Methods for configuring read transaction reordering for multi-level cascaded on-chip networks.
By reallocating transaction identifiers and mapping relationships in a multi-level cascaded on-chip network and configuring reordering buffers only at the source interface unit, the problem of mismatched order-preserving resource configuration in the prior art is solved, achieving efficient response sorting and performance improvement.
Patent Information
- Application Number
- CN202611631264.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-21
- Publication Date
- 2026-11-17
AI Technical Summary
In a multi-level cascaded on-chip network architecture, the existing order-preserving resource configuration is expensive and inefficient, and cannot effectively match the transaction emission characteristics of the master device, resulting in poor orderliness of response returns.
By acquiring the transaction emission characteristics of the overall master device, the target master device is determined, and a unique transaction identifier is reassigned in the interface unit to establish a mapping relationship. Only the source interface unit is configured with a reordering buffer, and the response is rearranged according to the mapping relationship to ensure that the response is returned in the emission order.
It achieves the matching of order-preserving resources with actual needs, reduces ineffective order-preserving overhead, improves the performance and response correctness of on-chip networks, and avoids redundant configuration of hierarchical order-preserving resources.
Smart Images

Figure CN123115052A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of system-on-a-chip (SoC) technology, and for example to a read transaction reordering configuration method for a multi-level cascaded SoC network. Background Technology
[0002] As the scale of AI inference workloads continues to grow, the number of computing engines integrated within a single chip is constantly increasing. To support large-scale engine expansion while controlling interconnect complexity, these chips generally adopt a multi-level cascaded on-chip network architecture of "backbone network + subsystem network": the backbone NoC (Network-on-Chip) serves as the switching hub for the entire chip, while subsystems such as computing engines, processors, and external interfaces each form their own internal networks and are connected to the backbone NoC through boundary interfaces. An end-to-end transaction typically needs to traverse multiple network layers, including the source subsystem NoC, the backbone NoC, and the destination subsystem NoC.
[0003] On-chip interconnects commonly employ the ARM AMBA AXI series protocols. The AXI (Advanced eXtensible Interface) protocol allows responses corresponding to different transaction identifiers to be returned out of order, but requires responses with the same identifier to be returned strictly in the order they were transmitted. Therefore, when a master device concurrently transmits multiple incomplete read transactions with the same identifier, the interconnect network needs to guarantee the order of these responses; conversely, if the master device assigns a unique identifier to each transaction or transmits them strictly serially, the interconnect network does not need to provide additional ordering mechanisms.
[0004] In related technologies, one approach is to set up a reordering buffer on the master device side. However, this requires configuring a buffer at each master device that may face out-of-order responses, resulting in a large total buffer area overhead when there are many master devices within the chip. Another approach is to set up a request parallelizer at the on-chip network entry point, renaming transaction identifiers and using it in conjunction with the reordering buffer. Related research shows that in this type of scheme, the reordering buffer area can account for more than 30% of the total NoC area, and the buffer resource utilization is low. Yet another approach is hop-by-hop end-to-end order preservation: each interface unit at each level of the interconnect network enables the order preservation transmission mechanism, and the order preservation intention issued by the master device is passed level by level. Each level assigns the same identifier to transactions with dependencies and configures corresponding ordering logic and buffer resources. In a multi-level cascaded architecture, this approach requires intermediate and boundary interface units to also configure ordering resources. The area overhead increases linearly with the number of cascaded levels, the order preservation logic introduces additional waiting delays, and order preservation constraints are imposed on traffic that does not originally require order preservation. Therefore, in a multi-level cascaded on-chip network architecture, there is a mismatch between the current order preservation overhead and the actual order preservation requirements: the order preservation resources are either configured one by one for each master device or configured level by level for each network layer, without being configured differently according to the transaction transmission characteristics of the master device, resulting in poor current order preservation performance. Summary of the Invention
[0005] The present invention aims to provide a method, apparatus, electronic device and storage medium for read transaction reordering configuration of multi-level cascaded on-chip networks.
[0006] According to one aspect of the present invention, a read transaction reordering configuration method for a multi-level cascaded on-chip network is proposed, comprising: acquiring the transaction transmission characteristics of the overall master device, and determining the target master device in the overall master device that concurrently transmits multiple incomplete read transactions under the same transaction identifier according to the transaction transmission characteristics; during the forward transmission of read transactions along the on-chip network, using the interface unit receiving the read transaction to reallocate the unique transaction identifier on the link connected to the interface unit, and establishing a mapping relationship between the reallocated transaction identifier and the original transaction identifier of the read transaction, so that the read transaction response can be directly forwarded along a preset response return path according to the mapping relationship; when the read transaction response returns to the source interface unit directly connected to the target master device, using the reordering buffer configured in the source interface unit, and reordering the returned read transaction response according to the mapping relationship, so as to return in the order in which the target master device transmits read transactions.
[0007] According to one aspect of the present invention, a read transaction reordering configuration apparatus for a multi-level cascaded on-chip network of a chip is provided, comprising: The device determination module is used to obtain the transaction emission characteristics of the overall master device, and determine the target master device in the overall master device that concurrently emits multiple incomplete read transactions under the same transaction identifier based on the transaction emission characteristics; The mapping relationship determination module is used to reallocate a unique transaction identifier on the link connected to the interface unit by the interface unit receiving the read transaction during the forward transmission of the read transaction along the on-chip network, and establish a mapping relationship between the reallocated transaction identifier and the original transaction identifier of the read transaction, so that the read transaction response can be directly forwarded along the preset response return path according to the mapping relationship. The reordering module is used to reorder the returned read transaction responses according to the mapping relationship when the read transaction response is returned to the source interface unit directly connected to the target master device, using the reordering buffer configured in the source interface unit. This ensures that the read transactions are returned in the order they were sent by the target master device.
[0008] According to one aspect of the present invention, an electronic device is provided, comprising: a processor; and a memory storing a computer program, which, when executed by the processor, causes the processor to perform the method described above.
[0009] According to one aspect of the present invention, a non-transitory computer-readable medium is provided, on which readable instructions are stored, which, when executed by a processor, cause the processor to perform the method described above.
[0010] It should be understood that the above general description and the following detailed description are merely exemplary and do not limit the invention.
[0011] Beneficial effects: Through the embodiments provided by this invention, reordering buffers are configured only on the source interface units directly connected to the target master device determined according to the transaction transmission characteristics. Intermediate and boundary interface units in the on-chip network, excluding the source interface units, are not configured with reordering buffers. Furthermore, each level of interface unit reassigns a unique transaction identifier and establishes a mapping relationship to allow the response to be directly delivered. Therefore, ordering resources are no longer configured one by one for each master device or layer by layer in the network, but are configured differently according to the actual transaction transmission characteristics of the master devices. The ordering overhead matches the actual ordering requirements. Simultaneously, since there are no reordering resources on the entire cascaded path except for the source interface units, transactions can be delivered out of order on each level of the link. The path performance is not limited by the ordering logic. When the response is returned, it is only rearranged by the source interface units according to the mapping relationship and returned in the transmission order. Thus, the correctness of the transaction response return is guaranteed without relying on hierarchical ordering resources and per-master device buffers. Attached Figure Description
[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without exceeding the scope of protection claimed by the present invention.
[0013] Figure 1 A flowchart illustrating a read transaction reordering configuration method for a multi-level cascaded on-chip network provided in an embodiment of the present invention; Figure 2 A schematic diagram of a multi-level cascaded on-chip network architecture provided in an embodiment of the present invention; Figure 3 A block diagram of a read transaction reordering configuration device for a multi-level cascaded on-chip network provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation
[0014] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, they are provided so that the invention will be thorough and complete, and the concept of the exemplary embodiments will be fully conveyed to those skilled in the art. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted.
[0015] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a full understanding of embodiments of the invention. However, those skilled in the art will recognize that the technical solutions of the invention can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of the invention.
[0016] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.
[0017] The flowcharts shown in the accompanying drawings are only illustrative, and do not necessarily include all contents and operations / steps, nor do they necessarily have to be executed in the described order. For example, some operations / steps may be decomposed, and some operations / steps may be merged or partially merged, so the actual execution order may be changed according to actual situations.
[0018] It should be understood that although the terms first, second, third, etc. may be used herein to describe various components, these components should not be limited by these terms. These terms are only used to distinguish one component from another. Therefore, the first component discussed below may be called the second component without departing from the teachings of the concept of the present invention. As used herein, the term "and / or" includes any one and all combinations of one or more of the associated listed items.
[0019] English abbreviations used in the present invention and their corresponding Chinese translations: English abbreviation: NoC, Chinese translation: 片上网络, full English name: Network-on-Chip; English abbreviation: AXI, Chinese translation: 高级可扩展接口, full English name: Advanced eXtensible Interface; English abbreviation: DMA, Chinese translation: 直接存储器访问, full English name: Direct Memory Access; English abbreviation: SRAM, Chinese translation: 静态随机存取存储器, full English name: Static Random Access Memory; English abbreviation: CPU, Chinese translation: 中央处理器, full English name: Central Processing Unit; English abbreviation: DSP, Chinese translation: 数据信号处理器, full English name: Digital Signal Processor; English abbreviation: ASIC, Chinese translation: 专用集成电路, full English name: Application Specific Integrated Circuit; English abbreviation: FPGA, Chinese translation: 现场可编程门阵列, full English name: Field Programmable Gate Array; English abbreviation: PCI, Chinese translation: 外设部件互连标准, full English name: Peripheral Component Interconnect; English abbreviation: EISA, Chinese translation: 扩展工业标准结构, full English name: Extended Industry Standard Architecture; English abbreviation: ROM, Chinese translation: Read-Only Memory, full English name: Read-Only Memory; English abbreviation: RAM, Chinese translation: Random Access Memory, full English name: Random Access Memory; English abbreviation: EEPROM, Chinese translation: Electrically Erasable Programmable Read-Only Memory, full English name: Electrically Erasable Programmable Read-Only Memory; English abbreviation: CD ROM, Chinese translation: Compact Disc Read-Only Memory, full English name: Compact Disc Read Only Memory.
[0020] For specific implementation manners, reference may be made to the following embodiments.
[0021] Figure 1 is a flowchart of a read transaction reordering configuration method for a multi-stage cascaded Network-on-Chip provided by an embodiment of the present invention. As Figure 1 shown, the method includes Step S10, Step S11 and Step S12.
[0022] In Step S10, the transaction emission characteristics of the overall master device are acquired, and a target master device that concurrently transmits a plurality of outstanding read transactions under the same transaction identifier in the overall master device is determined according to the transaction emission characteristics.
[0023] In the present invention, the overall master device refers to all master devices connected to the multi-stage cascaded Network-on-Chip, including both real master devices such as processors and DMA (Direct Memory Access) engines that directly initiate transactions, and interface units that serve as the outlet of the upper-level network in the cascaded structure and appear as master devices from the perspective of the current-level network. The transaction emission characteristics refer to the behavior characteristics of the master device in allocating transaction identifiers when emitting transactions and whether allowing a plurality of outstanding transactions to exist simultaneously, and such characteristics can be directly obtained from the specification of the master device, or can be obtained by analyzing and counting the historical transaction flows issued by the master device. Concurrently transmitting a plurality of outstanding read transactions under the same transaction identifier can be directly referred to as a transaction, which means that the master device issues two or more transactions in sequence, these transactions carry the same transaction identifier, and the latter transaction is issued before the response of the previous transaction is returned, so that under the AXI protocol, the responses of these transactions with the same identifier must be returned in the emission order. The target master device is a master device that has the above behavior characteristics, so that its responses have an out-of-order risk and require order preservation.
[0024] In some implementations, a classification module can be configured to perform this step. The classification module reads the transaction issuance characteristics of each master device and determines, one by one, whether the master device assigns a unique transaction identifier to each transaction or whether it issues the same transaction identifier serially. For multi-core processors and their storage interfaces within the processor subsystem, they typically issue multiple read transactions concurrently under the same transaction identifier, and are therefore identified as target master devices. For DMA engines within each computing engine subsystem, they typically assign a unique identifier to each transaction, and are therefore identified as non-target master devices. After classification, the results are written to the buffer configuration module as configuration parameters to determine which source interface units to configure reordering buffers for. (See reference) Figure 2 The multi-level cascaded on-chip network architecture shown has a backbone NoC as the switching hub for the entire chip. Subsystems such as computing engines, processors, and external interfaces are each networked internally and connected to the backbone NoC through boundary interfaces. The classification module performs the above judgment on all master devices connected to this architecture.
[0025] In step S11, during the forwarding of the read transaction along the multi-level cascaded on-chip network, the interface unit receiving the read transaction reassigns a unique transaction identifier on the link connected to the interface unit, and establishes a mapping relationship between the reassigned transaction identifier and the original transaction identifier of the read transaction, so that the read transaction response corresponding to the read transaction can be forwarded directly along the preset response return path according to the mapping relationship.
[0026] In this invention, an interface unit refers to a connecting component in a multi-level cascaded network-on-a-chip (hereinafter referred to as network-on-a-chip) located between the master device and the network, between the network and the slave device, and between two cascaded networks. Based on location, it can be divided into source interface units directly connected to the master device, target interface units directly connected to the slave device, boundary interface units located between two cascaded networks, and intermediate interface units other than the above three categories. "Uniqueness on the link connected to the interface unit" means that the reallocated transaction identifier only needs to remain unique on the link segment connected to that interface unit, and does not need to be globally unique across the entire cascaded path; therefore, different links can be numbered independently. The original transaction identifier refers to the transaction identifier carried by the master device when transmitting the transaction, i.e., the transaction identifier from the perspective of the target master device. The mapping relationship refers to the correspondence between the reallocated transaction identifier and the transaction identifier before this reallocation; this correspondence is used to restore the transaction identifier level by level when the response is returned. The "preset response return path" refers to the response return path corresponding to the transaction forward path, composed of interface units at each level.
[0027] In some implementations, an identifier reassignment module can be set up in each target interface unit. When a transaction is forwarded from the current network layer to the next network layer, the boundary interface unit between the current network layer and the next network layer, or the entry interface unit of the next network layer, reassigns a unique transaction identifier for the transaction on the link of the next network layer. When a transaction is forwarded from a multi-level cascaded on-chip network to a slave device, the target interface unit directly connected to the slave device reassigns a unique transaction identifier for the transaction on the link connected to that target interface unit. Each time a reassignment is performed, the identifier reassignment module records the correspondence between the transaction identifier after reassignment and the transaction identifier before reassignment. In this way, multiple transactions that originally carried the same transaction identifier are assigned different transaction identifiers after entering the next level link. Since the AXI protocol allows transactions with different identifiers to be completed out of order, these transactions can be forwarded out of order on each level of the link, maximizing path performance without violating the protocol agreement. Figure 2 In the two-level cascaded structure of "backbone + subsystem" shown, the exit target interface unit of the subsystem NoC and the corresponding entry interface unit of the backbone NoC are mutually bounded, and the exit target interface unit of the backbone NoC and the entry interface unit of the destination subsystem NoC are mutually bounded. Identifier reassignment is performed at these boundaries and at the final target interface unit.
[0028] In step S12, when the read transaction response is returned to the source interface unit directly connected to the target master device, the reordering buffer configured in the source interface unit is used, and the returned read transaction response is rearranged according to the mapping relationship so as to return in the order in which the target master device issued the read transaction; wherein, the multiple overall master devices are devices that initiate read transactions in a multi-level cascaded on-chip network, the source interface unit is the interface unit through which the target master device accesses the multi-level cascaded on-chip network, and the target master device is directly connected to the source interface unit.
[0029] In this invention, the source interface unit refers to the network interface unit directly connected to the target master device, which connects the transactions sent by the target master device to the on-chip network. The reordering buffer is a buffer resource configured within the source interface unit for temporarily storing out-of-order read transaction responses and reordering them. It is implemented using SRAM (Static Random Access Memory), with one buffer entry storing the data of a transaction within a single bus cycle width. Reordering refers to rearranging the multiple read transaction responses stored in the reordering buffer according to the order in which the target master device sends read transactions, so that the response order returned to the target master device is consistent with the sending order. Since out-of-order reordering of the same identifier from the master device's perspective may only occur at the source interface unit throughout the entire cascade path, and all other levels have been assigned different identifiers in step two without requiring reordering resources, the reordering buffer only needs to be set up at the source interface unit.
[0030] In some implementations, along the response return path, each level of interface unit, according to the correspondence recorded in step two, restores the transaction identifier in the read transaction response level by level to the transaction identifier used by the previous level link, ultimately restoring it to the original transaction identifier carried by the target master device when it sent the read transaction. The source interface unit stores multiple read transaction responses with the same original transaction identifier after restoration by each level of interface unit into a reordering buffer, and then returns the read transaction responses stored in the reordering buffer to the target master device in the order in which these read transactions were sent by the target master device, thereby ensuring protocol correctness. For second-type master devices that are not part of the target master device, their source interface unit does not configure a reordering buffer, because these master devices do not have the situation of concurrently sending with the same identifier, their responses do not need to be reordered, and transactions can be sent and returned directly out of order. (Reference) Figure 2 No buffers are configured on the entire path from the master device through the source subsystem NoC, the backbone NoC, the destination subsystem NoC to the slave device. Transactions are sent out of order directly on the path, with a reordering buffer only at the actual source master device.
[0031] Because this invention configures reordering buffers only on the source interface units directly connected to the target master device determined by the transaction transmission characteristics, while intermediate and boundary interface units in the on-chip network, excluding the source interface units, do not configure reordering buffers, and each level of interface unit enables the response to be directly delivered by reallocating unique transaction identifiers and establishing mapping relationships, the ordering resources are no longer configured one by one for each master device or one level by one for each network layer, but are configured differently according to the actual transaction transmission characteristics of the master devices, and the ordering overhead matches the actual ordering requirements. At the same time, since there are no reordering resources on the entire cascaded path except for the source interface units, transactions can be delivered out of order on each level of the link, and the path performance is not limited by the ordering logic. When the response is returned, it is only rearranged by the source interface units according to the mapping relationship and returned in the transmission order, thus ensuring the correctness of the transaction response return without relying on the hierarchical ordering resources and the per-master device buffer.
[0032] According to some embodiments, it can be determined whether the overall master device assigns a unique transaction identifier to read transactions; the master device that does not assign a unique transaction identifier to read transactions and concurrently transmits multiple incomplete read transactions under the same transaction identifier is identified as the target master device; the master device that assigns a unique transaction identifier to read transactions or serially transmits transactions for the same transaction identifier is identified as the non-target master device.
[0033] In this invention, whether a unique transaction identifier is assigned to a read transaction is the specific criterion for distinguishing the type of master device. If the master device assigns a different unique transaction identifier to each transaction, there is no order constraint under the same identifier between different transactions, and the interconnection network does not need to provide an additional order preservation mechanism for it; if the master device does not assign a unique identifier to each transaction, there may be a situation where multiple transactions share the same identifier, and further judgment is needed in combination with whether concurrent transmission occurs.
[0034] In some implementations, the classification module performs three actions sequentially. First, it determines whether the overall master device assigns a unique transaction identifier to read transactions. This determination can be based on the master device's specification or by monitoring whether the identifier field of transactions issued by the master device is repeated over a period of time. Second, master devices that do not assign a unique transaction identifier to read transactions and concurrently issue multiple incomplete read transactions under the same transaction identifier are identified as target master devices. For example, multi-core processors and their storage interfaces within the processor subsystem fall into this category, as they will concurrently issue multiple read transactions with the same identifier. Finally, master devices that assign a unique transaction identifier to read transactions or serially issue transactions for the same transaction identifier are identified as non-target master devices (meeting either of these two conditions is sufficient to identify a non-target master device). For example, DMA engines within each computing engine subsystem assign a unique identifier to each transaction, and configuration interface master devices serially issue transactions for the same identifier; these are all non-target master devices. After classification, the classification result is passed to the buffer configuration module as the configuration basis. Source interface units directly connected to the target master device are configured with reordering buffers, while source interface units directly connected to non-target master devices are not configured with reordering buffers. It should be noted that the above classification rules do not impose any special restrictions on the type of master device. Any third-party master device can be connected after being classified according to the transaction transmission characteristics, which has strong universality.
[0035] This invention uses "whether a unique transaction identifier is assigned to a read transaction" and "whether multiple incomplete read transactions are concurrently issued under the same transaction identifier" as classification criteria. This allows the target master device and non-target master devices to be accurately distinguished based on the master device's own verifiable specification characteristics. Therefore, the configuration object of the order-preserving resource can be accurately locked to the master device that actually has the risk of out-of-order traffic with the same identifier, while the non-target master device does not consume reordering resources at all. This avoids the waste of resources that need to be reserved for buffering regardless of whether the master device actually generates concurrent traffic with the same identifier.
[0036] According to some embodiments, the present invention can also determine the depth of the reordering buffer by dividing the product of the maximum concurrent read transactions and the maximum single read transaction data volume corresponding to the source interface unit by the bus single-cycle data volume; configure the reordering buffer in the source interface unit according to the depth; reserve a corresponding number of buffer entries in the reordering buffer according to the data volume of the read transaction; and suspend receiving read transactions when there are insufficient buffer entries until the corresponding number of buffer entries are released in the reordering buffer.
[0037] In this invention, the maximum concurrent read transaction count refers to the upper limit of the number of read transactions that the source interface unit is allowed to be in an incomplete state at the same time, which is determined by the Outstanding capability (the ability of the master device to continuously initiate multiple new transactions before receiving a response from the slave device for the previous transaction) supported by the interface unit; the maximum single read transaction data size refers to the maximum amount of data that a single read transaction may request; the bus single-cycle data size refers to the single-cycle data transmission width of the bus connected to the interface unit; and the buffer entry refers to the smallest storage unit in the reordering buffer used to store the data of a single transaction within the bus single-cycle width.
[0038] In some implementations, the buffer configuration module can determine the depth of the reordering buffer by dividing the product of the maximum concurrent read transactions and the maximum single read transaction data size corresponding to the source interface unit by the bus single-cycle data size. For example, assuming the maximum concurrent read transactions for each level of interface unit on the path are 16, the maximum single transaction data size is 512 bytes, and the bus single-cycle data size is 128 bytes, then the depth of each reordering buffer is 16 × 512 ÷ 128 = 64 entries, and the buffer capacity is 16 × 512 bytes = 8KB. After configuring the reordering buffer in the source interface unit according to this depth, before receiving a read transaction, a corresponding number of buffer entries are pre-allocated in the reordering buffer based on the data size of the read transaction. For example, a read transaction with a total volume of 1KB requires 2 pre-allocated buffer entries when the buffer width is 512 bytes. If there are insufficient buffer entries, receiving the transaction is paused until the corresponding number of buffer entries are released in the reordering buffer before continuing to receive. This pre-allocation and pause mechanism achieves flow control, preventing buffer overflow. When the buffer depth of the source interface unit is configured too shallow, it only manifests as transaction reception throttling rather than functional error, and has graceful degradation capability, that is, it adopts a gentle control method rather than forced interruption.
[0039] The depth of the reordering buffer in this invention is quantitatively determined by dividing the product of the maximum number of concurrent read transactions and the maximum data volume of a single read transaction by the data volume of a single bus cycle. Therefore, the buffer capacity exactly covers the maximum amount of read data that may occur in transit for the interface unit, so that there is no loss of response due to insufficient depth, nor is there a waste of area due to excessive depth. Furthermore, since the buffer entries are pre-occupied according to the data volume before receiving read transactions and reception is suspended when there are insufficient entries, the buffer will never overflow, and the flow control is performed smoothly in a throttling manner, avoiding forced interruption of the upstream master device.
[0040] According to some embodiments, the multi-level cascaded on-chip network also includes slave devices, which are devices for read transaction access. The slave devices access the multi-level cascaded on-chip network through a target interface unit, and the slave devices are directly connected to the target interface unit. The multi-level cascaded on-chip network includes multiple network layers that are cascaded sequentially. The current network layer and the next network layer are two adjacent levels in the read transaction forward direction among the multiple network layers. This invention can, when a read transaction is forwarded from the current network layer to the next network layer, utilize the boundary interface unit between the current network layer and the next network layer, or utilize the entry interface unit of the next network layer, to reallocate the unique transaction identifier of the read transaction on the link of the next network layer. Specifically, one end of the boundary interface unit is connected to the exit interface unit of the current network layer, and the other end is connected to the entry interface unit of the next network layer, which is the interface unit for receiving read transactions at the next network layer. Furthermore, when a read transaction is forwarded from a multi-level cascaded on-chip network to a slave device, the unique transaction identifier of the read transaction on the link connected to the target interface unit is reallocated using the target interface unit.
[0041] In this invention, the current network layer and the next network layer refer to two adjacent network layers that a transaction traverses sequentially on the cascading path, such as the source subsystem NoC and the backbone NoC, or the backbone NoC and the destination subsystem NoC; the boundary interface unit refers to the interface unit located between two cascading networks that forwards cross-network transactions; the entry interface unit refers to the interface unit in the next network layer that receives transactions from the previous level, and from the perspective of this network layer, it appears as a passing master device; the target interface unit refers to the network interface unit that is directly connected to the slave device.
[0042] In some implementations, the identifier reallocation module performs reallocation operations at two types of locations. First, when a read transaction is forwarded from the current network layer to the next network layer, it reallocates the unique transaction identifier for that transaction on the next network layer link, using either the boundary interface unit between the current and next network layers, or the entry interface unit of the next network layer. (See reference) Figure 2The exit target interface unit of the subsystem NoC and the corresponding entry interface unit of the backbone NoC are mutually bounded, and the exit target interface unit of the backbone NoC and the entry interface unit of the destination subsystem NoC are mutually bounded. Reassignment is performed at these boundaries. Additionally, when a read transaction is forwarded from a multi-level cascaded on-chip network to a slave device, the unique transaction identifier for that transaction on the link connected to the target interface unit can be reassigned using the target interface unit directly connected to the slave device. Based on this, only the source interface unit may experience out-of-order reassignment of the same identifier from the master device's perspective; all other levels do not require any reordering resources because the identifier has already been reassigned. Each target interface unit, when forwarding a transaction to the next level network or slave device, reassigns a unique transaction identifier for the transaction on that level's link. Transactions with different identifiers are allowed to complete out of order under the AXI protocol; therefore, out-of-order forwarding does not violate the protocol.
[0043] This invention explicitly sets the location of the identifier reassignment at the cascade boundary and at the target interface unit between the network and the slave device, so that each link uses a unique transaction identifier within the link segment. Therefore, transactions with the same identifier are split into different identifiers after crossing the boundary, and each level of the link can legally forward transactions out of order. This explains why intermediate interface units and boundary interface units do not need to be configured with reordering buffers, and the reordering resources on the entire cascade path can be concentrated at the source interface unit.
[0044] According to some embodiments, a read transaction uses a unique transaction identifier for the current link corresponding to the current network layer, and a unique transaction identifier for the next link corresponding to the next network layer. The current link and the next link are sequentially connected along the forward direction of the read transaction, and the current link is the next-level link of the next link. This invention can record the correspondence between the transaction identifier after the read transaction is reassigned and the transaction identifier before the reassignment when the transaction identifier is reallocated for the read transaction. During the process of the read transaction response returning along the response return path, the transaction identifier after the reallocation in the read transaction response is restored level by level to the transaction identifier used by the corresponding next-level link according to the correspondence.
[0045] In this invention, the transaction identifier after this reallocation refers to the new identifier used by the transaction on the next level link, and the transaction identifier before this reallocation refers to the old identifier used by the transaction on the previous level link. The corresponding relationship is the pairing record between the new identifier and the old identifier. Step-by-step restoration means that when the response is returned, the identifier in the response is replaced with the identifier used by the previous level link at each level interface unit, and finally restored to the original identifier when the master device sent the transaction.
[0046] In some implementations, when a transaction identifier is reallocated for a read transaction, the identifier reallocation module records the correspondence between the reallocated transaction identifier and the original transaction identifier. This correspondence can be stored in the interface unit performing the reallocation in the form of a mapping table or register pairs. During the read transaction response's return along the response return path, each level of the interface unit restores the reallocated transaction identifier in the read transaction response to the transaction identifier used by the previous level link, based on the recorded correspondence. In a specific implementation, the master device within the processor subsystem issues read transactions R1 and R2 with identifier A. The source interface unit pre-occupies the reordering buffer entries for R1 and R2 before sending the transactions into the processor subsystem NoC. When the transaction enters the backbone NoC via the boundary interface unit, the backbone NoC's exit target interface units R1 and R2 reallocate unique identifiers B1 and B2 respectively. The target interface unit of the destination compute engine subsystem NoC reallocates the identifier again before sending the transaction to the storage controller. If the response data of R2 is returned before that of R1, each level of interface unit directly forwards the response data according to the identifier mapping. Only the source interface unit rearranges the data according to the transmission order corresponding to identifier A and returns it to the master device.
[0047] This invention records the correspondence between the old and new identifiers each time they are redistributed, and restores the identifiers level by level according to the correspondence when the response is returned. Therefore, the response data can be directly forwarded along the response return path without having to look up the order information at each intermediate level. The intermediate interface units and the boundary interface units only need to perform identifier replacement to complete the forwarding. The order judgment required for reordering is postponed to be completed uniformly by the source interface unit, thereby simplifying the processing logic of each level of interface unit while ensuring the traceability of the identifier.
[0048] According to some embodiments, in the source interface unit, multiple read transaction responses with the same original transaction identifier after being restored by each level of interface unit can be stored in a reordering buffer; according to the order in which the corresponding read transactions were sent by the target master device, the read transaction responses stored in the reordering buffer are returned to the target master device in sequence.
[0049] In this invention, multiple read transaction responses with the same original transaction identifier refer to the response data corresponding to multiple read transactions concurrently emitted by the target master device with the same identifier. These responses may be returned out of order on the cascading path, but they have been restored to the same original identifier at the source interface unit level by level. Returning in sequence means returning these read transactions to the target master device one by one in the order in which they were emitted by the target master device.
[0050] In some implementations, after receiving a read transaction response, the source interface unit first restores the transaction identifier in the response to the original transaction identifier when the target master device issued the read transaction, based on the mapping relationship. Then, it stores multiple read transaction responses with the same original transaction identifier after being restored by each level of interface unit into a reordering buffer. Subsequently, the source interface unit returns the read transaction responses stored in the reordering buffer to the target master device in the order in which the corresponding read transactions were issued by the target master device. Thus, even if R2's response data is returned before R1, the source interface unit will return R1's response first, and then R2's response, ensuring that the return order matches the issuance order. For non-target master devices, there is no situation where the same identifier is issued concurrently, so responses do not need to be reordered, and transactions can be sent out of order. Therefore, their source interface units do not configure a reordering buffer.
[0051] The source interface unit of this invention first stores multiple out-of-order read transaction responses with the same identifier into a reordering buffer, and then returns them sequentially according to the order of transmission. Therefore, the response order of transactions with the same identifier is restored from the perspective of the master device, and the requirement of the AXI protocol to return responses with the same identifier in the order of transmission is met. At the same time, since the reordering is only performed at the source interface unit, and the other interface units at all levels directly forward according to the mapping relationship, the reordering logic will not be repeatedly arranged in multi-level networks, and the order-preserving waiting delay only exists within the source interface unit.
[0052] According to some embodiments, the present invention can also disable the order-keeping intent transmission configuration of multi-level interface units in a multi-level cascaded on-chip network; during the process of read transactions being forwarded along the on-chip network, each level of interface unit independently forwards the read transaction and sends back the corresponding read transaction response according to the reallocated transaction identifier.
[0053] In this invention, the order preservation intent transmission configuration refers to the configuration item in the on-chip network interface unit used to pass the order preservation intent of the master device to the next level interface unit level by level, and to require the next level to impose order constraints on transactions with dependencies. In the hop-by-hop end-to-end order preservation scheme, after this configuration is enabled, each level interface unit must allocate the same identifier for transactions with dependencies and configure the corresponding sorting logic and buffer resources. Independent forwarding means that each level interface unit forwards and responds to the transaction only according to the reallocated transaction identifier without receiving the order preservation intent of the previous level.
[0054] In some implementations, the ordering intent transmission configuration of multi-level interface units in a multi-level cascaded on-chip network is disabled. The responsibility for ordering is shared by the reordering buffer in the source interface unit directly connected to the target master device, the mechanism for reallocating unique transaction identifiers, and the inherent sequential return mechanism of responses with the same identifier in each level of interface unit. During the forwarding of a read transaction along the on-chip network, each level of interface unit independently forwards the transaction and returns the corresponding read transaction response according to the reallocated transaction identifier. That is, intermediate and boundary interface units only perform forwarding based on the identifier of their own link segment and do not concern themselves with dependencies between transactions. Dependencies are only resolved by the reordering buffer in the source interface unit at the actual target master device location. For write transactions, the inherent response tracking mechanism of each level of the on-chip network ensures that write responses with the same identifier are returned to the initiator in the order they were transmitted. For write transaction sequences with dependencies, such as writing data first and then writing the completion flag or doorbell register, the initiator transmits the next write transaction only after receiving the response of the previous write transaction. This order is naturally guaranteed by the aforementioned inherent mechanism, requiring no additional ordering logic or buffer configuration. In other implementations, the number of intermediate interface units and boundary interface units increases as the number of cascade levels or subsystems increases. However, since these interface units are not configured with reordering buffers and all order preservation intent transmission configurations are turned off, the number of configuration points for the reordering buffer is only related to the number of target master devices and does not increase with the number of cascade levels or interface units.
[0055] In this invention, the order preservation intent transmission configuration of the multi-level interface units is completely disabled, and dependencies are no longer passed level by level. Each level of interface unit independently reads transactions and sends back responses based on the reassigned identifier. Therefore, the sorting logic, buffer resources, and order preservation waiting delays at each level in the hop-by-hop order preservation scheme are eliminated, and unnecessary constraints are no longer imposed on traffic that does not require order preservation. At the same time, the correctness of the transaction response is jointly guaranteed by the reordering buffer and identifier reassignment mechanism of the source interface unit and the in-order return mechanism of the same identifier response inherent in each level of the network. No additional software intervention is required throughout the entire link, and there is no need to pass the order preservation intent level by level. Thus, low-overhead order preservation is achieved in a structural way in a multi-level cascaded architecture.
[0056] The following describes an embodiment of the apparatus of the present invention, which can be used to perform the embodiment of the method of the present invention. For details not disclosed in the embodiment of the apparatus of the present invention, please refer to the embodiment of the method of the present invention.
[0057] Figure 3 This is a block diagram of a read transaction reordering configuration device for a multi-level cascaded on-chip network provided in an embodiment of the present invention. Figure 3 As shown, the read transaction reordering configuration device 300 for a multi-level cascaded on-chip network includes a device determination module 301, a mapping relationship determination module 302, and a reordering module 303.
[0058] The device determination module 301 is used to obtain the transaction emission characteristics of multiple overall master devices, and determine the target master device among the multiple overall master devices that concurrently emits multiple incomplete read transactions under the same transaction identifier based on the transaction emission characteristics; The mapping relationship determination module 302 is used to reallocate a unique transaction identifier on the link connected to the interface unit by the interface unit receiving the read transaction during the forward delivery of the read transaction along the multi-level cascaded on-chip network, and establish a mapping relationship between the reallocated transaction identifier and the original transaction identifier of the read transaction, so that the read transaction response corresponding to the read transaction can be directly forwarded along the preset response return path according to the mapping relationship. The reordering module 303 is used to reorder the returned read transaction responses according to the mapping relationship when the read transaction response is returned to the source interface unit directly connected to the target master device, using the reordering buffer configured in the source interface unit. This reorders the responses so that they are returned in the order in which the target master device issued the read transactions. Here, the multiple overall master devices are devices that initiate read transactions in a multi-level cascaded on-chip network, and the source interface unit is the interface unit through which the target master device accesses the multi-level cascaded on-chip network. The target master device is directly connected to the source interface unit.
[0059] Optionally, the device determination module 301 is specifically used for: Determine whether each of the multiple master devices in the overall system has assigned a unique transaction identifier for the read transaction; The master device that has not assigned a unique transaction identifier to a read transaction and has concurrently issued multiple incomplete read transactions under the same transaction identifier is identified as the target master device; The master device that assigns a unique transaction identifier to a read transaction or serially transmits transactions for the same transaction identifier is identified as a non-target master device.
[0060] Optionally, the read transaction reordering configuration device 300 for the multi-level cascaded on-chip network of the chip also includes a reordering buffer configuration module 304, used for: The depth of the reordering buffer is determined by multiplying the maximum concurrent read transactions and the maximum single read transaction data volume corresponding to the source interface unit and dividing by the bus single-cycle data volume. Configure reordering buffers in the source interface unit according to depth; Based on the amount of data read in the transaction, reserve a corresponding number of buffer entries in the reordering buffer; If there are insufficient buffer entries, pause read transactions until the corresponding number of buffer entries are released from the reorder buffer.
[0061] Optionally, the multi-level cascaded on-chip network also includes slave devices, which are devices accessing read transactions. The slave devices access the multi-level cascaded on-chip network through a target interface unit, and are directly connected to the target interface unit. The multi-level cascaded on-chip network includes multiple network layers cascaded sequentially, where the current network layer and the next network layer are two adjacent layers in the read transaction forwarding direction. The mapping relationship determination module 302, during the forwarding of the read transaction along the on-chip network, utilizes the interface unit receiving the read transaction to reallocate the unique transaction identifier on the link connected to the interface unit. Specifically, it is used for: When a read transaction is forwarded from the current network layer to the next network layer, the unique transaction identifier of the read transaction on the link of the next network layer is reallocated by using the boundary interface unit between the current network layer and the next network layer, or by using the entry interface unit of the next network layer. One end of the boundary interface unit is connected to the exit interface unit of the current network layer, and the other end of the boundary interface unit is connected to the entry interface unit of the next network layer. The entry interface unit of the next network layer is the interface unit for receiving read transactions in the next network layer. When a read transaction is forwarded to a slave device from a multi-level cascaded on-chip network, the unique transaction identifier of the read transaction on the link connected to the target interface unit is reallocated using the target interface unit.
[0062] Optionally, the read transaction uses a unique transaction identifier for the current link corresponding to the current network layer, and a unique transaction identifier for the next link corresponding to the next network layer. The current link and the next link are sequentially connected along the forward direction of the read transaction, and the current link is the parent link of the next link. The mapping relationship determination module 302, in establishing the mapping relationship between the reallocated transaction identifier and the original transaction identifier of the read transaction, is specifically used for: When a transaction identifier is reassigned for a read transaction, the correspondence between the transaction identifier after the reassignment and the transaction identifier before the reassignment is recorded. During the process of returning the read transaction response along the response return path, the transaction identifier in the read transaction response after this reallocation is restored level by level to the transaction identifier used by the corresponding previous level link, according to the correspondence.
[0063] Optionally, the rearrangement module 303 is specifically used for: In the source interface unit, multiple read transaction responses with the same original transaction identifier after being restored by each level of interface unit are stored in the reordering buffer; According to the order in which the corresponding read transactions were sent by the target master device, the read transaction responses stored in the reordering buffer are returned to the target master device in sequence.
[0064] Optionally, the read transaction reordering configuration device 300 for the multi-level cascaded on-chip network of the chip also includes a configuration transfer module 305, used for: Disable the order-preserving intent transmission configuration of multi-level interface units in a multi-level cascaded on-chip network; During the forwarding of a read transaction along the on-chip network, each level of interface unit independently forwards the read transaction and sends back the corresponding read transaction response according to the reallocated transaction identifier.
[0065] The device performs functions similar to those described above; other functions are described in the preceding descriptions and will not be repeated here.
[0066] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention, such as... Figure 4 As shown, the electronic device 400 of this embodiment may include a memory 401 and a processor 402.
[0067] The memory 401 stores a computer program, which, when executed by the processor 402, causes the processor 402 to perform the method described in the above embodiment.
[0068] The processor 402 and the memory 401 are connected, for example, via a bus.
[0069] Optionally, the electronic device 400 may also include a transceiver. It should be noted that in practical applications, the transceiver is not limited to one, and the structure of the electronic device 400 does not constitute a limitation on the embodiments of the present invention.
[0070] Processor 402 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 402 may also be a combination that implements computational functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.
[0071] A bus can include a pathway for transmitting information between the aforementioned components. The bus can be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, only one thick line is used in the diagram, but this does not imply that there is only one bus or one type of bus.
[0072] The memory 401 can be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or it can be EEPROM (Electrically Erasable Programmable Read Only Memory), CD. ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed discs, laser discs, optical discs, digital universal discs, Blu-ray discs, etc.), disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.
[0073] The memory 401 stores application code that executes the present invention, and its execution is controlled by the processor 402. The processor 402 executes the application code stored in the memory 401 to implement the content shown in the foregoing method embodiments.
[0074] Electronic devices include, but are not limited to: mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (such as in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Servers can also be included. Figure 4 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.
[0075] The electronic device in this embodiment can be used to execute the method of any of the above embodiments, and its implementation principle and technical effect are similar, so they will not be described again here.
[0076] The present invention also provides a non-transitory computer-readable storage medium having stored computer-readable instructions thereon, which, when executed by a processor, cause the processor to perform the method as described in the above embodiments.
[0077] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a non-transitory computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.
[0078] The embodiments of the present invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, any changes or modifications made by those skilled in the art based on the ideas of the present invention, its specific implementation methods, and its application scope, are all within the scope of protection of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.
Claims
1. A method for read transaction reordering configuration of a multi-level cascaded on-chip network, characterized in that, include: Obtain the transaction launch characteristics of multiple master devices, and determine the target master device among the multiple master devices that concurrently launches multiple incomplete read transactions under the same transaction identifier based on the transaction launch characteristics; During the forwarding of the read transaction along the multi-level cascaded on-chip network, the interface unit receiving the read transaction is used to reallocate a unique transaction identifier on the link connected to the interface unit, and a mapping relationship is established between the reallocated transaction identifier and the original transaction identifier of the read transaction, so that the read transaction response corresponding to the read transaction can be directly forwarded along the preset response return path according to the mapping relationship. When the read transaction response returns to the source interface unit directly connected to the target master device, the reordering buffer configured in the source interface unit is used, and the returned read transaction response is rearranged according to the mapping relationship, so as to return in the order in which the target master device issued the read transaction; wherein, the plurality of overall master devices are devices that initiate the read transaction in the multi-level cascaded on-chip network, the source interface unit is the interface unit through which the target master device accesses the multi-level cascaded on-chip network, and the target master device is directly connected to the source interface unit.
2. The method according to claim 1, characterized in that, The step of acquiring transaction launch characteristics of multiple overall master devices and determining the target master device among the multiple overall master devices that concurrently launches multiple incomplete read transactions under the same transaction identifier based on the transaction launch characteristics includes: Determine whether each of the plurality of overall master devices has assigned a unique transaction identifier to the read transaction; The master device that has not assigned a unique transaction identifier to the read transaction and concurrently sends out multiple incomplete read transactions under the same transaction identifier is identified as the target master device; The master device that assigns a unique transaction identifier to the read transaction or serially transmits transactions for the same transaction identifier is identified as a non-target master device.
3. The method according to claim 1, characterized in that, Also includes: The depth of the reordering buffer is determined by dividing the product of the maximum concurrent read transactions and the maximum single read transaction data volume corresponding to the source interface unit by the bus single-cycle data volume. Configure the reordering buffer in the source interface unit according to the stated depth; Based on the amount of data in the read transaction, a corresponding number of buffer entries are reserved in the reordering buffer. If the buffer entries are insufficient, the reception of read transactions is suspended until the corresponding number of buffer entries are released from the reordering buffer.
4. The method according to claim 1, characterized in that, The multi-level cascaded on-chip network also connects to slave devices, which are the devices accessed by the read transactions. The slave devices are connected to the multi-level cascaded on-chip network through a target interface unit, and the slave devices are directly connected to the target interface unit. The multi-level cascaded on-chip network includes multiple network layers that are cascaded sequentially. The current network layer and the next network layer are two adjacent network layers in the read transaction forward direction. Wherein, during the process of the read transaction being forwarded along the multi-level cascaded on-chip network, the method of reallocating a unique transaction identifier on the link connected to the interface unit by using the interface unit receiving the read transaction includes: When a read transaction is forwarded from the current network layer to the next network layer, the unique transaction identifier of the read transaction on the link of the next network layer is reallocated using the boundary interface unit between the current network layer and the next network layer, or using the entry interface unit of the next network layer. Wherein, one end of the boundary interface unit is connected to the exit interface unit of the current network layer, and the other end of the boundary interface unit is connected to the entry interface unit of the next network layer. The entry interface unit of the next network layer is the interface unit of the next network layer that receives the read transaction. When the read transaction is forwarded to the slave device from the multi-level cascaded on-chip network, the unique transaction identifier of the read transaction on the link connected to the target interface unit is reallocated using the target interface unit.
5. The method according to claim 4, characterized in that, The read transaction uses the unique transaction identifier of the current link in the current link corresponding to the current network layer, and uses the unique transaction identifier of the next link in the next link corresponding to the next network layer. The current link and the next link are sequentially connected in series along the forward direction of the read transaction, and the current link is the link above the next link. The step of establishing a mapping relationship between the reallocated transaction identifier and the original transaction identifier of the read transaction includes: When a transaction identifier is reassigned for a read transaction, the correspondence between the transaction identifier after the reassignment and the transaction identifier before the reassignment is recorded. During the process of the read transaction response returning along the response return path, according to the correspondence, the transaction identifier in the read transaction response after this reallocation is restored level by level to the transaction identifier used by the corresponding previous link.
6. The method according to claim 1, characterized in that, When the read transaction response returns to the source interface unit directly connected to the target master device, the method of reordering the returned read transaction response according to the mapping relationship, utilizing the reordering buffer configured in the source interface unit, includes: In the source interface unit, multiple read transaction responses with the same original transaction identifier after being restored by multiple interface units are stored in the reordering buffer. According to the order in which the corresponding read transactions were sent by the target master device, the read transaction responses stored in the reordering buffer are returned to the target master device in sequence.
7. The method according to claim 1, characterized in that, Also includes: Disable the order-preserving intent transmission configuration of the multi-level interface units in the multi-level cascaded on-chip network; During the forwarding of the read transaction along the on-chip network, the multi-level interface unit independently forwards the read transaction and returns the corresponding read transaction response according to the reallocated transaction identifier.
8. A read transaction reordering configuration device for a multi-level cascaded on-chip network of a chip, characterized in that, include: The device determination module is used to acquire the transaction emission characteristics of multiple overall master devices, and determine the target master device among the multiple overall master devices that concurrently emits multiple incomplete read transactions under the same transaction identifier based on the transaction emission characteristics; The mapping relationship determination module is used to, during the forwarding of the read transaction along the multi-level cascaded on-chip network, reallocate a unique transaction identifier on the link connected to the interface unit that receives the read transaction, and establish a mapping relationship between the reallocated transaction identifier and the original transaction identifier of the read transaction, so that the read transaction response corresponding to the read transaction can be directly forwarded along a preset response return path according to the mapping relationship. The reordering module is used to, when the read transaction response returns to the source interface unit directly connected to the target master device, utilize the reordering buffer configured in the source interface unit and reorder the returned read transaction response according to the mapping relationship, so as to return it in the order in which the target master device issued the read transaction; wherein, the plurality of overall master devices are devices that initiate the read transaction in the multi-level cascaded on-chip network, the source interface unit is the interface unit through which the target master device accesses the multi-level cascaded on-chip network, and the target master device is directly connected to the source interface unit.
9. An electronic device, characterized in that, include: processor; A memory storing a computer program that, when executed by the processor, causes the processor to perform the method as described in any one of claims 1-7.
10. A non-transitory computer-readable storage medium, characterized in that, It stores computer-readable instructions that, when executed by a processor, cause the processor to perform the method as described in any one of claims 1-7.
