forums.silverfrost.com

[Anonymous]

Hello everybody,

I am puzzled with the following.
Trying to get an accurate estimation of time consumption of a certain algorithm, I was using two possible measurements: DATE_AND_TIME and SYSTEM_CLOCK.

Consider the following piece of code:

CALL DATE_AND_TIME (start_date,start_time,zone,values)
CALL SYSTEM_CLOCK (count_start,count_rate,count_max)
millisecs_start = CALC_MILLISECS (values) ! user-defined function

<STUFF TO DO>

CALL DATE_AND_TIME (stop_date,stop_time,zone,values)
CALL SYSTEM_CLOCK (count_stop,count_rate,count_max)
millisecs_stop = CALC_MILLISECS (values) ! user-defined function
WRITE(6,*) millisecs_stop-millisecs_start, count_start-count_stop ! count seems to be decreasing, but hey, no problem.

The results produced were:
elapsed millisecs = 44312
elapsed clock tics = 829019

The funny thing is that COUNT_RATE = 10000, and this is not in accordance with the above results (which rather indicate a CLOCK_RATE of about 18700...)
Any explanation, anybody?

Many thanks!
Lucas.

[PaulLaidler]

Lucas

A question like yours about SYSTEM_CLOCK has already been raised on a different forum thread and is still outstanding. In the mean time...

COUNT_RATE is an internal fixed scaling factor that is used when calculating the results.
I cannot think of a way that it might be used at programming level.

The arguments of SYSTEM_CLOCK should be of the same INTEGER KIND and there may well be cases when KIND=4 (INTEGER* is required.

The /TIMING compiler option may be of interest and relevance. See the help file for further details.