WO2022148359A1

WO2022148359A1 - Multi-core system-based broadcasting method, electronic device, and computer readable medium

Info

Publication number: WO2022148359A1
Application number: PCT/CN2022/070211
Authority: WO
Inventors: 何伟; 沈杨书; 祝夭龙
Original assignee: 北京灵汐科技有限公司
Priority date: 2021-01-08
Filing date: 2022-01-05
Publication date: 2022-07-14
Also published as: CN112887214B; CN112887214A

Abstract

The present disclosure provides a multi-core system-based broadcasting method, an electronic device, and a computer readable medium. The multi-core system comprises a plurality of processing cores, and the plurality of processing cores are communicatively connected to each other by means of a network on chip. The method comprises: when a preset broadcast condition is satisfied, sending a first broadcast signal to each processing core of the multi-core system for the first time, the first broadcast signal comprising the maximum transmission step size; and when a first feedback signal for the first broadcast signal is not received within a first time after the first broadcast signal is sent for the i-th time, sending the first broadcast signal to the processing core of the multi-core system for the (i+1)-th time, i being an integer greater than or equal to 1.

Description

Broadcasting method, electronic device, and computer-readable medium based on many-core system

technical field

The present disclosure relates to the field of computer technology, and in particular, to a broadcasting method, electronic device, computer-readable medium and computer program product based on a many-core system.

Background technique

With the development of technology (especially artificial intelligence technology), compared with single-core chips with limited computing power, such as CPU (Central Processing Unit, central processing unit), ALU (Arithmetic and Logic Unit, arithmetic logic unit), etc. Core (including multi-core) chips are increasingly being used.

In some many-core chips, a Network On Chip (NOC) is used to implement communication between processing cores.

SUMMARY OF THE INVENTION

The present disclosure provides a broadcasting method, an electronic device, a computer-readable medium, and a computer program product based on a many-core system.

In a first aspect, the present disclosure provides a broadcasting method based on a many-core system, the many-core system includes a plurality of processing cores, and the plurality of processing cores are communicated and connected through an on-chip network, and the method includes: when a preset is satisfied In the case of the broadcast condition of , the first broadcast signal is sent to the processing core of the many-core system for the first time, wherein the first broadcast signal includes the maximum transmission step; In the first time after the signal, if the first feedback signal for the first broadcast signal is not received, send the first broadcast signal to the processing core of the many-core system for the i+1th time, i is an integer greater than or equal to 1.

In a second aspect, the present disclosure provides an electronic device comprising: a plurality of processing cores; and an on-chip network configured to exchange data among the plurality of processing cores and external data; wherein the processing Instructions are stored in the core, and the instructions are executed by the processing core to enable the processing core to execute the above-mentioned broadcast method.

In a third aspect, the present disclosure provides a computer-readable medium on which a computer program is stored, wherein the computer program implements the above-mentioned broadcasting method when executed by a processor.

In a fourth aspect, the present disclosure provides a computer program product comprising computer-readable codes, or a non-volatile computer-readable storage medium carrying computer-readable codes, wherein, when the computer-readable codes are stored in an electronic device When running in the processor of the electronic device, the processor in the electronic device executes the method for implementing the above-mentioned broadcasting.

According to the embodiments of the present disclosure, a broadcast signal can be sent to the processing core when the broadcast conditions are satisfied, and the broadcast signal can be sent again when no feedback signal is received within the first time after sending. By introducing a replay mechanism and limiting The range of broadcast signal transmission can improve the overall processing efficiency in the many-core system and improve the utilization rate of transmission resources in the system.

It should be understood that what is described in this section is not intended to identify key or critical features of embodiments of the disclosure, nor is it intended to limit the scope of the disclosure. Other features of the present disclosure will become readily understood from the following description.

Description of drawings

The accompanying drawings are used to provide a further understanding of the present disclosure and constitute a part of the specification, and together with the embodiments of the present disclosure, they are used to explain the present disclosure, and are not intended to limit the present disclosure. The above and other features and advantages will become more apparent to those skilled in the art by describing detailed example embodiments with reference to the accompanying drawings, in which:

FIG. 1 is a schematic diagram of a many-core system according to an embodiment of the present disclosure.

FIG. 2 is a flowchart of a broadcasting method based on a many-core system according to an embodiment of the present disclosure.

FIG. 3 is a block diagram of a broadcast apparatus based on a many-core system according to an embodiment of the present disclosure.

FIG. 4 is a block diagram of an electronic device according to an embodiment of the present disclosure.

Detailed ways

In order for those skilled in the art to better understand the technical solutions of the present disclosure, the exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, and they should be considered to be exemplary only. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted from the following description for clarity and conciseness.

Various embodiments of the present disclosure and various features of the embodiments may be combined with each other without conflict.

As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.

The terminology used herein is used to describe particular embodiments only and is not intended to limit the present disclosure. As used herein, the singular forms "a" and "the" are intended to include the plural forms as well, unless the context clearly dictates otherwise. It will also be understood that when the terms "comprising" and/or "made of" are used in this specification, the stated features, integers, steps, operations, elements and/or components are specified to be present, but not precluded or Add one or more other features, integers, steps, operations, elements, components, and/or groups thereof. Words like "connected" or "connected" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect.

Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in common dictionaries should be construed as having meanings consistent with their meanings in the context of the related art and the present disclosure, and will not be construed as having idealized or over-formal meanings, unless expressly so limited herein.

Many Core system refers to a processing core collection system composed of a large number of processing cores connected together in a preset manner. Among them, each processing core (core, Core) is the smallest unit that can be independently scheduled and has independent computing capabilities, that is, each processing core has its own independent storage resources and computing resources, so that it can independently perform required operations (for example, neural network operations). There are various specific forms of the many-core system. For example, the many-core system may be a multi-core chip, a combination of multiple single-core chips, or a combination of multiple multi-core chips, which is not limited in the present disclosure.

Different processing cores of the many-core system can be connected to each other through routing (the routing can be, for example, in the form of a bus, an on-chip network, etc.), thereby realizing information exchange between any two processing cores. Thus, multiple processing cores in the many-core system can cooperate with each other to perform certain operations. In addition to processing cores and routing, the many-core system may also include some other units, such as a scheduler that controls each processing core, an on-chip storage space accessible to each processing core, etc., which is not limited in the present disclosure.

According to the embodiments of the present disclosure, multiple processing cores in the many-core system are connected through a Network On Chip (NOC) communication to implement communication between the processing cores and electronic devices outside the many-core system. Communication between (eg a compiler) and a processing core.

FIG. 1 is a schematic diagram of a many-core system according to an embodiment of the present disclosure. As shown in Figure 1, the multiple processing cores in the many-core system (that is, the dots in Figure 1, each dot in Figure 1 represents a processing core of the many-core system) are arranged in an array, adjacent ( The processing cores including adjacent in the column direction and adjacent in the row direction can directly communicate signals to each other. As shown in FIG. 1, the processing core N0 and the processing core N1 are adjacent processing cores in the row direction. The processing core N0 can send a signal to the processing core N1, and the processing core N1 can also send a signal to the processing core N0.

Through signal transmission between adjacent processing cores, signals can reach other processing cores in the many-core system from one processing core in the many-core system that are not adjacent to it. The processing core N0 sends the signal to the processing core N1, and the processing core N1 sends the signal to the processing core N2, so that the signal reaches the processing core N2 from the processing core N0.

In some embodiments, the processing core can send the data to be sent to the destination processing core to adjacent processing cores in the form of signals, and forward it step by step through the on-chip network, and finally reach the destination processing core, thereby realizing any processing core and processing Communication between cores.

In some embodiments, the processing core may add flag bit information to the data (eg, in the data header), where the flag bit information is used to indicate the location of the destination processing core to which the data needs to be sent, and send the data in the form of a signal to the destination processing core.

In some embodiments, the processing cores can also broadcast data. Since broadcasting is to send data to all possible receiving processing cores as much as possible, it does not have a destination. Therefore, when broadcasting data, it is not necessary to add flag information to the data. A processing core that needs to broadcast data directly sends the data to all adjacent processing cores in the form of signals, and these adjacent processing cores then send the received signals to other adjacent processing cores except the processing core.

Referring to FIG. 1 , it is assumed that the processing core N0 is a processing core that needs to broadcast data, and it sends the data to all its adjacent processing cores, such as the processing core N1 , in the form of a signal. After the processing core N1 receives the signal, it sends it to all other adjacent processing cores except the processing core N0, such as the processing core N2. Similarly, after the processing core N2 receives the signal, it also sends the signal to all other adjacent processing cores except the processing core N1.

In this way, the signal can traverse the processing cores of the entire many-core system "from near to far" from the processing core N0 (ie, the processing cores that are close to the processing core N0 are received before the processing cores that are farther away from the processing core N0). Signal).

At the same time, when the processing core sends a signal, it does not "turn back", that is, after the processing core receives the signal, it does not send a signal to the processing core that sends the signal to itself. For example, after the processing core N1 receives the signal sent by the processing core N0, The signal will not be sent to the processing core N0 again, so the signal will "go farther and farther", and the processing core that has already received the signal will not receive the signal again.

When a signal is transmitted in an unaddressed broadcast mode, the data packets of the signal may be propagated unrestrictedly, wasting transmission resources. Moreover, the processing core that has received the signal will not receive the signal again, which may cause the processing core to fail to respond to the signal, reducing the processing efficiency of the system. A possible situation is, for example, that the processing core is in a busy state when it receives a signal for issuing a task, but enters an idle state after the signal has passed, resulting in failure to respond to the signal.

According to the embodiments of the present disclosure, a broadcasting method based on a many-core system is provided, which can limit the range of signal transmission and introduce a replay mechanism, which can improve the utilization rate of transmission resources and improve the overall processing efficiency of the system.

The broadcasting method based on the many-core system according to the embodiment of the present disclosure can be applied to any processing core of the many-core system, or an electronic device outside the many-core system, which is not limited in the present disclosure.

FIG. 2 is a flowchart of a broadcasting method based on a many-core system according to an embodiment of the present disclosure. As shown in Figure 2, the method includes:

In step S21, when a preset broadcast condition is satisfied, a first broadcast signal is sent to the processing core of the many-core system for the first time, wherein the first broadcast signal includes a maximum transmission step size;

In step S22, if the first feedback signal for the first broadcast signal is not received within the first time after the i-th transmission of the first broadcast signal, the processing core of the many-core system is sent to the The first broadcast signal is sent i+1 times, where i is an integer greater than or equal to 1.

For example, broadcast conditions may be preset. For a processing core or an external electronic device of a many-core system, the broadcast condition may include: there is a task to be allocated; for a processing core, the broadcast condition may include: there is a task to be allocated in the processing core, the processing core cannot process the current Task (for example, an operation error occurs in the processing core, the temperature of the processing core exceeds a temperature threshold, etc.), the processing of the current task of the processing core has been completed, and the like. It should be understood that those skilled in the art can set the broadcast condition according to the actual situation, which is not limited in the present disclosure.

In some embodiments, if the preset broadcast conditions are met, a broadcast signal (referred to as a first broadcast signal) may be generated in step S21, and the first broadcast signal is sent to the processing core of the many-core system for the first time Signal. Wherein, the first broadcast signal may be sent in the above-mentioned broadcast manner. That is, for an external electronic device, the first broadcast signal can be sent to the first processing core directly connected to the external electronic device in the many-core system, and then sent by the first processing core to its adjacent processing cores. The first broadcast signal; for any processing core (which can be referred to as the first processing core), the first broadcast signal can be directly sent to its adjacent processing cores, and these processing cores adjacent to the first processing core will then receive the first broadcast signal. The broadcast signal is sent to other processing cores adjacent to it except the first processing core. By analogy, the step-by-step transmission of the first broadcast signal can be realized.

In some embodiments, the first broadcast signal refers to a signal that needs to be fed back, that is, after other processing cores receive the signal, if the processing core satisfies the condition for feeding back the first broadcast signal, it can send the signal to the external electronic device through the on-chip network. The device or the first processing core sends a feedback signal. Since the communication between the processing cores is through the processing core of the many-core system, it can also be said that the processing core sends the feedback signal to the external electronic device or the first processing core.

In some embodiments, when a certain processing core receives a broadcast signal, it does not meet the feedback condition, for example, it is in a "busy" state, then the feedback signal will not be sent.

Since the broadcast signal of the network on chip "doesn't look back", once a processing core receives the first broadcast signal, it cannot receive the first broadcast signal again, so even after the processing core has been satisfied to feed back the first broadcast signal condition, the feedback signal can no longer be sent according to the first broadcast signal.

When a processing core receives the first broadcast signal, and it satisfies the condition for feeding back the first broadcast signal, the feedback signal can be sent to the external electronic device or the first processing core. The process of sending the feedback signal may specifically be as follows: the processing core sends a feedback signal to the adjacent processing core that is closest to the external electronic device or the first processing core among its adjacent processing cores according to the position of the external electronic device or the first processing core, the Neighboring processing cores continue to send feedback signals to their neighboring processing cores that are closer to the external electronic device or the first processing core until the feedback signal reaches the external electronic device or the first processing core.

That is to say, it also takes time for the processing core that sends the feedback signal to send the feedback signal until the external electronic device or the first processing core receives the feedback signal, and the processing core that sends the feedback signal is farther away from the external electronic device or the first processing core. The shorter the time from sending the feedback signal to the external electronic device or the first processing core receiving the feedback signal, the farther the processing core sending the feedback signal is from the external electronic device or the first processing core, the longer the time from sending the feedback signal to the external electronic device The longer the device or the first processing core receives the feedback signal.

In some embodiments, a first time may be preset to indicate a time interval for replay. If the feedback signal (referred to as the first feedback signal) sent by other processing cores is received within the first time, the broadcasting process can be ended and the first feedback signal can be processed. For example, according to the received first feedback signal, a connection signal is sent to the processing core that sent the first feedback signal, where the connection signal is used to indicate that the first feedback signal has been received.

In some embodiments, if no feedback signals sent by other processing cores are received within the first time, the next first broadcast signal is sent to the processing cores of the many-core system. That is, in step S22, for any i-th (i is an integer greater than or equal to 1) transmission, if the first broadcast signal is not received within the first time after the i-th transmission of the first broadcast signal When a first feedback signal of the broadcast signal is sent, the first broadcast signal is sent to the processing core of the many-core system for the i+1th time. Until a feedback signal is received, or a condition for stopping replay is reached (for example, a preset threshold of replay times is reached), the broadcasting process can be ended.

In some embodiments, within the first time after sending the first broadcast signal, the first feedback signal for the first broadcast signal is not received, which may be the processing core that receives the first broadcast signal, The feedback condition is not met when the broadcast signal is broadcast; it may also be that the processing core that receives the first broadcast signal and meets the feedback condition is far away, although it has already sent the feedback signal, but the feedback signal is still in the process of being transmitted, and it fails to reach the first broadcast signal. reach the external electronic device or the first processing core within a time.

In some embodiments, if the feedback signal is not received within the first time, the broadcast signal is sent again, so that the processing core that does not meet the feedback condition when receiving the last first broadcast signal can receive the The first broadcast signal, and the feedback signal is sent if the feedback condition is satisfied. In the case where the processing core is relatively close, the feedback signal sent by the processing core may reach the external electronic device or the first processing core faster than the feedback signal sent by the processing core at a farther distance in response to the last first broadcast signal. nuclear.

That is to say, in contrast, in the case of replay, the received feedback signal is more likely to be sent by the processing core closer to itself, so that the external electronic device or the first processing core is processing the feedback signal. In the process and subsequent processing, when it is necessary to communicate with the processing core, due to the short distance, the communicated data can reach the processing core faster, which is convenient for processing feedback signals, and the number of processing cores occupied in the communication process (also It can be said that resources) are also less.

In some embodiments, when the method is applied to the first processing core, referring to FIG. 1 , the processing core N0 is, for example, the first processing core, and sends the first broadcast signal to the adjacent processing core; when the processing core N1 receives the processing When the first broadcast signal sent by the core N0, if the processing core corresponding to the processing core N1 does not satisfy the feedback condition, the processing core N1 sends the broadcast signal to the processing core adjacent to it except the processing core N0, such as the processing core N2; The processing core corresponding to the core N2 also does not meet the conditions for feeding back broadcast signals, and the processing core N2 continues to send the broadcast signal to the processing cores adjacent to it except the processing core N1, and so on, the broadcast signal may reach the processing core N3.

In some embodiments, if the processing core N3 satisfies the feedback condition, the processing core N3 sends a feedback signal to the processing core N0 according to the position of the processing core N0, and the process of sending the feedback signal is the same as the first broadcast signal from the processing core N0 to the processing core N3. The process is similar, that is, the feedback signal needs to pass through multiple processing cores such as the processing core N3, the processing core N2, and the processing core N1 to reach the processing core N0. In the process of the feedback signal reaching the processing core N0, since the distance between the processing core N0 and the processing core N3 is relatively long, the feedback signal may not reach the processing core N0 in the first time.

In some embodiments, if the processing core N0 does not receive the feedback signal within the first time, the first broadcast signal is sent again. The processing core N1 sends a feedback signal to the processing core N0. The processing core N1 is relatively close to the processing core N0. It may be that the feedback signal sent by the processing core N3 to the processing core N0 in the last broadcast process has not reached the processing core N0. The feedback signal sent by the processing core N1 to the processing core N0 during the broadcast process has reached the processing core N0, so that the processing core N0 preferentially receives the feedback signal sent by the processing core N1. Among them, the distance between the processing core N1 and the processing core N0 is obviously shorter than the distance between the processing core N3 and the processing core N1, and the communication speed between the processing core N1 and the processing core N0 is obviously higher than the communication speed between the processing core N3 and the processing core N0. .

In this way, it is beneficial for the close-range processing to check the broadcast signal for feedback, so that it can give priority to handshake with the close-range processing core and perform corresponding processing, which can improve the overall processing efficiency of the many-core system.

In addition, the many-core system has many processing cores. During the process of transmitting the broadcast signal to the processing cores farther away, the external electronic device or the first processing core may have received the feedback signal sent by the processing cores that are closer. At this time, it is meaningless to continue to transmit the broadcast signal, but it will occupy the transmission resources of the processing core and affect the transmission of other signals in the on-chip network.

At the same time, the broadcast signal is sent to the processing core with a long distance. Even if the processing core can process the broadcast signal, the feedback signal sent by the processing core is also received. Signal transmission between multiple processing cores may also cause signal loss, affect communication quality, and reduce the processing efficiency of processing signals by the processing cores.

By sending broadcast signals only to the processing cores within a predetermined range, on the one hand, it can avoid the waste of transmission resources caused by unrestricted broadcasting; The communication between the processing cores that send the feedback signal only passes through the number of processing cores within a predetermined number, so as to ensure the quality of communication with the processing cores that send the feedback signal, and improve the efficiency of processing the feedback signal.

According to an embodiment of the present disclosure, a maximum transmission step size is included in the first broadcast signal. That is to say, by adding a predetermined step size limit to the first broadcast signal, the propagation range of the first broadcast signal is limited, so as to ensure that the first broadcast signal is only sent to the processing core whose distance from the first processing core is within the predetermined range. nuclear. In this way, the utilization rate of transmission resources in the many-core system can be improved.

In some embodiments, every time the first broadcast signal passes through a transfer between processing cores, the value of counting the number of processed cores in the first broadcast signal is incremented by 1, and when this value is consistent with the predetermined maximum transmission step size, the Delete the broadcast signal.

In some embodiments, each time the first broadcast signal is transferred between processing cores, the maximum transmission step size of the first broadcast signal is decreased by 1, and when the maximum transmission step size becomes 0, the maximum transmission step size of the first broadcast signal is reduced by 1. The first broadcast signal is forwarded and the first broadcast signal is deleted.

In some embodiments, the processing core that receives the first broadcast signal may determine the distance to the external electronic device or the first processing core, and compare it with the maximum transmission step size. If the distance does not reach the maximum transmission step size, it will be forwarded; if the distance reaches the maximum transmission step size, it will not be forwarded and deleted. The present disclosure does not limit the specific implementation.

According to the broadcasting method of the embodiment of the present disclosure, a broadcasting signal can be sent to the processing core under the condition that broadcasting conditions are satisfied, and the broadcasting signal can be sent again under the condition that no feedback signal is received within the first time after sending, by introducing a replay mechanism And limiting the range of broadcast signal transmission can improve the overall processing efficiency in the many-core system and improve the utilization rate of transmission resources in the system.

In some embodiments, the broadcasting method according to the embodiment of the present disclosure may further include:

If a first feedback signal for the first broadcast signal is received within a first time after the i-th transmission of the first broadcast signal, use the first feedback signal as a target feedback signal;

Corresponding processing is performed according to the target feedback signal.

That is to say, within the first time after sending the first broadcast signal for the i-th time, if the first feedback signal sent by other processing cores is received, the first feedback signal can be directly used as the target feedback signal, and the broadcasting process is ended. . Further, corresponding processing is performed on the target feedback signal, such as shaking hands with the processing core corresponding to the first feedback signal, sending task data to the processing core, or receiving task data from the processing core, etc. The present disclosure does not limit the specific processing method. .

In this way, the judgment process can be simplified, and the processing efficiency of the feedback signal can be improved.

In the case of receiving a first feedback signal for the first broadcast signal within a first time after sending the first broadcast signal for the i-th time, buffering the first feedback signal;

If the second feedback signal for the first broadcast signal is not received within a second time after the first feedback signal is received, the first feedback signal is used as the target feedback signal;

Corresponding processing is performed according to the target feedback signal.

That is to say, within the first time after the i-th first broadcast signal is sent, if the first feedback signal sent by other processing cores is received, the first feedback signal can be buffered and waited for a certain period of time (that is, the second time). Wherein, the second time is smaller than the first time, that is, the time interval from receiving the feedback signal for the first time to stopping receiving the feedback signal is smaller than the time interval between repeatedly sending the broadcast signal.

In some embodiments, if no other feedback signal (referred to as a second feedback signal) is received within the second time, the first feedback signal may be used as the target feedback signal, and the broadcasting process is ended. Further, corresponding processing is performed on the target feedback signal, such as shaking hands with the processing core corresponding to the first feedback signal, sending task data to the processing core, or receiving task data from the processing core, etc. The present disclosure does not limit the specific processing method. .

In this way, after receiving the first feedback signal and waiting for a certain period of time, other processing cores can be given more feedback opportunities, so as to find a processing core with a closer distance, thereby improving the efficiency of task processing.

In the case where a second feedback signal for the first broadcast signal is received within a second time after the first feedback signal is received, from all feedback signals received up to the end of the second time Determine the target feedback signal; wherein, the target processing core that sends the target feedback signal is the processing core with the smallest signal transmission step;

Corresponding processing is performed according to the target feedback signal.

That is to say, within the first time after the i-th first broadcast signal is sent, if the first feedback signal sent by other processing cores is received, the first feedback signal can be buffered and waited for a certain period of time (that is, the second time).

In some embodiments, if other feedback signals (referred to as second feedback signals) are received within the second time, all feedback signals (including the first feedback signal and at least one of the feedback signals received by the end of the second time) may be The target feedback signal is determined in the second feedback signal), and then the corresponding processing is performed on the target feedback signal,

In some embodiments, the positions of the processing cores that send the feedback signals may be determined from the feedback signals received, the distances of the processing cores may be calculated, and the processing core with the shortest distance may be selected from all the processing cores corresponding to the feedback signals received. The processing core is the target processing core, and the feedback signal sent by the processing core is the target feedback signal.

In some embodiments, the distance of a certain processing core specifically refers to the signal transmission step size between the processing core and the external electronic device or the first processing core, that is, the signal is transmitted from the processing core to the external electronic device or the first processing core The length of the shortest path traversed. Determine the processing core with the shortest distance as the target processing core. That is to say, determine the processing core from which all received feedback signals come from, and determine the processing core with the smallest signal transmission step among these processing cores (or at the same transmission speed, send The signal can reach the external electronic device or the processing core of the first processing core as soon as possible as the target processing core, and the feedback signal sent by the processing core is the target feedback signal.

In some embodiments, after the target feedback signal has been determined, other feedback signals may be received, and if other feedback signals are received, the feedback signal may be ignored.

In this way, the target processing core with the smallest signal transmission step size can be selected according to the multiple received feedback signals, and the feedback signal of the target processing core can be processed, so that it can be paired with a processing core with a closer distance and improve communication. Efficiency and efficiency of task handling.

The broadcasting method according to the embodiment of the present disclosure can be used for task scheduling, for example, issuing a new task from an external electronic device or the first processing core to other processing cores; or when the first processing core is in an idle state, actively Other processing cores ask if there are new tasks to be processed.

In some embodiments, the broadcasting method according to the embodiment of the present disclosure may be applied to a first processing core of the many-core system, or an electronic device outside the many-core system, where the first processing core is a plurality of the process one of the cores,

Wherein, the broadcast condition includes: there is a first target task to be assigned; the first broadcast signal includes a handshake signal for assigning a task;

The performing corresponding processing according to the target feedback signal includes:

Sending task data corresponding to the first target task to the target processing core corresponding to the target feedback signal, so that the target processing core executes the first target task.

For example, when performing task scheduling, the external electronic device or the first processing core may issue a new task to other processing cores by way of broadcasting.

In this case, the preset broadcast condition may include: there is a first target task to be allocated. That is, when there is a first target task to be allocated, it is determined that the broadcast conditions are met, and the first broadcast signal can be sent to the processing core of the many-core system in steps S21-S22, and the feedback signal is not received within the first time. replay.

In this case, the first broadcast signal includes a handshake signal for assigning tasks, that is, handshakes can be performed with other processing cores to assign tasks to other processing cores.

In some embodiments, the processing core that satisfies the feedback condition may send a feedback signal to the external electronic device or the first processing core; after receiving the feedback signal, the external electronic device or the first processing core may determine whether to The feedback signal is determined as the target feedback signal, and the description will not be repeated here.

In some embodiments, when the feedback signal is determined to be the target feedback signal, corresponding processing is performed according to the target feedback signal. That is, communicate with the target processing core that sends the target feedback signal through the on-chip network to complete the handshake; and after completing the handshake, send the task data corresponding to the first target task to the target processing core, so that the target processing core executes the first target processing core. a target task.

In this way, decentralized task scheduling can be achieved, and through dynamic task distribution, the balance between the processing cores in the many-core system can be achieved, and the overall processing efficiency of the system can be improved.

In some embodiments, the broadcasting method according to the embodiment of the present disclosure may be applied to a first processing core of the many-core system, where the first processing core is one of a plurality of the processing cores,

Wherein, the broadcast condition includes any one of the following: there is a second target task to be allocated in the first processing core, an operation error occurs during the processing of the second target task by the first processing core, the first processing the temperature of the core during processing of the second target task exceeds a temperature threshold; the first broadcast signal includes a handshake signal for assigning tasks;

Sending task data corresponding to the second target task to the target processing core corresponding to the target feedback signal, so that the target processing core executes the second target task.

For example, when the first processing core needs to issue a new task, or the first processing core itself cannot process the current task, the first processing core may issue a corresponding task to other processing cores by broadcasting.

In this case, the preset broadcast conditions may include any of the following: there is a second target task to be allocated in the first processing core, an operation error occurs during the processing of the second target task by the first processing core, the first processing core The temperature during processing of the second target task exceeds the temperature threshold.

In some embodiments, if an operation error occurs during the processing of the second target task by the first processing core, the first processing core cannot obtain the correct processing result of the second target task, and other processing cores need to execute the second target task; If the temperature of the first processing core exceeds the temperature threshold during processing of the second target task, the first processing core cannot continue to process the second target task, and other processing cores need to execute the second target task. It should be understood that there may be other situations where the first processing core itself cannot process the current second target task, which is not limited in the present disclosure.

When the above situation occurs in the first processing core, it is determined that the broadcast conditions are met, and the first processing core may send the first broadcast signal to the processing cores of the many-core system in steps S21-S22, and does not receive the feedback signal within the first time replay.

In some embodiments, the processing core that satisfies the feedback condition may send a feedback signal to the first processing core; after receiving the feedback signal, the first processing core may determine whether to determine the feedback signal as the target feedback signal according to the aforementioned processing procedure , the description will not be repeated here.

In some embodiments, when the feedback signal is determined to be the target feedback signal, corresponding processing is performed according to the target feedback signal. That is, the on-chip network communicates with the target processing core that sends the target feedback signal to complete the handshake; and after the handshake is completed, the task data corresponding to the second target task is sent to the target processing core, so that the target processing core executes the first task. Two objective tasks.

In this way, decentralized task scheduling can be achieved, and through dynamic task distribution, the balance between the processing cores in the many-core system can be achieved, and the overall processing efficiency of the system can be improved. In addition, it can automatically broadcast when the first processing core has an operation error, the temperature is too high, etc., so that other processing cores can be used to perform corresponding tasks, which reduces the probability of task errors and improves the temperature uniformity of the many-core system.

Wherein, the broadcast condition includes: the first processing core is in an idle state; the first broadcast signal includes a handshake signal for requesting a task;

pairing with the target processing core corresponding to the target feedback signal to claim the third target task to be allocated in the target processing core,

The third target task is executed when task data corresponding to the third target task is received.

For example, when the first processing core is in an idle state, other processing cores may be actively inquired about whether there are new tasks to be processed. That is, the first processing core can send a broadcast signal to other processing cores in the many-core system to notify other processing cores that the first processing core is in an idle state and can "help" to process tasks to be processed. Wherein, processing the task to be processed may include, for example, storing data, performing calculation processing according to the data, etc., which is not limited in the present disclosure.

In this case, the preset broadcast condition may include: the first processing core is in an idle state. That is, when the first processing core is in an idle state, it is determined that the broadcast conditions are met, and the first broadcast signal can be sent to the processing core of the many-core system in steps S21-S22, and replayed when the feedback signal is not received within the first time. .

In this case, the first broadcast signal includes a handshake signal for requesting a task, that is, a handshake can be performed with other processing cores, so as to "help" other processing cores to process tasks to be processed.

In some embodiments, the first broadcast signal may further include state information of the first processing core.

In some embodiments, the state information includes the occupancy state of the first processing core, that is, whether the first processing core has been "occupied" and is in an idle state that can "help" to process tasks to be processed. When the first processing core is not "occupied" by a task and is in an idle state and can help other processing cores to process tasks to be processed, the occupied state of the first processing core is unoccupied. When the first processing core is in a working state and is processing tasks, and there is no extra computing power and storage space to help other processing cores process tasks to be processed, the occupied state of the first processing core is occupied.

It should be noted that when the first processing core receives feedback signals from other processing cores and decides to "help" process the pending tasks of this processing core, although the first processing core has not yet started to process the pending tasks of this processing core, That is, when the first processing core already has a task to be processed, but has not yet started to process the task, the occupied state of the first processing core is still occupied.

When the broadcast method of the embodiment of the present disclosure is used in the above task scheduling process, the first broadcast signal sent by the first processing core is to notify other processing cores that the first processing core is in an idle state, that is, the first processing core sends When the first broadcast signal is used, it should be in an idle state, and the occupied state of the first broadcast signal should be "unoccupied".

In some embodiments, the state information of the first processing core may further include a computing power state and/or a storage state of the first processing core. The computing power status of the first processing core refers to computing power status information of the first processing core, that is, the computing power available to the first processing core. The storage state of the first processing core refers to storage state information of the first processing core, that is, the storage capacity (or storage space) usable by the first processing core. The computing power state and storage state of the first processing core can be used as a basis for other processing cores to judge whether they meet the conditions for feeding back the first broadcast signal.

For example, after receiving the first broadcast signal sent by the first processing core, if other processing cores have no tasks to be processed, the processing core may directly forward the first broadcast signal to its adjacent processing core. If the processing core has tasks to be processed that need to be assigned to other processing cores for "help" processing, the processing core can determine whether the first processing core Satisfy the requirements for processing pending tasks.

In some embodiments, the computing power and storage space required for processing the task to be processed may be compared with the computing power and storage space available for the first processing core. If the computing power available to the first processing core is greater than the computing power required to process the tasks to be processed, and the storage space available to the first processing core is greater than the storage space required to process the tasks to be processed, it means that the first processing core satisfies the processing requirements. The requirement of the task to be processed means that the processing core satisfies the feedback condition, and the processing core sends a feedback signal to the first processing core.

In some embodiments, the feedback signal sent by the processing core that satisfies the feedback condition to the first processing core may include: the address of the processing core and the address of the first processing core, wherein the address of the processing core serves as the source of the feedback signal The address is convenient for the first processing core to quickly know the location of the processing core that sent the feedback signal after receiving the feedback signal. The address of the first processing core is used as the destination address of the feedback signal. After receiving the feedback signal, other processing cores can forward the feedback signal to the processing core closer to the first processing core according to the address of the first processing core, so that the feedback signal can be Faster to the first processing core.

In some embodiments, after receiving the feedback signal, the first processing core may determine whether to determine the feedback signal as the target feedback signal according to the foregoing processing procedure, and the description will not be repeated here.

In some embodiments, when the feedback signal is determined to be the target feedback signal, corresponding processing is performed according to the target feedback signal. That is, it handshakes with the processing core that sends the feedback signal, communicates with the processing core that sends the feedback signal through the on-chip network, and receives and processes the pending tasks.

By using the broadcast method of the embodiment of the present disclosure to send a broadcast signal, the feedback signal received by the first processing core is more likely to be sent by a processing core that is closer to itself. When the broadcasting method of the embodiment of the present disclosure is used for task scheduling, the first processing core selects and processes the pending task more likely to be the pending task of the processing core that is closer to itself. In this way, the first processing core is receiving pending tasks. When processing tasks, the pending tasks can reach the first processing core faster, which is convenient for the first processing core to process the pending tasks faster. At the same time, the pending tasks occupy the process of reaching the first processing core from the processing core sending the feedback signal. The number of processing cores is also small, and the transmission resources occupied are relatively small.

It should be emphasized that the broadcasting method of the embodiment of the present disclosure is used for the first processing core of the many-core system, which can be realized by directly modifying the control program of the first processing core of the many-core system, that is, the first processing core of the many-core system can Generate broadcast data, broadcast the broadcast data in the form of broadcast signals, the first processing core does not delete the broadcast signal after the broadcast, if it does not receive the feedback signal sent by other processing cores within the preset first time, then again Send a broadcast signal.

According to an embodiment of the present disclosure, the first processing core of the many-core system may serve as the transmitting side of the broadcast signal, transmit the broadcast signal in steps S21-S22, and replay the broadcast signal when no feedback signal is received within the first time. In addition, the first processing core can also serve as the receiving side of the broadcast signal, receive the broadcast signal, give a corresponding feedback signal, and/or forward the broadcast signal.

In the case of receiving the second broadcast signal sent by the second processing core of the many-core system for the kth time, determine the first broadcast signal according to the second broadcast signal and the processing state of the first processing core Whether the processing core satisfies the feedback condition, k is an integer greater than or equal to 1;

In the case that the first processing core satisfies the feedback condition, a third feedback signal is sent to the second processing core, so that the first processing core is paired with the second processing core and performs corresponding processing,

Wherein, the performing corresponding processing includes:

The first processing core receives the task data corresponding to the fourth target task sent by the second processing core, and executes the fourth target task; or

The first processing core sends task data corresponding to the fifth target task to the second processing core, so that the second processing core executes the fifth target task.

For example, if the first processing core receives the second broadcast signal sent by the second processing core for the kth time (k is an integer greater than or equal to 1), the content of the second broadcast signal and the Processing status, to determine whether the first processing core satisfies the feedback condition. For example, the second broadcast signal includes a handshake signal for assigning tasks. If the current processing state or the predicted processing state of the first processing core is idle, it can be determined that the feedback condition is satisfied; otherwise, it can be determined that the feedback condition is not satisfied.

In some embodiments, if the first processing core satisfies the feedback condition, a third feedback signal may be sent to the second processing core, so that the second processing core determines whether to use the third feedback signal as the target feedback signal. If the second processing core takes the third feedback signal as the target feedback signal, the first processing core shakes hands with the second processing core to realize pairing, and performs corresponding processing.

In some embodiments, if the second broadcast signal includes a handshake signal for assigning tasks, performing corresponding processing may include: the first processing core receiving task data corresponding to the fourth target task sent by the second processing core, and Execute the fourth target task; if the second broadcast signal includes a handshake signal for requesting a task, executing the corresponding processing may include: the first processing core sends the task data corresponding to the fifth target task to the second processing core, so that the The second processing core executes the fifth target task.

In this way, the processing procedure when the first processing core acts as the receiving side of the broadcast signal can be implemented, thereby improving the overall processing efficiency of the system.

In some embodiments, the second broadcast signal includes a maximum transmission step size, and the broadcast method according to an embodiment of the present disclosure may further include:

In the case of receiving the second broadcast signal sent by the second processing core for the kth time, according to the maximum transmission step size in the second broadcast signal, determine whether the second broadcast signal satisfies the forwarding condition;

If the second broadcast signal satisfies the forwarding condition, the second broadcast signal is forwarded to a third processing core far away from the second processing core.

That is, when the first processing core receives the second broadcast signal sent by the second processing core for the kth time, it can determine whether the second broadcast signal needs to be forwarded according to the maximum transmission step size in the second broadcast signal.

In some embodiments, if a method is adopted in which the number of processed cores counted in the second broadcast signal is incremented by 1 each time a transfer between processing cores is passed, it is judged that the forwarding is satisfied when the value does not reach the maximum transmission step size. condition, forward the second broadcast signal to the third processing core far away from the second processing core; otherwise, when the value reaches the maximum transmission step size, it is judged that the forwarding condition is not satisfied, and the second broadcast signal is deleted.

In some embodiments, if the maximum transmission step size is reduced by 1 after each transfer between the processing cores, when the maximum transmission step size is greater than 0, it is judged that the forwarding condition is satisfied, and the transfer to the third processing core far away from the second processing core is adopted. The core forwards the second broadcast signal; otherwise, when the maximum transmission step size is 0, it is determined that the forwarding condition is not satisfied, and the second broadcast signal is deleted.

In some embodiments, the distance between the second processing core and the first processing core may be determined if a comparison is used. When the maximum transmission step size is greater than the distance, it is judged that the forwarding condition is satisfied, and the second broadcast signal is forwarded to the third processing core far away from the second processing core; on the contrary, when the maximum transmission step size is equal to the distance, it is judged that the forwarding condition is not satisfied, and the first broadcast signal is deleted. 2. Broadcast signal.

In this way, the transmission range of the broadcast signal can be limited, thereby improving the utilization rate of transmission resources in the system.

In some embodiments, when the first processing core sends the broadcast signal to other processing cores in the many-core system, the first processing core may not send the broadcast signal to all other processing cores in the many-core system, but to the many-core system Multiple alternative processing cores of the system transmit broadcast signals.

Wherein, the candidate processing core refers to the processing core whose number of processing cores passed by the first processing core when sending the broadcast signal to it is less than the predetermined number of processing cores, that is, the processing core whose distance from the first processing core is within the predetermined range is reserved for Select the processing core.

Referring to FIG. 1 , the processing core N0 is the first processing core, and if the predetermined number of processing cores is 1, the alternative processing core refers to the processing core adjacent to the processing core N0; if the predetermined number of processing cores is 2, the alternative processing core The processing cores refer to the processing cores adjacent to the processing core N0 and the processing cores adjacent to the adjacent processing cores of the processing core N0.

Since the broadcast is to send data to all possible receiving processing cores as much as possible, it has no destination. In the process of the first processing core sending the broadcast signal to other processing cores of the on-chip network, if there is no restriction, let the signal have no destination. If the broadcast is stopped, the broadcast process will only stop when the broadcast signal has traversed all the processing cores of the on-chip network.

If the first processing core receives the feedback signal sent by the alternative processing core within the preset first time, it can end the broadcasting process and process the feedback signal (for example, the first processing core sends the feedback signal to the processing of the feedback signal according to the received feedback signal) The core sends a connect signal, which is used to indicate that the first processing core has received the sent feedback signal).

If the first processing core does not receive the feedback signal sent by the alternative processing core within the preset first time, the first processing core sends a broadcast signal to the alternative processing core of the many-core system again.

In some embodiments, the first processing core sends a broadcast signal to other processing cores of the many-core system. The broadcast signal may specifically be a handshake signal. Handshake signals are signals for communication between different devices (or devices).

For example, the handshake signal in this embodiment of the present disclosure may be a handshake signal used for task scheduling. For example, when the first processing core is in an idle state, the first processing core sends a handshake signal to other processing cores of the on-chip network to notify Other processing cores, the first processing core is in an idle state and can "help" to process signals to be processed. When other processing cores have tasks to be processed, the processing core sends a feedback signal to the first processing core, so that the processing core Handshake with the first processing core, so that the first processing core can "help" process the pending task.

In some embodiments, if the first processing core does not receive feedback signals sent by other processing cores within a preset first time, the first processing core sends a broadcast signal to other processing cores in the many-core system again.

In some embodiments, if the first processing core receives feedback signals sent by other processing cores within a preset first time, the first processing core may end the broadcasting process when receiving the first feedback signal , take the feedback signal as the target feedback signal, and process the target feedback signal.

In some embodiments, after the target feedback signal is determined, the first processing core may send the broadcast signal to other processing cores again. Different from the previous broadcast signal, after the target feedback signal is determined, the first processing core sends the broadcast signal. The occupancy status of the first processing core in the broadcast signal is "occupied". By sending a broadcast signal whose occupancy status is "occupied", other processing cores can be notified that the first processing core has been occupied, and there is no need to report to the first processing core. Send feedback.

After determining the target feedback signal, the first processing core may also send a connection signal to the processing core that sent the target feedback signal, where the connection signal is used to indicate that the first processing core has received the feedback signal sent by the first processing core.

In some embodiments, after receiving the feedback signal for the first time, the first processing core may also receive other feedback signals, and in the case of receiving other feedback signals, the first processing core ignores the feedback signal.

Referring to FIG. 1 , the processing core N0 is the first processing core, and both the processing core N1 and the processing core N3 send feedback signals to the processing core N0, but since the processing core N1 is closer to the processing core N0, the feedback signal sent by the processing core N1 The signal arrives at the processing core N0 preferentially. After receiving the feedback signal from the processing core N1 for a period of time, the processing core N0 will also receive the feedback signal sent by the processing core N3. After receiving the feedback signal from the processing core N1, the processing core N0 automatically The feedback signal sent by the processing core N3 is ignored or deleted.

In some embodiments, the broadcast signal is a handshake signal. After the first processing core receives the feedback signal for the first time, the first processing core shakes hands with the processing core corresponding to the processing core (ie, the processing core that sends the feedback signal). After a processing core receives a feedback signal sent by other processing cores after the handshake is successful, the first processing core cannot process the feedback signal. That is to say, it is meaningless for the first processing core to receive the feedback signal after the handshake is successful. It will only result in a waste of resources.

In some embodiments, the broadcasting method of the embodiments of the present disclosure can also be used in the case where the first processing core has a new task or a task that cannot be completed by itself needs to be processed by other processing cores: the first processing core has a new task or cannot complete it by itself When the task needs to be processed by other processing cores, the first processing core broadcasts the handshake signal to other processing cores, and when the processing core that receives the handshake signal has the ability to process the task, the processing core sends feedback to the first processing core. signal, the processing core shakes hands with the first processing core, receives and processes the task issued by the first processing core.

Similarly, if the first processing core handshakes with the processing core that sent the target feedback signal, and then receives feedback signals sent by other processing cores, it may cause tasks to be repeatedly issued to multiple processing cores, resulting in a waste of computing power and storage resources of waste.

In some embodiments, while the first processing core is in an idle state, other processing cores may be actively inquired whether there are new tasks to be processed. In this case, the first processing core sends a broadcast signal to other processing cores in the many-core system to notify other processing cores that the first processing core is in an idle state and can "help" to process tasks to be processed.

The broadcast signal may include the occupancy state of the first processing core, that is, whether the first processing core has been "occupied", and whether it is in an idle state that can "help" to process tasks to be processed.

The first processing core sends a broadcast signal to notify other processing cores that the first processing core is in an idle state, that is, when the first processing core broadcasts a broadcast signal, its corresponding processing core should be in an idle state, and the broadcast signal is occupied. Should be "unoccupied".

The state information of the first processing core may further include a computing power state and/or a storage state of the first processing core. The computing power status of the first processing core refers to computing power status information of the first processing core, that is, the computing power available to the first processing core. The storage state of the first processing core refers to storage state information of the first processing core, that is, the storage capacity (or storage space) usable by the first processing core.

In some embodiments, if the first processing core receives the feedback signal for the first time within a preset first time, the first processing If no other feedback signal is received by the core, the first processing core takes the received feedback signal as the target feedback signal, and processes the target feedback signal.

The second time is less than the first time, that is, the time between when the first processing core receives the feedback signal for the first time and when the first processing core stops receiving the feedback signal is less than when the first processing core sends the broadcast signal to the first processing core again. The time between sending broadcast signals.

In some embodiments, if the first processing core receives the feedback signal for the first time within the preset first time, if the first processing core receives the feedback signal for the first time within the second time If the processing core receives other feedback signals, the first processing core starts from all the feedback signals received (including the feedback signal received for the first time and from the feedback signal received for the first time to the end of the second time). feedback signal), determine a feedback signal as the target feedback signal, and process the target feedback signal.

Wherein, determining a feedback signal as the target feedback signal from all the received feedback signals may specifically be: determining the position of the processing core that sends the feedback signal from the received feedback signals, and calculating the difference between the processing core and the first processing core The processing core with the shortest distance from the first processing core is selected from all the processing cores corresponding to the received feedback signals as the target processing core, and the feedback signal sent by the processing core is the target feedback signal.

The distance between a certain processing core and the first processing core specifically refers to the step size between the processing core and the first processing core, that is, the length of the shortest path that the signal passes through from the processing core to the first processing core. The processing core with the shortest processing core distance is the target processing core. That is to say, determine the processing core from which all received feedback signals come from, and determine the processing core that is closest to the first processing core among these processing cores (or at the same transmission speed, the sent signal can reach the first processing core the fastest The processing core of the core) is the target processing core, and the feedback signal sent by the processing core is the target feedback signal.

In some embodiments, when the first processing core does not determine the target feedback signal within the preset first time, the first processing core still needs to send the broadcast signal (the occupancy state of the broadcast signal is still unoccupied).

The first processing core does not determine the target feedback signal within the preset first time, it may be that the first processing core does not receive the feedback signal within the preset first time; it may also be that the first processing core is within the preset first time. Multiple feedback signals are received within the first time, but the time when the first processing core receives the feedback signal for the first time is too close to the preset first time, so that the first time is located since the first time the first processing core received the feedback signal During the second time since the feedback signal, the first processing core cannot determine whether other feedback signals will be received between the first time and the second time since the first processing core first received the feedback signal, so The first processing core cannot determine the target feedback signal.

After the first processing core receives the feedback signal for the first time, it does not immediately use the feedback signal as the target feedback signal, but continues to wait for a period of time, and continues to receive the feedback signal during this period, so that the broadcast signal is sent last time. When a broadcast signal is received during the process, the corresponding processing core does not meet the conditions for feeding back the broadcast signal, and when a broadcast signal is received during the process of sending the broadcast signal, the corresponding processing core meets the conditions for feeding back the broadcast signal. There may be more time to send the feedback signal to the first processing core. Meanwhile, after receiving multiple feedback signals, the first processing core selects the feedback signal sent by the processing core closest to the first processing core as the target feedback signal. The two-pronged approach improves the probability that a processing core closer to the first processing core will be the target processing core.

Referring to FIG. 1 , the processing core N0 is the first processing core, the processing core N3 sends a feedback signal to the processing core N0 according to the broadcast signal transmitted at the Lth time, and the processing core N1 sends a feedback signal to the processing core according to the broadcast signal broadcast at the L+1th time. N0 sends a feedback signal.

Since the processing core N3 sends the feedback signal earlier than the processing core N1 sends the feedback signal, although the processing core N1 is closer to the processing core N0, the feedback signal sent by the processing core N3 may still arrive at the processing core N0 before the feedback signal sent by the processing core N1.

If the processing core N0 directly uses the processing core N3 as the target processing core after receiving the feedback signal from the processing core N3, the feedback signal sent by the processing core N1 will also be ignored by the processing core N0 after reaching the processing core N0, which will cause Although the processing core N1 is closer to the processing core N0 and satisfies the conditions for feeding back the broadcast signal, the processing core N0 cannot select the processing core N1 as the target processing core.

If the processing core N0 waits for a period of time after receiving the feedback signal from the processing core N3, within the waiting time, the processing core N0 may receive the feedback signal sent by the processing core N1, and the distance between the processing core N1 and the processing core N0 is obviously The distance between the processing core N3 and the processing core N0 is closer, and the processing core N0 selects the processing core N1 as the target processing core.

Since the distance between the processing core N1 and the processing core N0 is obviously closer than the distance between the processing core N3 and the processing core N1, the communication speed between the processing core corresponding to the processing core N1 and the processing core N0 is obviously higher than that of the processing core N3 Regarding the communication speed between the corresponding processing core and the processing core corresponding to the processing core N0, selecting the processing core N1 as the target processing core is obviously a better choice than selecting the processing core N3 as the target processing core.

According to an embodiment of the present disclosure, there is also provided a broadcasting apparatus based on a many-core system, wherein the many-core system includes a plurality of processing cores, and the plurality of processing cores are connected through an on-chip network communication.

FIG. 3 is a block diagram of a broadcast apparatus based on a many-core system according to an embodiment of the present disclosure. As shown in Figure 3, the device includes:

A broadcast module 31, configured to send a first broadcast signal to the processing core of the many-core system for the first time when a preset broadcast condition is met, wherein the first broadcast signal includes a maximum transmission step size;

The replay module 32 is configured to, within the first time after the i-th transmission of the first broadcast signal, in the case of not receiving the first feedback signal for the first broadcast signal, to the The processing core sends the first broadcast signal for the i+1th time, where i is an integer greater than or equal to 1.

Wherein, the broadcasting apparatus is applied to the processing core of the many-core system or the external electronic equipment of the many-core system.

According to an embodiment of the present disclosure, a signal transmission method for a many-core system is also provided. The method is used in the many-core system provided by the embodiment of the present disclosure, that is, one or more processing cores in the processing core of the many-core system are the first one. Processing cores that broadcast signals using the broadcast method provided by the embodiments of the present disclosure.

The many-core system signal transmission method according to the embodiment of the present disclosure includes:

S801. The first processing core sends a broadcast signal to other processing cores.

A certain first processing core of the many-core system sends a broadcast signal to other processing cores of the many-core system. The broadcasting method of the embodiment of the present disclosure may be used for the first processing core to send the broadcast signal to other processing cores, that is, the first processing core sends the broadcast signal again if it does not receive the feedback signal within the preset first time.

The other processing cores of the many-core system refer to other processing cores other than the first processing core that sends the broadcast signal, that is, during the execution of the signal transmission method of the many-core system according to the embodiment of the present disclosure, only one processing core is the first processing core. , even if other processing cores of the many-core system have the capability of executing the broadcasting method of the embodiment of the present disclosure, they are still other processing cores during the execution of the signal transmission method of the many-core system.

S802. At least one second processing core receives a broadcast signal, and the second processing core that satisfies the condition for feeding back the broadcast signal sends a feedback signal to the first processing core.

After receiving the broadcast signal, the other processing cores (ie the second processing core) of the many-core system judge whether the processing core or the processing core satisfies the feedback broadcast signal according to the actual state of the processing core or the processing core and the broadcast signal. condition, and send a feedback signal to the first processing core when the processing core or the actual state of the processing core satisfies the feedback broadcast signal condition.

The second processing core refers to the processing core that receives the broadcast signal, that is, the second processing core does not specifically refer to a certain processing core, and all the processing cores that receive the broadcast signal can be called the second processing core. Also, not all the second processing cores send feedback signals to the first processing cores after receiving the broadcast signals, and only the second processing cores that satisfy the conditions for feeding back the broadcast signals send the feedback signals to the first processing cores.

The many-core system signal transmission method according to the embodiment of the present disclosure can be used for resource scheduling.

For example, when the first processing core has free resources (such as computing resources, storage resources, etc.) that can "help" other processing cores to process tasks to be processed, the first processing core provides other processing cores (ie, the second processing core) to the many-core system. The processing core) sends a broadcast signal, and the broadcast signal is used to notify other processing cores in the many-core system that the first processing core has idle resources and can "help" to process tasks to be processed.

The broadcast signal sent by the first processing core may include state information of the first processing core. The state information of the first processing core is essentially the state information of the first processing core.

The state information of the first processing core includes the occupancy state of the first processing core, that is, whether the first processing core has been "occupied" and whether it is in an idle state that can "help" to process tasks to be processed. When the first processing core is not "occupied" by a task and is in an idle state, and can help the second processing core to process tasks to be processed, the occupied state of the first processing core is unoccupied. When the first processing core is in a working state and is processing tasks, and there is no extra computing power and storage space to help the second processing core process tasks to be processed, the occupied state of the first processing core is occupied.

It should be noted that, when the first processing core decides to "help" process the pending tasks of a certain second processing core, although the first processing core has not yet started to process the pending tasks of the processing core, the occupation of the first processing core The status is still occupied.

When the signal transmission method of the many-core system according to the embodiment of the present disclosure is used for task scheduling, the first processing core sends a broadcast signal to notify other processing cores that the first processing core is in an idle state, that is, at this time, the broadcast signal of the first processing core is in an idle state. The occupancy status of the broadcast signal shall be "unoccupied".

One or more second processing cores of the many-core system receive the broadcast signal sent by the first processing core, and the second processing cores satisfying the condition for feeding back the broadcast signal send the feedback signal to the first processing core.

The broadcast signal broadcast by the first processing core includes state information of the first processing core, and after receiving the broadcast signal broadcast by the first processing core, the second processing core can judge whether it satisfies the feedback condition according to the state information of the first processing core .

Wherein, after the second processing core receives the broadcast signal sent by the first processing core, the second processing core first determines whether it has tasks to be processed that need the assistance of the first processing core. If a task is to be processed, the second processing core can directly forward the broadcast signal to the adjacent second processing core, so that the broadcast signal can continue to be "broadcast".

If the second processing core has a pending task that needs to be assigned to other processing cores for "help" processing, the second processing core determines whether the first processing core can process the pending task according to the state information of the first processing core.

Wherein, when the status information of the first processing core includes the computing power status and/or storage status of the first processing core, the second processing core determines whether the first processing core can process the first processing core according to the status information of the first processing core The to-be-processed task may include: the second processing core judging whether the computing power state of the first processing core satisfies the condition for processing the to-be-processed task. And/or, the second processing core determines whether the storage state of the first processing core satisfies the condition for processing the to-be-processed task.

That is, the second processing core judges whether the first processing core meets the requirements for processing tasks to be processed according to the computing power status and storage status of the first processing core in the state information of the first processing core. Specifically, the second processing core can process The computing power and storage space required by the task to be processed are compared with the computing power and storage space that can be used by the first processing core.

If the computing power available to the first processing core is less than the computing power required to process the tasks to be processed, or the storage space available to the first processing core is less than the storage space required to process the tasks to be processed, it means that the first processing core does not satisfy the If the request for processing the pending task cannot be processed, it means that the second processing core does not meet the condition for feeding back the broadcast signal. Then, the second processing core can continue to forward the broadcast signal to the adjacent second processing core, so that the broadcast signal can continue to be "broadcast".

If the computing power available to the first processing core is greater than the computing power required to process the tasks to be processed, and the storage space available to the first processing core is greater than the storage space required to process the tasks to be processed, it means that the first processing core satisfies the processing requirements. The requirement of the to-be-processed task and the ability to process the to-be-processed task means that the second processing core satisfies the feedback broadcast signal condition, and the second processing core that satisfies the feedback broadcast signal condition sends a feedback signal to the first processing core.

Since the feedback signal is sent to the first processing core, the feedback signal is purposeful, so the feedback signal can be sent from the second processing core that satisfies the feedback broadcast signal condition to the first processing core in a unicast manner.

Before the feedback signal is sent from the second processing core satisfying the feedback broadcast signal condition to the first processing core in a unicast manner, the feedback signal needs to be processed. The processing process specifically includes: the second processing core satisfying the feedback broadcast signal condition sends the The address of the first processing core is used as the destination address to write the feedback signal. The second processing core that satisfies the condition for feeding back the broadcast signal writes the address of the second processing core into the feedback signal as a source address.

In some embodiments, the address of the first processing core is written into the feedback signal as the destination address of the feedback signal. During the process of sending the feedback signal from the second processing core to the first processing core, the processing core that receives the feedback signal may The feedback signal is forwarded to the processing core closer to the first processing core according to the destination address in the feedback signal. Compared with broadcasting without a destination address, this forwarding method not only saves transmission resources, but also allows the feedback signal to pass through the shortest path quickly. to the first processing core.

The address of the second processing core that sends the feedback signal is written into the feedback signal as the source address of the feedback signal. After the feedback signal reaches the first processing core, the first processing core can obtain the first processing core sent to the feedback signal through the source address of the feedback signal. The address of the second processing core.

In the many-core system, there may be multiple second processing cores that satisfy the feedback condition, and the multiple second processing cores that satisfy the feedback condition all send feedback signals to the first processing core.

The distances between the multiple second processing cores that satisfy the feedback condition and the first processing core may be different, which results in that the feedback signals sent by the multiple second processing cores that satisfy the feedback condition do not reach the first processing core at the same time.

In a special case, that is, the first processing core does not receive the feedback signal within the preset first time, and the first processing core sends the broadcast signal again. The processing core that feeds back the broadcast signal also sends feedback signals to the first processing core, so that the first processing core may receive more feedback signals.

The first processing core may, based on the priority principle, only receive the first feedback signal that reaches the first processing core, take the first feedback signal that reaches the first processing core as the target feedback signal, and will send the second processing core of the feedback signal. as the target second processing core.

The first processing core may also wait for a period of time after receiving the first feedback signal that arrives, that is, within the second time since the first processing core receives the feedback signal for the first time, the first processing core continues to receive and send the The feedback signal of the first processing core, and all the feedback signals received during this period (including the feedback signal received for the first time and the feedback signal received within the second time since the feedback signal was first received ), determine a feedback signal as the target feedback signal.

Wherein, determining one feedback signal as the target feedback signal from all the received feedback signals may specifically be: determining the position of the second processing core that sends the feedback signal according to the source address of the feedback signal from the received feedback signals, and calculating the The distance between the second processing core and the first processing core, select the processing core with the shortest distance from the first processing core from all the second processing cores corresponding to the received feedback signals as the target processing core, and the feedback signal sent by the processing core It is the target feedback signal.

The distance between the second processing core and the first processing core specifically refers to the step size between the second processing core and the first processing core, that is, the length of the shortest path that the signal travels from the second processing core to the first processing core, Determine the second processing core with the shortest distance from the first processing core as the target processing core. That is, determine the second processing core from which all received feedback signals come, and determine the processing core that is closest to the first processing core among these second processing cores. The core (or in other words, at the same transmission speed, the processing core of which the transmitted signal can reach the first processing core the fastest) is the target second processing core, and the feedback signal sent by the second processing core is the target feedback signal.

It should be emphasized that the second time is shorter than the first time, that is, the time between when the first processing core receives the feedback signal for the first time and when the first processing core stops receiving the feedback signal is shorter than the time between when the first processing core sends the broadcast signal to the first processing core. The time at which the processing core retransmits the broadcast signal interval.

After the first processing core determines the target feedback signal, the first processing core may send the broadcast signal again, and the occupied state of the broadcast signal is occupied.

The first processing core determines the target feedback signal, which is equivalent to the first processing core deciding to “help” the second processing core sending the target feedback signal to process the pending task. The first processing core has no extra resources to “help” other processing cores, so After the first processing core determines the target feedback signal, the occupied state of the broadcast signal broadcast by the first processing core should be occupied.

The broadcast signal is used to notify other processing cores of the network-on-chip that the first processing core is already occupied, so there is no need to send a feedback signal to the first processing core to avoid waste of transmission resources. After receiving the broadcast signal, other processing cores can know that The first processing core is already occupied, and no feedback signal will be sent to the first processing core.

After the first processing core determines the target feedback signal, the first processing core may also send a connection signal to the target second processing core, where the connection signal is used to indicate that the first processing core has received the feedback signal sent by the target second processing core.

After receiving the connection signal, the target second processing core can send the task to be processed to the first processing core, and the first processing core receives the task to be processed and processes it.

After the first processing core determines the target feedback signal, other feedback signals may also be received. If other feedback signals are received after the target feedback signal is determined, the received feedback signal may be ignored.

In the signal transmission method for a many-core system according to the embodiment of the present disclosure, the first processing core resends the broadcast signal if the feedback signal is not received within the preset first time, and at the same time, the first processing core receives the feedback signal for the first time. After the feedback signal is received, the feedback signal is not used as the target feedback signal immediately, but it continues to wait for a period of time, and continues to receive feedback signals during this period. If multiple feedback signals are received, select the one closest to the first processing core. The feedback signal sent by the second processing core is used as the target feedback signal, which increases the probability of the second processing core closer to the first processing core as the target second processing core, so that the first processing core is processing the target second processing core. During the task to be processed, when it is necessary to communicate with the target second processing core, the communicated data can reach the target second processing core faster, which improves the efficiency of the first processing core and reduces the occupation of the communication process. transmission resources.

4, an embodiment of the present disclosure provides an electronic device, the electronic device includes multiple processing cores 901 and an on-chip network 902, wherein the multiple processing cores 901 are all connected to the on-chip network 902, and the on-chip network 902 is used for interacting multiple data between cores and external data.

The on-chip network 902 is an on-chip network in the above-mentioned many-core system, and each processing core 901 corresponds to one processing core of the many-core system.

In some embodiments, at least a portion of the first processing core of the many-core system corresponds to processing core 901 .

In addition, an embodiment of the present disclosure also provides a computer-readable medium on which a computer program is stored, wherein the computer program implements the above-mentioned broadcasting method when executed by a processing core of a many-core system.

In addition, embodiments of the present disclosure also provide a computer program product, comprising computer-readable codes, or a non-volatile computer-readable storage medium carrying computer-readable codes, wherein, when the computer-readable codes are stored in an electronic When running in the processor of the device, the processor in the electronic device executes the above-mentioned broadcasting method.

Those of ordinary skill in the art can understand that all or some of the steps in the methods disclosed above, functional modules/units in the systems, and devices can be implemented as software, firmware, hardware, and appropriate combinations thereof. In a hardware implementation, the division between functional modules/units mentioned in the above description does not necessarily correspond to the division of physical components; for example, one physical component may have multiple functions, or one function or step may be composed of several physical components Components execute cooperatively. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor or microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit . Such software may be distributed on computer-readable media, which may include computer storage media (or non-transitory media) and communication media (or transitory media). As is known to those of ordinary skill in the art, the term computer storage media includes both volatile and nonvolatile implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data flexible, removable and non-removable media. Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cartridges, magnetic tape, magnetic disk storage or other magnetic storage devices, or may Any other medium used to store desired information and which can be accessed by a computer. In addition, communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and can include any information delivery media, as is well known to those of ordinary skill in the art .

Example embodiments have been disclosed herein, and although specific terms are employed, they are used and should only be construed in a general descriptive sense and not for purposes of limitation. In some instances, it will be apparent to those skilled in the art that features, characteristics and/or elements described in connection with a particular embodiment may be used alone or in combination with other embodiments, unless expressly stated otherwise. Features and/or elements are used in combination. Accordingly, it will be understood by those of ordinary skill in the art that various changes in form and details may be made therein without departing from the scope of the present disclosure as set forth in the appended claims.

Claims

A broadcasting method based on a many-core system, the many-core system includes a plurality of processing cores, and the plurality of the processing cores are communicated and connected through an on-chip network, the method comprising:

In the case that the preset broadcast condition is satisfied, send a first broadcast signal to the processing core of the many-core system for the first time, wherein the first broadcast signal includes a maximum transmission step size;

If the first feedback signal for the first broadcast signal is not received within the first time after the i-th transmission of the first broadcast signal, send the message to the processing core of the many-core system for the i+1th The first broadcast signal is sent for times, and i is an integer greater than or equal to 1.
The method of claim 1, wherein the method further comprises:

If a first feedback signal for the first broadcast signal is received within a first time after the i-th transmission of the first broadcast signal, use the first feedback signal as a target feedback signal;

Corresponding processing is performed according to the target feedback signal.
The method of claim 1, wherein the method further comprises:

In the case of receiving a first feedback signal for the first broadcast signal within a first time after sending the first broadcast signal for the i-th time, buffering the first feedback signal;

If the second feedback signal for the first broadcast signal is not received within a second time after the first feedback signal is received, the first feedback signal is used as the target feedback signal;

Corresponding processing is performed according to the target feedback signal.
The method of claim 3, wherein the method further comprises:

In the case where a second feedback signal for the first broadcast signal is received within a second time after the first feedback signal is received, from all feedback signals received up to the end of the second time Determine the target feedback signal; wherein, the target processing core that sends the target feedback signal is the processing core with the smallest signal transmission step;

Corresponding processing is performed according to the target feedback signal.
The method according to any one of claims 2-4, wherein the method is applied to a first processing core of the many-core system, or an electronic device external to the many-core system, the first processing core is one of the plurality of processing cores,

Wherein, the broadcast condition includes: there is a first target task to be assigned; the first broadcast signal includes a handshake signal for assigning a task;

The performing corresponding processing according to the target feedback signal includes:

Sending task data corresponding to the first target task to the target processing core corresponding to the target feedback signal, so that the target processing core executes the first target task.
The method according to any one of claims 2-4, wherein the method is applied to a first processing core of the many-core system, and the first processing core is one of a plurality of the processing cores,

The broadcast condition includes any one of the following: a second target task to be allocated exists in the first processing core, an operation error occurs during the processing of the second target task by the first processing core, the first processing the temperature of the core during processing of the second target task exceeds a temperature threshold; the first broadcast signal includes a handshake signal for assigning tasks;

The performing corresponding processing according to the target feedback signal includes:

Sending task data corresponding to the second target task to the target processing core corresponding to the target feedback signal, so that the target processing core executes the second target task.
The method according to any one of claims 2-4, wherein the method is applied to a first processing core of the many-core system, and the first processing core is one of a plurality of the processing cores,

Wherein, the broadcast condition includes: the first processing core is in an idle state; the first broadcast signal includes a handshake signal for requesting a task;

The performing corresponding processing according to the target feedback signal includes:

pairing with the target processing core corresponding to the target feedback signal to claim the third target task to be allocated in the target processing core,

The third target task is executed when task data corresponding to the third target task is received.
The method of claim 1, wherein the method is applied to a first processing core of the many-core system, the first processing core being one of a plurality of the processing cores, the method further comprising:

In the case of receiving the second broadcast signal sent by the second processing core of the many-core system for the kth time, determine the first broadcast signal according to the second broadcast signal and the processing state of the first processing core Whether the processing core satisfies the handshake condition, k is an integer greater than or equal to 1;

In the case that the first processing core satisfies the handshake condition, a third feedback signal is sent to the second processing core, so that the first processing core is paired with the second processing core and performs corresponding processing,

Wherein, the performing corresponding processing includes:

The first processing core receives the task data corresponding to the fourth target task sent by the second processing core, and executes the fourth target task; or

The first processing core sends task data corresponding to the fifth target task to the second processing core, so that the second processing core executes the fifth target task.
The method according to claim 8, wherein the second broadcast signal includes a maximum transmission step size, and the method further comprises:

In the case of receiving the second broadcast signal sent by the second processing core for the kth time, according to the maximum transmission step size in the second broadcast signal, determine whether the second broadcast signal satisfies the forwarding condition;

If the second broadcast signal satisfies the forwarding condition, the second broadcast signal is forwarded to a third processing core far away from the second processing core.
An electronic device comprising:

multiple processing cores; and

an on-chip network configured to exchange data and external data among the plurality of processing cores;

Wherein, instructions are stored in the processing core, and the instructions are executed by the processing core, so that the processing core can execute the broadcasting method according to any one of claims 1-9.
A computer-readable medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the broadcasting method of any one of claims 1-9.
A computer program product comprising computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code, wherein when the computer-readable code is executed in a processor of an electronic device, all The processor in the electronic device executes the broadcasting method for implementing any one of claims 1-9.