[英]what is a good way to run experiments for the memory usage of an algorithm in C++?
I have algorithm A
and algorithm B
that were implemented in C++. 我有用C ++实现的算法
A
和算法B
A
uses more space than B
in theory, and it turns out this is also the case in the practice. 理论上,
A
比B
使用更多的空间,事实证明,在实践中也是如此。 I would like to generate some nice graphs to illustrate this. 我想生成一些漂亮的图形来说明这一点。 Both algorithms receive an input
n
and I would like my experiments to vary for different n
, so the x axis of the graph must be something like n = 10^6, 2*10^6, ...
两种算法都接收输入
n
,我希望我的实验针对不同的n
进行变化,因此图的x轴必须类似n = 10^6, 2*10^6, ...
Usually when it comes to data like time or cache misses, my most preferred way of setting up the experiments is as follows. 通常,当涉及到时间或缓存丢失之类的数据时,我最喜欢的实验设置方法如下。 Inside a C++ file I have the algorithm that is implemented like this:
在C ++文件中,我具有如下实现的算法:
#include <iostream>
using namespace std;
int counters[1000];
void init_statistics(){
//use some library for example papi (http://icl.cs.utk.edu/papi/software/)
//to start counting, store the results in the counters array
}
void stop_statistics(){
//this is just to stop counting
}
int algA(int n){
//algorithm code
int result = ...
return result;
}
void main(int argc, const char * argv[]){
int n = atoi(argv[1]);
init_statistics(); //function that initializes the statistic counters
int res = algA(n);
end_statistics(); //function that ends the statistics counters
cout<<res<<counter[0]<<counter[1]<<....<<endl;
}
I would then create a python script that for different n
calls result = subprocess.check_output(['./algB',...])
. 然后,我将创建一个python脚本,用于不同的
n
调用result = subprocess.check_output(['./algB',...])
。 After that, parse the result string in python and print it in a suitable format. 之后,用python解析结果字符串并以适当的格式打印。 For example if I used R for the plots, I could print the data to an external file, where each counter is separated by a
\\t
. 例如,如果将R用于绘图,则可以将数据打印到外部文件中,其中每个计数器用
\\t
分隔。
This has worked very well for me, but now is the first time that I need data about the space used by the algorithm, and I am not sure how to count this space. 这对我来说效果很好,但是现在是我第一次需要有关算法使用的空间的数据,而且我不确定如何计算该空间。 One way would be to use valgrind, this is a possible output by valgrind:
一种方法是使用valgrind,这是valgrind可能的输出:
==15447== Memcheck, a memory error detector
==15447== Copyright (C) 2002-2015, and GNU GPL'd, by Julian Seward et al.
==15447== Using Valgrind-3.11.0 and LibVEX; rerun with -h for copyright info
==15447== Command: ./algB 1.txt 2.txt
==15447==
==15447==
==15447== HEAP SUMMARY:
==15447== in use at exit: 72,704 bytes in 1 blocks
==15447== total heap usage: 39 allocs, 38 frees, 471,174,306 bytes allocated
==15447==
==15447== LEAK SUMMARY:
==15447== definitely lost: 0 bytes in 0 blocks
==15447== indirectly lost: 0 bytes in 0 blocks
==15447== possibly lost: 0 bytes in 0 blocks
==15447== still reachable: 72,704 bytes in 1 blocks
==15447== suppressed: 0 bytes in 0 blocks
==15447== Rerun with --leak-check=full to see details of leaked memory
==15447==
==15447== For counts of detected and suppressed errors, rerun with: -v
==15447== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 0 from 0)
The interesting number is 471,174,306 bytes
. 有趣的数字是
471,174,306 bytes
。 However, valgrind slows down the execution time a lot, and at the same time doesn't just return this number but this large string. 但是,valgrind大大减慢了执行时间,同时,不仅返回此数字,而且返回了这个大字符串。 And I am not sure how to parse it because for some reason if with python I call
result = subprocess.check_output(['valgrind','./algB',...])
, the result
string only stores the output by ./algB
and completely ignores what valgrind returns. 而且我不确定如何解析它,因为由于某种原因,如果使用python我调用
result = subprocess.check_output(['valgrind','./algB',...])
,则result
字符串仅存储by的输出./algB
并且完全忽略valgrind返回的内容。
thank you in advace! 谢谢你的放心!
memcheck
是用于发现内存泄漏的工具,您应该使用massif
(valgrind中的另一种工具)进行内存分配分析。
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.