Patent Application
20250117333
This application relates to the field of chip technologies, and in particular, to a data processing method and apparatus, an electronic device, and a computer-readable storage medium.
In an electronic device, a processor (for example, a central processing unit (CPU)) may generally access storage spaces of a main memory and a cache. The main memory has a large capacity and a high access latency, and the cache has small space but a low access latency. The CPU usually copies a part of data in the main memory to the cache. When a to-be-processed request needs to access the data, the CPU can quickly obtain the data from the cache. A probability of hitting the data to be accessed by the to-be-processed request in the cache may also be referred to as a cache hit rate, and the cache hit rate affects performance of the electronic device.
However, the to-be-processed request usually comes from an application that runs in the CPU. When the application accesses the data in the cache through the to-be-processed request, a cache hit rate is low, and performance of the electronic device is poor. Therefore, a method for improving the performance of the electronic device is urgently needed.
This application provides a data processing method and apparatus, an electronic device, and a computer-readable storage medium, to improve a cache hit rate.
According to a first aspect, a data processing method is provided. The method is applicable to an electronic device, where the electronic device includes a first storage medium and a second storage medium, the first storage medium is a memory of the electronic device, and the second storage medium is a cache of the memory. An implementation process of the method includes: first obtaining a data processing request of an application, then obtaining, based on the data processing request, a memory bypass flag corresponding to a first storage space, and processing associated data based on the memory bypass flag. The data processing request indicates to process the associated data of the first storage space in the first storage medium. The memory bypass flag indicates a storage manner of using the second storage medium, and the storage manner is set based on an access requirement of the application for the first storage space.
According to this method, when the data processing request of the application for the first storage space in the memory is received, the memory bypass flag corresponding to the first storage space is obtained. The memory bypass flag indicates the storage manner of using the cache, and the storage manner is set based on the access requirement of the application for the first storage space. Therefore, when the associated data of the first storage space is processed based on the obtained memory bypass flag, the storage manner of using the cache for the associated data may be determined based on the access requirement of the application for the first storage space, so that a cache behavior is controlled based on the access requirement of the application, thereby improving a cache hit rate.
In an implementation, the second storage medium is a high bandwidth memory HBM, a use manner of the HBM is a cache mode, and the cache mode indicates that the HBM is used as the cache of the memory.
Based on the foregoing implementation, in an application scenario in which the HBM is a cache of the first storage medium, the memory bypass flag indicates that a storage space in the memory uses the HBM, so that the cache behavior of the HBM is controlled based on the access requirement of the application.
In another implementation, before the memory bypass flag corresponding to the first storage space is obtained based on the data processing request, the implementation process of the method further includes: first invoking a first application programming interface API in an operating system, setting a bypass flag field in a page table entry of a page table of the first storage medium, and then adding, based on the access requirement recorded in the application, the memory bypass flag to the bypass flag field in the page table entry corresponding to the first storage space, where the bypass flag field is for storing the memory bypass flag.
Based on the foregoing implementation, the first API is invoked to set the bypass flag field in the page table, so that a memory bypass attribute is added to the page table. In this way, the bypass flag field can be added to the corresponding page table entry based on the access requirement of the application for the storage space of the memory. A processor can sense an access memory feature of the application through a specified memory bypass flag corresponding to each storage space in the first storage medium. When accessing the storage space in the memory, the processor can determine, based on the memory bypass flag, whether to copy data in the storage space to the cache, so that the cache behavior is controlled based on the access requirement of the application. In this way, the processor subsequently can obtain, from the cache, data required by the application, thereby improving the cache hit rate.
In another implementation, an implementation of the invoking a first API in an operating system, and setting a bypass flag field in a page table entry of a page table of the first storage medium is: When the application applies to the operating system for allocating a storage space in the first storage medium or after the application completes applying for allocating the storage space, the first API is invoked and the bypass flag field is set in the page table entry of the page table.
Based on the foregoing implementation, when the application applies to the operating system for allocating the storage space in the first storage medium, only a part of code of the application needs to be modified, so that the application can invoke the first API in the operating system through the processor, to set the bypass flag field in the page table entry of the page table. This is simple to implement.
In another implementation, an implementation of the invoking a first application programming interface API in an operating system, and setting a bypass flag field in a page table entry of a page table of the first storage medium is: in a process of compiling the application by a compiler, invoking the first API through a second API or a directive provided by the compiler, and setting the bypass flag field in the page table entry of the page table.
Based on the foregoing implementation, by modifying a part of code of the application and the compiler, during compilation, the application can invoke the first API in the operating system by invoking a compiling interface in the compiler, to set the bypass flag field for the page table entry of the page table. This is simple to implement.
In another implementation, an implementation of the adding, based on the access requirement recorded in the application, the memory bypass flag to the bypass flag field in the page table entry corresponding to the first storage space includes: first querying, based on the access requirement recorded in the application and a virtual address of the first storage space, the page table entry corresponding to the first storage space, and then adding the memory bypass flag to the found page table entry.
In another implementation, before the memory bypass flag corresponding to the first storage space is obtained based on the data processing request, an implementation process of the method further includes: predicting, based on a data transmission attribute of at least one of the first storage medium and the second storage medium, the access requirement of the application for the first storage space, and then setting the memory bypass flag for the first storage space based on the predicted access requirement, where the data transmission attribute indicates impact of the storage medium on efficiency of data transmission.
Based on the foregoing implementation, the access requirement of the application for the first storage space is predicted based on a current data transmission attribute of at least one storage medium, so that the access requirement of the application for the first storage space does not need to be recorded in the application, and a real-time access requirement of the application for the storage space in the memory can be satisfied. Therefore, the cache behavior can be controlled in real time by predicting the memory bypass flag set for the access requirement of the application for the first storage space is located.
In another implementation, an implementation process of the predicting, based on a data transmission attribute of at least one of the first storage medium and the second storage medium, the access requirement of the application for the first storage space includes: inputting the data transmission attribute into a prediction network, and predicting, based on the input data transmission attribute using the prediction network, the access requirement of the application for the first storage space.
Based on the foregoing implementation, a user does not need to modify code of the application, and the operating system does not need to provide the first API, provided that the prediction network can be obtained, and the bypass flag field can be set for the page table entry in the page table of the first storage medium. This is simple to operate.
In another implementation, an implementation of the setting the memory bypass flag for the first storage space based on the predicted access requirement includes: first querying, based on the predicted access requirement and a physical address of the first storage space, a page table entry corresponding to the first storage space, and then adding the memory bypass flag to the found page table entry.
In another implementation, the data processing request includes the virtual address of the first storage space, and on this basis, an implementation process of the obtaining, based on the data processing request, a memory bypass flag corresponding to the first storage space includes: querying, based on the virtual address of the first storage space, the page table entry corresponding to the first storage space, and then obtaining the memory bypass flag from the found page table entry.
In another implementation, an implementation process of the querying, based on the virtual address, the page table entry corresponding to the first storage space includes: querying, based on the virtual address, the page table of the first storage medium for the page table entry corresponding to the first storage space; or querying, based on the virtual address, a translation lookaside buffer TLB of a processor for the page table entry corresponding to the first storage space.
In another implementation, an implementation process of the processing associated data based on the memory bypass flag includes the following two cases: if the memory bypass flag indicates to use the second storage medium to store data in the first storage medium, processing the associated data of the first storage space, and copying the associated data of the first storage space to the second storage medium; or if the memory bypass flag indicates not to use the second storage medium to store data in the first storage space, processing the associated data of the first storage space, and skipping performing the step of copying the associated data of the first storage space to the second storage medium.
Based on the foregoing implementation, based on the storage manner that is of using the cache and that is indicated by the memory bypass flag, a copy of data in the storage space of the memory is stored in the cache or is not stored in the cache, so that the cache behavior is controlled based on the memory bypass flag.
According to a second aspect, a data processing apparatus is provided. The apparatus includes all modules configured to perform the data processing method according to any one of the first aspect or the implementations of the first aspect.
According to a third aspect, an electronic device is provided. The electronic device includes a processor. The processor is configured to execute program code, to enable the electronic device to perform operations performed according to the foregoing data processing method.
According to a fourth aspect, a chip is provided. The chip is configured to implement operation steps of the data processing method according to any one of the first aspect or the implementations of the first aspect.
According to a fifth aspect, a processor is provided. The processor is configured to implement operation steps of the data processing method according to any one of the first aspect or the implementations of the first aspect.
According to a sixth aspect, a computer-readable storage medium is provided. The storage medium stores at least one piece of program code, and the program code is read by a processor, to enable an electronic device to perform operations according to the foregoing data processing method.
According to a seventh aspect, a computer program product or a computer program is provided. The computer program product or the computer program includes program code, and the program code is stored in a computer-readable storage medium. A processor of an electronic device reads the program code from the computer-readable storage medium, and the processor executes the program code, to enable the electronic device to perform the method provided in any one of the first aspect or the optional implementations of the first aspect.
In this application, the implementations according to the foregoing aspects, may be further combined to provide more implementations.
For an electronic device including a first storage medium and a second storage medium, this application provides a data processing method in the case that the first storage medium is a memory of the electronic device, and the second storage medium is a cache of the memory. In the method, when an application requests to process associated data of a storage space in the first storage medium, a storage manner of using the second storage medium for the storage space is first obtained. The storage manner is set based on an access requirement of the application for the storage space, and then the obtained storage manner is used to process the associated data.
In this case, a use manner of the second storage medium is a cache mode, and the cache mode indicates that the second storage medium is used as the cache of the memory (that is, the first storage medium). In an implementation, the method is performed by a processor (for example, a CPU) in the electronic device. With reference to
The second storage medium 102 is used as the cache of the memory. In this case, the use manner of the second storage medium 102 is a cache mode.
For either of the first storage medium 101 and the second storage medium 102, the storage medium includes a plurality of storage spaces, and the storage spaces are for storing data. When the processor 103 (for example, a CPU) runs an application, the application requests to process associated data of a storage space (for example, data read from the storage space, or data written into the storage space) in the first storage medium 101. Based on the request of the application, the to-be-processed associated data is first queried from the cache of the first storage medium 101. If the processor finds the associated data of the storage space in the cache, it is considered that a cache hit occurs. If the processor does not find the associated data of the storage space in the cache, it is considered that a cache miss occurs. If the cache hit occurs, the processor 103 processes the associated data in the cache. If the cache miss occurs, the processor 103 processes the associated data in the first storage medium 101.
In some embodiments, the data processing system 100 includes a plurality of second storage media 102, and the plurality of second storage media 102 are used as multi-level caches of the first storage medium 101. For example,
As shown in
For any cache in the multi-level caches of the second storage media 102a to 102d, the cache is configured to store a copy of a part of data stored in a next-level storage medium. When obtaining a request of the application for processing the associated data of the storage space in the first storage medium 101, the processor 103 preferentially queries, in descending order of levels, higher-level caches for whether the associated data of the storage space is stored. For any cache, during a query process, the processor 103 processes the associated data in the cache. If the cache miss occurs, the processor 103 continues to query a next-level storage medium of the cache until the associated data of the storage space is found.
The foregoing data processing system architectures can be deployed in an electronic device. Based on the foregoing data processing system architectures, the following describes, with reference to the accompanying drawings, a process of performing the data processing method by the electronic device.
When an application requests to process associated data of a storage space in the first storage medium, a memory bypass flag indicates a storage manner of using the second storage medium for the storage space, and the storage manner is obtained by the processor, and a processor processes the associated data based on the memory bypass flag. To describe the process, refer to
Step 301: The application invokes a first application programming interface (API) in an operating system, and sets a bypass flag field in a page table entry of a page table of the first storage medium. The bypass flag field is for storing the memory bypass flag.
The memory bypass flag indicates the storage manner of using the second storage medium, and the storage manner is set based on an access requirement of the application for the storage space in the first storage medium. The access requirement of the application for the storage space in the first storage medium includes an access requirement that is recorded in the application and that the application has for the storage space in the first storage medium, or a predicted access requirement of the application for the storage space in the first storage medium.
The memory bypass flag includes a first memory bypass flag or a second memory bypass flag. The first memory bypass flag indicates to use the second storage medium to store data in the first storage medium, and the second memory bypass flag indicates not to use the second storage medium to store the data in the first storage medium. The first memory bypass flag and the second memory bypass flag are represented in different manners. For example, the first memory bypass flag is 1, and the second memory bypass flag is 0. Alternatively, the first memory bypass flag is 0, and the second memory bypass flag is 1. Herein, the representation manners of the first memory bypass flag and the second memory bypass flag are not limited in embodiments of this application.
The page table shown in Table 1 is used as an example. The page table includes N+1 page table entries, and each page table entry includes one page number and one page frame number, to indicate that a page indicated by the page number is mapped to a page frame corresponding to the page frame number, where N is a positive integer.
In embodiments of this application, a function of a page table entry in a page table is further extended. A page table shown in Table 2 is used as an example. A bypass flag field is extended in the page table entry, and the bypass flag field is for storing a memory bypass flag, so that the page table has a memory bypass attribute.
By invoking the first API in the operating system, the application sets the bypass flag field in the page table entry of the page table of the first storage medium, to add the memory bypass attribute to the page table. In an implementation, the first API is a user-defined API for setting the memory bypass attribute, or is an existing API extended for setting the memory bypass attribute. Herein, a type of the first API is not limited in embodiments of this application.
When the first API is an existing API extended for setting the memory bypass attribute, the first API includes a memory-mapped file function, a memory-mapped protection function, and another type of function. Herein, a function type of the first API is not limited in embodiments of this application.
The memory-mapped file function is represented as: void *mmap (void *addr, size_t length, int prot, int flags, int fd, off_t offset). “addr” in the memory-mapped file function indicates a hint during memory allocation. “length” indicates a length of data that is in the file and that is mapped to the memory. “prot” indicates a protection manner of a mapping region of the file in the memory. In embodiments of this application, the prot function is extended, and a bypass attribute setting flag is added to the prot function, where the bypass attribute setting flag indicates that a bypass flag field is set for a page table entry in the page table. “flags” indicate various features of the mapping area. “fd” is a file descriptor, indicating a file mapped to the memory. “offset” indicates a file-mapped offset.
The memory mapping protection function is represented as: int mprotect (void *addr, size_t len, int prot). “addr” in the memory mapping protection file function indicates a to-be-modified start address of a protection attribute region in the memory. “len” indicates a length of the protection attribute region. “prot” indicates various attributes of the protection attribute region. In embodiments of this application, a prot function is extended, and a bypass attribute setting flag is added to the prot, where the bypass attribute setting flag indicates that a bypass flag field is set for a page table entry in the page table.
In an implementation, the application invokes the first API in the operating system in either of the following manners 1 and 2.
Manner 1: When the application applies to the operating system for allocating the storage space in the first storage medium or after the application completes applying for allocating the storage space, the application invokes the first API, and sets the bypass flag field in the page table entry of the page table.
The application is any application installed in the electronic device.
After the application completes applying to the operating system for allocating a storage space,
When the application applies to the operating system for allocating the storage space in the first storage medium, a part of code of the application is modified, so that the application can set the bypass flag field for the page table entry in the page by invoking the first API in the operating system through the processor. This is simple to implement.
Manner 2: The first API is a compiling interface encapsulated in a compiler. In a process of compiling the application by the compiler, the application invokes the first API through a second API or a directive in the compiler, to set the bypass flag field for the page table entry of the page.
The second API may be an API extended based on an existing API, or may be a newly added API. Herein, an implementation of the second API is not limited in embodiments of this application.
In this implementation, a part of code of the application and the compiler is modified, so that the application can invoke the first API in the operating system by invoking the compiling interface in the compiler during compilation, to set the bypass flag field for the page table entry in the page. This is simple to implement.
In another implementation, a target application is installed in the electronic device, and the target application is for invoking the first API. For example, in a process in which the processor in the electronic device runs the target application, the target application invokes the first API to set the bypass flag field for the page table entry of the page table, without invoking the first API interface through each application.
Step 301 is an optional step. In some embodiments, if the bypass flag field is already set in the page table, the operating system does not need to perform step 301. In some other embodiments, the operating system has a function of actively setting the bypass flag field in the page table. For example, when the electronic device is powered on and the operating system is initialized, the operating system sets the bypass flag field in the page table. In this case, step 301 does not need to be performed. In addition, step 301 may be performed only once, and does not need to be performed for more times.
Step 302: Based on the access requirement of the application that is recorded in the application and that the application has for a first storage space, the operating system adds the memory bypass flag to a bypass flag field in a page table entry corresponding to the first storage space.
The first storage space is any storage space in the first storage medium. In embodiments of this application, data stored in the first storage space is referred to as associated data, and the access requirement of the application for the first storage space in the first storage medium is an access requirement of the application for the associated data. The access requirement of the application for the associated data is, for example, that the application needs to access the associated data for at least a target quantity of times in a first time period. The first time period is a time period after a current moment, that is, a future time period. Herein, duration of the first time period is not limited in embodiments of this application.
In an implementation, if the first storage space has been allocated to the application, the operating system adds the memory bypass flag to the page table entry based on the access requirement recorded in the application. The following steps 3021 to 3022 are used as an example.
Step 3021: Based on the access requirement recorded in the application and a virtual address of the first storage space, the operating system queries the page table entry corresponding to the first storage space.
Step A1: If the first storage space in the first storage medium is already allocated to the application, the application sends a bypass flag setting request to the operating system, where the bypass flag setting request indicates to set a memory bypass flag for the first storage space.
The bypass flag setting request includes a virtual address where the associated data of the first storage space is located, or the bypass flag setting request includes a start address of a virtual address segment of the to-be-set memory bypass flag and a length of the virtual address segment. A storage space indicated by the virtual address segment includes the first storage space, or the bypass flag setting request includes the virtual address segment.
As shown in
When the application receives the first virtual address segment, all related data used in a process of running the application by the processor can be stored in a storage space corresponding to the first virtual address segment.
For at least one piece of data to be accessed by the application, if the application records a requirement flag corresponding to any piece of data, and the any piece of data is stored at a virtual address in the first virtual address segment, a page frame corresponding to a page in which the virtual address is located is used as the first storage space, and the any piece of data is used as the associated data of the first storage space, to generate the bypass flag setting request including the virtual address. The requirement flag indicates an access requirement of the application for the at least one piece of data.
In some embodiments, a plurality of pieces of data to be accessed by the application all correspond to the requirement flag. In this case, the application uses a virtual address that is in the first virtual address segment and that is for storing the plurality of pieces of data as a virtual address segment (that is, a second virtual address segment) of the to-be-set memory bypass flag, and generates a bypass flag setting request including the second virtual address segment. Alternatively, if virtual addresses in the second virtual address segment are continuous virtual addresses, the bypass flag setting request does not include the second virtual address segment, but includes a start address of the second virtual address segment and a length of the second virtual address segment.
In another implementation, the bypass flag setting request further includes at least one of the following: a memory bypass flag and a requirement flag of the second storage medium. Certainly, the bypass flag setting request may not include the memory bypass flag or the requirement flag.
Step A2: The operating system receives the bypass flag setting request, and queries, based on the bypass flag setting request and the virtual address of the first storage space, the page table entry corresponding to the first storage space.
After the operating system receives the bypass flag setting request, if the bypass flag setting request includes the virtual address of the first storage space, the operating system obtains a page number in the virtual address, where the page number indicates a page in which the first storage space is located, and then the operating system queries the page table for a page table entry including the page number.
If the bypass flag setting request includes the second virtual address segment, the operating system determines, based on the second virtual address segment, a page number of each page corresponding to the second virtual address segment, to align address pages. For ease of description, each page corresponding to the second virtual address segment is referred to as a first page. Alternatively, if the bypass flag setting request includes the start address and the length of the second virtual address segment, the operating system determines, based on the start address, a page indicated by the start address, and determines, based on the length of the second virtual address segment, a page indicated by the start address and a target quantity of pages following the page as the first page. If the length of the second virtual address segment is an integer multiple of a length of a single page, the target quantity is equal to the multiple, or if the multiple is not an integer, the target quantity is one greater than the multiple.
After determining each first page corresponding to the second virtual address segment, for any first page, the operating system queries the page table for a page table entry including a page number of the first page. If the first page is the page in which the first storage space is located, the operating system queries the page table for the page table entry including the page number of the first page, that is, queries the page table entry corresponding to the first storage space.
Step 3022: The operating system adds a memory bypass flag to the page table entry that is in the page table and that corresponds to the first storage space.
For any first page, if any page table entry in the page table includes the page number of the first page, the operating system adds a first memory bypass flag to a bypass flag field of the any page table entry. The following cases 1 and 2 are used as an example.
Case 1: When creating the page table, the operating system adds a second memory bypass flag to a bypass flag field of each page table entry in the page table. If any page table entry in the page table includes the page number of the first page, the second memory bypass flag in the bypass flag field of the any page entry is modified to the first memory bypass flag.
Case 2: When creating the page table, the operating system does not add any content to a bypass flag field of each page table entry in the page table, so that the bypass flag field is empty. If any page table entry in the page table includes the page number of the first page, the operating system adds the first memory bypass flag to the bypass flag field of the any page entry.
In a first time period after current time, the application may further access some storage spaces in the first storage medium for a quantity of times less than a target quantity of times, and the operating system sets the second memory bypass flag for a page (that is, a second page) in which the some storage spaces are located. It may be understood that the second page is each page except the first page in the at least one page created for the application. For the case 1, each page table entry in which the memory bypass flag is not modified in the page table is a page table entry corresponding to the second page. Because the memory bypass flag in the page table entry corresponding to the second page is the second memory bypass flag, the operating system does not need to set the second memory bypass flag again for the page table entry corresponding to the second page. For the case 2, because the bypass flag field of the page table entry corresponding to the second page is empty, the operating system adds the second memory bypass flag to the bypass flag field of the page table entry corresponding to the second page.
In another implementation, the application sends the bypass flag setting request to the operating system based on the first virtual address segment. In this case, the bypass flag setting request includes the second virtual address segment and a third virtual address segment. The second virtual address segment corresponds to the requirement flag or the first memory bypass flag, and the third virtual address segment includes a virtual address other than the second virtual address segment in the first virtual address segment. Then, based on the second virtual address segment in the bypass flag setting request and the corresponding requirement flag or the first memory bypass flag, the operating system adds the first memory bypass flag to the page table entry corresponding to each first page in the page table. Based on the third virtual address segment in the bypass flag setting request, the operating system determines a page number of the second page, and adds the second memory bypass flag to a page table entry that is in the page table and includes the page number of the second page.
As shown in
Step 303: The processor obtains a data processing request of the application, where the data processing request indicates to process the associated data of the first storage space in the first storage medium.
The first storage space is a storage space that is in the first storage medium and includes the associated data, and the data processing request includes the virtual address of the first storage space. In an implementation, the data processing request is a data read request or a data write request. The data read request indicates to read the associated data in the first storage space, the data write request indicates to write the associated data to the first storage space, and the data write request further includes the associated data.
In a process of running the application by the processor, if the application has a requirement for accessing the first storage space, the processor generates the data processing request. For example, in a process of running the application, if the application needs to write data into the first storage space, the data processing request generated by the processor is a data write request; or if the application needs to read the data in the first storage space, the data processing request generated by the processor is a data read request.
Step 304: Based on the data processing request, the processor obtains the memory bypass flag corresponding to the first storage space. The memory bypass flag indicates a storage manner of using the second storage medium, and the storage manner is set based on the access requirement of the application for the first storage space.
In an implementation, the processor queries, based on the virtual address of the first storage space in the data processing request, the page table entry corresponding to the first storage space, and obtains the memory bypass flag of the first storage space from the found page table entry (for example, obtains the memory bypass flag from the memory bypass field of the page table entry).
For a process in which the processor queries the page table entry corresponding to the first storage space, the processor queries, based on the virtual address, the TLB of the processor for the page table entry corresponding to the first storage space, or queries, based on the virtual address, the page table of the first storage medium for the page table entry corresponding to the first storage space.
For example, the processor determines, based on the virtual address, a page number of a page in which the first storage space is located and an intra-page offset. Then, based on the page number of the page, the TLB is queried for a TLB entry including the page number. If the TLB entry including the page number is found, the memory bypass flag is obtained from the bypass flag field of the TLB entry. If the TLB entry including the page number is not found in the TLB, the processor reads, from the page table, the page table entry including the page number to the TLB, and then obtains the memory bypass flag from the page table entry in the TLB. Alternatively, the processor first obtains the memory bypass flag from the page table entry, and then reads the page table entry to the TLB.
Step 305: The processor processes the associated data based on the memory bypass flag.
For different memory bypass flags, processes of the associated data processed by the processor are different. For example, if the memory bypass flag indicates to use the second storage medium to store data in the first storage medium, the processor processes the associated data of the first storage space, and copies the associated data of the first storage space to the second storage medium. If the memory bypass flag indicates not to use the second storage medium to store data in the first storage medium, the processor processes the associated data of the first storage space, and skips performing the step of copying the associated data of the first storage space to the second storage medium.
In addition, the processor determines, by querying the TLB, that the data in the first storage space is already stored in the second storage medium. If it is found that a TLB entry includes a page number of a page in which the first storage space is located, it indicates that the data in the first storage space is already stored in the second storage medium. The processor processes the associated data of the first storage space by accessing the second storage medium. Otherwise, the processor processes the associated data of the first storage space by accessing the first storage medium.
The following describes step 305 based on the found TLB entry or page table entry that includes the page number.
If the memory bypass flag is obtained from the TLB entry, the processor obtains a page frame number of the first storage space from the found TLB entry, and concatenates the page frame number and the intra-page offset in the virtual address of the first storage space, to obtain a physical address of the first storage space. The processor queries, based on the physical address, the second storage medium for a cache line that includes the data in the first storage space, and accesses, based on the data processing request, a cache space that is in the cache line and that corresponds to the first storage space, to process the associated data of the first storage space. For example, if the data processing request is a data read request, the processor reads data from the cache space. In this case, the read data is the associated data of the first storage space. If the data processing request is a data write request, the processor writes the associated data carried by the data write request to the cache space.
In this case, the second storage medium already stores the associated data of the first storage space in the first storage medium, and the processor does not perform any operation on the data of the first storage space in the first storage medium based on the found memory bypass flag, but processes the data write request.
If the processor obtains the memory bypass flag from the page table entry in the page table, the processor obtains the page frame number of the first storage space from the found page table entry, and concatenates the page frame number and the intra-page offset in the virtual address of the first storage space, to obtain the physical address of the first storage space. The processor accesses, based on the data processing request, the first storage space that is in the first storage medium and indicated by the physical address, to process the associated data of the first storage space. For example, if the data processing request is a data read request, the processor reads data from the first storage space. If the data processing request is a data write request, the processor writes the associated data carried by the data write request to the first storage space.
In this case, if the memory bypass flag is the first memory bypass flag, the processor copies the data (that is, the associated data) in the first storage space to the cache line in the second storage medium. If the memory bypass flag is the second memory bypass flag, the processor does not perform the step of copying the data in the first storage space to the cache line in the second storage medium.
In an implementation, when a size of the first storage space is less than a size of the cache line in the second storage medium, if the data in the first storage space is copied to the cache line in the second storage medium, the processor further copies data in other storage spaces in the first storage medium to the cache line, to fill the cache line. In addition, when an idle cache line of the second storage medium is insufficient, the processor evicts data in a non-idle cache line of the second storage medium from the second storage medium, so that the non-idle cache line becomes an idle cache line. Then, the processor copies the data in the first storage space and the data in the other storage spaces in the first storage medium together to the idle cache line.
The following describes the foregoing steps 304 and 305 with reference to
According to the method provided in this application, when the data processing request of the application for the first storage space in the memory is received, the memory bypass flag corresponding to the first storage space is obtained. The memory bypass flag indicates the storage manner of using the cache, and the storage manner is set based on the access requirement of the application for the first storage space. Therefore, when the associated data of the first storage space is processed based on the obtained memory bypass flag, the storage manner of using the cache for the associated data can be determined based on the access requirement of the application for the first storage space. In this way, a cache behavior is controlled based on the access requirement of the application, thereby improving a cache hit rate. In addition, only when the memory bypass flag corresponding to the first storage space is the first memory bypass flag, the processor copies the associated data of the first storage space to the cache, thereby avoiding frequently copying data from the first storage medium to the cache, and reducing memory bandwidth consumption of the first storage medium.
In another implementation, the operating system may further add the memory bypass flag to a corresponding bypass flag field based on a predicted access requirement of the application for the first storage space. For example,
Step 801: An operating system in the electronic device sets a bypass flag field for a page table entry in a page table of the first storage medium.
In an implementation, when the operating system in the electronic device starts to run, the bypass flag field is set in the page table entry of the page table of the first storage medium, and a processor is triggered to set the bypass flag field in a TLB entry in a TLB.
Alternatively,
The prediction network is for predicting an access requirement of an application for each storage space in the first storage medium. The prediction network is an artificial intelligence (AI) model, and the AI model includes a machine learning model, a deep learning model, a reinforcement learning model, or the like. For example, the prediction network is a convolutional neural network or a decision tree.
As shown in
In this implementation, a user does not need to modify code of the application, and the operating system does not need to provide the first API, provided that the prediction network can be obtained, and the bypass flag field can be set for the page table entry in the page table of the first storage medium. This is simple to operate.
In addition, step 801 is an optional step. In some embodiments, if the bypass flag field is already set in the page table, the operating system does not need to perform step 801. In addition, step 801 may be performed once, and does not need to be performed for more times.
Step 802: The operating system obtains a data transmission attribute of at least one of the first storage medium and the second storage medium, and the data transmission attribute indicates impact of the storage medium on efficiency of data transmission.
The data transmission attribute includes at least one of the following: a memory bound (Mem bound) of the processor, a data read/write latency of the processor, a read bandwidth of the first storage medium, a write bandwidth of the first storage medium, a read bandwidth of the second storage medium, a write bandwidth of the second storage medium, and a write-back ratio of the second storage medium.
The memory bound is a memory bandwidth between the processor and a cache and a memory of each level. The data read/write latency is a latency (for example, an average latency) that occurs when the processor reads/writes data in the cache and the memory of each level. The read bandwidth of the first storage medium is a bandwidth occupied for reading data from the first storage medium by the processor. The write bandwidth of the first storage medium is a bandwidth occupied for writing data to the first storage medium by the processor. The read bandwidth of the second storage medium is a bandwidth occupied for reading data from the second storage medium by the processor. The write bandwidth of the second storage medium is a bandwidth occupied for writing data to the second storage medium by the processor. The write-back ratio of the second storage medium is a ratio of data that is written back to the first storage medium to evicted data of the second storage medium.
When the processor runs the operating system, the processor collects the data transmission attribute of the at least one storage medium at an interval of a second time period, to periodically obtain the data transmission attribute of the at least one storage medium, thereby reducing occupied resources of the processor. Duration of the second time period may be set based on an implementation scenario. Herein, the duration of the second time period is not limited in embodiments of this application.
For example, each time the second time period elapses, the operating system samples at least one of following: the memory bound of the processor, the data read/write latency of the processor, the read bandwidth of the first storage medium, the write bandwidth of the first storage medium, the read bandwidth of the second storage medium, the write bandwidth of the second storage medium, and the write-back ratio of the second storage medium in the second time period, to obtain the data transmission attribute.
Step 803: The operating system predicts the access requirement of the application for the first storage space based on the data transmission attribute of the at least one of the first storage medium and the second storage medium.
In an implementation, the operating system inputs the data transmission attribute to the prediction network, and predicts, based on the input data transmission attribute using the prediction network, the access requirement of the application for the first storage space.
There are two cases for the predicted access requirement of the application for a single storage space (for example, a single page): The application needs to access the storage space in a first time period after current time for a quantity of times less than a target quantity of times; and the application needs to access the storage space in the first time period for at least the target quantity of times. For ease of description, the two cases are respectively referred to as: The application has no access requirement for the storage space, and the application has an access requirement for the storage space.
Step 804: Based on the predicted access requirement, the operating system sets the memory bypass flag for the first storage space.
In an implementation, the operating system queries, based on the predicted access requirement of the application for the first storage space and a physical address of the first storage space, a page table entry corresponding to the first storage space, and then adds the memory bypass flag corresponding to the first storage space to the page table entry.
For example, if the prediction network predicts, by using a page frame as a unit, an access requirement of the application for a page frame, at least one prediction result output by the prediction network corresponds to at least one page frame number of a page frame. For a page frame number corresponding to any prediction result, the operating system queries a page table (or a TLB) for a page table entry including the page frame number.
For example, the first storage space is any page frame in the first storage medium. The physical address of the first storage space is a page frame number of the any page frame, and the operating system queries a page table (or a TLB) for a page table entry including the page frame number. After that, if the prediction result corresponding to the page frame to which the first storage space belongs indicates that the application has an access requirement for a corresponding storage space, the operating system adds the first memory bypass flag to the found page table entry. If the prediction result indicates that the application has no access requirement for the corresponding storage space, the operating system adds the second memory bypass flag to the found page table entry.
Step 805: The processor obtains a data processing request of the application, where the data processing request indicates the processor to process associated data of the first storage space in the first storage medium.
Step 805 is similar to step 303. Herein, step 805 is not described again in embodiments of this application.
Step 806: Based on the data processing request, the processor obtains the memory bypass flag corresponding to the first storage space. The memory bypass flag indicates a storage manner of using the second storage medium, and the storage manner is set based on the access requirement of the application for the first storage space.
Step 806 is similar to step 304. Herein, step 806 is not described again in embodiments of this application.
Step 807: The processor processes the associated data based on the memory bypass flag.
Step 807 is similar to step 305. Herein, step 807 is not described again in embodiments of this application.
According to the method provided in embodiments of this application, when the data processing request of the application for the first storage space in the memory is received, the memory bypass flag corresponding to the first storage space is obtained. The memory bypass flag indicates the storage manner of using the cache, and the storage manner is set based on the access requirement of the application for the first storage space. Therefore, when the associated data of the first storage space is processed based on the obtained memory bypass flag, the storage manner for the associated data to use the cache can be determined based on the access requirement of the application for the first storage space. In this way, a cache behavior is controlled based on the access requirement of the application, thereby improving a cache hit rate. In addition, the access requirement of the application for the first storage space is predicted based on the current data transmission attribute of the at least one storage medium. This avoids use of the application to record the access requirement of the application for the first storage space, and can satisfy a real-time access requirement of the application for a storage space in the memory. Therefore, the cache behavior can be controlled in real time by predicting the memory bypass flag set for the access requirement of the application for the first storage space is located.
The foregoing steps 302 and 804 are both processes in which the operating system sets the memory bypass flag for the first storage space based on the access requirement of the application for the first storage space. This enables the processor to sense an access memory attribute of the application through the set memory bypass flag corresponding to each storage space in the first storage medium, thereby overcoming a problem that the processor cannot sense the access memory attribute of the application. When accessing a storage space in the memory, the processor can determine, based on the memory bypass flag, whether to copy data in the storage space to the cache. Therefore, the cache behavior is controlled based on the access requirement of application, so that the processor can subsequently obtain from the cache data required by the application, thereby improving the cache hit rate.
In addition, in this application, to reflect an interaction process in the electronic device, step 301 is performed by the application, and step 302 and step 801 to step 804 are performed by the operating system for description. It should be understood that these steps should be implemented by hardware (for example, the processor) by running the application or the operating system.
The foregoing method embodiments are described by using an example in which the bypass flag field is for storing a memory bypass flag of a single cache. In another implementation, the bypass flag field includes a plurality of indication bits. The plurality of indication bits may be bits reserved in a page table entry, or may be bits extended in a page table entry. The plurality of indication bits are in a one-to-one correspondence with a plurality of levels of caches (that is, a plurality of second storage media) of the first storage medium, and each indication bit is for storing a memory bypass flag of a corresponding cache. The page table shown in Table 3 is used as an example. The bypass flag field has M indication bits, and the M indication bits are in a one-to-one correspondence with M caches of levels L1 to LM, where M is a positive integer.
In this case, the memory bypass flag indicates a storage manner of using a corresponding cache, and the storage manner is set based on an access requirement of the application for a storage space in the first storage medium. In this case, the first storage medium is a memory or a cache of the electronic device.
For this case, when querying the memory bypass flag corresponding to the first storage medium, the processor may find a plurality of memory bypass flags. In this case, step 305 is described as follows:
For any one of the plurality of memory bypass flags, a cache corresponding to an indication bit in which the memory bypass flag is located in the bypass flag field is the second storage medium.
After obtaining a physical address of a storage space in a to-be-accessed memory, the processor queries, based on the physical address, the second storage medium for a cache line including data in the first storage space. If it is found that the cache line exists in the second storage medium, the processor accesses, based on the data processing request, a cache space that is in the cache line and that corresponds to the first storage space. If the cache line corresponding to the first storage space does not exists in the second storage medium and the first storage medium is a cache, the processor queries, based on the physical address, the first storage medium for the cache line including the data in the first storage space. If it is found that the cache line exists in the first storage medium, the processor accesses, based on the data processing request, the cache space that is in the cache line of the first storage medium and that corresponds to the first storage space. If any memory bypass flag is a first memory bypass flag, the processor copies the data in the first storage space in the cache line to the second storage medium. If any memory bypass flag is a second memory bypass flag, the processor does not perform the step of copying the data in the first storage space in the cache line to the second storage medium. If it is found that the cache line does not exists in the first storage medium, the processor continues to query, based on the physical address, a next-level cache of the first storage medium for the data in the first storage space. Until in a cache or memory whose level is lower than that of the first storage medium, a target storage space storing the data in the first storage space is find. The processor accesses the target storage space based on the data processing request. If any memory bypass flag is the first memory bypass flag, the processor copies the data in the target storage space to the second storage medium and a cache whose level is lower than that of the second storage medium. If any memory bypass flag is the second memory bypass flag, the processor does not perform the step of copying the data in the target storage space to the second storage medium and a cache whose level is lower than that of the second storage medium.
The foregoing describes the data processing methods provided in this application. With reference to the accompanying drawings, the following further describes an apparatus, a chip, and an electronic device provided in this application. It should be understood that the apparatus, the chip, and the electronic device that are described below may perform any function of the electronic device or a part of the electronic device in any one of the foregoing methods.
Refer to
The first obtaining unit 1103 is configured to obtain a data processing request of an application. The data processing request indicates to process associated data of a first storage space in the first storage medium 1101. The first storage medium 1101 is a memory of the electronic device. The second storage medium 1102 is a cache of the memory;
The second obtaining unit 1104 is configured to obtain, based on the data processing request, a memory bypass flag corresponding to the first storage space. The memory bypass flag indicates a storage manner of using the second storage medium 1102, and the storage manner is set based on an access requirement of the application for the first storage space.
The execution unit 1105 is configured to process the associated data based on the memory bypass flag.
It should be understood that the apparatus 1100 in this embodiment of this application may be performed through a CPU, or may be performed through an application-specific integrated circuit (ASIC), or may be performed through a programmable logic device (PLD). The PLD may be a complex programmable logical device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), a data processing unit (DPU), a system on chip (SoC), or any combination thereof. The apparatus 1100 may also perform the data processing methods shown in
In an implementation, the second storage medium 1102 is a high bandwidth memory HBM, a use manner of the HBM is a cache mode, and the cache mode indicates that the HBM is used as the cache of the memory.
In an implementation, the apparatus 1100 further includes:
In an implementation, the invoking unit is configured to:
In an implementation, the invoking unit is configured to:
In an implementation, the adding unit is configured to:
In an implementation, the apparatus 1100 further includes:
In an implementation, the prediction unit is configured to:
In an implementation, the setting unit is configured to:
In an implementation, the data processing request includes the virtual address of the first storage space, and the second obtaining unit 1104 is configured to:
In an implementation, the second obtaining unit 1104 is further configured to:
In an implementation, the execution unit 1105 is configured to:
It should be understood that the apparatus 1100 corresponds to the electronic device in the foregoing method embodiments, and the modules in the apparatus 1100 and the foregoing other operations and/or functions are respectively used to perform various steps and methods implemented by the electronic device in the method embodiments. For details, refer to the foregoing method embodiments. For brevity, details are not described herein again.
It should be understood that, when the apparatus 1100 processes data, division of the foregoing functional modules is merely used as an example for description. In actual application, the foregoing functions may be allocated to different functional modules for implementation based on requirements. That is, an internal structure of the apparatus 1100 is divided into different functional modules, to implement all or some of the functions described above. In addition, the apparatus 1100 provided in the foregoing embodiment shares a same concept with the foregoing method embodiments. For an implementation process of the apparatus 1100, refer to the foregoing method embodiments. Details are not described herein again.
It should be understood that the apparatus 1100 may be equivalent to the data processing system 100 (or 200), or equivalent to an execution component in the data processing system 100 (or 200).
Optionally, there is at least one processor 1201 in the chip system. It should be understood that in embodiments of this application, the processor 1201 may be a CPU or another general purpose processor, or the processor 1201 may be one or more integrated circuits for performing the solutions of this application, for example, a digital signal processor (DSP), an ASIC, a PLD, an FPGA or another programmable logic device, a discrete gate or a transistor logic device, and a discrete hardware component. The general purpose processor may be a microprocessor, any conventional processor, or the like.
Optionally, there may also be one or more memories 1203 in the chip system. The memory 1203 may be integrated with the processor 1201, or may be disposed separately from the processor 1201. This is not limited in this application. For example, the memory 1203 and the processor 1201 may be integrated in a same chip. As shown in
The memory 1203 may include a read-only memory and a random access memory, and provide instructions and data for the processor 1201. The memory 1203 may further include a nonvolatile random access memory. For example, the memory 1203 may further store information about a device type. Alternatively, the memory 1203 may be a volatile memory, or may include both a volatile memory and a nonvolatile memory.
The nonvolatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), used as an external cache. Through illustrative but not limited description, many forms of RAMs may be used, for example, a static random access memory (SRAM), a dynamic random access memory (DRAM), a synchronous dynamic random access memory (SDRAM), a double data rate synchronous dynamic random access memory (DDR SDRAM), an enhanced synchronous dynamic random access memory (ESDRAM), a synchlink dynamic random access memory (SLDRAM), and a direct rambus RAM (DR RAM).
For example, the chip system may be an FPGA, an ASIC, a SoC, a CPU, a network processor (NP), a digital signal processor (DSP), a micro controller unit (MCU), a PLD, or another integrated chip.
For example, the processor 1301 may include one or more CPUs, for example, a CPU 0 and a CPU 1 shown in
For example, the electronic device 1300 may include a plurality of processors, for example, the processor 1301 and a processor 1306 shown in
It should be understood that, in this application, an implementation of the processor 1301 is similar to an implementation of the processor 1201 in
The communication interface 1303 is configured to communicate with another device or a communication network through any apparatus such as a transceiver. The communication interface 1303 includes a wired communication interface, and may further include a wireless communication interface. The wired communication interface may be, for example, an Ethernet interface. The Ethernet interface may be an optical interface, an electrical interface, or a combination thereof. The wireless communication interface may be a wireless local area network (WLAN) interface, a cellular network communication interface, or a combination thereof.
The bus 1304 is configured to transmit information between the foregoing components. In addition to the communication bus, the bus 1304 may further include a power bus, a status signal bus, and the like. However, for clear description, various types of buses are marked as the bus 1304 in the figure. The communication bus may be classified into an address bus, a data bus, a control bus, and the like. For example, the communication bus may be a peripheral component interconnect express (PCIe) bus, an extended industry standard architecture (EISA) bus, a unified bus (Ubus or UB), a compute express link (CXL), a cache coherent interconnect for accelerators (CCIX), or the like.
The storage device 1305 may be a ROM or another type of static storage device that can store static information and instructions, or may be a RAM or another type of dynamic storage device that can store information and instructions, or may be an EEPROM, a compact disc read-only memory (CD-ROM) or another compact disc storage, an optical disc storage (including a compact disc, a laser disc, an optical disc, a digital versatile disc, a Blu-ray disc, or the like), a magnetic disk storage medium, or another magnetic storage device, or any other media that can be configured to carry or store expected program code in a form of instructions or a data structure and that can be accessed by the processor 1301. The storage device includes at least one memory 1302, but is not limited thereto. The storage device 1305 may exist independently, and be connected to the processor 1301 through the bus 1304. The storage device 1305 may also be integrated with the processor 1301.
For example, the memory 1302 and the storage device 1305 may be a first storage medium or a second storage medium in the electronic device 1300. At least one of the memory 1302 and the storage device 1305 may store program code for performing the data processing method in this application. The processor 1301 reads and runs the program code, to enable the electronic device 1300 to perform the data processing method in this application.
In some embodiments, the electronic device 1300 may further include an output device 1307 and an input device 1308. The output device 1307 communicates with the processor 1301, and may display information in a plurality of manners. For example, the output device 1307 may be a liquid crystal display (LCD), a light emitting diode (LED) display device, a cathode ray tube (CRT) display device, a projector, or the like. The input device 1308 communicates with the processor 1301, and may receive an input of a user in a plurality of manners. For example, the input device 1308 may be a mouse, a keyboard, a touchscreen device, or a sensor device.
It should be understood that the data processing apparatus 1100 according to this application may correspond to the electronic device 1300 in this embodiment of this application, and may correspond to a corresponding body in the data processing methods according to embodiments of this application. In addition, the foregoing and other operations and/or functions of the modules in the electronic device 1300 are respectively used to perform corresponding procedures of the methods in
This application further provides a computer-readable storage medium, for example, a memory including program code. The program code may be executed by a processor in an electronic device (or a chip) to perform the data processing methods in the foregoing embodiments. For an implementation of the computer-readable storage medium, refer to the memory 1203 in
This application further provides a computer program product or a computer program. The computer program product or the computer program includes program code, and the program code is stored in the computer-readable storage medium. A processor of the electronic device reads the program code from the computer-readable storage medium, and the processor executes the program code, to enable the processor to perform the foregoing data processing method.
In addition, this application further provides an apparatus. The apparatus may be a chip, a component, or a module. The apparatus may include a processor and a memory that are connected to each other. The memory is configured to store computer-executable instructions. When the apparatus runs, the processor may execute the computer-executable instructions stored in the memory, to enable the chip to perform the data processing methods in the foregoing method embodiments.
In this application, the apparatus, the device, the computer-readable storage medium, the computer program product, or the chip is configured to perform the corresponding methods provided above. Therefore, for beneficial effects that can be achieved by the apparatus, the device, the computer-readable storage medium, the computer program product, or the chip, refer to the beneficial effects in the corresponding methods provided above. Details are not described herein again.
All or some of the foregoing embodiments may be implemented using software, hardware, firmware, or any combination thereof. When software is used to implement embodiments, the foregoing embodiments may be implemented completely or partially in a form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed in a computer, procedures or functions according to embodiments of this application are all or partially generated. The computer may be a general purpose computer, a dedicated computer, a computer network, or other programmable apparatuses. The computer instructions may be stored in a computer-readable storage medium or may be transmitted from a computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website, computer, server, or data center to another website, computer, server, or data center in a wired (for example, a coaxial cable, an optical fiber, or a digital subscriber line (DSL)) or wireless (for example, infrared, radio, or microwave) manner. The computer-readable storage medium may be any usable medium accessible by a computer, or a data storage device, such as a server or a data center, integrating one or more usable media. The usable medium may be a magnetic medium (for example, a floppy disk, a hard disk drive, or a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium may be a solid-state drive (SSD).
The foregoing descriptions are merely implementations of this application. Any variation or replacement readily figured out by a person skilled in the art based on the implementations provided in this application shall fall within the protection scope of this application.
| Number | Date | Country | Kind |
|---|---|---|---|
| 202210682111.X | Jun 2022 | CN | national |
| 202211060725.0 | Aug 2022 | CN | national |
This application is a continuation of International Application No. PCT/CN2023/100447, filed on Jun. 15, 2023, which claims priorities to Chinese Patent Application No. 202211060725.0, filed on Aug. 31, 2022, and Chinese Patent Application No. 202210682111.X, filed on Jun. 15, 2022. All of the aforementioned patent applications are hereby incorporated by reference in their entireties.
| Number | Date | Country | |
|---|---|---|---|
| Parent | PCT/CN2023/100447 | Jun 2023 | WO |
| Child | 18983046 | US |
