IBM 1620: Encyclopedia II - IBM 1620 - The 1620's architecture

IBM 1620 - The 1620's architecture

It was a variable "word" length decimal (BCD) computer with a memory that could hold anything from 20,000 to 60,000 decimal digits increasing in 20,000 decimal digit increments. While the 5-digit addresses could have addressed 100,000 decimal digits, no machine larger than 60,000 decimal digits was ever built.

Memory was accessed two decimal digits at the same time (even-odd digit pair for numeric data or one alphameric character for text data). Each decimal digit was 6 bits, composed of an odd parity Check bit, a Flag bit, and four BCD bits for the value of the digit in the following format:

  C F 8 4 2 1

The Flag bit had several uses:

• In the least significant digit it was set to indicate a negative number (signed magnitude).
• It was set to mark the most significant digit of a number (wordmark).
• In the least significant digit of 5-digit addresses it was set for indirect addressing.
• In the middle 3 digits of 5-digit addresses (on the 1620 II) they were set to select one of 7 index registers.

In addition to the valid BCD digit values there were three special digit values (these could NOT be used in calculations):

  C F 8 4 2 1
      1 0 1 0  -  Record Mark (right most end of record)
      1 1 0 0  -  Numeric Blank (blank for punched card output formatting)
      1 1 1 1  -  Group Mark (right most end of a group of records for disk I/O)

Instructions were fixed length (12 decimal digits), consisting of a 2-digit "op code", a 5-digit "P Address", and a 5-digit "Q Address".

Fixed-point data "words" could be any size from two decimal digits up to all of memory not used for other purposes.

Floating-point data "words" (using the hardware floating point option) could be any size from 4 decimal digits up to 102 decimal digits (2 to 100 digits for the mantissa and 2 digits for the exponent).

The machine had no programmer-accessible registers: all operations were memory to memory (including the index registers of the 1620 II).

IBM 1620 - Character and Op codes

The table below lists Alphameric mode Characters (and Op codes).

The table below lists Numeric mode Characters.

IBM 1620 - A flawed architecture

Although the IBM 1620's architecture was very popular in the scientific and engineering community, computer scientist Edsger Dijkstra pointed out several flaws in its design in EWD37, "A review of the IBM 1620 data processing system" (see http://www.cs.utexas.edu/users/EWD/index00xx.html at the Dijkstra archive at the University of Texas).

Dijkstra pointed out flaws including the fact that the 1620's Branch and Transmit instruction together with Branch Back allow a grand total of ONE level of nested subroutine call, forcing the programmer of any code with more than one level to decide where the use of this "feature" would be most effective. He also showed how the paper tape reading support of the 1620 could not properly read paper tapes with record marks on them, since record marks were used to terminate the characters read in storage (one effect of this, although he did not mention it, is that the 1620 cannot duplicate a tape with record marks: when punching a tape and the first record mark that was read in is encountered, the punch instruction punches an EOL on the tape instead and stops punching!).

These flaws (and similar ones in other machines) may have caused IBM's decline, which started in the 1960s, as a serious competitor in scientific computing, opening up a market which led to the development of the DEC PDP-8.

Adapted from the Wikipedia article "The 1620's architecture", under the G.N U Free Docmentation License. Please also see http://en.wikipedia.org/wiki