JPH02211589A - Data driving type control method - Google Patents

Abstract

PURPOSE:To curtail the overhead for a queue control and to contrive the improvement of the performance of a device by executing an instruction without allowing a queue to intervene, in the case when an instruction executing formation is a sequential processing type. CONSTITUTION:In the case of a sequential processing, a head address of an instruction of this program is extracted from an instruction address store area 21, its address is set as an instruction address, an arithmetic designation 11 is extracted from an instruction template instructed by the instruction address, a processing of an arithmetic part 5 corresponding to the designation 11 is determined, and an instruction execution control part 2 transfers the instruction address of an arithmetic object to this arithmetic processing. The arithmetic part 5 which receives a control extracts prescribed input data from an area indicated by the number of input data acquisition destinations 12 and an input data acquisition destination 13 in the instruction template, executes a prescribed operation, stores a result of operation in an output data store area 17, and returns the control to the control part 2. Subsequently, when the next instruction exists, the control part 2 extracts an address of the next instruction from an output data notice destination 15, and sets its address as a new instruction address and executes an instruction processing.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a data-driven control method, and in particular, to a method for omitting instruction queuing control when a sequential processing program is executed on a single Neumann-type machine. The present invention relates to a data-driven instruction control method that can improve execution performance.

The present invention also relates to a compiling device, in particular.

Analyzes the instruction execution order from the source code representation of the data-driven program or the diagrammatically represented data-driven program diagram, and identifies whether the instruction execution control method is sequential processing type or parallel processing type. It relates to a compile device that can generate .

[Conventional technology]

The features of data-driven programs are (1) the order of instruction execution is determined only by data dependencies, and the program itself is represented diagrammatically, which can be expected to improve understandability; (2) parallel processing is made explicit. It can be expressed, etc.

Conventionally, when a data-driven program is executed by a single Neumann-type processor, it may be possible to execute multiple instructions at the same time depending on the output result of the instruction.
As described in Japanese Patent No. 1-23239, it is necessary to wait for the execution of instructions using a FIFO queue.

According to this method, data-driven programs written in parallel or sequentially can be executed using the same control mechanism, but even in data-driven programs written sequentially that do not require waiting for instruction execution, instructions can be stored in the FIFO queue. The disadvantage is that commands cannot be executed unless they are registered.

[Problem to be solved by the invention]

As mentioned above, the conventional technology has the advantage of being able to execute a data-driven program written sequentially and a data-driven program written in parallel on the same data-driven controller, but it is difficult to wait for instruction execution. Even when executing instructions of data-driven programs written sequentially that are not necessary,
It is necessary to register the data in the FIFO queue once, which poses a problem of reducing processing performance. Therefore, if it is possible to identify whether a data-driven program is a sequential processing type or a parallel processing type, in the case of a sequential processing type, the overhead required for the queue control can be reduced.

Furthermore, in a data-driven program expressed in source code or diagrammatic form, if it were possible to remove unnecessary tokens and convert from a parallel processing type to a sequential processing type by analyzing the tokens exchanged between instructions. Improvements in processing capacity can be expected.

SUMMARY OF THE INVENTION An object of the present invention is to provide a data-driven control method that can minimize a decrease in processing performance due to overhead in controlling the execution of instructions of a data-driven program.

It is also an object of the present invention to remove redundant tokens from a data-driven program in source code or diagrammatic representation, so that the data-driven program is a sequential processing type that does not require waiting for instruction execution. An object of the present invention is to provide a compiling device for a data-driven program that can automatically generate identification information indicating whether the program is a parallel description that requires waiting for execution of instructions, or a parallel description that requires waiting for instruction execution.

[Means to solve the problem]

In order to achieve the above objects, 1 the data-driven control method of the present invention provides an operation name for instructing an operation corresponding to each instruction, an input data acquisition source, the number of input data acquisition sources, and an operation result using the input data. An instruction group consisting of an output data notification destination, data indicating the number of output data notification destinations, an input data arrival counter indicating that the input data has been prepared, and an output data storage area for storing the output data. a program instruction section, a first-in, first-out queue means for waiting for the execution of instructions, and an arithmetic processing section for performing operations;
In a data-driven control method having a storage means storing an instruction execution control unit that controls instructions based on data drive, the instruction execution control method is sequentially executed for each execution unit of a data-driven program or for each data-driven calling program. A means for identifying whether the program is a processing type or a parallel processing type is provided in the program instruction section, and the instruction execution control section determines the identification information when executing an instruction. Immediately execute an instruction activated by a result without registering it in the queue, and in the case of a parallel processing type, register an instruction activated by a result of instruction execution in the queue to wait for execution of the instruction. There are characteristics.

In addition, in order to achieve the above purpose, a compilation device that translates a data-driven program reads the source code or the diagrammatically expressed data-driven program diagram and analyzes the data flow exchanged between instructions in the program. , create intermediate data containing instruction connection information,
After analyzing the instruction execution order from the connection information included in this intermediate data and removing extra tokens, it is determined whether the execution control method of the data-driven program instructions is sequential processing type or parallel processing type. The feature is that it generates identification display information.

[Effect]

As described above, in the instruction execution control of a data-driven program, when executing instructions in a sequential processing type using information that identifies the execution fg control method of the data-driven program,
Since instructions can be executed directly without going through a queue for waiting for instruction execution, performance can be improved. Furthermore, in a data-driven program expressed as a source code or diagrammatically, the execution order of instructions is analyzed by a compiling device.

The redundant tokens are removed, so that the instruction execution control method of the data-driven program is sequential;
Alternatively, it is determined whether the program is a parallel processing type, and the control method for data-driven program instructions can be automatically optimized.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of an embodiment of a control device equipped with the data-driven control method of the present invention. In FIG. 1, numeral 3 is a program instruction section that is an instruction of a data-driven program or a data-driven calling program, 4 is a waiting section that waits for instructions using a FIFO queue, and 5 is an instruction in an arithmetic section within the instruction. 2 is an instruction execution control unit that controls the execution of a data-driven program or a data-driven calling program instruction based on data drive; The configured devices 3, 4, 5 are stored in storage means and controlled by a conventional Neumann type processor.

FIG. 3 shows a configuration example of identification display information for identifying the execution control format of instructions in the program instruction section 2 and an instruction template. Identification display information 2-0 is prepared for each data-driven program or data-driven calling program, and for example, if the instruction execution control method is sequential processing type, it is set to 0'.
, if the program is of the parallel processing type, the instruction address storage area 21 has a value of 1'.If the data-driven program or the calling program is of the sequential processing type, the instruction address storage area 21 stores the start address of the instruction of the program. The operation name 11 in the instruction template indicates the operation specified by the instruction, the input data source 13 indicates the input data source address, the number of input data sources 12 indicates the number of input data sources, and the output data Notification destination 15 indicates an instruction address for notifying the next instruction that output data has been prepared according to the calculation result of the instruction, and output data notification destination number 14 is an input data arrival counter indicating the number of registered output data notification destinations. Reference numeral 16 indicates the number of arrivals of input data necessary for executing an instruction. Output data storage area 17 is an area for storing calculation results.

Here, the number of input data acquisition sources 12 and the number of output data notification destinations 14 are both arbitrary integers greater than or equal to '0', for example, 0.
' indicates that there is no input data acquisition source 13 or no output data notification destination 15.

If it is an integer greater than or equal to 1, it indicates that an area for input data acquisition destination 13 or output data notification destination 15 exists.

Operation name in the instruction template 11, number of input data sources 12, input data sources 13. Number of output data notification destinations: 1
4, Output data notification destination 15, Input data arrival counter 1
6. The output data storage area 17 is prepared for each instruction in the data-driven program or the calling program, and is stored in the storage means. Furthermore, since these instruction groups are executed based on data drive, operations are executed only when all input data necessary for execution of the instruction is collected.

FIG. 4 is an example of an operation flow diagram of the instruction execution control section 2 in FIG. 1. FIG. 4 shows the functional blocks of FIG. 1, the identification display information 20 of FIG. 3, the instruction address storage area 21. The positioning and operation of command fields 11 to 17 will be explained. Here, instruction execution control for a data-driven program will be explained, but since the instruction execution control for a data-driven calling program is the same as that for a data-driven program, the explanation will be omitted.

The definition of the sequential processing type here is: (1) the number of input data (tokens) required to execute the instruction is ``l'', (2) the number of output data notification destinations is 1'', and (3) the end of the program instruction. From (1), the number of output data notification destinations is 0', (
The parallel processing type refers to all data-driven programs other than the sequential processing type, and the program ends when the queue in the waiting section 4 becomes empty. In other words, it is determined that the command to be executed is not registered.

Prior to program execution, the execution format of the data-driven program to be executed in the program instruction section 3, for example, 10' for a sequential processing type, t for a parallel processing type.
The values of l and I are stored in the identification display information 20. and,
When the execution format of the data-driven program is a sequential processing type, the start address of the instruction is stored in the instruction address storage area 21, and when the execution format is a parallel processing type, the start address of the instruction is stored in the queue of the waiting section 4. It is registered at the beginning, and nothing is stored in the instruction address storage area 21.

The program execution is as follows. Instruction execution control unit 2
First, the identification display information 20 indicating the execution format of the program is extracted, and a determination process is performed as to whether the execution format is a sequential processing type or a parallel processing type (step 30).As a result, the identification display information 20 If the value is 0', that is, if it is a sequential processing type, extract the start address of the instruction of the program from the instruction address storage area 21 (step 38).
, take that address as an instruction address, extract the operation name 11 from the instruction template specified by the instruction address,
Determine the calculation process of the calculation unit 5 corresponding to the calculation name 11,
The instruction execution control unit 2 takes over the instruction address of the calculation target to the calculation process and passes control to the calculation process, and the calculation process of the controlled calculation unit 5 executes the instruction template specified by the inherited instruction address. Predetermined input data is extracted from the area specified by the input data acquisition source number 12 and input data acquisition source 13, a predetermined operation is performed on the extracted input data, and the operation result is stored in the output data storage area 17. Zero or more processes for storing and returning control to the instruction execution control section 2 are performed in step 39.

The command execution control unit 2 that has received the control determines the number of output data notification destinations (14) in the command template specified by the command address.
is 1', that is, if the primary instruction exists, the address of the next instruction is extracted from the output data notification destination 15, and that address is set as a new instruction address.Step 40), Step 3
Return to 9.

As described above, steps 39 to 40 are repeated as long as the number of output data notification destinations 14 is 1', and if the number of output data notification destinations 14 is JOl, it means that there is no next instruction address to be executed. The instruction execution processing of the program ends.

In the case of the parallel processing type, it is determined whether or not there is an address registered in the queue of the waiting section 4 (step 31), and if the address exists, the address is taken out from the queue as an instruction address, and the instruction address is used as the instruction address. Extract operation name 1 from the instructed instruction template,
Determine the calculation process of the calculation unit 5 corresponding to the calculation name 11,
The instruction execution control unit 2 takes over the instruction address of the calculation target to the calculation process, and passes control to the calculation process. Then, the arithmetic processing of the controlled arithmetic unit 5 extracts predetermined input data from the area specified by the number of input data sources 12 and the input data source 13 in the instruction template specified by the inherited instruction address. Then, predetermined calculations are performed on the extracted input data.

The operation result is stored in the output data storage area 17 and the control is returned to the instruction execution control unit 2. Zero or more processes are performed in step 3.
It is done in 2.

Then, the command execution control unit 2 that has received the control determines the number of output data notification destinations 14 in the instruction template specified by the instruction address, and if it is 1, for example, t O', it is determined that there is no notification destination, and in step 31 If the number is positive, it is determined that the output data notification destination 15 exists (step 33) 6
Next, the address of the next instruction is set to 1 from the output data notification destination 15.
By extracting the input data and updating the input data arrival counter 16 in the instruction template specified by the address of the next instruction, the token is transmitted (step 34), and all the input data necessary for executing the next instruction is collected. (step 35). If the input data is complete,
The address of the next command is registered in the queue of the waiting section 4 (step 36), and if the input data is not complete, the process moves to step 37 without doing anything. The process from step 34 to step 37 is repeated as many times as the number of notification destinations specified by the predetermined number of output data notification destinations 14, and when notification of tokens to all output data notification destinations is completed, step 31
Return to

Then, the processing from step 32 to step 37 is repeated until there are no addresses registered in the queue of the waiting section 4. If the queue of the waiting section 4 is empty, that is, no instruction address to be executed is registered, the program The instruction execution process ends.

As described above, in the case of the sequential processing type, the registration process in the queue of the waiting section can be skipped.

Further, an embodiment of the present invention will be described with reference to other drawings. FIG. 2 is a block diagram of a compiling device showing one embodiment of the present invention. In FIG. 2, a compiling device 50
The input is either a data-driven program represented graphically or a wiring source code that is completely equivalent to the data-driven program represented graphically. 51 is a source code conversion unit that converts a diagrammatically described data-driven program into source code; 52 is a unit that analyzes the syntax of the source code and analyzes the flow of data exchanged between instructions; A syntax analysis unit 53 generates an intermediate data file 55 containing input/output data information and control flow information between instructions, based on the data-driven principle from the control flow information of the intermediate data file 55 generated by the syntax analysis unit 52. An instruction execution analysis unit 54 is included in the optimum connection information file 56, which analyzes the order of instruction execution by deleting redundant tokens, determines an instruction execution control method based on the result, and generates the optimum connection information file 56. This is a program instruction code generation unit that converts optimized connection information and identification display information of an instruction execution control method into an executable instruction template as shown in FIG.

FIG. 5 is an example of a diagrammatically represented data-driven program diagram that is input to the compiling device 50.

Instructions 60 for performing basic operations are represented by rectangular nodes,
Enter the operation name in it. Further, instructions 61 whose results are selectively output based on calculation results are represented by trapezoidal nodes, and data exchanged between instructions is represented by directed arcs 62. In the figure, the directed arc that enters from above the instruction 60 or the instruction 61 represents the input, and the directed arc that exits from the bottom represents the output. Note that data 6 that is not exchanged between instructions
3. For example, alphanumeric characters and structures are written as direct command input. Furthermore, there are control tokens 64 that do not serve as input data but are used to define the order of execution of instructions, and are entered from the left and right sides of the instructions.

In addition, in order to facilitate code conversion to the connection table, each instruction is given a unique number 65 to identify the instruction, an input pin number 66, and an output pin number 67. A plurality of input pin numbers 66 and output pin numbers 67 can be assigned.

FIG. 6 shows an example of a connection table source code that is input to the syntax analysis unit 51, and is completely equivalent to the data-driven program diagram shown in FIG. That is, the connection table includes a unique number description section 70. which describes the unique number assigned to each instruction. It consists of an instruction name description part 71 that describes the command name, an input data description part 72 that describes the input data, and a control token description part 73 that describes the control token. Also, the end display of the command is;
', and the input data description section 72 corresponds to pins 0, 1, . . . in order from the left, and each input pin is separated by a blank space. Each input data is written in a format that refers to the unique number of the previous instruction that is input and the output pin.For example, '1-3' means that the unique number of the instruction is 1 and the output pin number is 3. Control tokens are identified by 'sign' and are written in a format that refers to the instruction's unique number and output pin in the same way as input data, and if there is more than one, ' , ' delimited.

Furthermore, input data that is not referenced between instructions is written directly to the corresponding input pin.

Figure 7 shows the data-driven program in Figure 5 as input.
This is a control flow rotation stored in the intermediate data file 55 generated by the syntax analysis unit 52.

In this diagram, only the data and control tokens exchanged between instructions are described, and alphanumeric characters, structures, etc. that are directly input to the instructions are excluded.

FIG. 8 shows an example of a processing procedure when the instruction execution analysis unit 53 reduces redundant tokens.

The processing procedure is performed as follows.

(1) Mark all branches on the control flow diagram as 'undeleteable' (step 80).

(2) Mark all nodes on the control flow diagram as "unfinished". However, the entry node is marked with "processing completed" (step 81).

(3) Select one node marked as "unfinished" and a node having a branch from among the processing-completed nodes and set it as the processing target (step 83).

(4) Calculate the length of the bus starting from the input branch of the processing target node, passing through the non-deletable branch, and ending at the input node. This bus length is determined for all branches of the processing target node (step 84), and an arbitrary reference wave is selected from among the branches that cannot be deleted, and the bus length is compared with other branches that cannot be deleted. are compared, and if the path length of the reference wave is always shorter than other non-removable path lengths, it is marked as "deletable". The reference wave is selected for all input branches. and,
When all bus length comparisons are completed for the path lengths of all input branches, a "processing completed" mark is placed on the processing target node. However, if there is only one input branch, the processing target node is marked as "processing completed" immediately.

Furthermore, if there is a conditional branch node on each bus to be compared, it is determined that deletion is impossible (step 8
5).

(5) When processing is completed for all nodes (step 82), branches marked as ``Deletable'' are deleted from the control flow diagram.

(6) Determine the instruction execution format of the control flow diagram according to the definition of the sequential processing type described above, and generate the result as identification display information.

FIG. 9 shows the control flow of the data-driven program diagram shown in FIG. 5 with redundant tokens removed.

The processing flow of the compiling device 50 can be summarized as follows using FIGS. 2 to 3 and 5 to 9.

The data-driven program diagram shown in FIG. 5 is read by the wiring source code converter 51 and converted into the wiring diagram shown in FIG. 6. The connection code becomes an input to the syntax analysis unit 52, and after parsing the syntax, the control flow diagram shown in FIG. 7 is output to the intermediate data file 55. According to the processing procedure, redundant tokens are deleted from the control flow diagram, and the token-optimized control flow diagram shown in FIG. 9 is created in the optimal connection information file 56.
At the same time, the instruction execution format is determined and the result is set in the identification display information 20. Then, the program instruction code generation unit 54 reads the control flow diagram and input/output data information from the optimal connection information file 56 to be processed,
The executable instruction template and m sub-display information 20 shown in FIG.

An instruction address storage area 21 is generated.

〔Effect of the invention〕

According to the present invention, in the instruction execution control of a data-driven program, when the instruction execution format is a sequential processing type, the instructions can be executed without going through a waiting queue, so the overhead for queue control can be reduced and the performance is improved. can be improved.

In addition, when compiling a data-driven program expressed graphically or in code, redundant tokens can be removed and the instruction execution format can be automatically determined.
Optimization work for instruction execution format can be done efficiently.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing an example of the block configuration of a control device equipped with the data-driven control method of the present invention, and FIG. 2 is a diagram showing an example of the block configuration of a compiling device equipped with the compiling method of the present invention. 3 is a diagram showing an example of the configuration of an instruction template stored in the program instruction section in FIG. 1, FIG. 4 is a diagram showing an example of the operation flow of the instruction execution control section in FIG. 1, and FIG. The figure shows an example of a data-driven program diagram that serves as an input to the embodiment of the present invention, and FIG. 6 shows an example of a wiring source code equivalent to the data-driven program diagram that serves as an input to the embodiment of the present invention. , FIG. 7 is a diagram showing an example of a control flow diagram obtained by converting the data-driven program diagram of FIG. A diagram showing
FIG. 9 is a diagram showing an example of a data-driven program diagram optimized by the instruction execution analysis section of FIG. 2. DESCRIPTION OF SYMBOLS 1... Control device having data-driven control, 2... Instruction execution control unit, 3... Program instruction unit, 4... Queue unit, 5... Arithmetic unit, 20... Identification display information ,5
o... Compilation device, 51... Connection source code conversion unit. 52...Syntax analysis unit, 53...Instruction execution analysis unit, 5
4...Program instruction code generation section. Figure 1 Figure 3 Figure 4 Figure 7 Figure 6 Figure 8

Claims (1)

[Scope of Claims] 1. An input data acquisition source, the number of input data acquisition sources, an output data notification destination that notifies the next instruction of the operation result using the input data, data indicating the number of output data notification destinations, and the input data. A program instruction section consisting of an input data arrival counter that indicates that the program is ready and a storage area for output data, an arithmetic section that performs the operations specified in the instruction, and a first-in, first-out queue for waiting for execution of instructions. and an instruction execution control section that controls the execution of the instructions in a data-driven manner, in an information processing device that stores in a storage means whether the instruction execution control method is sequential processing type or parallel processing type for each execution unit of a data-driven program. A program instruction section is provided with an identification display means to indicate whether the program is a processing type, and the instruction execution control section determines the identification display information when executing an instruction. It is characterized in that the instruction is executed immediately without waiting for execution in a queue, and in the case of a parallel processing type, the next instruction activated by the result of instruction execution is registered in the queue, and the execution of the instruction is waited for. data-driven control method. 2. In claim 1, the program instruction section includes means for identifying whether the instruction execution control method is a sequential processing type or a parallel processing type for each calling program in which instructions are executed by data drive. The instruction execution control unit determines the identification display information when executing an instruction, and in the case of a sequential processing type, immediately executes the next instruction activated by the result of instruction execution without waiting for execution of the instruction in the queue. A data-driven control method, characterized in that, in the case of a parallel processing type, the next instruction activated according to the result of the instruction execution is registered in the queue, and the execution of the instruction is waited for. 3. In a compilation method for translating a data-driven program, the source program is read, the flow of data exchanged between instructions in the source program is analyzed, and intermediate information including instruction input/output data information and control flow information is analyzed. A syntax analysis unit that generates data and an instruction execution order are analyzed from the control flow information of instructions included in the intermediate data generated by the syntax analysis unit, and the instruction execution control method of the data-driven program is a sequential processing type. An instruction execution analysis unit that generates identification information indicating whether the source program is a parallel processing type or a parallel processing type, and converts intermediate data of the source program and identification information that instructs the execution control method of the corresponding process into machine instructions. A compiling device comprising: a program instruction code generation unit. 4. A connection table source that extracts connection information from the input program diagram in a diagrammatic representation and converts the connection information into code of a source program in a connection table format in a compiling device that translates a data-driven program represented in a diagram. a code conversion unit; a syntax analysis unit that analyzes the flow of data exchanged between instructions in a source program and generates intermediate data including instruction connection information and control flow information; and a syntax analysis unit that generates intermediate data including instruction connection information and control flow information; The instruction execution order is analyzed from the instruction control flow information included in the intermediate data, and identifying information indicating whether the instruction execution control method of the data-driven program is a sequential processing type or a parallel processing type is obtained. A compiling device comprising: an instruction execution analysis section that generates an instruction execution analysis section; and a program instruction code generation section that converts intermediate data of the source program and processing identification information corresponding thereto into machine instructions.