Processor and information processing apparatus
The processor design improves throughput by using sequence and lock registers to manage order guarantee across multiple memories, allowing simultaneous request issuance without waiting for previous requests to finish, thus addressing throughput reduction in distributed memory systems.
Patent Information
- Application Number
- JP2023082542
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-18
- Publication Date
- 2024-11-28
AI Technical Summary
In a configuration where memory areas are distributed across multiple memories, existing techniques face a decrease in throughput due to the need to maintain order guarantee, leading to suppressed issuance of read or write requests until previous requests are completed.
A processor design that includes an issuing unit, memory interfaces, sequence registers, and lock registers to manage and ensure order guarantee while allowing simultaneous issuance of requests, utilizing shared completion information across multiple memory interfaces to prevent overlapping access.
The solution enhances throughput by enabling the issuance of read or write requests without waiting for previous requests to complete, while maintaining order guarantee, even in distributed memory configurations.
Smart Images

Figure 2024165938000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to a processor and an information processing device. [Background technology]
[0002] In a processor, the order may be guaranteed so that the order in which read or write requests are issued matches the order in which read or write commands are executed based on the requests. To ensure the order, the issuer refrains from issuing the next request until it is confirmed that all preceding requests have been executed. This may result in a decrease in throughput, which is the number of read or write requests output per unit time. A technique has been proposed that uses a shared buffer to enable the following request to overtake if a buffer miss or the like occurs in the preceding request. [Prior art documents] [Patent documents]
[0003] [Patent Document 1] JP 2004-005710 A Summary of the Invention [Problem to be solved by the invention]
[0004] In a configuration in which memory areas are distributed across multiple memories, it is desirable to avoid a decrease in throughput when issuing read or write requests while maintaining order assurance.
[0005] In one aspect, the present invention aims to improve the throughput of issuing read or write requests while maintaining order in a configuration in which memory areas are distributed and arranged across multiple memories. [Means for solving the problem]
[0006] In one aspect, the processor includes an issuing unit that issues write or read requests, a plurality of memory interfaces respectively connected to a plurality of memories, one or more first registers that store the memory addresses of destinations of the requests and a value indicating an order of the requests, and a second register into which is written information indicating that a request has been completed for each order value, and the first register extends a lock on the memory addresses of destinations of the requests whose order value is the first value until it confirms, based on the second register, that all requests whose order value is less than or equal to a first value have been completed. Effect of the Invention
[0007] In one aspect, the present invention can improve the throughput of issuing read or write requests while maintaining order guarantees, even in a configuration in which memory areas are distributed and arranged across multiple memories. [Brief description of the drawings]
[0008] [Figure 1] 13 is an example of a timing chart for issuing a subsequent request after confirming that all preceding requests with guaranteed order have been executed. [Diagram 2] 2 is a flowchart showing an example of the process of FIG. 1. [Diagram 3] 1 is a block diagram showing an example of a hardware (HW) configuration of an information processing device according to a first embodiment. [Figure 4] FIG. 4 is a diagram illustrating an example of a format of a sequence register in the first embodiment. [Diagram 5] FIG. 4 is a diagram illustrating an example of a format of a lock register in the first embodiment. [Figure 6] FIG. 2 is a diagram illustrating an outline of a processing procedure in the information processing device according to the first embodiment. [Figure 7] 4 is an example of a timing chart showing a processing procedure performed by the information processing device according to the first embodiment. [Figure 8]11 is an example of a timing chart showing a procedure for resetting sequence information by the information processing device according to the first embodiment. [Figure 9] 11 is a flowchart showing an example of an IO request processing operation by the information processing device according to the first embodiment. [Figure 10] FIG. 11 is a block diagram showing an example of the HW configuration of an information processing device according to a second embodiment. [Figure 11] FIG. 11 is a diagram illustrating an example of a format of a sequence register in the second embodiment. [Figure 12] FIG. 11 is a diagram illustrating an outline of a processing procedure in an information processing device according to a second embodiment. [Figure 13] 10 is an example of a timing chart showing a processing procedure performed by the information processing device according to the second embodiment. [Figure 14] 13 is a flowchart illustrating an example of an IO request processing operation by the information processing device according to the second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0009] An embodiment will be described below with reference to the drawings. However, the embodiment described below is merely an example, and there is no intention to exclude the application of various modified examples and techniques not specified in the embodiment. In other words, the present embodiment can be implemented with various modifications within the scope of the gist of the embodiment. In addition, each figure does not intend to include only the components shown in the figure, but can include other functions, etc.
[0010] [A] Explanation of related methods Fig. 1 is an example of a timing chart for issuing a subsequent request after confirming that all preceding requests with guaranteed order have been executed. Fig. 2 is a flowchart showing an example of the process of Fig. 1.
[0011] 1 shows a case where there is an IO request (SEQ=0) and an IO request (SEQ=1) for which order guarantee is required. The IO request is an example of a write or read request.
[0012] When order assurance is required, the processor will not start executing the subsequent IO request until all previously issued IO requests have been executed. Therefore, in the example of Fig. 1, the processor issues (a) the IO request (SEQ=0) that precedes the previous one in the issue order, and refrains from issuing (c) the next IO request (SEQ=1) until it is confirmed that the request has been executed (b). The processor confirms the execution of the IO request (SEQ=1) (d) and continues to execute the same process.
[0013] As shown in Fig. 2, when an IO request arrives at a request issuer (step S10), the issuer of the IO request determines whether there is a preceding IO request that requires in-order guarantee and is not yet completed (step S11). If there is a preceding IO request that requires in-order guarantee and is not yet completed (see the Yes route in step S11), the issuer suppresses the issuance of the IO request. The issuer waits for all preceding IO requests that require in-order guarantee to be completed (see the No route in step S11), and then issues the IO request to the memory interface that is the destination of the IO request (step S12).
[0014] The lock register puts the destination address of the IO request into a locked state (step S13). In the locked state, the memory interface rejects other requests to the destination address.
[0015] The memory interface executes the command of the IO request, that is, a read command or a write command for the requested address (step S14).
[0016] When the completion of the IO request is confirmed, the lock register releases the lock state of the request destination address (step S15), and the memory interface notifies the issuer of the IO request that the IO request has been completed (step S16).
[0017] Even when the processor has multiple memories #1 and #2, the processor can maintain the order guarantee by the method shown in Fig. 1 and Fig. 2. However, since the issuance of subsequent IO requests must be suppressed while waiting for the execution completion response of the preceding IO request at the IO request issuer, there is a concern that the throughput of IO requests may decrease. Therefore, an information processing device of an embodiment is shown that improves the throughput of IO requests even in a configuration having multiple memories #1 and #2 and memory areas distributed and arranged in multiple memories.
[0018] [B] First embodiment [B-1] Hardware configuration example FIG. 3 is a block diagram showing an example of the hardware (HW) configuration of the information processing device 1 according to the first embodiment.
[0019] The information processing device 1 is a computer system. The information processing device 1 includes a processor 100 and a plurality of memories 3-1 to 3-4 (sometimes collectively referred to as memories 3). The information processing device 1 has a configuration in which a memory area is distributed among a plurality of memories 3. The number of memories 3 is not limited to that shown in FIG.
[0020] The processor 100 is a processing device that performs various controls and calculations. The processor 101 executes an Operating System (OS) and programs loaded into the memory 3 to realize various functions. The processor 100 may be a CPU (Central Processing Unit). The processor 100 may be a multiprocessor including multiple CPUs, a multicore processor having multiple CPU cores, or a configuration having multiple multicore processors. In this example, the processor 100 is a multicore processor having multiple CPU cores.
[0021] The processor 100 includes an IO request issuer 10, a plurality of memory interfaces 20-1 to 20-4 (sometimes collectively referred to as memory interfaces 20), and caches 30-1 to 30-4 (sometimes collectively referred to as caches 30). The processor 100 also includes CPU cores 31-1a, 31-1b to 31-4a, and 31-4b (sometimes collectively referred to as CPU cores 31) that can communicate with the cache 30.
[0022] The IO request issuer 10 issues an IO request. The IO request is an example of a write or read request. In this embodiment, the order of the IO requests is guaranteed. The IO request is a request to execute a read command or a write command to a request destination address in the memory 3. The IO request issuer 10 is an example of an issuer that issues a write or read request.
[0023] An in-order guaranteed IO request issued by the IO request issuer 10 includes order information SEQ_ID and a destination address. The order information SEQ_ID is an example of a value indicating the order of an IO request. The destination address means a memory address to which a read command or a write command based on an IO request is issued. The destination address may be called a reference address or a destination address. An in-order guaranteed IO request may include information about an in-order guarantee flag. The in-order guarantee flag is information indicating that an IO request is an in-order guaranteed IO request.
[0024] The IO request issuer 10 has a sequence register 11. The sequence register 11 records the current order (sequence) of the IO requests. The IO request issuer 10 refers to the sequence register 11 to confirm the current order, and attaches sequence information SEQ_ID to the IO request.
[0025] The memory interfaces 20-1 (#1), 20-2 (#2), 20-3 (#3), and 20-4 (#4) are provided corresponding to the memories 3-1 (#1), 3-2 (#2), 3-3 (#3), and 3-4 (#4), respectively.
[0026] The memory interface 20 writes or reads data to or from the requested address of the corresponding memory 3 based on the IO request.
[0027] The memory interface 20-1 (#1) has a sequence register 21-1 (#1) and a lock register 22-1 (#1). Similarly, the memory interfaces 20-2 (#2) to 20-4 (#4) have sequence registers 21-2 (#2) to 21-4 (#4) and lock registers 22-2 (#2) to 22-4 (#4), respectively.
[0028] The sequence registers 21-1 (#1) to 21-4 (#4) may be collectively referred to as the sequence register 21. The lock registers 22-1 (#1) to 22-4 (#4) may be collectively referred to as the lock register 22. The sequence register 21 is an example of one or more second registers in which completion of an IO request is written for each value indicating the order of the IO request (i.e., request). The lock register 22 is an example of a first register.
[0029] Information indicating whether an IO request including sequence information has been completed in the entire information processing device 1 (i.e., the system) is written to each sequence register 21. The completion of an IO request may mean the completion of a write command or a read command based on the IO request.
[0030] When the memory interface 20-1 receives a notification that the IO request having the sequence information SEQ_ID for the memory 3-1 has been completed, the memory interface 20-1 writes the completion information to its own sequence register 21-1. Furthermore, the memory interface 20-1 also notifies the other memory interfaces 20-2 to 20-4 that the IO request having the sequence information SEQ_ID has been completed. The memory interfaces 20-2 to 20-4 write the completion information to the sequence registers 21-2 to 21-4 included in the respective memory interfaces 20-2 to 20-4.
[0031] 4 is a diagram showing an example of the format of the sequence register 21 in the first embodiment. The format of the sequence register 11 is also the same as that shown in FIG.
[0032] The sequence register 21 records whether or not a plurality of pieces of sequence information (SEQ_ID) are completed. In one example, when an IO request with SEQ_ID=1 is issued and completed, 1 is recorded as the completion information for SEQ_ID_1.
[0033] The multiple sequence registers 21 share with each other completion information regarding IO requests having multiple pieces of sequence information (SEQ_ID).
[0034] Returning to FIG. 3, the lock registers 22-1 to 22-4 are provided corresponding to the plurality of memories 3-1 to 3-4, respectively.
[0035] 5 is a diagram showing an example of the format of the lock register 22 in the first embodiment. Each memory interface 20 writes a lock address 60 (i.e., a request destination address), an order guarantee flag 61, and order information 62 (SEQ_ID) to the lock register 22 based on an IO request from an IO request issuer 10.
[0036] The lock register 22-1 constantly monitors the corresponding sequence register 21-1. The lock register 22-1 confirms that all IO requests (e.g., SEQ_ID=0 and 1) whose value indicated by the sequence information SEQ_ID is equal to or less than a first value (e.g., SEQ_ID=1) have been completed. Until the confirmation is made, the lock register 22-1 extends and maintains the address lock without releasing the lock on the destination memory address of the IO request whose sequence is the first value (e.g., SEQ_ID=1). When the destination address is in a locked state, other IO requests to the destination address are not executed.
[0037] In other words, when the lock register 22-1 confirms that all write or read commands based on IO requests prior to the corresponding issuance order have been completed, it releases the lock state for the lock address 60. In one example, the sequence guarantee flag 61 may indicate whether the lock address 60 is in a locked state or an unlocked state. When the lock register 22-1 confirms that all write or read commands based on IO requests prior to the issuance order indicated by the corresponding sequence information SEQ_ID have been completed, it releases the lock state for the lock address 60.
[0038] Returning to Fig. 3, the cache 30 is a memory circuit that can be accessed at high speed by the CPU core 31. In Fig. 3, one cache 30 is associated with two CPU cores 31, but the information processing device 1 is not limited to this case.
[0039] Fig. 6 is a diagram for explaining an outline of a processing procedure in the information processing device 1 according to the first embodiment. Fig. 6 illustrates a case where an IO request is issued to a request destination address in the memory 3-1.
[0040] The memory interface 20-1 acquires an IO request issued by the IO request issuer 10. The memory interface 20-1 writes the lock address 60 (that is, the request destination address) and sequence information SEQ_ID to the lock register 22-1 (process (1)).
[0041] The memory interface 20-1 processes the IO request, i.e., executes a write command or a read command based on the IO request (process (2)). When the IO request is completed in the memory 3-1, a completion notice (e.g., a flag value of 1) is written to the corresponding sequence information SEQ_ID in the sequence register 21-1 (process (3)).
[0042] The memory interface 20-1 notifies the sequence registers 21-1 to 21-4 in the other memory interfaces 20-2 to 20-4 of the sequence information SEQ_ID of the completed IO request (process (4)). The memory interface 20-1 may also notify the sequence register 11 of the IO request issuer 10 of the sequence information SEQ_ID of the completed IO request.
[0043] The lock registers 22-1 to 22-4 refer to the corresponding sequence registers 21-1 to 21-4, and when it is confirmed that all IO requests prior to the issuance order indicated by SEQ_ID have been completed, release the address lock of the request destination address (process (5)).
[0044] Fig. 7 is an example of a timing chart showing a processing procedure by the information processing device 1 according to the first embodiment. In Fig. 7, an example will be described in which the IO request is a data write request. In Fig. 7, an IO request (sequence information SEQ_ID=0) and an IO request (sequence information SEQ_ID=1) are issued in this issuance order, and the IO request (sequence information SEQ_ID=1) is completed prior to the IO request (sequence information SEQ_ID=0).
[0045] 7, the IO request source 10 refers to the sequence register 11. The IO request source 10 confirms that SEQ_ID=0 is the current sequence based on the reference result of the sequence register 11. After the confirmation, the IO request source 10 issues an IO request with SEQ_ID=0, i.e., a data write request.
[0046] In the example of Fig. 7, the destination of the IO request with SEQ_ID=0 is the memory interface 20-2 (memory interface #2 in Fig. 7). The memory interface 20-2 executes a memory write command to the memory 3-2 (memory #2 in Fig. 7).
[0047] In parallel with the execution of the memory write command, the memory interface 20-2 writes each piece of information, that is, a lock address 60, an order guarantee flag 61, and order information 62 (SEQ_ID=0), to the lock register 22-2 (lock register #2 in FIG. 7). The lock address 60 is the request destination address.
[0048] The lock register 22-2 locks the destination address (ie, the lock address 60) of an IO request during instruction execution in the memory interface 20-2, and rejects other requests to that destination address.
[0049] Next, the IO request issuer 10 refers to the sequence register 11. The IO request issuer confirms that SEQ_ID=1 following SEQ_ID=0 is the current order, i.e., the sequence, based on the reference result of the sequence register 11. After this confirmation, the IO request issuer 10 issues an IO request with SEQ_ID=1.
[0050] The destination of the IO request with SEQ_ID=1 is the memory interface 20-1 (memory interface #1 in FIG. 7). The memory interface 20-1 executes a memory write command to the memory 3-1 (memory #1 in FIG. 7).
[0051] The memory interface 20-1 may write each piece of information, such as a lock address 60, an order guarantee flag 61, and order information 62 (SEQ_ID=1), to a lock register 22-1 (lock register #1 in FIG. 7) in parallel with the execution of the instruction.
[0052] The lock register 22-1 locks the destination address (ie, the lock address 60) of an IO request during instruction execution in the memory interface 20-1, and rejects other requests to that destination address.
[0053] When the processing of the IO request with SEQ_ID=1, that is, the data write command to the memory 3-1, is completed, the memory interface 20-1 notifies the sequence register 21-1 that the IO request with SEQ_ID=1 is completed. The memory interface 20-1 also transmits the information that the IO request with SEQ_ID=1 is completed to the multiple sequence registers 21-2 to 21-4 provided in the multiple other memory interfaces 20-2 to 20-4, respectively. As a result, the information that the IO request with SEQ_ID=1 is completed is written to each sequence register 21. In one example, 1 is recorded as the completion information for SEQ_ID 1. Even if the IO request with SEQ_ID=1 is completed earlier than the IO request with SEQ_ID=0, the processing is executed.
[0054] When a data write command to the memory 3-2, which is the processing of the IO request with SEQ_ID=0, is completed, the memory interface 20-2 notifies the sequence register 21-2 that the IO request with SEQ_ID=0 is completed. The memory interface 20-2 transmits a completion notification of the IO request with SEQ_ID=0 to the sequence registers 21-1, 21-3, and 21-4 in the other memory interfaces 20-1, 20-3, and 20-4.
[0055] The memory interface 20-2 checks the value of the sequence register 21-2, and when it is confirmed that the requests before SEQ_ID=0 have been completed, it sets the sequence guarantee flag 61 of the lock register 22-2 to 0. This causes the lock register 22-2 to unlock the lock address 60 and enter a state in which other requests to that address can be accepted. The memory interface 20-2 also notifies the sequence register 11 of the IO request issuer 10 of the completion of SEQ_ID=0.
[0056] The memory interface 20-1 checks the value of the sequence register 21-1, and when it is confirmed that the requests before SEQ_ID=1 have been completed, it sets the sequence guarantee flag 61 of the lock register 22-1 to 0. This causes the lock register 22-1 to unlock the lock address 60 and enter a state in which other requests to that address can be accepted. The memory interface 20-1 also notifies the sequence register 11 of the IO request issuer 10 of the completion of SEQ_ID=1.
[0057] According to the process shown in Fig. 7, the throughput of IO requests can be increased even in a configuration in which the destination addresses of IO requests are distributed among a plurality of memories 3-1 to 3-4. Specifically, an IO request (SEQ_ID = 1) can be issued without waiting for a completion notification of a preceding issued IO request (SEQ_ID = 0). Even in a situation in which an IO request has been completed, each memory interface can prevent execution of other commands, such as reading, for the destination address until a completion notification of all commands prior to the issuance order is received. Therefore, the order of IO requests can be guaranteed.
[0058] 7, each memory interface 20 also notifies the sequence register 11 of the IO request issuer 10 of the completion of an IO request including issued sequence information 62 (SEQ_ID). Therefore, by referring to the sequence register 11, the IO request issuer 10 can confirm that all IO requests for which sequence is guaranteed are completed.
[0059] When it is confirmed that all IO requests accompanied by sequence information 62 (SEQ_ID) have been completed in the IO request issuer 10, the processor 100 can reset the sequence information 62 (SEQ_ID) to an initial value (for example, SEQ_ID=0). This allows the IO request issuer 01 to avoid erroneous determination due to wrap-around of SEQ_ID. Wrap-around is a problem in which, in the case of the sequence register 21 shown in FIG. 4, when the sequence information 62 (SEQ_ID) exceeds the range from SEQ_ID=0 to 8, after reaching the end of the processable range, SEQ_ID=8, the initial SEQ_ID=0 also returns.
[0060] 8 is an example of a timing chart showing a procedure for resetting order information by the information processing device 1 according to the first embodiment. The processor 100 executes the reset procedure in a state where all IO requests for which order is guaranteed have been completed.
[0061] The sequence register 11 stops issuing IO requests that guarantee order until the reset procedure is completed. The sequence register 11 of the IO request issuer 10 issues a command to reset the values of the sequence registers 21-1 to 21-4 to all the sequence registers 21-1 to 21-4.
[0062] When a reset command arrives, the sequence registers 21-1 to 21-4 (sequence registers #1 and #2 in FIG. 8) clear the values of the sequence registers 21-1 to 21-4, respectively. The sequence registers 21-1 to 21-4 notify the sequence register 11 in the IO request issuer 10 of a SEQ_ID reset completion notification indicating that the reset of the values of the sequence registers 21-1 to 21-4 has been completed. When the SEQ_ID reset completion notification is returned from all destinations, the reset procedure is completed. As a result, the suspension of the issuance of IO requests whose order is guaranteed is lifted and resumed. Therefore, the sequence information 62 returns to the initial value SEQ_ID=0 again. In this way, it becomes possible to handle IO requests greater than the number of SEQ_IDs that can be stored in the sequence register.
[0063] [B-2] Example of operation FIG. 9 is a flowchart showing an example of an IO request processing operation by the information processing device 1 according to the first embodiment.
[0064] When an IO request arrives at the IO request source 10 (step S21), the IO request source 10 issues an IQ request including sequence information 62 (SEQ_ID) (step S22). The IO request includes a sequence guarantee flag 61 and a request destination address.
[0065] Each memory interface 20 writes a lock address 60 (i.e., a destination address), an order flag 61, and order information 62 (SEQ_ID) to the lock register 22 based on information included in the IO request from the IO request issuer 10. In the case of an order-guaranteed IO request, in one example, the initial order flag 61 is 1. In this embodiment, when the order flag 61 is 1, it means that order is guaranteed. In one example, the lock register 22 locks the lock address 60 when the order flag 61 is 1 (step S23).
[0066] Each memory interface 20 executes the command of the IO request (step S24). The memory interface 20 confirms the completion of the IO request (step S25).
[0067] The memory interface 20 notifies the IO request issuer 10 that the IO request has been completed (step S26). By receiving the notification of the completion of the IO request, the IO request issuer 10 can determine whether the issued IO request has been completed and execute the reset procedure (FIG. 8).
[0068] The memory interface 20 notifies all the sequence registers 21-1 to 21-4 that the IO request of the sequence method (SEQ_ID) is completed (step S27). In one example, when an IO request of SEQ_ID=1 is issued and completed, 1 is recorded as completion information for SEQ_ID 1.
[0069] The memory interface 20 waits for all IO requests having regular information before the issuance order indicated by the sequence information (SEQ_ID) to be completed (see the No route in step S28). If the memory interface 20 can confirm that all IO requests having sequence information before the issuance order have been completed (see the YES route in step S28), it proceeds to step S29.
[0070] In step S29, the memory interface 20 unlocks the lock address that is the destination address of the IO request of that issuing order.
[0071] According to the information processing device 1 of the first embodiment, even in a configuration in which the destination addresses of the IO requests are distributed among a plurality of memories 3-1 to 3-4, it is possible to increase the throughput of the IO requests. Furthermore, the processor 100 can guarantee the order without adding a separate order control for the IO requests in the IO request issuing source 10.
[0072] [C] Second embodiment [C-1] Hardware configuration example FIG. 10 is a block diagram showing an example of a HW configuration of an information processing device 1 according to the second embodiment. In the information processing device 1, the sequence registers 21-1 to 21-4 are omitted from the information processing device 1 shown in FIG. 3. The sequence register 70 referred to by the IO request issuer 10 is different from the sequence register 11 of the first embodiment. The sequence register 70 is an example of a first register. Other configurations of the information processing device 1 of the second embodiment are similar to those of the information processing device 1 of the first embodiment, and therefore repeated explanations will be omitted. Also, similar components will be explained using the same member numbers.
[0073] 11 is a diagram showing an example of the format of the sequence register 70 in the second embodiment. The sequence register 70 includes a completion information recording section 71 that records whether or not a plurality of pieces of sequence information (SEQ_ID) are completed, and a current sequence information section 72 that indicates the current sequence (SEQ_ID).
[0074] The completion information recording unit 71 has the same format as the sequence register 21 shown in Fig. 4. The completion information recording unit 71 records whether or not a plurality of sequences (SEQ_ID) are completed.
[0075] The current sequence information unit 72 stores sequence information indicating the current sequence. The IO request issuer 10 updates the sequence information in the current sequence information unit 72 every time it issues an IO request with sequence guarantee.
[0076] Fig. 12 is a diagram for explaining an outline of a processing procedure in the information processing device 1 according to the second embodiment. Fig. 12 illustrates a case where an IO request is issued to a request destination address in the memory 3-1.
[0077] The IO request source 10 issues an IO request with SEQ_ID=0. The IO request source 10 writes SEQ_ID=1 in the current sequence information section 72 of the sequence register 70.
[0078] The memory interface 20-1 acquires an IO request issued by the IO request issuer 10. The memory interface 20-1 writes a lock address 60 (i.e., a request destination address) and sequence information SEQ_ID to the lock register 22-1 (process (1)). The memory interface 20-1 may write a sequence guarantee flag to the lock register 22-1.
[0079] The memory interface 20-1 processes the IO request, i.e., executes a write command or a read command based on the IO request (process (2)). When the IO request is completed in the memory 3-1, the memory interface 20-1 writes a completion notice (e.g., a flag value of 1) to the corresponding sequence information SEQ_ID in the sequence register 70 referenced by the IO request issuer 10 (process (3)).
[0080] When the IO request issuer 10 receives a completion report of an IO request having sequence information SEQ_ID, it refers to the completion information recording unit 71. The IO request issuer 10 judges whether or not all IO requests of sequence information following the issuance sequence indicated by the sequence information SEQ_ID have been completed. If all IO requests following the issuance sequence indicated by the sequence information SEQ_ID have been completed, the IO request issuer 10 causes all lock registers 22-1 to 22-4 to release the lock on the lock address 60 (process (4)).
[0081] Fig. 13 is an example of a timing chart showing a processing procedure by the information processing device 1 according to the second embodiment. In Fig. 13, an example will be described in which the IO request is a data write request. In Fig. 13, an IO request (sequence information SEQ_ID=0) and an IO request (sequence information SEQ_ID=1) are issued in this issuance order, and the IO request (sequence information SEQ_ID=1) is completed prior to the IO request (sequence information SEQ_ID=0).
[0082] As shown in FIG. 13, the IO request issuer 10 refers to the sequence register 70. Based on the reference result of the sequence register 70, the IO request issuer 10 confirms that SEQ_ID=0 is the current sequence. The IO request issuer 10 issues an IO request with SEQ_ID=0, i.e., a data write request. In the second embodiment, the IO request issuer 10 writes the current SEQ_ID=1 to the current sequence information section 72 of the sequence register 70.
[0083] In the example of Fig. 13, the destination of the IO request with SEQ_ID=0 is the memory interface 20-2 (memory interface #2 in Fig. 13). The memory interface 20-2 executes a memory write command to the memory 3-2 (memory #2 in Fig. 7).
[0084] In parallel with the execution of the memory write command, the memory interface 20-2 writes each piece of information, that is, a lock address 60, an order guarantee flag 61, and order information 62 (SEQ_ID=0), to the lock register 22-2 (lock register #2 in FIG. 7). The lock address 60 is the request destination address.
[0085] The lock register 22-2 locks the destination address (ie, the lock address 60) of an IO request during instruction execution in the memory interface 20-2, and rejects other requests to that destination address.
[0086] Next, the IO request issuer 10 refers to the sequence register 70. Based on the reference result of the sequence register 70, the IO request issuer 10 confirms that SEQ_ID=1 is the current order, i.e., the sequence. After the confirmation, the IO request issuer 10 issues an IO request with SEQ_ID=1. In addition, the IO request issuer 10 writes the current SEQ_ID=2 to the current order information section 72 of the sequence register 70.
[0087] The destination of the IO request with SEQ_ID=1 is the memory interface 20-1 (memory interface #1 in FIG. 13). The memory interface 20-1 executes a memory write command to the memory 3-1 (memory #1 in FIG. 13).
[0088] The memory interface 20-1 may write each piece of information, such as a lock address 60, an order guarantee flag 61, and order information 62 (SEQ_ID=1), to a lock register 22-1 (lock register #1 in FIG. 13) in parallel with the execution of the instruction.
[0089] The lock register 22-1 locks the destination address (ie, the lock address 60) of an IO request during instruction execution in the memory interface 20-1, and rejects other requests to that destination address.
[0090] When the processing of the IO request with SEQ_ID=1, i.e., the command to write data to the memory 3-1, is completed, the memory interface 20-1 notifies the sequence register 70 that the IO request with SEQ_ID=1 is completed. As a result, the completion of the IO request with SEQ_ID=1 is written in the sequence register 70. In one example, a flag value of 1 is written in the location of SEQ_ID=1. The memory interface 20-1 sends a completion notification to the IO request issuer 10.
[0091] Since the IO request with SEQ_ID=0 has not been completed, it is not confirmed that all IO requests with SEQ_ID=1 or earlier have been completed in the sequence register 70. Therefore, the IO request issuing source 10 does not issue a command to release the lock for all the lock registers 22-1 to 22-4.
[0092] Thereafter, when the processing of the IO request with SEQ_ID=0 is completed, the memory interface 20-2 notifies the sequence register 70 that the IO request with SEQ_ID=0 is completed. As a result, the completion of the IO request with SEQ_ID=0 is written in the sequence register 70. In one example, a flag value of 1 is written in the location of SEQ_ID=0. The memory interface 20-2 sends a completion notification to the IO request issuer 10.
[0093] The IO request issuer 10 refers to the sequence register 70. The IO request issuer 10 confirms that the IO requests with SEQ_ID=1 or less have been completed. When the IO request issuer 10 confirms the completion, the IO request issuer 10 notifies a release command to set the order guarantee flag of the lock register to 0 (no order guarantee) for the IO requests with SEQ_ID=1 or less. The command may be a completion notification of the IO requests with SEQ_ID=1 or less. In other words, when the IO request issuer 10 confirms that all requests before the issuance order have been completed, it causes each of the lock registers 22-1 to 22-4 to release the lock for all write or read requests before the issuance order. Each memory interface 20 releases all lock addresses 60 with SEQ_ID=1 or less based on the command.
[0094] As a result, the processor 100 is put into a state in which it can accept other IO requests to the locked address 60. However, IO requests issued before the notification of the release command must not overtake the notification of this release command.
[0095] In the second embodiment, the IO request issuer 10 can confirm that all IO requests have been completed. In this case, the processor 100 can reset the sequence information 62 (SEQ_ID) to an initial value (for example, SEQ_ID=0). This process is similar to the case described in FIG. 8 of the first embodiment. Therefore, repeated description will be omitted.
[0096] [C-2] Example of operation FIG. 14 is a flowchart showing an example of an IO request processing operation by the information processing device 1 according to the second embodiment.
[0097] The processing in steps S31 to S36 in Fig. 14 is basically similar to the processing in steps S21 to S26 in Fig. 9. Therefore, a repeated description of these processing will be omitted.
[0098] When the IO request issuer 10 receives a completion report of an IO request having sequence information SEQ_ID, it refers to the completion information recording unit 71. The IO request issuer 10 checks whether all IO requests of sequence information following the issuance sequence indicated by the sequence information SEQ_ID are completed (step S37). The IO request issuer 10 waits until it is confirmed that all IO requests following the issuance sequence indicated by the sequence information SEQ_ID are completed (see the No route of step S37). When the IO request issuer 10 checks that all IO requests following the issuance sequence indicated by the sequence information SEQ_ID are completed (see the Yes route of step S37), the process proceeds to step S38.
[0099] In step S38, the IO request source 10 notifies all of the IO request destination memory interfaces 20-1 to 20-4 of the SEQ_ID of the IO request whose completion has been confirmed. The notification is a command to all of the lock registers 22-1 to 22-4 to release the lock on the lock address 60.
[0100] Each memory interface 20 unlocks (releases) the lock address 60 based on the notification from the IO request issuer 10 (step S39). In one example, each memory interface 20 sets the order guarantee flag corresponding to the SEQ_ID of the IO request whose completion has been confirmed in each lock register 22 to 0.
[0101] According to the information processing device 1 of the second embodiment, even in a configuration in which the destination addresses of the IO requests are distributed among a plurality of memories 3-1 to 3-4, the throughput of the IO requests can be increased. Furthermore, since the sequence register 21 in each memory interface 20 can be omitted, the control can be simplified.
[0102] [D] Effects of the embodiment According to one example of the above-described first and second embodiments, for example, the following advantageous effects can be achieved.
[0103] The processor 100 includes an IO request source 10 that issues a write or read request. The processor 100 also includes a memory interface 20 that is connected to a plurality of memories 3. The processor 100 includes one or more lock registers 22 that store a memory address of a destination of the IO request and a value indicating the order of the IO request. The processor 100 also includes a sequence register 21 in which completion of an IO request is written for each value indicating the order. The lock address 60 also extends a lock on the memory address of a destination of a request whose order value is a first value based on the sequence register 21 until it is confirmed that all requests whose order values are equal to or less than a first value have been completed.
[0104] As a result, although the IO request is completed, it is possible to prevent the execution of other commands for the corresponding address until the completion notification of all commands below the sequence value (SEQ ID) arrives. Therefore, it is possible to improve the throughput of issuing IO requests, which are read or write requests, while maintaining the order guarantee. In particular, even in a configuration in which a shared buffer that guarantees the order cannot be provided because the memory area is distributed and arranged in multiple memories, it is possible to improve the throughput of issuing IO requests, which are read or write requests.
[0105] Each of the memory interfaces 20 is provided with a sequence register 21 and a lock register 22. When a request whose sequence number is a first value (SEQ_ID_N) is completed, the first value is notified to the memory interfaces 20. Completion of the request whose sequence number is the first value is written to each of the sequence registers 21. The lock register 22 monitors the corresponding second register, and extends the lock on the memory address that is the destination of the request whose sequence number is the first value until it is confirmed that all requests whose sequence number is equal to or less than the first value have been completed.
[0106] This allows the processor 100 to guarantee the order without adding a separate order control for IO requests in the IO request issuer 10 .
[0107] The sequence register 70 is provided in the IO request issuer 10. When a request whose sequence value is the first value is completed, the IO request issuer 10 is notified of the first value, and the completion of the request whose sequence value is the first value is written in the sequence register 70. The IO request issuer 10 monitors the sequence register 70, and when it is confirmed that all requests whose sequence values are equal to or less than the first value are completed, it notifies the lock register 22. Through the notification, the lock register 22 extends the lock on the memory address that is the destination of the request whose sequence value is the first value until it is confirmed that all requests whose sequence values are equal to or less than the first value are completed.
[0108] This makes it possible to omit the sequence register 21 in each memory interface 20, thereby simplifying the control.
[0109] After confirming that all of the already issued IO requests have been completed, the IO request issuer 10 issues a command to one or more sequence registers 21 to initialize a value indicating the order of the IO requests.
[0110] This makes it possible to handle more IO requests than the number of sequences that can be stored in the sequence registers 21 and 70. This makes it possible to avoid erroneous determination of sequence values due to wraparound.
[0111] One anticipated benefit in business situations is that it can contribute to further improvements in computational performance in the field of computing required in various fields such as industry, education, and commercial transactions.
[0112] [E] Notes Regarding the above embodiment, the following supplementary notes are further disclosed.
[0113] (Appendix 1) an issuing unit for issuing a write or read request; A plurality of memory interfaces respectively connected to a plurality of memories; one or more first registers for storing a memory address of a destination of the request and a value indicating an order of the request; a second register in which completion of the request is written for each value indicating the order; Equipped with The processor extends the lock on the memory address that is the destination of the request whose order value is the first value until the first register confirms, based on the second register, that all requests whose order value is less than or equal to the first value have been completed.
[0114] (Appendix 2) the first register and the second register are provided in each of the memory interfaces; when the request having the order number of the first value is completed, the first value is communicated to the plurality of memory interfaces; a first value indicating that the request is completed is written to each of the second registers; 2. The processor of claim 1, wherein the first register monitors the corresponding second register and extends a lock on a memory address that is the destination of a request whose order value is the first value until it determines that all requests whose order value is less than or equal to the first value have been completed.
[0115] (Appendix 3) the second register is provided in the issuing unit, when the request whose sequence number is the first value is completed, the first value is notified to the issuing unit, and the completion of the request whose sequence number is the first value is written to the second register; the issuing unit monitors the second register, and when it confirms that all of the requests whose sequence values are equal to or less than the first value have been completed, notifies the first register; 2. The processor of claim 1, wherein the first register extends a lock on a memory address that is a destination of the request whose order value is the first value until it confirms, through the notification, that all of the requests whose order value is less than or equal to the first value have been completed.
[0116] (Appendix 4) The processor according to any one of appendices 1 to 3, wherein the issuing unit issues an instruction to the second register to initialize a value indicating the order when it has confirmed that all of the requests that have already been issued have been completed.
[0117] (Appendix 5) An information processing device having a plurality of memories and a processor, The processor includes an issuing unit that issues a write or read request; A plurality of memory interfaces respectively connected to a plurality of memories; one or more first registers for storing a memory address of a destination of the request and a value indicating an order of the request; a second register in which completion of the request is written for each value indicating the order; Equipped with The information processing device, wherein the first register extends a lock on a memory address that is the destination of the request whose value indicating the order is the first value until it confirms, based on the second register, that all requests whose value indicating the order is less than or equal to the first value have been completed.
[0118] (Appendix 6) the first register and the second register are provided in each of the memory interfaces; when a request having the sequence number of the first value is completed, the first value is communicated to the plurality of memory interfaces; a first value indicating that the request is completed is written to each of the second registers; The information processing device described in Appendix 5, wherein the first register monitors the corresponding second register and extends a lock on a memory address that is the destination of the request whose order value is the first value until it confirms that all of the requests whose order value is less than or equal to the first value have been completed.
[0119] (Appendix 7) the second register is provided in the issuing unit, when the request whose sequence number is the first value is completed, the first value is notified to the issuing unit, and the completion of the request whose sequence number is the first value is written to the second register; the issuing unit monitors the second register, and when it confirms that all of the requests whose sequence values are equal to or less than the first value have been completed, notifies the first register; The information processing device described in Appendix 5, wherein the first register extends a lock on a memory address that is the destination of the request whose order value is the first value until it confirms through the notification that all of the requests whose order value is less than or equal to the first value have been completed.
[0120] (Appendix 8) The information processing device according to any one of appendices 5 to 7, wherein the issuing unit issues an instruction to the second register to initialize a value indicating the order of the requests when it has confirmed that all of the requests that have already been issued have been completed. [Explanation of symbols]
[0121] 1. Information processing device 3, 3-1~3-4 Memory 10 IO Request Issuer 11, 21, 21-1 to 21-4, 70 Sequence Register 20, 20-1 to 20-4 Memory Interface 22, 22-1 to 22-4 Lock Register 30, 30-1 to 30-4 Cash 31, 31-1a~31-4b CPU cores 61 Lock Address 61 Order Guarantee Flag 62 Order information 71 Completion Information Recording Section 72 Current Queue Information Department
Claims
1. an issuing unit for issuing a write or read request; A plurality of memory interfaces respectively connected to a plurality of memories; one or more first registers for storing a memory address of a destination of the request and a value indicating an order of the request; a second register in which completion of the request is written for each value indicating the order; Equipped with the first register extends a lock on a memory address that is a destination of the request whose order value is the first value until it is confirmed that all requests whose order value is equal to or less than a first value have been completed based on the second register; Processor.
2. The plurality of memory interfaces are each provided with the first register and the second register, when the request having the order number of the first value is completed, the first value is communicated to the plurality of memory interfaces; a first value indicating that the request is completed is written to each of the second registers; the first register monitors the corresponding second register and extends a lock on a memory address that is a destination of the request whose order value is the first value until it is determined that all of the requests whose order value is equal to or less than the first value have been completed; The processor of claim 1 .
3. the second register is provided in the issuing unit, when the request whose order is the first value is completed, the first value is notified to the issuing unit, and the completion of the request whose order is the first value is written to the second register; the issuing unit monitors the second register, and when it confirms that all of the requests whose sequence values are equal to or less than the first value have been completed, notifies the first register; the first register extends a lock on a memory address that is a destination of the request whose order value is the first value until it is confirmed through the notification that all of the requests whose order value is equal to or less than the first value have been completed; The processor of claim 1 .
4. 4. The processor according to claim 1, wherein the issuing unit issues an instruction to the second register to initialize a value indicating the order when the issuing unit has confirmed that all of the requests that have already been issued have been completed.
5. An information processing device having a plurality of memories and a processor, The processor, an issuing unit for issuing a write or read request; a memory interface connected to each of the plurality of memories; one or more first registers for storing a memory address of a destination of the request and a value indicating an order of the request; a second register in which completion of the request is written for each value indicating the order; Equipped with An information processing device, wherein the first register extends a lock on a memory address that is the destination of the request whose order value is the first value until it confirms, based on the second register, that all requests whose order value is less than or equal to the first value have been completed.
Citation Information
Patent Citations
Information processor
JP2004005710A
