Multiplier Circuit Suitable for Obtaining a Negative Product of a Multiplier and a Multiplicand

The present invention relates to the improvement of a parallel multiplier which is constituted on the basis of Booth's secondary algorithm.

Various methods for realizing the performing of high-speed parallel arithmetic operations of binary form have been proposed (e.g., refer to NIKKEI ELECTRONICS, May 29, 1978, pages 76 to 89). Use of Booth's secondary algorithm is known as one of these methods. In Booth's algorithm, the product X.multidot.Y is obtained in the following manner from the data X and Y which are expressed as the 2's complement. First, the input X and Y are modified as follows: ##EQU1## where, x.sub.m and y.sub.n are sign bits and n is an even number.

The product X.multidot.Y is then modified as shown in the following expression: ##EQU2## where,

y0=0, pi=y2i+y2i+1-2.multidot.y2i+2,i=0, 1 . . . (n-1)/2, and yn is most significant bit MSB.

pi is derived by decoding three continuous bits of the multiplier Y. Pi has a value of either one of 0, .+-.1, and .+-.2. Therefore, the value of the partial product X.multidot.pi is either one of 0, .+-.X, and .+-.2X. The partial product pi.multidot.X can be produced by a simple process of inverting and/or shifting the multiplicand X, on the basis of the value of pi. The number of partial products is n/2. Therefore, in the parallel multiplier using Booth's algorithm, it is sufficient to use n/2 partial product adding circuits to add the partial products Pi.multidot.X. Accordingly, the parallel multiplier using Booth's algorithm has an advantage in that the amount of hardware necessary is less than that required by the multiplier using other algorithms. Thus, the parallel multiplier using Booth's algorithm is suitable for the realizing of a parallel multiplier of a large scale of 8.times.8 bits or more in a one-chip integrating circuit.

A circuit for adding (.+-.X.multidot.Y+Z) or subtracting (.+-.X.multidot.Y-Z) another data Z, which is expressed as the 2's complement, to or from the result of the multiplication X.multidot.Y or -X.multidot.Y, is known. FIGS. 1 and 2 show examples of the constitution of such a circuit. In FIG. 1, multiplier 1 receives the multiplicand X and multiplier Y, and obtains the product X.Y. The output of multiplier 1 is converted into -X.multidot.Y by 2's complement circuit 2. The output X.multidot.Y of multiplier 1 is selected by switch 3 which is controlled by a signal OP. The output -X.multidot.Y of 2's complement circuit 2 is selected by switch 4 which is controlled by a signal OP. The selected output X.multidot.Y or -X.multidot.Y is supplied to adding/subtracting circuit 5. The data Z is supplied to adding/subtracting circuit 5. Adding/subtracting circuit 5 performs the adding or subtracting operation on the basis of a control signal CON. In this manner, four types of product and sum calculations (.+-.X.multidot.Y.+-.Z) are selectively performed under the control of the signals OP, OP, and CON.

On the other hand, in FIG. 2, the data Z is supplied to 2's complement circuit 2. Either one of the data Z and the output -Z of 2's complement circuit 2 is selected by switch 3, which is controlled by signal OP, and switch 4 which is controlled by signal OP. The selected output is supplied to adding/subtracting circuit 5. Adding/subtracting circuit 5 adds or subtracts the output X.multidot.Y of multiplier 1 to or from data Z or -Z, on the basis of the foregoing control signals. In this manner, the four types of product sum calculations are selectively executed.

It is not desirable to use 2's complement circuit 2 in order to allow the four types of product and sum calculations to be selectively performed as mentioned above, due to the increase in the amount of hardware required. Moreover, the arithmetic operating time of 2's complement circuit 2 is long. In particular, if the parallel multiplication-type multiplier is used in order to achieve high-speed arithmetic operation, 2's complement circuit 2 is also constituted as the parallel type, so that the operating time of complement circuit 2 is long because the propagation time of the carry signal is long. If a carry look-ahead circuit is used to reduce the propagation time of the carry signal, the amount of hardware required once more increases considerably, resulting in a "vicious circle" situation.

The foregoing problems, notably the increase in the amount of hardware and the increase in the operation time also arise when the circuit for obtaining the negative product of -X.multidot.Y from the multiplicand X and multiplier Y is constituted by a multiplier for obtaining the product X.multidot.Y on the basis of Booth's secondary algorithm and a 2's complement circuit which receives the output of the multiplier as shown in FIGS. 1 and 2.

An object of the present invention to provide a parallel multiplier which can efficiently perform the negative product calculation at high speed, using only a small amount of hardware.

To accomplish the above object, a parallel multiplier according to the invention comprises:

pi generating means (11) for receiving a multiplier Y and producing pi0 therefrom (in which it is defined that Pi=y2i+y2i+1-2.multidot.y2i+2, pi=y2i+y2i+1-2.multidot.y2i+2, and y0="0", and i=0, 1 . . . , (n-1)/2, and yj is bit data of the jth bit of the multiplier Y);

partial-product generating means (21) for receiving the pi from the pi generating means (11) and the multiplicand X, and producing the partial products pi.multidot.X as the products of the multiplicand X and the pi; and

partial-product adding means (21) for producing the negative produce ##EQU3## of the multiplicand X and multiplier Y, by weighting 2.sup.2i to the partial products pi.multidot.X from the partial-product producing means (21) and by adding the resultant data.

With the parallel multiplier having the above constitution according to the present invention, it is possible to obtain a parallel multiplier which can perform negative multiplication, using only a small amount of hardware in comparison to a conventional parallel multiplier. This is because the conventional pi generator and pi generator can be constituted so as to have an almost identical structure. Further, since the 2's complement circuit and the like are not used, the operating speed is almost equal to that of the conventional parallel multiplier. The function whereby the positive multiplication and negative multiplication can be selectively performed can be easily added as necessary.

FIGS. 1 and 2 are block diagrams showing examples of constitutions of circuits for performing the product/sum calculations;

FIG. 3 is a block diagram showing a structure of a parallel multiplier according to the first embodiment of the present invention;

FIG. 4 is a block diagram showing an example of practical constitution of the parallel multiplier shown in FIG. 3;

FIGS. 5A and 5B are block diagrams showing examples of constitutions of a Pi producing circuit shown in FIG. 3;

FIG. 6 is a circuit diagram showing practical a constitution of a Pi producing circuit for producing Pi by using a Pi producing circuit and a fundamental cell;

FIG. 7 is a circuit diagram showing an example of a modification of the Pi producing circuit shown in FIG. 6;

FIG. 8 is a circuit diagram showing another practical constitution of the Pi producing circuit and the fundamental cell;

FIG. 9 is a circuit diagram showing examples of modifications of the Pi producing circuit and the fundamental cell shown in FIG. 8;

FIG. 10 is a circuit diagram showing an example of a constitution of a Pi producing circuit including no Pi producing circuits, which is a modification of the Pi producing circuit shown in FIG. 6;

FIG. 11 is a circuit diagram showing an example of a modification of the Pi producing circuit shown in FIG. 10;

FIG. 12 is a circuit diagram showing an example of a constitution of a Pi producing circuit including no Pi producing circuits, which is a modification of the Pi producing circuit shown in FIG. 8;

FIG. 13 is a circuit diagram showing an example of a constitution of a Pi producing circuit including no Pi producing circuits, which is a modification of the Pi producing circuit shown in FIG. 9;

FIG. 14 is a block diagram showing a constitution of a parallel multiplier according to a second embodiment of the present invention;

FIGS. 15A and 15B are block diagrams showing constitutions of the Pi or Pi producing circuit shown in FIG. 14;

FIG. 16 is a circuit diagram showing an example of a practical constitution of a Pi or Pi producing circuit and a fundamental cell of the parallel multiplier shown in FIG. 14;

FIG. 17 is a circuit diagram showing a constitution of a Pi or Pi producing circuit formed by modifying the Pi producing circuit shown in FIG. 11;

FIG. 18 is a circuit diagram showing a constitution of a Pi or Pi producing circuit formed by modifying Pi producing circuit shown in FIG. 8; and

FIG. 19 is a block diagram showing an example of a circuit to which the parallel multiplier shown in FIG. 14 is applied.

An embodiment of the present invention will be described in detail hereinbelow with reference to the drawings.

FIG. 3 shows a multiplier to obtain the product -X.multidot.Y for the inputs X and Y. This multiplier consists of pi generator 11 and partial product generating/adding circuit 21. pi generator 11 decodes the multiplier Y (=yn, yn-1, . . . , y1) and produces the pi. Partial product generating/adding circuit 21 receives the pi and the multiplicand X (=x.sub.m, x.sub.m-1, . . . , x.sub.1) and produces the partial products (pi.multidot.X) on the basis of the Booth's algorithm and accumulates the partial products (.SIGMA.pi.multidot.X.2.sup.2i). In this manner, the negative product -X.multidot.Y is obtained with respect to the inputs X and Y.

The above constitution is based on the following principle. As mentioned above, the equation (1) is satisfied. ##EQU4## Therefore, the equation (2) is satisfied. By modifying the equation (2), the equation (3) is satisfied. ##EQU5## On the other hand, -pi=pi is satisfied. Thus, the equation (4) is satisfied. ##EQU6##

From the above equations, it will be understood that the negative product -X.multidot.Y is obtained by executing the following steps (1) to (3).

(1) pi generator 11 in FIG. 3 produces the pi.

(2) Partial product generating/adding circuit 21 produces the partial products pi.multidot.X.

(3) Partial product generating/adding circuit 21 adds the partial products produced. ##EQU7##

Therefore, in FIG. 3, a circuit to receive the multiplier Y and produce the pi is arranged in order to obtain the output of -X.multidot.Y.

Practical examples of pi generator 11 and partial product producing/adding circuit 21 will now be described with reference to FIG. 4. FIG. 4 shows a constitution in the case of performing the multiplication of 8 bits.times.8 bits. In FIG. 4, pi generator 11 consists of decoders DC.sub.1 to DC.sub.4. Decoders DC.sub.1 to DC.sub.4 receive continuous three bits of the multiplier Y. On the other hand, circuit 21 is constituted by a plurality of sign cells S and a plurality of basic cells B which are arranged in a matrix form. Decoders DC.sub.1 to DC.sub.4 decode three continuous bits of the multiplier Y on the basis of the Booth's algorithm and produce the pi (selection signal SE in the actual circuit). Decoders DC.sub.1 to DC.sub.4 supply the produced pi to the cells of the corresponding row. When the pi is "-1" and "-2", decoders DC.sub.1 to DC.sub.4 supply a carry CA to the basic cells B of the lowest column in order to obtain the 2's complement of the partial products pi.multidot.X. Each basic cell receives the pi from corresponding decoders DC.sub.1 to DC.sub.4, carry C from the adjacent cell, bit data xj of the multiplicand X corresponding to that cell, and bit data xj-1 which is lower than the corresponding bit by one digit. It is assumed that x0=0. Each cell supplies the sum data S to the cell of the lower row of the same digit and supplies the carry C to the adjacent cell which is upper by one digit. The ordinarily known sign cells may be used. The problems of sign processes in the Booth's algorithm have been described in, for example, "DIGITAL CIRCUITS FOR BINARY ARITHMETIC", the Machmillan Press, Ltd., pages 156 and 157; therefore, its description is omitted here. In FIG. 4, partial product generating/adding circuit 21 has been constituted by a plurality of cells arranged in a matrix form. The invention is not limited to this. FIG. 4 shows an example of each circuit. For example, circuit 21 can be also constituted by a partial product generator consisting of the cells of one row in such a manner that the output of this generator is shifted and again supplied as the input data to the partial product generator and the partial products are sequentially weighted by 2.sup.2i and then added. The arrangement of the cells, the supply of the carry and sum data, and the like are not limited to those shown in FIG. 4. For example, the Wallance Tree method, CSA method, or the like may be used.

The practical constitution of pi generator 11 will now be explained.

pi generator 11 shown in FIG. 5A comprises: ordinarily known circuit 111 to produce the pi; and inverters 112 to 114 to invert three bits y2i, y2i+1, and y2i+2 of the multiplier Y and supply the inverted data to circuit 111. pi generator 11 (DC.sub.1) corresponding to i=0 sets the bit data y0 to the fixed value of "1" as shown in FIG. 5B. With such a constitution, by inputting the inverted data y2i, . y2i+1, and y2i+2 ordinarily known Pi generator 111, the pi is produced.

A practical example of the circuit to produce the pi by inverting the multiplier Y and supplying to the Pi generator as shown in FIGS. 5A and 5B will be described with reference to FIG. 6. FIG. 6 shows a part 11A of the pi generator and a part 21A of the partial product generating/adding circuit. Part 11A corresponds to one of the decoders in FIG. 4. Part 21A corresponds to the cells of the lowest column among the cells arranged in a matrix form in FIG. 4. In the constitution shown in FIG. 6, three continuous bits y2i, y2i+1, and y2i+2 of the multiplier Y are supplied to inverter 31. The data y2i from inverter 31 is supplied to exclusive OR gate (hereinafter, abbreviated to as an EX OR gate) 112, AND gate 113, and an inverted input terminal of AND gate 114. The bit data y2+1 from inverter 31 is supplied to EX OR gate 112, AND gate 113, and an inverted input terminal of AND gate 114. The bit data y2i+2 from inverter 31 is supplied to the inverted input terminal of AND gate 113 and to AND gate 114. Further, the bit data y2i+2 is supplied as a carry CA to cell 21A of the lowest column. The output of EX OR gate 112 is output as the selection signal A. The outputs of AND gates 113 and 114 are input to OR ate 115. The output of OR gate 115 is output as the selection signal B. The signal A from pi generator 11A is supplied to AND gate 211. The bit data xj (in this case, x1) of the multiplicand X is also input to AND gate 211. The signal B is supplied to AND gate 212. The bit data xj-1 (in this case, x0) is also supplied to AND gate 212. The output of AND gates 211 and 212 are supplied to OR gate 213. The output of OR gate 213 is input to EX OR gate 214. Further, the carry CA (y2i+2) from generator 11A is supplied to EX OR gate 214. The output of EX OR gate 214 is set to one bit (pi.multidot.X)j of the partial product or 1's complementary number of (pi.multidot.X)j. One bit (pi.multidot.X)j of the partial product, the carry CA, and the sum data S.sub.in from the cells of the preceding row are supplied to full adder 215. Full adder 215 adds those input data and outputs the sum data S.sub.out and carry C. Namely, in FIG. 6, generator 11A outputs the signals A and B to instruct that partial product generating/adding circuit 21A produces the multiplicand X and data 2X or "0". Generator 11A outputs the signal CA to instruct that the produced data is inverted or directly output as the partial product and the partial product is converted to the 2's complement when the partial product becomes either one of the 1's complement of the multiplicand X, 1's complement of 2X, and 1's complement of "0". The truth table of the circuit shown in FIG. 6 is as shown in Table 1 and a desired partial product pi.multidot.X is obtained.

The circuits as shown in FIGS. 7 to 9 may be also used as pi generator 11 and partial product generating/adding circuit 21 (cell B). FIG. 7 shows a modified example 11B of pi generator 11A shown in FIG. 6. Therefore, only the structure of pi generator 11B will be explained. The bit data y2i from inverter 31 is input to EX OR gate 121, AND gate 122, and inverted input terminals of AND gate 123 and OR gate 124. The bit data y2i+1 is input to EX OR gate 121, AND gate 122, and inverted input terminals of AND gate 123 and OR gate 124. The bit data y2i+2 is input to the inverted input terminal of AND gate 122 and to AND gates 123 and 125. The output of OR gate 124 is input to AND gate 125. The outputs of AND gates 122 and 123 are input to OR gate 126. The output of EX OR gate 121 is set to the same signal as the control signal A. The output of OR gate 126 is set to the same signal as the control signal B. However, the carry CA which is output from AND gate 125 is not the same as the carry CA in FIG. 6. Namely, in the circuit of FIG. 6, when all of three bits of the multiplier Y are "0", the signal CA is also set to "1" in order to form the 2's complement from the 1's complement of "0". However, in FIG. 7, if all of three bits of the multiplier Y are "0", the carry CA is set to "0". However, there is no difference between the results of the arithmetic operations.

The structure of pi generator 11C in FIG. 8 will now be described. The bit data y2i from inverter 31 is input to EX OR gate 131, AND gate 132, and inverted input terminals of AND gate 133 and OR gate 134. The bit data y2i+1 is input to EX OR gate 131, AND gate 132, and inverted input terminals of AND gate 133 and OR gate 134. The bit data y2i+2 is input to inverted input terminals of AND gates 132 and 135 and to AND gates 133, 136, and 137. The output of EX OR gate 131 is input to AND gates 135 and 136. The output of OR gate 134 is input to AND gate 137. The structure of partial product generating/adding circuit 21B of FIG. 8 will now be described. The output .phi.d of AND gate 135 is supplied to AND gate 221. The bit data xj (in this case, x1) of the multiplicand X is also supplied to AND gate 221. The output .phi.e of AND gate 136 is supplied to AND gate 222. The bit data xj (in this case, x1) is also supplied to AND gate 222. The output .phi.f of AND gate 132 is supplied to AND gate 223. The bit data xj-1 (in this case x0) of the multiplicand X is also supplied to AND gate 223. The output .phi.g of AND gate 133 is supplied to AND gate 224 The bit data xj-1 (in this case, X0) is also supplied to AND gate 224. The outputs of AND gates 221 to 224 are supplied to OR gate 225. Full adder 226 receives the carry CA which is output from AND gate 137, one bit (pi.multidot.X)j of the partial product which is output from OR gate 225, and the sum data Sin from the cell of the preceding row. Full adder 226 then adds those input data and outputs the sum data S.sub.out an carry C. The truth table of the circuit shown in FIG. 8 is as shown in Table 2.

FIG. 9 shows a modified example of FIG. 8. First, the structure of generator 11D will be explained. The bit data y2i from inverter 31 is input to EX OR gate 141, AND gates 142 and 146, and inverted input terminals of AND gates 143 and 145 and OR gate 144. The bit data y2i+1 is input to EX OR gate 141, AND gate 142, inverted input terminals of AND gates 143 and 145, an inverted input terminal of OR gate 144, and AND gate 146. The bit data y2i+2 is input to inverted input terminals of AND gates 142 and 147 and to AND gates 143, 148, and 149. The output of EX OR gate 141 is input to AND gates 147 and 148. The output of OR gate 144 is input to AND gate 149. The outputs of AND gates 145 and 146 are supplied to OR gate 150. The structure of generating/adding circuit 21C will now be described. The output .phi.d of AND gate 147 is supplied to a control terminal of three state inverter 231. The bit data xj (in this case x1) of the multiplicand X is input to inverter 231. The output .phi.e of AND gate 148 is supplied to a control terminal of three-state inverter 232. The bit data xj (in this case, x1) is input to inverter 232. The output .phi.f of AND gate 142 is supplied to a control terminal of three-state inverter 233. The bit data xj-1 (in this case, x0) of the multiplicand X is input to inverter 233. The output .phi.g of AND gate 143 is supplied to a control terminal of inverter 234. The bit data xj-1 (in this case, x0) is input to inverter 234. The outputs .phi.h of OR gate 150 is supplied to a control terminal of inverter 235. The data "0" is input to inverter 235. The wired-OR is calculated from the outputs of inverters 231 to 235 and supplied to inverter 236. The output of inverter 236 is set to one bit (pi.multidot.X)j of the partial product. One bit (pi.multidot. X)j of the partial product, carry CA, and sum data S.sub.in are supplied to full adder 237. Full adder 237 adds those input data and outputs the sum data S.sub.out and carry C. The truth table of the circuit shown in FIG. 9 is as shown in Table 3. A desired partial product pi.X is derived.

The constitutions of pi generators 111A to 111D and partial product generating/adding circuit 21A to 21C are not limited to those shown in FIGS. 6 to 9. The conventionally known constitution is used.

Foregoing pi generator 11 obtains the pi by operating the multiplier Y. However, the invention is not limited to this method but may use a unique constitution as the pi generator. An embodiment with such a constitution will be described with reference to FIGS. 10 and 11.

pi generator 11E shown in FIG. 10 is obtained by modifying pi generator 111A shown in FIG. 6. pi generator 11E differs from pi generator 111A with respect to the point that the bit data y2i+2 is inverted by inverter and output as the carry CA. The constitution of partial product generating/adding circuit 21A is the same as the constitution shown in FIG. 6. Circuit 21A also operates in accordance with the truth table shown in Table 1 and the partial product (pi.multidot.X)j is derived. This partial product, carry CA, and sum data Sin from the preceding row are added by full adder 215 and the carry C and sum data S.sub.out are output.

It is a feature of the circuit shown in FIG. 11 that OR gate 162 and AND gate 163 are used in place of inverter 161 in Pi generator 11E shown in FIG. 10. OR gate 162 receives the bit data y2i and y2i+1. AND gate 163 receives the output of OR gate 162. The bit data y2i+2 is set to the inverted input.

On the other hand, pi generator 11G shown in FIG. 12 is constituted by modifying pi generator 111C shown in FIG. 8. Generator 11G has the following features.

(1) The output of AND gate 135 is supplied as the signal .phi.e to AND gate 222. The output of AND gate 136 is supplied as the signal .phi.d to AND gate 221. The output of AND gate 132 is supplied as the signal .phi.g to AND gate 224. The output of AND gate 133 is supplied as the signal .phi.f to AND gate 223.

(2) The carry CA is set to "1" when either the signal .phi.e or the signal .phi.g is set to "1".

These features will be explained hereinbelow by use of the practical circuit constitution.

(1) The signals .phi.d and .phi.e in FIG. 8 are exchanged and the signals .phi.f and .phi.g are exchanged.

(2) OR gate 134 is changed to OR gate 164 having no inverted input terminal. AND gate 137 is replaced by AND gate 165 which receives the output of OR gate 164 and the bit data y2i+2 as the inverted input.

pi generator 11H shown in FIG. 13 is formed by changing the structure of pi generator 111D in FIG. 9. As compared with pi generator 111D, generator 11H has the following features.

(1) The output of AND gate 147 is supplied as the signal .phi.e to a control terminal of inverter 232. The output of AND gate 148 is supplied as the signal .phi.d to a control terminal of inverter 231. The output of AND gate 143 is supplied as the signal .phi.f to a control terminal of inverter 233. The output of AND gate 142 is supplied as the signal .phi.g to a control terminal of inverter 234. (The outputs .phi.d and .phi.e in FIG. 9 are exchanged and the outputs .phi.f and .phi.g are exchanged).

(2) OR gate 166 is used in place of OR gate 144 having two inverted input terminals. AND gate 167 which receives the output of OR gate 166 and the bit data y2i+2 as the inverted input is used in place of AND gate 149. (The carry CA is set to "1" when either the output .phi.e or the output .phi.g is set to "1".

The generator shown in FIG. 13 obtains the negative product -X.multidot.Y in accordance with a truth table shown in Table 3.

As an example of application of the present invention, an explanation will be made with respect to a multiplier of the positive/negative switching type which can selectively obtain the negative multiplication output (-X.multidot.Y) or positive multiplication output (X.multidot.Y). FIG. 14 shows a conceptional constitution of such a circuit. The circuit shown in FIG. 14 comprises pi or pi generator 41 and partial product generating/adding circuit 21. pi or pi generator 41 selectively produces the pi or pi from the multiplier Y in accordance with the logic level of the control signal OP. Partial product generating/adding circuit 21 produces the partial products of the multiplicand X and the pi or pi from generator 41 and adds them. Circuit 21 may have the same constitution as circuit 21 in FIG. 3.

An embodiment of foregoing pi or pi generator 41 will now be described. pi or pi generator 41 shown in FIG. 15A uses EX OR gates 311 to 313 in place of inverters 112 to 114 in pi generators 11 in FIGS. 5A and 5B. EX OR gates 311 to 313 receive at one end the bit data y2i, y2i+1, and y2i+2 and receive at the other end the control signal OP. Therefore, the bit data y2i, y2i+1, and y2i+2 are directly input to pi generator 111 or inverted and input thereto in accordance with the level "0" or "1" of the control signal OP. Therefore, the output of circuit 41 is set to the pi or pi on the basis of whether the level of the signal OP is "0" or "1". In this case, p0 or p0 generator corresponding to i=0 can directly use the signal OP as the y.sub.0 input as shown in FIG. 15B. Therefore, pi or pi generator is obtained by using EX OR gates instead of inverter 31 in FIGS. 6 to 9.

pi or pi generator 41A in FIG. 16 uses EX OR gate 411 which receives the bit data y2i+2 and control signal OP in place of inverter 161 in pi generator 11E in FIG. 10. Therefore, when the level of the signal OP is set to "1", generator 41A and partial product generating. adding circuit 21A operate in accordance with Table 1 and one bit (pi.multidot.X).sub.j of the partial product is obtained. Thus, the negative product -X.multidot.Y is derived. When the level of the signal OP is set to "0", generator 41A and circuit 21A operate in accordance with Table 4 and one bit (pi.multidot.X).sub.j of the partial product is obtained. Thus, the positive product X.multidot.Y is obtained.

pi or pi generator 41B shown in FIG. 17 is constituted by modifying pi generator 11F in FIG. 11. As compared with generator 11F, generator 41B has such a feature that the carry CA is set to the level as shown in Table 4 or 5 in accordance with the level "0" or "1" of the signal OP. Practically speaking, OR gates 421, 422, and 425 and AND gates 423 and 424 are used in place of OR gate 162 and AND gate 163 in generator 11F.

pi or pi generator 41D can be constituted by modifying pi generator 11C shown in FIG. 8 as shown in FIG. 18. When the signal OP is set to the "0" level, generator 41D operates in accordance with Table 6. When the signal OP is set to the "1" level, generator 41D operates in accordance with Table 3. Namely, on the basis of the level of the signal OP, the signals .phi.d and .phi.e are exchanged and the signals .phi.f and .phi.g are exchanged and the level of the carry CA is inverted. Generator 41D is constituted by adding EX OR gate 431, AND gates 432 to 440, and OR gates 441 to 444 to generator 11G shown in FIG. 12.

Similarly, pi generator 11H shown in FIG. 13 can be also changed to pi or pi generator. In this case, the generator operates as shown in Table 3 or 7.

With the foregoing multiplier of the positive/negative switching type, it is possible to constitute the product sum circuit for selectively performing four kinds of calculations of .+-.X.multidot.Y.+-.Z between the output (X.multidot.Y or -X.multidot.Y) of the multiplier and another data Z which is expressed as a 2's complement. FIG. 19 shows an example of such a circuit. In FIG. 19, adding/subtracting circuit 5 receives the data from multiplier 51 of the positive/negative switching type and the data Z and adds or subtracts the data Z to or from the output data of multiplier 51.

As described above, with the parallel multiplier according to the invention, the negative product (-X.multidot.Y) of the multiplicand data X and the multiplier data Y is obtained by merely slightly changing the ordinary multiplying decoder in the parallel multiplier to which the secondary Booth's algorithm was applied. Therefore, as compared with the constitution of the conventional circuit to obtain the negative product, the circuit of 2's complement is not used. Therefore, the amount of hardware is little and the problems such that the high speed operation is obstructed due to the propagation of the carry and the like are not caused. Moreover, the number of circuits which are added when the multiplying decoder is modified is very small as shown in each of the foregoing embodiments. The delay time of the operation which is caused by adding the circuits is up to the time corresponding to only a few gates and is very small as compared with the delay time of the operation which is caused by the carry propagation of the circuit of 2's complement.

According to the invention, the circuit can be easily changed so as to selectively obtain the positive product (X.multidot.Y) or negative product (-X.multidot.Y) in response to the control signal. In this case, the number of circuits which are added is small as mentioned above and there is such an advantage that the delay time of the operation is small. Moreover, by combining the adding/subtracting circuit to the parallel multiplier according to the invention, four kinds of calculations (.+-.X.multidot.Y.+-.Z) can be selectively easily executed between the input data and another input data Z. The flexibility in digital signal process can be increased.

In the foregoing embodiments, the pi was produced and the negative product -X.multidot.Y was obtained on the basis of the pi for convenience of explanation. However, as mentioned above, -pi=pi. Therefore, the embodiments can be also considered such that -pi is produced and then the partial product is obtained.

On the other hand, the circuit diagrams shown in the diagrams are nothing but the examples. The invention is not limited to only those circuits.