JPH03189868A - Data processor - Google Patents

Data processor

Info

Publication number
JPH03189868A
JPH03189868A JP33008889A JP33008889A JPH03189868A JP H03189868 A JPH03189868 A JP H03189868A JP 33008889 A JP33008889 A JP 33008889A JP 33008889 A JP33008889 A JP 33008889A JP H03189868 A JPH03189868 A JP H03189868A
Authority
JP
Japan
Prior art keywords
data
arithmetic
external memory
buses
address
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
JP33008889A
Other languages
Japanese (ja)
Inventor
Akira Iwata
彰 岩田
Masao Yoshida
征夫 吉田
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.)
MAITETSUKU KK
Original Assignee
MAITETSUKU KK
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 MAITETSUKU KK filed Critical MAITETSUKU KK
Priority to JP33008889A priority Critical patent/JPH03189868A/en
Publication of JPH03189868A publication Critical patent/JPH03189868A/en
Pending legal-status Critical Current

Links

Landscapes

  • Complex Calculations (AREA)

Abstract

PURPOSE:To accelerate data processing by providing two address buses and data buses to be connected to external memories, respectively, and connecting two external memories in parallel. CONSTITUTION:The data processor 1 contains two registers 2, 3 for multiplica tion, a floating point multiplier 4, a data register group 5, and an arithmetic and logic computing element 6, and couples them in pipeline constitution, and performs parallel processing by performing the same arithmetic operation on a large amount of data successively. The data buses 7, 8, the registers 2, 3 for multiplication, and the data register group 5 are connected mutually, and also, the data bus 7 is connected to the external memory 9, and the data bus 8 on one side is connected to a separate external memory 10. Address values to designate the addresses of the external memories 9, 10 are transmitted from an address arithmetic part 13 to the external memories 9, 10, however, the two address buses 11, 12 are provided, and they are connected to the external memories 9, 10 separately. Thereby, the number of steps of the arithmetic processing can be reduced, which accelerates data processing speed.

Description

【発明の詳細な説明】 「産業上の利用分野] この発明は、データ処理プロセッサに関するものであり
、特に、データ処理の効率を向上させたデータ処理プロ
セッサに関するものである。
DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a data processing processor, and particularly to a data processing processor with improved data processing efficiency.

[従来の技術1 従来のデジタル信号を取扱うデータ処理プロセッサは、
一般に乗算器と算術論理演算器と内部メモリを内蔵し、
之等をバイブライン的データフロー構成としているため
、積和演算をバイブライン的並列処理によって高速に行
うことができる。
[Conventional technology 1 A conventional data processing processor that handles digital signals is
Generally, it has a built-in multiplier, arithmetic and logic unit, and internal memory.
Since these have a Vibrine-like data flow configuration, product-sum operations can be performed at high speed by Vibrine-like parallel processing.

又、プログラムメモリとデータメモリとをデータ処理プ
ロセッサから分離し、夫々別個のアドレスバスとデータ
バスを設け、命令の読出しとデータの転送を独立に行え
るという特徴を有する。
Another feature is that the program memory and data memory are separated from the data processor, and a separate address bus and data bus are provided, respectively, so that instruction reading and data transfer can be performed independently.

従って、信号処理や画像処理、或はニューラルネットワ
ーク等の大量のデータに対して積和演算を行う分野では
データ処理プロセッサが広く使用されている。
Therefore, data processing processors are widely used in fields such as signal processing, image processing, neural networks, and the like, where sum-of-products operations are performed on large amounts of data.

[発明が解決しようとする課題] 前述した大11【のデータに対する積和演算を要する分
野等に於ては、取扱うデータが膨大なためデータ処理の
より一層の高速化が要望されている。
[Problems to be Solved by the Invention] In fields such as those requiring product-sum operations on data of the above-mentioned size 11, there is a demand for even faster data processing because the amount of data to be handled is enormous.

そこで、データ転送に要する時間を短縮し、演算時間の
効率を向上して、データ処理を高速化するために解決せ
られるべき技術的課題が生じてくるのであり、本発明は
該課題を解決することを目的とする。
Therefore, a technical problem arises that needs to be solved in order to shorten the time required for data transfer, improve the efficiency of calculation time, and speed up data processing, and the present invention solves this problem. The purpose is to

[課題を解決するための手段] この発明は、上記目的を達成するために提案されたもの
であり、乗算器と算術論理演算器とを内装するプロセッ
サであって、外部メモリから内部レジスタにデータを読
込み、前記データに対し所定の乗算と算術論理演算とを
施し、演算結果を前記外部メモリに書込むデータ処理機
能を有するプロセッサに於て、外部メモリに接続するア
ドレスバス並びにデータバスを夫々2つ設け、2つの外
部メモリを並列に接続したことを特徴とするデータ処理
プロセッサを提供せんとするものである。
[Means for Solving the Problems] The present invention has been proposed to achieve the above object, and is a processor incorporating a multiplier and an arithmetic and logic unit, which transfers data from an external memory to an internal register. In a processor having a data processing function of reading the data, performing predetermined multiplications and arithmetic and logical operations on the data, and writing the operation results to the external memory, an address bus and a data bus connected to the external memory are connected to each other. It is an object of the present invention to provide a data processing processor characterized by having two external memories connected in parallel.

[作用コ この発明は外部メモリを2つの群に分割し、夫々)外部
メモリに対応するアドレスバス並びニデータパスを別個
に設けているので、外部メモリと内部レジスタとのL7
g、或は内部レジスタ間で2種類のデータを同時に転送
できる。
[Function] In this invention, the external memory is divided into two groups, and address buses and two data paths corresponding to each external memory are separately provided.
g, or two types of data can be transferred simultaneously between internal registers.

従って、乗算器或は演算器で必要とされる2つのデータ
は同時に外部メモリから読出される。又、演算に必要な
データを一方のデータバスを用いて読出すと同時に、並
列処理によって演算された他の演算結果を他方のデータ
バスによって外部メモリに書込むこともできる。
Therefore, two pieces of data required by the multiplier or arithmetic unit are read from the external memory at the same time. Further, while data necessary for an operation is read using one data bus, other operation results calculated by parallel processing can be written to an external memory using the other data bus.

[実施例] 以下、本発明の一実施例を別紙添付図面に従って説明す
る。第1図はデータ処理プロセッサ(+)の構成を示す
ブロック図である。前記データ処理プロセッサ(1)は
2つの乗算用レジスタ(2)(3)、浮動小数点乗算器
(4)、データレジスタ群(5)及び算術論理演算器(
6)を内装し、同図に示すように之等をバイブライン構
成として結合しているため、同一の演算を多数のデータ
に連続して行い並列処理ができる。そして、データバス
(η(8)と乗算用レジスタ(2)(3)とデータレジ
スタ群(5)とを接続するとともに、データバス(7)
は外部メモリ(9)に接続し、一方のデータバス(8)
は別個の外部メモ1バ10)に接続されている。
[Example] An example of the present invention will be described below with reference to the accompanying drawings. FIG. 1 is a block diagram showing the configuration of a data processor (+). The data processing processor (1) includes two multiplication registers (2) (3), a floating point multiplier (4), a data register group (5), and an arithmetic and logic unit (
6) and are connected in a vibe line configuration as shown in the same figure, so that the same operation can be performed on a large number of data in succession and processed in parallel. Then, the data bus (η(8)), the multiplication registers (2), (3), and the data register group (5) are connected, and the data bus (7)
is connected to external memory (9), and one data bus (8)
is connected to a separate external memory card 10).

そして、外部メモリ(9)(In)の番地を指定するた
めのアドレス値はアドレス演算部(+3から外部メモリ
(9)(10)へ伝送するが、アドレスバス(I 1)
(Uりを2つ設ケて外部メモリ(Q)Qo)に夫々別個
に接続している。又、データ処理プロセッサ(1)のプ
ログラムは、プログラムメモリ(ユに格納され、各命令
は制御回路部(1!9によって読出されて解釈され、制
御信号が生成される。
Then, the address value for specifying the address of the external memory (9) (In) is transmitted from the address calculation unit (+3) to the external memory (9) (10), but the address bus (I1)
(Two U-rings are provided and each is connected to the external memory (Q) and Qo) separately. Further, the program of the data processing processor (1) is stored in the program memory (1), and each instruction is read and interpreted by the control circuit unit (1!9) to generate a control signal.

上記パイプライン構成によってこのデータ処理プロセッ
サ(1)は、 ■ 外部メモ1バQ)(lr9から乗算用レジスタ(2
)(3)へのデータ読込み ■ 乗算用レジスタ(2)(3)に格納されたデータの
乗算、及び乗算結果のデータレジスタ群(5)への格納 ■ デルタレジスタ群(5)に格納されているデータに
対する算術論理演算器(6)による演算、及び演算結果
のデータレジスタ群(5)への格納以上の処理を同時に
実行できる。また、2つの外部メモ1バ9)(u’6を
別個のデータバス(7)(8)にて接続しているので、
上記■の代わりに一方の外部メモリからのデータ読出し
と他方の外部メモリへのデータ書込みを上記■■と同時
に実行できる。
With the above pipeline configuration, this data processing processor (1) is configured to: ■ External memory 1 (Q) (lr9 to multiplication register (2)
) (3) ■ Multiplying the data stored in the multiplication registers (2) and (3) and storing the multiplication results in the data register group (5)■ Storing the data in the delta register group (5) The arithmetic and logic unit (6) performs an operation on the existing data, and processes beyond storing the operation result in the data register group (5) can be executed simultaneously. Also, since the two external memory buses 1 and 9) (u'6) are connected by separate data buses (7) and (8),
Instead of (2) above, reading data from one external memory and writing data to the other external memory can be executed simultaneously with (2) above.

例えば、外部メモ1バ9)に数値列(alt82+・・
・・・・ an )、外部メモリ00)に数値列(b、
、b2、・・・・・・、bn)が順に格納されていると
して、両者の数値の積和演算 X=Σ m+’ll+ i=1 を実行する場合、2つのデータバス(7)(8)によっ
て乗算用レジスタ(2)(3)に前記数値a1st)i
 を同時に読出すことができる。そして、このデータ読
出しと、乗算並びに累積加算を並列に実行すると別表1
に示すように、n+3回のステップ数で前記積和演算を
完了する。
For example, a numerical string (alt82+...
...an), numerical string (b,
, b2, ..., bn) are stored in order, and when performing the product-sum operation X=Σ m+'ll+ i=1 of both numbers, two data buses (7) (8 ) to the multiplication registers (2) and (3).
can be read simultaneously. Then, if this data reading, multiplication, and cumulative addition are executed in parallel, Table 1
As shown in the figure, the product-sum operation is completed in n+3 steps.

データバスが1つのものの場合には数値の読出しをa、
とす、とて2ステツプ費やすのでbn+3回のステップ
数を必要とする。
If there is one data bus, read the numerical value by a,
Since it takes two steps, the number of steps is bn+3.

又、外部メモリ(9)に数値列(aIsa2+・・・・
・・a。)、外部メモリ00)に数値列(bl、b2.
・・・・・・ b、)が順に格納されている場合に、両
者の数値の夫々の積を求め、更に定数を加算し外部メモ
リ0〔0に格納する演算 Ci =3.  @ bl +d (+ = 19 2
1 ””n )を実行するときは、別表2に示すように
2つの数値a、、b、の読出しと、乗算及び加算をノで
イブライン的に並列実行すると、演算処理にn+1回の
ステップ数を要し、更にデータ格納にn回のステップ数
を要するので、合計20+1回のステップ数で演算を終
了する。
Also, a numerical string (aIsa2+...) is stored in the external memory (9).
・・a. ), numerical strings (bl, b2 .
. . . b, ) are stored in order, calculate the product of both numerical values, add a constant, and store the result in external memory 0 [0 [0]. Calculation Ci = 3. @ bl +d (+ = 19 2
1 ""n), as shown in Attached Table 2, if the reading of the two numerical values a, , b, and the multiplication and addition are performed in parallel in eveline manner, the number of steps in the arithmetic processing is n+1. Since n steps are required to store the data, the calculation is completed after a total of 20+1 steps.

データバスが1つのものに於ては、数値a1とす、とを
1ステツプで読出せず、2ステツプを要するのでbn+
1回のステップ数を必要とする。
In a system with one data bus, the numerical values a1 and 2 cannot be read in one step, but require two steps, so bn+
Requires one step.

そして、外部メモ1バ9)に数値列(al+  a2+
・・・・・・、a。)と(CI r  c、、l・・・
・・・Co )が格納され、他方の外部メモリ00)に
数値列(bl 、b2゜・・・・・・、bn )が順に
格納されていると仮定して、al とblの積を求め、
更にその結果に01を加算して順次外部メモ1バ10)
に格納する演算d1−ai  ” bl + CL  
(1”” L 2+ ”’−”n)を実行すると別表3
に示すように2つの数値ab、の読出しと乗算、加算を
行い、更に数値Cの読出しとデータ格納を2つのデータ
バス(7)(8)によって同時に実行するので、bn+
2回のステップ数で演算を完了する。
Then, write the numerical string (al+ a2+
・・・・・・、a. ) and (CI r c,, l...
...Co) is stored and the numerical string (bl, b2゜..., bn) is stored in order in the other external memory 00), calculate the product of al and bl. ,
Furthermore, add 01 to the result and sequentially write the external memory 1 bar 10)
The operation to be stored in d1-ai ” bl + CL
When you execute (1”” L 2+ “’-”n), Attached Table 3
As shown in , reading, multiplication, and addition of two numbers ab and 2 and reading and storing data of number C are performed simultaneously using two data buses (7) and (8), so bn+
The calculation is completed in two steps.

データバスが1つのものの場合は、数値ab  c の
読出しに3ステツプを要し、更にデータ格納に1ステツ
プを要するので演算完了にbn+2回のステップ数を必
要とする。このように、データバスを2つ設けたことに
よって1,5倍乃至2倍のデータ処理速度の高速化が達
成された。
If there is one data bus, three steps are required to read out the numerical value ab c and one step is required to store the data, so bn+2 steps are required to complete the calculation. In this way, by providing two data buses, the data processing speed has been increased by 1.5 to 2 times.

[発明の効果J この発明は、上記一実施例に詳述したように、外部メモ
リと内部レジスタとの間、或は内部レジスタ間で2つの
データを同時に転送できるので演算処理のステップ数が
著しく減少し、従来のものに比して1.5倍乃至2倍の
演算速度となる。
[Effect of the Invention J] As described in detail in the above-mentioned embodiment, this invention allows two pieces of data to be transferred simultaneously between an external memory and an internal register or between internal registers, so the number of steps in arithmetic processing is significantly reduced. The calculation speed is 1.5 to 2 times that of the conventional method.

従って、データ処理速度が飛躍的に高速化し、信号処理
、画像処理等に於て頻繁に行われる大hlのデータによ
る積和演算の処理能率の向上に寄与できる。
Therefore, the data processing speed is dramatically increased, and it is possible to contribute to improving the processing efficiency of sum-of-products operations using large HL data, which are frequently performed in signal processing, image processing, etc.

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

第1図は本発明のデータ処理プロセッサの一実施例を示
すブロック図である。 (1)・・・・・・データ処理プロセッサ(2)(3)
・・・・・・乗算用レジスタ(4)・・・・・・t7動
小数点乗算器(5)・・・・・・データレジスタ群 (6)・・・・・・算術論理演算器 (7)(8)・・・・・・データバス
FIG. 1 is a block diagram showing an embodiment of the data processing processor of the present invention. (1) Data processing processor (2) (3)
...Multiplication register (4) ...t7 Dynamic point multiplier (5) ...Data register group (6) ... Arithmetic logic unit (7 )(8)・・・Data bus

Claims (1)

【特許請求の範囲】[Claims] 乗算器と算術論理演算器とを内装するプロセッサであつ
て、外部メモリから内部レジスタにデータを読込み、前
記データに対し所定の乗算と算術論理演算とを施し、演
算結果を前記外部メモリに書込むデータ処理機能を有す
るプロセッサに於て、外部メモリに接続するアドレスバ
ス並びにデータバスを夫々2つ設け、2つの外部メモリ
を並列に接続したことを特徴とするデータ処理プロセッ
サ。
A processor incorporating a multiplier and an arithmetic and logic unit, which reads data from an external memory into an internal register, performs predetermined multiplication and arithmetic and logic operations on the data, and writes the result of the operation to the external memory. 1. A data processing processor having a data processing function, characterized in that two address buses and two data buses are provided to connect to external memories, and two external memories are connected in parallel.
JP33008889A 1989-12-20 1989-12-20 Data processor Pending JPH03189868A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP33008889A JPH03189868A (en) 1989-12-20 1989-12-20 Data processor

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
JP33008889A JPH03189868A (en) 1989-12-20 1989-12-20 Data processor

Publications (1)

Publication Number Publication Date
JPH03189868A true JPH03189868A (en) 1991-08-19

Family

ID=18228652

Family Applications (1)

Application Number Title Priority Date Filing Date
JP33008889A Pending JPH03189868A (en) 1989-12-20 1989-12-20 Data processor

Country Status (1)

Country Link
JP (1) JPH03189868A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6240492B1 (en) 1998-05-22 2001-05-29 International Business Machines Corporation Memory interface for functional unit of integrated system allowing access to dedicated memory and shared memory, and speculative generation of lookahead fetch requests
CN111213125A (en) * 2017-09-08 2020-05-29 甲骨文国际公司 Efficient direct convolution using SIMD instructions

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6240492B1 (en) 1998-05-22 2001-05-29 International Business Machines Corporation Memory interface for functional unit of integrated system allowing access to dedicated memory and shared memory, and speculative generation of lookahead fetch requests
CN111213125A (en) * 2017-09-08 2020-05-29 甲骨文国际公司 Efficient direct convolution using SIMD instructions
JP2020533691A (en) * 2017-09-08 2020-11-19 オラクル・インターナショナル・コーポレイション Efficient direct convolution using SIMD instructions
CN111213125B (en) * 2017-09-08 2023-11-07 甲骨文国际公司 Efficient direct convolution using SIMD instructions

Similar Documents

Publication Publication Date Title
US5473554A (en) CMOS multiplexor
KR20010031192A (en) Data processing system for logically adjacent data samples such as image data in a machine vision system
JPH10232779A (en) Method and device for parallel instruction processing
JP2620511B2 (en) Data processor
JPS63240664A (en) Fast processing computer
US4621324A (en) Processor for carrying out vector operation wherein the same vector element is used repeatedly in succession
JPS63167967A (en) Digital signal processing integrated circuit
US4631672A (en) Arithmetic control apparatus for a pipeline processing system
KR960024997A (en) Digital array processor for multi-instruction multi-data type neural network and system configured using same
JP2617974B2 (en) Data processing device
JP2993975B2 (en) Central processing unit
JP2690406B2 (en) Processor and data processing system
JPH03189868A (en) Data processor
JP3277399B2 (en) General-purpose processor for image processing
JPH01119861A (en) Digital signal processing lsi
JP2001256073A (en) Digital signal processor
JPS63136710A (en) Digital signal processing circuit
JPH0234058B2 (en)
JPH0926945A (en) Information processing device
KR920008212B1 (en) Two-dimensional fast fourier transform (FFT) processor using a mixed shuffle connection method (MSCM).
JPH05165875A (en) Vector arithmetic processor
JPS63197217A (en) data processing equipment
JPH01237766A (en) Data processing device and image data processing system using the same
JPH0344779A (en) Pattern recognizing processor
JPH04255064A (en) Parallel processor