FLOAT

The FLOAT option controls the use of the additional floating-point registers and whether Decimal Floating Point is supported.

             .-NODFP-.      
>>-FLOAT--(--+-DFP---+--)--------------------------------------><

DFP

The DFP facility is used: all DECIMAL FLOAT data will be held in the DFP format as described in the z/Architecture Principles of Operation manual, and operations using DECIMAL FLOAT will be carried using the DFP hardware instructions described therein.

NODFP

The DFP facility is not used.

Under FLOAT(DFP), the following applies:

• The maximum precision for extended DECIMAL FLOAT will be 34 (not 33 as it is for hex float).
• The maximum precision for short DECIMAL FLOAT will be 7 (not 6 as it is for hex float).
• The values for DECIMAL FLOAT for the following built-ins will all have the appropriate changes:
  • EPSILON
  • HUGE
  • MAXEXP
  • MINEXP
  • PLACES
  • RADIX
  • TINY
• The following built-in functions will all return the appropriate values for DECIMAL FLOAT (and values that will be much easier for a human to understand; for example, SUCC(1D0) will be 1.000_000_000_000_001, and the ROUND function will round at a decimal place of course.)
  • EXPONENT
  • PRED
  • ROUND
  • SCALE
  • SUCC
• Decimal floating-point literals, when converted to "right-units-view", that is, when the exponent has been adjusted, if needed, so that no nonzero digits follow the decimal point (for example, as would be done when viewing 3.1415E0 as 31415E-4), must have an exponent within the range of the normal numbers for the precision given by the literal. These bounds are given by the value of MINEXP-1 and MAXEXP-1. In particular, the following must hold:
  • For short float, -95 <= exponent <= 90
  • For long float, -383 <= exponent <= 369
  • For extended float, -6143 <= exponent <= 6111
• When a DECIMAL FLOAT is converted to CHARACTER, the string will hold 4 digits for the exponent (as opposed to the 2 digits used for hexadecimal float).
• The IEEE and HEXADEC attributes will be accepted only if applied to FLOAT BIN, and the DEFAULT(IEEE/HEXADEC) option will apply only to FLOAT BIN.
• The mathematical built-in functions (ACOS, COS, SQRT, etc) will accept DECIMAL FLOAT arguments and use corresponding Language Environment functions to evaluate them. DECIMAL FLOAT exponentiation will be handled in a similar way.
• Users of DFP need to be wary of the conversions that will arise in operations where one operand is FLOAT DECIMAL and the other is binary (that is, FIXED BINARY, FLOAT BINARY or BIT). In such operations, the PL/I language rules dictate that the FLOAT DECIMAL operand be converted to FLOAT BINARY, and that conversion will require a library call. For example, for an assignment of the form A = A + B;, where A is FLOAT DECIMAL and B is FIXED BINARY, three conversions will be done and two of those will be library calls:
  1. A will be converted through library call from FLOAT DECIMAL to FLOAT BINARY.
  2. B will be converted through inline code from FIXED BINARY to FLOAT BINARY.
  3. The sum A + B will be converted through library call from FLOAT BINARY to FLOAT DECIMAL.

  The use of the DECIMAL built-in function might help here: if the statement is changed to A = A + DEC(B);, the library calls will be eliminated. The library calls can also be eliminated by assigning B to a FLOAT DECIMAL temporary variable and then adding that temporary variable to A.

• The built-in function SQRTF is not supported for DECIMAL FLOAT arguments (because there is no hardware instruction to which it can be mapped).
• DFP is not supported by the CAST type function.