Soc module

The SoC module addresses cache memory conflicts in shared cache systems by prioritizing processor cores based on access patterns, simplifying control and reducing software costs and latency.

JP2024156298AActive Publication Date: 2024-11-06DENSO CORP +2
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
JP2023070645
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-04-24
Publication Date
2024-11-06
Estimated Expiration
2043-04-24

AI Technical Summary

Technical Problem

In SoC modules with shared cache memory among multiple processors, predefined access timing and addresses lead to complex control and increased software development costs due to potential conflicts.

Method used

An SoC module with processor cores having preset priorities, a profile acquisition unit to monitor access patterns, an access conflict adjustment unit to determine priority cores, and access suppression units to control access conflicts, ensuring simpler control and reduced contention.

Benefits of technology

Suppresses conflicts between processor cores with simpler control, reduces software development costs, and minimizes latency and power consumption by prioritizing core access to the cache memory.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024156298000001_ABST
    Figure 2024156298000001_ABST
Patent Text Reader

Abstract

To facilitate suppression of deterioration in efficiency of access to a cache memory in an SoC module.SOLUTION: An SoC module 100 includes: a plurality of processor cores 111, 112 each in which the priority of arithmetic processing is set; a cache memory 120 that is accessed by the plurality of processor cores to read and write data; a profile acquisition unit 130 that acquires access profile information; an access conflict adjustment unit 140 that executes determining whether or not accesses to the cache memory conflict with each other and determining a priority core; and a plurality of access suppression units 151, 152 that control execution and stop of access by each processor core. The access conflict adjustment unit determines a conflicting core with higher priority as the priority core. The access suppression units provided corresponding to non-priority cores stop accesses by the non-priority cores.SELECTED DRAWING: Figure 1
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present disclosure relates to an SoC module. [Background technology]

[0002] Conventionally, an SoC module having multiple processors is known (Patent Document 1). In such an SoC module, it is desirable to share a cache memory among the multiple processors in order to reduce the area of ​​the module. In order to reduce the area of ​​the module, the storage capacity of such a cache memory becomes smaller than the total capacity of the cache memory conventionally provided for each processor. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Special Publication No. 2022-524487 Summary of the Invention [Problem to be solved by the invention]

[0004] In such a cache memory, if the access timing and accessible addresses are specified in advance for each processor in order to suppress contention between processors when accessing the cache memory, complex control is required, which results in increased software development costs. [Means for solving the problem]

[0005] The present disclosure can be realized in the following forms.

[0006] According to one embodiment of the present disclosure, there is provided an SoC module (100). The SoC module includes a plurality of processor cores (111, 112), each of which has a preset priority for arithmetic processing, a cache memory (120) accessed by the plurality of processor cores to read and write data, and a profile acquisition unit (130) that acquires access profile information indicating a change over time in an access destination address of the cache memory by each of the plurality of processor cores, and a processor that uses the access profile information to determine whether or not accesses to the cache memory by at least two of the plurality of processor cores conflict with each other, and a processor that uses the access profile information to determine whether or not accesses to the cache memory by the other of the plurality of conflicting cores that are the plurality of processor cores that conflict with each other. and determining a priority core which is a competing core that will execute the access in priority over a competing core, and a plurality of access suppression units (151, 152) provided corresponding to each of the plurality of processor cores, the access conflict adjustment unit determines a competing core having a higher priority as the priority core among the plurality of competing cores, and the access suppression unit provided corresponding to a non-priority core which is a competing core that is not the priority core among the plurality of competing cores stops the access by the non-priority core.

[0007] According to this type of SoC module, among multiple competing cores, a competing core with a higher set priority is determined as a prioritized core, and access to the cache memory by non-prioritized cores is stopped, thereby suppressing competition between processor cores. Therefore, it is possible to suppress competition between processor cores with simpler control than a configuration in which access timing and accessible addresses are specified in advance for each processor core, and to suppress increases in software development costs. [Brief description of the drawings]

[0008] [Figure 1]FIG. 1 is an explanatory diagram showing a schematic configuration of an SoC module according to an embodiment of the present invention. [Diagram 2] 11 is a chart showing an example of a relationship between an access destination address and an access timing. [Diagram 3] 4 is a flowchart showing a procedure of access control according to the present embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] A. First embodiment: A-1.Device configuration: The SoC module 100 accesses data stored in the main memory device 200 and sequentially executes predetermined arithmetic processing. In this embodiment, the SoC module 100 executes arithmetic processing using a neural network. As shown in FIG. 1, the SoC module 100 of this embodiment includes processor cores 111 and 112, a cache memory 120, a bus BS, a profile acquisition unit 130, an access contention adjustment unit 140, and access inhibition units 151 and 152.

[0010] In this embodiment, the processor cores 111 and 112 are configured as NPUs (Neural network Processing Units) that execute arithmetic processing using a neural network. The processor cores 111 and 112 execute arithmetic processing for outputting feature quantities for discriminating pedestrians, structures, and the like included in image data captured by a camera mounted on a vehicle, for example. The processor cores 111 and 112 each operate according to a clock signal supplied via a bus BS from a clock generation circuit CLK provided outside the SoC module 100, and execute arithmetic processing. The processor cores 111 and 112 may be configured as CPUs (Central Processing Units) that execute predetermined arithmetic processing sequentially, without being limited to arithmetic processing using a neural network. The specific configuration of the processor cores 111 and 112 is similar to that of a known NPU, and therefore description thereof will be omitted. The number of processor cores included in the SoC module 100 is not limited to two, and may be three or more.

[0011] Further, the processor cores 111 and 112 are each set with a priority of arithmetic processing. In this embodiment, the "priority of arithmetic processing" refers to a degree indicating which of the processor cores 111 and 112 is to be given priority in executing arithmetic processing. The priority of arithmetic processing is determined according to the type of arithmetic processing executed in each processor core, and is determined so that, for example, the priority of a processor core that executes arithmetic processing that requires greater real-time performance is given a higher priority. In the following description, an example will be described in which the priority set for the processor core 111 is higher than the priority set for the processor core 112.

[0012] The main memory device 200 and the processor cores 111 and 112 are connected to each other via a bus BS. The main memory device 200 stores various data input / output via the bus BS in the arithmetic processing executed by the processor cores 111 and 112, such as image data and data such as coefficients used in the arithmetic processing. In this embodiment, the main memory device 200 is configured as a DRAM (Dynamic Random Access Memory).

[0013] The cache memory 120 is provided between the processor cores 111, 112 and the main storage device 200 on the bus BS. The cache memory 120 is accessed by the processor cores 111, 112 to read and write data. More specifically, when data is read from the main storage device 200 to the processor core 111 or the processor core 112, the cache memory 120 stores the data. When the processor core 111 or the processor core 112 requests to read the same data again, the processor core reads the data from the cache memory 120 instead of the main storage device 200. This reduces the frequency of access to the main storage device 200 from the processor cores that require a long time to access data, and reduces the decrease in data access efficiency. In this embodiment, the cache memory 120 is configured as an SRAM (Static Random Access Memory) that can read and write data at high speed.

[0014] The profile acquisition unit 130 is provided on the bus BS between the processor cores 111, 112 and the cache memory 120. The profile acquisition unit 130 acquires access profile information for each of the processor cores 111, 112. "Access profile information" refers to information indicating a change over time in the access destination address of the cache memory 120 accessed by each processor core. In this embodiment, the access profile information includes information indicating which processor core accesses which address of the cache memory 120 at what timing.

[0015] Furthermore, in this embodiment, the profile acquisition unit 130 uses the acquired access profile information to create a chart showing the relationship between the access destination address and the access timing for each processor core, as in the example shown in FIG. 2. In the chart, the vertical axis indicates the access destination address of the cache memory 120 by the processor core. The horizontal axis indicates the access timing. "Access timing" refers to the timing at which the processor core accesses the access destination address. Furthermore, a black circle indicates an access by the processor core 111, and a white dot indicates an access by the processor core 112. Note that in FIG. 2, for convenience of illustration, the chart showing the relationship between the access destination address and the access timing is shown in a schematic manner.

[0016] The access contention arbitration unit 140 shown in FIG. 1 acquires access profile information and determines whether or not the accesses by each processor core conflict with each other. "Conflict" means that the same address of the cache memory 120 is accessed by multiple processor cores. In the example shown in FIG. 2, contention occurs at access timings t1, t2, and t3. In addition, the access contention arbitration unit 140 determines a priority core when contention occurs. The "priority core" means a contention core that executes access with priority over other contention cores among multiple processor cores that compete with each other (also referred to as "contention cores" in the following description). The access contention arbitration unit 140 uses a priority previously set for each contention core to determine a contention core with a higher priority as a priority core. In addition, in the following description, a contention core that is not a priority core among multiple contention cores is also referred to as a "non-priority core."

[0017] 1 are provided corresponding to the processor cores 111 and 112, respectively. Each access inhibiting unit controls whether to enable or disable access to the cache memory 120 by each processor core. Each access inhibiting unit also controls whether to enable or disable supply of a clock signal to each processor core. Specific control by each access inhibiting unit will be described in the access control section below.

[0018] A-2. Access Control: As shown in Fig. 3, the SoC module 100 of this embodiment executes access control for controlling access from each processor core to the cache memory 120. The access control is started when the SoC module 100 is booted, and is executed repeatedly while the SoC module 100 is booted. As shown in Fig. 2, in the access control, the SoC module 100 executes the control of steps S10 and S20 and the control of steps S30 to S90 in parallel.

[0019] In step S10, the profile acquisition unit 130 acquires access profile information for each processor core. In this embodiment, the profile acquisition unit 130 can acquire the access profile information by monitoring the usage status of the bus BS by each processor core.

[0020] In step S20, the profile acquisition unit 130 uses the acquired access profile information to create a chart showing the relationship between the access destination address and the access timing for each processor core. The profile acquisition unit 130 repeatedly executes steps S10 and S20 at predetermined time intervals while access control is being performed.

[0021] In step S30, the access contention arbitration unit 140 uses the created chart to determine whether or not the accesses by the processor cores are in conflict with each other. More specifically, when a plurality of processor cores request access to the same address in the cache memory 120 at the same timing, the access contention arbitration unit 140 determines that the accesses by the processor cores are in conflict with each other.

[0022] If a conflict has occurred (step S30: Yes), the access conflict arbitration unit 140 determines a priority core (step S40). As described above, the access conflict arbitration unit 140 determines, among the multiple processor cores, a processor core having a higher preset priority as a priority core. As described above, in this embodiment, a higher priority is set for the processor core 111 than for the processor core 112, and therefore the processor core 111 is determined as the priority core.

[0023] On the other hand, if no contention has occurred (step S30: No), the access contention adjustment unit 140 repeats the determination of step S30. If the accesses by the processor cores are not in conflict with each other, each access inhibition unit may cause each processor core to execute arithmetic processing at any timing.

[0024] In step S50, the access suppression unit 152 stops the access by the processor core 112, which is a non-priority core. If the access by the non-priority core is not stopped, it may happen that immediately after data is read from the main memory device 200 to the cache memory 120 based on the access by the priority core, different data is read from the main memory device 200 to the same address of the cache memory 120 based on the access by the non-priority core. In such a case, even if the priority core reads data once again, the data has been overwritten by the access by the non-priority core, so the data is read from the main memory device 200 to the cache memory 120 again, and the latency increases. In order to avoid such an increase in latency, the access suppression unit stops the access by the non-priority core.

[0025] In step S60, the access inhibiting unit 152 stops the supply of a clock signal to the non-prioritized core, the processor core 112. This stops the operation of the non-prioritized core, and suppresses the operating power of the non-prioritized core.

[0026] In step S70, the access contention adjustment unit 140 determines whether the progress rate of the access by the priority core is equal to or greater than a predetermined threshold. The "progress rate" refers to the ratio of the arithmetic operations that have already been completed to the total arithmetic operations that are scheduled to be executed by the processor core. The access contention adjustment unit 140 can grasp the progress rate of the access by the priority core by referring to a chart showing the relationship between the access destination address and the access timing.

[0027] While the progress rate is less than a predetermined rate (step S70: No), access contention adjustment section 140 repeatedly executes the determination of step S70, and maintains the state in which access by non-prioritized cores and supply of clock signals to non-prioritized cores are stopped.

[0028] If the progress rate is equal to or greater than a predetermined rate (step S70: Yes), the access suppression unit 152 resumes the supply of a clock signal to the processor core 112, which is the non-priority core (step S80). Furthermore, in step S90, the access suppression unit 152 resumes access by the processor core 112, which is the non-priority core. This is because, in a processor core whose progress rate is equal to or greater than the threshold, the frequency of access to the cache memory 120 is reduced compared to the peak time, and therefore the frequency of occurrence of the increase in latency as described above is also suppressed. The SoC module 100 repeatedly executes the above-described steps S30 to S90.

[0029] According to the SoC module 100 of the embodiment described above, it is possible to suppress contention between processor cores by determining, among multiple competing cores, a competing core with a higher set priority as a prioritized core and stopping access by non-prioritized cores to the cache memory 120. Therefore, it is possible to suppress contention between processor cores with simpler control than in a configuration in which access timing and accessible addresses are specified in advance for each processor core, and it is possible to suppress increases in software development costs.

[0030] In addition, since access by the non-priority core is resumed when the progress rate of access by the priority core reaches or exceeds a preset threshold, access by the non-priority core can be resumed at a timing when the frequency of access to the cache memory 120 by the priority core drops compared to the peak time. This makes it possible to suppress an increase in latency due to the occurrence of contention.

[0031] Furthermore, while access by the non-prioritized core is stopped, the supply of a clock signal to the non-prioritized core is stopped, so that the operation of the non-prioritized core is stopped, and the operating power of the non-prioritized core can be reduced.

[0032] B. Other embodiments: (B1) In the above embodiment, when a conflict occurs, the access suppression unit corresponding to the non-priority core stops the access by the non-priority core and stops the supply of a clock signal to the non-priority core, but the present disclosure is not limited to this. The access suppression unit may continue the supply of a clock signal to the non-priority core and only stop the execution of the access by the non-priority core. The SoC module 100 configured in this way can also suppress the occurrence of conflicts in access to the cache memory 120 by multiple processor cores. In addition, since the access suppression unit 151 only needs to control the execution and stop of access by the processor core, the configuration of the access suppression unit can be prevented from becoming complicated.

[0033] (B2) In the above embodiment, the access conflict adjustment unit 140 determines whether or not the accesses conflict with each other for each processor core, but the present disclosure is not limited thereto. The access conflict adjustment unit 140 may determine whether or not the accesses conflict with each other for each of a plurality of instructions that realize the arithmetic processing executed in each processor. In this case, it is desirable that the priority of the arithmetic processing is also set for each instruction. The SoC module 100 having such a configuration also achieves the same effect as the above embodiment. In addition, since it is possible to determine whether or not a conflict occurs in an instruction unit smaller than a processor core unit, for example, while one processor core is executing arithmetic processing, the other processor core can execute an instruction that accesses an address that the processor core does not access, and the decrease in efficiency of the arithmetic processing can be suppressed.

[0034] (B3) In the above embodiment, when a contention occurs, only the access suppression units 151 and 152 provided corresponding to the non-priority cores stop access by the non-priority cores, but the present disclosure is not limited to this. When a contention occurs, all access suppression units may temporarily stop access by all processor cores and resume access in order starting from the processor core with the highest set priority. The SoC module 100 configured in this way also achieves the same effects as the above embodiment.

[0035] (B4) In the above embodiment, the access suppression units 151 and 152 resume the access by the non-priority core when the progress rate of the access by the priority core becomes equal to or greater than a preset threshold value, but the present disclosure is not limited to this. The access suppression units 151 and 152 may resume the access by the non-priority core after a predetermined time has elapsed after stopping the access by the non-priority core. The SoC module 100 having such a configuration can also resume the access by the non-priority core at a timing when the frequency of access to the cache memory 120 by the priority core decreases compared to the peak time. This makes it possible to suppress an increase in latency due to the occurrence of contention.

[0036] The present disclosure is not limited to the above-mentioned embodiment, and can be realized in various configurations without departing from the spirit of the present disclosure. For example, the technical features in each embodiment corresponding to the technical features in the form described in the Summary of the Invention column can be appropriately replaced or combined in order to solve some or all of the above-mentioned problems or to achieve some or all of the above-mentioned effects. Furthermore, if the technical feature is not described as essential in this specification, it can be appropriately deleted.

[0037] The SoC module 100 and the methods described herein may be implemented by a special purpose computer provided by configuring a processor and memory programmed to perform one or more functions embodied in a computer program. Alternatively, the SoC module 100 and the methods described herein may be implemented by a special purpose computer provided by configuring a processor with one or more dedicated hardware logic circuits. Alternatively, the SoC module 100 and the methods described herein may be implemented by one or more special purpose computers configured by a combination of a processor and memory programmed to perform one or more functions and a processor configured with one or more hardware logic circuits. The computer program may also be stored in a computer-readable non-transitory tangible storage medium as instructions executed by the computer. [Explanation of symbols]

[0038] 100... SoC module, 111, 112... processor core, 120... cache memory, 130... profile acquisition unit, 140... access competition adjustment unit, 151, 152... access suppression unit

Claims

1. A SoC module (100), A plurality of processor cores (111, 112) each having a predetermined priority for processing operations; A cache memory (120) that is accessed by the multiple processor cores to read and write data; a profile acquisition unit (130) that acquires access profile information indicating a change over time in an access destination address of the cache memory by each of the plurality of processor cores; an access contention adjustment unit (140) that uses the access profile information to determine whether or not accesses to the cache memory by at least two of the plurality of processor cores conflict with each other, and determines a priority core that is a contention core that is a plurality of processor cores that conflict with each other and that is a priority core that is to execute the access in priority over the other contention cores; a plurality of access inhibiting units (151, 152) provided corresponding to the plurality of processor cores, each of which controls execution and suspension of access to the cache memory by each of the processor cores; Equipped with the access contention arbitration unit determines a contention core having a higher priority among the plurality of contention cores as the prioritized core; the access suppression unit provided corresponding to a non-priority core that is a contention core other than the priority core among the plurality of contention cores stops the access by the non-priority core; SoC module.

2. 2. The SoC module of claim 1, the access suppression unit resumes the access by the non-prioritized core when a progress rate of the access by the priority core becomes equal to or greater than a preset threshold value. SoC module.

3. 3. The SoC module according to claim 1, the access suppression unit stops supply of a clock signal to the non-prioritized core while the access by the non-prioritized core is suspended; SoC module.

Citation Information

Patent Citations

  • Information processor

    JP1993012111A

  • Semiconductor device and bus generator

    JP2019114015A

  • Computational Data Storage System

    JP2022524487A