JPS63167971A - Arithmetic unit - Google Patents

Arithmetic unit

Info

Publication number
JPS63167971A
JPS63167971A JP31182486A JP31182486A JPS63167971A JP S63167971 A JPS63167971 A JP S63167971A JP 31182486 A JP31182486 A JP 31182486A JP 31182486 A JP31182486 A JP 31182486A JP S63167971 A JPS63167971 A JP S63167971A
Authority
JP
Japan
Prior art keywords
input
numerical value
output
vector
matrix
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
JP31182486A
Other languages
Japanese (ja)
Inventor
Yuji Okuto
奥戸 雄二
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
NEC Corp
Original Assignee
NEC Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by NEC Corp filed Critical NEC Corp
Priority to JP31182486A priority Critical patent/JPS63167971A/en
Publication of JPS63167971A publication Critical patent/JPS63167971A/en
Pending legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F15/00Digital computers in general; Data processing equipment in general
    • G06F15/76Architectures of general purpose stored program computers
    • G06F15/80Architectures of general purpose stored program computers comprising an array of processing units with common control, e.g. single instruction multiple data processors
    • G06F15/8053Vector processors

Abstract

PURPOSE:To multiply a matrix and a vector at a high speed by constituting the titled unit so that N pieces of operation units are all units, a first and fourth numerical values, and a second numerical value are inputted from the outside so as to be common, and different, respectively, and the fifth numerical value is connected so as to become the fourth numerical value. CONSTITUTION:Operation units [PU(1)-PU(N)] are all PUs, the first and the fourth numerical values, and the second numerical value are inputted from the outside so as to be common, and different, respectively, and a fifth numerical value becomes the fourth numerical value, and also, connected so as to be extracted as an output, as well. The product of a vector <BM> and a matrix CM,N is shown as <EN>=CM,N*<BM>, therefore, from an input (b), a component bm of <BM> is inputted to each PU at every unit operation time t1 in order of (m)=1, 2,...M, and when a component cm,n of the matrix CM,N is inputted from the upper part from an input an by in synchronism with said input, a vector <EN> is obtained as a group of an output en of PU(n) after t1 second after bM and cM,n have been inputted. In such a way, an operation speed of M times is obtained, comparing with a conventional method.

Description

【発明の詳細な説明】 (産業上の利用分野) 本発明は、ベクトルとマトリックスの相互の乗算を行う
為の演算装置、並びに、それを用いて、演算する方法に
関するものである。
DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to an arithmetic device for mutually multiplying vectors and matrices, and a method of performing arithmetic operations using the device.

(従来技術とその簡題点) 近年、集積回路技術の発達により、各種の演算装置が、
提供されるようになって来ている。例えば2つの数値を
乗算し、その結果を第3の数値に加えて出力する演算装
置や、二次元マトリックス同志を乗算する演算装置等が
、報告されているが、ベクトルとマトリックスを高速に
乗算する演算装置は、未だに報告されていす、現在では
ベクトルとマトリックスの乗算の場合は、その各々成分
同志を順次掛は合わせ、それらを順次足し合わせる事に
依って、行われて居る。その為に演算に要する時間が長
(、高速の演算は不可能であった。
(Prior art and its simple issues) In recent years, with the development of integrated circuit technology, various arithmetic devices have become
It is becoming available. For example, there have been reports of arithmetic devices that multiply two numbers, add the result to a third number, and output it, and arithmetic devices that multiply two-dimensional matrices. Arithmetic devices are still being reported, and at present, multiplication of vectors and matrices is carried out by multiplying their respective components in sequence and adding them in sequence. Therefore, the time required for calculation was long (and high-speed calculation was impossible).

本発明の目的は、このような従来の欠点を除き、ベクト
ルとマトリックスの乗算を、高速に行う事の可能な演算
装置、並びに、それを用いて演算する方法を提供する事
にある。
SUMMARY OF THE INVENTION An object of the present invention is to provide an arithmetic device capable of performing vector-matrix multiplication at high speed, as well as a method of arithmetic operations using the same, while eliminating such conventional drawbacks.

(問題点を解決するための手段) 本発明によると、入力として手えられる第一の数値と第
二の数値間で乗算を行う機能と、その結果得られる第三
の数値と、入力として与えられる第四の数値との間で加
算を行う機能と、その結果として得られる第五の数値を
、出力とするような演算機能を有する演算単位(以下P
Uと略記する)を有限の数N個に並べ、全てのPUで、
第1の数値が外部よりの共通の入力として、与えられ、
第2の数値は各々のPUに異なった物として外部より、
入力として与えられ、PUの第5の数値(出力)が第4
の数値として供給されると共に、出力としても取り出せ
る様に接続された事を特徴とする演算装置が得られる。
(Means for Solving the Problems) According to the present invention, there is provided a function of performing multiplication between a first numerical value and a second numerical value obtained as input, a third numerical value obtained as a result, and An arithmetic unit (hereinafter referred to as P
(abbreviated as U) are arranged into a finite number N, and for all PUs,
The first numerical value is given as a common input from the outside,
The second number is different for each PU from the outside,
given as input, the fifth number (output) of PU is the fourth
There is obtained an arithmetic device characterized in that the numeric value is supplied as a numerical value and is connected so that it can be taken out as an output.

更に、本発明によると、請求の範囲の、演算装置に、第
一の入力として、ベクトル〈BM〉の成分bm(mは、
1からM迄の整数)を、順次PUの単位演算時間、t1
、置きに入力し、その入力に同期させて、各々のPUの
第二の入力端子に、マトリックスCM、Nの成分cml
(1は、1からN迄の整数であり、mは、1からM迄の
整数)を、mの価がbmのそれと同期するようにして、
各々1番目のPUに入力し、m=Mの入力の後t1に、
PUの出力、en(nは、lからN迄の整数)を、ベク
トル、〈EN〉の形に、出力する事を特徴とするベクト
ルとマトリックスの乗算方法が得られる。
Furthermore, according to the present invention, the component bm (m is
(an integer from 1 to M) are sequentially calculated as the unit operation time of the PU, t1
, and in synchronization with the input, the component cml of the matrix CM,N is input to the second input terminal of each PU.
(1 is an integer from 1 to N, m is an integer from 1 to M), so that the value of m is synchronized with that of bm,
Input each to the first PU, and at t1 after inputting m=M,
A vector and matrix multiplication method is obtained, which is characterized in that the output of the PU, en (n is an integer from l to N), is output in the form of a vector, <EN>.

(従来技術との相違点) 本発明の従来技術に対する改革的な点は、従来技術に於
ては、N次元のベクトルとN*M次元のマトリックスの
乗算には、単位乗算時間のN*M倍の時間が必要であっ
たが、本発明に依れば、その乗算を単位乗算時間のM倍
の時間で行う事ができる、演算装置、及びその演算方法
を実現したことである。
(Differences from the prior art) The innovative point of the present invention over the prior art is that in the prior art, the multiplication of an N-dimensional vector and an N*M-dimensional matrix takes N*M unit multiplication time. However, according to the present invention, an arithmetic device and an arithmetic method thereof have been realized which can perform the multiplication in M times the unit multiplication time.

(実施例) 以下本発明の実施例について図面を参照して詳細に説明
する。
(Example) Examples of the present invention will be described in detail below with reference to the drawings.

第1図は、本発明の構成に使用される演算単位PUの機
能を説明する為の物である。PUの内部では、基本的に
は入力1(a)と入力2(b)の乗算がおこなわれ、そ
の結果Cに入力3(d)が加算され、その結果eが出力
として得られるようになっている。
FIG. 1 is a diagram for explaining the functions of a calculation unit PU used in the configuration of the present invention. Inside the PU, input 1 (a) and input 2 (b) are basically multiplied, and as a result, input 3 (d) is added to C, so that e is obtained as the output. ing.

a*b = c   c + d = eこの様な演算
単位は、集積回路として、現在では市販されている。第
2図は、本発明で、提供される演算装置の構造であり、
N個の演算単位(PU(1)−PU(N))が、全ての
PUで、第1第4の数値が外部より共通の入力として、
与えられ、第2の数値は各々のPUに異なった物として
外部より、入力として与えられ、全てのPUの第5の数
値(出力)が第4の数値となると共に、出力としても、
取り出せる様に接続されていることが特徴である。
a*b = c c + d = e Such arithmetic units are currently commercially available as integrated circuits. FIG. 2 shows the structure of the arithmetic device provided by the present invention,
N calculation units (PU (1) - PU (N)) are all PUs, and the first and fourth numerical values are input as a common input from the outside,
The second numerical value is given to each PU as a different input from the outside, and the fifth numerical value (output) of all PUs becomes the fourth numerical value, and also as an output.
It is characterized by being connected so that it can be taken out.

ベクトル、<BM>とマトリックスCM、Nの積は、<
EN> = CM、N  * < BM > と、あられされるので、第2図に示す様に、第2図の入
力すからベクトル〈BM〉の成分bmを、各PUに共通
にm=1.2,3.・・・Mの順にPUでの単位演算時
間t□、毎に入力し、これに同期して、第2図の入力a
n。
The product of vector <BM> and matrix CM, N is <
EN>=CM, N*<BM>, so as shown in FIG. 2, the component bm of the input vector <BM> in FIG. 2 is set to m=1. 2,3. ...Input in the order of M every unit calculation time t□ in the PU, and in synchronization with this, input a in Fig. 2.
n.

(nは1からN迄の整数)からマトリックスCM、Nの
成分cmn(mは1からM迄の整数nは1からN迄の整
数)を、各々図の上のほうから順次n=1.2,3.・
、N−2,N−1゜Nとなるように入力すれば、bM及
びCM、nが入力されてからt1秒後には、PU(n)
の出力en(nは1からN迄の整数)の組として、ベク
トル<EN>が得られる。
(n is an integer from 1 to N) to the matrix CM, and the component cmn of N (m is an integer from 1 to M, n is an integer from 1 to N) from n=1. 2,3.・
, N-2, N-1°N, t1 seconds after bM, CM, and n are input, PU(n)
A vector <EN> is obtained as a set of outputs en (n is an integer from 1 to N).

この様な観点から見れば、本発明に依って提供される演
算装置は、その構造成分である演算単位PUを、1つだ
け用いて、PUの演算に要する時間、tlごとにマトリ
ックス成分とベクトル成分を順次入力し、N*M番目の
演算の結果を出力するという、従来の方法に比べてM倍
の演算速度が得られることになる。
From this point of view, the arithmetic device provided by the present invention uses only one arithmetic unit PU, which is its structural component, and calculates the matrix components and vectors for each tl in the time required for the PU arithmetic operation. Compared to the conventional method of sequentially inputting components and outputting the result of the N*Mth operation, the calculation speed is M times faster.

以上本発明を一つの実施例について、説明したが、本発
明には、幾つもの、変形が可能であることは自明である
Although the present invention has been described above with reference to one embodiment, it is obvious that the present invention can be modified in many ways.

例えば、PUの数、N、よりも、マトリックスCM、N
並びに〈BM〉のサイズ、が小さい場合には、その時点
で演算をさせる必要のないPUには、入力として0を供
給すれば良い。
For example, the matrix CM, N, is larger than the number of PUs, N.
In addition, if the size of <BM> is small, 0 may be supplied as an input to the PU that does not need to perform calculations at that time.

また、順番に入力されるベクトルの供給の時間差がタイ
ミング、t工、より大きい場合には、入力が到着するタ
イミングに合わせてPUを動作させれば良い。
Furthermore, if the time difference between the supply of vectors that are sequentially input is larger than the timing, the PU may be operated in accordance with the timing at which the inputs arrive.

(発明の効果) この結果、本発明によって、従来の技術では不可能であ
ったマトリックスとベクトルの高速な乗算が可能と成っ
た。
(Effects of the Invention) As a result, the present invention enables high-speed multiplication of matrices and vectors, which was impossible with conventional techniques.

【図面の簡単な説明】[Brief explanation of the drawing]

第1図は、本発明の構成に使用される演算単位PUの機
能を説明する為の物である。PUの内部では、基本的に
は入力1(a)と入力2(b)の乗算がおこなわれ、そ
の結果Cに入力3(d)が加算され、その結果eが出力
となる。 第2図は本発明で提供される演算装置の構造であり、N
個の演算単位(PU(1)−PU(N))が、全てのP
Uで、第1第4の数値が外部より共通の入力として、与
えられ、第2の数値は各々のPUに異なった物として外
部より、入力として与えられ、全てのPUの第5の数値
(出力)が第4の数値となると共に、第1図 出力 入力b
FIG. 1 is a diagram for explaining the functions of a calculation unit PU used in the configuration of the present invention. Inside the PU, input 1 (a) and input 2 (b) are basically multiplied, and as a result, input 3 (d) is added to C, and as a result, e becomes the output. FIG. 2 shows the structure of the arithmetic device provided by the present invention, with N
operation units (PU(1)-PU(N)) are all P
In U, the first and fourth numerical values are given from the outside as a common input, the second numerical value is given to each PU as a different input from the outside, and the fifth numerical value ( output) becomes the fourth numerical value, and the output input b in Figure 1

Claims (1)

【特許請求の範囲】 1)入力として与えられる第一の数値と第二の数値間で
乗算を行う機能と、その結果得られる第三の数値と、入
力として与えられる第四の数値との間で加算を行う機能
と、その結果として得られる第五の数値を、出力とする
ような演算機能を有する演算単位(以下PUと略記する
)を有限の数N個に並べ、全てのPUで、第1の数値が
外部よりの共通の入力として与えられ、第2の数値は各
々のPUに異なった物として外部より、入力として与え
られ、PUの第5の数値(出力)が第4の数値として供
給されると共に、出力としても取り出せる様に接続され
た事を特徴とする演算装置。 2)第一項記載の請求の範囲の、演算装置に、第一の入
力として、ベクトル〈B_M〉の成分b_m(mは、1
からM迄の整数)を、順次PUの単位演算時間、t_1
、置きに入力し、その入力に同期させて、各々のPUの
第二の入力端子に、マトリックスCM、Nの成分c_m
_l(lは、1からN迄の整数であり、mは、1からM
迄の整数)を、mの価がb_mのそれと同期するように
して、各々1番目のPUの入力し、m=Mの入力の後t
_1に、PUの出力、e_n(nは、1からN迄の整数
)を、ベクトル、〈E_N〉の形に、出力する事を特徴
とするベクトルとマトリックスの乗算方法。
[Claims] 1) A function of performing multiplication between a first numerical value given as an input and a second numerical value, and a third numerical value obtained as a result and a fourth numerical value given as an input. Arrange a finite number N of arithmetic units (hereinafter abbreviated as PU) that have the function of performing addition and the function of outputting the fifth numerical value obtained as a result, and in all PUs, The first numerical value is given as a common input from the outside, the second numerical value is given as an input from the outside as a different thing to each PU, and the fifth numerical value (output) of the PU is the fourth numerical value. An arithmetic device characterized in that it is connected so that it can be supplied as an output and can also be taken out as an output. 2) The component b_m of the vector <B_M> (m is 1
to M) in sequence, the unit operation time of PU, t_1
, and in synchronization with the input, the component c_m of the matrix CM,N is input to the second input terminal of each PU.
_l (l is an integer from 1 to N, m is an integer from 1 to M
(an integer up to ) are input to the first PU so that the value of m is synchronized with that of b_m, and after the input of m=M, t
A vector and matrix multiplication method characterized by outputting the output of the PU, e_n (n is an integer from 1 to N), to _1 in the form of a vector, <E_N>.
JP31182486A 1986-12-29 1986-12-29 Arithmetic unit Pending JPS63167971A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP31182486A JPS63167971A (en) 1986-12-29 1986-12-29 Arithmetic unit

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
JP31182486A JPS63167971A (en) 1986-12-29 1986-12-29 Arithmetic unit

Publications (1)

Publication Number Publication Date
JPS63167971A true JPS63167971A (en) 1988-07-12

Family

ID=18021840

Family Applications (1)

Application Number Title Priority Date Filing Date
JP31182486A Pending JPS63167971A (en) 1986-12-29 1986-12-29 Arithmetic unit

Country Status (1)

Country Link
JP (1) JPS63167971A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH02150961A (en) * 1988-12-01 1990-06-11 Matsushita Electric Ind Co Ltd Parallel vector arithmetic unit

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS5793473A (en) * 1980-11-29 1982-06-10 Toshiba Corp Multiplexing convolution product sum calculating device

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPS5793473A (en) * 1980-11-29 1982-06-10 Toshiba Corp Multiplexing convolution product sum calculating device

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH02150961A (en) * 1988-12-01 1990-06-11 Matsushita Electric Ind Co Ltd Parallel vector arithmetic unit

Similar Documents

Publication Publication Date Title
US4692888A (en) Method and apparatus for generating and summing the products of pairs of numbers
JPH0640301B2 (en) Parallel multiplier circuit
JP3323312B2 (en) Accelerated test pattern generator
JPS63167971A (en) Arithmetic unit
Liu et al. A new systolic array algorithm for discrete Fourier transform
JPS63187366A (en) Arithmetic unit for moving average
JPH07141148A (en) Pipeline parallel multiplier
GB2226899A (en) An electronic circuit and signal processing arrangements using it
JPH0535773A (en) Method and system for vector division
JPS6259828B2 (en)
JPS63164640A (en) Cosine transformation device
JPS63167970A (en) Arithmetic unit
SU564638A1 (en) Device for solving linear algebraic equations systems
JPS6043743A (en) Signal processing circuit
JPH04364525A (en) Parallel arithmetic unit
JPS62160529A (en) Data processor
JPH0658671B2 (en) Vector processor
JP2000122844A (en) Semiconductor integrated circuit
JPS63167969A (en) Arithmetic unit
KR19990005455A (en) Multiplication Method and Apparatus Using Pipelined Booth Algorithm
JPH0773161A (en) Arithmetic unit
JPH0447454A (en) Method and device for discrete fourier transformation or cosine transformation of digital data
JPS62166424A (en) Wallace tree circuit
JPS63262760A (en) Fast fourier transforming device
JPH02201686A (en) Semiconductor integrated circuit device