Decimal computer

Early computers that were exclusively decimal include the ENIAC, IBM NORC, IBM 650, IBM 1620, IBM 7070, UNIVAC Solid State 80. In these machines, the basic unit of data was the decimal digit, encoded in one of several schemes, including binary-coded decimal (BCD), bi-quinary and two-out-of-five code. Except for the IBM 1620 and 1710, these machines used word addressing. When non-numeric characters were used in these machines, they were encoded as two decimal digits.

Other early computers were character oriented, providing instructions for performing arithmetic on character strings of decimal numerals, using BCD or excess-3 (XS-3)[1] for decimal digits. On these machines, the basic data element was an alphanumeric character, typically encoded in six bits. UNIVAC I and UNIVAC II used word addressing, with 12-character words. IBM examples include IBM 702, IBM 705, the IBM 1400 series,[2] IBM 7010, and the IBM 7080.

The IBM System/360, introduced in 1964 to unify IBM's product lines, used per character binary addressing, and also included instructions for packed decimal arithmetic as well as binary integer arithmetic, and binary floating point. It used 8-bit characters and introduced EBCDIC encoding, though ASCII was also supported.[3] The Burroughs B2500 introduced in 1966 also used 8-bit EBCDIC or ASCII characters and could pack two decimal digits per byte, but it did not provide binary arithmetic, making it a decimal architecture.

Support for BCD was common in early microprocessors, which were often used in roles like electronic calculators and cash registers where the math was all decimal. Examples of such support can be found in the Intel 8080, MOS 6502, Zilog Z80, Motorola 6800/6809 and most other designs of the era. In these designs, BCD was directly supported in the ALU, allowing it to perform operations on decimal data directly.

As part of the 8080 ISA, Intel BCD opcodes have remained in the x86 family to this day, although they are not supported in long mode. These instructions convert one-byte BCD numbers (packed and unpacked) to binary format before or after arithmetic operations.[4] These operations were not extended to wider formats and hence are now slower than using 32-bit or wider BCD 'tricks' to compute in BCD.[5] The x87 FPU has instructions to convert 10-byte (18 decimal digits) packed decimal data, although it then operates on them as floating-point numbers.

The Motorola 68000 series offered both conversion utilities as well as the ability to directly add and subtract in BCD.[6] These instructions were removed when the Coldfire instruction set was defined.

One of the few RISC designs to directly support decimal is the IBM POWER architecture, which added support for a full decimal implementation in hardware starting with the POWER6 design. Decimal support had been part of their mainframe line, and as part of the broader effort to merge the iSeries and zSeries it was added to the POWER line so that a single processor could support workloads from these older machines with full performance. The IBM POWER6 processor and the IBM System z9 have implemented these types using the densely packed decimal binary encoding rather than BCD,[7] the first in hardware and the second in microcode.

Decimal arithmetic is now becoming more common; for instance, three decimal types with two binary encodings were added to the 2008 IEEE 754r standard, with 7-, 16-, and 34-digit decimal significands.[8]

• Ternary computer

• Schmid, Hermann (1974). Decimal Computation (1 ed.). Binghamton, New York, USA: John Wiley & Sons, Inc. ISBN 0-471-76180-X. Retrieved 2016-01-03. (NB. This title provides detailed description of decimal calculations, including explanation of binary-coded decimals and algorithms.)
• Schmid, Hermann (1983) [1974]. Decimal Computation (1 (reprint) ed.). Malabar, Florida, USA: Robert E. Krieger Publishing Company. ISBN 0-89874-318-4. Retrieved 2016-01-03. (NB. At least some batches of this reprint edition were misprints with defective pages 115–146.)