C program execution time OpenMP/Sequential-CodePudding

I am trying to measure the execution time of a method (C program) that I have parallelized with OpenMP. I need to compare the execution times of the parallelized algorithm (with OpenMP) and its sequential version. I am using the method omp_get_wtime() in the parallel case and gettimeofday() in the sequential case to be able to do the comparison. I would like to know if this approach to making the comparison is correct or not. Thanks for your help.

Here the code i'm using to define macros:

#ifdef _OPENMP
    #include <omp.h>
    #define STARTTIME(id)                           \
    double start_time_42_##id, end_time_42_##id; \
    start_time_42_##id = omp_get_wtime();

    #define ENDTIME(id, x)        \
    end_time_42_##id = omp_get_wtime(); \
    x = end_time_42_##id - start_time_42_##id

#else   
    #include <sys/time.h>
    #include <time.h>
    #include <math.h>
    #define STARTTIME(id) \
    struct timeval start_##id; \
    gettimeofday(&start_##id, NULL);
    
    #define ENDTIME(id, x) \
    struct timeval end_##id; \
    gettimeofday(&end_##id, NULL); \
    x = ((end_##id.tv_sec  - start_##id.tv_sec) * 1000000u   end_##id.tv_usec - start_##id.tv_usec) / 1.e6;

#endif

CodePudding user response：

Sure, why not. The last line looks a bit cumbersome. Why not like this:

x = (end_##id.tv_sec - start_##id.tv_sec) (end_##id.tv_usec - start_##id.tv_usec) * 1.e-6;

Other than that, if wall clock time is all you need, your solution will work just fine.