linux下应用程序性能剖分神器gprofiler-tools-安装和使用

  最近在摆弄算法的的优化,需要剖分一下算法的瓶颈,就找了一些代码剖分工具,其中

gprofileer-tools是很不错的工具,gperftools时google开源的一款C++性能分析分析工具,github项目地址如下:

   https://github.com/gperftools/gperftools

  一 下载和安装:

  1.git clone: 

       2 . 安装:

   根目录下执行:sh autogen.sh

          然后执行:./configure

            最后:make all && sudo make install

          安装成功。

   如果没有安装libunwind会出现如,configure执行后报警告:   

configure: WARNING: No frame pointers and no libunwind. Using experimental backtrace capturing via libgcc. Expect crashy cpu profiler.

   解决方法:

使用命令apt-get install libunwind8-dev安装libunwind即可。

  3 最后使用ldconfig更新一下库文件即可。

  二 使用和实例:

  1 使用gprofiler-tools需要在的代码中添加如下几段代码:

   A头文件:

    #include <gperftools/profiler.h>

   B 起止函数:

    修改程序的源代码使得要profiler的代码段包含在ProfilerStart("name");ProfilerStop();

   C链接文件添加需要链接的宏:

    lprofiler //cpu性能

    -ltcmalloc  //heap资源

  2 实例:

#include <stdio.h>
#include <time.h>
#include <stdlib.h>
#include <gperftools/profiler.h>
#include <unistd.h>

void consumeSomeCPUTime1(int input){ 
    int i = 0;  
    input++; 
    while(i++ < 10000){ 
        i--;  i++;  i--;  i++; 
    }   
}; 

void consumeSomeCPUTime2(int input){ 
    input++; 
    consumeSomeCPUTime1(input); 
    int i = 0;  
    while(i++ < 10000){ 
        i--;  i++;  i--;  i++; 
    }   
}; 

int stupidComputing(int a, int b){ 
    int i = 0;  
    while( i++ < 10000){  
        consumeSomeCPUTime1(i); 
    }   
    int j = 0;  
    while(j++ < 5000){ 
        consumeSomeCPUTime2(j); 
    }   
    return a+b; 
}; 

int smartComputing(int a, int b){ 
    return a+b; 
}; 

int main()
{ 
    int i = 0;
    printf("reached the start point of performance bottle neck
"); 
    sleep(5);
    ProfilerStart("CPUProfile");
    while( i++ < 10){ 
        printf("Stupid computing return : %d
",stupidComputing(i, i+1)); 
        printf("Smart computing return %d
",smartComputing(i+1, i+2)); 
    }   
    printf("should teminate profiling now.
");  
    ProfilerStop();
    sleep(5);

    return 0;
}

  编译运行:

gcc -g test.c -o prog -lprofiler

  

export CPUPROFILE=info.prof

  执行:

pprof -text prog info.prof

  结果:

Using local file prog.
Using local file info.prof.
Total: 1348 samples
    1017  75.4%  75.4%     1017  75.4% consumeSomeCPUTime1
     331  24.6% 100.0%      674  50.0% consumeSomeCPUTime2
       0   0.0% 100.0%     1348 100.0% __libc_start_main
       0   0.0% 100.0%     1348 100.0% _start
       0   0.0% 100.0%     1348 100.0% main
       0   0.0% 100.0%     1348 100.0% stupidComputing

 

报错一:

Google perftool cannot read file “libprofiler.so.0”

解决方法:

sudo /sbin/ldconfig 

错误二:

Using local file prog.
Use of uninitialized value $host in substitution (s///) at /usr/local/bin/pprof line 3366.
Use of uninitialized value $hostport in concatenation (.) or string at /usr/local/bin/pprof line 3368.
Use of uninitialized value $prefix in concatenation (.) or string at /usr/local/bin/pprof line 3368.
Use of uninitialized value $host in substitution (s///) at /usr/local/bin/pprof line 3366.

解决方法:

env CPUPROFILE=/tmp/mybin.prof /home/gprofiler-tools/prog

参考资料:

1 https://www.kancloud.cn/subin/blog/619133

2 https://github.com/gperftools/gperftools

3 https://stackoverflow.com/questions/1581494/google-perftool-cannot-read-file-libprofiler-so-0

4     https://www.ibm.com/developerworks/cn/linux/l-cn-googleperf/index.html

5   https://stackoverflow.com/questions/24086867/cant-analyse-the-output-the-profile-file-of-gperftools-profiler

原文地址:https://www.cnblogs.com/dylancao/p/7683960.html