JPS6368932A - Information processor - Google Patents

Abstract

PURPOSE:To perform proper input limitation by providing an input control part and sending external input data corresponding to an input permission token generated according to the execution of an input instruction to an information processor main body. CONSTITUTION:A processing element 1 equipped with the information processor main body 8, a network 6 in a processing element, etc., is provided with an input control part 8 which receives the external input data through the network 6. This control part 6 controls the external input data to the main body 8 by receiving the input permission token generated when the main body 8 executes an instruction indicated in the driving program of a program storage part 3 by an arithmetic part 5 for operand data passed from a queuing part 4. Therefore, input limitation is performed properly even for a program which varies in processed variable with an input data value to prevent the processor main body from stopping owing to an excessive input and the throughput from decreasing owing to the input limitation.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to a data-driven information processing device that can execute the same program in a superimposed manner using a tagged token control method. This invention relates to a data-driven information processing device that can prevent the information processing device from stopping due to a lack of hardware resources such as hardware resources such as hardware resources and memory caused by excessive input of data. .

Here, the tagged token control method means that when multiple different data groups are processed in parallel by the same program, a tag that is an identifier for each program execution is attached to the data packet. , a method that distinguishes the data used to execute individual programs and allows programs to be executed simultaneously without confusion.

[Conventional technology]

FIG. 6 is a configuration diagram showing an example of a conventional data-driven information processing device (processing element). 6th
In the figure, 3 is a program storage unit that stores a program, 4 is a waiting unit that performs the function of aligning two operand data of a binary instruction, and 5 is an arithmetic unit that performs instruction processing on operand data. 6 is a network within the processing element, which includes a program storage section 3, a waiting section 4. Information is exchanged between each functional unit of the calculation unit 5 and with the outside of the processing element.

In such conventional data-driven information processing devices, if the amount of data in the information processing device increases excessively, the information processing device's finite hardware resources cannot process all the data and the information processing device may stop. There were times when I ended up doing this.

Possible reasons for the increase in data amount are as follows.

1) During the program execution process, a large amount of data is
When data is generated one after another in excess of the data processing capacity of the information processing device.

2) When the amount of data input from outside is too large.

In the former case, the execution program itself exceeds the processing capacity of the information processing device that processes it, and countermeasures using software techniques such as modifying the program are required. Therefore, this case will not be considered a problem at this time.

On the other hand, in the latter case, by limiting the input according to the amount of data in the data-driven information processing device, the program can be executed without running out of hardware resources. However, in the case of a data-driven information processing device, each functional unit operates independently of each other, so it is very difficult to effectively limit inputs by reflecting the internal state of the information processing device. One conventional input restriction method involves constantly monitoring the status of each functional unit and prohibiting input when a certain functional unit is about to run out of resources. but,
In this method, a large amount of data has already been input into the information processing apparatus at the time when input is restricted, and as the data processing progresses, there is a possibility that resources will become insufficient. It may be possible to prohibit input as early as possible, but
In that case, overcontrol always occurs, resulting in a decrease in processing efficiency of the data-driven information processing device.

Furthermore, as mentioned above, in a data-driven information processing device, it is difficult to perform input control that directly reflects the internal state of the information processing device. There is an example in which input is restricted by adjusting the time interval at which data is input into an information processing device.

[Problem that the invention seeks to solve]

Generally, in a data-driven information processing device, the processing efficiency improves as the amount of data that can be processed within the information processing device increases. However, if the amount of data increases too much, hardware resources become insufficient and the information processing device stops, as described above.

When input is restricted by the input interval described above, if the input interval is too short, the amount of data inside the information processing device increases too much, resulting in a shortage of hardware resources. On the other hand, if the input interval is too long, the processing efficiency of the information processing device will drop significantly. The optimal data input interval exists between these values, but finding the optimal value requires repeated trial and error. That is, the optimum input interval varies depending on the program, and it is impossible to know whether the information processing device will operate without stopping at a certain input interval unless the program is actually executed.

The above is the first problem.

Second, this input restriction has no effect on programs whose processing amount varies depending on the value of input data. A program that includes repetitive processing or conditional branch processing has a processing amount that fluctuates depending on the value of input data. When the amount of processing is large, the amount of data generated during processing is also large. If you continue to input input values that generate such a large amount of data at the same input interval as input values when the amount of processing is small and the generated data is consumed one after another, you will eventually run out of hardware resources. ,
The information processing device stops. On the other hand, if an input value that causes only a small amount of processing is input at an input interval long enough to avoid running out of hardware resources even if the amount of processing is large, the processing efficiency will drop significantly.

These problems occur because the input interval does not reflect the internal state of the information processing device; the amount of processing changes depending on the executed program and input data, and the amount of data inside the information processing device when the program is executed changes. This is due to the inability to respond to changes in

[Means for solving problems]

An information processing device of the present invention is an information processing device that performs processing using a data-driven execution control method, and includes a program storage unit that stores a data-driven program including input instructions;
A waiting section that waits for operand data according to the data-driven program and prepares necessary operand data, and an arithmetic section that executes instructions instructed by the data-driven program on the operand data passed from the waiting section. an information processing apparatus main body, which receives input data sent from the outside, receives an input permission token generated according to execution of an input command set in the data-driven program, and transmits input data corresponding to the input permission token to the data-driven program; and an input control section for sending data to the main body of the information processing device.

[Effect]

According to the configuration of the present invention, the following effects are produced by installing the input control section and introducing a new input command as a command for the information processing apparatus main body.

By executing the input command, data (referred to as an input permission token) indicating that input data from the outside may be input into the information processing apparatus main body is generated. The input authorization token is sent to the input control unit. The input control unit sends input data input from the outside and temporarily stored to the information processing apparatus main body. Generally, since a program has a plurality of inputs, the input control unit selects the input data permitted to be input by the input permission token and sends the selected input data to the information processing apparatus main body. Also, when the input permission token is generated, the corresponding input data is
If the input data has not yet arrived at the input control unit, the input data is sent to the information processing device main body as soon as it is input.

The above input commands are written in a data-driven program in the same way as general commands.

When executing a program that restricts input using input commands, even if input data is input one after another from the outside, the input data will not be input until the program execution progresses and the input command is executed. , subsequent input data is not input into the main body of the information processing apparatus, but is kept waiting in the input control section.

Therefore, no matter how large the amount of input data from the outside is, there will be no excessive input inside the information processing apparatus main body, and it is possible to prevent the information processing apparatus main body from stopping.

Furthermore, the input restriction based on the input command is an input restriction that accurately reflects the amount of data inside the information processing apparatus main body. In general, when the amount of data inside the information processing apparatus main body is large due to the overlapping execution of programs, processing of a certain program takes longer than when executed with a small amount of data. Therefore, even if the program is the same, if the amount of data inside the information processing device is large, the processing time required to execute the input command will be long, and the interval between input permissions will be long. When the amount of data inside the device main body is small, the time required for processing up to the execution of an input command is shortened, and the interval between input permissions is also shortened.

Furthermore, even for programs whose processing amount varies depending on input data values, effective input restrictions can be applied in accordance with the variation in processing amount. For example, in the case of a program that includes repetitive processing where the number of repetitions varies greatly depending on the input data value, conventional input restrictions have a fixed input interval, so they cannot cope with changes in the amount of processing that varies greatly depending on the number of repetitions, and the effectiveness is reduced. Certain input restrictions were not possible. However, when input is restricted by an input command, the influence of the repeated processing can be absorbed by programming so that an input command that permits the next input is executed after the repeated processing ends. That is, when the number of repetitions is large, the time required for the repetition process becomes longer, and the execution of the input command becomes slower accordingly. Conversely, when the number of repetitions is small, the time required for the repetition process is short, and the input command is executed quickly. In this way, the timing of permission for the next input changes according to changes in the amount of processing due to changes in the number of repetitions of repeated processing, so this method is effective even for programs where the amount of processing changes depending on the input data value. Input restrictions can be implemented.

Furthermore, since the increase or decrease in the amount of data can be seen on the program, there is no need to search for the optimal value through trial and error, unlike conventional input restrictions based on input intervals. Furthermore, since the input command is written in the program, the input command is always executed before the processing of the program is completed. At that time, input for the next execution is permitted, so input is not delayed for no reason, and processing efficiency is not excessively reduced due to input restrictions.

〔Example〕

FIG. 1 is a configuration diagram of an embodiment of a data-driven information processing device (processing element PE) of the present invention. In the figure, 2 is an input control section that limits input. 3 is a program storage unit that stores programs, and 4 is 2
A waiting section 5 performs a function of aligning two operand data of a term instruction, and a calculation section 5 performs instruction processing on operand data, and these constitute an information processing device main body 8. Reference numeral 6 denotes a network within the processing element, which exchanges information between each functional unit of the input control unit 2, program storage unit 3, waiting unit 4, and calculation unit 5 and with the outside of the processing element.

Figure 2 shows the processing element internal network 6.
This is an example of a packet used within. A code indicating the functional unit to which the packet is to be sent is set in the functional unit selection code field. The tag field stores a tag for distinguishing data of each superimposed execution of a program. The destination field stores the address of the execution instruction, and the instruction code field stores the execution instruction code. There are two data fields that store one or two operand data.

When only one of the operand data of a binary operation is stored in the data field, a flag indicating whether the operand data is left operand data or right operand data is stored in the left/right flag field. In addition, even in communication with outside the processing element,
Packets with the same configuration shall be used.

Here, the operation of the processing element shown in FIG. 1 will be briefly explained. Input data for input instructions of the program is input to the processing elements in the form of input data packets. The input data packet is sent to the input control unit 2 through the processing element internal network 6 because the function unit selection code field is set to the code “to input control unit”. Input control section 2
When both the input data packet from the outside and the corresponding input permission token packet are received, the packet in which the input data value is embedded in the data 1 field of the input permission token packet is sent to the program storage unit through the processing element internal network 6. Sent to 3. Note that the input permission token is generated within the processing element according to the execution of an input command in the program, and is sent to the input control unit 2 through the intra-processing element network 6.

The program storage unit 3 reads the program storage using the contents of the destination field of the received packet as an address, and updates the contents of the destination field, left/right operand flag field, instruction code field, and function section selection code field of the received packet. , and sends it out again to the processing element internal network 6.

The packet sent from the program storage unit 3 to the intra-processing element network 6 is sent to each functional unit according to the contents of the functional unit selection code field. Additionally, some are output to the outside of the processing element. The input permission token packet is sent to the input control section 2.

Furthermore, in the case of a unary operation packet, there is no need to wait for operand data, so it is sent directly to the operation unit 5.

In the case of a binary operation packet, it is sent to the waiting section 4 in order to wait for a packet holding paired operand data.

The waiting unit 4 waits for two packets in which the contents of the destination field and the contents of the tag field of the received packets match, and stores right operand data, that is, right operand data, in the data 2 field of the left operand data packet. data packet data 1
The content of the field is stored and the packet is sent to the calculation unit 5 through the processing element internal network 6.
Send to. Note that whether a packet is a left operand data packet or a right operand data packet is determined by
Determined by the contents of the packet's left/right operand flag fields.

The calculation unit 5 performs calculations on the received packet.

If the instruction code in the instruction code field of the received packet is a unary operation instruction, perform the operation indicated by the contents of the instruction code field on the contents of the data 1 field of the packet, write the resulting data to the data 1 field, and The packet 2 is transmitted to the program storage unit 3 through the processing element internal network 6. If the instruction code in the instruction code field of the received packet is a binary operation instruction, the operation indicated by the contents of the instruction code field is performed on the contents of the data 1 field and data 2 field of the packet, and the resulting data is The data 1 field is written and the packet is sent to the program storage unit 3 through the processing element internal network 6.

By repeating the above operations, the processing element executes the program stored in the program storage section 3.

FIG. 3 is a configuration diagram of an example of the input control section 2 in FIG. 1. The operation of the input control section 2 will be explained in detail with reference to the drawings.

The input control unit 2 receives two types of packets, an input data packet and an input permission token packet, through the intra-processing element network 6. These two types of packets are distinguished in the intra-processing element network 6 by the contents of the left/right operand flag fields of the packets, and are sent to the respective entrances of the input control section 2. Note that in this embodiment, left/right operand flag fields are used to identify input data packets and input permission token packets.
A field storing both identification codes may be provided separately from the left/right operand flag field.

The input data packet is a packet that inputs input data corresponding to an input command of a program to a processing element, and the input data is stored in the data 1 field of the packet. The address of the input command from which the input data should be read is set in the destination field. Further, in the case of this embodiment, a tag is set in the tag field to distinguish data in order to execute a program in a superimposed manner. The contents of each of these fields, including the functional unit selection code field, are set when the input data packet is generated outside the processing element.

On the other hand, the input permission token packet is a packet generated within the processing element by executing a program. Like the input data packet, the input permission token packet also has a tag set in the tag field, and the address of the input command whose input is to be permitted is set in the destination field. However, unlike input data packets, the contents of the data fields of input authorization token packets have no meaning.

The input data bucket (20) received by the input control unit 2 is temporarily stored in the data buffer memory (input data buffer function) 10. On the other hand, the input permission token packet 21 is sent to the comparison section (comparison function) 11.

The comparison unit 11 compares the contents of the destination field and tag field of the sent input permission token packet with the contents of the destination field and tag field of the input data packet stored in the input data buffer memory 10, and determines the destination. Search for input data packets with matching field and tag field contents. If any of the input data packets stored in the input data panzer memory 10 has the destination field and tag field with the same content as the input permission token packet sent, that input data packet cannot be read. (22), and the packet recombination unit (packet) lff
(replacement function) 12. In this case, the packet recombination unit 12 embeds the contents (input data) of the data 1 field of the read input data packet in the data 1 field of the input permission token packet 23 sent from the comparison unit 11, and The contents of the selection code are updated and sent to the branching unit 13 as a packet destined for the program storage unit (24). If there is no input data packet in the input data panzer memory 10 that has the same destination field and tag field as the input permission token packet, the packet recombination unit 12
The input permission token packet 23 sent from 1 is output as is (24).

If the sent packet has a functional unit selection code destined for the input control unit (in other words, there is no corresponding input data packet in the input data buffer memory 10), the branching unit 13 determines what processing the input permission token packet should do. ), the packet is sent to the input permission token buffer memory (input permission token buffer function) 14 (26), and if the sent packet has a function unit selection code other than input control unit 2 (i.e.
When the input authorization token packet and the corresponding input data packet are complete (in this case, the packet has a function selection code destined for the program storage), the packet is sent to the processing element internal network 6 (step 25). ).

sent to the input permission token buffer memory 14.
Although the input permission token packet 26 with no corresponding input data packet is temporarily stored, the input control unit 2
If there is no new input permission token packet arriving from outside, the packets are sequentially read out (27) and sent to the comparison unit 11, and the above operation is repeated again.

Through the above operations, the function of the input command can be realized, and the input can be restricted by the program.

In the above embodiment, it is assumed that a tag has already been set in the tag field of the input data packet received by the input control unit.

A second example will be described below, which is an example in which no tag is set in the tag field of the input data packet received by the input control unit. In a second embodiment, untagged input data packets are input to the processing element. The input data packets are sent to the input control unit through the network within the processing element, and the input data packets are sent to the input control unit according to the order in which they were input to the processing elements.
Tags are set.

The configurations of the processing element and input control section of the second embodiment are the same as those of the first embodiment shown in FIGS. 1 and 3, respectively.
Although the configuration is the same as that of the embodiment, there are some differences in the functions of the functional parts. The intra-processing element network 6 needs to be such that at least the input data packet does not overtake the route from the processing element input to the input control unit 2. When the intra-processing element network 6 satisfies this requirement, it is guaranteed that the input data packets are received by the input control unit 2 while maintaining the order in which they were input to the processing elements. Further, in the first embodiment, the comparison unit 11 of the input control unit 2 compares the contents of the destination field and the tag field of the received input permission token packet 21,
In the second embodiment, the contents of the destination field and the tag field of the input data packet in the input data buffer memory 10 are compared to find the input data packet corresponding to the input permission token packet. The difference is that the input data packet corresponding to the input permission token packet is searched for by comparing only the contents. Furthermore, when the comparison unit 11 searches for input data packets in the input data buffer memory 10, the input data packets that arrive at the input control unit 2 first are searched first, and the input data packets whose destination fields match first Read out. Therefore, in the input date batho memory 10,
Even if there are multiple packets with the same destination, the input data packet input first is always read out first. By performing the same operation as in the first embodiment on the read input data packet 22 in the bucket/node recombination unit 12, the input data (the contents of the data 1 field of the input data packet/node) is Embedded in the input authorization token packet. In other words, the input data is tagged with an input permission token.

In this second embodiment, since a tag is set in the input control unit 2 for an input data packet input from outside the processing element, when a data packet is input from outside the processing element to the processing element, the data packet There is no need to tag.

In addition, in this embodiment, if a plurality of input permission token packets with the same destination field contents and different tags exist in the input control unit z at the same time, which input permission token packet's tag is attached to the input data first. I cannot guarantee that. Therefore, it is necessary to ensure in the program that a plurality of input permission token packets with the same destination field contents and different tags do not exist in the input control unit 2 at the same time.

Furthermore, the input control unit of the present invention can also implement input restrictions for a data-driven information processing device having a multiprocessor configuration. Next, an embodiment in the case of a multiprocessor configuration will be described.

FIG. 4 is a configuration diagram of a data-driven information processing device having a multiprocessor configuration. In the figure, 11 to 1ゎ are processing elements, and 7 is a network between processing elements. Each processing element I
I to I11 exchange information with each other or with the outside of the information processing device via the inter-processing element network 7. Each processing element 11
~1. 1 has the configuration shown in FIG. 1, as in the case of a single processing element. In this embodiment, it is assumed that input data is input to the data-driven information processing device in the form of an input data packet provided with a tag. Therefore, the input control section of each processing element is the same as the input control section of the first embodiment.

Figure 5 shows the network 7 between processing elements.
This is an example of the field configuration of a packet used in the packet. Packets with the same configuration are also used in the intra-processing element network 6. This field configuration is
The configuration is the same as that shown in FIG. 2, but the destination field is divided into two fields: a PE# field indicating the processing element number and an instruction address program indicating the address when reading the program memory in the program storage section 3 of each processing element. has been used.

Here, the operation of the data-driven information processing apparatus having a multiprocessor configuration will be briefly described.

Input data for input commands is input to the inter-processing element network 7 in the form of input data packets. An input data packet input into the inter-processing element network 7 is sent to the processing element indicated by the PE# field in the destination field of the packet. The input data packet received by the corresponding processing element is sent to the input control section 2 through the intra-processing element network 6. Thereafter, in the processing element, the processing described in the single processor example is repeated.

In the case of a data-driven information processing device having a multiprocessor configuration, the program storage unit 3 may transmit packets destined for other processing elements. When the intra-processing element network 6 receives a packet destined for another processing element, it sends the packet to the inter-processing element network 7. The packet sent to the inter-processing element network 7 is sent to the corresponding processing element according to the contents of the PE# field of the destination field. When the packet sent to another processing element in this way is received by the target processing element, it is sent to the program storage unit 3 through the intra-processing element network 6, and from then on, the packet is sent to the program storage unit 3 in that processing element as well. The process described in the example is repeated.

Input restriction using an input command restricts input by the program itself executed by the data-driven information processing device, so input restriction can be implemented regardless of the configuration of the information processing device itself.

Note that when restricting only input from outside the data-driven information processing device, only the processing element that receives input data packets from the outside and performs input control, that is, the processing element that assigns input commands, performs input control. It is sufficient that the input control section is provided, and the other processing elements may be conventional processing elements not provided with the input control section shown in FIG.

Further, the following problems can also be solved using the input control section of the present invention.

In a data-driven information processing device with a multiprocessor configuration, if there is a large imbalance in the amount of program processing allocated to each processing element, the processing element with a large amount of processing may run out of hardware resources, which causes The entire system may stop.

In such a case, there is no effect unless the input data to the data-driven information processing device is limited, and the input data to the processing element is also limited for each individual processing element. If each processing element of the data-driven information processing device is configured with a processing element equipped with the input control section of the present invention,
If data sent from other processing elements is programmed to be taken into the processing element by an input command in the same way as input data from the outside, input restrictions for each processing element can be realized.

〔Effect of the invention〕

According to the information processing device of the present invention, since the input control unit is provided and the input command is introduced, it is possible to implement input restrictions in the input control unit by a program using the input command. Therefore, even if data is input one after another to the information processing device from the outside, the amount of data that can be taken in is limited according to the progress state of the program being executed, so the information processing device itself can be prevented from stopping due to excessive input. It can be prevented. Furthermore, in the case of input restriction by a program, processing efficiency does not decrease significantly due to input restriction. Furthermore, input restrictions can now be applied to programs whose processing amount changes depending on input data values, which was previously impossible to do.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of an embodiment of the data-driven information processing device of the present invention, FIG. 2 is a diagram showing the field structure of a packet, FIG. 3 is a block diagram of an example of the input control section, and FIG. 4 is a block diagram of an example of the input control section. A configuration diagram of an example of a data-driven information processing device with a multiprocessor configuration. FIG. 5 is a diagram showing the field configuration of a packet used in a data-driven information processing device with a multiprocessor configuration. FIG. 6 is a configuration of a conventional processing element. It is a diagram. 1.11-1o...processing element, 2.
...Input control section, 3...Program storage section, 4...
Waiting unit, 5... Calculation unit, 6... Network within processing elements, 7... Network between processing elements, 10... Input data buffer memory, 11... Comparison unit, 12... Packet recombination Part, 13... Branch part, 14... Input permission token buffer memory Patent applicant Matsushita Electric Industrial Co., Ltd. Sanyo Electric Co., Ltd. Mitsubishi Electric Corporation

Claims (2)

[Claims] (1) An information processing device that performs processing using a data-driven execution control method, and includes a program storage unit that stores a data-driven program including input instructions, and a program storage unit that stores operand data according to the data-driven program and performs necessary operations. An information processing device main body consisting of a waiting section that arranges operand data, an arithmetic section that executes instructions instructed by the data-driven program on the operand data passed from the waiting section, and input data sent from the outside. an input control unit that receives an input permission token generated in accordance with execution of an input command set in the data-driven program and sends input data corresponding to the input permission token to the information processing device main body. Information processing device. (2) The input control unit has an input data buffer function for temporarily storing input data from the outside, and receives the input permission token and detects input data corresponding to the input permission token from the input data buffer function. a comparison function, a packet recombination function that incorporates the detected input data into a packet processed within the information processing device main body, and a temporary storage of the input permission token when the input data corresponding to the input permission token is not input. The information processing apparatus according to claim 1, further comprising an input permission token buffer function for sending the input permission token to the comparison function.