ABAP Keyword Documentation → ABAP Programming Guidelines → Robust ABAP → Assignments, Calculations, and Other Types of Data Access
Selecting the Numeric Type
Other versions: 7.31 | 7.40 | 7.54
Background
Several numeric types with various properties and value ranges are available in ABAP, which you can use for storing numbers and for calculations:
- Signed 4-byte and 8-byte integers (type
i
). There are also 1-byte and 2-byte integers with the internal typesb
ands
. However, these data objects cannot be generated based on a predefined ABAP type. Instead, the predefined ABAP Dictionary types INT1 and INT2 are required.
- Packed numbers in BCD format (binary coded decimals, type
p
)
- Binary floating point numbers (type
f
)
- Decimal floating point numbers (types
decfloat16
,decfloat34
)
The decimal floating point numbers meet the requirements for highly exact processing of decimal numbers
in a large value ranges, where the data types p
and f
only cover a single aspect each.
Rule
Select suitable numeric types for numbers and calculations
Select a numeric type to suit the values that you want to map, in order to achieve the highest possible speed and accuracy. Here is a general rule of thumb:
i
for integers
p
for fixed point numbers
decfloat16
ordecfloat34
for floating point numbers
f
in exceptional cases only
Details
The calculation speed and accuracy are generally contradictory requirements and depend on the data type of the objects to be processed. Therefore, when selecting the numeric type, you must weigh up these two requirements. These considerations must also include the value range to be mapped:
- If the values to be mapped are integers, type
i
must usually be used. This guarantees the highest possible calculation speed. Examples of these integers include counters, indexes, offsets and time intervals. If the values to be mapped exceed the value range of type i, you can use typep
without decimal places instead. The calculation speed is slower, but decimal places are still mapped accurately (except for rounding errors). If this value range is still not sufficient, you can use a floating point type (decfloat16
anddecfloat34
) instead.
- If you have to map nonintegral values that have a fixed number of decimal places, you can use type
p
. However, calculations with typep
are executed in the ABAP kernel and not by the hardware. Examples of non-integral values include lengths, weights, or monetary amounts. If this value range is still not sufficient, you can use a decimal floating point type (decfloat16
anddecfloat34
) instead. The binary floating point typef
is less suitable because it cannot map all decimal fractions. This is a further impediment for calculation accuracy (in addition to the unavoidable rounding errors).
- For nonintegral values with a variable number of decimal places or a large value range, you should
use the decimal floating point types
decfloat16
ordecfloat34
. Decfloat34 has a larger number of decimal places and a larger value range. However, this also leads to increased memory consumption.
Thanks to the hardware support available on all platforms, binary floating point type f
allows fast calculations. However, it is inferior to the new decimal floating point types due to the following reasons:
- Type
f
can only accuately map fractions with the power of two in the denominator (1/2, 1/4, 1/8, etc.) and totals. Other values are rounded according to this mapping. This rounding process does not correspond to the decimal rounding process (and usually does not meet the expectations of the developer or user). For example, the value 0.815 is approximated internally as 8.1499999999999995E-01. This means when the value is rounded to two decimal places, 0.81 is returned instead of the expected value 0.82.
- Yery large numbers can no longer be exactly mapped (working in line with IEEE 754), if the difference between the largest and smallest exponent is larger than 52 (in the total of powers of two). For example, 1E+23 is mapped as 9.9999999999999992E+22.s
- A number of type
f
cannot be rounded to a specific number of decimal places, if the result needs to be assigned to another number of typef
.
- Divisions by powers of 10, which often occur when converting metric units, for example, are not exact. 8.0500000000000005E-01 is returned for 805/1000, for example.
- Simple calculations often produce unexpected results. 123456.15 - 123455 returns 1,1499999999941792, for example.
- The conversion of binary floating point numbers to other number formats is not clearly defined in
line with IEEE 754. Consequently, when data is stored in the database, the rounding behavior depends on the platform and how numbers of type
f
are mapped in the database.
- The decimal floating point types
decfloat16
anddecfloat34
do not have these problems. Both have a larger value range than typef
, anddecfloat34
has 34 decimal places instead of 16 decimal places. However, the following restrictions apply:
- Calculations with decimal floating point types are currently slower than calculations with type
f
(the speed is similar to calculations with typep
). Currently, only the IBM Power6 architecture provides hardware support for decimal floating point calculations that fulfill IEEE-754-2008. On other platforms, you must perform calculations with decimal floating point numbers in a similar way to how calculations with typep
are performed by the software in the ABAP kernel.
- The decimal floating point types are not supported by corresponding data types in all database platforms.
As a workaround, the ABAP Dictionary provides a set of predefined data types (DF16_..., DF34_...). These
are based on existing types (DEC and RAW) and offer various processing options. In most cases, the benefits
of the decimal floating point types compensate for the current slow calculation speed. However, you
might still need to use type
f
, if you have stringent requirements for performance and less stringent requirements for calculation accuracy. Remember that the speed advantage currently possible for calculations withf
, may be outweighed by the fact that conversions fromf
to other numeric types are relatively slow.
Note
For programs that are currently created with decimal floating point types, the performance is increased as soon as the processor architecture supports decimal floating point calculations and the ABAP runtime environment starts using this hardware support. Calculations with decimal floating point numbers then become faster than calculations with packed numbers.
Bad Example
The following source code shows a declaration of a binary floating point number. The start value 0.815 is assigned. The true start value, however, is 8.1499999999999995E-01.
DATA number TYPE f VALUE '0.815'.
Good Example
The following source code shows a declaration of a decimal floating point number. The start value 0.815 is assigned. The true start value is actually 8.15E-01.
DATA number TYPE decfloat34 VALUE '0.815'