PE>
  >                                                 This program
  > when profiled registers 95% of the time spent in DosSubSet!!!
  >
  > #include 
  >
  > static char buf[30000];
  >
  > int main(void)
  > {
  >     int x;
  >     FILE *fq;
  >
  >     fq = fopen("temp.txt", "wb");
  >     for (x = 0; x < 2000; x++)
  >     {
  >         fwrite(buf, sizeof buf, 1, fq);
  >     }
  >     fclose(fq);
  >     return (0);
  > }
PE>

  Stop and think.  Your profiler cannot register time spent in a device
  or filesystem driver, nor does it register time spent blocked waiting
  for I/O requests to complete in the bottom half of the kernel.

  Therefore the 95% figure for DosSubSet is a percentage of the time
  spent outside of I/O, not for the total execution time of your
  application.

  Considering that your program does little else *other* than I/O, what
  you are actually measuring is the profiling information for the C++
  runtime startup and cleanup code and the buffering in the 
  library.  The library initialisation and terminating code does little
  else than allocate buffers (for things like `stdin' and the argument
  list) so it's unsurprising that profiling it reveals that most of its
  time is spent in memory management routines.  The same goes for the
   library.

  To "unskew" your profiling measurements, try putting some non-I/O code
  in the body of your program (generate a series of 1000 random numbers
  in every iteration of the loop, or something) and then profiling it.

  > JdeBP <
___
 X MegaMail 2.10 #0:

--- Maximus/2 3.00