Re: AAI and AAO Record support

[Andrew Johnson <anj@aps.anl.gov>]

Hi David,

I'll try and catch some of the things Ralph didn't mention and add to
his ideas as well...

David Dudley wrote:
> I'm currently starting to work on adding support into my modbus
> device driver for Analog Arrays.  Is there any documentation on the
> AAI and AAO records?

Not that I'm aware of.  I think the original author was Johnny Tang 
<jtang@ornl.gov> when he was working at JLAB, and either he or they 
might have some (if you're lucky).  You can always just look at the 
aa*Record.dbd file and the associated .c file (in base/src/rec) to get 
some idea what they do.

If anyone wants to resurrect the AAI/AAO records, at minimum they need 
soft device support and some documentation.  Personally I want to 
overhaul our complete set of record types, but I think that's a big task 
that might best wait for any changes we make to the record support API.

> It has been suggested before that perhaps I should consider the
> Waveform record, but I have a little problem.  I need to read a
> string of 16 bit registers in a single group (sort of a 'simultaneous
> sample' thing), and I don't see the capability of doing this will a
> standard AI record (do I need to write special device support for
> Waveform?  Maybe that's the problem).

You should be able to do this in your AI device support if you are so 
inclined - look at the technique in directNetAsyn where I batch together 
a number of input operations from adjacent or nearby addresses and put 
the result into a cache from which any of the input records can read 
without kicking off another I/O operation (if they're quick enough).

Alternatively there's a technique that is used in the AllenBradley 
support where the I/O operation for the block is done by a Waveform 
record and there are individual input records that extract data from the 
waveform data.

> Also, even if I generate whatever is needed for Waveform records for
> input, how do I manage output?  I need to be able to write a 120
> register block out at the same time in order to insure consistency.
> Seems to me, that's an Analog Array Output (AAO) record, if I ever
> saw one.

Again this might be a device support implementation - you would have one 
special address that causes the block to be written out, and all of the 
other related output records just write data to the memory block.  You 
could do this with a waveform record again -- I don't think there would 
be any signficicant point in using the AAO record since the two are 
basically very similar, and neither give you the ability to say anything 
like 'put this value into the 13th array element'.

> On an unrelated note, I'm curious about a few things.  I don't see
> any way to do tests with the database.  I have a need to do things
> like "read an input - Test for a value - output a trigger if true",
> as a simple grouping.  I don't see any way to programmatically decide
> whether or not to execute a FWDLNK, which I think is what I want to
> do (or is that what the SEQ block can do?  Documentation is kind of
> weak.).

Look at the calcout record for conditional processing, and don't forget 
the SDIS link for disabling processing of your trigger record.

> Finally, I don't see any way to do 'discrete PID' controls.  Guess I
> should just shut up and write blocks to do what I want, which I think
> consist of the following (any way to do these already would be
> appreciated).

Ralph mentioned the PID record; the latest and best version of that is 
probably the epid which is part of the synApps/std package at 
http://www.aps.anl.gov/aod/bcda/synApps/std/std.html (with documentation 
too!).

> 1. Multiplex a input array into a single value (think I found that in
> the SEL block, but why is the limit 12 on so many of these blocks?).

That's a good question, my guess is that they were made as big as the 
original author needed, or they started from some common record type 
that only had 12 and so the meme just spread.

> 2. Demultiplex a value to a selected FWDLNK (don't see this one,
> unless the SEQ block does it.  But it doesn't have multiple FWDLNKs).

In addition to the SEQ record (point its ten OUTx links to the .PROC 
fields of the target records if you don't them want to pass a value), 
the fanout record can very easily be extended to almost as many forward 
links as you desire (if you want more than 16, change the SELN field to 
a DBF_ULONG; if you need more than 32 you'll have to take out the 
bit-mask mechanism).

>  3. IF or other conditional test block (if condition meets criteria,
> do FWDLNK, else don't).

See the SDIS mechanism - you always do the forward link, but the target 
doesn't always process if it's been disabled.

> 4. PID Control with Discrete output controls (jog up/jog down, for
> PID Control).

I guess I'm not completely sure what you're after here, but you should 
look at the capabilities of the motor record (another part of synApps 
which is shipped separately), which may do some of what you need: 
http://www.aps.anl.gov/upd/people/sluiter/epics/modules/mechanism/motor/index.html

> 5. DB Log command to output current value to a database point.

Hmm, not sure if you're talking about an archive here - check out the 
Channel Archiver, or expand on your needs.

> Also, is there any structure built into the system to provide a
> 'orderly startup / orderly shutdown' operation?  The stuff I have
> background with all provide the capability of loading an initial
> state and storing the current state when the system is started up and
> shut down, as well as the capability of doing a 'checkpoint' type
> operation.  Need that to do continuous control type operations.
> Can't have a remote IOC defaulting everything to invalid conditions
> if it has a short power failure or other problem that causes a
> reload.

This sounds like a combination of one or more sequence programs and the 
autosave/restore subsystem, another part of synApps (find all of it at 
http://www.aps.anl.gov/aod/bcda/synApps/) which you configure to 
checkpoint the current value of selected fields at programmed intervals 
and which can restore those values at IOC initialization time.

> Lastly, do most IOC's have the SEQUENCER software incorporated in
> them?  Seems to me that a lot of the things I would expect to do in
> combinational-logic don't seem to *exactly* exist in the DB.  Do all
> logic things need SEQ support?

Probably a large proportion of IOCs have sequencers in them.  However 
the sequencer is better for doing state transition logic; interlocks and 
the like are probably best done in the database itself.  The calc record 
can do combinatorial logic, and also see the transform record for more 
complex expressions.  There was a PAL record in R3.13 which could take 
the JEDEC file for a PAL16V8 or PAL22V10 chip and implement the logic 
found therein, but I don't think it was ever used outside of Los Alamos. 
 You can also write your logic in subroutine record code, and you 
should also look at the gensub record for whole-array handling.

Hope this helps,

- Andrew
--
The right to be heard does not automatically include
the right to be taken seriously. -- Hubert H. Humphrey