Chapter 28 Microprogramming and the design of the control circuits in an electronic digital computer 339

operations called for in the control register unit. The fourth column shows which conditional flip-flop, if any, is to be set and the digit which is to be used to set it; for example, (1)C₈ means that flip-flop number 1 is set by the sign digit of the number in register C, while (2)G₁ means that flip-flop number 2 is set by the least significant digit of the number in register C. In the case of unconditional micro-orders columns 5 and 7 are blank and column 6 contains the address of the next micro-order to be executed. In the case of conditional micro-orders column 5 shows which flip-flop is used to operate the conditional switch and columns 6 and 7 give the alternative addresses to which control is to be sent when the conditional flip-flop contains a 0 or a 1 respectively.

Micro-orders 0 to 4 are concerned with the extraction of orders from the store. They serve to bring about the transfer of the order from the store to register E and then cause the five most significant digits of the order to be placed in register II with the result that control is transferred to one of the micro-orders 5 to 15, each of which corresponds to a distinct order in the machine order code. In this way the sequence of micro-orders needed to perform the particular operation called for is begun.

The way in which the various operations are performed can be followed from Table 2. In the section dealing with multiplication, it is assumed that numbers lie in the range -1 < x < 1 and that negative numbers are represented in the machine by their complements with respect to 2. It will be noted that the process of drawing up a micro-programme is very similar to that of drawing up an ordinary programme for an automatic computing machine and the problems involved are very much alike.

The assumption that all micro-operations take the same length of time to perform is not likely to be borne out in practice. In particular in a parallel machine it may not be possible to design an adder in which the carry propagation time is sufficiently short to enable an addition to be performed in substantially the same length of time as that taken for a simple transfer. It will be necessary, therefore, to arrange that the wave-form generator feeding the decoding tree should, when suitably stimulated by a pulse from one of the outputs from matrix A, supply a somewhat longer pulse than that normally required. Other operations may take many times as long to perform as an ordinary micro-order; for example, access to and from the store (particularly if a delay store is used) and operation of the input and output devices of the machine. The sequence of operations in the micro-programme must therefore be interrupted. One way of doing this is to prevent pulses from the wave-form generator reaching the decoding tree during the waiting period. This method, although quite feasible, appears to involve just the kind of complication which the present system is designed to avoid. A more attractive system is to make the machine wait on a conditional micro-order which transfers control back to itself unless the associated conditional flip-flop is set. Setting of this flip-flop takes place when the operation is completed, and control then goes to the next micro-order in the sequence. The machine is thus in a condition of 'dynamic stop' while waiting for the operation to be completed. This system has the advantage that no complication is introduced into the units supplying the wave-forms to the decoding tree and that the control equipment required is similar to that already provided for other purposes.

It will be seen that the equipment needed to execute a complicated order in the machine order code is of the same form as that required for a simple one, namely outlets from the decoding tree and diodes in the matrices. Quite complicated orders can, therefore, be built into the machine without difficulty. In particular, arithmetical operations on numbers expressed in floating binary form and other similar operations can be micro-programmed and it is found that they do not involve very large numbers of micro-orders. For example, a micro-programme providing for the floating-point operations of addition, subtraction, and multiplication needs about 70 micro-orders. The switching system in the arithmetical unit must, of course, be designed with these operations in view. The decoding tree and matrices of a parallel machine with 40 digits in the arithmetical unit and provision for 256 micro-orders would only amount to about 15% of the total equipment in the machine, so that it appears that such a machine can well be provided with built-in facilities of considerable complexity.

The number of micro-orders needed in a complicated micro-programme can sometimes be reduced by making use of what might be called micro-subroutines. For example, when two numbers have to be added together in a floating binary machine, some shifting of one of them is usually necessary before the addition can take place. By making the micro-orders for this shifting operation serve also when a multiplication is called for, considerable saving is effected.

Four registers is the bare minimum needed in the arithmetical unit in order to enable the basic arithmetical operations to be performed. If any extension or refinement of the facilities provided is required, it may be necessary to increase the number of registers.