Chapter 9½ Design of the B 5000 System 131
on how the program references the data word. If the data word is a single variable and not an element of an array, the flag identifies the word as being operand, that is, a data word. If the word is an element of an array, the flag may be used to identify this particular element as an element of data which is not to be processed by the normal program (for example, a boundary point in mesh calculations).
When operating in the character mode, each data word consists of eight alphanumeric characters as illustrated in Fig. 3. Programs in the character mode can address any character in a word. Fields can start at any position in a word. A processor in a single operation can operate on fields of any length up to 63 characters long; operations on fields of greater length can easily be programmed. For example, two 57 character fields could be compared in a single operation.
There are two instances when the character mode operates with words of the type used in the word mode. Operations are provided in the character mode for converting numeric information in the alphanumeric representation to the standard word type of the word mode and vice versa. In both of these instances, the length of the alphanumeric fields being converted to or from the word mode type of word can be no greater than eight characters long. Again, conversion of fields of greater length can easily be programmed.
The purpose of the word mode is to provide the advantages of high-speed parallel operations, floating-point abilities and the inherent information density possible in a binary machine. In the first case, it is economically feasible to provide parallel operations in a word machine; the cost of parallel operations on variable length fields would be prohibitive. In the last case, a given size memory can contain over twenty percent more numeric information if that information is expressed in binary rather than binary-coded decimal, and over eighty percent more information than can be expressed in six-bit alphanumeric representation.
The purpose of the character mode is to provide editing, scanning, comparison and data manipulative abilities (although addition and subtraction are also provided). The type of editing facilities provided obviate the need for the artificial "add-shift-extract-store type of editing. For example, operations are provided for generalized insertion of editing symbols (such as blanks, decimal points, floating dollar signs, etc.) and for the substitution or suppression of any unwanted characters. For those interested in the new area of Information Processing Languages, the character mode is particularly well suited to list structures.
Programs in the B 5000 are composed of strings of syllables. A syllable
is the basic unit of the program and is twelve bits in length. The term
"syllable" is used rather than instruction to distinguish it from conventional
single-address or multi-address instructions. Each program word contains
four syllables and they are executed sequentially in a left-to-right order
within the program word, and sequentially by word. Branching is allowed
to any syllable within a word. Before delving into some of the details
of the internal operation of the B 5000 processor, it is necessary to discuss
stacks, Polish notation, and the Program Reference Table.
The internal organization of single-address computers forces the wasting of both programming and running time for the storage and recall of the intermediate results in the sequence of computation. The data must be placed into the proper registers and memory cells before the operation can be executed, and their contents must often be completely rearranged before the next operation can be performed. Multi-address computers are constructed to make the execution of a few selected operations more efficient, but at the expense of building inefficiencies into all the rest. Automatic programming aids attack this problem indirectly: they relieve the programmer of the need to laboriously code his way around machine design, but they still must provide object coding to accomplish the storage and recall functions. In brief, conventionally designed computers, with or without automatic programming aids, require the wasteful expenditure of programming effort, memory capacity, and running time to overcome the limitations of their internal organization.
The problem is attacked directly in the B 5000 by incorporation of a "pushdown" stack, which completely eliminates the need for instructions (coded or compiled) to store or recall intermediate results.
In a B 5000 processor, the stack is composed of a pair of registers,
the A and B registers, and a memory area. As operands are picked up by
the programs, they are placed in the A register. If the A register already
contains a word of information, that word is transferred to the B register
prior to loading the operand into the A register. If the B register is
also occupied by information, then the word in B is stored in a memory
area defined by an address register S. Then the word A can be transferred
to B and the operand brought into the A register. The new word coming into
the stack has pushed down the information previously held in the registers.
As each pushdown occurs, the address in the S register is automatically
increased by one. The information contained in the registers is the last
information entered into the stack; the stack operates on a "last in-first
out" principle. As
previous | contents | next