clock() precision may not be good enough for what I need

[LMHmedchem]

Hello,

I am trying to time a section of code where records are processed. I have tried the following, which seems too simple to not work,

#include <ctime>

// to output row number
 int record_count = 0;
// time when processing starts
clock_t start_clock, finish_clock;
// process time for current compound
long unsigned int clocks_to_finish = 0;

while( ***process record*** ) {

   // increment record count
   record_count++;
   // record clock at start of record processing
   start_clock = clock();

   ***

   // record clock at end of record processing
   finish_clock = clock();
   // time to finish current record
   clocks_to_finish = finish_clock - start_clock;

   // print difference
  cout << record_count << " clocks to finish record = " << clocks_to_finish << endl;

 }

In most cases, clocks_to_finish is 0. In some cases it is 1. I know that clock() precision isn't great but this program processes 1700+ records in about 1 minute 15 seconds. There should be at least 0.3 seconds or so per record so I was hoping that it would be good enough to provide some information. It may be that there are a couple of records that are taking up allot of the total time and most of them are quite quick. Even so, I should still be able to see a large value for some records.

Is there a solution like GetTickCount() that would provide better precision (I am in Linux)? Is the problem my implementation?

Thanks,

LMHmedchem

[salem_c]

Possibly the wrong tool for the job.

#include <stdio.h>
#include <time.h>
#include <unistd.h>

int main ( ) {
    struct timespec ts_start, ts_end;
    printf("Sleep for 2 seconds\n");
    clock_t start = clock();
    clock_gettime(CLOCK_REALTIME,&ts_start);
    sleep(2); // Your workload
    clock_gettime(CLOCK_REALTIME,&ts_end);
    clock_t end = clock();
    double elapsed = (double)(end - start) / CLOCKS_PER_SEC;
    printf("Elapsed CPU seconds=%f\n", elapsed);
    double wall = (ts_end.tv_sec-ts_start.tv_sec) * 1000000000.0 +
                  (ts_end.tv_nsec-ts_start.tv_nsec);
    wall /= 1000000000;   // back to seconds
    printf("Elapsed wall seconds=%f\n", wall);
}

$ ./a.out 
Sleep for 2 seconds
Elapsed CPU seconds=0.000044
Elapsed wall seconds=2.000088

The clock() function returns an approximation of processor time used by the program.

If your "***" work involves sending some kind of message to another program to do the actual work, then your program is going to effectively go to sleep and the clock() stops for the duration.

There are many clock types that clock_gettime can use, but all come with various caveats.
You should read the manual page to find out which clock type is best for you.

[2kaud]

@DerellLicht - The OP specially mentions that the solution is for Linux.

[DerellLicht]

@DerellLicht - The OP specially mentions that the solution is for Linux.

OOPS!! Sorry, I missed that entirely...
nevermind...

[Philip Nicoletti]

Try the functions in <chrono>

#include <chrono>
using namespace std::chrono


//
auto start = high_resolution_clock::now();

// code to time

auto finish = high_resolution_clock::now();

auto diff = finish - start;

// get times in desired units ... examples

auto millisec = std::chrono::duration_cast<std::chrono::milliseconds>(diff);
auto microsec = std::chrono::duration_cast<std::chrono::microseconds>(diff);
auto nanosec = std::chrono::duration_cast<std::chrono::nanoseconds>(diff);

// number of milli seconds in : millisec.count()
// number of micro seconds in : microsec.count()
// number of nano seconds in : nanosec.count()

// for my test routine ...
// number of millisec : 6
// number of microsec : 6752
// number of nanosec : 6752200