WO2021103618A1 - Configuration of operating frequency of chip - Google Patents
Configuration of operating frequency of chip Download PDFInfo
- Publication number
- WO2021103618A1 WO2021103618A1 PCT/CN2020/105195 CN2020105195W WO2021103618A1 WO 2021103618 A1 WO2021103618 A1 WO 2021103618A1 CN 2020105195 W CN2020105195 W CN 2020105195W WO 2021103618 A1 WO2021103618 A1 WO 2021103618A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- chip
- subtask
- frequency
- task
- data
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
- G06F9/5044—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals considering hardware capabilities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F1/00—Details not covered by groups G06F3/00 - G06F13/00 and G06F21/00
- G06F1/26—Power supply means, e.g. regulation thereof
- G06F1/32—Means for saving power
- G06F1/3203—Power management, i.e. event-based initiation of a power-saving mode
- G06F1/3234—Power saving characterised by the action undertaken
- G06F1/324—Power saving characterised by the action undertaken by lowering clock frequency
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F1/00—Details not covered by groups G06F3/00 - G06F13/00 and G06F21/00
- G06F1/26—Power supply means, e.g. regulation thereof
- G06F1/32—Means for saving power
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Definitions
- the present disclosure relates to smart device technology, in particular to a method and device for setting the operating frequency of a chip.
- Frequency reduction is one of the main methods of preventing excessive power consumption of the chip.
- the frequency reduction technology mainly reduces the power consumption of the chip by temporarily reducing the operating frequency of the chip (frequency).
- an initial operating frequency can be set before the application program is executed in the terminal device, and the chip runs at the initial operating frequency to perform calculation processing on the application program.
- a fixed frequency reduction strategy can be adopted for the chip, for example, the working frequency of the chip is reduced by a fixed ratio, or the working frequency is reduced to A fixed value.
- the embodiments of the present disclosure provide at least a method and device for setting the operating frequency of a chip.
- a method for setting the operating frequency of a chip includes: acquiring a plurality of subtasks of a target task and task parameters of each subtask, the task parameters including parameters used to indicate the operation scale of the subtasks Based on the task parameters of each subtask in the multiple subtasks, determine the target chip frequency of each subtask; according to the determined target chip frequency of each subtask, set the operating frequency of the chip to execute each subtask.
- the method further includes: performing task analysis processing on the target task to obtain the multiple subtasks and the task parameters of each subtask; storing each subtask and each subtask in the multiple subtasks The corresponding relationship of the task parameters of each subtask; the obtaining multiple subtasks of the target task and the task parameters of each subtask includes: searching the task parameters of each subtask from the stored corresponding relationships.
- the task parameter includes at least one of the following: a calculation amount of the subtask, and a memory access amount of the subtask.
- the determining the target chip frequency of each subtask based on the task parameter of each subtask of the plurality of subtasks includes: acquiring device information of the device where the chip is located, and The device information includes device resource information; the target chip frequency of each subtask is determined based on the device information and the task parameter of each subtask of the multiple subtasks.
- the device resource information includes any one or more of the following: the number of computing units, bandwidth, and memory capacity.
- the device information further includes: the chip temperature of the chip; the determining each subtask based on the device information and the task parameter of each subtask of the plurality of subtasks
- the target chip frequency of includes: determining the target of the first subtask based on the task parameters of the first subtask among the plurality of subtasks, the device resource information, and the chip temperature during execution of the first subtask Chip frequency.
- the determining the target chip frequency of each subtask based on the device information and the task parameter of each subtask of the plurality of subtasks includes: acquiring a plurality of first data and A preset mapping relationship between a plurality of second data, the first data includes preset task parameters and preset device information, the second data includes a preset chip frequency; according to the preset mapping relationship, the Device information and task parameters of each subtask determine the target chip frequency of each subtask.
- the determining the target chip frequency of each subtask according to the preset mapping relationship, the device information, and the task parameters of each subtask includes: determining the third data and the task parameters of each subtask.
- the distance between each first data in the plurality of first data in the preset mapping relationship, the third data includes task parameters and device information of the first subtask among the plurality of subtasks; It is assumed that the preset chip frequency corresponding to the target first data closest to the third data in the mapping relationship is used as the target chip frequency of the first subtask.
- the method before the obtaining the preset mapping relationship between the plurality of first data and the plurality of second data, the method further includes: obtaining a plurality of sets of optional chip frequencies, and obtaining samples obtained A plurality of discrete first data; for each of the first data in the plurality of first data, a set of chip frequencies is selected from the plurality of sets of optional chip frequencies as the The data corresponds to the second data, and a mapping relationship between each of the first data and the selected second data is established.
- the preset chip frequency corresponding to the first data in the preset mapping relationship is based on the second chip frequency under the condition of each group of selectable chip frequencies in the multiple sets of selectable chip frequencies.
- the performance evaluation parameter corresponding to one data is selected from the multiple sets of optional chip frequencies.
- the performance evaluation parameters include task processing performance parameters and chip operating power consumption;
- the preset chip frequency corresponding to the first data in the preset mapping relationship is: the multiple groups Among the optional chip frequencies, the chip operating power consumption is lower than the preset power consumption and the optional chip frequency with the best task processing performance parameters.
- the method further includes: receiving configuration information input by a user for the preset mapping relationship.
- the determining the target chip frequency of each subtask based on the task parameter of each subtask in the plurality of subtasks includes:
- the task parameters are selected from multiple sets of selectable chip frequencies, and the chip frequency that enables the chip to achieve the lowest task running time under the restriction of chip power consumption is selected as the target chip frequency.
- the method further includes: receiving frequency setting strategy information input by a user; and determining the target chip of each subtask based on the task parameter corresponding to each subtask in the plurality of subtasks
- the frequency includes: determining the target chip frequency of each subtask based on the frequency setting strategy information and the task parameter of each subtask of the multiple subtasks.
- the frequency setting strategy information includes enabling or disabling the chip frequency dynamic setting function for subtasks.
- the operating frequency includes at least one of the following: a core frequency of the chip or a memory frequency.
- a device for setting the operating frequency of a chip includes: an acquisition module for acquiring multiple subtasks of a target task and task parameters of each subtask.
- the task parameters include A parameter of the calculation scale of the task; a frequency control module for determining the target chip frequency of each subtask based on the task parameters of each of the multiple subtasks; a frequency setting module for determining each subtask according to the task parameters
- the target chip frequency is to set the working frequency of the chip to execute each subtask.
- an electronic device including: a memory and a processor, the memory is configured to store machine-readable instructions, and the processor is configured to invoke the machine-readable instructions to implement the first aspect of the present disclosure method.
- the device further includes: a chip configured to process each subtask in the target task based on the operating frequency set by the processor.
- a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the processor is prompted to implement the method described in the first aspect of the present disclosure.
- the method and device for setting the working frequency of a chip provided by the embodiments of the present disclosure determine the target chip frequency of each subtask in the target task based on the task parameters of the subtask, so that the chip can use the subtask when executing each subtask.
- Fig. 1 shows a flowchart of a method for setting a chip operating frequency provided by some embodiments of the present disclosure.
- Fig. 2 shows another flowchart of a method for setting a chip operating frequency provided by some embodiments of the present disclosure.
- FIG. 3 shows another flowchart of the process of establishing a preset mapping relationship provided by some embodiments of the present disclosure.
- FIG. 4 shows another flowchart of a method for setting the operating frequency of a chip provided by some embodiments of the present disclosure.
- Fig. 5 shows a block diagram of a device for setting the operating frequency of a chip provided by some embodiments of the present disclosure.
- FIG. 6 shows another block diagram of the device for setting the operating frequency of a chip provided by some embodiments of the present disclosure.
- Fig. 7 shows a block diagram of an electronic device with a chip operating frequency provided by some embodiments of the present disclosure.
- the embodiments of the present disclosure provide a method for setting the operating frequency of a chip in a terminal device.
- the method aims to make the operating performance of the device better when the chip runs at the set operating frequency, for example, tasks running on the device It can be processed at a faster speed.
- the chip can be an artificial intelligence (AI) chip, for example, an AI chip used in a smartphone, an AI chip used in an autonomous vehicle, an AI chip used in edge devices of the Internet of Things, etc. Wait. It may also be other types of chips, such as a CPU (Central Processing Unit) chip, a DSP (Digital Signal Processing), a memory chip, etc., which are not limited in the embodiment of the present disclosure.
- AI artificial intelligence
- Fig. 1 shows a flow chart of a method for setting the operating frequency of a chip provided by some embodiments of the present disclosure. As shown in Fig. 1, the method may include the following processing.
- step 100 a plurality of subtasks of the target task and task parameters of each subtask are acquired, the task parameters including parameters used to indicate the operation scale of the subtasks.
- the target task may be a task to be executed and processed on the device.
- the task may be a training task of a deep learning model, or an inference task of a neural network model, or running an application program, etc.
- the execution of the target task requires the use of a chip on the device, and the chip is responsible for part or all of the processing work such as calculations during the execution of the task.
- the subtask may be a relatively independent execution unit included in the target task.
- the target task can include multiple functions, which can be divided according to functions, and each function is used as an independent subtask in the target task.
- the task code can also be divided into multiple code blocks and subtasks are divided according to the code blocks, and each piece of code is relatively complete.
- a relatively complete or independent code block as a subtask.
- each layer or layers of the neural network model can be regarded as a subtask; or, it can also be the code segment of the realization process of each layer As a subtask.
- one or at least two of the multiple functional modules can also be used as a subtask, and so on.
- the embodiments of the present disclosure do not limit the way of dividing subtasks, and there may be multiple ways of dividing subtasks, such as the above-mentioned function as a unit, code block as a unit, network layer or operator as a unit, or function module as a unit. It should be noted that although dividing the target task into multiple sub-tasks can achieve fine-grained frequency settings and improve equipment operating performance, too many sub-tasks may cause more frequent frequency settings. It has a certain impact on the operating performance of the equipment. Therefore, you can balance the settings of the number of subtasks. For example, try not to divide the inside of the loop body included in the target task into subtasks to reduce the number of frequency settings. The specific settings can be based on actual conditions. Needs to proceed, so I won’t go into details here.
- the task parameter of the subtask may include a parameter used to indicate the calculation scale of the subtask, for example, may include a parameter used to indicate one or more resources that the subtask needs to occupy.
- the task parameters may include but are not limited to at least one of the following: calculation amount of the subtask, memory access (memory access is the number of times the chip accesses the memory and/or the total amount of data accessed when the subtask is executed), Or, it may also include other types of task parameters, which are not limited in the embodiment of the present disclosure.
- the task parameters of the subtasks can be obtained in various ways. For example, obtain task parameters of subtasks from other devices or from other functional modules in the system, or obtain task parameters of subtasks from local storage, or obtain task parameters of each subtask by performing task analysis on the target task. and many more.
- the correspondence between each subtask and the respective task parameter may be stored in advance.
- task analysis processing may be performed on the target task in advance to obtain multiple subtasks included in the target task and task parameters of each subtask, and each subtask and its task parameters are stored. In this way, when the target task needs to be run, the task parameters of each subtask can be quickly found based on the stored information.
- the correspondence between the subtasks and task parameters can be stored in the form of key-value.
- corresponding task IDs can be set for multiple subtasks, and the task IDs of the subtasks can be stored as keys , And store the task parameters of this subtask as value.
- the key-value storage method is suitable for querying through the primary key, with fast query speed and large amount of stored data. But the present disclosure is not limited to this.
- device information of the device where the chip is located can be further obtained, and the target chip frequency of each subtask can be jointly determined based on the device information and task parameters.
- the aforementioned device information may include device resource information of the device where the chip is located.
- the device resource information is used to indicate information about the available resources of the device.
- the device resource information may include, but is not limited to, any of the following or Multiple device resource information: the number of computing units, bandwidth, memory capacity, etc.
- the target chip frequency of each subtask may also depend on the dependency between multiple subtasks or the task execution mode, such as serial, parallel, or serial between partial subtasks and partial subtasks. Parallel and so on.
- the chip can occupy all available device resources on the device when executing each subtask, such as occupying all computing units and all bandwidth on the device. And so on, but the embodiments of the present disclosure are not limited thereto.
- step 102 the target chip frequency of each subtask is determined based on the task parameter of each subtask in the plurality of subtasks.
- the target chip frequency of each subtask can be determined according to the task parameters of the subtask.
- the target chip frequency of each subtask may be jointly determined based on the device information of the device where the chip is located and the task parameters of each subtask.
- multiple sets of optional chip frequencies may be determined, and a corresponding optional chip frequency may be selected for each subtask from the multiple sets of optional chip frequencies based on a certain strategy.
- the multiple sets of optional chip frequencies are a specific number of discrete frequencies that can be set by the chips in the device.
- the core frequency may include m frequency values such as x1, x2...xm.
- the determination of the target chip frequency can be transformed into an optimization problem.
- the task running time can be used as the main reference factor for selection. As an example, it can be based on each sub The task parameters of the task, among the multiple sets of optional chip frequencies that can be set, select the chip frequency that enables the chip to achieve the lowest task running time under the restriction of chip power consumption, as the target chip frequency of each subtask .
- multiple subtasks can be analyzed as a whole to determine the overall task running time of the target task. At this time, under the condition of determining the shortest task running time, each of the multiple subtasks can be obtained at the same time. The target chip frequency of each subtask.
- the task running time of each subtask can be analyzed separately, and the target chip frequency of the subtask can be determined based on the shortest task running time of each subtask as a condition, and so on.
- the system resources consumed by the task, chip operating power consumption, task operating speed, and task execution result accuracy can also be selected as reference factors. This is not limited.
- the target chip frequency is determined according to the task parameters and device information of the subtasks as an example to illustrate the principle of solving an optimization problem.
- the task parameter of the subtask is k n (n represents the n-dimensional space, such as the amount of memory access, the amount of calculation, etc.), and the device information corresponding to the subtask is d m (m represents the m-dimensional space, Such as memory capacity, etc.), and assume that the chip frequency is x, where the frequency can be a core frequency, a memory frequency, or a frequency combination of a core frequency and a memory frequency.
- P(k n , d m , x) represents the operating power consumption of the chip.
- the operating power consumption of the chip is related to task parameters, device information and chip frequency. Even if the task parameters and device information are fixed and the chip frequency changes, P will Will change accordingly. T(k n , d m , x) represents the task running time. Similarly, even if the task parameters and device information are fixed, T will change when the chip frequency changes. Then you can combine the task parameters and device information of the above subtasks with multiple chip frequencies to obtain the corresponding chip operating power consumption and task operating time, and select from multiple chip frequencies based on the chip operating power consumption and task operating time The optimal chip frequency for each subtask.
- the optimal chip frequency can be selected according to the following formula (1):
- the above formula (1) indicates that the lowest task running time is achieved under the condition of chip power consumption limit (not exceeding Power Limited ), which is the optimization goal of the optimization problem. Find the optimal chip frequency with this optimization goal.
- the multiple optional chip frequencies that the chip can set if a certain chip frequency is used, the subtask can be processed with the shortest task running time under the condition of chip power consumption.
- the chip frequency is the target chip frequency of the subtask.
- the target chip frequency of each subtask may also be different. In this step, the target chip frequency that best matches each subtask can be found, so that the subtask is completed at the fastest speed and the chip does not exceed the power consumption.
- step 104 according to the determined target chip frequency of each subtask, the operating frequency of the chip when each subtask is executed is set.
- the chip frequency can be set as the target chip frequency of the subtask, or it can be based on the target chip frequency and the device or chip executing
- the current status information of the subtask for example, the status information of the device or the chip when the chip starts to execute the subtask, such as the chip temperature, is used to set the operating frequency of the chip when the subtask is executed.
- the above process of determining the target chip frequency may be performed uniformly before executing the target task, and the target chip frequency of each of the multiple subtasks is stored, and then, in the process of running each subtask, Set the chip frequency to the target chip frequency of the subtask.
- the target chip frequency of each subtask may also be determined before the execution of each subtask.
- the target chip frequency of the subtask may be determined based on the task parameters of the subtask and the current state information of the device or chip. Frequency.
- the determination of the target chip frequency of each subtask may be performed at different times, which is not limited in the embodiment of the present disclosure.
- the operating frequency of the chip may include at least one of the following: the core frequency of the chip or the memory frequency.
- the method of the embodiment of the present disclosure can be used only to set the core frequency of the chip, where the memory frequency can be set to a fixed value or set in the unit of the target task, or the method is only used to set the memory frequency, where the core
- the frequency can be set as a fixed value or set in the unit of the target task, or this method is used to set the core frequency and memory frequency of the chip, for example, the core frequency and the memory frequency are set as a combination of frequencies, and the target chip of each subtask Frequency is the combination of frequencies.
- the frequency of the memory may be the frequency of a volatile memory, such as the frequency of SDRAM (Synchronous Dynamic Random Access Memory).
- SDRAM Serial Dynamic Random Access Memory
- DDR SDRAM Double Data Rate SDRAM
- DDR2 SDRAM Double Data Rate SDRAM
- DDR3 SDRAM DDR4 SDRAM
- DDR5 SDRAM DDR5 SDRAM
- the frequency of the memory may also be the frequency of a non-volatile memory, such as the frequency of a flash memory.
- the method for setting the working frequency of the chip in the embodiment of the present disclosure determines the target chip frequency of each subtask based on the task parameters of each subtask in the target task, so that the chip can use the target chip of the subtask when executing each subtask. Frequency operation.
- This refined frequency setting method can achieve the optimal operating performance as much as possible when executing each subtask, thereby increasing the operating speed of the entire task.
- Figure 2 provides another flow chart of a method for setting chip operating frequency in some embodiments of the present disclosure.
- the method takes the target chip frequency jointly determined according to the task parameters and device information of the subtasks as an example, and exemplarily shows A way to determine the frequency of the target chip.
- the method may include the following processing, wherein the same steps as those in FIG. 1 will not be described in detail.
- step 200 task analysis is performed on the target task to be run on the device to obtain multiple subtasks and task parameters of each of the subtasks.
- the task parameter of each subtask may include at least one of the following: the calculation amount and the memory access amount of the subtask.
- step 202 the device information of the device where the chip is located is obtained.
- the device information of the device where the chip is located may include device resource information, and the device resource information may include at least one of the following: bandwidth, number of computing units, memory capacity, and so on.
- step 204 a preset mapping relationship between a plurality of first data and a plurality of second data is acquired.
- the device may pre-store a plurality of mapping relationships between a plurality of first data and a plurality of second data, where the first data includes preset task parameters and preset device information of the subtask, and the second data includes preset device information.
- Set the chip frequency may be the core frequency of the chip.
- the preset chip frequency may be a combination of the core frequency of the chip and the memory frequency.
- Table 1 illustrates the preset mapping relationship between multiple first data and multiple second data:
- the following example illustrates how to establish the above-mentioned mapping relationship.
- the values of the task parameters and device information will be within a certain range.
- the value of the task parameter is within the range F1
- the value of the device information is within the range F2.
- Both F1 and F2 can be referred to as the preset first data value range.
- k1, k2, and k3 represent task parameters
- d1, d2, and d3 represent device information.
- the chip frequency to be set is the core frequency, it may include multiple sets of optional chip frequencies that the chip may set, for example, x1, x2, x3, and so on.
- a group of chip frequencies can be selected from the above-mentioned multiple groups of optional chip frequencies as the second data corresponding to the first data, and the first data and the second data can be established.
- the mapping relationship between the data may be based on performance evaluation parameters.
- the performance evaluation parameters of the first data under the condition of each group of optional chip frequencies can be determined separately, for example, a performance evaluation parameter can be obtained according to the first data and one group of optional chip frequencies; according to the first data and Another set of optional chip frequencies obtains another performance evaluation parameter, where the performance evaluation parameter can be obtained through simulation, inference, or mathematical formula operation, which is not limited in the embodiment of the present disclosure.
- a group of chip frequencies can be selected from the multiple sets of optional chip frequencies as the first data according to the performance evaluation parameters
- the corresponding second data may include task processing performance parameters and chip operating power consumption, and the chip operating power consumption of the above multiple sets of optional chip frequencies may be lower than the preset power consumption and the optional chip frequency with the optimal task processing performance parameters As the second data corresponding to the first data.
- the aforementioned task processing performance parameters include but are not limited to task processing time.
- the task processing performance parameter is the task running time
- the chip frequency is the core frequency as an example.
- step 2041 traverse the multiple sets of optional chip frequencies, and obtain the operating time under the conditions of each set of optional chip frequencies and the first data according to each set of optional chip frequencies and a certain set of first data.
- the operating power consumption and task running time of the chip is not limited to, but not limited to, but not limited to, but not limited to, but not limited to, but not limited to, but not limited to, but not limited to, but not limited to, but not limited to, a certain set of first data.
- the ten sets of chip frequencies can be combined respectively to obtain ten chip operating power consumption P and ten task operating time T.
- the chip frequency is x1
- the corresponding chip operating power consumption of the subtask is P1
- the task running time is T1
- the chip frequency is x2
- the corresponding chip operating power consumption of the subtask is P2
- the task is running Time
- P may be determined according to the power consumption model
- T may be determined according to the runtime model.
- the model input of the power consumption model is task parameters, device information and chip frequency
- the model output is P.
- the model inputs of the runtime model are task parameters, device information and chip frequency, and the model outputs T.
- the specific structure of the above-mentioned power consumption model and runtime model can be obtained in a variety of ways, for example, a support vector machine, a feedback neural network, a K-means aggregation algorithm, and so on.
- both the power consumption model and the runtime model are neural network models as an example, and the neural network model can be obtained through model training.
- these sets are used as the training set of the neural network model.
- the task parameters of the task can be obtained, and the device information of the device can be obtained, and then a frequency x 1 is selected from the above device frequency set, and the three are used as the input of the power consumption model.
- the predicted value of the operating power consumption of the chip is output.
- the predicted value is a difference between the predicted value and the real value of the chip's operating power consumption under the condition of input task parameters, device information, and frequency x 1 (the result of real device operation).
- backpropagation is performed to train the function.
- Consumption model In the same way, the runtime model can also be trained in the above manner, but the output is changed to the task runtime, which will not be described in detail here.
- the first data and the chip frequency to the power consumption model to obtain the running power consumption of the chip. For example, by inputting the first data (k1, d1) and the chip frequency xi into the power consumption model, the operating power consumption Pi of the chip can be obtained. Input the first data and chip frequency to the runtime model to get the task runtime. For example, by inputting the first data (k1, d1) and the chip frequency xi into the runtime model, the task runtime Ti can be obtained.
- step 2042 the chip frequency that determines the optimal chip running power consumption and task running time is selected as the second data corresponding to the first data.
- the chip frequency with the lowest task running time within the limit of chip running power consumption can be selected as the second data.
- the chip frequency with the smallest Ti, such as x1 is selected as the second data corresponding to the first data (k1, d1).
- step 2043 a mapping relationship between the first data and the second data is established.
- each set of first data can obtain the second data corresponding to the first data, that is, the optimal chip frequency, according to the process shown in FIG. 3.
- the second data corresponding to each first data in the multiple first data can be obtained, and each mapping relationship is established accordingly, thereby obtaining a set of mapping relationships.
- the set may include multiple sets of mapping relationships, and each set of mapping relationships The relationship includes a first data and a corresponding second data.
- the set of mapping relationships may be the preset mapping relationships described in step 204, as shown in Table 1 as an example.
- step 206 the chip frequency corresponding to the subtask in the preset mapping relationship is determined as the target chip frequency according to the preset mapping relationship, the device information, and the task parameters of the subtask.
- the task parameter of a certain subtask is k1
- the device information corresponding to the subtask is d1.
- the target chip frequency can be obtained as x1.
- This subtask can be called the first subtask.
- the first data in the mapping relationship table is discrete, sometimes the task parameters and device information of the first subtask are not completely consistent with the first data in the mapping relationship. In this case, as an optional implementation In this way, the first data closest to the device information and task parameters of the first task can be found in the mapping relationship table, and the chip frequency corresponding to the closest first data is used as the target chip frequency of the first subtask.
- the distance between the third data and each first data in the mapping relationship can be calculated.
- the distance can be calculated by taking the third data and each first data as a vector, and calculating the distance between the vectors, for example, it can be (Euclidean distance between the vector of the third data and the vector of each first data Or other distances.
- the chip frequency corresponding to the first data closest to the third data distance can be used as the target chip frequency of the first subtask.
- the above-mentioned first data closest to the third data may be the closest downward, and the closest downward refers to taking the first data corresponding to the lower chip frequency as much as possible.
- the first The chip frequency corresponding to the data Y1 is lower than the chip frequency corresponding to the second data Y2, so you can choose to use the chip frequency corresponding to Y1. If the distances between the first data Y1 and the first data Y2 and the third data are not equal, the first data with a relatively close distance can still be selected as the target first data.
- the first data corresponding to the lower chip frequency is preferentially selected.
- the preset range may be based on the actual The requirement setting is not limited in the embodiment of the present disclosure.
- step 208 according to the determined target chip frequency of each subtask, the operating frequency of the chip to execute each subtask is set.
- the frequency of the chip can be set as the target chip frequency of the subtask.
- the task to be executed is decomposed to obtain multiple subtasks, and the target chip frequency of each subtask is obtained respectively.
- This refined frequency setting method can make each task The sub-task part is as far as possible to achieve the best operating performance, thereby improving the running speed of the entire task.
- by pre-establishing the mapping relationship between task parameters, device information and chip frequency it is possible to speed up the determination of the chip frequency and improve the performance of the device.
- FIG. 4 provides another flow chart of the method for setting the working frequency of the chip according to some embodiments of the present disclosure.
- setting the combined frequency of the chip is taken as an example for description, and the obtained device information may also include the chip of the chip Temperature, the chip temperature can be collected by the temperature sensor in the device where the chip is located.
- the target task to be performed is a three-layer neural network model inference task.
- the power consumption model and runtime model obtained by training may be completely different.
- the model can be determined by offline training for the specific end device where the target task is running. After the power consumption model and the runtime model are determined, the determined model and the optimization problem solving engine can be stored in the end device.
- the processing performed by the optimization problem solving engine can be referred to the process of establishing the preset mapping relationship in the process description shown in Figure 2.
- the optimization problem solving engine can obtain the result according to the sampled multiple sets of first data and the determined model Each chip frequency corresponds to the chip running power consumption and task running time, and then the optimal chip frequency is selected according to formula (1), thereby establishing the mapping relationship between the first data and the second data.
- the device may execute the establishment of the preset mapping relationship, or another device may execute the process of establishing the preset mapping relationship, and store the pre-established mapping relationship in the device where the target task is running. The device can directly look up the table when running the target task.
- step 400 task analysis is performed on the target task to be run on the device, and multiple subtasks and task parameters of each of the subtasks are obtained.
- the target task is an inference task of a three-layer neural network model, each of which can be a subtask.
- a subtask list can be obtained.
- the subtask list includes three subtasks: subtask 1, subtask 2, and subtask 3.
- this step also analyzes and obtains the task parameters of each subtask, for example, the amount of memory access and the amount of calculation of the subtask.
- the subtask list may include: ⁇ subtask 1, task parameter 1>, ⁇ subtask 2, task parameter 2>, ⁇ subtask 3, task parameter 3>.
- the task parameters can be expressed as k n (n represents n-dimensional space, such as the amount of memory access and the amount of calculation).
- Subtask 1, subtask 2, and subtask 3 can be regarded as the first subtask, respectively.
- step 402 the device information of the device where the chip is located is obtained.
- the device information may include device resource information.
- the subtask 1, the subtask 2 and the subtask 3 are in a serial relationship, and the device resource information occupied by these subtasks during operation may be the same.
- the bandwidth, the number of computing units, etc. can be the same.
- Device information can be expressed as d m (m represents m-dimensional space, such as bandwidth, memory capacity, etc.).
- step 404 before it is determined that the execution of subtask 1 is about to start, the chip temperature C1 on the chip is collected, and according to the task parameters of subtask 1, device resource information and chip temperature C1, it is determined that the device of subtask 1 is running.
- the device information may also include the chip temperature of the chip, because an excessively high temperature during the execution of the task will also cause the chip to take measures to reduce the frequency.
- the device information in the embodiments of the present disclosure not only includes device resource information such as the number of computing units, bandwidth, and memory capacity, but also includes chip temperature.
- the chip temperature may be dynamically acquired, that is, before each subtask in the running process of the target task starts to execute, the chip temperature corresponding to the subtask is acquired, and the target chip frequency of the subtask is determined accordingly.
- step 406 the working frequency of the chip is set to the target chip frequency of subtask 1, and the execution of subtask 1 is started.
- step 408 before it is determined that the execution of subtask 2 is about to start, the chip temperature C2 on the chip is collected, and according to the task parameters of subtask 2, device resource information and chip temperature C2, it is determined that the device of subtask 2 is running.
- the temperature of the chip changes during the running process of the target task, then the latest chip temperature can be collected and combined with the chip before each subtask is executed during the running process of the target task.
- the temperature determines the target chip frequency for this subtask.
- the chip frequency in the embodiment of the present disclosure may be a combined frequency or one of the combined frequencies, that is, the core frequency and the memory frequency of the chip are set at the same time.
- the core frequency is x and the memory frequency is y.
- the corresponding chip frequency can be (x1, y1); when the subtask is When the task parameters and device information are (k2, d2), the corresponding chip frequency can be (x2, y2).
- the input of the power consumption model can include: k1, d1, x1, y1, and the output of the model can be the chip's operating power consumption P; the same way, the input of the runtime model It can include: k1, d1, x1, y1, and the output of the model can be the task running time T.
- the device information may include the chip temperature.
- step 410 the working frequency of the chip is set to the target chip frequency of subtask 2, and the execution of subtask 2 is started.
- step 412 before it is determined that subtask 3 will be executed, the chip temperature C3 of the chip on the device is collected, and the task parameters of subtask 3, device resource information, and chip temperature C3 are used to determine where subtask 3 is running.
- the chip on the device corresponds to the set target chip frequency.
- step 414 the operating frequency of the chip is set to the target chip frequency of subtask 3, and the execution of subtask 3 is started.
- the operating frequency of the device chip can be reset to the default value, and the execution of the target task ends.
- the task to be executed is decomposed to obtain multiple subtasks, and the target chip frequency of each subtask is obtained respectively.
- This refined frequency setting method can make each subtask of the task Some parts are as far as possible to achieve the best operating performance, thereby improving the running speed of the entire task.
- dynamically collecting the chip temperature during the execution of the first subtask, and synthesizing the chip temperature to determine the target chip operating frequency of the first subtask which can make the determination of the chip operating frequency more comprehensive. Therefore, the frequency setting is more reasonable, and the operation performance of the equipment is further improved.
- the end device may also provide a user interface through which the input configuration information for the preset mapping relationship can be received.
- the user can calculate or estimate the target chip frequency that each subtask should use offline, and store the preset mapping relationship between each subtask and the corresponding target chip frequency in the device, so that the device runs according to the The configured preset mapping relationship only needs to set the working frequency of the corresponding chip when the subtask is executed.
- the aforementioned user interface can also be used to receive frequency setting strategy information.
- the frequency setting strategy information may include, for example, a model for determining chip operating power consumption and task operating time according to task parameters and device information, or how to select and determine the target chip frequency of a certain subtask according to the power consumption model and task operating time model. According to the frequency setting strategy information, based on the frequency setting strategy information and the task parameters of each of the multiple subtasks, the target chip frequency corresponding to the chip on the device when each subtask is running can be determined.
- the frequency setting strategy information may also include: turning on or turning off the chip frequency dynamic setting function for subtasks.
- the device can determine the target chip frequency for each subtask of the target task according to the method for setting chip operating frequency described in the foregoing embodiment of the present disclosure. For example, you can collect task parameters, chip temperature, and device resource information of subtasks, and select and determine the target chip frequency to be used based on these information, power consumption model and task runtime model.
- the frequency setting strategy information includes turning off the chip frequency dynamic setting function for subtasks, the device can directly obtain the target chip frequency that each subtask should use according to the preset mapping relationship configured offline and calculated through the user interface.
- a more flexible chip frequency setting method can be provided, and the user can update the chip frequency determination method more conveniently, making the chip frequency determination more reasonable and faster.
- FIG. 5 is an exemplary structure diagram of a device for setting the operating frequency of a chip provided by an embodiment of the present disclosure. As shown in FIG. 5, the device may include: an acquisition module 51, a frequency control module 52, and a frequency setting module 53.
- the obtaining module 51 is configured to obtain multiple subtasks of the target task and task parameters of each subtask, where the task parameters include parameters for representing the operation scale of the subtasks.
- the task parameter includes at least one of the following: a calculation amount of the subtask, and a memory access amount of the subtask.
- the frequency control module 52 is configured to determine the target chip frequency of each subtask based on the task parameter of each subtask in the multiple subtasks.
- the frequency setting module 53 is configured to set the operating frequency of the chip to execute each subtask according to the determined target chip frequency of each subtask.
- the device may further include: a task analysis module 54 for performing task analysis processing on the target task to obtain the multiple subtasks and task parameters of each subtask; and storing Correspondence between each subtask and the task parameter of each subtask in the plurality of subtasks.
- the obtaining module 51 when used to obtain multiple subtasks of the target task and the task parameters of each subtask, includes: searching the task parameters of each subtask from the stored correspondence relationship.
- the task analysis module 54 can first perform task analysis processing on the target task, such as parsed to obtain multiple subtasks, and then analyze each of the multiple subtasks separately Obtain the task parameters of each subtask.
- the task analysis module 54 can store the corresponding relationship between each subtask obtained by the analysis and its task parameters.
- the obtaining module 51 may be responsible for controlling the running process of the target task. For example, the obtaining module 51 may obtain the corresponding relationship parsed and stored by the task analyzing module 54 described above, and control the execution of each subtask therein one by one. Exemplarily, assuming there are three subtasks, when the first subtask is to be executed, the obtaining module 51 can obtain the task parameters of the first subtask from the corresponding relationship, and combine the task parameters and the device of the device where the chip is located. The information is sent to the frequency control module 52, and the frequency control module 52 determines the target chip frequency of the first subtask according to task parameters and device information.
- the frequency control module 52 may send the determined target chip frequency to the frequency setting module 53, and the frequency setting module 53 may set the operating frequency of the chip to the target chip frequency.
- the frequency setting module 53 may send a feedback signal to the acquiring module 51 after setting the working frequency of the chip to notify the acquiring module 51 that the frequency setting has been completed. Then, the acquisition module 51 can start to execute the first subtask according to the feedback signal.
- the acquisition module 51 can start to prepare to execute the second subtask. Similarly, before the execution of the second subtask, the acquisition module 51 can obtain the first subtask from the correspondence obtained by the task analysis module 54.
- the task parameters of the two subtasks are sent to the frequency control module 52 to determine the target chip frequency of the second subtask. After the frequency setting module 53 feedbacks that the chip frequency setting is completed, the acquiring module 51 starts to execute the second subtask.
- the implementation of the third subtask is similar, and will not be detailed again.
- the acquiring module 51 After the acquiring module 51 confirms that all the subtasks of the target task have been executed, it can send a frequency reset signal to the frequency control module 52, and the frequency control module 52 then sends a frequency reset signal to the frequency setting module 53, and the frequency setting module 53
- the operating frequency of the chip can be set to the default value. At this point, the execution of the target task is over.
- the frequency control module 52 when used to determine the target chip frequency of each subtask based on the task parameters of each of the multiple subtasks, it includes: acquiring device information of the device where the chip is located, The device information includes device resource information; based on the device information and the task parameter of each subtask of the multiple subtasks, the target chip frequency of each subtask is determined.
- the device resource information includes any one or more of the following: the number of computing units, bandwidth, and memory capacity.
- the device information further includes: the chip temperature of the chip; the frequency control module 52 is further configured to: based on the task parameters of the first subtask among the plurality of subtasks, the device resource information, and The chip temperature during execution of the first subtask determines the target chip frequency of the first subtask.
- the frequency control module 52 is configured to: obtain a preset mapping relationship between a plurality of first data and a plurality of second data, the first data includes preset task parameters and preset device information, so The second data includes a preset chip frequency; the target chip frequency of each subtask is determined according to the preset mapping relationship, the device information, and the task parameters of each subtask.
- the frequency control module 52 when used to determine the target chip frequency of each subtask according to the preset mapping relationship, the device information, and the task parameters of each subtask, it includes: determining a third The distance between the data and each first data in the plurality of first data in the preset mapping relationship, the third data includes task parameters and device information of the first subtask among the plurality of subtasks; The preset chip frequency corresponding to the target first data closest to the third data in the preset mapping relationship is used as the target chip frequency of the first subtask.
- the frequency control module 52 is further configured to obtain multiple sets of selectable chip frequencies before obtaining the preset mapping relationship between the multiple first data and the multiple second data, and obtain the discrete discrete data obtained by sampling.
- a plurality of first data for each of the first data in the plurality of first data, a set of chip frequencies is selected from the plurality of sets of selectable chip frequencies as the chip frequency corresponding to each of the first data Second data, and establish a mapping relationship between each of the first data and the selected second data.
- the preset chip frequency corresponding to the first data in the preset mapping relationship is preset based on the first data under the condition of each group of selectable chip frequencies in a plurality of sets of selectable chip frequencies
- the corresponding performance evaluation parameters are selected from the multiple sets of optional chip frequencies.
- the performance evaluation parameters include task processing performance parameters and chip operating power consumption;
- the preset chip frequency corresponding to the first data in the preset mapping relationship is: the multiple sets of selectable chip frequencies
- the operating power consumption of the medium chip is lower than the preset power consumption and the optional chip frequency with the best task processing performance parameters.
- the device may further include: an interface module 55, configured to receive configuration information of the preset mapping relationship input by the user.
- the frequency control module 52 when used to determine the target chip frequency of each subtask based on the task parameters of each subtask of the plurality of subtasks, it includes: The task parameters of the task are selected from multiple sets of selectable chip frequencies, and the chip frequency that enables the chip to achieve the lowest task running time under the limitation of chip power consumption is selected as the target chip frequency.
- the interface module 55 is further configured to: receive frequency setting strategy information input by the user; the frequency control module 52 is further configured to: based on the frequency setting strategy information and each of the multiple subtasks The task parameters to determine the target chip frequency of each subtask.
- the frequency setting strategy information includes enabling or disabling the chip frequency dynamic setting function for subtasks.
- the operating frequency includes at least one of the following: a core frequency of the chip or a memory frequency.
- the above-mentioned apparatus may be used to execute any corresponding method described above, and for the sake of brevity, it will not be repeated here.
- the electronic device 700 includes a memory 710 and a processor 720.
- the memory 710 is used to store machine-readable instructions 711, and the processor 720 is used to call
- the machine-readable instruction 711 implements the method for setting the operating frequency of the chip in any embodiment of this specification.
- the target chip for chip operating frequency setting may be the memory 710 and/or the processor 720.
- the electronic device 700 may further include a communication interface 730 and a bus 740.
- the memory 710, the processor 720, and the communication interface 730 are connected to each other through a bus 740.
- the electronic device 700 may further include a chip 750 configured to process each subtask in the target task based on the operating frequency set by the processor 720.
- the embodiment of the present disclosure also provides a computer-readable storage medium on which a computer program is stored.
- the program is executed by a processor, the processor is prompted to implement the chip operating frequency setting method of any embodiment of this specification .
- the computer-readable storage medium may be the memory 710 in FIG. 7.
- one or more embodiments of the present disclosure may be provided as a method, a system, or a computer program product. Therefore, one or more embodiments of the present disclosure may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, one or more embodiments of the present disclosure may adopt computer programs implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The form of the product.
- computer-usable storage media including but not limited to disk storage, CD-ROM, optical storage, etc.
- “and/or” in the embodiments of the present disclosure means having at least one of the two, for example, “multi and/or B” includes three schemes: multi, B, and "multi and B".
- the embodiments of the subject and functional operations described in the present disclosure can be implemented in the following: digital electronic circuits, tangible computer software or firmware, computer hardware including the structures disclosed in the present disclosure and structural equivalents thereof, or among them A combination of one or more.
- the embodiments of the subject matter described in the present disclosure may be implemented as one or more computer programs, that is, one or one of the computer program instructions encoded on a tangible non-transitory program carrier to be executed by a data processing device or to control the operation of the data processing device Multiple modules.
- the program instructions may be encoded on artificially generated propagated signals, such as machine-generated electrical, optical, or electromagnetic signals, which are generated to encode information and transmit it to a suitable receiver device for data transmission.
- the processing device executes.
- the computer storage medium may be a machine-readable storage device, a machine-readable storage substrate, a random or serial access memory device, or a combination of one or more of them.
- the processing and logic flow described in the present disclosure can be executed by one or more programmable computers executing one or more computer programs to perform corresponding functions by operating according to input data and generating output.
- the processing and logic flow can also be executed by a dedicated logic circuit, such as FPGA (Field Programmable Gate Array) or ASIC (Application Specific Integrated Circuit), and the device can also be implemented as a dedicated logic circuit.
- FPGA Field Programmable Gate Array
- ASIC Application Specific Integrated Circuit
- Computers suitable for executing computer programs include, for example, general-purpose and/or special-purpose microprocessors, or any other type of central processing unit.
- the central processing unit will receive instructions and data from a read-only memory and/or a random access memory.
- the basic components of a computer include a central processing unit for implementing or executing instructions and one or more memory devices for storing instructions and data.
- a computer will also include one or more mass storage devices for storing data, such as magnetic disks, magneto-optical disks, or optical disks, or the computer will be operatively coupled with this mass storage device to receive data from or send data to it. It transmits data, or both.
- the computer does not have to have such equipment.
- the computer can be embedded in another device, such as a mobile phone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a global positioning system (GPS) receiver, or, for example, a universal serial bus (USB ) Flash drives are portable storage devices, just to name a few.
- PDA personal digital assistant
- GPS global positioning system
- USB universal serial bus
- Computer-readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media, and memory devices, including, for example, semiconductor memory devices (such as EPROM, EEPROM, and flash memory devices), magnetic disks (such as internal hard disks or Removable disks), magneto-optical disks, CD ROM and DVD-ROM disks.
- semiconductor memory devices such as EPROM, EEPROM, and flash memory devices
- magnetic disks such as internal hard disks or Removable disks
- magneto-optical disks CD ROM and DVD-ROM disks.
- the processor and the memory can be supplemented by or incorporated into a dedicated logic circuit.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Power Sources (AREA)
Abstract
Description
Claims (35)
- 一种芯片工作频率的设置方法,其特征在于,所述方法包括:A method for setting the operating frequency of a chip, characterized in that the method includes:获取目标任务的多个子任务以及每个子任务的任务参数,所述任务参数包括用于表示所述子任务的运算规模的参数;Acquiring multiple subtasks of the target task and task parameters of each subtask, where the task parameters include parameters used to indicate the operation scale of the subtasks;基于所述多个子任务中每个子任务的任务参数,确定所述每个子任务的目标芯片频率;Determine the target chip frequency of each subtask based on the task parameter of each subtask in the plurality of subtasks;根据确定的每个子任务的目标芯片频率,设置芯片执行所述每个子任务的工作频率。According to the determined target chip frequency of each subtask, the working frequency of the chip to execute each subtask is set.
- 根据权利要求1所述的方法,其特征在于,所述方法还包括:The method according to claim 1, wherein the method further comprises:对所述目标任务进行任务解析处理,得到所述多个子任务以及每个子任务的任务参数;Performing task analysis processing on the target task to obtain the multiple subtasks and the task parameters of each subtask;存储所述多个子任务中每个子任务与所述每个子任务的任务参数的对应关系;Storing the correspondence between each subtask and the task parameter of each subtask among the plurality of subtasks;所述获取目标任务的多个子任务以及每个子任务的任务参数包括:从存储的所述对应关系中查找每个子任务的任务参数。The obtaining the multiple subtasks of the target task and the task parameters of each subtask includes: searching the task parameters of each subtask from the stored correspondence relationship.
- 根据权利要求1或2所述的方法,其特征在于,所述任务参数包括如下至少一项:所述子任务的计算量、所述子任务的访存量。The method according to claim 1 or 2, wherein the task parameter includes at least one of the following: a calculation amount of the subtask, and a memory access amount of the subtask.
- 根据权利要求1至3中任一项所述的方法,其特征在于,所述基于所述多个子任务中每个子任务的任务参数,确定所述每个子任务的目标芯片频率,包括:The method according to any one of claims 1 to 3, wherein the determining the target chip frequency of each subtask based on the task parameter of each subtask in the plurality of subtasks comprises:获取所述芯片所在设备的设备信息,所述设备信息包括设备资源信息;Acquiring device information of the device where the chip is located, where the device information includes device resource information;基于所述设备信息和所述多个子任务中每个子任务的任务参数,确定所述每个子任务的目标芯片频率。Determine the target chip frequency of each subtask based on the device information and the task parameter of each subtask of the multiple subtasks.
- 根据权利要求4所述的方法,其特征在于,所述设备资源信息包括以下中的任意一项或多项:计算单元的数量、带宽、存储器容量。The method according to claim 4, wherein the device resource information includes any one or more of the following: the number of computing units, bandwidth, and memory capacity.
- 根据权利要求4或5所述的方法,其特征在于,所述设备信息还包括:所述芯片的芯片温度;The method according to claim 4 or 5, wherein the device information further comprises: the chip temperature of the chip;所述基于所述设备信息和所述多个子任务中每个子任务的任务参数,确定所述每个子任务的目标芯片频率,包括:The determining the target chip frequency of each subtask based on the device information and the task parameter of each subtask of the multiple subtasks includes:基于所述多个子任务中第一子任务的任务参数、所述设备资源信息和所述第一子任务执行时的芯片温度,确定所述第一子任务的目标芯片频率。The target chip frequency of the first subtask is determined based on the task parameters of the first subtask among the plurality of subtasks, the device resource information, and the chip temperature when the first subtask is executed.
- 根据权利要求4至6任一项所述的方法,其特征在于,所述基于所述设备信息和所述多个子任务中每个子任务的任务参数,确定所述每个子任务的目标芯片频率,包括:The method according to any one of claims 4 to 6, wherein the determining the target chip frequency of each subtask based on the device information and the task parameters of each subtask of the plurality of subtasks, include:获取多个第一数据和多个第二数据之间的预设映射关系,所述第一数据包括预设任务参数和预设设备信息,所述第二数据包括预设芯片频率;Acquiring a preset mapping relationship between a plurality of first data and a plurality of second data, where the first data includes preset task parameters and preset device information, and the second data includes a preset chip frequency;根据所述预设映射关系、所述设备信息以及每个子任务的任务参数,确定所述每个子任务的目标芯片频率。Determine the target chip frequency of each subtask according to the preset mapping relationship, the device information, and the task parameters of each subtask.
- 根据权利要求7所述的方法,其特征在于,所述根据所述预设映射关系、所述设备信息以及每个子任务的任务参数,确定所述每个子任务的目标芯片频率,包括:The method according to claim 7, wherein the determining the target chip frequency of each subtask according to the preset mapping relationship, the device information, and the task parameters of each subtask comprises:确定第三数据与所述预设映射关系中的多个第一数据中每个第一数据之间的距离,所述第三数据包括所述多个子任务中第一子任务的任务参数和设备信息;Determine the distance between the third data and each first data in the plurality of first data in the preset mapping relationship, where the third data includes task parameters and equipment of the first subtask among the plurality of subtasks information;将所述预设映射关系中与所述第三数据距离最近的第一数据对应的预设芯片频率,作为所述第一子任务的目标芯片频率。The preset chip frequency corresponding to the first data closest to the third data in the preset mapping relationship is used as the target chip frequency of the first subtask.
- 根据权利要求7或8所述的方法,其特征在于,所述获取多个第一数据和多个第二数据之间的预设映射关系之前,所述方法还包括:The method according to claim 7 or 8, characterized in that, before the obtaining the preset mapping relationship between the plurality of first data and the plurality of second data, the method further comprises:获取多组可选芯片频率,并获取采样得到的离散的多个第一数据;Obtain multiple sets of optional chip frequencies, and obtain multiple discrete first data obtained by sampling;对于所述多个第一数据中的每个所述第一数据,由所述多组可选芯片频率中选择一组芯片频率作为与所述每个第一数据对应的第二数据,并建立所述每个第一数据和选择的所述第二数据之间的映射关系。For each of the plurality of first data, select a set of chip frequencies from the plurality of sets of selectable chip frequencies as the second data corresponding to each of the first data, and establish The mapping relationship between each of the first data and the selected second data.
- 根据权利要求7至9中任一项所述的方法,其特征在于,所述预设映射关系中与所述第一数据对应的预设芯片频率是基于在多组可选芯片频率中每组可选芯片频率的条件下所述第一数据对应的性能评估参数,从所述多组可选芯片频率中选择的。The method according to any one of claims 7 to 9, wherein the preset chip frequency corresponding to the first data in the preset mapping relationship is based on each group of multiple sets of selectable chip frequencies The performance evaluation parameter corresponding to the first data under the condition of the selectable chip frequency is selected from the multiple sets of selectable chip frequencies.
- 根据权利要求10所述的方法,其特征在于,所述性能评估参数包括任务处理性能参数和芯片运行功耗;The method according to claim 10, wherein the performance evaluation parameters include task processing performance parameters and chip operating power consumption;所述预设映射关系中与所述第一数据对应的预设芯片频率为:所述多组可选芯片频率中芯片运行功耗低于预设功耗且任务处理性能参数最优的可选芯片频率。The preset chip frequency corresponding to the first data in the preset mapping relationship is: among the multiple sets of selectable chip frequencies, the operating power consumption of the chip is lower than the preset power consumption and the task processing performance parameter is optimal. Chip frequency.
- 根据权利要求7至11中任一项所述的方法,其特征在于,所述方法还包括:The method according to any one of claims 7 to 11, wherein the method further comprises:接收用户输入的对于所述预设映射关系的配置信息。Receiving configuration information input by the user for the preset mapping relationship.
- 根据权利要求1至12任一项所述的方法,其特征在于,所述基于所述多个子任务中每个子任务的任务参数,确定所述每个子任务的目标芯片频率,包括:The method according to any one of claims 1 to 12, wherein the determining the target chip frequency of each subtask based on the task parameter of each subtask in the plurality of subtasks comprises:基于所述多个子任务中每个子任务的任务参数,由多组可选芯片频率中,选择能够使得所述芯片在芯片功耗限制条件下实现任务运行时间最低的芯片频率,作为所述目标芯片频率。Based on the task parameters of each of the multiple subtasks, from multiple sets of optional chip frequencies, select the chip frequency that enables the chip to achieve the lowest task running time under the restriction of chip power consumption, as the target chip frequency.
- 根据权利要求1至13任一项所述的方法,其特征在于,所述方法还包括:The method according to any one of claims 1 to 13, wherein the method further comprises:接收用户输入的频率设置策略信息;Receive the frequency setting strategy information input by the user;所述基于所述多个子任务中每个子任务的任务参数,确定所述每个子任务的目标芯片频率,包括:基于所述频率设置策略信息和所述多个子任务中每个子任务的任务参数,确定所述每个子任务的目标芯片频率。The determining the target chip frequency of each subtask based on the task parameter of each subtask of the plurality of subtasks includes: setting the strategy information based on the frequency and the task parameter of each subtask of the plurality of subtasks, Determine the target chip frequency of each subtask.
- 根据权利要求14所述的方法,其特征在于,所述频率设置策略信息包括打开或关闭针对子任务的芯片频率动态设置功能。The method according to claim 14, wherein the frequency setting strategy information comprises turning on or turning off the chip frequency dynamic setting function for subtasks.
- 根据权利要求1至15任一项所述的方法,其特征在于,所述工作频率包括如下至少一项:所述芯片的核心频率、或者存储器频率。The method according to any one of claims 1 to 15, wherein the operating frequency includes at least one of the following: a core frequency of the chip or a memory frequency.
- 一种芯片工作频率的设置装置,其特征在于,所述装置包括:A device for setting the operating frequency of a chip, characterized in that the device comprises:获取模块,用于获取目标任务的多个子任务以及每个子任务的任务参数,所述任务参数包括用于表示所述子任务的运算规模的参数;An obtaining module, configured to obtain multiple subtasks of the target task and task parameters of each subtask, the task parameters including parameters used to indicate the operation scale of the subtasks;频率控制模块,用于基于所述多个子任务中每个子任务的任务参数,确定所述每个子任务的目标芯片频率;A frequency control module, configured to determine the target chip frequency of each subtask based on the task parameter of each subtask in the plurality of subtasks;频率设置模块,用于根据确定的每个子任务的目标芯片频率,设置芯片执行所述每个子任务的工作频率。The frequency setting module is used to set the working frequency of the chip to execute each subtask according to the determined target chip frequency of each subtask.
- 根据权利要求17所述的装置,其特征在于,所述装置还包括:The device according to claim 17, wherein the device further comprises:任务解析模块,用于对所述目标任务进行任务解析处理,得到所述多个子任务以及每个子任务的任务参数;并存储所述多个子任务中每个子任务与所述每个子任务的任务参数的对应关系;The task analysis module is used to perform task analysis processing on the target task to obtain the multiple subtasks and the task parameters of each subtask; and store the task parameters of each subtask and each subtask in the multiple subtasks Correspondence;所述获取模块还用于:从存储的所述对应关系中查找每个子任务的任务参数。The acquiring module is further used for searching the task parameters of each subtask from the stored corresponding relationship.
- 根据权利要求17或18所述的装置,其特征在于,所述任务参数包括如下至少一项:所述子任务的计算量、所述子任务的访存量。The device according to claim 17 or 18, wherein the task parameter includes at least one of the following: a calculation amount of the subtask, and a memory access amount of the subtask.
- 根据权利要求17至19任一项所述的装置,其特征在于,所述频率控制模块还用于:The device according to any one of claims 17 to 19, wherein the frequency control module is further configured to:获取所述芯片所在设备的设备信息,所述设备信息包括设备资源信息;Acquiring device information of the device where the chip is located, where the device information includes device resource information;基于所述设备信息和所述多个子任务中每个子任务的任务参数,确定所述每个子任务的目标芯片频率。Determine the target chip frequency of each subtask based on the device information and the task parameter of each subtask of the multiple subtasks.
- 根据权利要求20所述的装置,其特征在于,所述设备资源信息包括以下中的任意一项或多项:计算单元的数量、带宽、存储器容量。The apparatus according to claim 20, wherein the device resource information includes any one or more of the following: the number of computing units, bandwidth, and memory capacity.
- 根据权利要求20或21所述的装置,其特征在于,所述设备信息还包括:所述芯片的芯片温度;The apparatus according to claim 20 or 21, wherein the device information further comprises: the chip temperature of the chip;所述频率控制模块还用于:基于所述多个子任务中的第一子任务的任务参数、所述设备资源信息和所述第一子任务执行时的芯片温度,确定所述第一子任务的目标芯片频率。The frequency control module is further configured to determine the first subtask based on the task parameters of the first subtask among the plurality of subtasks, the device resource information, and the chip temperature during execution of the first subtask The target chip frequency.
- 根据权利要求20至22任一项所述的装置,其特征在于,所述频率控制模块还用于:The device according to any one of claims 20 to 22, wherein the frequency control module is further configured to:获取多个第一数据和多个第二数据之间的预设映射关系,所述第一数据包括预设任务参数和预设设备信息,所述第二数据包括预设芯片频率;Acquiring a preset mapping relationship between a plurality of first data and a plurality of second data, where the first data includes preset task parameters and preset device information, and the second data includes a preset chip frequency;根据所述预设映射关系、所述设备信息以及每个子任务的任务参数,确定所述每个子任务的目标芯片频率。Determine the target chip frequency of each subtask according to the preset mapping relationship, the device information, and the task parameters of each subtask.
- 根据权利要求22所述的装置,其特征在于,所述频率控制模块还用于:The device according to claim 22, wherein the frequency control module is further configured to:确定第三数据与所述预设映射关系中的多个第一数据中每个第一数据之间的距离,所述第三数据包括所述多个子任务中第一子任务的任务参数和设备信息;Determine the distance between the third data and each first data in the plurality of first data in the preset mapping relationship, where the third data includes task parameters and equipment of the first subtask among the plurality of subtasks information;将所述预设映射关系中与所述第三数据距离最近的第一数据对应的预设芯片频率,作为所述第一子任务的目标芯片频率。The preset chip frequency corresponding to the first data closest to the third data in the preset mapping relationship is used as the target chip frequency of the first subtask.
- 根据权利要求23或24所述的装置,其特征在于,所述频率控制模块,还用于:The device according to claim 23 or 24, wherein the frequency control module is further configured to:在获取多个第一数据和多个第二数据之间的预设映射关系之前,获取多组可选芯片频率,并获取采样得到的离散的多个第一数据;Before acquiring the preset mapping relationship between the plurality of first data and the plurality of second data, acquiring a plurality of sets of optional chip frequencies, and acquiring a plurality of discrete first data obtained by sampling;对于所述多个第一数据中的每个所述第一数据,由所述多组可选芯片频率中选择一组芯片频率作为与所述每个第一数据对应的第二数据,并建立所述每个第一数据和选择的所述第二数据之间的映射关系。For each of the first data in the plurality of first data, select a set of chip frequencies from the plurality of selectable chip frequencies as the second data corresponding to each of the first data, and establish The mapping relationship between each of the first data and the selected second data.
- 根据权利要求23至25中任一项所述的装置,其特征在于,预设所述预设映射关系中与所述第一数据对应的预设芯片频率是基于在多组可选芯片频率中每组可选芯片频率的条件下所述第一数据对应的性能评估参数,从所述多组可选芯片频率中选择的。The device according to any one of claims 23 to 25, wherein the preset chip frequency corresponding to the first data in the preset mapping relationship is based on a plurality of sets of selectable chip frequencies The performance evaluation parameter corresponding to the first data under the condition of each set of selectable chip frequencies is selected from the multiple sets of selectable chip frequencies.
- 根据权利要求26所述的装置,其特征在于,The device of claim 26, wherein:所述性能评估参数包括任务处理性能参数和芯片运行功耗;The performance evaluation parameters include task processing performance parameters and chip operating power consumption;所述预设映射关系中与所述第一数据对应的预设芯片频率为:所述多组可选芯片频率中芯片运行功耗低于预设功耗且任务处理性能参数最优的可选芯片频率。The preset chip frequency corresponding to the first data in the preset mapping relationship is: among the multiple sets of selectable chip frequencies, the operating power consumption of the chip is lower than the preset power consumption and the task processing performance parameter is optimal. Chip frequency.
- 根据权利要求23至27中任一项所述的装置,其特征在于,所述装置还包括:The device according to any one of claims 23-27, wherein the device further comprises:接口模块,用于接收用户输入的对于所述预设映射关系的配置信息。The interface module is used to receive the configuration information of the preset mapping relationship input by the user.
- 根据权利要求17至28中任一项所述的装置,其特征在于,所述频率控制模块还用于:The device according to any one of claims 17 to 28, wherein the frequency control module is further configured to:基于所述多个子任务中每个子任务的任务参数,由多组可选芯片频率中,选择能够使得所述芯片在芯片功耗限制条件下实现任务运行时间最低的芯片频率,作为所述目标芯片频率。Based on the task parameters of each of the multiple subtasks, from multiple sets of optional chip frequencies, select the chip frequency that enables the chip to achieve the lowest task running time under the restriction of chip power consumption, as the target chip frequency.
- 根据权利要求28所述的装置,其特征在于,The device of claim 28, wherein:所述接口模块还用于:接收用户输入的频率设置策略信息;The interface module is further configured to: receive frequency setting strategy information input by the user;所述频率控制模块还用于:基于所述频率设置策略信息和所述多个子任务中每个子任务的任务参数,确定所述每个子任务的目标芯片频率。The frequency control module is further configured to determine the target chip frequency of each subtask based on the frequency setting strategy information and the task parameter of each subtask of the multiple subtasks.
- 根据权利要求30所述的装置,其特征在于,所述频率设置策略信息包括打开或关闭针对子任务的芯片频率动态设置功能。The device according to claim 30, wherein the frequency setting strategy information comprises enabling or disabling the chip frequency dynamic setting function for subtasks.
- 根据权利要求17至31任一项所述的装置,其特征在于,所述工作频率包括如下至少一项:所述芯片的核心频率、或者存储器频率。The device according to any one of claims 17 to 31, wherein the operating frequency comprises at least one of the following: a core frequency of the chip or a memory frequency.
- 一种电子设备,其特征在于,包括:存储器、处理器,所述存储器用于存储机器可读指令,所述处理器用于调用所述机器可读指令,实现权利要求1至16任一项所述的方法。An electronic device, comprising: a memory and a processor, the memory is used to store machine-readable instructions, and the processor is used to call the machine-readable instructions to implement any one of claims 1 to 16 The method described.
- 根据权利要求33所述的设备,其特征在于,还包括:The device according to claim 33, further comprising:芯片,用于基于所述处理器设置的工作频率对目标任务中的每个子任务进行处理。The chip is used to process each subtask in the target task based on the operating frequency set by the processor.
- 一种计算机可读存储介质,其上存储有计算机程序,其特征在于,所述程序被处理器执行时,促使所述处理器实现权利要求1至16任一项所述的方法。A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the processor is prompted to implement the method according to any one of claims 1 to 16.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
KR1020217020535A KR20210098508A (en) | 2019-11-29 | 2020-07-28 | Setting the chip operating frequency |
JP2021538698A JP2022516549A (en) | 2019-11-29 | 2020-07-28 | Chip operating frequency setting |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911201455.9 | 2019-11-29 | ||
CN201911201455.9A CN112882819B (en) | 2019-11-29 | 2019-11-29 | Method and device for setting chip working frequency |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2021103618A1 true WO2021103618A1 (en) | 2021-06-03 |
Family
ID=76038734
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2020/105195 WO2021103618A1 (en) | 2019-11-29 | 2020-07-28 | Configuration of operating frequency of chip |
Country Status (5)
Country | Link |
---|---|
JP (1) | JP2022516549A (en) |
KR (1) | KR20210098508A (en) |
CN (1) | CN112882819B (en) |
TW (1) | TWI743934B (en) |
WO (1) | WO2021103618A1 (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114785376B (en) * | 2022-05-06 | 2023-07-21 | Oppo广东移动通信有限公司 | Frequency-voltage pre-configuration method and related device |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2362297A1 (en) * | 2010-02-25 | 2011-08-31 | Telefonaktiebolaget L M Ericsson (publ) | Technique for selecting a frequency of operation in a processor system |
CN102169357A (en) * | 2011-02-23 | 2011-08-31 | 北京大学深圳研究生院 | DSP (Digital Signal Processor) capable of regulating working voltage and clock frequency and regulating method thereof |
CN103955264A (en) * | 2014-05-15 | 2014-07-30 | 乐视致新电子科技(天津)有限公司 | Method and system for dynamically regulating working frequency of processor |
CN104407690A (en) * | 2014-12-19 | 2015-03-11 | 中科创达软件股份有限公司 | Method, device and mobile terminal for regulating operating frequency of CPU (Central Processing Unit) |
US9400518B2 (en) * | 2013-06-05 | 2016-07-26 | Qualcomm Innovation Center, Inc. | Temporary frequency adjustment of mobile device processors based on task migration |
CN107357658A (en) * | 2017-06-29 | 2017-11-17 | 上海斐讯数据通信技术有限公司 | A kind of method and system for adjusting mobile terminal cpu frequency |
Family Cites Families (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPWO2003083693A1 (en) * | 2002-04-03 | 2005-08-04 | 富士通株式会社 | Task scheduling device in distributed processing system |
US7206950B2 (en) * | 2004-06-16 | 2007-04-17 | Matsushita Electric Industrial Co., Ltd. | Processor system, instruction sequence optimization device, and instruction sequence optimization program |
WO2006011189A1 (en) * | 2004-07-26 | 2006-02-02 | Mitsubishi Denki Kabushiki Kaisha | Parallel computer |
US20100064124A1 (en) * | 2006-11-16 | 2010-03-11 | Karl Rinne | Digital power controller |
TWI336453B (en) * | 2006-12-18 | 2011-01-21 | Asustek Comp Inc | Method for adjusting working frequency of chip |
WO2009047853A1 (en) * | 2007-10-11 | 2009-04-16 | Fujitsu Limited | Information processor, operation control method, and operation control program |
US8381004B2 (en) * | 2010-05-26 | 2013-02-19 | International Business Machines Corporation | Optimizing energy consumption and application performance in a multi-core multi-threaded processor system |
KR101885857B1 (en) * | 2012-01-04 | 2018-08-06 | 삼성전자주식회사 | Temperature management unit, system on chip including the same and method of managing temperature in a system on chip |
JP5898342B2 (en) * | 2012-12-21 | 2016-04-06 | ルネサスエレクトロニクス株式会社 | Semiconductor device and control method thereof |
US9494996B2 (en) * | 2013-03-15 | 2016-11-15 | Intel Corporation | Processor having frequency of operation information for guaranteed operation under high temperature events |
CN105045702B (en) * | 2015-07-22 | 2018-03-20 | Tcl移动通信科技(宁波)有限公司 | It is a kind of that the method for chip, system and chip are protected according to chip temperature |
CN106527653A (en) * | 2016-10-12 | 2017-03-22 | 东软集团股份有限公司 | CPU frequency adjusting method and apparatus |
JP2018106591A (en) * | 2016-12-28 | 2018-07-05 | ルネサスエレクトロニクス株式会社 | Semiconductor device, operation control method, and program |
CN108334405A (en) * | 2017-01-20 | 2018-07-27 | 阿里巴巴集团控股有限公司 | Frequency isomery CPU, frequency isomery implementation method, device and method for scheduling task |
US10921874B2 (en) * | 2017-03-06 | 2021-02-16 | Facebook Technologies, Llc | Hardware-based operating point controller for circuit regions in an integrated circuit |
CN109212306B (en) * | 2017-07-06 | 2021-02-26 | 龙芯中科技术股份有限公司 | Method, circuit and device for adjusting chip power consumption |
CN107562682A (en) * | 2017-08-01 | 2018-01-09 | 华南理工大学 | Communication means based on many-core chip coker heat energy in the heart |
CN107678855B (en) * | 2017-09-19 | 2020-06-12 | 中国电子产品可靠性与环境试验研究所 | Processor dynamic adjustment method and device and processor chip |
CN108268265A (en) * | 2018-01-24 | 2018-07-10 | 深圳市道通科技股份有限公司 | Link mapping method, software code method for burn-recording and the burning host in common code library |
-
2019
- 2019-11-29 CN CN201911201455.9A patent/CN112882819B/en active Active
-
2020
- 2020-07-28 KR KR1020217020535A patent/KR20210098508A/en not_active Application Discontinuation
- 2020-07-28 WO PCT/CN2020/105195 patent/WO2021103618A1/en active Application Filing
- 2020-07-28 JP JP2021538698A patent/JP2022516549A/en active Pending
- 2020-08-11 TW TW109127158A patent/TWI743934B/en active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2362297A1 (en) * | 2010-02-25 | 2011-08-31 | Telefonaktiebolaget L M Ericsson (publ) | Technique for selecting a frequency of operation in a processor system |
CN102169357A (en) * | 2011-02-23 | 2011-08-31 | 北京大学深圳研究生院 | DSP (Digital Signal Processor) capable of regulating working voltage and clock frequency and regulating method thereof |
US9400518B2 (en) * | 2013-06-05 | 2016-07-26 | Qualcomm Innovation Center, Inc. | Temporary frequency adjustment of mobile device processors based on task migration |
CN103955264A (en) * | 2014-05-15 | 2014-07-30 | 乐视致新电子科技(天津)有限公司 | Method and system for dynamically regulating working frequency of processor |
CN104407690A (en) * | 2014-12-19 | 2015-03-11 | 中科创达软件股份有限公司 | Method, device and mobile terminal for regulating operating frequency of CPU (Central Processing Unit) |
CN107357658A (en) * | 2017-06-29 | 2017-11-17 | 上海斐讯数据通信技术有限公司 | A kind of method and system for adjusting mobile terminal cpu frequency |
Also Published As
Publication number | Publication date |
---|---|
CN112882819B (en) | 2022-03-08 |
KR20210098508A (en) | 2021-08-10 |
TW202121116A (en) | 2021-06-01 |
CN112882819A (en) | 2021-06-01 |
TWI743934B (en) | 2021-10-21 |
JP2022516549A (en) | 2022-02-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11586473B2 (en) | Methods and apparatus for allocating a workload to an accelerator using machine learning | |
US20170286484A1 (en) | Graph Data Search Method and Apparatus | |
WO2018102240A1 (en) | Joint language understanding and dialogue management | |
CN109558945A (en) | The method and device that artificial neural network and floating-point neural network are quantified | |
KR102521054B1 (en) | Method of controlling computing operations based on early-stop in deep neural network | |
CN112513886B (en) | Information processing method, information processing apparatus, and information processing program | |
CN103309738A (en) | User job scheduling method and device | |
TWI775210B (en) | Data dividing method and processor for convolution operation | |
CN113645637B (en) | Method and device for unloading tasks of ultra-dense network, computer equipment and storage medium | |
WO2021103618A1 (en) | Configuration of operating frequency of chip | |
CN112862112A (en) | Federal learning method, storage medium, terminal, server, and federal learning system | |
CN113159188B (en) | Model generation method, device, equipment and storage medium for image classification | |
CN116501505B (en) | Method, device, equipment and medium for generating data stream of load task | |
CN108897619B (en) | Multi-level resource flexible configuration method for super computer | |
WO2018168695A1 (en) | Distributed machine learning device, distributed machine learning method, and distributed machine learning recording medium | |
CN112115667A (en) | FPGA layout method, device, electronic equipment and computer readable medium | |
CN112988275B (en) | Task perception-based mobile edge computing multi-user computing unloading method | |
CN113961267B (en) | Service processing method, device and equipment | |
US11475311B2 (en) | Neural network instruction streaming | |
CN114675975A (en) | Job scheduling method, device and equipment based on reinforcement learning | |
WO2022178731A1 (en) | Operating method and apparatus for accelerator | |
CN117407155A (en) | Resource scheme determining method and device, storage medium and electronic equipment | |
CN117667383A (en) | Resource-constrained-edge-oriented distributed data stream optimization and training optimization method | |
TW202328984A (en) | Control method and system based on layer-wise adaptive channel pruning | |
JP2016162400A (en) | Mapping information generation program, mapping information generation method, and mapping information generation device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
ENP | Entry into the national phase |
Ref document number: 2021538698 Country of ref document: JP Kind code of ref document: A Ref document number: 20217020535 Country of ref document: KR Kind code of ref document: A |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 20892245 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 20892245 Country of ref document: EP Kind code of ref document: A1 |
|
32PN | Ep: public notification in the ep bulletin as address of the adressee cannot be established |
Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 16.01.2023) |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 20892245 Country of ref document: EP Kind code of ref document: A1 |