Experimental Physics and
Industrial Control System

Benjamin Franksen <[email protected]> · Thu, 12 Mar 1998 15:03:53 +0100

Jeff Hill wrote:
> 
> On Tuesday, March 10, 1998 8:46 AM, Andrew Johnson wrote:
> > Benjamin Franksen wrote:
> >
> > > another proposal (suggested by Goetz Pfeiffer): Currently there is a
> > > 'Symb' device support for ai, ao, longin, longout, stringin and
> > > stringout records.
> > ...
> > > (1) To protect the data transport, simply add some
> > >
> > >         intLock()
> > >         ...
> > >         intUnlock()
> >
> > This is unnecessary as long as the value is read in a single cycle - for
> > longs certainly, and possibly for doubles too (I'd need to check the
> > compiler output to be sure of this one).  It may be worth doing this for
> > strings and waveforms though (added by me, see below).
> >
> 
> I agree that this is normally only required for multi-element data such
> as waveforms and strings.
> 
> It is almost always preferable to use a semTake()/semGive() of a vxWorks
> mutual exclusion semaphore instead of the intLock()/intUnlock() in task
> level code. I prefer not to cause time critical ISRs to wait while we are
> copying an array in task level code. Frequently, it is also preferable to
> specify the task delete safe and priority inheritance options when creating
> a mutex semaphore in a code that will be used at many different sites.

I was primarily thinking of scalar data (probably no more than a hand
full of machine instructions to copy them), so intLock/Unlock would be
acceptable and more efficient than using a semaphore. But I don't want
to insist on using intLock instead of semaphores. The more important
point is:

I would rather not depend on machine and compiler dependent details such
as how many processor instructions are generated for a given code line
(although I agree that in _most_ cases you are _probably_ right). This
applies especially to doubles (64 bits). Is it _guaranteed_ that 

	*((double *)(*private->ppvar) + private->index) = pao->val;

(such as happens to be in devAoSymb.c) compiles into non-interruptable
machine code? Is it even guaranteed for the simpler line

	vxvar = pao->val;

where vxvar is a globally visible variable of type double?

	Ben
-- 
The Notorious Neb Nesknarf
-----------------------------------------------------------------------
BESSY II
Berliner Elektronenspeicherring-                phone: +49 30 6392-4865
Gesellschaft fuer Synchrotronstrahlung            fax: +49 30 6392-4859
Rudower Chaussee 5                         email: [email protected]
D-12489 Berlin                       web: http://www.bessy.de/~franksen
-----------------------------------------------------------------------
HOME: Benjamin Franksen, Fichtestr. 19a, D-10967 Berlin, +49(30)6913756
-----------------------------------------------------------------------

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System