WO2020093205A1 - Deep learning computation method and related device - Google Patents

Abstract

Disclosed in the present application are a deep learning computation method and a related device, which are applied to an electronic device; the method comprises: determining a target model required for the present instance of deep learning computation, the target model comprising multiple layers of model data, and the multiple layers of model data being arranged in sequential order in the target model; on a first thread, sequentially reading the multiple layers of model data according to the set sequence; and when a layer of model data is completely read, performing computation on a second thread on the basis of the layer of model data that is completely read, and repeating said step until all computation is complete. By employing the embodiments of the present application, the phenomenon of running slowly is resolved.

Description

Deep learning calculation method and related equipment Technical field

This application relates to the field of electronic technology, in particular to a deep learning calculation method and related equipment.

Background technique

With the continuous development of electronic technology, electronic devices (such as smart phones, tablet computers, etc.) are becoming more and more intelligent. The intelligentization of electronic devices depends on the deep learning calculations of electronic devices. The electronic device needs to read the model for deep learning calculation, and then calculate based on the read model. At present, when the read model is large, it is easy to cause the electronic device to freeze.

Summary of the invention

Embodiments of the present application provide a deep learning calculation method and related equipment, which are used to solve the lag phenomenon.

In a first aspect, an embodiment of the present application provides a deep learning calculation method, which is applied to an electronic device and includes:

Determine the target model required for this deep learning calculation, the target model includes multi-layer model data, and the multi-layer model data is arranged in the target model according to a set order;

Sequentially read in the multi-layer model data according to the set order on the first thread;

When the layer of model data is read in, the calculation is performed on the second thread based on the layer of model data read in, and this step is repeated until all calculations are completed.

In a second aspect, an embodiment of the present application provides a deep learning computing device, which is applied to an electronic device and includes:

A model determining unit, configured to determine a target model required for this deep learning calculation, the target model includes multi-layer model data, and the multi-layer model data is arranged in the target model according to a set order;

A data reading unit for sequentially reading the multi-layer model data according to the set sequence on the first thread;

The calculation unit is used to perform calculation on the second thread based on the completed layer of model data when the layer of model data is read in, and repeat this step until all calculations are completed.

In a third aspect, an embodiment of the present application provides an electronic device, including a processor, a memory, a communication interface, and one or more programs, where the one or more programs are stored in the memory and configured to be processed by the above The above program includes instructions for performing the steps in the method described in the first aspect of the embodiments of the present application.

According to a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program for electronic data exchange, wherein the computer program causes the computer to execute the first embodiment of the present application. Part or all of the steps described in the method described in one aspect.

According to a fifth aspect, an embodiment of the present application provides a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing the computer program, and the computer program is operable to cause the computer to execute as implemented in the present application Examples of some or all of the steps described in the method described in the first aspect. The computer program product may be a software installation package.

It can be seen that in the embodiment of the present application, the model includes multiple layers of model data, a layer of model data is read on the first thread, and then after reading a layer of model data, based on the second thread The model data after reading is calculated until all calculations are completed. Compared with the prior art, the entire model data is read at one time and then calculated, and the application layer-by-layer model calculation and calculation are performed in layers to avoid the phenomenon of jamming caused by the one-time reading of the model.

BRIEF DESCRIPTION

In order to more clearly explain the embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings required in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only These are some embodiments of the present application. For those of ordinary skill in the art, without paying any creative work, other drawings can be obtained based on these drawings.

1 is a schematic structural diagram of an electronic device provided by an embodiment of the present application;

2 is a schematic flowchart of a deep learning calculation method provided by an embodiment of the present application;

3 is a schematic flowchart of another deep learning calculation method disclosed in an embodiment of the present application;

4 is a schematic structural diagram of an electronic device disclosed in an embodiment of the present application;

FIG. 5 is a block diagram of functional units of a deep learning computing device disclosed in an embodiment of the present application.

detailed description

In order to enable those skilled in the art to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only It is a part of the embodiments of this application, but not all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the scope of protection of this application.

The details are described below.

The terms "first", "second", "third", and "fourth" in the description and claims of the present application and the accompanying drawings are used to distinguish different objects, not to describe a specific order . In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally also includes Other steps or units inherent to these processes, methods, products or equipment.

Reference herein to "embodiments" means that specific features, structures, or characteristics described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art understand explicitly and implicitly that the embodiments described herein can be combined with other embodiments.

Electronic devices can include various handheld devices with wireless communication capabilities, in-vehicle devices, wearable devices, computing devices or other processing devices connected to wireless modems, as well as various forms of user equipment (User Equipment, UE), Mobile Station (MS), terminal device (terminal) and so on.

As shown in FIG. 1, FIG. 1 is a schematic structural diagram of an electronic device according to an embodiment of the present application. The electronic device includes a processor, a memory, a signal processor, a transceiver, a display screen, a speaker, a microphone, a random access memory (Random Access Memory, RAM), a camera, a sensor, and so on. Among them, the memory, signal processor, display screen, speaker, microphone, RAM, camera and sensor are connected to the processor, and the transceiver is connected to the signal processor.

Among them, the display screen may be a liquid crystal display (Liquid Crystal Display), organic or inorganic light-emitting diode (Organic Light-Emitting Diode, OLED), active matrix organic light-emitting diode panel (Active Matrix / Organic Light Emitting Diode, AMOLED )Wait.

Wherein, the camera may be an ordinary camera or an infrared camera, which is not limited herein. The camera may be a front camera or a rear camera, which is not limited herein.

Among them, the sensor includes at least one of the following: a light sensor, a gyroscope, an infrared proximity sensor, a fingerprint sensor, a pressure sensor, and so on. Among them, the light sensor, also known as the ambient light sensor, is used to detect the ambient light brightness. The light sensor may include a photosensitive element and an analog-to-digital converter. Among them, the photosensitive element is used to convert the collected optical signal into an electrical signal, and the analog-to-digital converter is used to convert the aforementioned electrical signal into a digital signal. Optionally, the light sensor may further include a signal amplifier, and the signal amplifier may amplify the electrical signal converted by the photosensitive element and output it to the analog-to-digital converter. The photosensitive element may include at least one of a photodiode, a phototransistor, a photoresistor, and a silicon photovoltaic cell.

Among them, the processor is the control center of the electronic device, using various interfaces and lines to connect the various parts of the entire electronic device, by running or executing the software programs and / or modules stored in the memory, and calling the data stored in the memory, Perform various functions and process data of electronic devices to monitor the electronic devices as a whole.

Among them, the processor may integrate an application processor and a modem processor, wherein the application processor mainly handles an operating system, a user interface, an application program, etc., and the modem processor mainly handles wireless communication. It can be understood that the foregoing modem processor may not be integrated into the processor.

Among them, the memory is used to store software programs and / or modules, and the processor executes various functional applications and data processing of the electronic device by running the software programs and / or modules stored in the memory. The memory may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, software programs required for at least one function, and the like; the storage data area may store data created according to the use of electronic devices and the like. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one magnetic disk storage device, a flash memory device, or other volatile solid-state storage devices.

Please refer to FIG. 2. FIG. 2 is a schematic flowchart of a deep learning calculation method provided by an embodiment of the present application, which is applied to an electronic device. The method includes:

Step 201: Determine a target model required for this deep learning calculation. The target model includes multi-layer model data, and the multi-layer model data is arranged in the target model according to a set order.

Among them, the model refers to the problem-solving steps of a class of problems, that is, the algorithm of a class of problems. The model data included in the model refers to the data of the algorithm of a type of problem, then the layer of model data refers to the data of the algorithm to solve a problem.

The target model may be stored in the electronic device or in a storage device associated with the electronic device, which is not limited herein.

Step 202: Read the multi-layer model data in sequence on the first thread according to the set order.

Step 203: When the first layer of model data is read in, calculation is performed on the second thread based on the completed layer of model data read in, and this step is repeated until all calculations are completed.

For example, suppose that the target model includes three layers of model data. The three layers of model data include the first layer model data, the second layer model data, and the third layer model data. The arrangement order of the three layer model data is the first layer model data From the second layer model data to the third layer model data, then read the first layer model data on the first thread, and read the second layer model on the first thread when the first layer model data is completed. The data is calculated on the second thread based on the first layer model data, when the second layer model data is read in, the third layer model data is read on the first thread and the second layer model is based on the second thread The data is calculated, and when the third layer model data is read in, the calculation is performed on the second thread based on the third layer model data.

Some existing deep learning computing frameworks, such as Xiaomi's MACE and Tencent's NCNN, first read all model data synchronously, and then perform calculation after the data reading is completed. This has a drawback. When the model data volume When it is very large (for example, some model data files may have hundreds of MB), it may take several hundred ms to read all the model data at one time. In some scenarios, such as an application that depends on a deep learning computing framework, When the model data is read in, there will be a significant stuck.

In the embodiment of the present application, a layer of model data is read on the first thread, and then after reading the layer of model data, calculation is performed on the second thread based on the read model data. Until all calculations are completed, the phenomenon of jamming caused by reading the model at one time is avoided.

In an implementation manner of the present application, the determining the target model required for this deep learning calculation includes: determining a first scenario of a first application running in the foreground, and determining this time based on the first scenario of the first application The target model required for deep learning calculations.

The models used in different scenes are different, for example, in the preview photo scene, the model is the preview model, in the photo scene, the model is the photo model, and so on.

In an implementation manner of the present application, when the first application meets the first condition, the resources occupied by the first thread and the second thread are resources in a reserved resource pool, and the first condition At least one of the following: the first application is a setting application, the usage priority of the first application is greater than or equal to the setting priority, and the number of times the first application is used in the first setting period is greater than or equal to Equal to the set number of times.

Among them, the setting application refers to an application previously set by the user, for example, one or more game applications, one or more video applications, one or more shopping applications, one or more instant messaging applications, One or more news applications, one or more office applications, etc.

The first set time period is, for example, 1 hour, 5 hours, 1 day, 1 week, 1 month, or other values. The set number of times is, for example, 10 times, 15 times, 21 times, 30 times, or other values.

The application priority may be set by the user in advance or determined by the electronic device, which is not limited herein. The specific method for the electronic device to determine the use priority of the first application includes: the electronic device determining the target number of uses of the first application in the second set period, and determining the location based on the mapping relationship between the use number and the use priority The use priority corresponding to the target use count; or, the electronic device determines the target download count of the first application to be downloaded, and determines the use priority corresponding to the target download count according to the mapping relationship between the download count and the use priority. The second setting period may be the same as the first setting period, or may be different from the first setting period, which is not limited herein.

It can be seen that, in the case where the first application is a more important application, the resources in the reserved resource pool are used for deep learning calculation and learning, thus avoiding other applications from seizing resources with the first application, and further avoiding the phenomenon of jam.

In an implementation manner of the present application, before the sequentially reading in the multi-layer model data according to the set order on the first thread, the method further includes:

It is determined that the size of the target model is greater than or equal to the first threshold, and / or the current memory occupancy rate of the electronic device is greater than or equal to the second threshold.

The first threshold is, for example, 50MB, 60MB, 100MB, 150MB or other values. The second threshold is, for example, 60%, 70%, 75%, 78%, 80%, or other values.

In an implementation manner of the present application, the method further includes:

When the size of the target model is less than the first threshold, and / or the current memory occupancy rate of the electronic device is less than the second threshold, reading the multi-layer model data on a third thread, And after the multi-layer model data is read in, calculation is performed on the third thread based on the multi-layer model data.

The resource occupied by the third thread may be a resource in the reserved resource pool, or may not be a resource in the reserved resource pool, which is not limited herein.

It can be seen that only when the target model is large and / or the current memory occupancy rate of the electronic device is large, the model reading and calculation are performed in layers. When the target model is small and / or the current memory occupancy rate of the electronic device is small, the model can be directly read and then calculated, which realizes the dynamic adjustment of the deep learning calculation method and improves the intelligence of the electronic device.

In an implementation manner of the present application, during the deep learning calculation process of the first scenario of the first application, the method further includes:

If it is detected that the scene running in the foreground is switched from the first scene of the first application to the second scene of the second application, the deep learning calculation of the first scene of the first application is suspended, and the second application is performed The deep learning calculation of the second scene;

If it is not detected that the scene running in the foreground is switched to the first scene of the first application within the set duration, the first thread and the second thread are released;

If it is detected that the scene running in the foreground is switched to the first scene of the first application within the set duration, the deep learning calculation of the first scene of the first application is continued.

Wherein, the setting time period is, for example, 6s, 10s, 30s, 1min or other values. The setting time corresponding to different scenes is different, for example, the setting time corresponding to the preview photo scene is duration 1, the setting time corresponding to the photographing scene is duration 2, the setting duration corresponding to the game battle scene is duration 3, duration 1, Duration 2 and duration 3 are different from each other. Of course, the setting duration corresponding to different scenes can also be the same.

It can be seen that in the process of performing deep learning calculations, if the scene changes, the deep learning calculations performed this time are suspended first. If the scene does not switch back within the set duration, the first thread and the second thread are released directly to The release of this deep learning calculation prevents resources from being occupied and affects the running rate of other applications. If the scene is switched back within a set time period, continue the deep learning calculation to avoid the deep learning calculation from re-running and improve the electronic device Intelligence.

In an implementation manner of the present application, before the suspending the deep learning calculation of the first scene of the first application, the method further includes:

It is determined that the importance level of the second scene of the second application is less than the importance level of the first scene of the first application, and / or the second scene of the second application is a setting scene.

Among them, the setting scene includes, for example, a system desktop scene, a system setting scene, an SMS viewing scene, and so on.

Among them, the importance level of the scene can be customized by the user in advance, or can be determined by the electronic device (for example, the electronic device determines the target number of occurrences of the scene in the third set period, and is determined according to the mapping relationship between the number of occurrences and the important level The importance level corresponding to the number of occurrences of the target) is not limited herein.

It can be seen that when it is determined that the importance level of the second scene of the second application is less than the importance level of the first scene of the first application, and / or the second scene of the second application is a set scene The following indicates that the first scenario of the first application is more important than the second scenario of the second application, and priority is given to ensuring the running experience of the first scenario of the first application to improve the experience of important scenarios.

In an implementation manner of the present application, the method further includes:

When the importance level of the second scene of the second application is greater than the importance level of the first scene of the first application, and / or the second scene of the second application is not a set scene, release the first A thread and the second thread, and performing deep learning calculation of the second scenario of the second application.

It should be noted that the deep learning calculation of the second scene of the second application is the same as the deep learning calculation of the first scene of the first application, and will not be described here.

It can be seen that the importance level of the second scene of the second application is determined to be greater than the importance level of the first scene of the first application, and / or the second scene of the second application is not a set scene In this case, the deep learning calculation of the first scene of the first application is directly released to provide the second scene of the second application with sufficient resources for deep learning calculation, so as to avoid the phenomenon that the resource is occupied and the card is stuck.

Consistent with the embodiment shown in FIG. 2, please refer to FIG. 3. FIG. 3 is a schematic flowchart of a deep learning calculation method provided by an embodiment of the present application. The method is applied to the above electronic device. The method includes:

Step 301: Determine the first scenario of the first application running in the foreground.

Step 302: Determine a target model required for this deep learning calculation based on the first scenario of the first application, the target model includes multi-layer model data, and the multi-layer model data is set according to the target model The order is one after the other.

Step 303: Determine whether the size of the target model is greater than or equal to the first threshold, and / or determine whether the current memory occupancy rate of the electronic device is greater than or equal to the second threshold.

If yes, go to step 304.

If not, go to step 306.

Step 304: Read the multi-layer model data in sequence on the first thread according to the set order.

Step 305: When the reading of the model data of the first layer is completed, calculation is performed on the second thread based on the model data of the layer of the reading, and this step is repeated until all calculations are completed.

Step 306: Read in the multi-layer model data on the third thread, and after the multi-layer model data is read in, perform calculation on the third thread based on the multi-layer model data.

It should be noted that, for the specific implementation process of this embodiment, reference may be made to the specific implementation process described in the foregoing method embodiments, and details are not described herein.

Consistent with the above embodiments shown in FIG. 2 and FIG. 3, please refer to FIG. 4, which is a schematic structural diagram of an electronic device provided by an embodiment of the present application. As shown in the figure, the electronic device includes a processor and a memory , A communication interface, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor, and the program includes instructions for performing the following steps:

Determine the target model required for this deep learning calculation, the target model includes multi-layer model data, and the multi-layer model data is arranged in the target model according to a set order;

Sequentially read in the multi-layer model data according to the set order on the first thread;

When the layer of model data is read in, the calculation is performed on the second thread based on the layer of model data read in, and this step is repeated until all calculations are completed.

It can be seen that in the embodiment of the present application, the model includes multiple layers of model data, a layer of model data is read on the first thread, and then after reading a layer of model data, based on the second thread The model data after reading is calculated until all calculations are completed. Compared with the prior art, the entire model data is read at a time and then calculated, and the application layer-by-layer model reading and calculation are performed in layers to avoid the phenomenon of jamming caused by reading the model at one time.

In an implementation of the present application, in terms of determining the target model required for this deep learning calculation, the above program includes instructions specifically for performing the following steps:

The first scene of the first application running in the foreground is determined, and the target model required for this deep learning calculation is determined based on the first scene of the first application.

In an implementation manner of the present application, when the first application is a setting application, and / or the use priority of the first application is greater than or equal to the setting priority, the first thread and the first application The resources occupied by the two threads are the resources in the reserved resource pool.

In an implementation of the present application, before sequentially reading in the multi-layer model data according to the set order on the first thread, the above program includes instructions that are further used to perform the following steps:

It is determined that the size of the target model is greater than or equal to the first threshold, and / or the current memory occupancy rate of the electronic device is greater than or equal to the second threshold.

In an implementation of the present application, the above program includes instructions that are also used to perform the following steps:

When the size of the target model is less than the first threshold, and / or the current memory occupancy rate of the electronic device is less than the second threshold, reading the multi-layer model data on a third thread, And after the multi-layer model data is read in, calculation is performed on the third thread based on the multi-layer model data.

In an implementation manner of the present application, during the deep learning calculation process of the first scenario where the first application is performed, the foregoing program includes instructions that are further used to perform the following steps:

If it is detected that the scene running in the foreground is switched from the first scene of the first application to the second scene of the second application, the deep learning calculation of the first scene of the first application is suspended;

If it is not detected that the scene running in the foreground is switched to the first scene of the first application within the set duration, the first thread and the second thread are released;

If it is detected that the scene running in the foreground is switched to the first scene of the first application within the set duration, the deep learning calculation of the first scene of the first application is continued.

In an implementation manner of the present application, before the deep learning calculation of the first scenario of the first application is suspended, the above program includes instructions that are further used to perform the following steps:

It is determined that the importance level of the second scene of the second application is less than the importance level of the first scene of the first application, and / or the second scene of the second application is a setting scene.

It should be noted that, for the specific implementation process of this embodiment, reference may be made to the specific implementation process described in the foregoing method embodiments, and details are not described herein.

The above embodiments mainly introduce the solutions of the embodiments of the present application from the perspective of the execution process on the method side. It can be understood that, in order to realize the above-mentioned functions, the electronic device includes a hardware structure and / or a software module corresponding to each function. Those skilled in the art should easily realize that, in combination with the exemplary units and algorithm steps described in the embodiments disclosed herein, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed by hardware or computer software driven hardware depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered beyond the scope of this application.

The embodiments of the present application may divide the functional unit of the electronic device according to the method example, for example, each functional unit may be divided corresponding to each function, or two or more functions may be integrated into one processing unit. The integrated unit may be implemented in the form of hardware or a software functional unit. It should be noted that the division of the units in the embodiments of the present application is schematic, and is only a division of logical functions. In actual implementation, there may be another division manner.

The following is a device embodiment of the present application. The device embodiment of the present application is used to execute the method implemented by the method embodiment of the present application. Please refer to FIG. 5. FIG. 5 is a deep learning computing device provided by an embodiment of the present application, which is applied to an electronic device. The deep learning computing device includes:

The model determination unit 501 is used to determine a target model required for this deep learning calculation. The target model includes multi-layer model data, and the multi-layer model data is arranged in the target model according to a set order;

The data reading unit 502 is configured to sequentially read in the multi-layer model data according to the set order on the first thread;

The calculation unit 503 is configured to perform calculation on the second thread based on the completed layer of model data read in when the layer of model data is read in, and repeat this step until all calculations are completed.

It can be seen that in the embodiment of the present application, the model includes multiple layers of model data, a layer of model data is read on the first thread, and then after reading a layer of model data, based on the second thread The model data after reading is calculated until all calculations are completed. Compared with the prior art, the entire model data is read at one time and then calculated, and the present application performs model reading and calculation in layers to avoid the phenomenon of jamming caused by the one-time reading of the model.

In an implementation manner of the present application, in terms of determining a target model required for this deep learning calculation, the model determining unit 501 is specifically used to:

The first scene of the first application running in the foreground is determined, and the target model required for this deep learning calculation is determined based on the first scene of the first application.

In an implementation manner of the present application, when the first application is a setting application, and / or the use priority of the first application is greater than or equal to the setting priority, the first thread and the first application The resources occupied by the two threads are the resources in the reserved resource pool.

In an implementation manner of the present application, the device further includes:

The first determining unit 504 is configured to determine that the size of the target model is greater than or equal to the first threshold before the data reading unit 502 sequentially reads the multi-layer model data according to the set order on the first thread, And / or the current memory occupancy rate of the electronic device is greater than or equal to the second threshold.

In an implementation manner of the present application, the data reading unit 502 is further used when the size of the target model is less than the first threshold, and / or the current memory occupancy rate of the electronic device is less than the second threshold In the case of, read the multi-layer model data on the third thread;

The calculation unit 503 is further configured to perform calculation based on the multi-layer model data on the third thread after the multi-layer model data is read in.

In an implementation manner of the present application, the device further includes:

The pause unit 505 is configured to pause the deep learning calculation of the first scene of the first application if it is detected that the scene running in the foreground is switched from the first scene of the first application to the second scene of the second application;

The releasing unit 506 is configured to release the first thread and the second thread if the scene running in the foreground is not detected to switch to the first scene of the first application within a set duration;

The execution unit 507 is configured to continue the deep learning calculation of the first scene of the first application if it is detected that the scene running in the foreground is switched to the first scene of the first application within the set duration.

In an implementation manner of the present application, the device further includes:

The second determining unit 508 is configured to determine that the importance level of the second scene of the second application is less than the first of the first application before the pause unit 505 pauses the deep learning calculation of the first scene of the first application The importance level of the scene, and / or the second scene of the second application is a setting scene.

It should be noted that the model determination unit 501, the data reading unit 502, the calculation unit 503, the first determination unit 504, the pause unit 505, the release unit 506, the execution unit 507, and the second determination unit 508 may be implemented by a processor.

An embodiment of the present application further provides a computer storage medium, wherein the computer storage medium stores a computer program for electronic data exchange, and the computer program causes the computer to perform part or all of the steps of any method described in the foregoing method embodiments , The aforementioned computer includes electronic devices.

An embodiment of the present application also provides a computer program product, the computer program product includes a non-transitory computer-readable storage medium that stores the computer program, and the computer program is operable to cause the computer to perform any of the methods described in the foregoing method embodiments Some or all steps of the method. The computer program product may be a software installation package, and the computer includes an electronic device.

It should be noted that, for the sake of simple description, the foregoing method embodiments are all expressed as a series of action combinations, but those skilled in the art should know that this application is not limited by the described action sequence, Because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the involved actions and modules are not necessarily required by this application.

In the above embodiments, the description of each embodiment has its own emphasis. For a part that is not detailed in an embodiment, you can refer to related descriptions in other embodiments.

In the several embodiments provided in this application, it should be understood that the disclosed device may be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above-mentioned units is only a division of logical functions. In actual implementation, there may be other division methods, for example, multiple units or components may be combined or integrated To another system, or some features can be ignored, or not implemented. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections may be indirect couplings or communication connections through some interfaces, devices or units, and may be in electrical or other forms.

The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware or software functional unit.

If the above integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it may be stored in a computer readable memory. Based on this understanding, the technical solution of the present application essentially or part of the contribution to the existing technology or all or part of the technical solution can be embodied in the form of a software product, the computer software product is stored in a memory, Several instructions are included to enable a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the above methods in various embodiments of the present application. The aforementioned memory includes: U disk, Read-Only Memory (ROM, Read-Only Memory), Random Access Memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

A person of ordinary skill in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by a program instructing relevant hardware. The program may be stored in a computer-readable memory, and the memory may include: a flash disk , Read-Only Memory (English: Read-Only Memory, abbreviation: ROM), Random Access Device (English: Random Access Memory, abbreviation: RAM), magnetic disk or optical disk, etc.

The embodiments of the present application are described in detail above, and specific examples are used to explain the principle and implementation of the present application. The descriptions of the above embodiments are only used to help understand the method and core idea of the present application; at the same time, Those of ordinary skill in the art, according to the ideas of the present application, may have changes in specific implementations and application scopes. In summary, the content of this specification should not be construed as limiting the present application.

Claims (10)

A deep learning calculation method characterized by being applied to electronic equipment, including: Determine the target model required for this deep learning calculation, the target model includes multi-layer model data, and the multi-layer model data is arranged in the target model according to a set order; Sequentially read in the multi-layer model data according to the set order on the first thread; When the layer of model data is read in, the calculation is performed on the second thread based on the layer of model data read in, and this step is repeated until all calculations are completed. The method according to claim 1, wherein the determining the target model required for the deep learning calculation includes: The first scene of the first application running in the foreground is determined, and the target model required for this deep learning calculation is determined based on the first scene of the first application. The method according to claim 2, wherein when the first application is a setting application, and / or the use priority of the first application is greater than or equal to the setting priority, the first thread And the resources occupied by the second thread are resources in the reserved resource pool. The method according to any one of claims 1 to 3, wherein before the sequentially reading the multi-layer model data on the first thread according to the set order, the method further comprises: It is determined that the size of the target model is greater than or equal to the first threshold, and / or the current memory occupancy rate of the electronic device is greater than or equal to the second threshold. The method according to claim 4, wherein the method further comprises: When the size of the target model is less than the first threshold, and / or the current memory occupancy rate of the electronic device is less than the second threshold, reading the multi-layer model data on a third thread, And after the multi-layer model data is read in, calculation is performed on the third thread based on the multi-layer model data. The method according to any one of claims 2-5, wherein during the deep learning calculation process of the first scene of the first application, the method further comprises: If it is detected that the scene running in the foreground is switched from the first scene of the first application to the second scene of the second application, the deep learning calculation of the first scene of the first application is suspended; If it is not detected that the scene running in the foreground is switched to the first scene of the first application within the set duration, the first thread and the second thread are released; If it is detected that the scene running in the foreground is switched to the first scene of the first application within the set duration, the deep learning calculation of the first scene of the first application is continued. The method according to claim 6, wherein before the suspending the deep learning calculation of the first scene of the first application, the method further comprises: It is determined that the importance level of the second scene of the second application is less than the importance level of the first scene of the first application, and / or the second scene of the second application is a setting scene. A deep learning computing device characterized by being applied to electronic equipment, including: A model determining unit, configured to determine a target model required for this deep learning calculation, the target model includes multi-layer model data, and the multi-layer model data is arranged in the target model according to a set order; A data reading unit for sequentially reading the multi-layer model data according to the set sequence on the first thread; The calculation unit is used to perform calculation on the second thread based on the completed layer of model data when the layer of model data is read in, and repeat this step until all calculations are completed. An electronic device, characterized in that it includes a processor, a memory, a communication interface, and one or more programs, the one or more programs are stored in the memory, and are configured to be executed by the processor, The program includes instructions for performing the steps in the method of any one of claims 1-7. A computer-readable storage medium characterized by storing a computer program for electronic data exchange, wherein the computer program causes a computer to execute the method according to any one of claims 1-7.

Images