Errata exists for this version of the document.
All floating-point values shall be encoded with the appropriate IEEE-754 binary representation which has three basic components: the sign, the exponent, and the fraction. The bit ranges assigned to each component depend on the width of the type. Table 4 lists the bit ranges for the supported floating point types.
Table 4 – Supported Floating Point Types
Name |
Width (bits) |
Fraction |
Exponent |
Sign |
Float |
32 |
0-22 |
23-30 |
31 |
Double |
64 |
0-51 |
52-62 |
63 |
In addition, the order of bytes in the stream is significant. All floating point values shall be encoded with the least significant byte appearing first (i.e. little endian).
Figure 3 illustrates how the value −6,5 (Hex: C0D00000) should be encoded as a Float.
The floating-point type supports positive and negative infinity and not-a-number (NaN). The IEEE specification allows for multiple NaN variants; however, the encoders/decoders may not preserve the distinction. Encoders shall encode a NaN value as an IEEE quiet-NAN (000000000000F8FF) or (0000C0FF). Any unsupported types such as denormalized numbers shall also be encoded as an IEEE quiet-NAN. Any test for equality between NaN values always fails.
Figure 3 – Encoding Floating Points in a binary stream