Quick Sort II
Len of array: 100 Time =0.000 seconds.
Fortran call C version of qsort
Len of array: 100 Time =0.000 seconds.
Fortran version of Quick Sort I
Len of array: 1000 Time =0.000 seconds.
Fortran version of Quick Sort II
Len of array: 1000 Time =0.000 seconds.
Fortran call C version of qsort
Len of array: 1000 Time =0.010 seconds.
Fortran version of Quick Sort I
Len of array: 10000 Time =0.010 seconds.
Fortran version of Quick Sort II
Len of array: 10000 Time =0.020 seconds.
Fortran call C version of qsort
Len of array: 10000 Time =0.010 seconds.
Fortran version of Quick Sort I
Len of array: 100000 Time =0.180 seconds.
Fortran version of Quick Sort II
Len of array: 100000 Time =0.160 seconds.
Fortran call C version of qsort
Len of array: 100000 Time =0.110 seconds.
Fortran version of Quick Sort I
Len of array: 1000000 Time =1.500 seconds.
Fortran version of Quick Sort II
Len of array: 1000000 Time =1.750 seconds.
Fortran call C version of qsort
Len of array: 1000000 Time =1.370 seconds.
Fortran version of Quick Sort I
Len of array:10000000 Time = 17.660 seconds.
Fortran version of Quick Sort II
Len of array:10000000 Time = 20.340 seconds.
Fortran call C version of qsort
Len of array:10000000 Time = 16.110 seconds.
*******************************************************************************************************************************
初步意见：
1、ISO_C_BINDING对于混合编程相当便捷，fortran可以方便的使用成熟的C代码；
2、大数组排序，在linux或mingw环境下直接调用qsort更具优势
********************************************************************************************************************************
主程序
program SortTest
use QuickSortMod
use qsort_c_module
implicit none
interface
subroutine qsort(a,length)bind(c,name="sort_integer")
use,intrinsic::iso_c_binding
integer(c_int),value::length
integer(c_int)::a(length)
end subroutine
end interface
integer::N,i
integer,allocatable::IntDat(:),intDat2(:),intdat3(:)
real,allocatable::dat(:)
real::start,finish
do i=1,7
N=10**i
if(allocated(intdat))deallocate(intdat)
if(allocated(intdat2))deallocate(intdat2)
if(allocated(dat))deallocate(dat)
if(allocated(intdat3))deallocate(intdat3)
allocate(IntDat(n),intDat2(n),dat(n),intdat3(n))
call random_seed()
call random_number(dat)
intDat = int((2*Dat-1)*100)
intdat2 = intdat
intdat3 = intdat
call cpu_time(start)
call quick_sort_i(intdat)
call cpu_time(finish)
write(*,*)"Fortran version of Quick Sort I"
write(*,'("Len of array:",i10," Time = ",f6.3," seconds.")')n,finish-start
call cpu_time(start)
call QsortC(intdat3)
call cpu_time(finish)
write(*,*)"Fortran version of Quick Sort II"
write(*,'("Len of array:",i10," Time = ",f6.3," seconds.")')n,finish-start
call cpu_time(start)
call qsort(intdat2,n)
call cpu_time(finish)
write(*,*)"Fortran call C version of qsort"
write(*,'("Len of array:",i10," Time = ",f6.3," seconds.")')n,finish-start
deallocate(intdat,intdat2,dat,intdat3)
enddo
end program
c接口程序代码：
#include <stdlib.h>
#include <stdio.h>
/* qsort int comparison function */
int int_cmp(const void *a, const void *b)
{
const int *ia = (const int *)a; // casting pointer types
const int *ib = (const int *)b;
return *ia- *ib;
/* integer comparison: returns negative if b > a
and positive if a > b */
}
/* integer array printing function */
void print_int_array(const int *array, size_t len)
{
size_t i;
for(i=0; i<len; i++)
printf("%d | ", array);
putchar('\n');
}
void sort_integer(int *a,size_t a_len)
{
/* print original integer array */
//print_int_array(a, a_len);
/* sort array using qsort functions */
qsort(a, a_len, sizeof(int), int_cmp);
}
pasuka 发表于 2014-3-10 13:25
简单测试了一下，分别调用QuickSortMod和Juli Rew 的qsort（修改为整数排序）以及gcc自带的qsort，随机整 ...
从结果来看，gcc自带的qsort貌似效率最高，不知fortran的代码编译时是否优化？
但QuickSortMod和Juli Rew的QSortC跟石头测出来的效率对比图恰恰相反，是否修改时出了点问题？
aliouying 发表于 2014-3-11 18:57
从结果来看，gcc自带的qsort貌似效率最高，不知fortran的代码编译时是否优化？
但QuickSortMod和Juli Rew ...
gcc和gfortran都是默认的优化设置，没有修改过
我只是去掉了QuickSortMod的interface，因为gfotran编译报错
aliouying 发表于 2014-3-11 18:57
从结果来看，gcc自带的qsort貌似效率最高，不知fortran的代码编译时是否优化？
但QuickSortMod和Juli Rew ...
如果有兴趣，还可以gfortran调用gsl再测试一下
pasuka 发表于 2014-3-11 20:26
如果有兴趣，还可以gfortran调用gsl再测试一下
我在写毕业论文，所以没多少时间，等过段时间测试下
排序是个很好的课题，方法应该有很多种，期待看到别的排序方法，都拿来比较下
快排纵然好理解，但是个人总感觉递归的效率可能不高，正如Davis将所有递归利用压栈弹栈来操作一样，虽然在汇编级递归确实也是不断的弹栈压栈过程。。。
不知诸君有何见解？