RRM 3-14 Compression
compress - Compression
The data compression record is used to collect and compress data from arrays. When the INP field references a data array field, it immediately compresses the entire array into an element of an array using one of several algorithms, overwriting the previous element. If the INP field obtains its value from a scalar-value field, the compression record will collect a new sample each time the record is processed and add it to the compressed data array as a circular buffer.
The INP link can also specify a constant; however, if this is the case, the compression algorithms are ignored, and the record support routines merely return after checking the FLNK field.
Parameter Fields
The data compression fields fall into the following categories:
- scan parameters
- read parameters
- operator display parameters
- run-time parameters
Scanning Parameters
The compression record has the standard fields for specifying under what circumstances the record will be processed. These fields are listed in Scan Fields. In addition, Scanning Specification explains how these fields are used. Since the compression record supports no direct interfaces to hardware, its SCAN field cannot specify I/O Intr
.
Read Parameters and Algorithm Parameters
These fields determine what channel to read and how to compress the data. The user specifies the algorithm to be used in the ALG field. There are six possible algorithms which can be specified as follows:
Circular Buffer
Average
N to 1 Low Value
N to 1 High Value
N to 1 Average
N to 1 Median
These algorithms are explained in one of the sections below.
The RES field can be accessed at run time to cause the algorithm to reset itself.
Field | Summary | Type | DCT | Initial | Access | Modify | Rec Proc Monitor | PP |
---|---|---|---|---|---|---|---|---|
RES | Reset | SHORT | No | 0 | Yes | Yes | No | No |
ALG | Algorithm | RECCHOICE | Yes | 0 | Yes | Yes | No | No |
NSAM | Number in Sample | ULONG | Yes | 1 | Yes | No | No | No |
N | Number | ULONG | Yes | 1 | Yes | Yes | No | No |
ILIL | Initial Low Interest Value | DOUBLE | Yes | 0 | Yes | Yes | No | No |
IHIL | Initial High Interest Value | DOUBLE | Yes | 0 | Yes | Yes | No | No |
Input Specification
The input specification should be a database or channel access link. Though INP can be a constant, the data compression algorithms are supported only when INP is a database link. See Address Specification for information on specifying links.
Algorithms and Related Fields
As stated above, the ALG field specifies which algorithm to be performed on the data. The rest of the fields--NSAM, N, ILIL, IHIL, and OFF--are used in the compressions, though N is not used in the Circular Buffer
algorithm and ILIL, IHIL, and OFF are used neither in the Circular Buffer
algorithm nor in the Average
algorithm.
The Circular Buffer
algorithm keeps a circular buffer of length NSAM. Each time the record is processed, it gets the data referenced by INP and puts it into the circular buffer referenced by VAL. Note that when INP refers to a scalar, VAL is just a time ordered circular buffer of values obtained from INP.
Average
takes an average of every element of the array obtained from INP over time; that is, the entire array referenced by INP is retrieved, and for each element the new average is calculated and placed in the corresponding element of the value buffer. The retrieved array is truncated to be of length NSAM. N successive arrays are averaged and placed in the buffer. Thus, VAL[0] holds the average of the first element of INP over N samples, VAL[1] holds the average of the next element of INP over N samples, and so on. The following shows the equation:
If any of the N to 1
algorithms are chosen, then VAL is a circular buffer of NSAM samples.
The actual algorithm depends on whether INP references a scalar or an array. If INP refers to a scalar, then N successive time ordered samples of INP are taken. After the Nth sample is obtained a new value, determined by the algorithm (Lowest, Highest, or Average), is written to the circular buffer referenced by VAL. If Low Value
the lowest value of all the samples is written; if High Value
the highest value is written; and if Average
, the average of all the samples are written. The Median
setting behaves like Average
with scalar input data.
If INP refers to an array, then the following applies:
N to 1 Low Value
- Compress N to 1 samples, keeping the lowest value.
N to 1 High Value
- Compress N to 1 samples, keeping the highest value.
N to 1 Average
- Compress N to 1 samples, taking the average value.
N to 1 Median
- Compress N to 1 samples, taking the median value.
The compression record keeps NSAM data samples.
The field N determines the number of elements to compress into each result.
Thus, if NSAM was 3, and N was also equal to 3, then the algorithms would work as in the following diagram:
IHIL and ILIL can be set to provide an initial value filter on the input array. If ILIL < IHIL input elements will be skipped until a value is found that is in the range ILIL to IHIL.
RES resets the algorithm before the maximum number of samples are reached.
Operator Display Parameters
These parameters are used to present meaningful data to the operator. They display the value and other parameters of the record either textually or graphically.
The HOPR and LOPR fields only specify the range for VAL, HIHI, HIGH, LOLO and LOW fields.
PREC controls the floating-point precision whenever get_precision
is called, and the field being referenced is the VAL field (i.e., one of the values contained in the circular buffer).
The EGU field should be given a string that describes the value of VAL, but is used whenever the get_units
record support routine is called.
See Fields Common to All Record Types for more on the record name (NAME) and description (DESC) fields.
Field | Summary | Type | DCT | Initial | Access | Modify | Rec Proc Monitor | PP |
---|---|---|---|---|---|---|---|---|
EGU | Engineering Units | STRING [16] | Yes | null | Yes | Yes | No | No |
HOPR | High Operating Range | DOUBLE | Yes | 0 | Yes | Yes | No | No |
LOPR | Low Operating Range | DOUBLE | Yes | 0 | Yes | Yes | No | No |
PREC | Display Precision | SHORT | Yes | 0 | Yes | Yes | No | No |
NAME | Record Name | STRING [29] | Yes | 0 | Yes | No | No | |
DESC | Description | STRING [29] | Yes | Null | Yes | Yes | No | No |
Alarm Parameters
The compression record has the alarm parameters common to all record types. Alarm Fields lists other fields related to a alarms that are common to all record types.
Run-time Parameters
These parameters are used by the run-time code for processing the data compression algorithm. They are not configurable by the user, though some are accessible at run-time. They can represent the current state of the waveform or of the record whose field is referenced by the INP field.
NUSE holds the number of elements currently stored in VAL.
BPTR is a pointer that refers to the buffer referenced by VAL.
The SPTR field pointer to an array that is used for array averages.
WPTR is used by the dbGetlinks routines.
Field | Summary | Type | DCT | Initial | Access | Modify | Rec Proc Monitor | PP |
---|---|---|---|---|---|---|---|---|
NUSE | Number Used | ULONG | No | 0 | Yes | No | No | No |
BPTR | Buffer Pointer | NOACCESS | No | 0 | No | No | No | |
SPTR | Summing Buffer Pointer | NOACCESS | No | 0 | No | No | No | |
WPTR | Work Buffer Pointer | NOACCESS | No | 0 | No | No | No | |
CVB | Compress Value Buffer | DOUBLE | No | 0 | Yes | No | No | No |
INX | Current Index of Circular Buffer | ULONG | No | 0 | Yes | No | No | No |
Record Support
Record Support Routines
init_record
Space for all necessary arrays is allocated. The addresses are stored in the appropriate fields in the record.
process
See next section.
special
This routine is called when RSET, ALG, or N are set. It performs a reset.
get_value
Fills in the values of struct valueDes so that they refer to VAL.
cvt_dbaddr
This is called by dbNameToAddr. It makes the dbAddr structure refer to the actual buffer holding the result.
get_array_info
Obtains values from the circular buffer referenced by VAL.
put_array_info
Writes values into the circular buffer referenced by VAL.
get_units
Retrieves EGU.
get_precision
Retrieves PREC.
get_graphic_double
Sets the upper display and lower display limits for a field. If the field is VAL, the limits are set to HOPR and LOPR, else if the field has upper and lower limits defined they will be used, else the upper and lower maximum values for the field type will be used.
get_control_double
Sets the upper control and the lower control limits for a field. If the field is VAL, the limits are set to HOPR and LOPR, else if the field has upper and lower limits defined they will be used, else the upper and lower maximum values for the field type will be used.
Record Processing
Routine process implements the following algorithm:
- If INP is not a database link, check monitors and the forward link and return.
- Get the current data referenced by INP.
- Perform the appropriate algorithm:
- Average: Read N successive instances of INP and perform an element by element average. Until N instances have been obtained it just return without checking monitors or the forward link. When N instances have been obtained complete the algorithm, store the result in the VAL array, check monitors and the forward link, and return.
- Circular Buffer: Write the values obtained from INP into the VAL array as a circular buffer, check monitors and the forward link, and return.
- N to 1 xxx and INP refers to a scalar: Obtain N successive values from INP and apply the NTO1xxx algorithm to these values. Until N values are obtained monitors and forward links are not checked. When N successive values have been obtained, complete the algorithm, check monitors and the forward link, and return.
- N to 1 xxx and INP refers to an array: The ILIL and IHIL are honored if ILIL < IHIL. The input array is divided into subarrays of length N. The specified N to 1 xxx compression algorithm is applied to each sub-array and the result stored in the array referenced by VAL. The monitors and forward link are checked.
- If success, set UDF to FALSE.
- Check to see if monitors should be invoked:
- Alarm monitors are invoked if the alarm status or severity has changed.
- NSEV and NSTA are reset to 0.
- Scan forward link if necessary, set PACT FALSE, and return.
EPICS Record Reference Manual - 19 MAY 1998