CN-121210276-B - UCIe performance information acquisition method, core particle, chip and data processing equipment

CN121210276BCN 121210276 BCN121210276 BCN 121210276BCN-121210276-B

Abstract

The present disclosure relates to a UCIe performance information acquisition method, a core, a chip and a data processing device, the method includes collecting, by a DMA module built in a first core, transmission performance information of a UCIe module of the first core, forming at least one performance analysis data chain based on time ordering, wherein the transmission performance information includes at least one of first transmission performance information of the first core accessing a second core, second transmission performance information of the second core accessing the first core, and abnormal transmission performance information, and transmitting, by the DMA module, the at least one performance analysis data chain to a system bus to store the at least one performance analysis data chain in a system bus memory. The method and the device realize the real-time acquisition and storage of the transmission performance information in time sequence, and the obtained UCIe performance data change data detail along with time is more detailed, so that the accuracy and the reliability of UCIe performance state analysis results are improved.

Inventors

Request for anonymity
Request for anonymity
Request for anonymity
Request for anonymity

Assignees

上海壁仞科技股份有限公司

Dates

Publication Date: 20260508
Application Date: 20251127

Claims (12)

1. A method for obtaining UCIe performance information, comprising: Acquiring transmission performance information of UCIe modules of a first core particle through a DMA module arranged in the first core particle to form a plurality of performance analysis data chains based on time sequencing, wherein the transmission performance information comprises at least one of first transmission performance information of the first core particle accessing a second core particle, second transmission performance information of the second core particle accessing the first core particle and abnormal transmission performance information; Transmitting the plurality of performance analysis data chains to a system bus through the DMA module so as to store the plurality of performance analysis data chains in a system bus memory, wherein the stored plurality of performance analysis data chains form a complete dynamic change record of the UCIe performance information; Wherein the performance analysis data chain comprises: a start identification field for identifying a header of the performance analysis data chain; an end identification field for identifying a tail of the performance analysis data chain; a timestamp identification field located after the start identification field for identifying a time of generation of the performance analysis data chain, and At least one of the first transmission performance information, the second transmission performance information, and the abnormal transmission performance information is located between the timestamp identification field and the end identification field.
2. The UCIe performance information acquisition method according to claim 1, wherein: The first transmission performance information includes at least one of a first average delay, a first maximum delay, a first minimum delay, and a first bandwidth efficiency of a plurality of access requests sent from UCIe to UCIe of the first core to the second core within a set time range after a time identified by the timestamp identification field; The second transmission performance information includes at least one of a second average delay, a second maximum delay, a second minimum delay, and a second bandwidth efficiency of the second core for multiple accesses to the first core from receiving UCIe the access request to issuing UCIe the access response within a set time range after the time identified by the timestamp identification field; The abnormal transmission performance information comprises abnormal transmission performance data of at least zero abnormal accesses in a set time range after the moment identified by the timestamp identification field, wherein the abnormal transmission performance data comprises at least one of delay information and access addresses of the abnormal accesses.
3. The UCIe performance information acquisition method according to claim 1, wherein: The starting identification field comprises index information, and the index information is used for representing the position of the performance analysis data chain where the starting identification field is located in the plurality of performance analysis data chains; The end identification field comprises the quantity information of the abnormal transmission performance data contained in the abnormal transmission performance information in the performance analysis data chain where the end identification field is located.
4. The method of claim 1, wherein the UCIe performance information obtaining method further comprises: Recording the request sending time of each access of the first core grain to the second core grain; Acquiring response receiving time of each access of the first core particle to the second core particle; obtaining the delay of each access of the first core particle to the second core particle according to the obtained response receiving time and the recorded request sending time; Obtaining at least one of a first average delay, a first maximum delay and a first minimum delay of multiple accesses of the first core grain to the second core grain according to the delay of each access of the first core grain to the second core grain; recording the byte number carried by each access request of the first core grain to the second core grain; Obtaining a first bandwidth efficiency of the first core particle to the multi-access of the second core particle according to the byte number carried by the request of the first core particle to the whole access of the second core particle completed in a preset time period and the duration of the preset time period; And obtaining the first transmission performance information through at least one of a first average delay, a first maximum delay, a first minimum delay and a first bandwidth efficiency of multiple accesses of the first core grain to the second core grain.
5. The method of claim 1, wherein the UCIe performance information obtaining method further comprises: Recording UCIe request receipt times for each access of the second core to the first core; UCIe response sending time of each access of the second core grain to the first core grain is obtained; obtaining the local delay of each access of the second core grain to the first core grain according to the obtained UCIe response sending time and the recorded UCIe request receiving time; Obtaining at least one of second average delay, second maximum delay and second minimum delay of multiple accesses of the second core grain to the first core grain according to the local delay of each access of the second core grain to the first core grain; Recording the byte number carried by UCIe requests of each access of the second core grain to the first core grain; obtaining second bandwidth efficiency of multiple accesses of the second core particle to the first core particle according to the byte number carried by UCIe requests of all accesses of the second core particle to the first core particle completed in a preset time period and the duration of the preset time period; and obtaining the second transmission performance information through at least one of second average delay, second maximum delay, second minimum delay and second bandwidth efficiency of the plurality of accesses of the second core grain to the first core grain.
6. The UCIe performance information acquisition method according to claim 1, wherein: The system bus includes at least one of a memory bus and a control bus within the first die.
7. The UCIe performance information obtaining method of claim 6, further comprising: configuring a target system bus object sent by the DMA module, so that the DMA module sends the performance analysis data chains to a configured main bus, and sends the plurality of performance analysis data chains to a configured auxiliary bus under the condition that the performance analysis data chains are failed to be sent to the main bus; Wherein the main bus is one of the memory bus and the control bus, and the auxiliary bus is the other of the memory bus and the control bus.
8. The UCIe performance information acquisition method according to claim 1, wherein: The system bus memory is provided with a performance analysis data storage space, and the performance analysis data chains are stored in the performance analysis data storage space; the storage mode of the plurality of performance analysis data chains in the performance analysis data storage space is configured to be one of an uncovered storage mode, a ring covered storage mode and a specified number of storage modes; wherein, in a case where the storage mode is configured as the uncovered storage mode, the plurality of performance analysis data chains are sequentially stored in the performance analysis data storage space in order, and when the performance analysis data storage space is full, the storage is stopped; in the case where the storage mode is configured as the ring overlay storage mode, the plurality of performance analysis data chains are sequentially stored in the performance analysis data storage space in order, and when the performance analysis data storage space is full, the latest performance analysis data chain overlays the earliest performance analysis data chain position stored in the performance analysis data storage space; In the case where the storage mode is configured as the specified number of storage modes, the plurality of pieces of performance analysis data chains are sequentially stored in the performance analysis data storage space in order until the storage of the last piece of performance analysis data chain of the specified number is completed, the storage is stopped.
9. The UCIe performance information acquisition method according to claim 1, wherein: The access includes at least one of a read operation, a write operation, an atomic operation.
10. A core particle, comprising: UCIe modules; The DMA module is used for collecting transmission performance information of the UCIe module, forming a plurality of performance analysis data chains based on time sequence, and sending the plurality of performance analysis data chains to a system bus so as to store the plurality of performance analysis data chains in a system bus memory, wherein the stored plurality of performance analysis data chains form a complete dynamic change record of UCIe performance information; Wherein the performance analysis data chain comprises: a start identification field for identifying a header of the performance analysis data chain; an end identification field for identifying a tail of the performance analysis data chain; a timestamp identification field located after the start identification field for identifying a time of generation of the performance analysis data chain, and At least one of the first transmission performance information, the second transmission performance information, and the abnormal transmission performance information is located between the timestamp identification field and the end identification field.
11. A chip, comprising: A first core particle and a second core particle interconnected using UCIe protocol, at least one of the first core particle and the second core particle being the core particle of claim 10.
12. A data processing device comprising a chip as claimed in claim 11.

Description

UCIe performance information acquisition method, core particle, chip and data processing equipment Technical Field The disclosure relates to the technical field of UCIe performance analysis, in particular to a UCIe performance information acquisition method, a core particle, a chip and data processing equipment. Background UCIe (Universal Chiplet Interconnect Express, universal core interconnect protocol) is an emerging standard for core (Chiplet) interconnect, providing a high bandwidth, low latency, low power interconnect scheme that aims to solve the interconnect problem between cores produced by different vendors. UCIe is used as an interface of the core particles, and because the matching relation of request response among different core particles or crystal grains (Die) is easy to analyze, when the performance analysis and measurement are carried out, the overall performance and bandwidth delay performance of the core particles to the system are easy to analyze from UCIe, and the transmission efficiency of the core particle connection is provided for analysis and test staff. Disclosure of Invention In view of this, the disclosure provides a UCIe performance information acquisition method, a core, a chip and a data processing device, so as to realize real-time recording of UCIe performance information of the core in a longer period, and further help to improve accuracy and reliability of UCIe performance analysis results based on UCIe performance information recorded in the longer period. According to an aspect of the embodiments of the present disclosure, there is provided a UCIe performance information acquisition method, including: acquiring transmission performance information of UCIe modules of a first core particle through a DMA module arranged in the first core particle to form at least one performance analysis data chain based on time sequencing, wherein the transmission performance information comprises at least one of first transmission performance information of the first core particle accessing a second core particle, second transmission performance information of the second core particle accessing the first core particle and abnormal transmission performance information; the at least one performance analysis data chain is sent to a system bus by the DMA module to store the at least one performance analysis data chain in a system bus memory. In one possible embodiment, the performance analysis data chain comprises: a start identification field for identifying a header of the performance analysis data chain; an end identification field for identifying a tail of the performance analysis data chain; a timestamp identification field located after the start identification field for identifying a time of generation of the performance analysis data chain, and At least one of the first transmission performance information, the second transmission performance information, and the abnormal transmission performance information is located between the timestamp identification field and the end identification field. In one possible implementation, the first transmission performance information includes at least one of a first average delay, a first maximum delay, a first minimum delay, and a first bandwidth efficiency of a plurality of access requests from UCIe to UCIe of the first core to the second core within a set time range after the time identified by the timestamp identification field; The second transmission performance information includes at least one of a second average delay, a second maximum delay, a second minimum delay, and a second bandwidth efficiency of the second core for multiple accesses to the first core from receiving UCIe the access request to issuing UCIe the access response within a set time range after the time identified by the timestamp identification field; The abnormal transmission performance information comprises abnormal transmission performance data of at least zero abnormal accesses in a set time range after the moment identified by the timestamp identification field, wherein the abnormal transmission performance data comprises at least one of delay information and access addresses of the abnormal accesses. In a possible implementation manner, the start identification field includes index information, where the index information is used to characterize a position of a performance analysis data chain where the start identification field is located in the at least one performance analysis data chain; The end identification field comprises the quantity information of the abnormal transmission performance data contained in the abnormal transmission performance information in the performance analysis data chain where the end identification field is located. In one possible implementation manner, the UCIe performance information acquisition method further includes: Recording the request sending time of each access of the first core grain to the second core grain; Acquiring r