上一篇实现，是统计了每一次函数调用的层次关系和耗时。如果在函数调用比较多的情况下，更关心的是减少输出，只需要总计的耗时即可。另外如果是后台程序，输出到文件会更好。

成都创新互联坚持“要么做到，要么别承诺”的工作理念，服务领域包括：做网站、网站设计、企业官网、英文网站、手机端网站、网站推广等服务，满足客户于互联网时代的上党网站设计、移动媒体设计的需求，帮助企业找到有效的互联网解决方案。努力成为您成熟可靠的网络建设合作伙伴！

因此，对程序做了改进。当设置TRACELOG环境变量，则输出到这个环境变量指定的文件中。如果设置了DETAILF，则会输出调用层次关系和每一次的耗时，否则只输出总计时间。

tr.cpp改进后的程序如下。

#include

#include

#include

#include

#include

#include

using std::vector;

using std::map;

using std::string;

using std::clog;

using std::endl;

using std::ofstream;

using std::ostream;

extern "C" void print_trace(const char * const procname);

struct Stat

{

    int lvl;

    string name;

    long stm;

    long etm;

    long oitv;

    Stat(int l, const string& s, long st) : lvl(l), name(s), stm(st), etm(0), oitv(0) {}

};

struct Sum

{

    int cnt;

    long itv;

    Sum() : cnt(0), itv(0) {}

    Sum operator+=(long i)

    {

        ++cnt;

        itv+=i;

        return *this;

    }

};

namespace

{

vector tracev;

map itvm;

int clvl = 0;

bool detailf = (getenv("DETAILF")!=NULL);

}

extern "C" void

__func_trace_enter(const char * const proc_name,

                   const char * const file_name,

                   const int line_no,

                   void ** const id)

{

// printf("{ %s (%s:%d) %p %s\n", proc_name, file_name, line_no, id[0], (char*)*id);

    struct timeval nowtm;

    gettimeofday(&nowtm, NULL);

    ++clvl;

    tracev.push_back(Stat(clvl, string(proc_name)+"() : "+file_name, nowtm.tv_sec * 1000000 + nowtm.tv_usec));

}

extern "C" void

__func_trace_exit(const char * const proc_name,

                  const char * const file_name,

                  const int line_no,

                  void ** const id)

{

// printf("} %s (%s:%d) %p %s\n", proc_name, file_name, line_no, id[0], (char*)*id);

    struct timeval nowtm;

    int itv;

    gettimeofday(&nowtm, NULL);

    auto iter = tracev.end() - 1;

    while (iter->etm != 0)

        --iter;

    iter->etm = nowtm.tv_sec * 1000000 + nowtm.tv_usec;

    itv = iter->etm - iter->stm - iter->oitv;

    itvm[proc_name]+=itv;

    for (auto s = tracev.begin(); s!=iter; ++s)

    {

        if (s->etm == 0)

            s->oitv += itv;

    }

    --clvl;

    if (!detailf)

        tracev.erase(iter, tracev.end());

    if (clvl == 0)

    {

        print_trace(proc_name);

        tracev.clear();

        itvm.clear(); 

    }

}

extern "C" void print_trace(const char * const procname)

{

    time_t t;

    char buf[30];

    ofstream logf(getenv("TRACELOG"), std::ios::app);

    ostream &log = logf ? logf : clog;

    if (detailf)

    {

        log<<"Detail of "<< procname <<"() :\n";

        for (auto s = tracev.begin(); s!=tracev.end(); ++s)

        {

            log << s->lvl << "\t";

            t=s->stm/1000000;

            strftime(buf, sizeof(buf), "%Y%m%d%H%M%S", localtime(&t));

            log << buf << "." << s->stm % 1000000 << "\t";

            t=s->etm/1000000;

            strftime(buf, sizeof(buf), "%Y%m%d%H%M%S", localtime(&t));

            log << buf << "." << s->etm % 1000000 << "\t";

            log << s->etm-s->stm << "\t" << s->oitv << "\t" << s->etm-s->stm-s->oitv << "\t";

            log << string(s->lvl-1, ' ') << s->name << "\n";

        }

    }

    log<<"Total of "<< procname <<"() :\n";

    for (auto s = itvm.begin(); s!=itvm.end(); ++s)

    {

        log << s->first << "()\t count "<< (s->second).cnt << "\ttime " << (s->second).itv << "\n";

    }

    log<

}

当前文章：AIX下C/C++函数性能统计实现方法--改进版-创新互联
标题来源：http://scyanting.com/article/cshdjh.html