JPH03282641A - Measuring method of executing time of data driving processor - Google Patents

Abstract

PURPOSE:To easily measure the executing time of the processing carried out based on a program of a data driving device by applying a dummy packet having a processor No as the processor information before application of a data packet to be processed. CONSTITUTION:A dummy packet is applied to a data driving processor 1 and then the data packets which are processed by the processor 1 are successively applied to the processor 1 via an interface part 3 at the fixed time intervals. These data packets circulate each processing part of the processor 1 and the processing is carried out based on a program. The executing time of the processing is calculated from the time when the dummy packet is outputted from the processor 1 and the time when the packet processed by the processor 1 is outputted last from the processor 1. Thus the executing time of the processing carried out based on the program of the processor 1 can be easily measured.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a method for determining the execution time i11+1 of a data-driven device, and more particularly to a method for measuring the execution time of processing according to a program in a data-driven device.

[Prior Art] In a conventional Neumann computer, various instructions are stored in a program memory in advance as a program, and the instructions are sequentially read by sequentially specifying addresses in the program memory by a program counter. executed.

On the other hand, a data-driven device is a type of non-Neumann computer that does not have the concept of sequential instruction execution using a program counter. Such data-driven devices employ an architecture based on parallel processing of instructions. In data-driven devices, instructions can be executed as soon as the data to be operated on is available, and multiple instructions are simultaneously driven by the data, so programs are executed in parallel according to the natural flow of data. . As a result, the time required for calculations is considered to be significantly reduced.

[Problems to be Solved by the Invention] As mentioned above, in a Neumann computer, program instructions are sequentially executed by a program counter that operates in response to the fundamental frequency, so it is difficult to determine the program execution time. It's easy.

However, in data-driven devices, the concept of sequential instruction execution using a program counter does not exist.
Since multiple instructions are simultaneously driven by data and programs are executed in parallel according to the natural flow of data, it is difficult to determine the program execution time.

Therefore, an object of the present invention is to provide a method that can easily measure the execution time of processing according to a program in a data-driven device.

[Means for Solving the Problems] A method for measuring the execution time of a data-driven device according to the present invention is that after a dummy packet that passes through the data-driven device without being processed is input into the data-driven device, the execution time is measured according to a program. The packets to be processed are sequentially injected into the data-driven device, and the time when the dummy packet was output from the data-driven device and the time when the packet processed by the data-driven device was last output from the data-driven device The execution time is calculated based on the .

[Operation] Let a be the time from when a dummy packet is received by the data-driven device by 1+-1 until the packet processed by the data-driven device is finally output from the data-driven device. Further, let b be the time from when a packet is input to the data-driven device until the packet is output without being processed by the data-driven device. This time is known. Further, let C be the time interval between packet inputs to the data-driven device. This time C is also known based on the set parameters.

Using these times a, b, and c, the execution time according to the program in a data-driven device is a+b-
It can be determined by c.

Furthermore, if the execution time is long, the execution time can be regarded as a.

[Example] Hereinafter, an example of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of a data-driven device development apparatus to which the execution time measuring method of the present invention is applied.

This development device includes a data-driven processor 1, a control computer 2, an interface section 3, and a tracer section 4.
．． Consists of 5. The control computer 2 controls the interface section 3 and the tracers 4 and 5 according to a control program, inputs various data to the interface section 3, and collects various data from the interface section 3. The interface section 3 is the control computer 2
The data-driven processor loads program data and injects data packets according to instructions from the data-driven processor. The tracer units 4 and 5 have a timer and a tracer memory, and store input data packets together with time information in the tracer memory. Here, the time information is information representing the time when the data packet was input to the tracer section. Further, the tracer units 4 and 5 have a function of dumping trace results using the interface unit 3.

FIG. 2 is a block diagram showing the configuration of the data-driven processor of FIG. 1. Further, FIG. 3 is a diagram showing the field structure of a data packet processed by the data-driven processor.

The data packet shown in FIG. 3 has a destination field,
Contains instruction and data fields. Destination information is stored in the destination field. This destination information is
Mainly contains processor information and node information. The processor information is information representing the processor number of the data-driven processor. Node information is information used for addressing when reading a data flow program.

Instruction information is stored in the instruction field, and operand data is stored in the data field.

In FIG. 2, a data packet is given to the branching section 11 from the interface section 3 of FIG. If the processor information in the destination field of the data packet matches the processor number given to the data-driven processor 1, the data packet is input to the program storage unit 1B through the merging unit 12. If the processor information does not match the processor number of the data-driven processor 1, the data packet is output from the data-driven processor 1 through the confluence section 18.

As shown in FIG. 4, the program storage unit 13 stores a data flow program consisting of multiple sets of destination information and command information. The program storage unit 13 reads destination information and instruction information from the data flow program by addressing based on the node information in the destination field of the input data packet, and stores the information in the destination field and instruction field of the pig packet, respectively. and outputs the data packet.

The data pair detection unit 14 waits for data packets output from the program storage unit 13.

That is, the paired data detection unit 14 detects two data packets having the same destination information, stores the operand data of one data packet in a predetermined data field of the other pig packet, and outputs the other data packet. do. Note that at this time, one of the data packets disappears. The data packet output from the data detection section] 4 is processed by the arithmetic processing section 15 and the I/Seaser section 4 (Fig. 1).
The flow is divided into both.

The arithmetic processing unit 15 decodes the instruction information of the data packet output from the paired data detection unit 14, performs the arithmetic processing specified by the instruction information on the two operand data, and uses the result as the data of the data packet. field and outputs the data packet to the branching unit 16.

The branching unit 6 outputs the data packet to the internal data buffer unit 7 or the merging unit 18 based on the destination information of the data packet. The internal data buffer unit 17 outputs data packets to the merging unit 12 on a first-come, first-served basis. The merging unit 12 transfers the input data packet to the program storage unit 13 without performing any processing on the input data packet.
Output to.

The merging unit 18 outputs the input data packet I to the outside of the data-driven processor without performing any processing on the input data packet.

In this way, the data packet is stored in the program storage unit 13,
Processing based on the program stored in the program storage section 13 by continuing to sequentially go through the pair data detection section 14, the arithmetic processing section 15, the branching section] 6, the internal data buffer section 17, the merging section 12, and the program storage section 13. progresses.

Data packets output from the data pair detection section 14 of the data-driven processor 1 to the outside are input to the tracer section 4 shown in FIG. Furthermore, the data packets output from the merging section 18 of the data-driven processor to the outside are
The signal is input to the tracer section 5 shown in the figure.

Next, a method for measuring execution time according to an embodiment of the present invention will be explained. Here, the execution time refers to the time from when the first data packet to be executed according to the data flow program is input to the data-driven processor 1, until those data packets are processed by the data-driven processor 1, and the data-driven This is the time until the last data packet is output from the processor 1. That is, it refers to the time from when a data packet is first output from the interface unit 3 in FIG. 1 until the last data packet is input to the tracer unit 5. Further, it is assumed that the data-driven processor 1 in FIG. 1 is given a processor number 0.

First, before a data packet to be processed is input to the data-driven processor 1, a dummy packet is input from the interface section 3 to the data-driven processor 1. This dummy packet, like the data packet to be processed, has the field configuration shown in FIG. However, the destination field of the dummy packet stores the processor number FF as processor information.

This processor number FF is data driven processor 1
This indicates that the signal passes through and is directly input to the I/laser section 5. The information stored in the instruction and data fields of the dummy packet does not affect the execution time measurement method of this embodiment. Therefore, it is sufficient that some information is stored in the command field and the data field.

After the dummy packet is input to the data-driven processor 1, data packets to be processed by the data-driven processor 1 are sequentially input from the interface section 3 to the data-driven processor 1 at regular time intervals.

As these data packets circulate through each processing section of the data-driven processor 1 shown in FIG. 2, processing according to the program progresses.

The dummy 1 packet input to the data-driven processor 1 is transferred to the branch section 1 and the confluence section 18 shown in FIG.
is applied to the outside through the , and is input to the tracer section 5 .

Data packets that are sequentially inputted from the merging section 18 of the data-driven processor are also inputted to the tracer section 5.

FIG. 5 shows the contents of the trace memory of the l-laser section 5. When a dummy packet is input to the tracer unit 5, the contents of the dummy packet and time information representing the input time are stored in the trace memory. Next, when the first data packet output from the data-driven processor 1 is input to the tracer section 5, the contents of the first data packet and its time information are stored in the trace memory. Similarly, the contents and time information of the data packets outputted from the data-driven processor 1 are sequentially stored in the trace memory.

Assuming that the time information of the dummy packet is T1 and the time information of the last data packet is T2, the data-driven processor 1 outputs the dummy packet, and then the data-driven processor 1 outputs the last two data packets. The time a is calculated by the following formula.

a=T2-Tl-(]) Also,
The time it takes for a dummy packet or data packet to pass through the data-driven processor 1 is as follows: When there is only one data-driven processor 1, the data packet or dummy packet passes through the branch section [1] and the confluence section]8 in Fig. 2. It's time to go through. This time is known from the structure of the data-driven processor 1.

The time C between the input of a dummy packet and a data packet input from the interface section 3 to the data-driven processor 1 is known from the parameter settings of the control computer 2.

Using the time a obtained from the above equation (1) and the known times and times C, the processing time T of the data-driven processor 1 according to the program can be calculated.

T=a+b-c (2) Note that if the execution time is long, the execution time 3 can be considered to be a.

FIG. 6 is a block diagram showing a development device for a data-driven device consisting of a multiprocessor.

In FIG. 6, the data-driven device consists of three data-driven processors 1. These data-driven processors 1 are sequentially given processor numbers 0 to 2.

The data packets output from the paired data detection unit 14 of the data-driven processor with processor number 0 are input to the corresponding tracer unit 4, and the data packets I output from the merging unit]8 are data-driven with processor number 1. Branch section of type processor 1] 1 is input. Similarly, data packets output from the paired data detection unit 14 of the data-driven processor 1- with processor number 1- are input to the corresponding tracer unit 4, and data packets output from the merging unit 18 are output from the processor number 2-. branch unit 1 of the data-driven processor 1]. Data packets output from the paired data detection unit 14 of the data-driven processor 1 with processor number 2 are input to the corresponding tracer unit 4, and data packets output from the merging unit]8 are input to the tracer unit 5. .

A processor number FF is stored in the destination field of the dummy packet as processor information. Therefore, the dummy packet input from the interface section 3 first passes through the branching section 11 and merging section 18 of the data-driven processor 1 with processor number O, and passes through the branching section 11 and merging section 18 of the data-driven processor 1 with processor number 1. passes through the confluence section 18, passes through the branch section 11 of the data-driven processor 1 of processor number 2 and the confluence section]8,
Tracer 5 is used as it is.

When measuring the execution time according to a program in a data-driven device consisting of three data-driven processors 1 shown in FIG. 6, as in the case of the development device shown in FIG.
The execution time T can be determined from 2). However, the time for a dummy packet or data packet to pass through the data-driven processor 1 is three times the time for it to pass through one data-driven processor 51. If N data-driven devices are connected, the time is N times the time to pass through one data-driven processor 1.

As described above, according to the execution time measuring method according to the above embodiment, data driving can be easily performed by simply inputting a dummy packet having the process number FF as processor information before inputting the data packet to be processed. It is possible to measure the execution time of processing according to the program of the molding device.

[Effects of the Invention] As described above, according to the present invention, the execution time of a data-driven device according to a program can be easily measured without changing the configuration of the data-driven device. Therefore, the program can be easily evaluated.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an example of the configuration of a data-driven device development apparatus to which the present invention is applied. FIG. 2 is a block diagram showing the configuration of a data-driven processor. FIG. 3 is a diagram showing a field structure of a data packet processed by a data-driven processor. FIG. 4 is a diagram showing a data flow program stored in the program storage section of the data-driven processor. FIG. 5 is a diagram showing an example of the contents stored in the tracer section. FIG. 6 is a block diagram showing the configuration of a data-driven device development device comprising a multiprocessor. In the figure, ] indicates a data-driven processor, 2 a control computer, 3 an interface section, and 4.5 a tracer section.

Claims (1)

[Claims] A method for measuring the execution time of processing according to a program in a data-driven device, the method comprising: injecting into the data-driven device a dummy packet that passes through the data-driven device without being processed; Packets to be processed according to the program are sequentially inputted to the data-driven device, and the time when the dummy packet is output from the data-driven device and the packet processed by the data-driven device is output from the data-driven device. An execution time measuring method for a data-driven device, the execution time being calculated based on the last output time.