A structure and method for cross-chip transmission

By designing the method of early response and ID remapping in cross-chip transmission, the communication delay and resource waste during cross-chip access in large-scale computing clusters are solved, and efficient cross-chip transmission and bandwidth utilization are achieved.

CN118193444BActive Publication Date: 2025-04-22BEIJING TSINGMICRO INTELLIGENT TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202410139667.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-02-01
Publication Date
2025-04-22
Estimated Expiration
2044-02-01

AI Technical Summary

Technical Problem

The prior art lacks early response design for write transmission when cross-chip access in large-scale computing clusters, resulting in large communication delays, waste of resources and high power consumption.

Method used

A structure and method for cross-chip transmission is designed, including a master agent and a slave agent, an early response write request detector, an early response generator, a send multiplexer, a response multiplexer, an early response parser and an ID remapping table to realize early response and ID remapping of cross-chip transmission, and reduce the inter-chip communication bandwidth occupied by the response.

Benefits of technology

Through early response and ID remapping, the processing efficiency of the write request end is significantly improved, the utilization of cross-chip bandwidth is improved, the response delay of cross-chip requests is reduced, and non-blocking cross-chip transmission is realized.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118193444B_ABST
    Figure CN118193444B_ABST
Patent Text Reader

Abstract

The present invention belongs to the field of chip communication technology, and specifically discloses a structure and a transmission method for cross-chip transmission. The method provided by the present invention realizes the early response of write requests during cross-chip transmission, and releases the resources occupied by the main chip for issuing requests in advance, which not only improves the utilization rate of cross-chip bandwidth, reduces the response delay of cross-chip requests and improves the processing efficiency of the write request end; it can also filter out requests that have been responded to in advance by the source device, thereby reducing the inter-chip communication bandwidth occupied by the response. In addition, the method provided by the present invention can realize non-blocking execution of some requests, effectively improving the concurrency of cross-chip transmission.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of chip communication, and in particular relates to a structure and a transmission method for cross-chip transmission. Background Art

[0002] Since some on-chip memories take a long time to complete a transfer, the master device that issues the transfer request needs to experience a relatively large delay before it can actually receive a response indicating that the transfer is complete. By implementing a circuit that can respond to transfers in advance on the chip, the master device will issue a write transfer that can be responded to in advance, and can send a response to the master device without waiting until the data is actually written to the destination. Therefore, this method can significantly improve chip transmission efficiency.

[0003] However, on the one hand, the current designs for early response to write transfers are basically aimed at on-chip memories, and there is a lack of early response designs for write transfers for cross-chip accesses. On the other hand, due to the very large cross-chip access traffic in large-scale computing clusters, the current cross-chip access is to reduce communication delays by increasing inter-chip bandwidth and the number of channels, which will greatly waste chip resources and increase power consumption. Summary of the invention

[0004] In order to solve at least one of the problems mentioned in the above background technology, the present invention proposes a structure and a transmission method for cross-chip transmission.

[0005] A cross-chip transmission structure, comprising:

[0006] The chip is configured to send requests, receive requests and process these requests, and specifically includes a master agent and a slave agent, the master agent has two channels, namely, a channel for sending requests and a channel for receiving responses; the slave agent has two channels, namely, a channel for receiving requests from a peer chip and a channel for sending responses to the peer chip;

[0007] A write request detector that can respond in advance is configured to parse the Response field in the request and forward the request according to the result of the parsing, wherein the Response field stores information on whether the request is responded in advance;

[0008] an early response generator configured to generate an early response based on a received request ID;

[0009] A sending multiplexer configured to send the request and the real response generated by the chip to the data sending module, and to limit only one of them to be sent at the same time;

[0010] A response multiplexer, which is configured to send the responses corresponding to the requests to the chip and limit only one of them to be sent at a time;

[0011] an advance response parser configured to parse a response sent by the chip and distinguish whether the response is a real response or an advance response;

[0012] The ID remapping table is configured to create remapping entries of the original ID and the new ID for the received request and modify the corresponding status; receive the real response and the early response and modify the corresponding status. Its structure is as follows:

[0013] {"Remapped ID", "Original ID", "Early Response(0 / 1)", "In-use(0 / 1)"}

[0014] in,

[0015] Remapped ID indicates the remapped ID, i.e., the "new ID". Original ID indicates the original ID before remapping, i.e., the "original ID". Early Response indicates whether the request is an early response request. In-use indicates whether the remapped entry is being used.

[0016] A data sending module, configured to send the request and response generated by the chip to the opposite chip;

[0017] The data receiving module is configured to receive requests and responses generated by the opposite chip.

[0018] A method for cross-chip transmission, comprising the steps of:

[0019] Step S1, the chip 1 master agent sends a request to the early response request detector;

[0020] Step S2, the early response request detector receives the request and parses the Response field of the request, wherein the Response field stores information on whether the request is an early response;

[0021] Step S3, executing different steps according to the analysis result. Specifically, if it is a request that can be responded to in advance, executing step A first and then executing step B, otherwise executing step B directly;

[0022] Implementing a circuit that can respond to transmissions in advance on the chip can significantly improve chip transmission efficiency. However, the current designs for early response to write transmissions are basically for on-chip memory, and there is a lack of early response designs for write transmissions for cross-chip accesses. The present invention designs a chip cross-chip transmission structure and a corresponding transmission method, which can release the resources occupied by the main chip to issue a request in advance, greatly improve the processing efficiency of the write request end, improve the utilization of the cross-chip bandwidth, and reduce the response delay of the cross-chip request.

[0023] Step A includes 2 routes, where the execution process of route 1 is as follows:

[0024] First, the early response request detector sends the request to the transmission multiplexer; then, the transmission multiplexer receives the request sent by the early response request detector and sends the request to the chip 1 data transmission module;

[0025] The execution process of route 2 is as follows:

[0026] First, the request can be copied in the early response request detector and the copied request is sent to the early response generator; then, the early response generator receives the copied request and generates an early response based on the copied request; then, the early response generator sends the generated early response to the response multiplexer, and finally, the response multiplexer receives the early response sent by the early response generator and sends the early response to the chip 1 master agent.

[0027] After executing the above steps, the request sent by the chip 1 master agent has been responded to and processed, and the chip 2 still needs to receive and process the request sent by the chip 1 master agent.

[0028] Currently, cross-chip access is achieved by increasing the inter-chip bandwidth and the number of channels to reduce communication delay, which greatly wastes chip resources and increases power consumption. The method provided by the present invention can implement non-blocking execution for some requests, effectively improving the concurrency of cross-chip transmission.

[0029] The execution process of step B is as follows:

[0030] First, the chip 1 data sending module sends the request to the chip 2 data receiving module; then, the chip 2 data receiving module receives the request and sends the request to the demultiplexer; then, the demultiplexer receives the request and sends the request to the ID remapping table; finally, the ID remapping table performs ID remapping.

[0031] The ID remapping process is as follows:

[0032] First, parse out the original ID of each request, further, record the parsed original ID in the ID remapping table, and search for an unused new ID in the ID remapping table; then, the new ID agent that is found is used to parse out the original ID, mark the status of the new ID as being in use, and send a request to chip 2 from the agent through the new ID; then, chip 2 receives the request from the agent, processes the request and generates a response according to the request, and further, chip 2 sends the generated response to the early response parser; finally, the early response parser receives the response and determines whether the request corresponding to the response has been generated for an early response according to the response. If an early response has been generated, the status of the parsed original ID and the new ID found in the ID remapping table is changed from being used to being unused, and the response process ends without executing step C;

[0033] Otherwise, the new ID is replaced with the original ID, the status of the original ID and the new ID is changed from used to unused, and step C is executed.

[0034] According to the designed chip cross-chip transmission structure and the corresponding transmission method, the present invention designs an ID remapping table to execute the ID remapping process on the basis of realizing early response to write transmission, which can filter out requests that have been responded to in advance by the source device, thereby reducing the inter-chip communication bandwidth occupied by the response.

[0035] After executing the above steps, the chip 2 slave agent has processed the request and generated a corresponding response, which will be sent to the chip 1 master agent through step C.

[0036] The execution process of step C is as follows:

[0037] First, after the original ID replaces the new ID, the ID remapping table sends the response of the original ID to the sending multiplexer, the sending multiplexer receives the response and sends the response to the chip 2 data sending module; then, the chip 2 data sending module receives the response and sends the response to the chip 1 data receiving module, the chip 1 data receiving module receives the response and sends the response to the demultiplexer; finally, the demultiplexer receives the response and sends the response to the response multiplexer, the response multiplexer receives the response and sends the response to the chip 1 master agent.

[0038] After executing the above steps, the chip 1 master agent has processed the requests that cannot be responded to in advance.

[0039] The present invention proposes a cross-chip transmission structure and a transmission method, which have the following beneficial effects compared with the existing technology:

[0040] The present invention designs a chip cross-chip transmission structure and a corresponding transmission method, which can release the resources occupied by the main chip to issue a request in advance, greatly improve the processing efficiency of the write request end, improve the utilization rate of the cross-chip bandwidth, and reduce the response delay of the cross-chip request;

[0041] The method provided by the present invention can filter out requests that have been responded to in advance by the source device, thereby reducing the inter-chip communication bandwidth occupied by the response;

[0042] The method provided by the present invention can realize non-blocking execution of some requests, effectively improving the concurrency of cross-chip transmission. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is a schematic diagram of the structure and process of the present invention. DETAILED DESCRIPTION

[0044] In order to make the purpose and features of the present invention more obvious and easy to understand, the present technical solution is described in detail below through embodiments and in conjunction with the accompanying drawings.

[0045] according to Figure 1 The structure and corresponding process shown in the figure are as follows:

[0046] A cross-chip transmission structure, comprising:

[0047] The chip is configured to send requests, receive requests and process these requests, and specifically includes a master agent and a slave agent, the master agent has two channels, namely, a channel for sending requests and a channel for receiving responses; the slave agent has two channels, namely, a channel for receiving requests from a peer chip and a channel for sending responses to the peer chip;

[0048] In the actual implementation process, the master agent's responsibility is to issue requests and receive responses to these requests, and the slave agent's responsibility is to receive requests, process them, and then issue responses to these requests.

[0049] A write request detector that can respond in advance is configured to parse the Response field in the request and forward the request according to the result of the parsing, wherein the Response field stores information on whether the request is responded in advance;

[0050] An early response generator is configured to generate an early response based on a received request ID.

[0051] The sending multiplexer is configured to send the request and the real response generated by the chip to the data sending module, and limit only one of them to be sent at the same time.

[0052] The response multiplexer is configured to send the responses corresponding to the requests to the chip and limit only one of them to be sent at the same time.

[0053] The advance response parser is configured to parse the response sent by the chip and distinguish whether the response is a real response or an advance response.

[0054] The ID remapping table is configured to create remapping entries of the original ID and the new ID for the received request and modify the corresponding status; receive the real response and the early response and modify the corresponding status. Its structure is as follows:

[0055] {"Remapped ID", "Original ID", "Early Response(0 / 1)", "In-use(0 / 1)"}

[0056] in,

[0057] Remapped ID indicates the remapped ID, that is, the "new ID", Original ID indicates the original ID before remapping, that is, the "original ID", Early Response indicates whether the request is an early response request, and In-use indicates whether the remap entry is being used.

[0058] The data sending module is configured to send the request and response generated by the chip to the opposite chip.

[0059] The data receiving module is configured to receive requests and responses generated by the opposite chip.

[0060] A method for cross-chip transmission, comprising the steps of:

[0061] Step S1, the chip 1 master agent sends a request to the early response request detector.

[0062] Step S2: The early response request detector receives the request and parses the Response field of the request, wherein the Response field stores information on whether the request is an early response.

[0063] Step S3, executing different steps according to the analysis result. Specifically, if it is a request that can be responded to in advance, execute step A before executing step B, otherwise execute step B directly.

[0064] Step A includes 2 routes, where the execution process of route 1 is as follows:

[0065] First, the early response request detector sends the request to the transmission multiplexer; then, the transmission multiplexer receives the request sent by the early response request detector and sends the request to the chip 1 data transmission module.

[0066] The execution process of route 2 is as follows:

[0067] First, the request can be copied in the early response request detector and the copied request is sent to the early response generator; then, the early response generator receives the copied request and generates an early response based on the copied request; then, the early response generator sends the generated early response to the response multiplexer, and finally, the response multiplexer receives the early response sent by the early response generator and sends the early response to the chip 1 master agent.

[0068] After executing the above steps, the request sent by the chip 1 master agent has been responded to and processed, and the chip 2 still needs to receive and process the request sent by the chip 1 master agent.

[0069] The execution process of step B is as follows:

[0070] First, the chip 1 data sending module sends the request to the chip 2 data receiving module; then, the chip 2 data receiving module receives the request and sends the request to the demultiplexer; then, the demultiplexer receives the request and sends the request to the ID remapping table; finally, the ID remapping table performs ID remapping.

[0071] The ID remapping process is as follows:

[0072] First, parse out the original ID of each request, further, record the parsed original ID in the ID remapping table, and search for an unused new ID in the ID remapping table; then, the new ID agent that is found is used to parse out the original ID, mark the status of the new ID as being in use, and send a request to chip 2 from the agent through the new ID; then, chip 2 receives the request from the agent, processes the request and generates a response according to the request, and further, chip 2 sends the generated response to the early response parser; finally, the early response parser receives the response and determines whether the request corresponding to the response has been generated for an early response according to the response. If an early response has been generated, the status of the parsed original ID and the new ID found in the ID remapping table is changed from being used to being unused, and the response process ends without executing step C;

[0073] Otherwise, the new ID is replaced with the original ID, the status of the original ID and the new ID is changed from used to unused, and step C is executed.

[0074] After executing the above steps, the chip 2 slave agent has processed the request and generated a corresponding response, which will be sent to the chip 1 master agent through step C.

[0075] The execution process of step C is as follows:

[0076] First, after the original ID replaces the new ID, the ID remapping table sends the response of the original ID to the sending multiplexer, the sending multiplexer receives the response and sends the response to the chip 2 data sending module; then, the chip 2 data sending module receives the response and sends the response to the chip 1 data receiving module, the chip 1 data receiving module receives the response and sends the response to the demultiplexer; finally, the demultiplexer receives the response and sends the response to the response multiplexer, the response multiplexer receives the response and sends the response to the chip 1 master agent.

[0077] After executing the above steps, the chip 1 master agent has processed the requests that cannot be responded to in advance.

[0078] So far, according to the method disclosed in the present invention, the working process of the present invention has been implemented once.

[0079] It should be noted that the serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments. And the terms "including", "comprising" or any other variants thereof in this article are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "including a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0080] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0081] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for cross-chip transmission, characterized in that: Includes steps: Step S1, the chip 1 master agent sends a request to the early response request detector; Step S2, the early response request detector receives the request and parses the Response field of the request, wherein the Response field stores information on whether the request is an early response; Step S3, executing different steps according to the analysis result. Specifically, if it is a request that can be responded to in advance, executing step A first and then executing step B, otherwise executing step B directly; Wherein, step A described in step S3 includes 2 routes, wherein route 1 includes the steps: Step A11, the request detector may be responded to in advance to send the request to the transmission multiplexer; Step A12, the sending multiplexer receives the request sent by the early response request detector, and sends the request to the chip 1 data sending module; Route 2 includes the steps: Step A21, the request may be copied in the early response request detector, and the copied request may be sent to the early response generator; Step A22, the early response generator receives the duplicate request and generates an early response according to the duplicate request; Step A23, the early response generator sends the generated early response to the response multiplexer, Step A24, the response multiplexer receives the early response sent by the early response generator, and sends the early response to the chip 1 master agent; Step B described in step S3 comprises the steps of: Step B1, the chip 1 data sending module sends a request to the chip 2 data receiving module; Step B2, the data receiving module of chip 2 receives the request and sends the request to the demultiplexer; Step B3, the demultiplexer receives the request and sends the request to the ID remapping table; Step B4, ID remapping is performed by the ID remapping table, the steps are as follows: Step B401, parse out the original ID of each request; Step B402, recording the parsed original ID in the ID remapping table, and searching for an unused new ID in the ID remapping table; Step B403, the proxy resolves the original ID found by the new ID, marks the status of the new ID as in use, and sends a request to the chip 2 from the proxy through the new ID; Step B405, chip 2 receives the request from the agent, processes the request and generates a response according to the request; Step B406, chip 2 sends the generated response to the early response parser; Step B407: The early response parser receives the response, parses the response, and executes different steps according to the parsed result.

2. The method for cross-chip transmission according to claim 1, characterized in that: The different steps described in step B407 are executed according to the parsing result, specifically including: First, determine whether the request corresponding to the response has been generated in advance. If so, change the status of the parsed original ID and the new ID found from used to unused in the ID remapping table, and the response process ends without executing step C. Otherwise, the new ID is replaced with the original ID, the status of the original ID and the new ID is changed from used to unused, and step C is executed; 3. The method for cross-chip transmission according to claim 2, characterized in that: The step C comprises the steps of: Step C1, after the original ID replaces the new ID, the ID remapping table sends the response of the original ID to the transmission multiplexer; Step C2, the sending multiplexer receives the response and sends the response to the chip 2 data sending module; Step C3, the chip 2 data sending module receives the response and sends the response to the chip 1 data receiving module; Step C4, the chip 1 data receiving module receives the response and sends the response to the demultiplexer; Step C5, the demultiplexer receives the response and sends the response to the response multiplexer; Step C6, the response multiplexer receives the response and sends the response to the chip 1 master agent.

4. A cross-chip transmission structure, characterized in that: include: a chip configured to issue requests, receive requests, and process those requests; A write request detector that can respond in advance is configured to parse the Response field in the request and forward the request according to the result of the parsing, wherein the Response field stores information on whether the request is responded in advance; an early response generator configured to generate an early response based on a received request ID; A sending multiplexer configured to send the request and the real response generated by the chip to the data sending module, and to limit only one of them to be sent at the same time; A response multiplexer, which is configured to send the responses corresponding to the requests to the chip and limit only one of them to be sent at a time; an advance response parser configured to parse a response sent by the chip and distinguish whether the response is a real response or an advance response; The ID remapping table is configured to create remapping entries of the original ID and the new ID for the received request and modify the corresponding status; receive the real response and the early response and modify the corresponding status. Its structure is as follows: {"Remapped ID", "Original ID", "Early Response(0 / 1)", "In-use(0 / 1)"} where, Remapped ID indicates the remapped ID, i.e., the "new ID", Original ID indicates the original ID before remapping, i.e., the "original ID", Early Response indicates whether the request is an early response request, and In-use indicates whether the remap entry is being used; A data sending module, configured to send the request and response generated by the chip to the opposite chip; The data receiving module is configured to receive requests and responses generated by the opposite chip.

5. The cross-chip transmission structure according to claim 4, characterized in that: The chip specifically includes a master agent and a slave agent. The master agent has two channels, namely, a channel for sending requests and a channel for receiving responses; the slave agent has two channels, namely, a channel for receiving requests from the opposite chip and a channel for sending responses to the opposite chip.

Citation Information

Patent Citations

  • Data transmission method of multi-core processor on-chip network system

    CN116303221A

  • Chip response method, target chip and computer readable storage medium

    CN116484406A