使用dtime或time命令打印总运行时间（挂钟）？

Question

我试图使用ifort来评估模型性能，但是打印输出没有正确显示。

这是我的代码。请告诉我如何完成这个任务。

! Time the run 

     call dtime(timearray, telapse) 

! End of the simulation 

     call dtime(timearray, telapse) 
     call print_runtime(telapse, ctime) 

     subroutine print_runtime(telapse, ctime) 

     implicit none 
     real*4 telapse 
     character*8 ctime(2) 
     integer*2 RunDays, 
    >   RunHours, 
    >   RunMins, 
    >   RunSecs 
     character*40 msgstr, rtfname 
     parameter (msgstr = ' *** Total run time (wallclock) was ') 
     parameter (rtfname = 'runtime.txt') 

! Now convert telapse from seconds to DD HH:MM:SS 

     RunDays = INT (telapse/86400.0) 
     telapse = telapse - (RunDays * 86400.0) 
     RunHours = INT (telapse/3600.0) 
     telapse = telapse - (RunHours * 3600.0) 
     RunMins = INT (telapse/60.0) 
     RunSecs = NINT (telapse - (RunMins * 60.0)) 

     if (RunDays .GT. 0) then 
      write (*,1) msgstr, RunDays, RunHours, RunMins, RunSecs 
1  format (A36, I2, 'days, ', I2.2, ':', I2.2, ':', I2.2, ' hh:mm:ss ***') 

     else if (RunHours .GT. 0) then 
      write (*,2) msgstr, RunHours, RunMins, RunSecs 
2  format (A36, I2, ':', I2.2, ':', I2.2, ' hh:mm:ss ***') 

     else if (RunMins .GT. 0) then 
      write (*,3) msgstr, RunMins, RunSecs 
3  format (A36, I2, ':', I2.2, ' mm:ss ***') 

     else 
      write (*,4) msgstr, telapse 
4  format (A36, F7.4, ' seconds ***') 

     endif 
     return 
     end 

目前，我的打印输出显示“*** Total run time (wallclock) was ******* seconds ***” 如何显示一个适当的总运行时间（挂钟）在这里？ 谢谢。 Michael

Answer 1

以下代码片段显示如何以秒为单位输出经过的挂钟时间。这与消耗的CPU时间不同，可能不是衡量代码性能的好方法。 DTIME是运行时间，而不是挂钟时间，它也是一个扩展。如果你想cpu rutime，语言标准提供了cpu_time。你从秒转换的代码看起来没问题。

integer :: clck_counts_beg, clck_counts_end, clck_rate 
call system_clock (clck_counts_beg, clck_rate) 
your code 
call system_clock (clck_counts_end, clck_rate) 
write (*, *) (clck_counts_end - clck_counts_beg)/real (clck_rate) 

CPU时间例子，也以秒

real :: beg_cpu_time, end_cpu_time 
call cpu_time (beg_cpu_time) 
your code 
call cpu_time (end_cpu_time) 
write (*, *) end_cpu_time - beg_cpu_time