Hi Dominique,
I
just put together a test database to convert a 32-bit ao record to 2 16-bit ao
records. Here is the database:
record(ao, "test:ao") { field(FLNK,
"test:calcLSW.PROC") }
record(calcout, "test:calcLSW") { field(INPA,
"test:ao.VAL NPP NMS") field(CALC,
"A&65535") field(OUT,
"test:aoLSW") field(FLNK,
"test:calcMSW.PROC") }
record(calcout, "test:calcMSW") { field(INPA,
"test:ao.VAL NPP NMS") field(CALC,
"(A>>16)&65535") field(OUT,
"test:aoMSW") }
record(ao, "test:aoLSW") { }
record(ao, "test:aoMSW") { }
I
then tested this database with various 32-bit input values to test:ao, and
looked at the 16-bit output values in test:aoLSW and
test:aoMSW.
Here are the results:
corvette> caput -s test:ao "0x1" Old :
test:ao
2.41592e+09 New :
test:ao
1 corvette> caget test:ao test:aoLSW
test:aoMSW test:ao
1
test:aoLSW
1
test:aoMSW
0
corvette> caput -s test:ao "0xffff" Old :
test:ao
1 New :
test:ao
65535 corvette> caget test:ao test:aoLSW
test:aoMSW test:ao
65535
test:aoLSW
65535
test:aoMSW
0
corvette> caput -s test:ao "0x7fffffff" Old :
test:ao
2.68435e+08 New :
test:ao
2.14748e+09 corvette> caget test:ao test:aoLSW
test:aoMSW test:ao
2.14748e+09
test:aoLSW
65535
test:aoMSW
32767
corvette> caput -s test:ao "0x8fffffff" Old :
test:ao
2.14748e+09 New :
test:ao
2.41592e+09 corvette> caget test:ao test:aoLSW
test:aoMSW test:ao
2.41592e+09
test:aoLSW
0
test:aoMSW
32768
I
conclude that the database works OK until the most-significant bit in the 32-bit
test:ao record is set. The database then fails, probably because the calc
parser is converting to a 32-bit signed integer, rather than unsigned. For
numbers up to 0x7fffffff ( 2147483647 decimal) it works OK.
Mark
Mark Rivers a écrit :
Hi Dominique,
Perhaps I discovered a modbus driver "feature" ?
I created one modbus number 4 read function which read 11 bytes. I tested three waveform input records.
These three records read 3 different values on different offsets. These 3 records read the offset 0 ???!!!
That is the documented behavior for asynInt32Array support. Look at
http://cars9.uchicago.edu/software/epics/modbusDoc.html#EPICS%20device%20support
in the section on asynInt32Array support. You will see that the "offset" field is NA (not applicable) for all Modbus functions for asynInt32Array.
You are totally right. Thanks. I'll take back my glasses.
;-)
About ai/ao records, I will need to write and read float values with the Modbus driver.
I saw integer examples. Ai/ao examples work with asyn32int type.
Did you test asynFloat64 values ?
Do I have to use waveform records (ie 32bits : sign:1bit exponent:8bits fraction:23bits) ?
Float values are not a standard Modbus data type; they are not documented in the Modbus standard. Again, because they are more than 16 bits long, they require multiple Modbus words. I think you need to do the same thing you do for 32-bit integers, i.e. use a calc, Gensub or SNL code to convert to two 16-bit integers and then use two longout records or a waveform record.
I have a beginner's question : does the bit left shift work like C code?
Can I write something like A << 16 in the calc _expression_?
Yes, that is documented under Bitwise Operators here:
http://www.aps.anl.gov/epics/EpicsDocumentation/AppDevManuals/RecordRef/Recordref-13.html#HEADING13-0
I tried to convert the 32 bit integer value into 2 16 bits values with a calc record.
Because of float conversion troubles, I don't get through to get the right values.
It seems to me like it should work, since I think any 32-bit integer can be represented exactly in a 64-bit double. Can you send an example that did not seem to work?
I broke the EPICS database to make a SNL code. I haven't
got any example. I tried to use the 0x0000FFFF bit mask. I thought this value
was converted in a float value and the bitwise operation didn't give the
right result when it was converted back to an integer. The 0x0001000 bit
mask worked. Perhaps I made a mistake.
I will tried again with two
shifts (one 16 bits left shift to lose MS bits and and one right to give back
the right bit position).
Thanks again for your help. Dominique.
Mark
________________________________
From: Touchard Dominique [mailto:[email protected]]
Sent: Tue 10/9/2007 3:50 AM
To: Mark Rivers
Cc: [email protected]
Subject: Re: Use of waveform records
Hi Mark,
Thanks for your answer. I agree with you. We mustn't modify the driver.
The power supply could work with two longout records. In the future, I will need to know how to work with a waveform, an SNL or Gensub code. So I decided to try these tools. I have written an SNL code which does this conversion. I know this is not the best solution. I will try a gensub record soon.
I tried to convert the 32 bit integer value into 2 16 bits values with a calc record. Because of float conversion troubles, I don't get through to get the right values. I have a beginner's question : does the bit left shift work like C code? Can I write something like A << 16 in the calc _expression_?
Perhaps I discovered a modbus driver "feature" ? I created one modbus number 4 read function which read 11 bytes. I tested three waveform input records. These three records read 3 different values on different offsets. These 3 records read the offset 0 ???!!!
So I decided to create only one input waveform which read the whole bytes. It works fine. At present, the SNL code converts the values.
About ai/ao records, I will need to write and read float values with the Modbus driver. I saw integer examples. Ai/ao examples work with asyn32int type. Did you test asynFloat64 values ? Do I have to use waveform records (ie 32bits : sign:1bit exponent:8bits fraction:23bits) ?
Thanks for your help.
Dominique.
Mark Rivers a écrit :
One thought on this.
Because you must do 2 Modbus operations to write a 32-bit value with
Modbus, I suspect that the power supply only updates the value when you
write one of the 16-bit words, i.e. either the most-significant or
least-significant 16 bits. You should read the documentation to find
out which one actually updates the power supply and make your database
write that value last. Otherwise you will get a momentary garbage
output if you do it in the wrong order.
Mark
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Mark Rivers
Sent: Friday, October 05, 2007 11:23 AM
To: Touchard Dominique; [email protected]
Subject: RE: Use of waveform records
Hi Dominique,
Sorry for the delay in replying to this, I was on vacation.
I think an easier solution to this problem is to use 2 calc records to
convert the 32-bit integer to 2 16-bit integers, and write
those to the
Modbus driver using function 6. Then you don't have to write any C
code, just a simple database.
PowerSupply (longout) -> calc1 (16 LSB) -> longout1 -> Modbus
-> calc2 (16 MSB) -> longout2 -> Modbus
I don't recommend modifying the driver, because a 32-bit
integer is not
a standard Modbus data type, and we could end up with all sorts of
modifications for non-standard data types.
Mark
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of
Touchard Dominique
Sent: Wednesday, September 19, 2007 8:44 AM
To: [email protected]
Subject: Use of waveform records
Hi,
we are working on/with the Epics modbus tcp driver written by Mark
Rivers. We need to write a 32 bits integer value in a power supply
command control interface. First we thought to write an asyn32Int
record with the number 6 modbus function. Immediatly, we
saw that the
driver uses the number 6 modbus function which works only on
a 16 bits word.
After reading the Modbus Driver Support documentation, we
saw that we
have to use a two value length waveform record.
First question : is this correct ?
Second question : do we have to use the gensub record to manipulate
values inside a waveform record ? In this case, is there any example
that shows how to do this ? Could somebody send us an example
on this ?
Third question : Is a bad idea to modify the modbus driver
in the goal
to write a asyn32Int 32 bits integer record with the
number 16 modbus
function ?
Thanks for your help .
Dominique Touchard.
|