Re: timer delay compensation

[Eric Norum <wenorum@lbl.gov>]

On Jun 14, 2012, at 3:52 PM, Andrew Johnson wrote:

……...

Lets go with 0.999999999, chosen since we only represent time down to the
nanosecond anyway.

The units of the addition are 'ticks' so 9 9's is actually a fraction of a nanosecond (likely 1/50, 1/60 or 1/100).  

But even so the '1 extra tick' problem wouldn't show up until around 8 million ticks (10^9 ticks ~= 30 bits, 53 bit double mantissa) so 9 9's it is.

Here's the RTEMS implementation:

void

epicsThreadSleep (double seconds)

{

    rtems_status_code sc;

    rtems_interval delay;

    extern double rtemsTicksPerSecond_double;

   

    if (seconds <= 0.0) {

        delay = 0;

    }

    else {

        /*

         * Low-overhead approximation of ceil()

         */

        delay = seconds * rtemsTicksPerSecond_double + 0.999999999;

    }

    sc = rtems_task_wake_after (delay);

    if(sc != RTEMS_SUCCESSFUL)

        errlogPrintf("epicsThreadSleep: %s\n", rtems_status_text (sc));

}

-- 
Eric Norum
wenorum@lbl.gov