Task scheduling method for computing resource collaborative management of heterogeneous multi-core processor

By designing CMCR components on heterogeneous multi-core processors, using the collaborative management of ARM core and DSP core, the problems of complexity of task scheduling strategies and low resource utilization are solved, efficient task processing and resource management are achieved, and the system is improved real-time and development efficiency.

CN118916148APending Publication Date: 2024-11-08BEIJING INST OF COMP TECH & APPL
View PDF 0 Cites 5 Cited by

Patent Information

Application Number
CN202411233771.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-04
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The task scheduling strategies of existing heterogeneous multi-core processors have problems such as high programming complexity and insufficient resource utilization, which limits their full use of applications and performance in a wider range of scenarios.

Method used

Design a CMCR component, including the main control core component, the computing core component and the multi-core communication component. The ARM core deploys a real-time operating system, and realizes seamless communication and resource management between the ARM core and the DSP core through the multi-core communication component, intelligently implements the Tasks are allocated to the most suitable DSP core, and the shared memory and inter-core interrupt mechanisms are used to optimize task execution and resource utilization.

Benefits of technology

It realizes the immediacy and stability of task processing, maximizes the utilization of computing resources, reduces development complexity, improves resource utilization and system real-timeness, and ensures timely task response and optimal resource allocation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118916148A_ABST
    Figure CN118916148A_ABST
Patent Text Reader

Abstract

The invention relates to a task scheduling method for computing resource collaborative management of a heterogeneous multi-core processor, and belongs to the field of heterogeneous scheduling of embedded operating systems. The invention designs and realizes an efficient CMCR component. The efficient CMCR component comprises a main control core assembly, a calculation core assembly and a multi-core communication assembly. A real-time operating system is deployed in an ARM core, a main control core assembly is designed to manage a DSP calculation core, the instantaneity and stability of task processing are ensured, the heterogeneous resource advantage of HMP is fully utilized, and calculation resources are utilized to the maximum extent; the ARM core provides a calculation task affinity service interface for a user, so that the programming efficiency is improved, and the complexity and difficulty of application development under the HMP are reduced; through the design of a multi-core communication component, rapid data transmission is realized by using inter-core interruption and a shared memory mechanism, and data between different processor cores is mapped to an exclusive address space, so that resource access conflicts are avoided. According to the invention, optimization of task scheduling is realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of heterogeneous scheduling of embedded operating systems, and in particular relates to a task scheduling method for collaborative management of computing resources of heterogeneous multi-core processors. Background Art

[0002] Heterogeneous multi-core processors (HMPs) integrate various types of processor cores with different characteristics. These cores have their own advantages in computing power, control functions, etc. By working together, they can not only effectively cope with the increasingly complex and changing performance requirements of applications, but also show significant advantages in energy efficiency. However, the complexity and diversity of the HMP architecture also bring unprecedented challenges to its software task scheduling strategy. Hardware design, as the cornerstone, needs to be carefully planned to support diverse application scenarios; and the task scheduling technology at the software level becomes the key. It requires the ability to efficiently and intelligently assign tasks to the most suitable processor cores to give full play to the heterogeneous resource advantages of HMP, maximize the use of each computing resource, and achieve a double leap in performance and efficiency. Therefore, in the development process of HMP, continuously optimizing the software scheduling strategy will be an important driving force to push its performance limit and broaden the application boundary.

[0003] Early software was in the form of bare metal. HMP did not run an operating system. The application was artificially split into multiple modules and distributed to different cores of HMP using a static binding mode. Each core executed independently.

[0004] With the promotion of embedded operating system applications, HMP runs more operating systems, and scheduling algorithms become the focus of research. The application is split into multiple tasks, and multiple tasks are scheduled on HMP through scheduling algorithms. The existing scheduling algorithms are subdivided into two categories: independent task scheduling algorithms and non-independent task scheduling algorithms.

[0005] For independent task scheduling algorithms, there are two main strategies: online mode and batch mode. The scheduling algorithm in online mode focuses on real-time response and resource optimization; while the scheduling algorithm in batch mode focuses more on global optimization and resource balancing, and achieves overall performance improvement of the task set through batch processing and intelligent allocation.

[0006] Non-independent task scheduling algorithms are also divided into two categories: deterministic and non-deterministic. Deterministic scheduling algorithms are known for their predictability and efficiency. Through precise task division and scheduling strategies, they ensure the timing of task execution and maximize resource utilization. On the other hand, non-deterministic scheduling algorithms simulate the optimization mechanism of nature or physical processes to explore better scheduling solutions in a probabilistic way, showing strong flexibility and optimization capabilities when dealing with complex scheduling problems.

[0007] Without an embedded operating system, the HMP architecture faces significant challenges: although each core can execute independently, task allocation is limited to a static binding mode, making it difficult to achieve dynamic load balancing. This rigid allocation not only requires developers to intervene deeply and manually bind tasks to specific cores, increasing development complexity and the risk of errors, but also results in a significant waste of computing resources due to the inability to flexibly respond to changes in system requirements.

[0008] In the embedded operating system environment, although HMP technology shows great potential, it needs to rely on heterogeneous scheduling algorithms. Although independent task scheduling algorithms on heterogeneous multi-cores are flexible and can be dynamically adjusted according to system resource conditions and task characteristics, these algorithms often face the problem of unsatisfactory resource utilization, especially in the processing of emergency tasks, which may seem incapable, resulting in idleness and waste of precious computing resources; in contrast, non-independent task scheduling algorithms significantly improve resource utilization through more sophisticated resource management and task execution order optimization. However, this type of algorithm is also accompanied by a significant increase in computational complexity and an increase in the difficulty of parameter configuration, which to a certain extent limits their application in environments that require high real-time performance. Summary of the invention

[0009] 1. Technical issues to be resolved

[0010] The technical problem to be solved by the present invention is how to provide a task scheduling method for collaborative management of computing resources of heterogeneous multi-core processors to solve the problem that current scheduling strategies have certain limitations, which lead to high programming complexity and insufficient utilization of processor resources for HMP, restricting the application and promotion of HMP technology in a wider range of scenarios and limiting the full utilization of its effectiveness.

[0011] (II) Technical solution

[0012] In order to solve the above technical problems, the present invention proposes a task scheduling method for collaborative management of computing resources of heterogeneous multi-core processors, designs and implements an efficient CMCR component, including: a main control core component, a computing core component and a multi-core communication component, the main control core component is a main control core CMCR component, the computing core component is a computing core CMCR component, and the multi-core communication component is a CMCR multi-core communication component;

[0013] Deploy an embedded real-time operating system on the ARM core while keeping the DSP core in bare metal operation mode. The main control core component runs on the ARM core, the computing core component runs on the DSP core, and the multi-core communication component runs on the ARM core and the DSP core.

[0014] The DSP core and the ARM core are started in parallel. The DSP core completes the initialization of the computing core component first and enters the standby state, waiting for the task scheduling instruction of the ARM core. At the same time, the ARM core starts and loads the embedded real-time operating system to execute the initialization process of the main control core component. Once the main control core component is ready, the entire CMCR component starts working.

[0015] When a computing task request is triggered at the application layer, the embedded real-time operating system intelligently retrieves available DSPs from the DSP free resource pool, writes the task data securely into the shared memory area through the multi-core communication component, and then triggers an inter-core interrupt to wake up the standby DSP to process the task;

[0016] After receiving the interrupt signal, the DSP immediately extracts the task data from the shared memory area accurately and updates its own status synchronously. After the computing core component is efficiently executed, the multi-core communication component stores the processing results back to the shared memory and triggers an inter-core interrupt to notify the embedded real-time operating system that the task has been completed, thus achieving a seamless task execution and status feedback loop.

[0017] If the DSP idle resources are insufficient, the embedded real-time operating system will temporarily store the pending tasks in the DSP task queue and continuously monitor the resource status. Once it detects that the DSP resources are released, it will immediately schedule the task execution from the queue.

[0018] (III) Beneficial effects

[0019] The present invention proposes a task scheduling method for collaborative management of computing resources of heterogeneous multi-core processors. The present invention proposes a CMCR mechanism, which has the following advantages:

[0020] 1. Deploy a real-time operating system on the ARM core and design the main control core component to manage the DSP computing core to ensure the immediacy and stability of task processing, fully utilize the heterogeneous resource advantages of HMP, and maximize the use of computing resources;

[0021] 2. ARM core provides computing task affinity service interface for users, which improves programming efficiency and reduces the complexity and difficulty of application development under HMP;

[0022] 3. The design of multi-core communication components uses inter-core interrupts and shared memory mechanisms to achieve fast data transmission. Data between different processor cores is mapped to exclusive address spaces to avoid resource access conflicts.

[0023] The present invention proposes a task scheduling method based on HMP CMCR. An embedded real-time operating system is deployed on the ARM core, and the CMCR component is carefully designed. During operation, the ARM core can intelligently analyze task characteristics and resource status by virtue of its comprehensive control over system resources, thereby accurately allocating tasks to the most suitable DSP core. This process not only optimizes the execution path of the task and ensures the maximum utilization of resources, but also optimizes task scheduling through management strategies. More importantly, this mechanism effectively guarantees the real-time performance of the system and ensures timely response of tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is the overall architecture diagram of the CMCR component system of the present invention;

[0025] Figure 2 It is a schematic diagram of the working of the CMCR component of the present invention. DETAILED DESCRIPTION

[0026] In order to make the purpose, content and advantages of the present invention more clear, the specific implementation methods of the present invention are further described in detail below in conjunction with the drawings and examples.

[0027] The present invention belongs to the field of heterogeneous scheduling of embedded operating systems, and in particular relates to a task scheduling method based on collaborative management of computing resources (CMCR) of heterogeneous multi-core processors (HMP).

[0028] Current scheduling strategies have certain limitations, which lead to high programming complexity and insufficient utilization of processor resources for HMP, restricting the application and promotion of HMP technology in a wider range of scenarios and limiting the full realization of its effectiveness.

[0029] The present invention proposes a task scheduling method based on HMP CMCR, and uses the method to design and implement an efficient CMCR component, including: a main control core component, a computing core component and a multi-core communication component, the main control core component is a main control core CMCR component, the computing core component is a computing core CMCR component, and the multi-core communication component is a CMCR multi-core communication component. For the convenience of design and verification, the domestic Jingdaoxin No. 2 processor (ARM+DSP heterogeneous processor) is selected as the core instance, and this design framework model is also applicable to other heterogeneous platforms.

[0030] Deploy an embedded real-time operating system on the ARM core while keeping the DSP core in bare metal operation mode. Design the CMCR component, which consists of three core components: the main control core component, the computing core component, and the multi-core communication component. The main control core component runs on the ARM core, the computing core component runs on the DSP core, and the multi-core communication component runs on the ARM core and the DSP core. The multi-core communication component acts as a bridge to achieve seamless communication between the ARM core and the DSP core, using the shared memory mechanism to efficiently transmit data. At the same time, through fine access control, it ensures accurate synchronization of multi-core states. The overall system architecture of the CMCR component is as follows: Figure 1 shown.

[0031] As the main control end of the system, the ARM core leads the scheduling of computing tasks on the DSP side to ensure optimal resource utilization, while the DSP core focuses on efficiently executing the assigned computing tasks. The workflow and collaboration mechanism of the CMCR components are intuitively displayed in Figure 2 The schematic diagram clearly depicts the collaborative operation and resource allocation strategy among multiple cores.

[0032] The overall design idea of ​​the present invention is: the DSP core and the ARM core are started in parallel, the DSP core first completes the initialization of the computing core component, enters the standby state, and waits for the task scheduling instruction of the ARM core. At the same time, the ARM core starts and loads the embedded real-time operating system to execute the initialization process of the main control core component. Once the main control core component is ready, the entire CMCR component starts working. When the application layer triggers a computing task request, the embedded real-time operating system intelligently retrieves the available DSP from the DSP idle resource pool, writes the task data safely into the shared memory area through the multi-core communication component, and then triggers the inter-core interrupt to wake up the standby DSP to process the task. After receiving the interrupt signal, the DSP immediately extracts the task data accurately from the shared memory area and synchronously updates its own state; after the computing core component is efficiently executed, the processing result is stored back to the shared memory through the multi-core communication component, and the inter-core interrupt is triggered to notify the embedded real-time operating system that the task execution is completed, so as to realize a seamless task execution and state feedback cycle; if the DSP idle resources are insufficient, the embedded real-time operating system will temporarily store the task to be processed in the DSP task queue, continuously monitor the resource status, and once the DSP resource release is detected, the task execution is immediately scheduled from the queue.

[0033] 1. Main control core component design

[0034] The ARM0 core of the Jingdaoxin No. 2 processor is designed as the main control core. When starting up, the system is reset first to ensure that the DSP core and the ARM core are reset and started synchronously. After the non-main control processor core completes the program boot, it writes the startup completion flag into SYS_REG and then enters a silent waiting state, waiting for further instructions from the main control core.

[0035] The main control core components mainly include: main control core initialization submodule, computing task affinity service interface, command operation mapping submodule, idle DSP resource stack management submodule, computing task queue, main control core interrupt processing submodule, computing resource scheduler, main control core input submodule and main control core output submodule.

[0036] The main control core initialization submodule consists of Figure 2 The computing resource initialization, computing task initialization, and interrupt initialization are composed of computing resource initialization, computing resource initialization, the main control core is responsible for fully registering DSP computing resources to ensure that each idle DSP resource is orderly included in the idle resource stack management; computing task initialization starts to build a computing task queue to lay a solid foundation for subsequent task allocation; interrupt initialization realizes the precise connection and immediate enabling of the core interrupt from the DSP core to the ARM core to ensure smooth communication. After this series of initialization processes are completed, the CMCR components are fully activated and enter an efficient operation state.

[0037] The computing task affinity service interface is used to provide a user interface. It will detect idle DSP computing resources internally. Users only need to use the simple P (IN, CMD, OUT) interface to easily implement dynamic core allocation of DSP tasks, which greatly simplifies the programming complexity of heterogeneous multi-core systems. The interface has three parameters. The IN parameter is used as an input item to carry the original data called by the user; the OUT parameter is used as an output carrier to feedback the result of the calculation and processing; the CMD parameter plays the role of an operation instruction. Its diversity (such as FFT, FIR, mean calculation, modulus calculation, etc.) is accurately parsed through the command operation mapping submodule to ensure that each CMD value can correspond to the corresponding operation. The main control core input submodule and the main control core output submodule work together efficiently to accurately write the IN and OUT parameters into the data transmission address array of the shared memory, realizing seamless data flow and accurate execution of operation instructions.

[0038] The command operation mapping submodule is used to parse the CMD parameters to obtain operation operations such as FFT, FIR, mean calculation, and modulus calculation.

[0039] The idle DSP resource stack management submodule is specifically used to store the DSP computing resources to be allocated, and the computing task queue orderly arranges the tasks that have not been executed due to resource shortage. When the user triggers the P interface, the system will automatically detect the status of the idle DSP resource stack: if the stack is empty, that is, there are no available resources, the new task will be parsed by the command operation mapping submodule and added to the computing task queue for processing; if the stack is not empty, a DSP computing resource core number will be quickly selected from the top of the stack, and the stack popping operation will be performed synchronously for immediate allocation of execution tasks to ensure maximum resource utilization.

[0040] The computing resource scheduler is used to schedule DSP tasks to the DSP with the specified core number for execution. Through the efficient shared memory mechanism, the task information is transmitted to the target DSP to ensure the real-time and accuracy of data transmission. Subsequently, the computing resource scheduler immediately triggers the inter-core interrupt to notify the corresponding DSP to immediately process the assigned DSP task, thereby achieving rapid response and efficient execution of task processing.

[0041] The main control core interrupt processing submodule is used to monitor the dynamic changes of idle DSP computing resources. Once idle resources are detected, it will take action immediately and first examine the fullness of the computing task queue: if the queue is empty, that is, there are no tasks to be processed, then the newly discovered idle DSP resources will be quickly included in the resource stack for subsequent allocation; if the queue is not empty, the first task in the queue will be processed first, and through the careful deployment of the computing resource scheduler, the task will be accurately assigned to the idle DSP for execution, ensuring the dual efficiency of resource utilization and task processing.

[0042] 2. Computing core component design

[0043] The computing core components include: computing core initialization submodule, computing core interrupt processing submodule, operator processing tasks, computing core input submodule and computing core output submodule, etc., which fully support the collaborative operation of computing resources.

[0044] The computing core initialization submodule focuses on interrupt initialization configuration and realizes the inter-core interrupt hooking and activation from ARM to DSP. Once ready, the computing core component enters the standby state, waiting for the ARM core to issue computing tasks.

[0045] The computing core interrupt processing submodule monitors the DSP task scheduling instructions in real time. Once a DSP task scheduling instruction is captured, it accurately extracts the task data from the shared memory area, synchronously updates its own status, and accurately schedules the operator processing task according to the interrupt category.

[0046] The operator processing task is used to undertake ARM's FFT, FIR, mean and modulus operations. It efficiently extracts input data IN from the shared memory through the calculation core input submodule and the calculation core output submodule. After being processed by the operator processing task, the result is directly written into the output data OUT.

[0047] 3. Multi-core communication components for data interaction between ARM and DSP

[0048] The multi-core communication component for data interaction between ARM and DSP is mainly composed of ARM and DSP inter-core interrupts and shared memory. ARM and DSP use inter-core interrupts to achieve efficient interrupt management and allocation, which has been described in interrupt processing modules 1 and 2; shared memory covers the task information storage area and input and output data area to ensure smooth and efficient data exchange.

[0049] The task information storage area is mainly used to store the instructions and status information of the task, and update the DSP working status in real time; the input and output data areas are cleverly designed, and only the data address is transmitted through the two-dimensional memory array shared by ARM and DSP to avoid redundant copying and improve efficiency. This array is carefully laid out to allocate independent input and output data areas for each DSP resource, including input data, input data length, input data type, output data, etc., to ensure that data between different processor cores is mapped to a dedicated address space to achieve efficient data exchange.

[0050] Embodiment 1:

[0051] To evaluate the efficiency of ARM core scheduling of DSP tasks and the superiority of dynamic scheduling over static scheduling, we designed two comparative experiments. In the experiment, the single FFT, mean (MEAN), and modulus (MAG) operation time was too short. To ensure that the single calculation time was significant enough for analysis, each task was set to perform 10,000 repeated operations: FFT operation performed 10,000 64-point FFTs on the sequence from 0 to 63; mean calculation performed 10,000 128-length average calculations on the sequence from 0 to 127; modulus calculation also performed 10,000 128-length modulus calculations on the sequence from 0 to 127. The three satisfied t FFT >t MAG >t MEAN This setting ensures sufficient computation to clearly reveal the performance differences under different operations and scheduling strategies.

[0052] Experiment 1: Comparative test of scheduling different DSP resource numbers for FFT calculation tasks

[0053] The ARM master control end played the role of the core scheduler in the experiment. It scheduled 60 computing tasks for the single-core DSP (in the worst case, all 60 computing tasks were scheduled to the same DSP for processing), scheduled two DSPs to perform 60 computing tasks in parallel, scheduled four DSPs to perform 60 computing tasks, and scheduled six DSPs to perform 60 computing tasks in parallel, and counted the execution time respectively.

[0054] Experiment 2: Comparative test of static scheduling and dynamic scheduling computing tasks

[0055] The computing tasks are allocated as follows: In static binding mode, DSP core 0 focuses on performing 60 FFT calculations, DSP core 1 undertakes 60 mean calculations, and DSP core 2 performs 60 modulus calculations. In dynamic scheduling, three DSPs are first used to flexibly complete a total of 180 tasks, including 60 FFTs, 60 mean calculations, and 60 modulus calculations. Subsequently, six DSPs are expanded to perform the same calculations to explore the efficiency of task execution under higher parallelism.

[0056] In Experiment 1, the execution time data for the FFT calculation task is recorded in detail in Table 1. It can be seen that as the number of DSP resources increases, the execution time of the FFT calculation task gradually shortens, which strongly reflects the role of resource expansion in improving computing efficiency. During the scheduling of 60 DSP similar calculation tasks, the time consumption increases linearly with the number of calculations. Scheduling two DSPs to perform 60 DSP calculation tasks takes nearly half the time of a single-core DSP, and as the number of DSP cores increases, the time consumption continues to decrease, verifying the significant effect of the CMCR component in efficiently scheduling DSP core resources.

[0057] Table 1 Experimental results of scheduling different DSP resource numbers for FFT calculation tasks

[0058]

[0059] The data of Experiment 2 is shown in Table 2, which clearly reveals the difference between static scheduling and dynamic scheduling in the execution time of heterogeneous computing tasks. The dynamic scheduling strategy significantly shortens the overall execution time of computing tasks through a flexible task balancing mechanism. More importantly, as the number of DSP resources increases, the advantages of dynamic scheduling become more prominent, effectively demonstrating its superior performance in complex computing environments.

[0060] Table 2 Experimental results of static scheduling and dynamic scheduling computing tasks

[0061]

[0062]

[0063] The experimental results of the HMP-based CMCR method proposed in the present invention show that for similar tasks, resource expansion directly leads to a leap in processing rate and a sharp reduction in task execution time, which strongly confirms the excellent performance of the component in task scheduling, effectively balances resource load, and achieves a qualitative leap in resource utilization. In the face of heterogeneous tasks, although the overall progress is limited by the most time-consuming tasks, the processing speed still steadily increases with the growth of resources. It is particularly noteworthy that in this complex situation, the dynamic scheduling advantage of the component is particularly significant, which not only effectively balances the DSP resource load of HMP, but also ensures the high real-time response capability of the system, and fully demonstrates its core value of efficient management. At the same time, it reduces the difficulty of application development and improves development efficiency.

[0064] The key points of the present invention are:

[0065] The HMP CMCR-based task scheduling method proposed in the present invention fully considers the advantages of the main control core and the computing core, realizes the task scheduling and execution of the computing core through the CMCR component, and realizes the efficient collaborative management of computing resources.

[0066] The present invention proposes a CMCR mechanism, which has the following advantages:

[0067] 1. Deploy a real-time operating system on the ARM core and design the main control core component to manage the DSP computing core to ensure the immediacy and stability of task processing, fully utilize the heterogeneous resource advantages of HMP, and maximize the use of computing resources;

[0068] 2. ARM core provides computing task affinity service interface for users, which improves programming efficiency and reduces the complexity and difficulty of application development under HMP;

[0069] 3. The design of multi-core communication components uses inter-core interrupts and shared memory mechanisms to achieve fast data transmission. Data between different processor cores is mapped to exclusive address spaces to avoid resource access conflicts.

[0070] The present invention proposes a task scheduling method based on HMP CMCR. An embedded real-time operating system is deployed on the ARM core, and the CMCR component is carefully designed. During operation, the ARM core can intelligently analyze task characteristics and resource status by virtue of its comprehensive control over system resources, thereby accurately allocating tasks to the most suitable DSP core. This process not only optimizes the execution path of the task and ensures the maximum utilization of resources, but also optimizes task scheduling through management strategies. More importantly, this mechanism effectively guarantees the real-time performance of the system and ensures timely response of tasks.

[0071] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A task scheduling method for collaborative management of computing resources of heterogeneous multi-core processors, characterized in that: Design and implement an efficient CMCR component, including: a main control core component, a computing core component and a multi-core communication component, the main control core component is a main control core CMCR component, the computing core component is a computing core CMCR component, and the multi-core communication component is a CMCR multi-core communication component; Deploy an embedded real-time operating system on the ARM core while keeping the DSP core in bare metal operation mode. The main control core component runs on the ARM core, the computing core component runs on the DSP core, and the multi-core communication component runs on the ARM core and the DSP core. The DSP core and the ARM core are started in parallel. The DSP core completes the initialization of the computing core component first and enters the standby state, waiting for the task scheduling instruction of the ARM core. At the same time, the ARM core starts and loads the embedded real-time operating system to execute the initialization process of the main control core component. Once the main control core component is ready, the entire CMCR component starts working. When a computing task request is triggered at the application layer, the embedded real-time operating system intelligently retrieves available DSPs from the DSP free resource pool, writes the task data securely into the shared memory area through the multi-core communication component, and then triggers an inter-core interrupt to wake up the standby DSP to process the task; After receiving the interrupt signal, the DSP immediately extracts the task data from the shared memory area accurately and updates its own status synchronously. After the computing core component is efficiently executed, the multi-core communication component stores the processing results back to the shared memory and triggers an inter-core interrupt to notify the embedded real-time operating system that the task has been completed, thus achieving a seamless task execution and status feedback loop. If the DSP idle resources are insufficient, the embedded real-time operating system will temporarily store the pending tasks in the DSP task queue and continuously monitor the resource status. Once it detects that the DSP resources are released, it will immediately schedule the task execution from the queue.

2. The task scheduling method for collaborative management of computing resources of heterogeneous multi-core processors according to claim 1, characterized in that: The heterogeneous multi-core processor is the Jingdaoxin No. 2 processor, which is an ARM+DSP heterogeneous processor.

3. The task scheduling method for collaborative management of computing resources of heterogeneous multi-core processors according to claim 1, characterized in that: As the system master, the ARM core dominates the scheduling of computing tasks on the DSP side, while the DSP core focuses on efficiently executing the assigned computing tasks.

4. The task scheduling method for collaborative management of computing resources of heterogeneous multi-core processors according to claim 1, characterized in that: The ARM0 core is designed as the main control core. When starting up, the system is reset first to ensure that the DSP core and the ARM core are reset and started synchronously. After completing the program boot, the non-master processor core writes the startup completion flag into SYS_REG and then enters a silent waiting state, waiting for further instructions from the main control core.

5. The task scheduling method for collaborative management of computing resources of heterogeneous multi-core processors according to any one of claims 1 to 4, characterized in that: The main control core components include: main control core initialization submodule, computing task affinity service interface, command operation mapping submodule, idle DSP resource stack management submodule, computing task queue, main control core interrupt processing submodule, computing resource scheduler, main control core input submodule and main control core output submodule; The main control core initialization submodule consists of computing resource initialization, computing task initialization, and interrupt initialization. In the computing resource initialization stage, the main control core is responsible for fully registering DSP computing resources to ensure that each idle DSP resource is orderly included in the idle resource stack management; computing task initialization starts to build a computing task queue; interrupt initialization realizes the precise connection and immediate enabling of inter-core interrupts from DSP core to ARM core; after this series of initialization processes are completed, the CMCR components are fully activated and enter an efficient operation state. The computing task affinity service interface is used to provide a user interface. It will detect idle DSP computing resources internally. Users can use the simple P (IN, CMD, OUT) interface to implement dynamic core allocation of DSP tasks. The interface has three parameters. The IN parameter is used as an input item to carry the original data called by the user. The OUT parameter is used as an output carrier to feed back the results of the computing process. The CMD parameter is the operation instruction. The main control core input submodule and the main control core output submodule work together efficiently to accurately write the IN and OUT parameters into the data transfer address array of the shared memory; The command operation mapping submodule is used to parse the CMD parameters and obtain the operation; The idle DSP resource stack management submodule is specifically used to store the DSP computing resources to be allocated, and the computing task queue orderly arranges the tasks that have not been executed due to resource shortage. When the user triggers the P interface, the system will automatically detect the status of the idle DSP resource stack: if the stack is empty, that is, there are no available resources, the new task will be parsed by the command operation mapping submodule and added to the computing task queue for processing; if the stack is not empty, a DSP computing resource core number will be quickly selected from the top of the stack, and the stack popping operation will be performed synchronously for immediate allocation of execution tasks. The computing resource scheduler is used to schedule DSP tasks to be executed on the DSP with the specified core number. The task information is passed to the target DSP through the shared memory mechanism. Subsequently, the computing resource scheduler immediately triggers the inter-core interrupt to notify the corresponding DSP to immediately start processing the assigned DSP task. The main control core interrupt processing submodule is used to monitor the dynamic changes of idle DSP computing resources. Once idle resources are detected, it will take action immediately and first examine the fullness of the computing task queue: if the queue is empty, that is, there are no tasks to be processed, then the newly discovered idle DSP resources will be quickly included in the resource stack for subsequent allocation; if the queue is not empty, the first task in the queue will be processed first, and the task will be accurately assigned to the idle DSP for execution through the careful deployment of the computing resource scheduler.

6. The task scheduling method for collaborative management of computing resources of heterogeneous multi-core processors as claimed in claim 5, characterized in that: The operations include: FFT, FIR, mean value calculation and modulus operation.

7. The task scheduling method for collaborative management of computing resources of heterogeneous multi-core processors according to claim 5, characterized in that: The computing core component includes: a computing core initialization submodule, a computing core interrupt processing submodule, an operator processing task and a computing core input submodule and a computing core output submodule; The computing core initialization submodule focuses on interrupt initialization configuration, realizes the inter-core interrupt hooking and activation from ARM to DSP. Once ready, the computing core component enters the standby state, waiting for the ARM core to issue computing tasks; The computing core interrupt processing submodule monitors the DSP task scheduling instructions in real time. Once a DSP task scheduling instruction is captured, it accurately extracts the task data from the shared memory area, synchronously updates its own status, and accurately schedules the operator processing task according to the interrupt category. The operator processing task is used to undertake the calculation operation of ARM. It efficiently extracts the input data IN from the shared memory through the calculation core input submodule and the calculation core output submodule. After being processed by the operator processing task, the result is directly written into the output data OUT.

8. The task scheduling method for collaborative management of computing resources of heterogeneous multi-core processors according to claim 7, characterized in that: The multi-core communication component for data interaction between ARM and DSP consists of interrupts between ARM and DSP cores and shared memory; the shared memory covers the task information storage area and the input and output data area.

9. The task scheduling method for collaborative management of computing resources of heterogeneous multi-core processors according to claim 8, characterized in that: The task information storage area stores the task instructions and status information, and updates the DSP working status in real time.

10. The task scheduling method for collaborative management of computing resources of heterogeneous multi-core processors according to claim 8, characterized in that: Only data addresses are transmitted through the two-dimensional memory array shared by ARM and DSP. This array allocates independent input and output data areas for each DSP resource, including input data, input data length, input data type and output data, ensuring that data between different processor cores is mapped to exclusive address space.

Citation Information

Cited By

  • Hybrid deployment method for heterogeneous multi-core chip application program

    CN119883650A

  • Satellite task planning computing power dynamic distribution system based on heterogeneous multi-core processor

    CN120803752A

  • Satellite task planning computing power dynamic allocation system based on heterogeneous multi-core processor

    CN120803752B

  • PMC parallel acquisition system and method of ARM heterogeneous multi-core processor

    CN121411983A

  • A PMC parallel acquisition system and method of an ARM heterogeneous multi-core processor

    CN121411983B