Programming

Hi all,

I just wrote a small problem to check gettimeofday in a multi-cores environment.

Code:

int timediff(double fTimeQvs, double fTimeTcp)
{
        int iTimeDiff;
        iTimeDiff = (((((int)fTimeQvs)/10000) - (((int)fTimeTcp)/10000)) * 3600) * 1000000;
        iTimeDiff += ((((((int)fTimeQvs)/100)%100) - ((((int)fTimeTcp)/100)%100)) * 60) * 1000000;
        iTimeDiff += ((((int)fTimeQvs)%100) - (((int)fTimeTcp)%100)) * 1000000;
        iTimeDiff += (((long long)(fTimeQvs*1000000))%1000000) - (((long long)(fTimeTcp*1000000))%1000000);

    return iTimeDiff;
}

int main()
{
        struct timeval cur;
        gettimeofday(&cur, 0);

        for(;;)
        {
                struct timeval new_time;
                gettimeofday(&new_time, 0);

                double diff = difftime(new_time.tv_sec,cur.tv_sec);
                if(diff > 1.0)
                        printf("Big:%ld.%ld,%ld.%ld,%lf\n",new_time.tv_sec,new_time.tv_sec, cur.tv_sec,cur.tv_usec,diff);
                else if(diff < 0.0)
                        printf("Small::%ld.%ld,%ld.%ld,%lf\n",new_time.tv_sec,new_time.tv_usec, cur.tv_sec,cur.tv_usec,diff);
                cur.tv_sec = new_time.tv_sec;
        }
        return 0;
}

Here is the result:
Small::1217485404.98014,1217489802.813903,-4398.000000
Big:1217489802.1217489802,1217485404.813903,4398.000000
Small::1217485404.99014,1217489802.813903,-4398.000000
Big:1217489802.1217489802,1217485404.813903,4398.000000
Small::1217485404.100014,1217489802.813903,-4398.000000
Big:1217489802.1217489802,1217485404.813903,4398.000000
Small::1217485404.101013,1217489802.813903,-4398.000000
Big:1217489802.1217489802,1217485404.813903,4398.000000
Small::1217485404.102014,1217489802.813903,-4398.000000
Big:1217489802.1217489802,1217485404.813903,4398.000000
Small::1217485404.103014,1217489802.813903,-4398.000000
Big:1217489802.1217489802,1217485404.813903,4398.000000
Small::1217485404.104014,1217489802.813903,-4398.000000
Big:1217489802.1217489802,1217485404.813903,4398.000000
Small::1217485404.105014,1217489802.813903,-4398.000000
Big:1217489802.1217489802,1217485404.813903,4398.000000
Small::1217485404.106014,1217489802.813903,-4398.000000
Big:1217489802.1217489802,1217485404.813903,4398.000000
Small::1217485404.107013,1217489802.813903,-4398.000000
Big:1217489802.1217489802,1217485404.813903,4398.000000
Small::1217485404.108013,1217489802.813903,-4398.000000
Big:1217489802.1217489802,1217485404.813903,4398.000000
Small::1217485404.109013,1217489802.813903,-4398.000000
Big:1217489802.1217489802,1217485404.813903,4398.000000
Small::1217485404.110014,1217489802.813903,-4398.000000
Big:1217489802.1217489802,1217485404.813903,4398.000000
Small::1217485404.111014,1217489802.813903,-4398.000000
Big:1217489802.1217489802,1217485404.813903,4398.000000
Small::1217485404.112014,1217489802.813903,-4398.000000
Big:1217489802.1217489802,1217485404.813903,4398.000000
Small::1217485404.113014,1217489802.813903,-4398.000000
Big:1217489802.1217489802,1217485404.813903,4398.000000
Small::1217485404.114014,1217489802.813903,-4398.000000

so is there any way to overcome it?

First off, try man timercmp. If you are on older systems it may not exist.
Otherwise -- Look for something like this in your sys/time.h file

Code:

#  define timerisset(tvp)               ((tvp)->tv_sec || (tvp)->tv_usec)
#  define timercmp(tvp, uvp, cmp) \
          ((tvp)->tv_sec cmp (uvp)->tv_sec || \
           (tvp)->tv_sec == (uvp)->tv_sec && (tvp)->tv_usec cmp (uvp)->tv_usec)
#  define timerclear(tvp)               ((tvp)->tv_sec = (tvp)->tv_usec = 0L)
#  define is_valid_timeval(tvp) \
          ((tvp)->tv_usec >= 0L && (tvp)->tv_usec < 1000000L)

Note - these are from an HPUX 11.23 implementation - use the ones written for your system....


These macros are defined to work with timeval structs and perform arithmetic operations on them.

Thank you first

Code:

# define timerisset(tvp)        ((tvp)->tv_sec || (tvp)->tv_usec)
# define timerclear(tvp)        ((tvp)->tv_sec = (tvp)->tv_usec = 0)
# define timercmp(a, b, CMP)                                                  \
  (((a)->tv_sec == (b)->tv_sec) ?                                             \
   ((a)->tv_usec CMP (b)->tv_usec) :                                          \
   ((a)->tv_sec CMP (b)->tv_sec))

these codes are in the kernel. Does it mean I need to add "is_valid_timeval"?
or I need modify it to ignore those values moving forward and backward?

You probably have others like timersub or timeradd that let you do actual arithmetic operations.

I think you may have some problems using (int) values on the time_t tv_sec - which is an unsigned int on most systems. Current epoch seconds can overflow overflow a signed integer, for example on subtraction. In other words, your timediff function is using wrong datatypes, use the standard difftime() it is declared in time.h and works correctly.

let us take this "Big:1217489802.1217489802,1217485404.813903,4398.000000" as an example.
1217489802-1217485404=4398

so, there are some problems for the difftime function?

No, your timediff function. Not the std C library function difftime.

Sorry, I think it is misleading. the function I used in Main is stardard function difftime, not my own function timediff.