NoC multi-stage out-of-order transmission packet processing method and device based on split prediction

Abstract

This invention relates to the field of system-on-a-chip (SoC) technology and discloses a method and apparatus for processing NoC (NoC) multi-level out-of-order transmission packets based on splitting prediction. The method includes: identifying a target slave device based on first transmission packet parameters of the host; predicting a downstream splitting strategy and a downstream caching strategy for the first transmission packet based on the current transmission packet receiving capability parameters of the determined target slave device, to split the first transmission packet into at least one level of second transmission packets matching the transmission packet receiving capability parameters; instructing the target slave device to determine the target cache space for each second transmission packet, and performing a cache rearrangement and merging operation on all second transmission packets based on all splitting level identifiers and all sequence identifiers, wherein each second transmission packet has a corresponding splitting level identifier and sequence identifier; and generating a target output instruction based on the downstream splitting strategy and the downstream caching strategy. Therefore, implementing this invention can improve the efficiency of NoC multi-level out-of-order transmission packet processing.

Description

Technical Field

[0001] This invention relates to the field of system-on-a-chip technology, and in particular to a method and apparatus for processing NoC multi-level out-of-order transmission packets based on split prediction. Background Technology

[0002] With the continuous improvement of system-on-chip (SoC) integration, network-on-chip (NoC) has become the mainstream solution for communication between core components in complex SoCs due to its high bandwidth, low latency and scalability.

[0003] However, in heterogeneous computing, AI acceleration, and other scenarios, the burst transmission capabilities between master devices (such as CPUs and GPUs) and slave devices (such as DDR controllers and peripherals) differ significantly. This necessitates that the original transmission transactions be split into multi-level sub-transmission packets in the NoC to accommodate the processing capabilities of the slave devices. This process involves first-level splitting by the host interface unit (NI) and second-level splitting by the slave interface unit. However, the split sub-transmission packets may be returned out of order due to factors such as routing strategies and multi-channel architectures, requiring merging and reordering to restore the original data order. Traditional solutions require deploying reordering modules (such as ROUs) on both the master and slave sides. For example, the slave-side reordering module needs to reserve cache space matching the DDR command queue (e.g., 8KB cache is required when each transmission is 64B). In multi-slave scenarios, this not only increases resource consumption exponentially, but such modules also need to additionally store the split sub-packet data and order information, leading to a significant increase in hardware overhead and easily increasing transmission latency, directly restricting system performance.

[0004] Therefore, improving the efficiency of NoC multi-level out-of-order packet processing is particularly important. Summary of the Invention

[0005] This invention provides a method and apparatus for processing NoC multi-level out-of-order transmission packets based on split prediction, which can improve the processing efficiency of NoC multi-level out-of-order transmission packets.

[0006] To address the aforementioned technical problems, the first aspect of this invention discloses a NoC multi-level out-of-order packet processing method based on split prediction, the method comprising:

[0007] Based on the first transmission packet parameters of the host, the target slave is identified, and the target slave is used to receive the first transmission packet corresponding to the first transmission packet parameters;

[0008] Determine the current packet reception capability parameters of the target slave device;

[0009] Based on the transmission packet receiving capability parameters, a downstream splitting strategy and a downstream caching strategy for the first transmission packet are predicted. The downstream splitting strategy is used to split the first transmission packet into at least one level of second transmission packets that match the transmission packet receiving capability parameters. Each second transmission packet has a corresponding splitting level identifier and sequence identifier. The downstream caching strategy is used to instruct the target slave to determine the target cache space for each second transmission packet and to perform a cache rearrangement and merging operation on all second transmission packets based on all the splitting level identifiers and all the sequence identifiers.

[0010] Based on the downstream splitting strategy and the downstream caching strategy, a target output instruction is generated. The target output instruction is used to instruct the target slave to output all the second transmission packets based on all the splitting level identifiers and all the sequence identifiers, and to release the target cache space corresponding to the downstream caching strategy.

[0011] As an optional implementation, in the first aspect of the present invention, the transmission packet receiving capability parameters include at least one of address alignment parameters, burst transmission data volume parameters, transmission type parameters, buffer capacity parameters, multi-channel configuration parameters, transmission priority parameters, response delay parameters, and transmission protocol parameters; the step of predicting the downstream splitting strategy of the first transmission packet based on the transmission packet receiving capability parameters includes:

[0012] Based on the transmission packet receiving capability parameters, calculate the downstream splitting level and downstream splitting granularity of the first transmission packet;

[0013] Based on the NoC routing policy between the host and the target slave, predict the downstream transmission path of the first transmission packet;

[0014] Based on the downstream splitting level, the downstream splitting granularity, and the downstream transmission path, predict the downstream splitting strategy for the first transmission packet.

[0015] As an optional implementation, in the first aspect of the present invention, calculating the downstream splitting level and downstream splitting granularity of the first transmission packet based on the transmission packet receiving capability parameters includes:

[0016] Based on the transmission packet receiving capability parameters, analyze the current first transmission timeliness parameters of the target slave device;

[0017] The second transmission timeliness parameter of the host is determined based on the target demand timeliness parameter of the first transmission packet and the first transmission timeliness parameter;

[0018] Based on the first transmission timeliness parameter and the second transmission timeliness parameter, calculate the downstream splitting level and downstream splitting granularity of the first transmission packet.

[0019] As an optional implementation, in the first aspect of the present invention, the step of analyzing the current first transmission timeliness parameter of the target slave device based on the transmission packet receiving capability parameter includes:

[0020] Based on the transmission packet receiving capability parameters, the splitting level capability parameters of the target slave device are analyzed to determine the first available splitting level parameters and the first splitting granularity parameters of the target slave device. The splitting level capability parameters include at least one of the following: splitting logic configuration parameters, hardware processing depth limit parameters, and protocol constraint parameters.

[0021] Based on the first available splitting level parameter and the first splitting granularity parameter, analyze the current first transmission timeliness parameter of the target slave device;

[0022] And, the step of calculating the downstream splitting level and downstream splitting granularity of the first transmission packet based on the first transmission timeliness parameter and the second transmission timeliness parameter includes:

[0023] Obtain the splitting strategy configuration parameters of the host, wherein the splitting strategy configuration parameters include at least one of the following: second available splitting level parameter, second splitting granularity parameter, and splitting priority parameter;

[0024] Based on the configuration parameters of the splitting strategy, analyze the current third transmission timeliness parameters of the host;

[0025] Calculate the target distance value between the second transmission timeliness parameter and the third transmission timeliness parameter;

[0026] Based on the target distance value, the first available splitting level parameter, the first splitting granularity parameter, and the splitting strategy configuration parameter, generate collaborative splitting matching parameters between the host and the target slave.

[0027] Based on the collaborative splitting matching parameters, the downstream splitting level and downstream splitting granularity of the first transmission packet are calculated. The downstream splitting level includes at least one splitting level, and the downstream splitting granularity includes the splitting granularity corresponding to each splitting level.

[0028] As an optional implementation, in a first aspect of the present invention, predicting the downstream caching strategy of the first transmission packet based on the transmission packet reception capability parameters includes:

[0029] Based on the transmission packet receiving capability parameters, determine whether the target slave device has a reserve buffer space required to satisfy the downstream splitting strategy;

[0030] When it is determined that the target slave device has the required reserve cache space to satisfy the downstream splitting strategy, the downstream caching strategy of the first transmission packet is predicted based on the reserve cache space.

[0031] When it is determined that the target slave does not have the required reserve cache space to satisfy the downstream splitting strategy, an upstream backpressure instruction for the target slave is generated according to the target priority parameter of the first transmission packet to adjust the current cache space of the target slave and re-trigger the operation of determining whether the target slave has the required reserve cache space to satisfy the downstream splitting strategy.

[0032] As an optional implementation, in the first aspect of the present invention, generating the target output instruction according to the downstream splitting strategy and the downstream caching strategy includes:

[0033] Based on the downstream splitting strategy and the downstream caching strategy, it is determined whether all the second transmission packets have completed the cache rearrangement and merging operation on the target slave.

[0034] When it is determined that all the second transmission packets have completed the buffer rearrangement and merging operation at the target slave, a target output instruction is generated based on all the split level identifiers and all the sequence identifiers.

[0035] And, before determining whether all the second transmission packets are present at the target slave device before the target slave device completes the cache rearrangement and merging operation, based on the downstream splitting strategy and the downstream caching strategy, the method further includes:

[0036] The cache address of each second transmission packet is determined based on a multi-level pointer linked list;

[0037] For each split-level identifier, based on the cache address and sequence identifier of all the second transmission packets corresponding to the split-level identifier, it is determined whether all the second transmission packets corresponding to the split-level identifier have been completely transmitted to the target slave and whether all the sequence identifiers match the preset sequence identifier;

[0038] When it is determined that all the second transmission packets corresponding to all split level identifiers have been completely transmitted to the target slave and all the sequence identifiers match the preset sequence identifiers, the operation of determining whether all the second transmission packets have completed the cache rearrangement and merging operation on the target slave is triggered according to the downstream splitting strategy and the downstream caching strategy.

[0039] A second aspect of this invention discloses a NoC multi-level out-of-order packet processing apparatus based on split prediction, the apparatus comprising:

[0040] The identification module is used to identify the target slave device based on the first transmission packet parameters of the host, wherein the target slave device is used to receive the first transmission packet corresponding to the first transmission packet parameters;

[0041] The determination module is used to determine the current transmission packet receiving capability parameters of the target slave device;

[0042] The prediction module is used to predict the downstream splitting strategy and the downstream caching strategy of the first transmission packet based on the transmission packet receiving capability parameters. The downstream splitting strategy is used to split the first transmission packet into at least one level of second transmission packets that match the transmission packet receiving capability parameters. Each second transmission packet has a corresponding splitting level identifier and sequence identifier. The downstream caching strategy is used to instruct the target slave to determine the target cache space for each second transmission packet and to perform a cache rearrangement and merging operation on all second transmission packets based on all the splitting level identifiers and all the sequence identifiers.

[0043] The generation module is used to generate a target output instruction based on the downstream splitting strategy and the downstream caching strategy. The target output instruction is used to instruct the target slave to output all the second transmission packets based on all the splitting level identifiers and all the sequence identifiers, and to release the target cache space corresponding to the downstream caching strategy.

[0044] As an optional implementation, in the second aspect of the present invention, the transmission packet receiving capability parameters include at least one of address alignment parameters, burst transmission data volume parameters, transmission type parameters, buffer capacity parameters, multi-channel configuration parameters, transmission priority parameters, response delay parameters, and transmission protocol parameters; the specific method by which the prediction module predicts the downstream splitting strategy of the first transmission packet based on the transmission packet receiving capability parameters includes:

[0045] Based on the transmission packet receiving capability parameters, calculate the downstream splitting level and downstream splitting granularity of the first transmission packet;

[0046] Based on the NoC routing policy between the host and the target slave, predict the downstream transmission path of the first transmission packet;

[0047] Based on the downstream splitting level, the downstream splitting granularity, and the downstream transmission path, predict the downstream splitting strategy for the first transmission packet.

[0048] As an optional implementation, in a second aspect of the present invention, the specific method by which the prediction module calculates the downstream splitting level and downstream splitting granularity of the first transmission packet based on the transmission packet receiving capability parameters includes:

[0049] Based on the transmission packet receiving capability parameters, analyze the current first transmission timeliness parameters of the target slave device;

[0050] The second transmission timeliness parameter of the host is determined based on the target demand timeliness parameter of the first transmission packet and the first transmission timeliness parameter;

[0051] Based on the first transmission timeliness parameter and the second transmission timeliness parameter, calculate the downstream splitting level and downstream splitting granularity of the first transmission packet.

[0052] As an optional implementation, in a second aspect of the present invention, the specific method by which the prediction module analyzes the current first transmission timeliness parameter of the target slave device based on the transmission packet reception capability parameter includes:

[0053] Based on the transmission packet receiving capability parameters, the splitting level capability parameters of the target slave device are analyzed to determine the first available splitting level parameters and the first splitting granularity parameters of the target slave device. The splitting level capability parameters include at least one of the following: splitting logic configuration parameters, hardware processing depth limit parameters, and protocol constraint parameters.

[0054] Based on the first available splitting level parameter and the first splitting granularity parameter, analyze the current first transmission timeliness parameter of the target slave device;

[0055] Furthermore, the specific method by which the prediction module calculates the downstream splitting level and downstream splitting granularity of the first transmission packet based on the first transmission timeliness parameter and the second transmission timeliness parameter includes:

[0056] Obtain the splitting strategy configuration parameters of the host, wherein the splitting strategy configuration parameters include at least one of the following: second available splitting level parameter, second splitting granularity parameter, and splitting priority parameter;

[0057] Based on the configuration parameters of the splitting strategy, analyze the current third transmission timeliness parameters of the host;

[0058] Calculate the target distance value between the second transmission timeliness parameter and the third transmission timeliness parameter;

[0059] Based on the target distance value, the first available splitting level parameter, the first splitting granularity parameter, and the splitting strategy configuration parameter, generate collaborative splitting matching parameters between the host and the target slave.

[0060] Based on the collaborative splitting matching parameters, the downstream splitting level and downstream splitting granularity of the first transmission packet are calculated. The downstream splitting level includes at least one splitting level, and the downstream splitting granularity includes the splitting granularity corresponding to each splitting level.

[0061] As an optional implementation, in a second aspect of the present invention, the specific method by which the prediction module predicts the downstream caching strategy of the first transmission packet based on the transmission packet receiving capability parameters includes:

[0062] Based on the transmission packet receiving capability parameters, determine whether the target slave device has a reserve buffer space required to satisfy the downstream splitting strategy;

[0063] When it is determined that the target slave device has the required reserve cache space to satisfy the downstream splitting strategy, the downstream caching strategy of the first transmission packet is predicted based on the reserve cache space.

[0064] When it is determined that the target slave does not have the required reserve cache space to satisfy the downstream splitting strategy, an upstream backpressure instruction for the target slave is generated according to the target priority parameter of the first transmission packet to adjust the current cache space of the target slave and re-trigger the operation of determining whether the target slave has the required reserve cache space to satisfy the downstream splitting strategy.

[0065] As an optional implementation, in the second aspect of the present invention, the specific method by which the generation module generates the target output instruction according to the downstream splitting strategy and the downstream caching strategy includes:

[0066] Based on the downstream splitting strategy and the downstream caching strategy, it is determined whether all the second transmission packets have completed the cache rearrangement and merging operation on the target slave.

[0067] When it is determined that all the second transmission packets have completed the buffer rearrangement and merging operation at the target slave, a target output instruction is generated based on all the split level identifiers and all the sequence identifiers.

[0068] Furthermore, the determining module is also configured to determine the cache address of each second transmission packet based on a multi-level pointer linked list before the generating module determines whether all second transmission packets are present in the target slave device before the target slave device completes the cache rearrangement and merging operation, according to the downstream splitting strategy and the downstream caching strategy.

[0069] The device also includes:

[0070] The judgment module is used to determine, for each split level identifier, whether all the second transmission packets corresponding to the split level identifier have been completely transmitted to the target slave and whether all the sequence identifiers match the preset sequence identifier, based on the cache address and sequence identifier of all the second transmission packets corresponding to the split level identifier; when it is determined that all the second transmission packets corresponding to all split level identifiers have been completely transmitted to the target slave and all the sequence identifiers match the preset sequence identifier, the generation module is triggered to perform the operation of determining whether all the second transmission packets have completed the cache rearrangement and merging operation on the target slave based on the downstream splitting strategy and the downstream caching strategy.

[0071] The third aspect of the present invention discloses an electronic unit, wherein the electronic unit is integrated in a host and is used to execute the NoC multi-level out-of-order transmission packet processing method based on split prediction disclosed in the first aspect of the present invention.

[0072] A fourth aspect of the present invention discloses another NoC multi-level out-of-order packet processing apparatus based on split prediction, the apparatus comprising:

[0073] Memory containing executable program code;

[0074] A processor coupled to the memory;

[0075] The processor calls the executable program code stored in the memory to execute the NoC multi-level out-of-order transmission packet processing method based on split prediction disclosed in the first aspect of the present invention.

[0076] The fifth aspect of the present invention discloses a computer storage medium storing computer instructions, which, when invoked, are used to execute the NoC multi-level out-of-order transmission packet processing method based on split prediction disclosed in the first aspect of the present invention.

[0077] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0078] In this embodiment of the invention, a target slave device is identified based on the first transmission packet parameters of the host. The target slave device is used to receive the first transmission packet corresponding to the first transmission packet parameters. The current transmission packet receiving capability parameters of the target slave device are determined. Based on the transmission packet receiving capability parameters, a downstream splitting strategy and a downstream caching strategy for the first transmission packet are predicted. The downstream splitting strategy is used to split the first transmission packet into at least one level of second transmission packets that match the transmission packet receiving capability parameters. Each second transmission packet has a corresponding splitting level identifier and sequence identifier. The downstream caching strategy is used to instruct the target slave device to determine the target cache space for each second transmission packet and to perform a cache rearrangement and merging operation on all second transmission packets based on all splitting level identifiers and all sequence identifiers. Based on the downstream splitting strategy and the downstream caching strategy, a target output instruction is generated. The target output instruction is used to instruct the target slave device to output all second transmission packets based on all splitting level identifiers and all sequence identifiers and to release the target cache space corresponding to the downstream caching strategy. As can be seen, implementing this invention can identify the target slave device to which the corresponding first transmission packet needs to be transmitted based on the parameters of the first transmission packet. Based on the current transmission packet receiving capability parameters of the determined target slave device, it can predict the downstream splitting strategy and downstream caching strategy of the first transmission packet. This enables unified prediction of multi-level transmission packet splitting strategies and unified management of caching strategies under multi-level transmission packet splitting between the host (master device, such as CPU, GPU, etc.) and the slave device (slave device, such as DDR controller, peripherals). It eliminates the dependence on the reordering module (such as ROU) on the slave side, reduces the hardware resource overhead and data transmission latency of NoC network from the hardware dimension, further removes the constraint of reordering processing on slave performance, ensures system communication efficiency in complex scenarios, eliminates resource redundancy in hierarchical architecture, and dynamically reserves cache space based on the receiving capability parameters of the slave device. To avoid the waste of static allocation based on maximum burst volume in traditional solutions, this approach improves resource utilization and achieves precise dynamic cache allocation. By unifying the management of split-level identifiers and sequence identifiers, it ensures that sub-packets after multi-level splitting can still be merged in their original order after out-of-order return, reducing the overhead of data retransmission or verification and effectively guaranteeing end-to-end order. By eliminating the dependence on backpressure on the slave side (such as the DDR command queue being shared by the host-side module), it can still guarantee full-pressure access of DDR even in complex scenarios (such as multi-channel high load), increasing the system bandwidth limit and achieving system performance decoupling. Through the combination of hardware and software, it comprehensively improves the processing efficiency of NoC multi-level out-of-order transmission packets. Compared with existing technical solutions, it can significantly reduce the hardware resource overhead and data transmission latency required for merging and rearranging, and further remove the constraint of rearrangement processing on slave performance, ensuring system communication efficiency in complex scenarios. Attached Figure Description

[0079] 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 creative effort.

[0080] Figure 1 This is a flowchart illustrating a NoC multi-level out-of-order transmission packet processing method based on split prediction disclosed in an embodiment of the present invention;

[0081] Figure 2 This is a schematic diagram of a raw segmented processing transmission packet rearrangement and merging process disclosed in an embodiment of the present invention;

[0082] Figure 3 This is a schematic diagram of an improved segmented processing transmission packet rearrangement and merging process disclosed in an embodiment of the present invention;

[0083] Figure 4 This is a schematic diagram of a NoC multi-level out-of-order transmission packet processing architecture based on split prediction disclosed in an embodiment of the present invention;

[0084] Figure 5 This is a flowchart illustrating another NoC multi-level out-of-order packet processing method based on split prediction disclosed in an embodiment of the present invention.

[0085] Figure 6 This is a schematic diagram of the structure of a NoC multi-level out-of-order transmission packet processing device based on split prediction disclosed in an embodiment of the present invention;

[0086] Figure 7 This is a schematic diagram of another NoC multi-level out-of-order transmission packet processing device based on split prediction disclosed in an embodiment of the present invention;

[0087] Figure 8 This is a schematic diagram of another NoC multi-level out-of-order transmission packet processing device based on split prediction disclosed in an embodiment of the present invention. Detailed Implementation

[0088] To enable those skilled in the art to better understand the present invention, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0089] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or end that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or ends.

[0090] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0091] This invention discloses a method and apparatus for processing multi-level out-of-order transmission packets in NoC based on split prediction. It can identify the target slave device to which the first transmission packet needs to be transmitted based on the parameters of the first transmission packet. Based on the determined target slave device's current transmission packet receiving capability parameters, it predicts the downstream splitting strategy and downstream caching strategy for the first transmission packet. This achieves unified prediction of multi-level transmission packet splitting strategies and unified management of caching strategies under multi-level transmission packet splitting between the host (master device, such as CPU, GPU, etc.) and slave devices (slave devices, such as DDR controllers, peripherals). It eliminates dependence on slave-side reordering modules (such as ROUs), reduces NoC network hardware resource overhead and data transmission latency from a hardware perspective, further removes the constraint of reordering processing on slave performance, ensures system communication efficiency in complex scenarios, eliminates resource redundancy in hierarchical architectures, and improves performance based on the slave device's receiving capability. The dynamic reservation of cache space based on force parameters avoids the waste of static allocation based on maximum burst volume in traditional solutions, improving resource utilization and achieving precise dynamic cache allocation. Through unified management of split-level identifiers and sequence identifiers, it ensures that sub-packets after multi-level splitting can still be merged in their original order after out-of-order return, reducing the overhead of data retransmission or verification and effectively guaranteeing end-to-end order. By eliminating the dependency on slave-side backpressure (such as the DDR command queue being shared by the master-side module), even in complex scenarios (such as multi-channel high load), it can still guarantee full-pressure DDR access, increasing the system bandwidth limit and achieving system performance decoupling. Through the combination of hardware and software, it comprehensively improves the processing efficiency of NoC multi-level out-of-order transmission packets. Compared with existing technologies, it can significantly reduce the hardware resource overhead and data transmission latency required for merging and rearranging, and further remove the constraint of rearrangement processing on slave performance, ensuring system communication efficiency in complex scenarios. These will be explained in detail below.

[0092] Example 1

[0093] Please see Figure 1 , Figure 1 This is a flowchart illustrating a NoC multi-level out-of-order packet processing method based on split prediction disclosed in an embodiment of the present invention. Figure 1 The described NoC multi-level out-of-order packet processing method based on split prediction can be applied to system-on-a-chip (SoC) and also to smart devices related to SoCs. These smart devices include, but are not limited to, one or more of battery devices, cloud devices, edge computing devices, relay devices, base station devices, urban management devices, smart connected devices, and smart home devices. This invention does not limit the scope of the application. Figure 1 As shown, the NoC multi-level out-of-order packet processing method based on split prediction can include the following operations:

[0094] 101. Identify the target slave device based on the first transmission packet parameters of the host. The target slave device is used to receive the first transmission packet corresponding to the first transmission packet parameters.

[0095] Optionally, in this embodiment of the invention, the executing entity of the method parses the identification information (such as DDR controller ID) of the target slave device based on the host's first transmission packet parameters (such as target address, transmission type, data volume, and target ID information). For example, if the target address of the transmission packet belongs to the DDR address space, the target slave device is identified as a DDR controller to determine the receiver of the transmission packet and provide a basis for subsequent splitting prediction.

[0096] 102. Determine the target slave device's current packet receiving capability parameters;

[0097] In this embodiment of the invention, optionally, parameters such as address alignment (e.g., 64-byte alignment), maximum burst data transmission volume (e.g., 128 bytes), transmission type (read / write), buffer capacity (e.g., 8KB), and multi-channel configuration (e.g., dual-channel support) can be obtained by querying the slave configuration register in order to dynamically monitor the current status of the slave (e.g., cache utilization, processing latency), thereby comprehensively evaluating the processing capability of the slave and providing data support for the splitting strategy.

[0098] 103. Based on the transmission packet receiving capability parameters, predict the downstream splitting strategy and downstream caching strategy of the first transmission packet. The downstream splitting strategy is used to split the first transmission packet into at least one level of second transmission packets that match the transmission packet receiving capability parameters. Each second transmission packet has a corresponding splitting level identifier and sequence identifier. The downstream caching strategy is used to instruct the target slave to determine the target cache space for each second transmission packet and to perform a cache rearrangement and merging operation on all second transmission packets based on all splitting level identifiers and all sequence identifiers.

[0099] In this embodiment of the invention, optionally, for the splitting strategy prediction, the splitting start address and sub-packet length can be determined based on the address alignment method of the slave device (e.g., the original transmission packet is 256 bytes, and it is split into 4 sub-packets according to 64-byte alignment); combined with the maximum burst data volume of the slave device, it is determined whether multi-level splitting is required (e.g., if the sub-packets still exceed the slave device capacity after the master device splits, then a second splitting on the slave device side is triggered).

[0100] For caching strategy prediction, the total amount of cache space required can be calculated (e.g., splitting into 4 sub-packets at the first level, splitting each sub-packet into 2 at the second level, requiring a total of 8 sub-packet cache spaces); dynamically allocate contiguous or non-contiguous cache addresses and mark them as "reserved" to ensure that the split sub-packets can be cached and avoid insufficient space or waste;

[0101] In embodiments of the present invention, such as Figure 2 , Figure 3 As shown, Figure 2 This is a schematic diagram of a raw segmented processing transmission packet rearrangement and merging process disclosed in an embodiment of the present invention; Figure 3 This is a schematic diagram of an improved segmented processing transmission packet rearrangement and merging process disclosed in an embodiment of the present invention; Figure 3 This paper describes a case study of unified handling of merging and reordering problems in a dual-channel NoC system: Based on the transmission splitting prediction and data order management scheme proposed in this invention, an improved reordering processing unit can be integrated at each host interface module to uniformly handle the reordering and merging of two levels of transmission packets. Compared with traditional hierarchical processing schemes, this invention does not require the integration of a reordering module (such as...) at the slave interface. Figure 2 The NoC system uses shaded modules ROU5 and ROU6 to handle the merging and rearrangement of secondary transmission packets, saving at least 16KB of storage resources (each DDR access data is 64B, and the slave-side rearrangement module requires 8KB of cache space to ensure full pressure on the DDR command queue). Furthermore, after removing the slave-side rearrangement module, the DDR command queue can be shared by the host-side rearrangement module. Therefore, even in complex scenarios, the NoC system can guarantee full pressure access to DDR, fully ensuring system communication performance.

[0102] In embodiments of the present invention, such as Figure 4 As shown, Figure 4 This is a schematic diagram of a NoC multi-level out-of-order transmission packet processing architecture based on split prediction disclosed in an embodiment of the present invention. In scenarios where transmission is split at multiple levels and returned out of order, it is necessary to further add indications for subsequent splits based on monitoring the order of transmission packets, and thereby carry out the merging management of sub-transmission packets. Figure 4This describes a data order management mechanism in a two-level splitting scenario: The host NI performs the initial splitting of the original transmission transaction, using the splitting end identifier (last) and sequence identifier-transmission sequence number (seqid) of the first splitting level to indicate the splitting and order of the first-level transmission packets, and packages them into the request delivery. The slave NI performs a second-level splitting of the transmission packets, adding a PTR pointer indicating the order of the second-level transmission packets and an end identifier indicating that the transmission packet splitting is complete, and merges them into the request delivery. These splitting-related indicators at each level are eventually returned with the response, and after being shaped into NoC internal packet format in the slave NI, they are forwarded to the return path. The host NI monitors the return status of each level of transmission packets based on the splitting indicators in the response packet, and tracks the sub-transmission packet cache addresses through a multi-level pointer linked list. Until a complete transmission return is detected and meets the reordering requirements, the transmission packets are returned sequentially according to the splitting order, while releasing the occupied cache space.

[0103] It should be noted that the current illustration shows the split prediction and order management mechanism in a two-level split scenario. For more complex split scenarios, it can be expanded by adding a new level of split indicator.

[0104] In summary, the splitting level identifier and sequence identifier mentioned above, in addition to representing the current level and sequence identifier, can also include the splitting end identifier (last) and sequence identifier - transmission sequence number (seqid) of the first splitting level, as well as the PTR pointer of the transmission packet sequence and the end identifier of the transmission packet splitting completion of the second splitting level; for splitting scenarios with more levels, expansion can be achieved by adding a new level splitting indicator identifier.

[0105] In this embodiment of the invention, as an optional implementation, the aforementioned transmission packet receiving capability parameters include at least one of address alignment parameters, burst transmission data volume parameters, transmission type parameters, buffer capacity parameters, multi-channel configuration parameters, transmission priority parameters, response delay parameters, and transmission protocol parameters; based on the transmission packet receiving capability parameters, predicting the downstream splitting strategy of the first transmission packet includes:

[0106] Based on the transmission packet receiving capability parameters, calculate the downstream splitting level and downstream splitting granularity of the first transmission packet;

[0107] Based on the NoC routing policy between the host and the target slave, predict the downstream transmission path of the first transmission packet;

[0108] Based on the downstream splitting level, downstream splitting granularity, and downstream transmission path, predict the downstream splitting strategy for the first transmission packet.

[0109] In this embodiment of the invention, optionally, for calculating the splitting level and granularity, the splitting level is calculated as follows: if the maximum burst data size of the slave device is 128 bytes and the original transmission packet is 512 bytes, then it needs to be split into at least 4 sub-packets (512 / 128 = 4); the splitting granularity is calculated as follows: according to the address alignment method (e.g., 64-byte alignment), the sub-packet length is adjusted to an integer multiple of 64 (e.g., 64, 128 bytes).

[0110] To predict downstream transmission paths, the path of sub-packets from the host to the target slave is determined based on the NoC routing table (such as the XY routing algorithm) (e.g., through channel A or B). Combined with multi-channel configuration, sub-packets are evenly distributed to different channels to improve throughput, predict the transmission path of sub-packets, avoid congestion, and optimize transmission efficiency.

[0111] For the comprehensive prediction of downstream splitting strategies, the splitting level, granularity, and path prediction results are integrated to generate the final splitting strategy. For example: the first level splits into 4 sub-packets (64 bytes), which are transmitted alternately through channels A and B; the slave side further splits into 2 sub-packets (32 bytes) to meet the burst capability of the DDR controller;

[0112] As can be seen, implementing this optional embodiment can accurately predict downstream splitting requirements (such as the necessity of secondary splitting) through comprehensive analysis of parameters such as address alignment, burst volume, and multi-channel configuration, avoiding bandwidth loss caused by insufficient or excessive splitting. Combined with NoC routing strategies (such as XY routing and channel load balancing), it can predict sub-packet transmission paths, reduce congestion and latency caused by path conflicts, support different slave characteristics (such as differences between DDR controllers and AI accelerators), dynamically generate differentiated splitting strategies, improve the versatility of the solution, and reduce repeated splitting operations caused by mismatched splitting granularity through granular calculation (such as matching slave burst capacity), thereby reducing processing overhead. It achieves a comprehensive prediction of the downstream splitting strategy for the first transmission packet by combining multi-dimensional parameter collaborative prediction, routing path optimization, flexible adaptation to heterogeneous slaves, and avoidance of invalid splitting.

[0113] In this optional embodiment, as an optional implementation, the above-mentioned calculation of the downstream splitting level and downstream splitting granularity of the first transmission packet based on the transmission packet receiving capability parameters includes:

[0114] Based on the transmission packet receiving capability parameters, analyze the current first transmission timeliness parameters of the target slave device;

[0115] Based on the target demand timeliness parameters and the first transmission timeliness parameters of the first transmission packet, determine the second transmission timeliness parameters of the host.

[0116] Based on the first transmission timeliness parameter and the second transmission timeliness parameter, calculate the downstream splitting level and downstream splitting granularity of the first transmission packet.

[0117] In this embodiment of the invention, optionally, for the first transmission timeliness parameter of the target slave device being analyzed, its current transmission timeliness (e.g., maximum throughput of 100Gb / s) can be calculated based on the slave device's processing latency (e.g., processing time per sub-packet is 10ns) and buffer idle rate. If the slave device's buffer occupancy rate is high, the timeliness parameter is reduced, and the number of splitting levels needs to be reduced;

[0118] To determine the second transmission timeliness parameter of the host, the maximum timeliness it can support (e.g., 150Gb / s) can be calculated based on the host's splitting capability (e.g., splitting engine frequency) and transmission priority.

[0119] Regarding the calculation of split levels and granularity, if the slave timeliness is 100Gb / s and the master timeliness is 150Gb / s, then the split strategy needs to balance the differences between the two. The split level should be selected as 2 levels (master level 1 split, slave level 2 split), with each level having a granularity of 64 bytes, so as to avoid the master or slave becoming a performance bottleneck through timeliness matching.

[0120] As can be seen, implementing this optional embodiment can adjust the splitting level according to the real-time processing capability of the slave (such as the current load of the DDR controller). When under high load, the splitting level is reduced to reduce processing pressure, and the level is increased when idle to improve bandwidth utilization. By coordinating the transmission timeliness of the master and slave (such as reducing the splitting granularity when the master timeliness is greater than the slave timeliness), the system load is balanced to prevent one side from becoming a performance bottleneck. In dynamically changing system environments (such as burst traffic and multi-task concurrency), the splitting strategy is automatically adjusted to ensure the stability of communication efficiency. It achieves a combination of timeliness-driven dynamic splitting, avoidance of single-sided performance bottlenecks, and adaptation to complex scenarios, dynamically adjusting the splitting level and granularity.

[0121] In this optional embodiment, as another optional implementation, the above-mentioned analysis of the target slave's current first transmission timeliness parameter based on the transmission packet reception capability parameter includes:

[0122] Based on the transmission packet receiving capability parameters, the splitting level capability parameters of the target slave device are analyzed to determine the first available splitting level parameters and the first splitting granularity parameters of the target slave device. The splitting level capability parameters include at least one of the following: splitting logic configuration parameters, hardware processing depth limit parameters, and protocol constraint parameters.

[0123] Based on the first available splitting level parameter and the first splitting granularity parameter, analyze the current first transmission timeliness parameter of the target slave device;

[0124] Optionally, the calculation of the downstream splitting level and downstream splitting granularity of the first transmission packet based on the first transmission timeliness parameter and the second transmission timeliness parameter includes:

[0125] Obtain the host's splitting strategy configuration parameters, which include at least one of the following: second available splitting level parameter, second splitting granularity parameter, and splitting priority parameter;

[0126] Based on the splitting strategy configuration parameters, analyze the host's current third transmission timeliness parameters;

[0127] Calculate the target distance value between the second transmission timeliness parameter and the third transmission timeliness parameter;

[0128] Based on the target distance value, the first available splitting level parameter, the first splitting granularity parameter, and the splitting strategy configuration parameter, generate collaborative splitting matching parameters between the host and the target slave.

[0129] Based on the collaborative splitting matching parameters, the downstream splitting level and downstream splitting granularity of the first transmission packet are calculated. The downstream splitting level includes at least one splitting level, and the downstream splitting granularity includes the splitting granularity corresponding to each splitting level.

[0130] In this embodiment of the invention, optionally, for parsing the splitting level capability of the target slave device, the splitting level configuration of the slave device can be read (e.g., supporting a maximum of 2 levels of splitting), its splitting logic constraints can be obtained (e.g., only supporting splitting by address alignment) and the hardware processing depth can be analyzed (e.g., buffer capacity limits the splitting level).

[0131] For the collaborative matching master-slave splitting strategy, if the master supports 3-level splitting but the slave only supports 2-level splitting, then the final splitting level will be 2 levels; if the master's splitting granularity (128 bytes) exceeds the slave's capacity (64 bytes), then the slave will be triggered to supplement the splitting; this ensures that the splitting strategy is adapted to the master-slave capabilities and avoids invalid splitting.

[0132] To generate collaborative splitting matching parameters, a matching parameter table can be generated based on the splitting level capability, granularity, and priority of the master and slave machines to guide the final splitting strategy.

[0133] As can be seen, implementing this optional embodiment can dynamically select the optimal splitting level based on the host's splitting capabilities (e.g., supporting 3-level splitting) and the slave's processing constraints (e.g., supporting only 2 levels), avoiding resource idleness or exceeding limits; by parsing the slave's protocol constraints (e.g., AXI burst length limit), it ensures that the splitting strategy conforms to the protocol specification, reducing transmission errors; based on transmission priority parameters (e.g., high real-time tasks), it prioritizes the splitting granularity of critical transmissions, ensuring low latency requirements; and achieves dynamic adaptation of the level and granularity by combining master-slave collaborative resource integration, enhanced protocol compatibility, and priority-driven splitting optimization.

[0134] In another optional implementation of this invention, the above-mentioned prediction of the downstream buffering strategy for the first transmission packet based on the transmission packet receiving capability parameters includes:

[0135] Based on the transmission packet receiving capability parameters, determine whether the target slave device has the reserve buffer space required to meet the downstream splitting strategy;

[0136] When it is determined that the target slave device has the reserve buffer space required to satisfy the downstream splitting strategy, the downstream buffering strategy of the first transmission packet is predicted based on the reserve buffer space.

[0137] When it is determined that the target slave does not have the reserve buffer space required to satisfy the downstream splitting strategy, an upstream backpressure instruction for the target slave is generated according to the target priority parameter of the first transmission packet to adjust the current buffer space of the target slave and re-trigger the operation of determining whether the target slave has the reserve buffer space required to satisfy the downstream splitting strategy.

[0138] In this embodiment of the invention, to determine the available buffer space, the cache status table of the target slave can be queried to check whether there is enough contiguous space required for the predicted split (e.g., 8 sub-packets × 64 bytes = 512 bytes); if the space is sufficient, it is marked as "allocated"; otherwise, backpressure is triggered.

[0139] For generating backpressure commands, if the cache is insufficient, a backpressure signal can be sent to the host to suspend the issuance of new requests until the slave releases space (e.g., 256 bytes are released after merging the output);

[0140] For the re-triggering judgment, the cache space check can be re-executed after the back pressure is released until the prepared space meets the requirements;

[0141] As can be seen, implementing this optional embodiment can accurately trigger backpressure when the cache is insufficient, suspend the issuance of non-critical requests (such as low-priority transmissions), and prioritize the resource allocation of high-priority tasks; through space reallocation after the backpressure is released, it avoids the "cache dead zone" caused by fixed reservation in traditional solutions, thereby improving resource turnover; in the event of sudden traffic surges, it prevents the system from crashing due to cache overflow by dynamically adjusting cache allocation, thereby enhancing fault tolerance; and it achieves dynamic cache control that combines intelligent backpressure mechanism, elastic cache reclamation, and improved system robustness.

[0142] 104. Based on the downstream splitting strategy and the downstream caching strategy, generate the target output instruction. The target output instruction is used to instruct the target slave to output all second transmission packets based on all splitting level identifiers and all sequence identifiers, and release the target cache space corresponding to the downstream caching strategy.

[0143] In this embodiment of the invention, optionally, after all the split sub-packets have completed cache rearrangement and merging, an output instruction is generated based on the splitting level identifier (such as Level 1 L1, Level 2 L2) and the sequence identifier (such as SEQ01, SEQ02); the output instruction triggers the slave device to send the merged data in sequence and release the occupied cache space; to ensure that the data is output in the original order and to reclaim resources for subsequent transmission.

[0144] As can be seen, implementing the embodiments of the present invention can identify the target slave device to which the corresponding first transmission packet needs to be transmitted based on the parameters of the first transmission packet. Based on the current transmission packet receiving capability parameters of the determined target slave device, the downstream splitting strategy and downstream caching strategy of the first transmission packet are predicted. This enables unified prediction of multi-level splitting strategies and unified management of caching strategies under multi-level splitting of transmission packets between the host (master device, such as CPU, GPU, etc.) and the slave device (slave device, such as DDR controller, peripherals). It eliminates the dependence on the reordering module (such as ROU) on the slave side, reduces the hardware resource overhead and data transmission latency of the NoC network from the hardware dimension, and further removes the constraint of reordering processing on slave performance, ensuring system communication efficiency in complex scenarios, eliminating resource redundancy in hierarchical architecture, and dynamically reserving cache space based on the receiving capability parameters of the slave device. This approach avoids the waste of static allocation based on maximum burst volume in traditional solutions, improves resource utilization, and achieves precise dynamic cache allocation. By unifying the management of split-level identifiers and sequence identifiers, it ensures that sub-packets after multi-level splitting can still be merged in their original order after out-of-order return, reducing the overhead of data retransmission or verification and effectively guaranteeing end-to-end order. By eliminating the dependence on backpressure on the slave side (such as the DDR command queue being shared by the host-side module), it can still guarantee full-pressure access of DDR even in complex scenarios (such as multi-channel high load), increasing the system bandwidth limit and achieving system performance decoupling. Through the combination of hardware and software, it comprehensively improves the processing efficiency of NoC multi-level out-of-order transmission packets. Compared with existing technical solutions, it can significantly reduce the hardware resource overhead and data transmission latency required for merging and rearranging, and further remove the constraint of rearrangement processing on slave performance, ensuring system communication efficiency in complex scenarios.

[0145] Example 2

[0146] Please see Figure 5 , Figure 5 This is a flowchart illustrating another NoC multi-level out-of-order packet processing method based on split prediction disclosed in an embodiment of the present invention. Figure 5The described NoC multi-level out-of-order packet processing method based on split prediction can be applied to system-on-a-chip (SoC) and also to smart devices related to SoCs. These smart devices include, but are not limited to, one or more of battery devices, cloud devices, edge computing devices, relay devices, base station devices, urban management devices, smart connected devices, and smart home devices. This invention does not limit the scope of the application. Figure 5 As shown, the NoC multi-level out-of-order packet processing method based on split prediction can include the following operations:

[0147] 201. Identify the target slave device based on the first transmission packet parameters of the host. The target slave device is used to receive the first transmission packet corresponding to the first transmission packet parameters.

[0148] 202. Determine the target slave device's current packet receiving capability parameters;

[0149] 203. Based on the transmission packet receiving capability parameters, predict the downstream splitting strategy and downstream caching strategy of the first transmission packet. The downstream splitting strategy is used to split the first transmission packet into at least one level of second transmission packets that match the transmission packet receiving capability parameters. Each second transmission packet has a corresponding splitting level identifier and sequence identifier. The downstream caching strategy is used to instruct the target slave to determine the target cache space for each second transmission packet and to perform a cache rearrangement and merging operation on all second transmission packets based on all splitting level identifiers and all sequence identifiers.

[0150] In this embodiment of the invention, for the supplementary explanation of steps 201-203, please refer to the supplementary explanation of steps 101-103 in Embodiment 1. This embodiment of the invention will not repeat the explanation here.

[0151] 204. Based on the downstream splitting strategy and the downstream caching strategy, determine whether all second transmission packets have completed the cache rearrangement and merging operation at the target slave.

[0152] In this embodiment of the invention, as an optional implementation, the method for determining whether all second transmission packets are processed before the target slave completes the cache rearrangement and merging operation, based on the downstream splitting strategy and the downstream caching strategy, further includes:

[0153] The buffer address of each second transmission packet is determined based on the multi-level pointer linked list;

[0154] For each split level identifier, based on the cache address and sequence identifier of all second transmission packets corresponding to that split level identifier, determine whether all second transmission packets corresponding to that split level identifier have been completely transmitted to the target slave and whether all sequence identifiers match the preset sequence identifier;

[0155] When it is determined that all second transmission packets corresponding to all split level identifiers have been completely transmitted to the target slave and all sequence identifiers match the preset sequence identifiers, the operation of determining whether all second transmission packets have completed the cache rearrangement and merging operation on the target slave is triggered according to the downstream splitting strategy and the downstream caching strategy.

[0156] 205. When it is determined that all second transmission packets have completed the cache rearrangement and merging operation on the target slave, a target output instruction is generated based on all split level identifiers and all sequence identifiers. The target output instruction is used to instruct the target slave to output all second transmission packets based on all split level identifiers and all sequence identifiers, and release the target cache space corresponding to the downstream cache strategy.

[0157] In this embodiment of the invention, optionally, for multi-level pointer linked list management, a linked list head node (such as first-level head node L1, second-level head node L2) can be created for each split level; each sub-package generates a linked list child node, recording its cache address, sequence identifier and next node pointer;

[0158] To determine if cache reordering is complete, the linked list nodes can be traversed to check if all sub-packages at the same level have been returned in the correct order (e.g., SEQ01→SEQ02); if a missing or out-of-order packet is detected, wait until the condition is met.

[0159] For generating the target output instruction, after all sub-packages at all levels have been merged, the data can be output in the order of the linked list, and the linked list nodes can be deleted to release the cache.

[0160] As can be seen, implementing this embodiment of the invention can dynamically associate the cache address and sequence identifier (such as seqid, PTR) of sub-packets through a multi-level pointer linked list. Even if sub-packets return out of order across levels, they can still be quickly located and merged. After detecting that all sub-packets at the same level have returned completely, the merging operation is automatically triggered, reducing manual intervention and improving processing efficiency. After merging, all associated cache space is released at once, avoiding resource waste caused by some cache residue in traditional solutions. By verifying the sequence identifier (such as the end marker of the second-level sub-packet) level by level, it is ensured that the merged data block is completely consistent with the original transmission packet, reducing the risk of data errors. It combines efficient out-of-order management, automated merging triggering, atomicity of cache release, and cross-level data integrity, significantly reducing the hardware resource overhead and data transmission latency required for merging and rearranging, and further removing the constraint of rearranging processing on slave performance, ensuring system communication efficiency in complex scenarios.

[0161] Example 3

[0162] Please see Figure 6 , Figure 6This is a schematic diagram of a NoC multi-level out-of-order packet processing device based on split prediction, as disclosed in an embodiment of the present invention. This NoC multi-level out-of-order packet processing device based on split prediction can be applied to system-on-a-chip (SoC) and also to smart devices related to SoCs. These smart devices include, but are not limited to, one or more of the following: battery devices, cloud devices, edge computing devices, relay devices, base station devices, urban management devices, smart connected devices, and smart home devices. The embodiments of the present invention do not impose any limitations on this. Figure 6 As shown, the NoC multi-level out-of-order packet processing apparatus based on split prediction may include:

[0163] The identification module 301 is used to identify the target slave device according to the first transmission packet parameters of the host, and the target slave device is used to receive the first transmission packet corresponding to the first transmission packet parameters;

[0164] The determination module 302 is used to determine the current transmission packet receiving capability parameters of the target slave device;

[0165] The prediction module 303 is used to predict the downstream splitting strategy and the downstream caching strategy of the first transmission packet according to the transmission packet receiving capability parameters. The downstream splitting strategy is used to split the first transmission packet into at least one level of second transmission packets that match the transmission packet receiving capability parameters. Each second transmission packet has a splitting level identifier and a sequence identifier corresponding to the second transmission packet. The downstream caching strategy is used to instruct the target slave to determine the target cache space for each second transmission packet and to perform a cache rearrangement and merging operation on all second transmission packets based on all splitting level identifiers and all sequence identifiers.

[0166] The generation module 304 is used to generate a target output instruction based on the downstream splitting strategy and the downstream caching strategy. The target output instruction is used to instruct the target slave to output all second transmission packets based on all splitting level identifiers and all sequence identifiers, and to release the target cache space corresponding to the downstream caching strategy.

[0167] As can be seen, implementing the embodiments of the present invention can identify the target slave device to which the corresponding first transmission packet needs to be transmitted based on the parameters of the first transmission packet. Based on the current transmission packet receiving capability parameters of the determined target slave device, the downstream splitting strategy and downstream caching strategy of the first transmission packet are predicted. This enables unified prediction of multi-level splitting strategies and unified management of caching strategies under multi-level splitting of transmission packets between the host (master device, such as CPU, GPU, etc.) and the slave device (slave device, such as DDR controller, peripherals). It eliminates the dependence on the reordering module (such as ROU) on the slave side, reduces the hardware resource overhead and data transmission latency of the NoC network from the hardware dimension, and further removes the constraint of reordering processing on slave performance, ensuring system communication efficiency in complex scenarios, eliminating resource redundancy in hierarchical architecture, and dynamically reserving cache space based on the receiving capability parameters of the slave device. This approach avoids the waste of static allocation based on maximum burst volume in traditional solutions, improves resource utilization, and achieves precise dynamic cache allocation. By unifying the management of split-level identifiers and sequence identifiers, it ensures that sub-packets after multi-level splitting can still be merged in their original order after out-of-order return, reducing the overhead of data retransmission or verification and effectively guaranteeing end-to-end order. By eliminating the dependence on backpressure on the slave side (such as the DDR command queue being shared by the host-side module), it can still guarantee full-pressure access of DDR even in complex scenarios (such as multi-channel high load), increasing the system bandwidth limit and achieving system performance decoupling. Through the combination of hardware and software, it comprehensively improves the processing efficiency of NoC multi-level out-of-order transmission packets. Compared with existing technical solutions, it can significantly reduce the hardware resource overhead and data transmission latency required for merging and rearranging, and further remove the constraint of rearrangement processing on slave performance, ensuring system communication efficiency in complex scenarios.

[0168] In this embodiment of the invention, as an optional implementation, the aforementioned transmission packet receiving capability parameters include at least one of the following: address alignment parameters, burst transmission data volume parameters, transmission type parameters, buffer capacity parameters, multi-channel configuration parameters, transmission priority parameters, response delay parameters, and transmission protocol parameters; the specific method by which the prediction module 303 predicts the downstream splitting strategy of the first transmission packet based on the transmission packet receiving capability parameters includes:

[0169] Based on the transmission packet receiving capability parameters, calculate the downstream splitting level and downstream splitting granularity of the first transmission packet;

[0170] Based on the NoC routing policy between the host and the target slave, predict the downstream transmission path of the first transmission packet;

[0171] Based on the downstream splitting level, downstream splitting granularity, and downstream transmission path, predict the downstream splitting strategy for the first transmission packet.

[0172] As can be seen, implementing this optional embodiment can accurately predict downstream splitting requirements (such as the necessity of secondary splitting) through comprehensive analysis of parameters such as address alignment, burst volume, and multi-channel configuration, avoiding bandwidth loss caused by insufficient or excessive splitting. Combined with NoC routing strategies (such as XY routing and channel load balancing), it can predict sub-packet transmission paths, reduce congestion and latency caused by path conflicts, support different slave characteristics (such as differences between DDR controllers and AI accelerators), dynamically generate differentiated splitting strategies, improve the versatility of the solution, and reduce repeated splitting operations caused by mismatched splitting granularity through granular calculation (such as matching slave burst capacity), thereby reducing processing overhead. It achieves a comprehensive prediction of the downstream splitting strategy for the first transmission packet by combining multi-dimensional parameter collaborative prediction, routing path optimization, flexible adaptation to heterogeneous slaves, and avoidance of invalid splitting.

[0173] In this optional embodiment, as an optional implementation, the specific method by which the prediction module 303 calculates the downstream splitting level and downstream splitting granularity of the first transmission packet based on the transmission packet receiving capability parameters includes:

[0174] Based on the transmission packet receiving capability parameters, analyze the current first transmission timeliness parameters of the target slave device;

[0175] Based on the target demand timeliness parameters and the first transmission timeliness parameters of the first transmission packet, determine the second transmission timeliness parameters of the host.

[0176] Based on the first transmission timeliness parameter and the second transmission timeliness parameter, calculate the downstream splitting level and downstream splitting granularity of the first transmission packet.

[0177] As can be seen, implementing this optional embodiment can adjust the splitting level according to the real-time processing capability of the slave (such as the current load of the DDR controller). When under high load, the splitting level is reduced to reduce processing pressure, and the level is increased when idle to improve bandwidth utilization. By coordinating the transmission timeliness of the master and slave (such as reducing the splitting granularity when the master timeliness is greater than the slave timeliness), the system load is balanced to prevent one side from becoming a performance bottleneck. In dynamically changing system environments (such as burst traffic and multi-task concurrency), the splitting strategy is automatically adjusted to ensure the stability of communication efficiency. It achieves a combination of timeliness-driven dynamic splitting, avoidance of single-sided performance bottlenecks, and adaptation to complex scenarios, dynamically adjusting the splitting level and granularity.

[0178] In this optional embodiment, as another optional implementation, the specific method by which the prediction module 303 analyzes the target slave's current first transmission timeliness parameter based on the transmission packet receiving capability parameter includes:

[0179] Based on the transmission packet receiving capability parameters, the splitting level capability parameters of the target slave device are analyzed to determine the first available splitting level parameters and the first splitting granularity parameters of the target slave device. The splitting level capability parameters include at least one of the following: splitting logic configuration parameters, hardware processing depth limit parameters, and protocol constraint parameters.

[0180] Based on the first available splitting level parameter and the first splitting granularity parameter, analyze the current first transmission timeliness parameter of the target slave device;

[0181] Optionally, the prediction module 303 calculates the downstream splitting level and downstream splitting granularity of the first transmission packet based on the first transmission timeliness parameter and the second transmission timeliness parameter in the following specific ways:

[0182] Obtain the host's splitting strategy configuration parameters, which include at least one of the following: second available splitting level parameter, second splitting granularity parameter, and splitting priority parameter;

[0183] Based on the splitting strategy configuration parameters, analyze the host's current third transmission timeliness parameters;

[0184] Calculate the target distance value between the second transmission timeliness parameter and the third transmission timeliness parameter;

[0185] Based on the target distance value, the first available splitting level parameter, the first splitting granularity parameter, and the splitting strategy configuration parameter, generate collaborative splitting matching parameters between the host and the target slave.

[0186] Based on the collaborative splitting matching parameters, the downstream splitting level and downstream splitting granularity of the first transmission packet are calculated. The downstream splitting level includes at least one splitting level, and the downstream splitting granularity includes the splitting granularity corresponding to each splitting level.

[0187] As can be seen, implementing this optional embodiment can dynamically select the optimal splitting level based on the host's splitting capabilities (e.g., supporting 3-level splitting) and the slave's processing constraints (e.g., supporting only 2 levels), avoiding resource idleness or exceeding limits; by parsing the slave's protocol constraints (e.g., AXI burst length limit), it ensures that the splitting strategy conforms to the protocol specification, reducing transmission errors; based on transmission priority parameters (e.g., high real-time tasks), it prioritizes the splitting granularity of critical transmissions, ensuring low latency requirements; and achieves dynamic adaptation of the level and granularity by combining master-slave collaborative resource integration, enhanced protocol compatibility, and priority-driven splitting optimization.

[0188] In this embodiment of the invention, as another optional implementation, the specific method by which the prediction module 303 predicts the downstream caching strategy of the first transmission packet based on the transmission packet receiving capability parameters includes:

[0189] Based on the transmission packet receiving capability parameters, determine whether the target slave device has the reserve buffer space required to meet the downstream splitting strategy;

[0190] When it is determined that the target slave device has the reserve buffer space required to satisfy the downstream splitting strategy, the downstream buffering strategy of the first transmission packet is predicted based on the reserve buffer space.

[0191] When it is determined that the target slave does not have the reserve buffer space required to satisfy the downstream splitting strategy, an upstream backpressure instruction for the target slave is generated according to the target priority parameter of the first transmission packet to adjust the current buffer space of the target slave and re-trigger the operation of determining whether the target slave has the reserve buffer space required to satisfy the downstream splitting strategy.

[0192] As can be seen, implementing this optional embodiment can accurately trigger backpressure when the cache is insufficient, suspend the issuance of non-critical requests (such as low-priority transmissions), and prioritize the resource allocation of high-priority tasks; through space reallocation after the backpressure is released, it avoids the "cache dead zone" caused by fixed reservation in traditional solutions, thereby improving resource turnover; in the event of sudden traffic surges, it prevents the system from crashing due to cache overflow by dynamically adjusting cache allocation, thereby enhancing fault tolerance; and it achieves dynamic cache control that combines intelligent backpressure mechanism, elastic cache reclamation, and improved system robustness.

[0193] In this embodiment of the invention, as another optional implementation, the specific method by which the generation module 304 generates the target output instruction according to the downstream splitting strategy and the downstream caching strategy includes:

[0194] Based on the downstream splitting strategy and the downstream caching strategy, determine whether all second transmission packets have completed the cache rearrangement and merging operation at the target slave.

[0195] When it is determined that all second transmission packets have completed the buffer rearrangement and merging operation at the target slave, the target output instruction is generated based on all split level identifiers and all sequence identifiers.

[0196] Optionally, the determining module 302 is also used to determine the cache address of each second transmission packet based on a multi-level pointer linked list before the generating module determines whether all second transmission packets are cached before the target slave completes the cache reordering and merging operation, according to the downstream splitting strategy and the downstream caching strategy.

[0197] Optional, such as Figure 7 As shown, the device also includes:

[0198] The judgment module 305 is used to determine, for each split level identifier, whether all second transmission packets corresponding to that split level identifier have been completely transmitted to the target slave and whether all sequence identifiers match the preset sequence identifier, based on the cache address and sequence identifier of all second transmission packets corresponding to that split level identifier; when it is determined that all second transmission packets corresponding to all split level identifiers have been completely transmitted to the target slave and all sequence identifiers match the preset sequence identifier, the generation module 304 is triggered to perform the operation of determining whether all second transmission packets have completed the cache rearrangement and merging operation at the target slave according to the downstream splitting strategy and the downstream caching strategy.

[0199] As can be seen, implementing this embodiment of the invention can dynamically associate the cache address and sequence identifier (such as seqid, PTR) of sub-packets through a multi-level pointer linked list. Even if sub-packets return out of order across levels, they can still be quickly located and merged. After detecting that all sub-packets at the same level have returned completely, the merging operation is automatically triggered, reducing manual intervention and improving processing efficiency. After merging, all associated cache space is released at once, avoiding resource waste caused by some cache residue in traditional solutions. By verifying the sequence identifier (such as the end marker of the second-level sub-packet) level by level, it is ensured that the merged data block is completely consistent with the original transmission packet, reducing the risk of data errors. It combines efficient out-of-order management, automated merging triggering, atomicity of cache release, and cross-level data integrity, significantly reducing the hardware resource overhead and data transmission latency required for merging and rearranging, and further removing the constraint of rearranging processing on slave performance, ensuring system communication efficiency in complex scenarios.

[0200] Example 4

[0201] This invention discloses an electronic unit integrated in a host computer, wherein the electronic unit is used to execute the NoC multi-level out-of-order packet processing method based on split prediction described in Embodiment 1 or Embodiment 2 of this invention.

[0202] Example 5

[0203] Please see Figure 8 , Figure 8 This is a schematic diagram of another NoC multi-level out-of-order packet processing device based on split prediction disclosed in an embodiment of the present invention. This NoC multi-level out-of-order packet processing device based on split prediction can be applied to system-on-a-chip (SoC) and also to smart devices related to SoCs. These smart devices include, but are not limited to, one or more of the following: battery devices, cloud devices, edge computing devices, relay devices, base station devices, urban management devices, smart connected devices, and smart home devices. The embodiments of the present invention do not impose any limitations on this. Figure 8 As shown, the NoC multi-level out-of-order packet processing apparatus based on split prediction may include:

[0204] Memory 401 that stores executable program code.

[0205] Processor 402 coupled to memory 401.

[0206] The processor 402 calls the executable program code stored in the memory 401 to execute the steps in the NoC multi-level out-of-order transmission packet processing method based on split prediction described in Embodiment 1 or Embodiment 2 of the present invention.

[0207] Example 6

[0208] This invention discloses a computer storage medium storing computer instructions. When these computer instructions are invoked, they are used to execute the steps in the NoC multi-level out-of-order transmission packet processing method based on split prediction described in Embodiment 1 or Embodiment 2 of this invention.

[0209] Example 7

[0210] This invention discloses a computer program product, which includes a non-transitory computer storage medium storing a computer program, and the computer program is operable to cause a computer to perform the steps in the NoC multi-level out-of-order transport packet processing method based on split prediction described in Embodiment 1 or Embodiment 2.

[0211] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0212] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0213] Finally, it should be noted that the NoC multi-level out-of-order transmission packet processing method and apparatus based on splitting prediction disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention, and are only used to illustrate the technical solutions of the present invention, not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for split-prediction based NoC multi-stage out-of-order packet processing, characterized in that, The method includes: Based on the first transmission packet parameters of the host, the target slave is identified, and the target slave is used to receive the first transmission packet corresponding to the first transmission packet parameters; Determine the current packet reception capability parameters of the target slave device; Based on the transmission packet receiving capability parameters, a downstream splitting strategy and a downstream caching strategy for the first transmission packet are predicted. The downstream splitting strategy is used to split the first transmission packet into at least one level of second transmission packets that match the transmission packet receiving capability parameters. Each second transmission packet has a corresponding splitting level identifier and sequence identifier. The downstream caching strategy is used to instruct the target slave to determine the target cache space for each second transmission packet and to perform a cache rearrangement and merging operation on all second transmission packets based on all the splitting level identifiers and all the sequence identifiers. Based on the downstream splitting strategy and the downstream caching strategy, a target output instruction is generated. The target output instruction is used to instruct the target slave to output all the second transmission packets based on all the splitting level identifiers and all the sequence identifiers, and to release the target cache space corresponding to the downstream caching strategy.

2. The split prediction based NoC multi-stage out-of-order packet processing method according to claim 1, characterized in that, The transmission packet receiving capability parameters include at least one of the following: address alignment parameter, burst transmission data volume parameter, transmission type parameter, buffer capacity parameter, multi-channel configuration parameter, transmission priority parameter, response delay parameter, and transmission protocol parameter. The step of predicting the downstream splitting strategy of the first transmission packet based on the transmission packet receiving capability parameters includes: Based on the transmission packet receiving capability parameters, calculate the downstream splitting level and downstream splitting granularity of the first transmission packet; Based on the NoC routing policy between the host and the target slave, predict the downstream transmission path of the first transmission packet; Based on the downstream splitting level, the downstream splitting granularity, and the downstream transmission path, predict the downstream splitting strategy for the first transmission packet.

3. The NoC multi-level out-of-order transmission packet processing method based on split prediction according to claim 2, characterized in that, The step of calculating the downstream splitting level and downstream splitting granularity of the first transmission packet based on the transmission packet receiving capability parameters includes: Based on the transmission packet receiving capability parameters, analyze the current first transmission timeliness parameters of the target slave device; The second transmission timeliness parameter of the host is determined based on the target demand timeliness parameter of the first transmission packet and the first transmission timeliness parameter; Based on the first transmission timeliness parameter and the second transmission timeliness parameter, calculate the downstream splitting level and downstream splitting granularity of the first transmission packet.

4. The NoC multi-level out-of-order transmission packet processing method based on split prediction according to claim 3, characterized in that, The step of analyzing the target slave's current first transmission timeliness parameter based on the transmission packet receiving capability parameter includes: Based on the transmission packet receiving capability parameters, the splitting level capability parameters of the target slave device are analyzed to determine the first available splitting level parameters and the first splitting granularity parameters of the target slave device. The splitting level capability parameters include at least one of the following: splitting logic configuration parameters, hardware processing depth limit parameters, and protocol constraint parameters. Based on the first available splitting level parameter and the first splitting granularity parameter, analyze the current first transmission timeliness parameter of the target slave device; And, the step of calculating the downstream splitting level and downstream splitting granularity of the first transmission packet based on the first transmission timeliness parameter and the second transmission timeliness parameter includes: Obtain the splitting strategy configuration parameters of the host, wherein the splitting strategy configuration parameters include at least one of the following: second available splitting level parameter, second splitting granularity parameter, and splitting priority parameter; Based on the configuration parameters of the splitting strategy, analyze the current third transmission timeliness parameters of the host; Calculate the target distance value between the second transmission timeliness parameter and the third transmission timeliness parameter; Based on the target distance value, the first available splitting level parameter, the first splitting granularity parameter, and the splitting strategy configuration parameter, generate collaborative splitting matching parameters between the host and the target slave. Based on the collaborative splitting matching parameters, the downstream splitting level and downstream splitting granularity of the first transmission packet are calculated. The downstream splitting level includes at least one splitting level, and the downstream splitting granularity includes the splitting granularity corresponding to each splitting level.

5. The NoC multi-level out-of-order transmission packet processing method based on split prediction according to any one of claims 1-4, characterized in that, The step of predicting the downstream caching strategy for the first transmission packet based on the transmission packet receiving capability parameters includes: Based on the transmission packet receiving capability parameters, determine whether the target slave device has a reserve buffer space required to satisfy the downstream splitting strategy; When it is determined that the target slave device has the required reserve cache space to satisfy the downstream splitting strategy, the downstream caching strategy of the first transmission packet is predicted based on the reserve cache space. When it is determined that the target slave does not have the required reserve cache space to satisfy the downstream splitting strategy, an upstream backpressure instruction for the target slave is generated according to the target priority parameter of the first transmission packet to adjust the current cache space of the target slave and re-trigger the operation of determining whether the target slave has the required reserve cache space to satisfy the downstream splitting strategy.

6. The NoC multi-level out-of-order transmission packet processing method based on split prediction according to any one of claims 1-4, characterized in that, The step of generating the target output instruction based on the downstream splitting strategy and the downstream caching strategy includes: Based on the downstream splitting strategy and the downstream caching strategy, it is determined whether all the second transmission packets have completed the cache rearrangement and merging operation on the target slave. When it is determined that all the second transmission packets have completed the buffer rearrangement and merging operation at the target slave, a target output instruction is generated based on all the split level identifiers and all the sequence identifiers. And, before determining whether all the second transmission packets are present at the target slave device before the target slave device completes the cache rearrangement and merging operation, based on the downstream splitting strategy and the downstream caching strategy, the method further includes: The cache address of each second transmission packet is determined based on a multi-level pointer linked list; For each split-level identifier, based on the cache address and sequence identifier of all the second transmission packets corresponding to the split-level identifier, it is determined whether all the second transmission packets corresponding to the split-level identifier have been completely transmitted to the target slave and whether all the sequence identifiers match the preset sequence identifier; When it is determined that all the second transmission packets corresponding to all split level identifiers have been completely transmitted to the target slave and all the sequence identifiers match the preset sequence identifiers, the operation of determining whether all the second transmission packets have completed the cache rearrangement and merging operation on the target slave is triggered according to the downstream splitting strategy and the downstream caching strategy.

7. A NoC multi-level out-of-order transmission packet processing device based on split prediction, characterized in that, The device includes: The identification module is used to identify the target slave device based on the first transmission packet parameters of the host, wherein the target slave device is used to receive the first transmission packet corresponding to the first transmission packet parameters; The determination module is used to determine the current transmission packet receiving capability parameters of the target slave device; The prediction module is used to predict the downstream splitting strategy and the downstream caching strategy of the first transmission packet based on the transmission packet receiving capability parameters. The downstream splitting strategy is used to split the first transmission packet into at least one level of second transmission packets that match the transmission packet receiving capability parameters. Each second transmission packet has a corresponding splitting level identifier and sequence identifier. The downstream caching strategy is used to instruct the target slave to determine the target cache space for each second transmission packet and to perform a cache rearrangement and merging operation on all second transmission packets based on all the splitting level identifiers and all the sequence identifiers. The generation module is used to generate a target output instruction based on the downstream splitting strategy and the downstream caching strategy. The target output instruction is used to instruct the target slave to output all the second transmission packets based on all the splitting level identifiers and all the sequence identifiers, and to release the target cache space corresponding to the downstream caching strategy.

8. An electronic unit, characterized in that, The electronic unit is integrated in the host and is used to perform the NoC multi-level out-of-order transmission packet processing method based on split prediction as described in any one of claims 1-6.

9. A NoC multi-level out-of-order transmission packet processing device based on split prediction, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the NoC multi-level out-of-order transmission packet processing method based on split prediction as described in any one of claims 1-6.

10. A computer storage medium, characterized in that, The computer storage medium stores computer instructions, which, when invoked, are used to execute the NoC multi-level out-of-order transmission packet processing method based on split prediction as described in any one of claims 1-6.

Images