##
Unisys (and Univac) Representations of Floating Point Numbers Using 36-bit
Words

Univac traditionally used 6-bit bytes (not 8-bit bytes), complicating
some translations. It is not unusual to see data which was originally 36
bits unpacked into six, 6-bit bytes. Then, each byte was packed with two
extra zeroes so that what was six, six-bit bytes (36 bits total) became six,
eight-bit bytes (48 bits total). This facilitated use by 32-bit hardware
and software. For example, 9-track tapes would then frequently have six
bits of data (on six tracks), two zero bits, or "zero fill" (on two tracks),
and a parity bit on the 9th track. Be aware of these possibilities when
reading old Univac/Unisys data.

*Single Precision (one 36-bit word) floating point*:

**Unisys/Univac Floating Point Representations:
"F_Floating" Structure (36 bits):** |

35 |
34 |
33 |
32 |
31 |
30 |
29 |
28 |
27 |
26 |
25 |
24 |
23 |
22 |
21 |
20 |
19 |
18 |
17 |
16 |
15 |
14 |
13 |
12 |
11 |
10 |
9 |
8 |
7 |
6 |
5 |
4 |
3 |
2 |
1 |
0 |

Sign Bit |
Exponent (bias: 128_{10}) |
Fraction: bit 26 is the most significant |

*Double Precision (two 36-bit words) floating point: *

**Unisys/Univac Floating Point Representations:
Double Precision (72 bits): First 36-bit Word:** |

35 |
34 |
33 |
32 |
31 |
30 |
29 |
28 |
27 |
26 |
25 |
24 |
23 |
22 |
21 |
20 |
19 |
18 |
17 |
16 |
15 |
14 |
13 |
12 |
11 |
10 |
9 |
8 |
7 |
6 |
5 |
4 |
3 |
2 |
1 |
0 |

Sign Bit |
Exponent (bias: 1024_{10}) |
Fraction (First, most significant part): bit 23
is the most significant |

**Unisys/Univac Floating Point Representations:
Double Precision (72 bits): Second 36-bit Word:** |

35 |
34 |
33 |
32 |
31 |
30 |
29 |
28 |
27 |
26 |
25 |
24 |
23 |
22 |
21 |
20 |
19 |
18 |
17 |
16 |
15 |
14 |
13 |
12 |
11 |
10 |
9 |
8 |
7 |
6 |
5 |
4 |
3 |
2 |
1 |
0 |

Fraction (Second, less significant part): bit 35
is the most significant bit in this word |

EXPONENT (**E**): To the base 2, with a bias in the single precision
floating point of 128 decimal (200 octal), and in the double precision
floating point of 1024 decimal (2000 octal).

FRACTION: Has no hidden bit.