Data processing system, method and device and storage medium

By designing a nearby read and write data processing system in the AI ​​chip, the problem of low data processing efficiency of AI chips is solved, and the effect of improving data processing efficiency and throughput is achieved.

CN118963648APending Publication Date: 2024-11-15BEIJING SUIYUAN INTELLIGENT TECH CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202410973764.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-19
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The data processing efficiency of existing AI chips is low, mainly because the memory read and write speed is much slower than the processor's calculation speed, resulting in limited data read and read in each working cycle, limiting the throughput of AI chips. and data processing efficiency.

Method used

Design a data processing system, including an instruction transmitting unit and an in-memory computing module, the distance between the arithmetic logic unit and the matching memory unit is smaller than the distance between the instruction transmitting unit and the arithmetic logic unit, so as to realize nearby reading and writing, and transmit through instruction The unit sends data processing instructions, and the arithmetic logic unit reads the operation data in the memory unit and sends the results to the instruction transmitting unit and the memory unit, optimizing the read and write efficiency of the memory unit.

Benefits of technology

Through nearby reading and writing, the read overhead of the memory unit is reduced, the read and write efficiency of the memory unit is improved, and the data throughput and data processing efficiency of the entire data processing system are improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118963648A_ABST
    Figure CN118963648A_ABST
Patent Text Reader

Abstract

The invention discloses a data processing system, method and device and a storage medium, and relates to the technical field of chips, the data processing system comprises an instruction transmitting unit, an arithmetic logic unit and a memory unit; the distance between the arithmetic logic unit and the matched memory unit is smaller than the distance between the arithmetic logic unit and the instruction transmitting unit; the instruction transmitting unit is used for transmitting the data processing instruction to the arithmetic logic unit; and the arithmetic logic unit is used for reading the operation data in the memory unit according to the data processing instruction transmitted by the instruction transmitting unit and transmitting a data processing result to the instruction transmitting unit and the memory unit. According to the technical scheme provided by the embodiment of the invention, the arithmetic logic unit realizes nearby reading and writing of the data stored in the memory unit, the reading overhead of the memory unit is reduced, the reading and writing efficiency of the memory unit is improved, the data throughput of the data processing system is further improved, and the data processing efficiency is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of chip technology, and in particular to a data processing system, method, device and storage medium. Background Art

[0002] With the continuous development of artificial intelligence (AI) technology, AI chips are gradually moving towards high performance, and data processing efficiency has become an important goal of AI chip design.

[0003] In the prior art, the traditional von Neumann architecture is usually used between the processor and memory of an AI chip. During the calculation process, the input data and weight data need to be written into the memory, and then read and transmitted to the processor for corresponding calculations, so as to realize the reciprocating transmission of data and instructions between the processor and the memory.

[0004] However, with this data processing method, since the read and write speed of the memory is much slower than the computing speed of the processor, the amount of data that can be read in each working cycle is limited, which limits the number of operations in each cycle, reduces the throughput of the AI ​​chip, and limits the data processing efficiency of the AI ​​chip. Summary of the invention

[0005] The present invention provides a data processing system, method, device and storage medium to solve the problem of low data processing efficiency of AI chips.

[0006] According to one aspect of the present invention, there is provided a data processing system, comprising: at least one instruction issuing unit and at least one in-memory computing module; wherein the in-memory computing module comprises an arithmetic logic unit and a memory unit; the arithmetic logic unit is connected to a matching memory unit and the at least one instruction issuing unit; the distance between the arithmetic logic unit and the matching memory unit is smaller than the distance between the arithmetic logic unit and the instruction issuing unit;

[0007] The instruction issuing unit is used to send the data processing instruction to the arithmetic logic unit;

[0008] The arithmetic logic unit is used to read the operation data in the memory unit according to the data processing instruction issued by the instruction issuing unit, and send the data processing result to the instruction issuing unit and the memory unit.

[0009] The instruction issuing unit is further used to obtain attribute information of the data processing instruction and send the data processing instruction configured with the attribute information to the arithmetic logic unit; wherein the attribute information includes an operation code and at least one operand.

[0010] The instruction emission unit is specifically used to determine a matching first arithmetic logic unit and a first memory unit matching the first arithmetic logic unit according to a data processing instruction, and to use an operation code in the data processing instruction and an operand whose storage address is located outside the first memory unit as attribute information of the data processing instruction.

[0011] The arithmetic logic unit includes an address comparison subunit and a sorting execution subunit; the address comparison subunit is connected to the sorting execution subunit, and is used to manage each data processing instruction through a window queue, so as to allocate each data processing instruction to a matching window queue according to an access address; wherein data processing instructions with different access addresses correspond to different window queues; the sorting execution subunit is used to sort the data processing instructions in each window queue.

[0012] The sorting execution subunit is specifically used to obtain the first receiving time and the first access address corresponding to the first data processing instruction if it is determined that the first data processing instruction in the window queue is in a queued state, and use the storage data of the first access address at the first receiving time as the data reading result of the first data processing instruction.

[0013] The sorting execution subunit is specifically used to configure the read and write result of the second data processing instruction to the third data processing instruction if there are second data processing instructions and third data processing instructions arranged in sequence in the current window queue.

[0014] The arithmetic logic unit is also used to enable the clock signal of the arithmetic logic unit if a data processing instruction is detected in the transmission bus, and enable the clock signal of the corresponding calculation subunit according to the operation type of the data processing instruction; wherein the calculation subunit includes a logic operation subunit, an integer operation subunit and a floating-point operation subunit.

[0015] According to another aspect of the present invention, there is provided a data processing method, which is applied to the data processing system according to any embodiment of the present invention, comprising:

[0016] The instruction issuing unit sends the data processing instructions to the arithmetic logic unit;

[0017] The arithmetic logic unit reads the operation data in the memory unit according to the data processing instruction issued by the instruction issuing unit;

[0018] The arithmetic logic unit sends the data processing results to the instruction issuing unit and the memory unit.

[0019] According to another aspect of the present invention, there is provided a data processing device, which is applied to the data processing system according to any embodiment of the present invention, comprising:

[0020] An instruction sending and executing module, configured in the instruction issuing unit, is used to send the data processing instruction to the arithmetic logic unit;

[0021] An operation data acquisition module, configured in the arithmetic logic unit, is used to read the operation data in the memory unit according to the data processing instruction issued by the instruction issuing unit;

[0022] The result sending execution module is configured in the arithmetic logic unit and is used to send the data processing result to the instruction issuing unit and the memory unit.

[0023] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the data processing method described in any embodiment of the present invention when executed.

[0024] According to another aspect of the present invention, a computer program product is provided, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the data processing method according to any embodiment of the present invention.

[0025] According to the technical solution of the embodiment of the present invention, after the instruction issuing unit issues a data processing instruction to the arithmetic logic unit, the arithmetic logic unit configured near the memory unit reads the operation data in the memory unit according to the data processing instruction issued by the instruction issuing unit, and then sends the data processing result based on the read data to the instruction issuing unit and the memory unit. The arithmetic logic unit realizes the nearby reading and writing of the data stored in the memory unit, reduces the reading overhead of the memory unit, improves the reading and writing efficiency of the memory unit, thereby improving the data throughput of the entire data processing system and improving the data processing efficiency.

[0026] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0028] Figure 1 is a structural diagram of a data processing system provided according to Embodiment 1 of the present invention;

[0029] Figure 2is a schematic diagram of a flow chart of a data processing system processing a data processing instruction according to Embodiment 1 of the present invention;

[0030] Figure 3 is a schematic diagram of a flow chart of another data processing system processing a data processing instruction according to the first embodiment of the present invention;

[0031] Figure 4 is a schematic diagram of the structure of an arithmetic logic unit provided according to the second embodiment of the present invention;

[0032] Figure 5 is a schematic diagram of a flow chart of processing a data processing instruction by another data processing system provided in Embodiment 2 of the present invention;

[0033] Figure 6 is a schematic diagram of a flow chart of controlling a clock signal by an arithmetic logic unit according to a third embodiment of the present invention;

[0034] Figure 7 is a flow chart of a data processing method provided according to a fourth embodiment of the present invention;

[0035] Figure 8 It is a structural diagram of a data processing device provided according to Embodiment 5 of the present invention. DETAILED DESCRIPTION

[0036] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0037] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. 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 necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0038] Embodiment 1

[0039] Figure 1 A structural diagram of a data processing system provided in Embodiment 1 of the present invention is shown in FIG. Figure 1 As shown, the data processing system includes at least one instruction emission unit 100 and at least one in-memory computing module 200; wherein the in-memory computing module 200 includes an arithmetic logic unit 210 and a memory unit 220; the arithmetic logic unit 210 is connected to the matching memory unit 220 and the at least one instruction emission unit 100; the distance between the arithmetic logic unit 210 and the matching memory unit 220 is smaller than the distance between the arithmetic logic unit 210 and the instruction emission unit 100.

[0040] The instruction issuing unit 100 is configured in a processor (for example, a central processing unit and a graphics processing unit, etc.) to facilitate timely acquisition of data processing instructions issued by the processor; the arithmetic logic unit 210 is configured near the memory unit 220, and an arithmetic logic unit 210 and a memory unit 220 constitute an in-memory computing module 200; obviously, the physical distance between the instruction issuing unit 100 and the arithmetic logic unit 210 is relatively far, and the physical distance between the arithmetic logic unit 210 and the matching memory unit 220 is relatively close.

[0041] The instruction emission unit 100 is used to send data processing instructions to the arithmetic logic unit 210; wherein the instruction emission unit 100 can be interconnected with each arithmetic logic unit 210 through a crossbar or other connection methods; for example, the instruction emission unit 100 and the arithmetic logic unit 210 can be configured as the fourth generation Advanced Extensible Interface 4th (AXI4) protocol, that is, a communication protocol based on the AXI bus; wherein the AW channel (write request channel) is used for the transmission of data processing instructions, and the B channel is used for the transmission of feedback data.

[0042] In particular, the in-memory computing module 200 in the present application may come from the same AI chip, and different memory units 220 may refer to different areas in the memory of the AI ​​chip; the in-memory computing module 200 may also come from different AI chips, and different memory units 220 refer to all or part of the areas in the memory of different AI chips; that is, the data processing system in the embodiment of the present invention may be configured inside an AI chip, or may be composed of multiple instruction emission units 100 and multiple in-memory computing modules 200 distributed in different AI chips.

[0043] The arithmetic logic unit 210 is used to read the operation data in the memory unit 220 according to the data processing instruction issued by the instruction issuing unit 100, and send the data processing result to the instruction issuing unit 100 and the memory unit 220. Since the arithmetic logic unit 210 is configured near the memory unit 220, after obtaining the data processing instruction issued by the instruction issuing unit 100, the arithmetic logic unit 210 reads the operation data in the memory unit 220 nearby, and then performs corresponding data calculation according to the read operation data and the operation code in the data processing instruction, and stores the data calculation result in the memory unit 220 nearby.

[0044] This not only reduces the reading overhead of the arithmetic logic unit 210 for the memory unit 220 and improves the data calculation efficiency of the arithmetic logic unit 210, but also when the data processing system includes multiple instruction issuance units 100 and multiple in-memory computing modules 200, the parallel computing architecture of the distributed system improves the data processing efficiency; in addition, different instruction issuance units 100 can be configured to issue different types of instructions according to different calculation methods (for example, multiplication and addition operations); because the opcode represents the calculation method, different instruction issuance units 100 can actually be configured to issue different types of instructions according to the different calculation methods represented by the opcode.

[0045] Optionally, in an embodiment of the present invention, the instruction emission unit 100 is also used to obtain attribute information of the data processing instruction, and send the data processing instruction configured with the attribute information to the arithmetic logic unit 210; wherein the attribute information includes an operation code and at least one operand.

[0046] Specifically, the instruction emission unit 100 determines to which arithmetic logic unit 210 to issue the data processing instruction according to the memory unit 220 where the first operand in the data processing instruction is located, that is, the data processing instruction is issued to the arithmetic logic unit 210 located in the same in-memory computing module 200 as the current memory unit 220, and the operation code and one or more operands are used as attribute information, and the data processing instruction and the attribute information are sent to the arithmetic logic unit 210 together, that is, the operand in the attribute information is the data to be calculated in the form of data, rather than the storage address in the form of an address, thereby improving the computing performance of the arithmetic logic unit 210 and optimizing the access energy efficiency of the memory unit 220.

[0047] In particular, the memory unit 220 can be a storage device such as a static random access memory (SRAM) or a synchronous dynamic random access memory (SDRAM). At the same time, the arithmetic logic unit 210 can also access the memory unit 220 through a cache memory (Cache); in addition, the instruction issuance unit 100 can select one or more operands as attribute information according to the instruction bandwidth, for example, all remaining operands except the first operand are used as attribute information.

[0048] like Figure 2 As shown, taking a three-operand data processing instruction as an example, when the instruction emission unit 100 sends the data processing instruction to the arithmetic logic unit 210, the operation code (Opcode), the second operand (Oprand1) and the third operand (Oprand2) are used as attribute information and sent to the arithmetic logic unit 210 together with the data processing instruction; wherein the second operand and the third operand in the attribute information are both in the form of data to be calculated.

[0049] The arithmetic logic unit 210 sends a read request of the first operand (Oprand0) to the memory unit 220, and then the arithmetic logic unit 210 reads the first operand from the memory unit 220. Then the arithmetic logic unit 210 performs corresponding data calculations on the first operand, the second operand, and the third operand according to the operation code; wherein the operation code (Opcode) supports integer operations, floating-point data operations, logical operations, and other types of operations executable by the AI ​​chip; the arithmetic logic unit 210 writes the calculation result into the memory unit 220, and feeds the calculation result back to the instruction issuing unit 100.

[0050] Optionally, in an embodiment of the present invention, the instruction emission unit 100 is specifically used to determine a matching first arithmetic logic unit 210 and a first memory unit 220 matching the first arithmetic logic unit 210 according to a data processing instruction, and use an opcode in the data processing instruction and an operand whose storage address is located outside the first memory unit 220 as attribute information of the data processing instruction.

[0051] Specifically, taking the above technical solution as an example, after the instruction emission unit 100 determines to issue a data processing instruction to the arithmetic logic unit 210 located in the same in-memory computing module 200 as the first memory unit 220 based on the memory unit 220 (i.e., the first memory unit 220) where the first operand in the data processing instruction is located, if a certain operand is located outside the first memory unit 220, the current arithmetic logic unit 210 needs to read the operand across different in-memory computing modules 200. At this time, the instruction emission unit 100 can use the operand whose storage address is located outside the first memory unit 220 as the attribute information of the data processing instruction, while the operand whose storage address is located within the first memory unit 220 is still retained in the data processing instruction in the form of an address, thereby avoiding the data reading operation of the arithmetic logic unit 210 across the in-memory computing module 200, reducing the data reading overhead and improving the data reading efficiency.

[0052] In addition, if Figure 3 As shown, the instruction emission unit 100 can also be directly connected to each memory unit 220. When the instruction emission unit 100 issues a data processing instruction to the arithmetic logic unit 210, it can issue the address information of the first operand to the memory unit 220, so that the memory unit 220 directly provides the first operand stored under the address information to the arithmetic logic unit 210; in addition, when the instruction emission unit 100 issues a data processing instruction to the arithmetic logic unit 210, it can also issue the address information of the first operand to the memory unit 220. After obtaining the address information of the first operand, the memory unit 220 directly provides the first operand to the arithmetic logic unit 210.

[0053] After obtaining the first operand, the arithmetic logic unit 210 performs data calculation based on other operands and operation codes in the attribute information, and writes the calculation results into the memory unit 220; thus, the memory unit 220 not only exists as a data storage medium, but directly participates in the calculation process. According to the instructions of the operation code, the data in the memory unit 220 is directly operated and updated in real time, skipping the loading process of the data processing instructions and the data storage process, reducing the number of data transfers between the memory unit 220 and other storage devices, not only significantly reducing the power consumption of the data processing system, but also reducing the execution delay of the data processing instructions, and improving the system performance and processing efficiency.

[0054] According to the technical solution of the embodiment of the present invention, after the instruction issuing unit issues a data processing instruction to the arithmetic logic unit, the arithmetic logic unit configured near the memory unit reads the operation data in the memory unit according to the data processing instruction issued by the instruction issuing unit, and then sends the data processing result based on the read data to the instruction issuing unit and the memory unit. The arithmetic logic unit realizes the nearby reading and writing of the data stored in the memory unit, reduces the reading overhead of the memory unit, improves the reading and writing efficiency of the memory unit, thereby improving the data throughput of the entire data processing system and improving the data processing efficiency.

[0055] Embodiment 2

[0056] Figure 4 A schematic diagram of the structure of an arithmetic logic unit provided in the second embodiment of the present invention is shown in FIG. Figure 4 As shown, the arithmetic logic unit 210 includes an address comparison subunit 211 and a sorting execution subunit 212; the address comparison subunit 211 is connected to the sorting execution subunit 212, and is used to manage each data processing instruction through a window queue, so as to allocate each data processing instruction to a matching window queue according to an access address; wherein data processing instructions with different access addresses correspond to different window queues; the sorting execution subunit 212 is used to sort the data processing instructions in each window queue.

[0057] Specifically, although parallel computing can improve efficiency and processing speed by executing multiple computing tasks simultaneously, there is a problem of execution order in parallel computing. When multiple computing tasks access or modify shared data at the same time, data inconsistency may occur due to data competition, thereby affecting the accuracy of the calculation results. In addition, when multiple computing tasks wait for each other to release resources, deadlock may occur. Therefore, data processing instructions issued by different instruction emission units 100 may have access address conflicts. It is necessary to ensure data consistency before or after the data processing instruction access operation at the same address so that the atomicity of the data processing instruction access is not affected.

[0058] The window queue is used to carry data processing instructions to which address allocation has been completed. Each data processing instruction in a window queue has the same address space, and the address spaces between different window queues are different. When the address comparison subunit 211 obtains a data processing instruction, the access address space of the data processing instruction (for example, the storage address where the first operand in the data processing instruction is located) is compared with the access address space of the currently existing window queue.

[0059] If the access address of the current data processing instruction matches the access address of a window queue, the current data processing instruction is assigned to the window queue; if the access address of the current data processing instruction does not match the access addresses of all window queues, the current data processing instruction is assigned to an idle window queue or a newly created window queue, and the access address of the newly created window queue or the idle window queue is configured as the access address of the current data processing instruction.

[0060] The sorting execution subunit 212 sorts each data processing instruction in sequence according to the receiving time of each data processing instruction; in particular, when there is no data processing instruction in a window queue, the window queue is regarded as an idle window queue, and then the idle window queue is allocated to other data processing instructions for use.

[0061] For data processing instructions in different window queues, since the access addresses do not interfere with each other, out-of-order reading and writing can be supported at this time, thereby improving the computational efficiency of parallel computing; for each data processing instruction in a window queue, since the access addresses interfere with each other, it is necessary to ensure that the data processing instructions are executed in order to avoid address conflicts that lead to errors in the calculation results and ensure the accuracy of the calculation results of the data processing instructions; Figure 5 For example, the address comparison subunit 211 obtains three data processing instructions (i.e., instruction 0, instruction 1, and instruction 2) in sequence; instruction 0 and instruction 1 have different access addresses, so the two are assigned to different window queues in sequence; instruction 2 has the same access address as instruction 0, so instruction 2 is assigned to the window queue where instruction 0 is located.

[0062] In particular, a maximum number threshold can be configured for the number of window queues. If the number of window queues reaches the maximum number threshold and there is no idle window queue, if the current data processing instruction does not belong to any window queue, the current data processing instruction will be put into waiting processing until an idle window queue appears (that is, a window queue without any data processing instructions). Then, the access address of the idle window queue is changed to the access address of the current data processing instruction, and the current data processing instruction is placed in the idle window queue.

[0063] Optionally, in an embodiment of the present invention, the sorting execution subunit 212 is specifically used to obtain the first receiving time and the first access address corresponding to the first data processing instruction if it is determined that the first data processing instruction in the window queue is in a queued state, and use the storage data of the first access address at the first receiving time as the data reading result of the first data processing instruction.

[0064] Specifically, when the first data processing instruction is obtained, the storage data in its access address (i.e., the first access address) may be data A; since the first data processing instruction is in a queue state, other data processing instructions that are arranged before it will update the storage data in the first access address. When the first data processing instruction is processed, the storage data in the first access address may have been changed to data B. In fact, for the first data processing instruction, if it is processed immediately after being obtained, its corresponding read data should be data A.

[0065] As described above, due to the influence of sorting, the data in the first access address has been updated. At this time, the receiving time of the first data processing instruction (i.e., the first receiving time) can be recorded, and the storage data of the first access address at the first receiving time (i.e., data A) can be used as the data reading result of the first data processing instruction; thereby, according to the receiving time of the data processing instruction, it is ensured that the data reading result matches the receiving time of the data processing instruction, thereby avoiding the write operation results of other data processing instructions in the window queue affecting the read operation results of subsequent data processing instructions.

[0066] Optionally, in an embodiment of the present invention, the sorting execution subunit 212 is specifically used to configure the read and write results of the second data processing instruction to the third data processing instruction if there are second data processing instructions and third data processing instructions arranged in sequence in the current window queue.

[0067] Specifically, each data processing instruction within the window queue corresponds to the same access address. For two adjacent data processing instructions, the read operation result of the latter data processing instruction (i.e., the third data processing instruction) is actually the result of the previous data processing instruction (i.e., the second data processing instruction) after executing the read operation and / or write operation. Therefore, the read and write results of the second data processing instruction are directly configured to the third data processing instruction, eliminating the operation process of writing the data of the second data processing instruction to the memory unit 220 and the operation process of the third data processing instruction reading the data of the memory unit 220, reducing the number of accesses to the memory unit 220, reducing the calculation delay, and avoiding the pipeline performance of the opcode from reducing the instruction throughput and calculation speed.

[0068] According to the technical solution of the embodiment of the present invention, the address comparison subunit in the arithmetic logic unit manages each data processing instruction through a window queue to allocate each data processing instruction to a matching window queue according to the access address, and the sorting execution subunit sorts the data processing instructions in each window queue. This ensures the data consistency before or after the data processing instructions at the same address execute the access operation, so that the atomicity of the data processing instruction access is not affected, and while improving the computing efficiency through parallel computing, it avoids address conflicts that lead to errors in the computing results, and ensures the accuracy of the computing results of the data processing instructions.

[0069] Embodiment 3

[0070] Figure 6 A schematic diagram of a flow chart of a clock signal control by an arithmetic logic unit provided in the third embodiment of the present invention, such as Figure 6 As shown, the arithmetic logic unit 210 is also used to enable the clock signal of the arithmetic logic unit 210 if a data processing instruction is detected in the transmission bus, and enable the clock signal of the corresponding calculation subunit according to the operation type of the data processing instruction; wherein the calculation subunit includes a logic operation subunit, an integer operation subunit and a floating-point operation subunit.

[0071] Specifically, two layers of clock signals are configured in the arithmetic logic unit 210. The first layer of clock signal is the clock signal of the arithmetic logic unit 210 itself. If it is detected that there are data processing instructions in the transmission bus between any instruction issuing unit 100 and the arithmetic logic unit 210, the first layer of clock signal is enabled; if it is detected that there are no data processing instructions in the transmission bus between each instruction issuing unit 100 and the arithmetic logic unit 210, the first layer of clock signal is not enabled.

[0072] The second layer of clock signals, namely, the clock signals of the various computing sub-units inside the arithmetic logic unit 210, for example, the clock signals of the computing sub-units such as the logic operation sub-unit, the integer operation sub-unit and the floating-point operation sub-unit, the above-mentioned computing sub-units are respectively used for logic operation, integer operation and floating-point data operation; after the first layer of clock signals are enabled, the computing task type is determined according to the operation code in the data transmission instruction, and then the matching computing sub-unit is determined according to the computing task type, and the clock signal of the corresponding computing sub-unit is enabled; the clock signals of other computing sub-units remain in the non-enabled state; in addition, the arithmetic logic unit 210 also supports mixed operations of scalars and vectors. If a computing sub-unit is not assigned a computing task, an empty instruction can be configured for the computing sub-unit.

[0073] According to the technical solution of the embodiment of the present invention, when the arithmetic logic unit detects the presence of a data processing instruction in the transmission bus, the clock signal of the arithmetic logic unit is enabled, and according to the operation type of the data processing instruction, the clock signal of the corresponding computing sub-unit is enabled, thereby realizing precise control of two-level clock signals and saving the computing power consumption of the arithmetic logic unit.

[0074] Embodiment 4

[0075] Figure 7 This is a flowchart of a data processing method provided in Embodiment 4 of the present invention. The method can be executed by a data processing device. The data processing device can be implemented in the form of hardware and / or software. The data processing device can be configured in the data processing system in any embodiment of the present invention. Figure 7 As shown, the method includes:

[0076] S401, the instruction issuing unit sends the data processing instruction to the arithmetic logic unit.

[0077] S402: The arithmetic logic unit reads the operation data in the memory unit according to the data processing instruction issued by the instruction issuing unit.

[0078] S403, the arithmetic logic unit sends the data processing result to the instruction issuing unit and the memory unit.

[0079] According to the technical solution of the embodiment of the present invention, after the instruction issuing unit issues a data processing instruction to the arithmetic logic unit, the arithmetic logic unit configured near the memory unit reads the operation data in the memory unit according to the data processing instruction issued by the instruction issuing unit, and then sends the data processing result based on the read data to the instruction issuing unit and the memory unit. The arithmetic logic unit realizes the nearby reading and writing of the data stored in the memory unit, reduces the reading overhead of the memory unit, improves the reading and writing efficiency of the memory unit, thereby improving the data throughput of the entire data processing system and improving the data processing efficiency.

[0080] Embodiment 5

[0081] Figure 8 : is a structural block diagram of a data processing device provided by Embodiment 5 of the present invention, and the device specifically includes:

[0082] An instruction sending and executing module 501, configured in the instruction issuing unit, is used to send the data processing instruction to the arithmetic logic unit;

[0083] The operation data acquisition module 502 is configured in the arithmetic logic unit and is used to read the operation data in the memory unit according to the data processing instruction issued by the instruction issuing unit;

[0084] The result sending execution module 503 is configured in the arithmetic logic unit and is used to send the data processing result to the instruction issuing unit and the memory unit.

[0085] According to the technical solution of the embodiment of the present invention, after the instruction issuing unit issues a data processing instruction to the arithmetic logic unit, the arithmetic logic unit configured near the memory unit reads the operation data in the memory unit according to the data processing instruction issued by the instruction issuing unit, and then sends the data processing result based on the read data to the instruction issuing unit and the memory unit. The arithmetic logic unit realizes the nearby reading and writing of the data stored in the memory unit, reduces the reading overhead of the memory unit, improves the reading and writing efficiency of the memory unit, thereby improving the data throughput of the entire data processing system and improving the data processing efficiency.

[0086] The above device can execute the data processing method provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method. For technical details not described in detail in this embodiment, please refer to the data processing method provided by any embodiment of the present invention.

[0087] Embodiment 6

[0088] In some embodiments, the data processing method may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on a heterogeneous hardware accelerator via a ROM and / or a communication unit. When the computer program is loaded into RAM and executed by a processor, one or more steps of the data processing method described above may be performed. Alternatively, in other embodiments, the processor may be configured to perform the data processing method in any other appropriate manner (e.g., by means of firmware).

[0089] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0090] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0091] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0092] To provide interaction with a user, the systems and techniques described herein may be implemented on a heterogeneous hardware accelerator having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and pointing device (e.g., a mouse or trackball) through which a user can provide input to the heterogeneous hardware accelerator. Other types of devices may also be used to provide interaction with a user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0093] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0094] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0095] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0096] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A data processing system, characterized in that: include: At least one instruction issuing unit and at least one in-memory computing module; wherein the in-memory computing module includes an arithmetic logic unit and a memory unit; the arithmetic logic unit is connected to a matching memory unit and the at least one instruction issuing unit; the distance between the arithmetic logic unit and the matching memory unit is smaller than the distance between the arithmetic logic unit and the instruction issuing unit; The instruction issuing unit is used to send the data processing instruction to the arithmetic logic unit; The arithmetic logic unit is used to read the operation data in the memory unit according to the data processing instruction issued by the instruction issuing unit, and send the data processing result to the instruction issuing unit and the memory unit.

2. The data processing system according to claim 1, characterized in that The instruction issuing unit is further used to obtain attribute information of the data processing instruction and send the data processing instruction configured with the attribute information to the arithmetic logic unit; wherein the attribute information includes an operation code and at least one operand.

3. The data processing system according to claim 2, characterized in that: The instruction emission unit is specifically used to determine a matching first arithmetic logic unit and a first memory unit matching the first arithmetic logic unit according to a data processing instruction, and to use an operation code in the data processing instruction and an operand whose storage address is located outside the first memory unit as attribute information of the data processing instruction.

4. The data processing system according to claim 1, characterized in that: The arithmetic logic unit includes an address comparison subunit and a sort execution subunit; The address comparison subunit is connected to the sorting execution subunit, and is used to manage each data processing instruction through a window queue, so as to allocate each data processing instruction to a matching window queue according to the access address; wherein data processing instructions with different access addresses correspond to different window queues; The sorting execution subunit is used to sort the data processing instructions in each window queue.

5. The data processing system according to claim 4, characterized in that: The sorting execution subunit is specifically used to obtain the first receiving time and the first access address corresponding to the first data processing instruction if it is determined that the first data processing instruction in the window queue is in a queued state, and use the storage data of the first access address at the first receiving time as the data reading result of the first data processing instruction.

6. The data processing system according to claim 4, characterized in that: The sorting execution subunit is specifically used to configure the read and write result of the second data processing instruction to the third data processing instruction if there are second data processing instructions and third data processing instructions arranged in sequence in the current window queue.

7. The data processing system according to claim 1, characterized in that: The arithmetic logic unit is also used to enable the clock signal of the arithmetic logic unit if a data processing instruction is detected in the transmission bus, and enable the clock signal of the corresponding calculation subunit according to the operation type of the data processing instruction; wherein the calculation subunit includes a logic operation subunit, an integer operation subunit and a floating-point operation subunit.

8. A data processing method, characterized in that: The data processing system applied to any one of claims 1 to 7 comprises: The instruction issuing unit sends the data processing instructions to the arithmetic logic unit; The arithmetic logic unit reads the operation data in the memory unit according to the data processing instruction issued by the instruction issuing unit; The arithmetic logic unit sends the data processing results to the instruction issuing unit and the memory unit.

9. A data processing device, characterized in that: The data processing system applied to any one of claims 1 to 7 comprises: An instruction sending and executing module, configured in the instruction issuing unit, is used to send the data processing instruction to the arithmetic logic unit; An operation data acquisition module, configured in the arithmetic logic unit, is used to read the operation data in the memory unit according to the data processing instruction issued by the instruction issuing unit; The result sending execution module is configured in the arithmetic logic unit and is used to send the data processing result to the instruction issuing unit and the memory unit.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the data processing method of claim 8 when executed.

Citation Information

Cited By

  • In-storage calculation module, in-storage calculation method, equipment and storage medium

    CN120540714A

  • In-memory computing module, in-memory computing method, device, storage medium

    CN120540714B